• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    The embodiments of the present invention provide a method and a device for predicting the instability stages of a rock mass based on a fusion of multiple features, and belong to the technical field of rock mass instability prediction. . The method includes: acquiring acoustic emission signals from a rock mass; performing hybrid domain characteristic extraction on the acquired acoustic emission signals; obtaining prediction errors according to each characteristic parameter and a predefined BP neural network model; establish a D-S proof theory identification framework, in which the identification framework includes multiple identification sub-frames in one-to-one correspondence with various stages of rock mass instability; constructing basic probability distribution functions of the characteristic parameters according to the prediction errors and calculating values of degree of confidence, corresponding to the various stages of rock mass instability, of the characteristic parameters; merge all feature parameters into the same identification subframe through DS proof theory according to obtained confidence level values, and predict rock mass instability stages based on feature parameters merged. The present invention achieves accurate prediction of the various stages of rock mass instability, and the problem of imprecise prediction through a single feature parameter is effectively solved.
    公开号:BE1027019B1
    申请号:E20205405
    申请日:2020-06-08
    公开日:2021-06-25
    发明作者:Huiming Chen；Qingling Wu；Fan Shao；Xiaoyan Luo；Xianghai Huang
    申请人:Univ Jiangxi Sci & Technology；
    IPC主号:
    专利说明:

    METHOD AND DEVICE FOR PREDICTING THE STAGES OF INSTABILITY OF A ROCKY MASS ON THE BASIS OF A FUSION
    OF MULTIPLE CHARACTERISTICS Reference to Related Applications This application claims the benefit of Chinese Patent Application No. 201910638983.4, filed on July 16, 2019, entitled "PROCESS AND DEVICE FOR PREDICTING THE STAGES OF INSTABILITY OF AN
    ROCKY MASS BASED ON A FUSION OF MULTIPLE CHARACTERISTICS ”, which is incorporated by reference as if set out therein at full length. Field of the Invention The present invention relates to the technical field of predicting the instability of a rock mass, in particular to a method for predicting the instability stages of a rock mass on the basis of a melting of rock. multiple characteristics, and a device for predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics. Background of the Invention As a heterogeneous and complex geological material, a rock mass exhibits non-linear and anisotropic mechanical properties, which makes it difficult to describe the relationship between the internal state of the rock mass and its mechanical characteristics. inherent. The bedrock loading instability fracture process is actually the continuous process whereby primary cracks widen into micro-cracks, and eventually macro-fracturing occurs. When rock mass instability occurs, the external representation shapes of the rock mass are simultaneously reflected on multiple feature domains. A classic prediction model usually extracts only one characteristic for a prediction analysis. The single characteristic cannot fully describe the internal state of rock mass instability. There is a certain feature that is accurate in describing a certain stage of rock mass instability, but cannot accurately describe another stage, resulting in a reduction in the accuracy rate of a prediction result. Summary of the Invention An object of the embodiments of the present invention is to provide a method and a device for predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics in order to solve the problem. of imprecise prediction of rock mass instability stages through a single feature.
    To achieve the above object, in a first aspect of the present invention, the method of predicting instability stages of a rock mass on the basis of a fusion of multiple characteristics is provided, the method comprising: acquiring acoustic emission signals from a rock mass; perform hybrid domain characteristic extraction on the acquired acoustic emission signals, and extract multiple characteristic parameters with significant category difference information from parameter characteristics, time domain characteristics and / or characteristics of frequency domain respectively; obtain prediction values representing the stages of rock mass instability according to each feature parameter and a predefined BP neural network model, and obtain prediction errors according to the prediction values and expected output values;
    establish a D-S proof theory identification framework, in which the identification framework includes multiple identification subframes in one-to-one correspondence with various stages of rock mass instability; construct basic probability distribution functions of feature parameters according to prediction errors, and calculate confidence values, corresponding to various stages of rock mass instability, of feature parameters through distribution functions basic probability of the characteristic parameters; merge all feature parameters into the same identification subframe through DS proof theory according to the degree of confidence values, corresponding to the various stages of rock mass instability, of the feature parameters to obtain merged characteristic parameters; calculating degree of confidence values, corresponding to the various stages of rock mass instability, of the merged feature parameters, and determining the stage of rock mass instability corresponding to the maximum value among the degree of confidence values as prediction result.
    In a second aspect of the present invention, a device for predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics is provided, the device comprising: a data acquisition module for acquiring the data. sound emission signals from the rock mass; a characteristic extraction module for performing hybrid domain characteristic extraction on the acquired acoustic emission signals, and extracting the multiple characteristic parameters with the significant category difference information from the parameter characteristics, the characteristics of time domain and / or frequency domain characteristics respectively; a first prediction module for obtaining the prediction values representing the stages of rock mass instability according to each feature parameter and the predefined BP neural network model, and obtaining the prediction errors according to the prediction values and the output values expected; an identification framework establishment module for establishing the identification framework of the DS proof theory, in which the identification framework comprises the multiple identification subframes in one-to-one correspondence with the various instability stages of rock mass; a confidence level calculator for constructing the basic probability distribution functions of the characteristic parameters according to the prediction errors, and calculating the confidence level values, corresponding to the various stages of rock mass instability, of the parameters characteristic through the basic probability distribution functions of the characteristic parameters; a feature fusion module to merge all feature parameters into the same identification subframe through DS proof theory according to the degree of confidence values, corresponding to the various stages of rock mass instability, characteristic parameters to obtain the merged characteristic parameters; and a second prediction module for calculating the degree of confidence values, corresponding to the various stages of rock mass instability, of the merged feature parameters, and determining the stage of rock mass instability corresponding to the maximum value among the values of degree of confidence as being the result of prediction.
    In a third aspect of the present invention, a computer readable storage medium is provided, in which instructions are stored on the readable storage medium, and when the instructions execute on a computer, the instructions allow the computer 5 to perform the method of predicting the instability stages of a rock mass on the basis of a fusion of multiple features.
    In the above technical solutions of the present invention, the basic probability distribution functions, corresponding to the various instability stages, of each of the characteristic parameters of the extracted rock mass instability acoustic emission signals are constructed. based on the prediction errors, obtained through the BP neural network model and each characteristic parameter; degrees of confidence, corresponding to the various stages of instability, of each characteristic parameter are obtained through the basic probability distribution functions; and based on the confidence degrees, degrees of support between all of the feature parameters are entered, the degrees of support are used as a weight to weight all of the feature parameters, and a merging of multiple features is performed through DS proof theory. Therefore, an accurate prediction of all stages of rock mass instability is achieved, and the problem of imprecise prediction through a single feature parameter is effectively solved.
    Other features and advantages of the embodiments of the present invention will be described in detail in the following section of the detailed description of the embodiments.
    Brief Description of the Drawings The detailed description of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the detailed description described herein is intended merely to illustrate and explain the present invention, rather than to limit the present invention.
    FIG. 1 is a flowchart of a method of predicting instability stages of a rock mass based on a fusion of multiple characteristics according to an embodiment of the present invention.
    FIG. 2 is a flowchart of construction of basic probability distribution functions of characteristic parameters according to an embodiment of the present invention.
    FIG. 3 is a flowchart of merging all feature parameters into the same identification subframe through D-S proof theory according to one embodiment of the present invention.
    FIG. 4 is a block diagram of a device for predicting the instability stages of a rock mass based on a fusion of multiple features according to an embodiment of the present invention.
    FIG. 5 is a block diagram of a confidence degree calculation module according to an embodiment of the present invention.
    FIG. 6 is a block diagram of a feature fusion module according to an embodiment of the present invention. Joint Drawing Marks Instructions Data Acquisition Module 100, Feature Extraction Module 200, First Prediction Module 300, Identification Frame Establishment Module 400, Confidence Calculation Module 500, feature fusion 600, second prediction module 700, first distance computation sub-module 510, correlation coefficient function construction sub-module 520, base probability distribution function construction sub-module 530, second Distance Calculation Submodule 610, Support Degree Function Building Submodule 620, Feature Parameter Update Submodule 630, Feature Merge Submodule 640.
    Detailed Description of Embodiments Hereinafter, specific embodiments of the present invention are described in detail with reference to the drawings.
    It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not intended to limit the present invention.
    In the embodiments of the present invention, the terms "comprising", "comprising" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, a method, a product or a A device comprising a series of elements does not include only the elements, but also includes other elements which are not clearly enumerated, or additionally comprises elements inherent in the process, method, product or device.
    In a case without further limitations, the elements defined by "including a ..." do not exclude the fact that the process, process, product or device comprising the element also has other elements. identical.
    As shown in FIG. 1, one embodiment of the present invention provides a method of predicting instability stages of a rock mass based on a fusion of multiple features, the method comprising: acquiring acoustic emission signals from a rock mass. rocky mass; perform hybrid domain characteristic extraction on the acquired acoustic emission signals, and extract multiple characteristic parameters with significant category difference information from parameter characteristics, time domain characteristics and / or characteristics of frequency domain respectively; obtain prediction values representing the instability stages of the rock mass according to each feature parameter and a predefined BP neural network model, and obtain prediction errors according to the prediction values and expected output values; establish an identification framework for D-S proof theory, in which the identification framework includes multiple identification subframes in one-to-one correspondence with the various stages of rock mass instability; construct basic probability distribution functions of feature parameters according to prediction errors, and calculate confidence values, corresponding to various stages of rock mass instability, of feature parameters through distribution functions basic probability of the characteristic parameters; merge all feature parameters into the same identification subframe through DS proof theory according to the degree of confidence values, corresponding to the various stages of rock mass instability, of the feature parameters to obtain merged characteristic parameters; calculating confidence level values, corresponding to the various stages of rock mass instability, of the merged feature parameters and determining a stage of rock mass instability corresponding to the maximum confidence level value as a prediction result.
    Thus, the basic probability distribution functions, corresponding to the various instability stages, of each characteristic parameter of the extracted rock mass instability acoustic emission signals are constructed according to the prediction errors of each characteristic parameter at the level of the corresponding instability stages, in which the prediction errors are obtained through the predefined BP neural network model based on the characteristic parameters; degrees of confidence, corresponding to the various stages of instability, of each characteristic parameter are obtained through the basic probability distribution functions; and, based on the degrees of confidence, degrees of support between any two parameters of all of the characteristic parameters are entered, and the degrees of support are used as a weight to weight the various characteristic parameters, and a merge of Multiple characteristics is carried out through the theory of evidence DS. Therefore, an accurate prediction of the various stages of rock mass instability is achieved by calculating the merged feature parameters, and the problem of imprecise prediction through a single feature parameter is effectively solved.
    Specifically, as a heterogeneous and complex geological material, the rock mass exhibits nonlinear and anisotropic mechanical properties, which makes it difficult to describe the relationship between the internal state of the rock mass and its inherent mechanical characteristics. The rock mass loading instability fracture process is actually the continuous process whereby primary cracks widen into micro-cracks, and eventually macro-fracturing occurs. A large number of experiments show that the whole process of rock mass instability fracture is accompanied by the phenomenon of acoustic emission, and there are different characteristics of acoustic emission signal in different stages of instability. . The stages of rock mass instability are generally divided into four stages, namely: the stage of compaction stage |, the stage of elastic deformation stage II, the stage of plastic deformation stage III and the stage of post-peak failure stage IV. . By analyzing the characteristic parameters of acoustic emission signals of rock mass instability and predicting the various stages of rock mass instability, the safety concerns caused by rock mass instability can be effectively eliminated. The acoustic emission signals at the four stages of rock mass instability are respectively selected for hybrid domain characteristic parameter extraction in order to avoid the problem of imprecise prediction caused by the lack of ability to characterize the various stages of rock. rock mass instability through a single characteristic.
    According to D-S proof theory, the identification framework is established, and the identification framework includes the multiple identification subframes.
    Each identification sub-frame includes a set of confidence values, corresponding to the stages of rock mass instability represented by the identification sub-frames, of all characteristic parameters.
    In the embodiment, the identification subframes further include values of the uncertainty basis probability distribution function. The identification frame is represented by:
    ‘Ol AV ml EN mm> © ì ml ml dr) m zj SN © = Al} (B), ak} m {P), ne}; where (a). () ae} AD) and ae) are the various identification subframes of the identification frame © in which a1) represents a basic probability distribution function value in the compaction stage, (8) represents a base probability distribution function value in the elastic strain stage, ae) represents a base probability distribution function value in the plastic strain stage, (2) represents a base probability distribution function value in the post-peak breakage stage, and no} represents the value of the basic uncertainty probability distribution function of the proof theory DS.
    All characteristic parameters are included in each identification sub-frame, and each characteristic parameter serves as evidence.
    With each characteristic parameter as input, the predefined BP neural network model outputs the prediction value of the stage of rock mass instability corresponding to each characteristic parameter; the prediction errors are obtained according to the prediction values and the expected output values; according to the prediction values and the prediction errors, the basic probability distribution functions, corresponding to the various identification subframes, of each characteristic parameter are constructed; according to the obtained basic probability distribution functions, the confidence level values, corresponding to the various identification subframes, of each characteristic parameter can be calculated; according to the confidence values of the various characteristic parameters in the same identification subframe, the degrees of support between any two parameters of all the characteristic parameters in the same identification subframe are obtained; the degrees of support are normalized and used as the weights of the basic probability distribution functions corresponding to the various characteristic parameters; each proof is updated through the weighted basis probability distribution functions to obtain new proofs; then all the evidence in the same identifying subframe is merged through D-S evidence theory; degrees of confidence between the merged evidence and each identification sub-frame are calculated, and the larger the confidence level values the higher the correlation between the merged evidence and the identification sub-frames; the stage of rock mass instability corresponding to the identification sub-frame with the highest confidence level of the merged evidence is selected as the prediction result to be output, as well as the stages of mass instability boulder are predicted by extracting the hybrid domain feature parameters from the acoustic emission signals of the rock mass, and greater accuracy is obtained compared to prediction through the feature alone.
    Further, the parameter characteristics include one or more of an overshoot count, a number of acoustic emission events, an energy count, a duration, a rise time, an RMS voltage value, and an effective amplitude value of the acoustic emission signals.
    The time domain characteristics include one or more of a peak value, an average value, a kurtosis value, a standard deviation, a pulse factor, a margin factor and a form factor. wave of acoustic emission signals.
    The frequency domain characteristics include one or more of a main frequency band position, signal energy, and a degree of spectral aggregation and dispersion of the acoustic emission signals.
    In this embodiment, the five parameters including the overshoot count, the energy count, the peak value, the signal energy and the degree of spectral aggregation and dispersion of the rock mass acoustic emission signals. are selected as the characteristic parameters, respectively, in which, the five characteristic parameters represent five sources of evidence denoted by = Es Far Vie U) ef In this way, the different characteristic parameters with the significant difference information from the Parameter characteristics, time domain characteristics and frequency domain characteristics of the rock mass acoustic emission signals are extracted as the evidence, thus effectively avoiding the problem of imprecise prediction caused by the lack of ability to characterize the various stages of rock mass instability through the sole characterist ic, and improving the prediction accuracy.
    In addition, the predefined BP neural network model is obtained by training a BP neural network and genetic algorithm through feature parameters in different stages of rock mass instability.
    More precisely, the BP neural network is built first; a single input-single output prediction model is established with a certain characteristic parameter of the input rock mass acoustic emission signals and the stage of rock mass instability corresponding to the output characteristic parameter; a large number of the acoustic emission signals of the rock mass in the different stages of instability are acquired in advance; the acquired acoustic emission signals of rock mass instability are divided into the four stages of rock mass instability; in each corresponding stage, 20 corresponding groups of acoustic emission signals are intercepted at equal time intervals, the length of each group of acoustic emission signals is 2048 data points, and each data point represents signal data impulse; overshoot count, energy count, peak value, signal energy and degree of spectral aggregation and dispersion are extracted for the 80 intercepted groups of acoustic emission signals correspondingly as being the characteristic parameters; all the extracted feature parameters are divided into a training set and a test set to obtain 80 groups of data samples; 40 groups of data samples are randomly selected as the training set for training the BP neural network, and the remaining 40 groups of data samples are used as the test set to verify the trained BP neural network model, in which the specific ratio of the training set to the test set can be adjusted according to actual situations.
    The BP neural network is trained with each characteristic parameter in the input training set to obtain a prediction value and a prediction error of each characteristic parameter; a BP neural network threshold and weight are adjusted through backpropagation until the output prediction error is less than a set value.
    A classical BP neural network is likely to fall into the local optimal solution in the learning process, which leads to the suboptimal parameter optimization effect of the BP neural network, makes the stability of the neural network poor and affects thus the prediction accuracy.
    Therefore, this embodiment further optimizes the weight and threshold of the BP neural network through the genetic algorithm, thereby effectively preventing the BP neural network from falling into the local optimal solution.
    The optimization process includes: defining population parameters, the population parameters including population size, number of evolutions, probability of crossing and probability of mutation; performing actual encoding of the weight and threshold of the BP neural network; calculate the fitness of individuals, select an optimal individual from the current population according to the fitness of the individuals, perform a crossbreeding and mutation operation until the optimized condition is reached, and select a weight and threshold optimal to update the initial BP neural network weight and threshold.
    As shown in FIG. 2, the construction of the basic probability distribution functions of the characteristic parameters includes: constructing a first distance function which takes the absolute values of the prediction errors as the first distances between the characteristic parameters and actual values, in which the value prediction output from the BP neural network model is> ‚and the actual values at the various stages of rock mass instability are x = fr Ta Na Va {= 11.234; NOT . . AREA, then the first distance function is ten rs = Hi. | . at ; ', which represents the distance between the proof and the real value: according to the first distance function, construct the correlation coefficient functions representing the degrees of correlation between the characteristic parameters and the identification subframes: Hed, Ÿ , IS gp> ad) Lu EE is construct the basic probability distribution functions of the characteristic parameters according to the correlation coefficient functions and the root mean square errors between the prediction values and the expected output values of the neural network model Predefined BP, so that the greater the degrees of correlation, the greater the confidence values obtained by the basic probability distribution functions of the characteristic parameters.
    More precisely, the basic probability distribution functions of the various identification subframes are as follows: ee; Se + E mir) „. The . where, wi represents the basic probability distribution functions of the various identification subframes characterizing the stages of rock mass instability to calculate ”- and zip) ©); OL ue a represents the basic probability distribution functions of uncertainty characterizing the description of the uncertainties of the evidence, and £ is the mean squared error between the prediction value and the expected output value ÏX Es yo of the network model neuronal BP predefined, =. The smaller the first distance obtained, the higher the reliability of the evidence.
    Based on the distances between the prediction values and the actual values of the BP neural network model, the degrees of correlation between the evidence and the identification subframes are obtained, and the reliability of the evidence can be effectively quantified.
    The basic probability distribution function values, corresponding to each stage of the stages | at IV of rock mass instability, of the five proofs are calculated by # 2) as shown in Table 1: Proof Stage m (A) m (B) m (C) m (D) m (©) ‚Result d 'identification OL 0.7077 0.2043 0.1025 0.0737 0.0441 02022 Il 0.2538 0.6242 0.1614 0.1064 0.0729 Il © Il 0.0392 0.0715 0.8207 0.0848 0.0081 Ill IV 0.0478 0.0683 0.1363 0.7868 0.0196 IV | 0.5800 0.3689 0.1846 0.1404 0.1369 | Il 0.2438 0.6316 0.1572 0.1026 0.0676 Il ° 2 II 0.0545 0.0952 0.7629 0.1212 0.0169 Ill IV 0.0737 0.1025 0.2043 0.7077 0 , 0441 IV | 0.5077 0.5706 0.2796 0.2252 0.2916 Il Il 0.0834 0.7868 0.1025 0.0515 0.0121 Il ° 3 I 0.0582 0.1007 0.7500 0.1303 0, 0196 Ill IV 0.0853 0.1174 0.2340 0.6785 0.0576 IV | 0.5861 0.3580 0.1792 0.1359 0.1296 | Il 0.1746 0.5830 0.2884 0.1462 0.0961 Il a I 0.1021 02020 0.6547 0.1379 0.0484 Ill IV 0.2362 0.2913 0.5970 0.5027 0, 3136 Ill | 0.4889 0.7099 0.3385 0.2819 0.4096 Il Il 0.1335 0.6644 0.1928 0.0974 0.0441 Il © II 0.0697 0.1384 0.7317 0.1052 0.0225 Ill IV 0.1359 0.1792 0.3580 0.5861 0.1296 IV Table As shown in Table 1, the maximum value of the basic probability distribution function of the proof “* at stage | is # 4B) and is equal to 0.5706, which indicates that the proof has extracted from the stage | has the highest level of confidence for stage II; the maximum value of the basic probability distribution function of the proof ©: at stage IV is PC) and is equal to 0.5970, which indicates that the proof “+ extracted from stage IV has the highest degree of confidence. high for stage III; the maximum value of the basic probability distribution function of the proof * at stage I is and is equal to 0.7099, indicating that the proof * extracted from stage | has the highest confidence level for stage II. In summary, there are cases of misidentification at the stadium | of proof * 3, at stage IV of proof * and at stage | evidence - indicating that there are instances of misidentification through the unique characteristic, thus failing to accurately identify the various stages of rock mass instability.
    Therefore, all feature parameters in the same identification subframe must be merged through D-S proof theory, as shown in Fig. 3, comprising: calculating the mathematical distance between any two parameters of all characteristic parameters in the same identification subframe according to the confidence values of all characteristic parameters in the identification subframe.
    More precisely, the set of results determined in the framework y. =} Ne x 3 | 25 3 | vo 4 Y his ŸÜTF 60 9 identification © is © = Im da albe Bl they ++, and if ‚,‚. # 7, represents the set of proofs composed of the basic probability distribution function values obtained from * proofs of * focal elements, where ‘+" <= Correspond to a De or °° respectively; and * "" ’represents the basic probability distribution function values of the proof * in 3 _ - x A - - - the identification sub-frame corresponding to, and so on.
    The mathematical distances between different proofs in the same identification subframe are calculated through a distance function Sk * where, ”and * represent the base probability distribution function values of two different proofs in the same identification sub-frame. . PR of The measure of similarity * between the two proofs is obtained according to a. Se.
    distance function above ° 7, in which * * a support degree function is constructed according to *, so that the smaller the mathematical distance between any two proofs, the greater the measure of similarity, that is - ie, the greater the value of degree of support between the Tim}: LL "ARE 3 two proofs, and the function of degree of support" * ** of the proofs is as follows: Tds Vs Feen Normalize the degree function of support to obtain;, Tim} OFEN, 1 A, jm S Tin) JE ‚add and average the degree of support of ©, use the normalized degree of support function as, At} 18,.. tn weight “- '' of 'to get a new distribution function of Rn = s Alm Je> 3 = L2 PS + 3 Sk Ke È 2; probability:>, and weight the base probability distribution function value of each proof in the same identification subframe through the new probability distribution function, and update l The confidence level values of the evidence in Table 1 to obtain new evidence.
    According to the DS proof theory, all new proofs in the same identification subframe are merged to obtain the merged probability distribution function Ju} in particular, so STE = 8 to + to Ss + 8, in which ” * and *: respectively represent the two new corresponding probability distribution functions in the same identification subframe, * * and '”are the focal elements corresponding to *: and # :, K is the conflict coefficient which reflects the degree of mutual conflict between each two tests.
    The higher the value of K, the greater the degree of mutual conflict between the two proofs.
    When K is 1, each two proofs are in complete conflict, the merge rule is not valid, and the two proofs cannot be merged. 1-K is a canonical number, which means that the distribution values of ie; SE LL. . base probability of the set VO} ge base probability distribution values of uncertainty of the merged evidence are assigned to mla) m) mic) m5}; “7,‘ and “respectively in proportion, and the proportion of the base probability distribution values can be, but not limited to, Æ => md {BB} evenly distributed, in which B,. The updated Table 1 is used to merge all evidence into the same identification subframe through the above merge rules, merge “and“:. Ef. . EK EN 2. 3 a. . . to get © “: | then merge Èzet © to obtain ©:, and so on, until all the evidence is merged.
    The result after weighted fusion of the D-S evidence is shown in Table 2.
    Stage Proof of m (A) m (B) m (C) m (D) m (O9) Identification fusion result 1 ene2 06727 0.2255 0.0835 0.0183 01 | e ,, e2, e3 0.6786 0.2508 0.0651 0.0056 0 | e ,, e2, e3, e4 0.7512 0.2108 0.0368 0.0013 0 | e1.82, e3, e4, es 0.7265 0.2440 0.0291 0.0004 0 | ee, 0.2047 0.6677 0.1078 0.0198 0 Il | e ,, e2, e3 0.0658 0.8901 0.0417 0.0023 0 Il € 1.62.6:, 64 0.0352 0.9338 0.0304 0.0005 0 Il € 1.62.6: , 8465 0.0140 0.9707 0.0152 0.0001 0 Il ee, 0.0263 0.0547 0.8953 0.0237 0 IN Ill €, 62.6; 0.0101 0.0241 0.9608 0.0049 0 IN € 1.62.6:, 64 0.0046 0.0143 0.9800 0.0011 0 IN € 1, € 2.6:, 64.65 0 , 0014 0.0059 0.9925 0.0002 0 IN ee, 0.1042 0.1500 0.3152 0.4305 0 IV €, 62.6; 0.0935 0.1315 0.2863 0.4987 0 IV V e1, e2, e3, e4 0.1355 0.1584 0.2409 0.4653 0 IV e1.82, e3, e4, es 0.1174 0, 1339 0.2358 0.5129 0 IV Table 2 As shown in Table 2, evidence subjected to weighted fusion also has good processing capacity in the state where the & and *% evidence at stage | in Table 1 cannot be identified with precision.
    In addition, for the base probability distribution functions after merging the evidence at Stage II and Stage III, the base probability distribution function values are gradually increased with the more merging evidence.
    This indicates that the weighted fusion of the D-S proofs is characterized in that the more the features merged, the higher the rate of prediction accuracy.
    When the high conflict * evidence in Stage IV is merged, the weighted merge of the DS evidence can effectively overcome the lack of unpredictability in a classic DS evidence merger due to the presence of high conflict evidence, and hence the rate of prediction accuracy is increased.
    As shown in FIG. 4, in a second aspect of the present invention, a device for predicting the instability stages of a rock mass based on a fusion of multiple characteristics is provided, the device comprising: a data acquisition module 100 to acquire the acoustic emission signals of the rock mass; a feature extraction module 200 for performing hybrid domain feature extraction on the acquired acoustic emission signals, and extracting the multiple feature parameters with the significant category difference information from the parameter features, characteristics time domain and / and frequency domain characteristics respectively;
    a first prediction module 300 for obtaining the prediction values representing the stages of rock mass instability according to each characteristic parameter and the predefined BP neural network model, and obtaining the prediction errors according to the prediction values and the values of expected output;
    an identification framework establishment module 400 for establishing the identification framework of the proof theory DS, wherein the identification framework comprises the multiple identification subframes in one-to-one correspondence with the various stages of rock mass instability;
    a confidence level calculator 500 for constructing the basic probability distribution functions of the characteristic parameters according to the prediction errors, and calculating the confidence level values, corresponding to the various stages of rock mass instability,
    characteristic parameters through the basic probability distribution functions of the characteristic parameters; a feature fusion module 600 to merge all feature parameters into the same identification subframe through DS proof theory according to the confidence level values, corresponding to the various stages of rock mass instability , characteristic parameters to obtain the merged characteristic parameters; and a second prediction module 700 for calculating the degree of confidence values, corresponding to the various stages of rock mass instability, of the merged feature parameters, and determining the stage of rock mass instability corresponding to the degree value of maximum confidence as the prediction result.
    Optionally, the parameter characteristics include one or more of overshoot count, number of acoustic emission events, energy count, duration, rise time, rms voltage value, and rms amplitude value of the acoustic emission signals.
    Time domain characteristics include one or more of peak value, mean value, kurtosis value, standard deviation, pulse factor, margin factor, and form factor. wave of acoustic emission signals.
    The frequency domain characteristics include one or more of the main frequency band position, the signal energy, and the degree of spectral dispersion and aggregation of the acoustic emission signals.
    Optionally, the predefined BP neural network model is obtained by training the BP neural network and the genetic algorithm through the characteristic parameters of the different stages of rock mass instability.
    Optionally, the identification sub-frames are the set of confidence values, corresponding to the stages of rock mass instability represented by the identification sub-frames, of all characteristic parameters.
    As shown in FIG. 5, the degree of confidence calculation module 500 further comprises: a first distance calculation submodule 510 for constructing the first distance function which takes the absolute values of the prediction errors as being the first distances between the parameters of characteristic and actual values; a correlation coefficient function construction sub-module 520 for constructing the correlation coefficient functions representing the degrees of correlation between the feature parameters and the identification subframes according to the first distance functions; a basic probability distribution function construction submodule 530 for constructing the basic probability distribution functions of the characteristic parameters according to the correlation coefficient functions and the mean squared errors between the prediction values and the values of expected output of the predefined BP neural network model, so that the greater the degrees of correlation, the greater the confidence degree values obtained by the basic probability distribution functions of the characteristic parameters.
    As shown in FIG. 6, the feature merging module 600 includes: a second distance calculating submodule 610 for calculating the mathematical distance between any two parameters of all feature parameters in the same identification subframe according to the degree values confidence of all characteristic parameters in the identification subframe; a support degree function construction submodule 620 for constructing the support degree function so that the smaller the mathematical distance between any two parameters of the feature parameters, the smaller the support degree value between the two parameters of characteristic is great; a feature parameter update submodule 630 to normalize the degree of support function, use the normalized degree of support function as the weight of the feature parameters, and perform a weighting average on each feature parameter in the same identification sub-frame to form the new characteristic parameters; a feature merge submodule 640 for merging all new feature parameters into the same identification subframe through D-S proof theory to obtain the merged feature parameters.
    The above modules are program modules and the division of modules is only a division of logic function. In a real implementation, other splitting processes can be done. For example, the multiple modules can be integrated into a single processing unit, or each unit can also physically exist separately, or the two or more modules can also be integrated into a single unit.
    In a third aspect of the present invention, the computer readable storage medium is provided, in which instructions are stored on the readable storage medium, and when the instructions execute on a computer, the instructions allow the computer to perform the above method of predicting the stages of rock mass instability based on a fusion of multiple features. In summary, in the above technical solutions of the present invention, the basic probability distribution functions corresponding to the various instability stages, of each characteristic parameter of the extracted rock mass instability acoustic emission signals are built on the basis of the prediction errors, obtained through the BP neural network model and each characteristic parameter; the degrees of confidence of each characteristic parameter corresponding to the various stages of instability are obtained through the basic probability distribution functions; on the basis of the confidence degrees, the distance function is introduced to obtain the distance between each two different proofs in the same identification subframe, and to obtain the measure of similarity between the proofs according to the distances between the proofs; according to the similarity measure, the degree of support function indicating the degree of support of other evidence in the same identification subframe with respect to a certain evidence is constructed, and the degrees of support are used as the weights for weight all the proofs, and a fusion of multiple features is performed through the DS proof theory; and by introducing the distance function, the degrees of support between any two proofs of all the proofs can be efficiently quantified. Therefore, an accurate prediction of the various stages of rock mass instability is made, and the problem of imprecise prediction through the single feature parameter is effectively solved; furthermore, the influences of the high conflict evidence on the prediction result are effectively eliminated on the basis of the evidence merging, which improves the accuracy of the prediction.
    The present application is described with reference to flowcharts and / or block diagrams of methods, devices and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in flowcharts and / or block diagrams, and the combination of processes and / or blocks in flowcharts and / or block diagrams can be implemented by instructions computer program. The computer program instructions may be provided for a processor of a universal computer, a specialized computer, an integrated processor, or other programmable data processing devices so as to produce a machine, so that the instructions executed by the processor computer or other programmable data processing devices generate a device to implement the functions specified in one or more processes in the flowcharts and / or one or more block (s) in the block diagrams.
    Computer program instructions may also be stored in computer readable memory capable of directing the computer or other programmable data processing devices to operate in a particular manner so that the instructions stored in computer readable memory produce an article of manufacture comprising an instructing device, in which the instructing device implements the functions specified in one or more processes in the flowcharts and / or one or more block (s) in the block diagrams.
    Computer program instructions can also be loaded onto the computer or other programmable data processing devices, so that a series of steps are performed on the computer or other programmable devices to produce processing carried out. by the computer. Thus, instructions executed on the computer or other programmable devices provide steps to implement the functions specified in one or more processes in the flowcharts and / or one or more block (s) in the block diagrams.
    The optional embodiments of the present invention are described in detail above with reference to the drawings, but the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and the simple modifications all fall within the scope of protection of the embodiments of the present invention. present invention.
    In addition, it should be noted that the specific technical features described in the specific embodiments above can be combined in any suitable manner without conflict. In order to avoid unnecessary repetitions, the various possible combinations are not further explained by the embodiments of the present invention.
    Those skilled in the art can understand that all or some of the steps of the methods in the above embodiments can be completed by instructing associated hardware through a program. The program is stored in a storage medium and includes several instructions for directing a single chip microcomputer, a chip or a processor to perform all or part of the steps of the methods in the various embodiments of the present invention. . The aforementioned storage medium has various types of media which can store program code, such as USB flash disk, mobile hard disk, ROM (Read-Only Memory), RAM (RAM), Magnetic disk or optical disc.
    权利要求:
    Claims (13)
    [1]
    A method of predicting the instability stages of a rock mass based on a fusion of multiple characteristics, comprising: acquiring acoustic emission signals from a rock mass; perform hybrid domain characteristic extraction on the acquired acoustic emission signals, and extract multiple characteristic parameters with significant category difference information from parameter characteristics, time domain characteristics and / or characteristics of frequency domain respectively; obtain prediction values representing the instability stages of the rock mass according to each feature parameter and a predefined BP neural network model, and obtain prediction errors according to the prediction values and expected output values; establish an identification framework for D-S proof theory, in which the identification framework includes multiple identification sub-frames in one-to-one correspondence with various stages of rock mass instability; construct basic probability distribution functions of feature parameters according to prediction errors, and calculate confidence values, corresponding to various stages of rock mass instability, of feature parameters through distribution functions basic probability of the characteristic parameters; merge all feature parameters into the same identification subframe through DS proof theory according to the degree of confidence values, corresponding to the various stages of rock mass instability, of the feature parameters to obtain merged characteristic parameters; and calculating degree of confidence values, corresponding to the various stages of rock mass instability, of the merged feature parameters, and determining a stage of rock mass instability corresponding to the maximum value among the degree of confidence values as that prediction result.
    [2]
    The method of predicting instability stages of a rock mass based on a fusion of multiple features according to claim 1, wherein the parameter features include one or more of a count of overshoots, an overshoot count. number of acoustic emission events, an energy count, a duration, a rise time, an RMS voltage value and an RMS amplitude value of the acoustic emission signals; the time domain characteristics include one or more of a peak value, an average value, a kurtosis value, a standard deviation, a pulse factor, a margin factor and a waveform factor acoustic emission signals; and the frequency domain characteristics include one or more of a main frequency band position, signal energy, and a degree of spectral aggregation and dispersion of the acoustic emission signals.
    [3]
    A method of predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics according to claim 1, wherein the predefined BP neural network model is obtained by training a BP neural network and of a genetic algorithm through characteristic parameters in different stages of rock mass instability.
    [4]
    The method of predicting the stage of instability of a rock mass based on a fusion of multiple features according to claim 1, wherein the identification subframes are a set of corresponding confidence level values. at the stages of rock mass instability represented by the identification sub-frames, of all characteristic parameters.
    [5]
    The method of predicting the stage of instability of a rock mass on the basis of a fusion of multiple characteristics according to claim 1, wherein the construction of the basic probability distribution functions of the characteristic parameters comprises: the construction a first distance function which takes absolute values of the prediction errors as the first distances between the characteristic parameters and actual values; constructing correlation coefficient functions representing degrees of correlation between the feature parameters and the identification subframes, according to the first distance functions; and constructing the basic probability distribution functions of the characteristic parameters according to the correlation coefficient functions and the mean squared errors between the prediction values of the predefined BP neural network model and the expected output values, so that more the greater the degrees of correlation, the greater the confidence values obtained by the basic probability distribution functions of the characteristic parameters.
    [6]
    The method of predicting instability stages of a rock mass based on a fusion of multiple features according to claim 1, wherein the fusion of all the feature parameters in the same identification sub-frame by. The DS proof theory bias includes: calculating a mathematical distance between any two parameters of all feature parameters in the same identification subframe according to the confidence values of all feature parameters in the identification subframe; constructing a degree of support function, so that the smaller the mathematical distance between any two parameters of the characteristic parameters, the larger the value of the degree of support between the two characteristic parameters; normalizing the degree of support function, using the normalized degree of support function as the weight of the characteristic parameters, and performing a weighting average on each characteristic parameter in the same sub-frame identification to form new feature parameters; and merging all new feature parameters into the same D-S proof theory identification subframe to obtain merged feature parameters.
    [7]
    7. Device for predicting the instability stages of a rock mass based on a fusion of multiple characteristics, comprising: a data acquisition module for acquiring the acoustic emission signals of the rock mass; a feature extraction module for performing hybrid domain feature extraction on the acquired acoustic emission signals, and extracting multiple feature parameters with significant category difference information from parameter features, characteristics of time domain and / or frequency domain characteristics respectively; a first prediction module for obtaining prediction values representing the stages of rock mass instability according to each feature parameter and a predefined BP neural network model, and obtaining prediction errors according to the prediction values and output values expected; an identification framework establishment module for establishing an identification framework of the DS proof theory, in which the identification framework comprises multiple identification subframes in one-to-one correspondence with the various stages of instability of rock mass; a confidence level calculator for constructing basic probability distribution functions of the characteristic parameters according to the prediction errors, and calculating confidence level values, corresponding to the various stages of rock mass instability, characteristic parameters through the basic probability distribution functions of the characteristic parameters;
    a feature fusion module to merge all feature parameters into the same identification subframe through DS proof theory according to the degree of confidence values, corresponding to the various stages of rock mass instability, characteristic parameters to obtain merged characteristic parameters; and a second prediction module for calculating degree of confidence values, corresponding to the various stages of rock mass instability, of the merged characteristic parameters, and determining a stage of rock mass instability corresponding to the maximum value among the values. of degree of confidence as being the result of prediction.
    [8]
    A device for predicting the stage of instability of a rock mass based on a fusion of multiple features according to claim 7, wherein the parameter features include one or more of a count of overshoots, an overshoot count. number of acoustic emission events, an energy count, a duration, a rise time, an RMS voltage value and an RMS amplitude value of the acoustic emission signals; the time domain characteristics include one or more of a peak value, an average value, a kurtosis value, a standard deviation, a pulse factor, a margin factor and a waveform factor acoustic emission signals; and the frequency domain characteristics include one or more of a main frequency band position, signal energy, and a degree of spectral dispersion and aggregation of the acoustic emission signals.
    [9]
    The device for predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics according to claim 7, wherein the predefined BP neural network model is obtained by training the BP neural network and a genetic algorithm through characteristic parameters in different stages of rock mass instability.
    [10]
    A device for predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics according to claim 7, wherein the identification subframes are a set of corresponding confidence level values. at the stages of rock mass instability represented by the identification sub-frames, of all characteristic parameters.
    [11]
    The device for predicting the instability stages of a rock mass on the basis of a fusion of multiple characteristics according to claim 7, wherein the degree of confidence calculation module further comprises: a first sub-module of calculating a distance to construct the first distance function which takes absolute values of the prediction errors as the first distances between the characteristic parameters and the actual values; a correlation coefficient function construction sub-module for constructing correlation coefficient functions representing degrees of correlation between the feature parameters and the identification subframes according to the first distance functions; and a basic probability distribution function construction submodule for constructing the basic probability distribution functions of the characteristic parameters according to the correlation coefficient functions and the mean squared errors between the prediction values of the network model predefined BP neural and expected output values, so that the greater the degrees of correlation, the greater the confidence degree values obtained by the basic probability distribution functions of the characteristic parameters.
    [12]
    The device for predicting the instability stages of a rock mass on the basis of a fusion of multiple features according to claim 7, wherein the feature fusion modulus comprises:
    a second distance calculating sub-module for calculating a mathematical distance between any two parameters of all the characteristic parameters in the same identification sub-frame according to the confidence values of all the characteristic parameters in the sub- identification framework; a support degree function construction submodule for constructing a support degree function, so that the smaller the mathematical distance between any two parameters of the feature parameters, the smaller the support degree value between the two parameters of characteristic is great; a feature parameter update submodule to normalize the degree of support function, use the normalized degree of support function as the weight of the feature parameters, and perform a weighting average on each feature parameter in the same identification sub-frame to form new characteristic parameters; and a feature merge submodule for merging all new feature parameters into the same identification subframe through D-S proof theory to obtain merged feature parameters.
    [13]
    13. A computer readable storage medium, in which instructions are stored on the readable storage medium, and when the instructions are executed on a computer, the instructions allow the computer to perform the process of predicting stages of development. An instability of a rock mass on the basis of a fusion of multiple characteristics according to any one of claims 1 to 6.
    类似技术:
    公开号 | 公开日 | 专利标题
    BE1027019B1|2021-06-25|METHOD AND DEVICE FOR PREDICTING THE STAGES OF INSTABILITY OF A ROCKY MASS BASED ON A FUSION OF MULTIPLE CHARACTERISTICS
    CN106156809A|2016-11-23|For updating the method and device of disaggregated model
    CN106203679A|2016-12-07|A kind of customer loss Forecasting Methodology and system
    CN108475393A|2018-08-31|The system and method that decision tree is predicted are promoted by composite character and gradient
    US10083415B2|2018-09-25|Multi-variable assessment systems and methods that evaluate and predict entrepreneurial behavior
    WO2018214895A1|2018-11-29|Data processing method, data processing apparatus, storage device and network device
    CN109964265A|2019-07-02|Language learning is measured using standardized fraction scale and self-adaptive estimation engine
    US10318671B2|2019-06-11|Predicting social, economic and learning outcomes
    AU2016247051A1|2018-01-04|Resource evaluation for complex task execution
    WO2019037202A1|2019-02-28|Method and apparatus for recognising target customer, electronic device and medium
    CN101447995B|2012-05-23|Method for identifying P2P data stream, device and system thereof
    Nayebi et al.2017|Which version should be released to app store?
    Sommestad et al.2012|Effort estimates for vulnerability discovery projects
    CN109934433A|2019-06-25|A kind of personnel ability&#39;s appraisal procedure, device and cloud service platform
    Mahmud et al.2013|When will you answer this? estimating response time in twitter
    CN106651574A|2017-05-10|Personal credit assessment method and apparatus
    Gupta et al.2016|Will I get in? modeling the graduate admission process for American universities
    Das et al.2011|Evaluation of liquefaction potential of soil using genetic programming
    WO2021129368A1|2021-07-01|Method and apparatus for determining client type
    Naseem et al.2019|Using ensemble decision tree model to predict student dropout in computing science
    Khezrimotlagh2015|How to detect outliers in data envelopment analysis by Kourosh and Arash method
    Kitsios et al.2015|New service development: Strategy and innovation process in cultural and creative industries using MCDA
    US20180068323A1|2018-03-08|Automated method for learning the responsiveness of potential consumers to different stimuli in a marketplace
    Bru et al.2018|The perceived quality of process discovery tools
    TWI634508B|2018-09-01|A system and method using multi-dimensional rating to determine an entity&#39;s future commercial viability
    同族专利:
    公开号 | 公开日
    CN110457757A|2019-11-15|
    BE1027019A1|2020-08-26|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    CN104062677A|2014-07-03|2014-09-24|中国科学院武汉岩土力学研究所|Multifunctional comprehensive integrated high-precision intelligent micro-seismic monitoring system|
    CN113063857B|2020-06-15|2022-02-18|中国科学院武汉岩土力学研究所|Acoustic emission identification method for rock structural surface tension-shear failure in direct shear test|
    CN113920172B|2021-12-14|2022-03-01|成都睿沿芯创科技有限公司|Target tracking method, device, equipment and storage medium|
    法律状态:
    2021-07-19| FG| Patent granted|Effective date: 20210625 |
    优先权:
    申请号 | 申请日 | 专利标题
    CN201910638983.4A|CN110457757A|2019-07-16|2019-07-16|Instability of Rock Body stage forecast method and device based on multi-feature fusion|
    [返回顶部]