• 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:
    In order to determine a model for an output quantity (y) of a technical system which is non-linearly dependent on a number of input variables in the form of an input variable vector (u), a target output variable range (COR) is defined and a model-based experimental design is determined with which the model in the target output variable range (COR) is parameterized by the selection of associated input variable vectors (Ucand.coR). For selecting the input variable vectors (Ucand.coR), a distance-based selection criterion is used.
    公开号:AT512977A2
    申请号:T50347/2013
    申请日:2013-05-22
    公开日:2013-12-15
    发明作者:
    申请人:Avl List Gmbh；
    IPC主号:
    专利说明:

    AV-3539 AT
    Method for determining a model of an output of a technical system
    The present invention relates to a method for determining a model for an output of a technical system which is non-linearly dependent on a number of input variables in the form of an input vector, wherein a model structure with model parameters is given as a model and the model parameters are determined in an iterative method be optimized by measured output values from the bench tests on the technical system and the test bench experiments follow a test plan, which is created for each iteration step from estimated outputs of the model.
    The output variables of many technical systems are multi-dimensionally non-linear and dependent on a large number of parameters. A typical example is an internal combustion engine, in which many parameters (outputs), e.g. NO * emissions, soot in the exhaust gas, consumption, etc., from a variety of input variables (or operating points), such as. Speed, torque, temperature, position of the swirl flap, fuel pressure, etc., are dependent. In general, a technical system is understood to be a system in which a number of specific input variables, such as e.g. Actuator settings, a number of particular outputs, e.g. Characteristics or measurable with sensors sizes cause so that there is an arbitrary functional relationship between input and output variables. In the development of such technical systems, it is very costly to carry out the development exclusively on a test bench, since the many nonlinear output variables which depend on one another and on different input variables can hardly be manually optimized, which is a possible development goal. Therefore, it has already been attempted to model such technical systems or specific output variables as a function of input variables and system parameters by means of mathematical models, by means of which one can simulate and study the influence of input variable variations or model parameter variations and make optimizations, which are then performed e.g. be verified on the test bench. For this purpose, so-called model-based design of experiments have become known, e.g. from WO 2012/163 972 A1, which uses a mathematical method and measured values to identify a model of a non-linear technical system. A model structure is chosen, e.g. a known Local Model Network (LMN) or a known Multilayer Perceptron (MLP), and model parameters are estimated from data obtained from bench tests (online). In doing so, the model iteratively generates output variables which are used in a test plan in order to generate new input variables for the next iteration step, which are traversed on the test bench for the technical system. The output quantities of inted: 23 ^ 05-2013
    AV-3539 AT
    Bench tests are then used, taking into account certain boundary conditions, to improve the model parameters. This is repeated until a predetermined termination criterion is met, that is, until the model parameters have been estimated with sufficient accuracy. It is important for the estimation of the model parameters that the test plan covers the entire dynamics and the entire output range of the technical system in order to achieve a sufficient model quality throughout the output range.
    However, many parameters of a technical system are interesting only in certain areas, be it because there are legal regulations that make specifications for parameters in certain parameter ranges, or because a parameter is meaningful only in certain 10 parameter areas. A typical example of this is the NO * or soot emissions of an internal combustion engine, which must comply with statutory limits in certain operating ranges of the internal combustion engine. A model for such a parameter would therefore only have to be accurate within this target output range, but not for the entire possible output range. However, the method for model-based model identification according to WO 2012/163 972 A1 creates a model that is valid for the entire output range, which on the one hand involves a high outlay, in particular for the necessary test bench tests. On the other hand, it is possible that the model does not deliver sufficiently accurate results in the target output range, because the model identification used too few training data 20 in the target output range. Thus, such a model would be only partially usable for this target output range.
    From Picheny, V., et al., "Adaptive Designs of Experiments for Accurate Approximation of a Target Region," Journal of Mechanical Design, Vol. 132, July 2010, 071008-1 to 071008-9, a method is disclosed in which An experimental design is created so that a meta-model in the vicinity of a narrow target range estimates output quantities as accurately as possible. The metodell is a Kriging model and the target area is defined as a contour line. The model should therefore estimate very accurately around a contour line, but not within a larger target exit range, for which this method is not suitable. In this case, a sequential design of experiments method is used in which a new input variable 30 is selected according to an optimization criterion as a function of the already known input variables and their associated calculated output variables. This method therefore does not require any bench tests. The aim of this method is therefore not the optimization based on concrete measured values from test bench tests on the real system, but rather the adaptation of a function to a given contour line. -2- fif HPüoür
    AV-3539 AT
    It is an object of the present invention to provide a method for determining a model for an output of a technical system, are used for the optimization test bench tests, the method with the least possible effort for test bench experiments and the estimated model in a large target 5 output range should have a high accuracy.
    This object is achieved by determining a set of output variables from a set of input variable vectors using the model valid for the current iteration step, from which set of output variables those target input variables vectors are determined that are to an output variable within a predetermined target value. 10 output range, from the particular target input vector vectors for the next iteration step a new input vector is selected to supplement the set of input vector vectors, the new input vector being selected based on a given distance based selection criterion, using such extended set of selected input vector vectors as a design plan, 15 to generate test data of the output variable by means of test bench tests, with which the model is optimized and the above method steps are repeated iteratively until a predefined even abort criterion is met. Through this method, the model is trained specifically in a defined target output range, for which fewer test bench tests are required and still a very high accuracy in the Zielausgangsgrößenbe-rich can be achieved.
    A distance-based selection criterion is a criterion in the input variable range, in the output variable range or in the input / output variable range.
    The use of a local model network as a model structure has proven to be particularly favorable since it can be used with low computational complexity and requires existing knowledge of the technical system and leads to model parameters which can be used in a robust manner for the experimental design.
    The subject invention will be explained in more detail below with reference to Figures 1 to 10, which show by way of example, schematically and not by way of limitation advantageous embodiments of the invention. FIG. 1 schematically shows the sequence of the method according to the invention,
    2 to 4 different iteration stages of the method according to the invention for a distance-based selection criterion in the input / output variable range,
    FIGS. 5 to 7 show a comparison of the various selection criteria with a conventional model identification method.
    P2013 / 50347
    AV-3539 AT
    8 shows a comparison of the mean square errors using the various selection criteria and a conventional method for model identification, and FIGS. 9 and 10 show the application of the method according to the invention to the determination of a model for the ΝΟχ emissions of an internal combustion engine. For the method according to the invention, a model with a model structure is first to be selected with which the behavior of the technical system for a parameter or output variables is to be estimated as a function of specific input variables. The following description of the method will be described using a so-called known local model network (LMN). It should be noted, however, that the method can be used in the same way also with other known model structures, such as e.g. Multilayer perceptrons, support vector machines or Gauss processes can be implemented. However, a local model network is characterized by low computing overhead, the ability to use existing knowledge of the technical system, and model parameters that can be used robustly in experimental design, and is therefore particularly suited to the method of the invention, therefore The invention will be described below with reference to a local model network LMN as a model structure.
    A local model network LMN consists of I local models y} ("; S;), which are valid only locally, ie for certain input variables, u designates the n-dimensional input variable vector of the technical system, which contains all input variables, and <9, the
    Model parameter of the jth local model. The local models ^ (";, £,) are here quadratic multiple regression models with interaction terms, although it should be noted that other model structures can also be used for the local models, e.g. linear or cubic multiple regression models. The output of the local Mo network network y (u) is the weighted sum of the outputs of the local models (»; £,) and is an estimate of the system output. For this purpose, a validity function Φ7 (χ) is defined, which defines the validity range of the local models yj (u; 3}), where x denotes selected input variables of the input quantity vector u. The output of the local 1
    Model network y (u) is then known to be y (u) =. The local model network LMN needs training data to estimate the relationship between input quantities u and output y of the technical system. By training in a training session 2 you can not only use the ModelIparametem but also the -4-
    AV-3539 AT
    Number I of the local models yj (u; Sj) and their validity functions Φ; (χ). This is well known in the art, e.g. from Hametner, C., et al., "Local Model Net-work Identification for Online Engine Modeling", 2013, Information Sciences 220 (0), p.210-225, so will not be discussed further here. The training data comes from a design plan, which is created as described below with reference to FIG.
    The starting point is that the model does not have to estimate accurately in the entire possible output variable range, but only in a defined target output variable range. For example, internal combustion engines are to be calibrated in such a way that certain output quantities, e.g.
    NO * or soot emissions, certain specifications for certain input variables. Thus, the target output range is not known but the associated input variables of the model. Therefore, an experimental design is needed which identifies these input variables and parameterizes the model parameters 9j of the model, and possibly also the number I of local models and their validity functions, for the target output variable range.
    The test plan is to be understood as a sequence, possibly also a time sequence, of input variables which are set on the test stand of the technical system and in the process the desired output variable of the technical system is observed or measured. On the test bench, e.g. an engine test bench, the technical system, e.g. an internal combustion engine arranged, optionally with a loading machine, e.g. an electric dynamometer, connected and controlled the test stand or the technical system or the loading machine according to the design.
    The experimental design is based on an iterative selection of input quantity vectors Uoand from an input quantity vector set Ucand. As input quantity vector set Ucand it is possible to use any desired set of input variable vectors U ^ which covers the entire input size range as uniformly as possible. The input quantity vector quantity Ucand is determined in advance and is assumed to be known for the method.
    An input vector Ucand can only be used once for the design. At the beginning of the method, a set U ^ n of input quantity vectors ustart is given. Thus, in each iteration step, a new input variable vector is unew least selected from the input vector vectors Ustart and then from the input vector vectors Ucand and the reaction of the technical system to this input vector Unew is measured in the form of an output variable ynew by a bench test on a test bench 1 for the technical system. The amount Uapp of already known-5-
    AV-3539 AT th or selected input variable vectors UapP (including the new input variable vector ιι ^) and the amount Yapp the associated measured output quantities yapp form the training data for the training of the local model network LMN in the training unit 2 (eg a suitable computer with software and implemented Algorithms for the execution of the training). In each iteration step k, therefore, the local model network LMN is trained with this training data. However, the training can only in some iteration steps a change of the local model network LMN, e.g. the network dimension I or the model parameter. In modifying the model, the model parameters, and possibly also the number I of local models 10, and their validity functions Φν (ϊ), are updated according to the rules of the local model network LMN, e.g. in Hametner, C., et al., "Local Model Network Identification for Online Engine Modeling", 2013, Information Sciences 220 (0), p.210-225.
    As a result, the amount Uapp of the already selected input variable vectors uapp and 15 the amount Yapp of the associated measured output quantities yapp constantly increases until a sufficiently accurate model LMNfin for the target output range has been reached. This can be determined by a suitable termination criterion, e.g. a maximum number of training data or a threshold below which the mean square model error must be in the destination output area. The new input variable vector unew 20 is selected according to the following procedure.
    First, the output variables y {ucand) for the input quantity vector quantity Ucand estimated by the current model are determined on the basis of the current local model network LMN in the respective iteration step k. Thereafter, all the input quantity vectors Ucand, coR are determined therefrom, their estimated outputs m are the target output ranges (COR) defined by its limits ymjn and ymax (COR = and these become in the target input quantity vector set
    Ucand, cor summarized.
    Ucand, COR = {Ucand € U!: An /} | ^ jK ^ cand) -> W}
    From this target input quantity vector set Ucand, cor, the new input variable vector iw is now selected in each iteration step k 30. The aim is to achieve as uniform a distribution as possible in the input variable space, in the output variable space or in the input output variable space. The background to this is that the local model network LMN, or more generally the model, is more accurate in the vicinity of the training data than further away.
    AV-3539 AT of the training data. Consequently, if the training data is distributed as evenly as possible, the model uncertainty is reduced. For this reason, a distance-based criterion (also known as S-optimal design or maximin distance design) is used to select the new input vector. The aim of the ab-5 state-based selection criterion is to maximize the minimum distance between design points (here input size and / or output size). For this purpose, three distance-based selection criteria can now be defined:
    Distance-based selection criterion in the input size range:
    It is selected from the target input quantity vector set Ucand.coR that input variable vector to Ucand.coR as a new input variable vector Unew, which has the smallest possible Euclidean distance to the already known input variable vectors uapp in the set Uapp, for which applies: u
    This criterion leads to a uniform distribution of the selected input quantity vectors in the input variable range.
    Distance-based Selection Criterion in the Ausaanasarross Range:
    First, the output quantities y (ucandC0R) are calculated using the currently valid model for the input quantity vector set Ucand, coR. The new input variable vector unew has the
    Output variable ynew = argmax min yj (yapp ~ yca) ldicoRj that is, the output quantity which has the largest possible smallest Euclidean distance to the already known output variables, and the associated new input variable vector u "ew then results
    This criterion leads to a uniform distribution of the output variables in the target output range. 25 Distance-based selection criterion in the single-pineapple / Ausaanasgrößenbereich:
    Here, the set of already selected input variable vectors Uapp is supplemented by the associated output variables Yapp, U * app = läpp 7 ^], and the target input vector quantity Ucand.coR is added by the associated estimated output variables
    AV-3539 AT) and COR = jcandiCOR y (Uca / ld, coR)] - The new input variable vector unew then results from the criterion
    argmax
    Here, therefore, the input variable with the smallest possible Euclidean distance from the already known input variables and the smallest possible Euclidean distance of the associated output variable to the already known output variables is searched. This criterion leads to a uniform distribution of the input variables in the input variable range and the output variables in the target output variable range.
    The local model network LMNfin trained in this way is then able to estimate the output y of the technical system within a target output range with the desired accuracy (predetermined by the selected termination criterion). The trained model network LMN may then be e.g. used for the development of the technical system, e.g. to calibrate control units of an internal combustion engine so that predetermined limit values for the output variable are met. The method according to the invention will be demonstrated on the basis of FIGS. 2 to 4 on the basis of a bathtub-like function f (y = f (u)), which is dependent on a one-dimensional input variable u, and on the basis of a distance-based selection criterion in the input / output variable range , This function f shows a strongly nonlinear behavior in the input variable ranges u = [0; 0.2] and u = [0.8; 1], and is almost linear in the input variable 20 range u = [0.2; 0.8]. The target output range, in which the model (a local model network LMN) is to be parametrized with sufficient accuracy, should be defined for the example by ymin = 0.15 and ymBX = 1. With the curve 3 in Figure 2, the initial model with three selected initial input variables ustarl in the set Uapp dargestelit. The two dashed lines mark the target output range. For each axis, the frequency of the points in the input and output area is shown in the form of bars. The new input quantity u 1 determined by means of the method is also shown in FIG. FIG. 3 shows the model after the selection of seven input variables in the set Uapp and the reaction of the current model in the form of the output variables yapp. 4 shows the model after the selection of eleven input quantities in the quantity Uapp, with which the procedure is stopped (the new input variable unew is no longer taken into account). It can be seen on the basis of FIGS. 2 to 4 that the method systematically parameterizes the model in the defined target output variable range, whereby the model therein is very accurate, but inaccurate outside this range. The frequency distributions also show a very even distribution of the points in the entrance and exit areas to interest.
    AV-3539 AT destination output range (according to the selected selection criterion). In addition, it can be seen that the method is able to cover both branches of the output function equally.
    FIGS. 5 to 7 compare the different distance-based selection criteria according to FIG. 5, each with eleven selected input variables. FIG. 5 shows the result with a method according to the prior art without target output variable range. It can be seen that, although the input quantities are distributed very uniformly, but no input variable with output variable has been parameterized in the target output range, the model is unusable in the target output range. FIG. 6 uses a distance-based selection criterion described above in the input variable range, which leads to the selection of four input quantities with output variable in the target output variable range. FIG. 7 uses a distance-based selection criterion described above in the output variable range, which allows for selection of six input variables with output variable in the target output variable range leads. Fig. 8 also shows the mean square error (MSE) of the different methods (on a logarithmic scale). Curve 4 shows the MSE for a method without target output range, curve 5 shows the method with distance-based selection criterion in the input variable range, curve 6 shows the method with distance-based selection criterion in the output variable range and curve 7 shows the method distance-based selection criterion in the input / output variable range. It can be seen that a distance-based selection criterion, while taking into account a target output size range, improves the model quality compared to a method without a target output range by significantly smaller model errors. Qualitatively, a distance-based selection criterion in the input / output variable range seems to be better than the other distance-based selection criteria.
    A model-based design with a target output range is therefore an efficient method that parametrizes the model of the engineering process in the target output range to reduce the input size range, resulting in improved model quality in the target output range. At the same time, the necessary number of required bench tests is reduced in order to achieve the same model quality.
    Below is the method of the invention is still a practical example in the form of NO * emissions of an internal combustion engine (output y) as a function of the input variables position of the swirl flap (S), fuel pressure (p) and exhaust gas recirculation 35 (EGR) (input variable vector u) for a certain speed and a certain load
    η.
    AV-3539 AT be cleared. As the target output range, the NCVEmission is defined in the range of ymin = 0.1x10'5Kg / s and ymax = 0.4x10 '£ kg / s. The underlying model, hereby. in the form of a local model network LMN, is systematically parameterized for the target output range using the method according to the invention. As can be seen in FIG. 9, the output variables are estimated in the target output range by the method according to the invention by the model. Curve 8 shows the result for five selected input variables for a method without target output range, curve 9 the result with distance-based selection criterion in the input variable range, curve 10 the result with distance-based selection criterion in the output variable range and Kur-10 ve 11 the result of distance-based selection criterion in input / output size area. Finally, FIG. 10 shows the NCVEmeasurement estimated with the determined model for a constant position of the swirl flap S = 0.1 and for a specific speed and a specific load
    -10-
    权利要求:
    Claims (5)
    [1]


    [Ϊ02Ο13 / $ Ο347 AV-3539 ΑΤ Claims 1. A method for determining a model for an output (y) of a technical system which is nonlinearly dependent on a number of inputs in the form of a single-input vector (u), a model structure with model parameter is given as model and the model parameters are optimized in an iterative method based on measured output values from bench tests on the technical system and the bench tests follow a test plan which for each iteration step (k) consists of estimated output quantities (j>) of the Model is generated, characterized in that with the model valid for the current iteration step (k), a quantity (y (Ucand)) of estimated output quantities (Kuc &lt;; md)) is determined that out of this quantity (y (Ucanel)) of output variables those target input quantity vectors n (Uc ^ .cor) of the target input vector quantity (Ucand, coR) resulting in an estimated output {y (ueond)) in -15 within a predetermined target output range (COR), from the target input vector quantity (Ucand, coR) a new input variable vector (unew) is selected for the next iteration step (k + 1) to supplement the set (Uapp) of the already selected input variable vectors {uapp), wherein the new input variable vector (unew) is selected on the basis of a predetermined distance-based selection criterion will use the so-called Uapp of selected input vector vectors (uapp) as an experimental design to generate benchmark data of the output (yapp) using bench tests to optimize the model and to iteratively repeat the above steps; until a predetermined termination criterion is met.
    [2]
    2. Method according to claim 1, characterized in that selected from the target input quantity vector set (Ucand.coiO that input variable vector (Ucand.coR) as a new input variable vector (Unew), which has the smallest possible Euclidean distance to the already known input variable vectors (uapp ) in the set (Uapp) of the selected input variable vectors.
    [3]
    3. Method according to claim 1, characterized in that with the currently valid model for the target input vector quantity (Ucend.coR) the output quantities (yiKtmdcoü)) are calculated and the new input variable vector (Unew) is selected as input vector (Ucend.coR), its estimated output size

    IIIMIif AV-3539 AT (y {uc. <M <i, cw)) has the smallest possible Euclidean distance to the already known output values (yapp),
    [4]
    4. Method according to claim 1, characterized in that the set of already selected input variable vectors (Uapp) is supplemented by the associated output quantities (Yapp) and 5 the target input variable vector set (Ucand.coiO by the associated estimated output variables c <mdjcoR)) and as new input variable vector (unew) that input variable vector (Ucand.coiO is selected, the smallest possible Euclidean distance to the already known input variable vectors (uapp) and its associated output (y (uCündC0R)) the largest possible Euclidean distance to the 10 already known output variables (yapp).
    [5]
    5. Method according to claim 1, characterized in that the model is a local model network (LMN) with a number (I) of local models (yfaSj)) with model parameter (3}) and associated validity functions (0, (5 ) ) and, in an iteration step (k), the number (I) of the local models (j> y (»; $, -)) from the selected input values (uapp) and the measured output variables (y ^ p), the model parameters (£;) and / or its associated validation functions (Φ, (*)). -12-
    类似技术:
    公开号 | 公开日 | 专利标题
    AT512977B1|2014-12-15|Method for determining a model of an output of a technical system
    EP1626317B1|2009-06-24|Method for developinga nonlinear, stationary or dynamical model of a control variable of aconstruction machine
    EP2519861B1|2015-05-20|Method for the computer-aided control of a technical system
    EP3132317B1|2021-03-31|Method for computer-aided plant control optimisation using a simulation module
    WO2014131661A1|2014-09-04|Method for designing a nonlinear controller for nonlinear processes
    EP3268713A1|2018-01-17|Method for generating a model ensemble for calibrating a control device
    DE102017110795A1|2017-11-30|Method for improved calibration of the control of an internal combustion engine
    DE102009021781A1|2010-11-25|Engine-operating method for calculating an engine-operating map for a vehicle&#39;s control device creates a map with a specified number of nodes while measuring data points to calculate a map value
    EP2943841B1|2019-08-28|Method for the computerized control and/or regulation of a technical system
    EP3022611A1|2016-05-25|Method for the computer-aided control and/or regulation of a technical system
    DE102013212889A1|2015-01-08|Method and device for creating a control for a physical unit
    WO2020187591A1|2020-09-24|Method and device for controlling a robot
    EP3617912A1|2020-03-04|Method and device for the computer-assisted generation of a component of a technical system
    WO2020207789A1|2020-10-15|Method and device for controlling a technical apparatus
    DE102017218851A1|2019-04-25|Method, device and computer program for creating a deep neural network
    DE10222699A1|2003-12-11|Computer-aided motor vehicle system optimization method for optimizing a system&#39;s parameters uses approximation models for forecasting
    DE102017213510A1|2019-02-07|Method and apparatus for generating a machine learning system, and virtual sensor device
    DE102017110065A1|2017-07-06|Method and computer program product for modeling a NOx storage behavior of a NOx storage catalyst
    WO2017178015A1|2017-10-19|Method and device for determining a technical system model
    AT522958B1|2021-04-15|Method and system for calibrating a control of a machine
    WO2016198588A1|2016-12-15|Method for generating characteristic maps of a physical technical system or process
    DE102019134053A1|2021-06-17|Process for the continuous validation of automated driving functions applied in driving tests
    DE102019128655B4|2021-11-25|Method for providing a computer-aided control for a technical system
    DE102019218841A1|2021-06-10|Method of making a device with an optimized design feature vector
    DE102019208121A1|2020-12-10|Active data generation taking into account uncertainties
    同族专利:
    公开号 | 公开日
    JP2016530585A|2016-09-29|
    US10331810B2|2019-06-25|
    US20160063151A1|2016-03-03|
    EP2999998B1|2018-08-08|
    KR102206324B1|2021-01-22|
    EP2999998A1|2016-03-30|
    WO2014187828A1|2014-11-27|
    AT512977A3|2014-11-15|
    CN105209984B|2018-01-19|
    AT512977B1|2014-12-15|
    JP6404909B2|2018-10-17|
    CN105209984A|2015-12-30|
    KR20160013012A|2016-02-03|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    JPH056202A|1991-06-27|1993-01-14|Matsushita Electric Ind Co Ltd|Learning controller|
    JPH0756608A|1993-08-17|1995-03-03|Kawasaki Steel Corp|Signal processing method|
    JP2003044827A|2001-07-30|2003-02-14|Haruo Ishikawa|Method for estimating characteristic|
    JP4274046B2|2004-05-25|2009-06-03|オムロン株式会社|Control device, control parameter adjustment device, control parameter adjustment method, program, and recording medium|
    JP5279299B2|2008-02-29|2013-09-04|キヤノン株式会社|POSITION CONTROL APPARATUS, EXPOSURE APPARATUS, AND DEVICE MANUFACTURING METHOD EQUIPPED WITH AN ITS|
    US20090319453A1|2008-06-24|2009-12-24|Livermore Software Technology Corporation|Sampling Strategy Using Genetic Algorithms in Engineering Design Optimization|
    AT10301U3|2008-09-01|2009-09-15|Avl List Gmbh|METHOD AND REGULATION FOR REGULATING A REGULAR TRACK WITH A RECYCLING WORKING CYCLE|
    WO2011029015A2|2009-09-03|2011-03-10|Larimore Wallace E|Method and system for empirical modeling of time-varying, parameter-varying, and nonlinear systems via iterative linear subspace computation|
    CN102122132A|2010-01-11|2011-07-13|北京航空航天大学|Intelligent control system for environmental simulation system based on a fuzzy neural network|
    CN101782743A|2010-02-11|2010-07-21|浙江大学|Neural network modeling method and system|
    CN101846979A|2010-06-29|2010-09-29|北京航空航天大学|Advanced iterative learning control method for accurate target tracking|
    US8548621B2|2011-01-31|2013-10-01|Xerox Corporation|Production system control model updating using closed loop design of experiments|
    JP5707230B2|2011-05-25|2015-04-22|新日鉄住金エンジニアリング株式会社|Process state prediction method|
    US8898096B2|2011-05-31|2014-11-25|Oracle International Corporation|Application configuration generation|
    AT511577B1|2011-05-31|2015-05-15|Avl List Gmbh|MACHINE IMPLEMENTED METHOD FOR OBTAINING DATA FROM A NON-LINEAR DYNAMIC ESTATE SYSTEM DURING A TEST RUN|
    CN102629104B|2011-12-01|2013-10-02|燕山大学|Calcination predictive control system and method for rotary cement kiln|
    AT512977B1|2013-05-22|2014-12-15|Avl List Gmbh|Method for determining a model of an output of a technical system|AT512977B1|2013-05-22|2014-12-15|Avl List Gmbh|Method for determining a model of an output of a technical system|
    DE102016104236A1|2016-03-09|2017-09-14|Iav Gmbh Ingenieurgesellschaft Auto Und Verkehr|Method of creating a model of a technical system|
    AT518676B1|2016-05-17|2018-02-15|Avl List Gmbh|Method for calibrating a technical system|
    CN106836389A|2017-01-15|2017-06-13|苏州大学|A kind of Intelligent water-saving type bathtub and its method for designing based on CFD|
    JP2018120327A|2017-01-24|2018-08-02|オムロン株式会社|Control apparatus, control program and control system|
    AT520827B1|2018-03-09|2019-08-15|Avl List Gmbh|A method of determining a vehicle parameter of a vehicle record of a vehicle and using the vehicle parameter on a test bench|
    AT521422B1|2018-06-21|2020-09-15|Avl List Gmbh|Procedure for calibrating a technical system|
    EP3845979A1|2019-12-30|2021-07-07|Siemens Aktiengesellschaft|Method and system for adjusting the parameters of a system model|
    WO2022015324A1|2020-07-17|2022-01-20|Gram Labs, Inc.|System, method, and server for optimizing deployment of containerized applications|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    ATA50347/2013A|AT512977B1|2013-05-22|2013-05-22|Method for determining a model of an output of a technical system|ATA50347/2013A| AT512977B1|2013-05-22|2013-05-22|Method for determining a model of an output of a technical system|
    CN201480028121.5A| CN105209984B|2013-05-22|2014-05-20|For the method for the model for determining technological system output valve|
    US14/784,171| US10331810B2|2013-05-22|2014-05-20|Method for determining a model of an output quantity of a technical system|
    PCT/EP2014/060350| WO2014187828A1|2013-05-22|2014-05-20|Methods for ascertaining a model of a starting variable of a technical system|
    KR1020157031506A| KR102206324B1|2013-05-22|2014-05-20|Methods for ascertaining a model of a starting variable of a technical system|
    JP2016514378A| JP6404909B2|2013-05-22|2014-05-20|How to calculate the output model of a technical system|
    EP14727174.6A| EP2999998B1|2013-05-22|2014-05-20|Methods for ascertaining a model of an output variable of a technical system|
    [返回顶部]