Modelica system simulation model parameter identification method based on CVM analysis

Abstract

The invention provides a Modelica system simulation model parameter identification method based on CVM analysis, and relates to the technical field of system modeling simulation. The CVM analysis-based Modelica system simulation model parameter identification method comprises the following steps: step 1, modeling a speed regulation system mechanism model based on Modelica; 2, analyzing the sensitivity of the mechanism model parameters of the speed regulation system; and step 3, analyzing the parameter sensitivity of the speed regulation system based on the CVM. According to the method, the machine learning method based on data driving is applied to the field of modeling simulation of the complex equipment system, and the electric power system model is continuously verified by using the physical test data of the electric power system to ensure the accuracy of model parameters, so that the simulation confidence of the electric power system is improved; the method can solve the problem of uncertainty of speed regulation system parameters in a power system, improves the accuracy of the model, reduces the difference between the mechanism model and the actual speed regulation system operation state, and better reflects the system characteristics.

Description

Technical Field

[0001] The invention relates to the technical field of system modeling and simulation, and in particular to a Modelica system simulation model parameter identification method based on CVM analysis. Background Art

[0002] The power system digital twin simulates the operating conditions of the actual physical system on a computer and provides decision-making for the operation of the power system by comparing it with the data of the actual power system. Therefore, the power system digital twin places higher requirements on the accuracy and adaptability of the model. The power system model must be accurate enough to simulate the dynamic characteristics of the real power system, and the model must adapt to the ever-changing needs of the power system.

[0003] The mechanism model of the power system is the basis for power system planning, design, operation and control. However, in actual simulation calculations, many model parameters cannot be provided due to various reasons. Even if they are available, typical parameters provided by the manufacturer are often used. Due to the influence of many factors such as magnetic saturation, nonlinearity, and operating environment, these parameters are somewhat different from the actual unit operating parameters. In particular, the speed control system and power load not only play an important role in the adjustment of power quality (frequency, voltage), but their dynamic characteristics are also crucial to the dynamic behavior of the power system during electromagnetic transients and electromechanical transients, including the stability of power angle, voltage, frequency, and power flow. Therefore, the speed control system and the accuracy of the detailed model and parameters of the motor load are particularly important for the stability of the power system.

[0004] In response to the above problems, data-driven machine learning methods are applied to the modeling and simulation of complex equipment systems. The power system model is continuously verified using the actual test data of the power system to ensure the accuracy of the model parameters, thereby improving the simulation confidence of the power system. Power system parameter identification is of great significance in practical applications and can improve the stability, reliability, economy and other indicators of the power system. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the shortcomings of the prior art, the present invention provides a Modelica system simulation model parameter identification method based on CVM analysis to solve the uncertainty problem of speed regulation system parameters in the power system, improve the accuracy of the model, narrow the gap between the mechanism model and the actual speed regulation system operating status, and better reflect the system characteristics.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] The parameter identification method of Modelica system simulation model based on CVM analysis includes the following steps:

[0010] Step 1. Modeling the speed control system mechanism model based on Modelica

[0011] For the speed control system, the actual diesel engine speed is measured by the speed sensor and compared with the speed setting potentiometer. The speed control unit outputs the control command to the electromagnetic actuator. After the electromagnetic actuator is actuated, it drives the adjustment mechanism. The adjustment mechanism controls the diesel engine fuel injection amount to accelerate or decelerate the diesel engine. Among them, the speed sensor can use the proportional link, the control unit uses PID control, the actuator can use the first-order inertia link, and the adjustment mechanism can use the delay link for simulation. Therefore, the principle is clear, and the relevant parameters of each link need to be obtained to accurately simulate the speed control system.

[0012] Step 2. Parameter sensitivity analysis of the speed control system mechanism model

[0013] For the speed control system mechanism model, the main parameters to be identified are the electromagnetic actuator gain Ka, the electromagnetic actuator time constant Ta, and the opening and closing speeds Tco of the adjustment mechanism. The MWORKS.Sysplorer simulation tool is used to establish a multidisciplinary unified model of the speed control system, and the sensitivity of the influence of different factors on the speed control system is quantitatively analyzed.

[0014] Step 3. CVM-based speed control system parameter sensitivity analysis

[0015] Firstly, the parameter sensitivity of the Modelica speed regulation system and power load model to be identified is analyzed, the parameter sensitivities are sorted, and the parameters with high sensitivity are identified first. Then, the initial values ​​and ranges of the speed regulation system parameters are given, and an optimization algorithm based on the constrained variable scaling method (CVM) is used. The variance between the measured data and the model simulation data of the speed regulation system and power load in the power system is used as the objective function to perform a one-dimensional search and iterative optimization to obtain the optimal identification value of each parameter, which is the final identification result of the speed regulation system and power load model parameters.

[0016] Furthermore, in step 1, the speed regulation system adopts a conventional electronic speed regulator model, which mainly includes speed measurement, controller, differential regulation, actuator, and adjustment mechanism; in the speed regulation system model, each link is modeled using a transfer function; the speed sensor can be simulated using a proportional link due to its fast response; the controller is modeled using conventional PID control; the actuator can be simulated using a first-order inertia link due to its response time; the adjustment mechanism can be simulated using a delay link, specifically including:

[0017] 1) Speed ​​measurement, using first-order inertia link simulation

[0018]

[0019] Where:

[0020] Vm—Sensor measuring speed signal; Vg—Generator speed; K—Sensor proportional coefficient;

[0021] Tr—sensor response time;

[0022] 2) Controller, using traditional PID control

[0023]

[0024] Where:

[0025] Vset—speed regulator set speed; S—PID control output signal; Kp—PID control proportional coefficient; Ti—PID integral time constant; Td—PID differential time constant; kd—PID differential proportional coefficient;

[0026] 3) Differential regulation, mainly simulating the static speed regulation characteristics of the generator, which is conducive to the automatic distribution of the active power of the generator

[0027] V set =V0-P·δ / P N

[0028] Where:

[0029] V0—generator no-load speed; P—generator active power; PN—generator rated active power; δ—generator active power adjustment coefficient;

[0030] 4) Electromagnetic actuator, mainly simulates the control signal power amplification function, using the first-order inertia link simulation

[0031]

[0032] Where:

[0033] Sa—actuator output signal; S—PID control output signal; Ka—actuator adjustment gain;

[0034] Ta—actuator response time;

[0035] 5) Adjustment mechanism, mainly simulates the valve opening and closing speed, using integral link to simulate

[0036]

[0037] Where:

[0038] Vo—valve opening signal; Tco—valve opening and closing time.

[0039] Furthermore, the step 1 also includes: for the speed regulator model, the main factors that dominate its dynamic characteristics are the electromagnetic actuator gain Ka, the electromagnetic actuator time constant Ta, and the adjustment mechanism opening and closing speed Tco;

[0040] Since the field test data of the power system does not have the main steam valve opening measurement data, it can provide voltage, current and speed. Therefore, it is necessary to process the speed control system mechanism model. The generator active power P can be calculated through the instantaneous values ​​of voltage and current. Adding the measured speed n can meet the input of the speed control system model. In order to solve the problem of no valve opening data, for the no-load starting process, the power system generator and prime mover are no-load, so only the steam turbine needs to be equivalent. Here, the first-order inertia link fitting is adopted for the steam turbine, and the first-order inertia link time constant is estimated through the external characteristic curve of the steam turbine as the prime mover model, which is subsequently verified through the steam turbine data model.

[0041] Furthermore, in step 2, for the speed control system model, a model test toolbox is used in the MWORKS.Sysplorer simulation tool to perform sensitivity analysis on each key parameter through batch simulation, wherein, taking Ka, Ta, and Tco parameters as an example, different parameter sets are set for batch simulation tests, and the parameters are sorted by sensitivity according to the parameter variation range and dynamic response results.

[0042] Furthermore, in step 3, the constrained variable scaling method CVM is a constrained nonlinear programming algorithm, which performs linear search, HESSE matrix correction, differential gradient and quadratic programming on the objective function based on the initial variable scaling algorithm, converts the optimization mathematical model into a quadratic programming sub-problem, and then performs optimization to effectively combine quadratic programming, linear approximation, Lagrange multipliers, penalty functions, variable scaling and inexact search, thereby forming the characteristics of less function calculation and superlinear convergence.

[0043] Furthermore, in step 3, the parameter identification method and algorithm process based on the optimization algorithm are as follows:

[0044] 1) First, the speed control system model parameter sensitivity analysis method is used to obtain the main parameter sensitivity ranking of the speed control system as X = {Tco, Ka, Ta}, and its initial value is given as {1, 10, 5};

[0045] 2) Establish the optimization problem objective function f(X) and its constraint function g(X). For the speed control system, the optimization goal is to minimize the variance between the simulated generator speed and the measured generator speed under different test conditions, that is,

[0046] min f(X)=1 / (n*Σ(Vf-Vm)2)

[0047]

[0048] Where: Vf is the simulated generator speed, Vm is the measured generator speed, f(X) is the objective function, hi(X) and gi(X) are the equality constraints and inequality constraints respectively, and X* is the formula min f(X)=1 / (n*Σ

[0049] A feasible solution of (Vf-Vm)2);

[0050] The set of tight constraint indices I(X*) = {i|gi(X*) = 0}, let {▽gi(X*),▽hj(X*)|i∈I(X*)} be linearly independent. If X* is a local minimum, then there exist Lagrange multipliers μi≥0(i∈I(X*)) and μj such that

[0051]

[0052] 3) Solving the positive quadratic programming formula and introducing the Lagrange auxiliary function

[0053]

[0054] According to the formula Mode Can be simplified to

[0055]

[0056] General Perform Taylor expansion at Xk and transform the original optimization problem 1 into a positive definite quadratic programming subproblem, which can be expressed as:

[0057]

[0058] Where ▽f is the gradient vector of the original objective function; ▽g and ▽h are the gradient matrices of the constraint conditions; Bk is the Lagrangian function A suitable approximation matrix of the second-order derivative matrix, which is similar to the HESSE matrix, also known as the variable scaling matrix, dk = Xk + 1-Xk is the change in variable X;

[0059] 4) Select the objective function of the one-dimensional search, take the solution dk of the quadratic programming subproblem as a search direction, and perform a one-dimensional search for the benefit function ω(t) along the direction dk starting from Xk:

[0060]

[0061] in, The iteration rule is

[0062]

[0063] 5) One-dimensional search

[0064] The one-dimensional search method uses the following two-point quadratic difference method, record

[0065] φ(t)=Φ(X k +td k ,μ k ,σ k )

[0066] And take

[0067] Finally, the one-dimensional search algorithm is as follows:

[0068] ① Set t = 1. If φ(λ) < φ(0) + 0.1δλ, then set λ k =λ, stop; otherwise, go to the next step;

[0069] ②Calculation Place Enter ①.

[0070] Obtain the optimal step length tk, iterate repeatedly, when the optimal solution dk = 0 of the quadratic programming subproblem, the corresponding Xk is the optimal solution X* of the original optimization problem, and there are optimal multipliers λ* and μ*. The optimal solution X* is the identified parameter set {Tco, Ka, Ta}.

[0071] (6) Bk iterative correction

[0072] When the optimal solution is not obtained, it is necessary to iteratively correct Bk so that the iteration can continue. The requirement for Bk+1 is that it is symmetric, positive definite and approximates the Lagrangian function. The second-order derivative matrix of

[0073]

[0074] ΔX k =X k+1 -X k

[0075] In order to make Bk+1 meet the requirements, take

[0076]

[0077] η k =θΔg k +(1-θ)B k ΔX k

[0078] in

[0079] Furthermore, the specific use process of the model test toolbox is as follows:

[0080] Source model: Select the single steam generator set model as the simulation instance for model calibration, and the system will automatically call the solver associated with the instance;

[0081] Adjustment parameters: Select the system adjustment mechanism opening and closing speed Tco, the adjustment parameters are the electromagnetic actuator gain Ka, and the electromagnetic actuator time constant Ta as the parameters to be adjusted. Obviously, the selected parameters must have an impact on the result variable to be feasible. If the result variable output remains unchanged after changing the adjustment parameters, the parameter calibration will fail;

[0082] Test data: Browse external measurement data. Here, the actual steam generator set data is used as the test data.

[0083] Fixed parameters: Select one or more fixed parameters. If the model parameters are associated with specific product parameters, operating condition options, etc., these parameters will remain unchanged in the actual physical scenario;

[0084] Result variables: Select one or more result variables and specify the associated measurement indicators. These variables will be displayed in the form of curves, from which the difference between the two can be intuitively seen. As mentioned above, the sum of squares of the residuals of the model simulation result variables and the corresponding measurement indicator variables at each measurement time point is used as the model validation criterion, and minimizing this value is used as the target of parameter calibration;

[0085] Solution settings: set the solution start and end time, step size, algorithm, error and other options. Options: set parameter calibration process options, such as convergence error, maximum number of iterations, etc.

[0086] (III) Beneficial effects

[0087] The present invention provides a method for identifying parameters of a Modelica system simulation model based on CVM analysis. It has the following beneficial effects:

[0088] 1. The present invention provides a Modelica system simulation model parameter identification method based on CVM analysis, applies a data-driven machine learning method to the field of complex equipment system modeling and simulation, and uses the actual test data of the power system to continuously verify the power system model to ensure the accuracy of the model parameters, thereby improving the simulation confidence of the power system.

[0089] 2. The present invention provides a Modelica system simulation model parameter identification method based on CVM analysis, which can solve the uncertainty problem of speed regulation system parameters in the power system, improve the accuracy of the model, narrow the gap between the mechanism model and the actual speed regulation system operating state, and better reflect the system characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A diagram of the mechanism model of the speed regulation system provided by an embodiment of the present invention;

[0091] Figure 2 The simulation results of the Ka parameter set provided by the embodiment of the present invention;

[0092] Figure 3 The simulation results of the Ta parameter set provided by the embodiment of the present invention;

[0093] Figure 4 The simulation results of the Tco parameter set provided by the embodiment of the present invention;

[0094] Figure 5 A flow chart of a constrained variable-scaling CVM algorithm provided by an embodiment of the present invention;

[0095] Figure 6 A toolbox interface diagram for parameter identification provided by an embodiment of the present invention;

[0096] Figure 7 This is a comparison chart between simulation results and actual measurements provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0097] The following will be combined with the drawings in the specific embodiments of the present invention to clearly and completely describe the technical solutions in the specific embodiments of the present invention. Obviously, the specific embodiments described are only part of the specific embodiments of the present invention, not all of the specific embodiments. Based on the specific embodiments of the present invention, all other specific embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0099] The specific implementation of the present invention provides a method for identifying parameters of a Modelica system simulation model based on CVM analysis, comprising the following steps:

[0100] Step 1. Modeling the speed control system mechanism model based on Modelica

[0101] For the speed control system, the actual diesel engine speed is measured by the speed sensor and compared with the speed setting potentiometer. The speed control unit outputs the control command to the electromagnetic actuator. After the electromagnetic actuator is actuated, it drives the adjustment mechanism. The adjustment mechanism controls the diesel engine fuel injection amount to accelerate or decelerate the diesel engine. Among them, the speed sensor can use the proportional link, the control unit uses PID control, the actuator can use the first-order inertia link, and the adjustment mechanism can use the delay link for simulation. Therefore, the principle is clear, and the relevant parameters of each link need to be obtained to accurately simulate the speed control system.

[0102] The speed control system adopts the conventional electronic speed regulator model, which mainly includes speed measurement, controller, differential regulation, actuator, and adjustment mechanism. Figure 1 As shown in the figure, in the speed control system model, each link is modeled by transfer function; the speed sensor can be simulated by proportional link due to its fast response; the controller is modeled by conventional PID control; the actuator can be simulated by first-order inertia link due to its response time; the adjustment mechanism can be simulated by delay link, including:

[0103] 1) Speed ​​measurement, using first-order inertia link simulation

[0104]

[0105] Where:

[0106] Vm—Sensor measuring speed signal; Vg—Generator speed; K—Sensor proportional coefficient;

[0107] Tr—sensor response time;

[0108] 2) Controller, using traditional PID control

[0109]

[0110] Where:

[0111] Vset—speed regulator set speed; S—PID control output signal; Kp—PID control proportional coefficient; Ti—PID integral time constant; Td—PID differential time constant; kd—PID differential proportional coefficient;

[0112] 3) Differential regulation, mainly simulating the static speed regulation characteristics of the generator, which is conducive to the automatic distribution of the active power of the generator

[0113] V set =V0-P·δ / P N

[0114] Where:

[0115] V0—generator no-load speed; P—generator active power; PN—generator rated active power;

[0116] δ—Generator active power regulation coefficient;

[0117] 4) Electromagnetic actuator, mainly simulates the control signal power amplification function, using the first-order inertia link simulation

[0118]

[0119] Where:

[0120] Sa—actuator output signal; S—PID control output signal; Ka—actuator adjustment gain;

[0121] Ta—actuator response time;

[0122] 5) Adjustment mechanism, mainly simulates the valve opening and closing speed, using integral link to simulate

[0123]

[0124] Where:

[0125] Vo—valve opening signal; Tco—valve opening and closing time;

[0126] For the speed regulator model, the main factors that dominate its dynamic characteristics are the electromagnetic actuator gain Ka, the electromagnetic actuator time constant Ta, and the opening and closing speed Tco of the adjustment mechanism;

[0127] Since the field test data of the power system does not have the main steam valve opening measurement data, it can provide voltage, current and speed. Therefore, it is necessary to process the speed control system mechanism model. The generator active power P can be calculated through the instantaneous values ​​of voltage and current. Adding the measured speed n can meet the input of the speed control system model. In order to solve the problem of no valve opening data, for the no-load starting process, the power system generator and prime mover are no-load, so only the steam turbine needs to be equivalent. Here, the first-order inertia link fitting is adopted for the steam turbine, and the first-order inertia link time constant is estimated through the external characteristic curve of the steam turbine as the prime mover model, which is subsequently verified through the steam turbine data model.

[0128] Step 2. Parameter sensitivity analysis of the speed control system mechanism model

[0129] For the speed control system mechanism model, the main parameters to be identified are the electromagnetic actuator gain Ka, the electromagnetic actuator time constant Ta, and the opening and closing speeds Tco of the adjustment mechanism. The MWORKS.Sysplorer simulation tool is used to establish a multidisciplinary unified model of the speed control system, and the sensitivity of the influence of different factors on the speed control system is quantitatively analyzed.

[0130] For the speed control system model, the model test toolbox in the MWORKS.Sysplorer simulation tool is used to perform sensitivity analysis on each key parameter through batch simulation. Taking Ka, Ta, and Tco parameters as examples, different parameter sets are set for batch simulation tests, and the parameters are ranked according to the parameter variation range and dynamic response results. The corresponding batch simulation results are shown in Figure 2. Figure 2-Figure 4 As shown, from the simulation results, it can be seen that the actuator opening and closing speed Tco parameter is the most sensitive and has the greatest impact on the speed characteristics, followed by the main regulator parameter Ka, and the main regulator time constant Ta has a slightly smaller impact on the speed characteristics;

[0131] Step 3. CVM-based speed control system parameter sensitivity analysis

[0132] Firstly, the parameter sensitivity of the Modelica speed regulation system and power load model to be identified is analyzed, the parameter sensitivity is sorted, and the parameters with high sensitivity are identified first. Then, the initial value and range of the speed regulation system parameters are given, and the optimization algorithm based on the constrained variable scaling method (CVM) is used. The variance between the measured data and the model simulation data of the speed regulation system and power load in the power system is used as the objective function to perform a one-dimensional search and iterative optimization to obtain the optimal identification value of each parameter, which is the final identification result of the speed regulation system and power load model parameters.

[0133] The constrained variable scaling method (CVM) is a constrained nonlinear programming algorithm. It performs linear search, HESSE matrix correction, differential gradient and quadratic programming on the objective function based on the initial variable scaling algorithm, transforms the optimization mathematical model into a quadratic programming sub-problem, and then performs optimization to effectively combine quadratic programming, linear approximation, Lagrange multipliers, penalty functions, variable scaling and inexact search, thus forming the characteristics of less function calculation and superlinear convergence.

[0134] The parameter identification method and algorithm process based on the optimization algorithm are as follows:

[0135] 1) First, the speed control system model parameter sensitivity analysis method is used to obtain the main parameter sensitivity ranking of the speed control system as X = {Tco, Ka, Ta}, and its initial value is given as {1, 10, 5};

[0136] 2) Establish the optimization problem objective function f(X) and its constraint function g(X). For the speed control system, the optimization goal is to minimize the variance between the simulated generator speed and the measured generator speed under different test conditions, that is,

[0137] min f(X)=1 / (n*Σ(Vf-Vm)2)

[0138]

[0139] Where: Vf is the simulated generator speed, Vm is the measured generator speed, f(X) is the objective function, hi(X) and gi(X) are the equality constraint and inequality constraint respectively, and X* is the feasible solution of min f(X) = 1 / (n*Σ(Vf-Vm)2);

[0140] The set of tight constraint indices I(X*) = {i|gi(X*) = 0}, let {▽gi(X*),▽hj(X*)|i∈I(X*)} be linearly independent. If X* is a local minimum, then there exist Lagrange multipliers μi≥0(i∈I(X*)) and μj such that

[0141]

[0142] 3) Solving the positive quadratic programming formula and introducing the Lagrange auxiliary function

[0143]

[0144] According to the formula Mode Can be simplified to

[0145]

[0146] General Perform Taylor expansion at Xk and transform the original optimization problem 1 into a positive definite quadratic programming subproblem, which can be expressed as:

[0147]

[0148] Where ▽f is the gradient vector of the original objective function; ▽g and ▽h are the gradient matrices of the constraint conditions; Bk is the Lagrangian function A suitable approximation matrix of the second-order derivative matrix, which is similar to the HESSE matrix, also known as the variable scaling matrix, dk = Xk + 1-Xk is the change in variable X;

[0149] 4) Select the objective function of the one-dimensional search, take the solution dk of the quadratic programming subproblem as a search direction, and perform a one-dimensional search on the benefit function ω(t) along the direction dk starting from Xk:

[0150]

[0151] in, The iteration rule is

[0152]

[0153] 5) One-dimensional search

[0154] The one-dimensional search method uses the following two-point quadratic difference method, record

[0155] φ(t)=Φ(X k +td k ,μ k ,σ k )

[0156] And take

[0157] Finally, the one-dimensional search algorithm is as follows:

[0158] ① Set t = 1. If φ(λ) < φ(0) + 0.1δλ, then set λ k=λ, stop; otherwise, go to the next step;

[0159] ②Calculation Place Enter ①.

[0160] Obtain the optimal step length tk, iterate repeatedly, when the optimal solution dk = 0 of the quadratic programming subproblem, the corresponding Xk is the optimal solution X* of the original optimization problem, and there are optimal multipliers λ* and μ*. The optimal solution X* is the identified parameter set {Tco, Ka, Ta}.

[0161] (6) Bk iterative correction

[0162] When the optimal solution is not obtained, it is necessary to iteratively correct Bk so that the iteration can continue. The requirement for Bk+1 is that it is symmetric, positive definite and approximates the Lagrangian function. The second-order derivative matrix of

[0163]

[0164] ΔX k =X k+1 -X k

[0165] In order to make Bk+1 meet the requirements, take

[0166]

[0167] η k =θΔg k +(1-θ)B k ΔX k

[0168] in

[0169] The specific algorithm flow of the constrained variable scaling method is as follows Figure 5 As shown;

[0170] Through the study of model parameter identification technology based on the above algorithm flow, a model calibration module was developed in the MWORKS.Sysplorer platform for ease of use. Figure 6As shown, this module can configure the parameter correction process, define the parameters to be corrected and the analysis objectives, and provide parameter correction algorithms, including efficient global optimization, evolutionary strategy, quasi-Newton method, conjugate gradient, pattern search and their combinations; support automatic iteration to obtain the parameter combination with the smallest error with the test data or design indicators; support constraints on the objective function. After setting the maximum and minimum values ​​of the objective function, the corrected or optimized optimal solution will meet the constraints. It supports adding multiple objectives at the same time for optimization calculations, can compare and display the error evaluation results before and after correction, and provides display and export functions for correction calculation results.

[0171] The specific use process of the model test toolbox is as follows:

[0172] Source model: Select the single steam generator set model as the simulation instance for model calibration, and the system will automatically call the solver associated with the instance;

[0173] Adjustment parameters: Select the system adjustment mechanism opening and closing speed Tco, the adjustment parameters are the electromagnetic actuator gain Ka, and the electromagnetic actuator time constant Ta as the parameters to be adjusted. Obviously, the selected parameters must have an impact on the result variable to be feasible. If the result variable output remains unchanged after changing the adjustment parameters, the parameter calibration will fail;

[0174] Test data: Browse external measurement data. Here, the actual steam generator set data is used as the test data.

[0175] Fixed parameters: Select one or more fixed parameters. If the model parameters are associated with specific product parameters, operating condition options, etc., these parameters will remain unchanged in the actual physical scenario;

[0176] Result variables: Select one or more result variables and specify the associated measurement indicators. These variables will be displayed in the form of curves, from which the difference between the two can be intuitively seen. As mentioned above, the sum of squares of the residuals of the model simulation result variables and the corresponding measurement indicator variables at each measurement time point is used as the model validation criterion, and minimizing this value is used as the target of parameter calibration;

[0177] Solution settings: set the solution start and end time, step size, algorithm, error and other options. Options: set parameter calibration process options, such as convergence error, maximum number of iterations, etc.

[0178] Based on the above toolbox, a single steam turbine generator model is used as the source model, the speed control system adjustment mechanism opening and closing speed Tco, the adjustment parameters are the electromagnetic actuator gain Ka, and the electromagnetic actuator time constant Ta are set as the adjustment parameters. The active, reactive, voltage, speed and other data calculated based on the measurement data of 10 steam generator sudden increase and sudden unloading conditions are imported. The iterative solution parameters and CVM optimization algorithm configuration are set, and the identified parameter values ​​are obtained. The results of parameter identification are assigned to the system model, and the operating conditions of the simulation model are kept consistent with the operating conditions of the measured data. The simulation curve is obtained by simulation. The simulation results of the speed control system model based on the parameter identification results are shown as follows: Figure 7 shown.

[0179] The wavy curve is the measured generator speed, and the smooth curve is the simulation result of the generator speed. Under the sudden load condition, the dynamic error of the generator speed is less than 0.3%. By comparing the simulation result curve of the identified parameters with the measured data, it can be seen that the simulation results of the parameters input to the speed control system after CVM identification can better fit the dynamic characteristics of the speed control system, verifying the effectiveness of the method.

[0180] Although specific embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the specific embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A Modelica system simulation model parameter identification method based on CVM analysis, characterized in that: The following steps are involved: Step 1. Modeling the speed control system mechanism model based on Modelica For the speed control system, the actual diesel engine speed is measured by the speed sensor and compared with the speed setting potentiometer. The speed control unit outputs the control command to the electromagnetic actuator. After the electromagnetic actuator is actuated, it drives the adjustment mechanism. The adjustment mechanism controls the diesel engine fuel injection amount to accelerate or decelerate the diesel engine. Among them, the speed sensor can use the proportional link, the control unit uses PID control, the actuator can use the first-order inertia link, and the adjustment mechanism can use the delay link for simulation. Therefore, the principle is clear, and the relevant parameters of each link need to be obtained to accurately simulate the speed control system. Step 2. Parameter sensitivity analysis of the speed control system mechanism model For the speed control system mechanism model, the main parameters to be identified are the electromagnetic actuator gain Ka, the electromagnetic actuator time constant Ta, and the opening and closing speeds Tco of the adjustment mechanism. The MWORKS.Sysplorer simulation tool is used to establish a multidisciplinary unified model of the speed control system, and the sensitivity of the influence of different factors on the speed control system is quantitatively analyzed. Step 3. CVM-based speed control system parameter sensitivity analysis Firstly, the parameter sensitivity of the Modelica speed regulation system and power load model to be identified is analyzed, the parameter sensitivities are sorted, and the parameters with high sensitivity are identified first. Then, the initial values ​​and ranges of the speed regulation system parameters are given, and an optimization algorithm based on the constrained variable scaling method (CVM) is used. The variance between the measured data and the model simulation data of the speed regulation system and power load in the power system is used as the objective function to perform a one-dimensional search and iterative optimization to obtain the optimal identification value of each parameter, which is the final identification result of the speed regulation system and power load model parameters.

2. The method for identifying parameters of a Modelica system simulation model based on CVM analysis according to claim 1, characterized in that: In step 1, the speed regulation system adopts a conventional electronic speed regulator model, which mainly includes speed measurement, controller, differential regulation, actuator, and adjustment mechanism; in the speed regulation system model, each link is modeled using a transfer function; the speed sensor can be simulated using a proportional link due to its fast response; the controller is modeled using conventional PID control; the actuator can be simulated using a first-order inertia link due to its response time; the adjustment mechanism can be simulated using a delay link, specifically including: 1) Speed ​​measurement, using first-order inertia link simulation Where: Vm—Sensor measuring speed signal; Vg—Generator speed; K—Sensor proportional coefficient; Tr—sensor response time; 2) Controller, using traditional PID control Where: Vset—speed regulator set speed; S—PID control output signal; Kp—PID control proportional coefficient; Ti—PID integral time constant; Td—PID differential time constant; kd—PID differential proportional coefficient; 3) Differential regulation, mainly simulating the static speed regulation characteristics of the generator, which is conducive to the automatic distribution of the active power of the generator V set =V0-P·δ / P N Where: V0—generator no-load speed; P—generator active power; PN—generator rated active power; δ—generator active power adjustment coefficient; 4) Electromagnetic actuator, mainly simulates the control signal power amplification function, using the first-order inertia link simulation Where: Sa—actuator output signal; S—PID control output signal; Ka—actuator adjustment gain; Ta—actuator response time; 5) Adjustment mechanism, mainly simulates the valve opening and closing speed, using integral link to simulate Where: Vo—valve opening signal; Tco—valve opening and closing time.

3. The method for identifying parameters of a Modelica system simulation model based on CVM analysis according to claim 1, characterized in that: The step 1 also includes: for the speed regulator model, the main factors that dominate its dynamic characteristics are the electromagnetic actuator gain Ka, the electromagnetic actuator time constant Ta, and the adjustment mechanism opening and closing speed Tco; Since the field test data of the power system does not have the main steam valve opening measurement data, it can provide voltage, current and speed. Therefore, it is necessary to process the speed control system mechanism model. The generator active power P can be calculated through the instantaneous values ​​of voltage and current. Adding the measured speed n can meet the input of the speed control system model. In order to solve the problem of no valve opening data, for the no-load starting process, the power system generator and prime mover are no-load, so only the steam turbine needs to be equivalent. Here, the first-order inertia link fitting is adopted for the steam turbine, and the first-order inertia link time constant is estimated through the external characteristic curve of the steam turbine as the prime mover model, which is subsequently verified through the steam turbine data model.

4. The method for identifying parameters of a Modelica system simulation model based on CVM analysis according to claim 1, characterized in that: In step 2, for the speed control system model, a model test toolbox is used in the MWORKS.Sysplorer simulation tool to perform sensitivity analysis on various key parameters through batch simulation. Taking Ka, Ta, and Tco parameters as examples, different parameter sets are set for batch simulation tests, and the parameters are sorted by sensitivity according to the parameter variation range and dynamic response results.

5. The method for identifying parameters of a Modelica system simulation model based on CVM analysis according to claim 1, characterized in that: In step 3, the constrained variable scaling method CVM is a constrained nonlinear programming algorithm, which performs linear search, HESSE matrix correction, differential gradient and quadratic programming on the objective function based on the initial variable scaling algorithm, converts the optimization mathematical model into a quadratic programming sub-problem, and then performs optimization to effectively combine quadratic programming, linear approximation, Lagrange multipliers, penalty functions, variable scaling and inexact search, thereby forming the characteristics of less function calculation and superlinear convergence.

6. The method for identifying parameters of a Modelica system simulation model based on CVM analysis according to claim 1, characterized in that: In step 3, the parameter identification method and algorithm process based on the optimization algorithm are as follows: 1) First, the speed control system model parameter sensitivity analysis method is used to obtain the main parameter sensitivity ranking of the speed control system as X = {Tco, Ka, Ta}, and its initial value is given as {1, 10, 5}; 2) Establish the optimization problem objective function f(X) and its constraint function g(X). For the speed control system, the optimization goal is to minimize the variance between the simulated generator speed and the measured generator speed under different test conditions, that is, min f(X)=1 / (n*Σ(Vf-Vm)2) Where: Vf is the simulated generator speed, Vm is the measured generator speed, f(X) is the objective function, hi(X) and gi(X) are the equality constraints and inequality constraints respectively, and X* is the formula min f(X)=1 / (n*Σ A feasible solution of (Vf-Vm)2); The set of tight constraint indices I(X*) = {i|gi(X*) = 0}, let {▽gi(X*),▽hj(X*)|i∈I(X*)} be linearly independent. If X* is a local minimum, then there exist Lagrange multipliers μi≥0(i∈I(X*)) and μj such that 3) Solving the positive quadratic programming formula and introducing the Lagrange auxiliary function According to the formula Mode It can be simplified to ▽L(X * , μ * ,λ * )=0 The formula ▽L(X * , μ * ,λ * )=0, and make Taylor expansion at Xk to transform the original optimization problem 1 into a positive definite quadratic programming subproblem, which can be expressed as: Where ▽f is the gradient vector of the original objective function; ▽g and ▽h are the gradient matrices of the constraint conditions; Bk is the Lagrangian function A suitable approximation matrix of the second-order derivative matrix, which is similar to the HESSE matrix, also known as the variable scaling matrix, dk = Xk + 1-Xk is the change in variable X; 4) Select the objective function of the one-dimensional search, take the solution dk of the quadratic programming subproblem as a search direction, and perform a one-dimensional search for the benefit function ω(t) along the direction dk starting from Xk: in, The iteration rule is 5) One-dimensional search The one-dimensional search method uses the following two-point quadratic difference method, record φ(t)=Φ(X k +td k ,μ k ,σ k ) And take Finally, the one-dimensional search algorithm is as follows: ① Set t = 1. If φ(λ) < φ(0) + 0.1δλ, then set λ k =λ, stop; otherwise, go to the next step; ②Calculation Place Enter ①. Obtain the optimal step length tk, iterate repeatedly, when the optimal solution dk = 0 of the quadratic programming subproblem, the corresponding Xk is the optimal solution X* of the original optimization problem, and there are optimal multipliers λ* and μ*. The optimal solution X* is the identified parameter set {Tco, Ka, Ta}. (6) Bk iterative correction When the optimal solution is not obtained, it is necessary to iteratively correct Bk so that the iteration can continue. The requirement for Bk+1 is that it is symmetric, positive definite and approximates the Lagrangian function. The second-order derivative matrix of Δg k =▽ X L(X k+1 ,m (k+1) ,λ(k+1))-▽ X L(X k ,m (k+1) ,l (k+1) ) ΔX k =X k+1 -X k In order to make Bk+1 meet the requirements, take or k =θΔg k +(1-θ)B k ΔX k in 7. The method for identifying parameters of a Modelica system simulation model based on CVM analysis according to claim 4, characterized in that: The specific use process of the model test toolbox is as follows: Source model: Select the single steam generator set model as the simulation instance for model calibration, and the system will automatically call the solver associated with the instance; Adjustment parameters: Select the system adjustment mechanism opening and closing speed Tco, the adjustment parameters are the electromagnetic actuator gain Ka, and the electromagnetic actuator time constant Ta as the parameters to be adjusted. Obviously, the selected parameters must have an impact on the result variable to be feasible. If the result variable output remains unchanged after changing the adjustment parameters, the parameter calibration will fail; Test data: Browse external measurement data. Here, the actual steam generator set data is used as the test data. Fixed parameters: Select one or more fixed parameters. If the model parameters are associated with specific product parameters, operating condition options, etc., these parameters will remain unchanged in the actual physical scenario; Result variables: Select one or more result variables and specify the associated measurement indicators. These variables will be displayed in the form of curves, from which the difference between the two can be intuitively seen. As mentioned above, the sum of squares of the residuals of the model simulation result variables and the corresponding measurement indicator variables at each measurement time point is used as the model validation criterion, and minimizing this value is used as the target of parameter calibration; Solution settings: set the solution start and end time, step size, algorithm, error and other options. Options: set parameter calibration process options, such as convergence error, maximum number of iterations, etc.

Images