• 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 system controls a plant process that includes manipulated variables (e.g., input state) and control variables (e.g., output state). The system includes sensor circuitry to provide measurements of control variables, memory to store correction time constants and the highest and lowest thresholds for at least one control variable. Include. The highest and lowest thresholds are separated by bands of values that can accommodate one control variable. The processor includes data describing the process model to relate the cost of the adjustment variable to the control variable and further provides predicted values for one control variable according to the solution. The logic circuitry within the processor adjusts the minimum cost of returning the predicted value of one control variable into an acceptable value band in response to the measured value function of one control variable that is outside the band of values. Determine the variable. Control means in the plant perform the operation of changing the adjustment variable (and the input state) in accordance with the signal from the processor.
    公开号:KR19990008140A
    申请号:KR1019970707665
    申请日:1996-04-26
    公开日:1999-01-25
    发明作者:바르투시악레이몬드도날드；폰테인로버트윌리암
    申请人:엑손케미칼패턴츠인코포레이티드；
    IPC主号:
    专利说明:

    Feedback method for nonlinear process control
    [2] US Pat. No. 4,349,869, entitled "Dynamic Matrix Control Method" by Prett et al. Describes a method and apparatus for controlling and optimizing the operation of a series of interdependent processes in a plant environment. In order to perform the control operation, attention is paid to the dynamic change in the output by carefully changing the input variables to the plant so that the future response of the process during the on-line operation can be predicted. To implement the control method, Fret et al. Created a table of values derived during the initial test phase. This table incorporates several inputs and the resulting outputs, which serve as important reference points during subsequent plant operation.
    [3] Fret et al. Are particularly adapted to control linear system behavior or behavior that can be simulated linearly. However, in the case of nonlinear plant operation, a method such as fretting cannot be performed properly, especially when there are a large number of control and adjustment variables. The control variable is a plant output that is affected by a change in input to one or more adjustment variables, for example a plant.
    [4] The application of the dynamic matrix control method to the polymerization process is described by Peterson et al. Under the name "A Non-linear DMC Algorithm and its Application to a Semibatch Polymerization Reactor", Chem. Eng. Science, Vol. 47, No. 4, pp. 737-753 (1992). Peterson et al. Employ non-linear controllers and numerical algorithms to find solutions, but their methods do not attempt to minimize input state costs when they reach a control solution. Is not. Brown et al. Are entitled "A Constrained Nonlinear Multivariable Control Algorithm," Trans I ChemE. Vol. 68 (A), Sept. 1990, pp. The document 464-476 describes a nonlinear controller that includes an acceptable output value of a specified level at which a control action is prohibited. However, Brown et al. Have not tested which inputs achieve the minimum cost and simultaneously control the output.
    [5] Prior art patents disclose several techniques that employ model-based control systems that employ both linear and nonlinear representations to relate control and adjustment variables. U.S. Patent No. 4,663,703 describes a reference predictive model controller by Axelby et al that employs a sub-system impulse model to simulate and predict future output. The system includes adjustable gain feedback and control loops that are adjusted to produce a dynamic system with constant characteristics even when the dynamic characteristics change.
    [6] U. S. Patent No. 5,260, 865 describes a nonlinear model based control system by Beauford et al. That employs a nonlinear model to calculate process vapor and distillate flow rates. Sanchez (No. 4,358,822) is an adaptive-predictive control system that determines the control vector to be applied to a process so that the model generates a process output that will be the desired value at a future time constant. It describes. The parameters of the model are updated in real time so that the output vector is close to the actual process vector. In US Pat. No. 5,268,834, Sanner et al. Employ a neural network to construct a control plant model.
    [7] When plant operation involves dynamic, nonlinear process operation, and includes many adjustment and control variables, the extension of the model-based control system for plant operation is not a simple matter. Until recently, process control computers of moderate size and cost lacked the processing power to simultaneously solve the solutions of several equations by modeling such dynamic plant processing.
    [8] Reference synthesis techniques have been developed for application to nonlinear control problems (eg pH control problems). In the reference system synthesis technique, it is desirable to have a non-linear plant system along a reference trajectory, and once the plant delay is over, it reaches a set point according to the first and second trajectories. Bartusiak et al., Entitled " Non-linear Feed Forward / Feedback Control Structures designed by Reference Systems Synthesis, " 44, No. 9, pages 1837-1851 (1989) describe a control process applicable to very nonlinear plant operation. In essence, Barthusiak et al. Describe a plant to be controlled by a series of differential equations. The integral-differential equation, which can be designed non-linearly, can represent the desired behavior of a closed loop control system. Certain operations are referred to as reference systems.
    [9] Bartosiac et al. Adjust the adjustment variables so that the system behaves as similar as possible to the reference system so that the desired closed loop operation is performed. The adjustment variable behavior is determined by equalizing or generally minimizing the difference between the open loop system and the desired closed loop system. Thereafter, the desired plant behavior is defined. Control variables are specified with tuning parameters that control the rate at which the control variable reaches the set point. In particular, the desired plant output parameters are set and the speed at which the control system reaches the desired output parameters in the control phase is indicated by the tuning parameters. Thus, a control function is driven that causes the output to reach a specified parameter value independent of each adjustment variable cost function. These results do not take into account changes in the cost of adjustment variables that can minimize costs as well as control plant efficiency efficiently.
    [10] It is therefore a primary object of the present invention to provide an improved method for nonlinear process control that allows tuning parameters to be applied to control variables.
    [11] It is a second object of the present invention to provide an improved method for nonlinear process control that can achieve the desired control variable while minimizing the cost of inputting the adjustment variable according to the control method.
    [12] Summary of the Invention
    [13] The system controls a plant process that includes manipulated variables (e.g., input state) and control variables (e.g., output state). The system includes sensor circuitry to provide measurements of control variables, memory to store correction time constants and the highest and lowest thresholds for at least one control variable. Include. The highest and lowest thresholds are separated by bands of values that can accommodate one control variable. The processor includes data describing the process model to relate the cost of the adjustment variable to the control variable and further provides predicted values for one control variable according to the solution. The logic circuitry within the processor adjusts the minimum cost of returning the predicted value of one control variable into an acceptable value band in response to the measured value function of one control variable that is outside the band of values. Determine the variable. Control means in the plant perform the operation of changing the adjustment variable (and the input state) in accordance with the signal from the processor.
    [1] FIELD OF THE INVENTION The present invention relates to process control systems, and in particular, model-based feedback in which there is a non-linear relationship between plant manipulated variables and plant control variables. A model-based feedback control system.
    [14] 1 is a block diagram of a system of the present invention.
    [15] 2 is a schematic diagram of a control function used in the present invention.
    [16] 3 and 4 are flow charts useful for understanding the operation of the present invention.
    [17] The following terms will then be employed to describe the invention.
    [18] Process Model A process model defines the behavior of a plant system and is formulated in the form of algebraic and differential equations in the continuous time domain.
    [19] Discretization of Manipulated Variables: Coordinated movement is a discrete time variable. The zero-order hold function is employed to provide discrete, adjusted shift variables for use in the process model.
    [20] Reference Trajectory: The reference trajectory specifies the performance of the controller as a rate of response of the control variable.
    [21] Objective Function: The objective function defines the optimal control performance. The objective function includes penalties for violating the control set point and economic cost (benefit) functions.
    [22] Manipulated Variable Limits: The adjusted variable limits are set to reflect secondary controller limits or conditions, such as range limits, set point limits, and anti wind-up conditions. .
    [23] Feedback: Feedback is integrated as a bias value that represents the error between process measurements and model prediction at the reference trajectory.
    [24] State Estimation: Predictions and outputs for process model states are derived from dynamic models based on adjustments and feedforward variables and current values from predictions derived during previous controller scan times. It is provided for each controller scan by integration.
    [25] Initialization: The controller's output is initialized by reading the value of the current adjustment variable in each scan and providing this value as the controller's moving increment. When the controller program is running (in a closed loop or an open loop), the model state and output are initialized to the values expected during the previous controller scan. The model state and output are initialized by obtaining a steady state model for the current adjustments and feedforward values when the program is first turned on.
    [26] 1, a digital computer based control system monitors the processes occurring in the plant 12. The process value is input to a nonlinear controller function 14 located within the digital control system 10. The process model 16 is stored in the digital control system 10 and specifies a series of nonlinear equations that provide a reference system to the nonlinear controller 14. The plurality of control parameters 18 provide constraints on the control values derived by the nonlinear controller 14. By comparing process values measurements with prediction values derived via solution model 16 (with control parameters 18), correction values are derived and applied to plant 12 as control inputs. .
    [27] In FIG. 2, the nonlinear controller 14 includes a dynamic process model 16 that defines the rate of change of process state with respect to changes in system adjustment variables, independent variables, and bias values. Nonlinear controller 14 further includes one or more tuning values that define closed loop process response characteristics. In particular, each process response characteristic defines the trajectory that the control variable will follow in response to a change in the adjustment variable. When there is a difference between the measured value and the predicted value derived from the process module 16, the optimization function 19 determines the minimized adjustment variable cost to achieve the desired response trajectory.
    [28] It can be seen that the nonlinear controller 14 sets the boundary of the limit value for one or more variables (eg, outputs) from the plant 12. Once the highest and lowest limit values for the control variable are set, the nonlinear controller 14 executes a control procedure that compares the measured rate of change between the desired rates of movement of the control variable with respect to the at least one limit value. If the control variable is within the maximum and minimum limits, no control action is taken. If the control variable is outside the limit value, the error rate of the change value is derived by comparing the dynamic rate of the measured change with the dynamic rate of the model change. The error rate of the change value is then employed by the objective function to determine the adjustment variable that will represent the minimum cost to obtain the return of the control variable within the highest and lowest limits. By defining the control variable values in the acceptable range using the highest and lowest limit values, the cost of the various adjustment variables to determine which combination can return the control variable to the limits and at the same time minimize the cost of the adjustment variable. You can test
    [29] 3 and 4, a description of the operation of the nonlinear controller 14 is shown. The nonlinear controller 14 runs on a general purpose computer integrated with the plant 12. The nonlinear controller 14 is run at a specified frequency or scan rate, for example one scan rate per minute, whereby the control variables are monitored and deriving movement for each to realize the control action. The adjustment variables are calculated.
    [30] The procedure begins by reading plant data into the digital control system 10 (box 30). Such data includes current values for control variables, adjustment variables and auxiliary or feedforward variables. The plant is provided by field instruments or off-line laboratory analyses. Next, the current measurement of each control variable is compared with the corresponding model prediction. The bias value indicative of plant / model mismatch is calculated as the difference between the measured value and the predicted value (box 32).
    [31] Then, as shown in box 34, the input data becomes valid (eg, abnormal conditions such as unavailable measurements or out of range values are discarded). In addition, condition setting of the data is performed, which includes filtering and setting of the adjustment variable boundary based on the operator-specified limit values and plant control system state values.
    [32] At the start of operation of the nonlinear controller 14, a cold start initialization is performed (see decision box 36). The values for the independent variables, adjustments or feedforward variables are read from a database stored in the digital control system 10 (box 38). The initialization operation calculates the plant state and model outputs that represent plant conditions such as temperature, configuration, and production characteristics.
    [33] Then, a state-space model will be used to describe the procedure. Each state is defined by an "x" vector value, and the plant output is represented by an "y" vector value. The independent variable is represented by the value "u".
    [34] o = F (x, u)
    [35] y = H (x)
    [36] The value for the plant state is then used as the initial value for the nonlinear controller 14 (see boxes 40 and 42). Thereafter, the status value is evaluated and written to memory (box 44). At this point, the nonlinear controller 14 begins operation of the process control algorithm (box 46).
    [37] As shown in FIG. 4, the control process reads process data from the plant control system hardware (box 48) to determine the current state of the process.
    [38] Process data includes the following:
    [39] Initial value for each model state.
    [40] Initial value for the predicted plant output.
    [41] Bias value indicating plant / model error.
    [42] Model parameters.
    [43] Current measurement of the independent variable.
    [44] Setpoint or target value for control variables and restrictions.
    [45] Boundaries for adjustment variables.
    [46] Input status condition.
    [47] The values for model state and predicted plant output are previous values from the last controller run or from the cold start initialization. The control variable (s) (eg the output to be controlled) and the threshold value set point are input by the operator. The set point is entered as the highest limit value and the lowest limit value. By using these values, adjustment variables (inputs) can be adjusted to minimize the cost of reaching control variable values within the highest and lowest limits. Model parameter values are predetermined. Current measurements of independent variables are derived by plant field instrumentation or laboratory analysis. The boundaries of the adjustment variables are based on the threshold values specified by the operator and the plant control system state values, as described above.
    [48] Thereafter, the operating mode of the controller is set (box 50). One controller mode enables model prediction to be calculated and control of derived signals without applying control signals to the plant. The digital control system is then assumed to be set to the full mode of operation, and the adjustment parameters are actively controlled in accordance with model calculations and measured system conditions.
    [49] The input data is converted into a form for use in the model / control system (box 52), and the condition evaluation procedure is started (box 54). Each condition is evaluated using a dynamic model of the plant. In the state / spatial models shown in equations (3) and (4) below, the state is represented by an "x" variable, the plant output is represented by a "y" variable, and the independent variable is represented by "y".
    [50] dx / dt = F (x, u)
    [51] y = H (x)
    [52] Equation 3 means that the rate of change of the model state is a function of the model state and the independent variable. Equation 4 means that the output is a function of the model state. The evaluation of the model is obtained by the integration of equations (3) and (4) from the last execution of the nonlinear controller 14 to the current time. Preferred calculation methods include orthogonal collocation, where equations 3 and 4 are divided into time segments, so that for the same time increment, the differential equations can be solved in parallel. .
    [53] Control calculations performed by the nonlinear controller 14 are performed by employing a sequential quadratic programming technique (box 56). Control calculations determine the future movement of the tuning variable that best matches to control future performance assignments on the time horizon. The nonlinear controller 14 uses the model of the plant, a reference trajectory defining the specified controller performance, the objective function (described below), and the boundaries of the adjustment variables. Movement of adjustment variables is separated in the future in the timeline.
    [54] The models shown in equations (3) and (4) are used. As mentioned above, the "u" variable means an independent variable, a subset of which is an adjustment variable (ie an input). The value for all independent variables is the "zero-order hold function" of the adjustment variable U k separated at step k of each time. The zero order hold function assumes that the value of the adjustment variable remains constant between program executions.
    [55] The reference trajectory specifies the controller performance upon change of the control variable according to the applied limit condition. Reference trajectories (5) and (6) show the relationship between the rate of change of the control variable and the error (difference) between the control variable set point and the measured control variable.
    [56] dy k / dt = (SPH k- (y k + b)) / T + Vhp k -Vhn k
    [57]
    [58] here,
    [59] SPH = highest limit for a control variable or restriction;
    [60] SPL = lowest limit value for the control variable or restriction;
    [61] y = predicted control variable;
    [62] b = bias associated with error in prediction and measurement;
    [63] Vhp = variation in amount of SP from measured SPH;
    [64] Vhn = negative variation from SPH of the measured variable;
    [65] Vln = positive variation from SPL of the measured variable;
    [66] Vln = negative variation from SPL of the measured variable;
    [67] k = time step into the future;
    [68] K = time step into the future on the timeline used by the controller,
    [69] T = time constant for the desired closed loop speed of the response of the controlled variable.
    [70] Each variable Vlp, Vhp, Vln, and Vhn is hereinafter referred to as a "violation" variable. Each violation variable allows for inequality to be converted into an equality relationship, and allows for prioritization of constraints through the application of a weighting function in the objective function. The objective function (ie, the relationship to be satisfied by the control operation) is given by the following equation.
    [71] MinSum (Wh * Vhp k + Wl * Vln k) + C (x, u)
    [72] here,
    [73] Wh, Wl = Penalty weights;
    [74] Vhp k , Vlp k = violation variables defined above;
    [75] C (x, u) = cost penalty function.
    [76] Equation 7 shows a sum minimization function for use when a violation occurs in the highest limit value or the lowest limit value of the control variable. Equation (7) applies a weighting factor that allows the positive or negative violation value to be emphasized (or not emphasized) in any case. Equation 7 also includes a term (ie, C (x, u)), which is an independent cost function for both adjustment variable u and model state x.
    [77] The control system solves Equation 7 and evaluates the summation from each solution when several changes in the adjustment variable are to occur. The purpose is to return the control variable y into the boundaries defined by the highest limit value (SHP) and the lowest limit value (SPL). Since SHP and SPL are separated by the span of the variable that defines the range of acceptable control variables, several possible changes within the adjustment variable are calculated so that any combination returns the control variable to the acceptable range while Determine the minimum cost for The adjustment variable (in any control action) can return the output of the plant within the width between SPH and SPL, with the first two representations in Equation 7 being zero, respectively, and the solution of the function It is closely related to the costs expressed.
    [78] The optimization solution of equation (7) depends on the boundary of additional adjustment variables as expressed in equations (8) and (9) below.
    [79] ulb <u k <uhp
    [80] ABS (u k -u ( k -1) <dub
    [81] here,
    [82] uhb = highest boundary of adjustment variable;
    [83] ulb = lowest boundary of the adjustment variable;
    [84] dub = boundary of change in u between time steps.
    [85] Once an acceptable solution is achieved, the output consisting of adjustment parameter values for each future time step is checked for system constraints (box 58). Assuming the validity of the output data, the data is written to memory (box 60) and the calculated adjustment variable is passed to the plant (box 62) to operate the field control element (ie the values).
    [86] It will be appreciated that the foregoing is merely illustrative of the present invention. Many alternatives and modifications may be devised by those skilled in the art without departing from the invention. Accordingly, it is an object of the present invention to cover all alternatives, modifications and variations that fall within the scope of the appended claims.
    权利要求:
    Claims (4)
    [1" claim-type="Currently amended] A plant process control system comprising an adjustment variable comprising an input state and a control variable comprising an output state,
    Sensor means for measuring at least the control variable;
    Storing, for at least one control variable, the highest and lowest limit values and a modified time constant separated by a band of values at which the at least one control variable can be accommodated Memory means for;
    A data describing the model of the plant process coupled to the sensor means and the memory means, the model of the plant process associating a cost of an adjustment variable with a control variable, the predicted value for the at least one control variable according to a solution And logic means to minimize the cost of the adjustment variable, wherein the adjustment variable is responsive to a measured value function of the at least one control variable that is out of the band of values. Processor means for generating a control signal to change in a predetermined direction such that the predicted value of the at least one control variable is within the band of values; And
    And control signal means for controlling the adjustment variable in response to the control signal to operate the means in the plant.
    [2" claim-type="Currently amended] The method of claim 1,
    The memory means includes an orbital response function for the model that prescribes a rate at which the control variable returns to the band of acceptable values when the control variable deviates from the highest threshold value, and the at least one Further storing data describing an orbital response function for the model that prescribes a rate of returning the one control variable to the band of acceptable values when the control variable deviates from the lowest limit value, wherein The two orbital response functions include a modified time constant, while representing a relationship between the measured rate of change of the at least one control variable and the desired rate of change, wherein the logic means uses the data to express the minimized cost input state. Determine plant process control system.
    [3" claim-type="Currently amended] The method of claim 2,
    The orbital response function for the at least one control variable is
    (Equation 5)
    dy k / dt = (SPH k- (y k + b)) / T + Vhp k -Vhn k
    (Equation 6)

    Is,
    here,
    SPH = highest limit for a control variable or restriction;
    SPL = lowest limit value for the control variable or restriction;
    y = predicted control variable;
    b = bias associated with error in prediction and measurement;
    Vhp = variation in amount of SP from measured SPH;
    Vhn = negative variation from SPH of the measured variable;
    Vln = positive variation from SPL of the measured variable;
    Vln = negative variation from SPL of the measured variable;
    k = time step into the future;
    K = time step into the future on the timeline used by the controller,
    T = plant process control system, which is a time constant for the desired closed loop speed of the response of the controlled variable.
    [4" claim-type="Currently amended] The method of claim 3, wherein
    The logic means is operative to provide a solution to a predetermined minimization relationship to determine a minimized cost adjustment variable for moving the at least one control variable into the band of values, the predetermined minimization relationship being:
    (Equation 7)
    MinSum (Wh * Vhp k + Wl * Vln k) + C (x, u)
    here,
    Wh, Wl = Penalty weights;
    Vhp k , Vlp k = violation variables;
    C (x, u) = cost penalty function
    Plant process control system represented by.
    类似技术:
    公开号 | 公开日 | 专利标题
    US8571691B2|2013-10-29|Scaling and parameterizing a controller
    JP6008898B2|2016-10-19|Online adaptive model predictive control in process control systems.
    Ge et al.2002|Robust PID controller design via LMI approach
    Qin et al.1997|An overview of industrial model predictive control technology
    US7315846B2|2008-01-01|Method and apparatus for optimizing a system model with gain constraints using a non-linear programming optimizer
    AU2001240799B2|2005-04-28|Process control system
    US6745088B2|2004-06-01|Multi-variable matrix process control
    US6493596B1|2002-12-10|Method and apparatus for controlling a non-linear mill
    Chen et al.2004|Applying neural networks to on-line updated PID controllers for nonlinear process control
    US5847952A|1998-12-08|Nonlinear-approximator-based automatic tuner
    US5568378A|1996-10-22|Variable horizon predictor for controlling dead time dominant processes, multivariable interactive processes, and processes with time variant dynamics
    Wojsznis et al.2003|Practical approach to tuning MPC
    Ou et al.2006|Comparison between PSO and GA for parameters optimization of PID controller
    US7630868B2|2009-12-08|Computer method and apparatus for constraining a non-linear approximator of an empirical process
    Fromion et al.1999|Nonlinear performance of a PI controlled missile: an explanation
    US7330767B2|2008-02-12|Configuration and viewing display for an integrated model predictive control and optimizer function block
    DE69909838T2|2004-04-22|Control units for setting optimal parameters in process control systems and methods for applying the same
    Marchetti et al.2014|Steady-state target optimization designs for integrating real-time optimization and model predictive control
    Ketata et al.1995|Fuzzy controller: design, evaluation, parallel and hierarchial combination with a pid controller
    De Keyser2009|Model based predictive control for linear systems
    KR940010393B1|1994-10-22|Process control device
    EP0496570B1|1998-06-03|Two-level system identifier apparatus with optimization
    US7653445B2|2010-01-26|Apparatus and method for model-based control
    KR900005546B1|1990-07-31|Adaptive process control system
    EP0624264B1|1997-11-26|Neuro-pid controller
    同族专利:
    公开号 | 公开日
    NO974959D0|1997-10-27|
    HU225571B1|2007-03-28|
    JP3949164B2|2007-07-25|
    NO318927B1|2005-05-23|
    DE69608796T2|2000-11-23|
    NO974959L|1997-12-19|
    CN1183148A|1998-05-27|
    BR9608042A|1999-01-26|
    DE69608796D1|2000-07-13|
    KR100371728B1|2003-03-15|
    HU9802089A3|1999-05-28|
    US5682309A|1997-10-28|
    CZ336797A3|1998-09-16|
    MX9708318A|1998-02-28|
    CA2217381A1|1996-10-31|
    AU702101B2|1999-02-11|
    AU5631496A|1996-11-18|
    EP0823078B1|2000-06-07|
    PL323049A1|1998-03-02|
    AT193771T|2000-06-15|
    CZ296539B6|2006-04-12|
    HU9802089A2|1998-12-28|
    EP0823078A1|1998-02-11|
    JPH11504454A|1999-04-20|
    CA2217381C|2005-06-14|
    WO1996034324A1|1996-10-31|
    PL182764B1|2002-02-28|
    TW297108B|1997-02-01|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    1995-04-28|Priority to US8/431,244
    1995-04-28|Priority to US08/431,244
    1995-04-28|Priority to US08/431244
    1996-04-26|Application filed by 엑손케미칼패턴츠인코포레이티드
    1999-01-25|Publication of KR19990008140A
    2003-03-15|Application granted
    2003-03-15|Publication of KR100371728B1
    优先权:
    申请号 | 申请日 | 专利标题
    US8/431,244|1995-04-28|
    US08/431,244|US5682309A|1995-04-28|1995-04-28|Feedback method for controlling non-linear processes|
    US08/431244|1995-04-28|
    [返回顶部]