Proportional integral derivative (PID) parameter setting method based on deaerator system

Abstract

The application provides a proportional integral differential (PID) parameter setting method based on a deaerator system, simulates a PID controller by constructing an action network based on a neural network, and constructs an evaluation network to evaluate the performance of the PID controller, so as to optimize and iterate the action network and the evaluation network by using the self-learning and self-adaptive characteristics of the neural network, and to realize the adaptive optimization of the action network and the evaluation network in combination with the running state data of the deaerator system, so as to achieve the purpose of optimizing the control performance of the PID controller simulated by the action network, and to run the deaerator system in a stable running state based on the optimized PID controller, so as to eliminate the defects of the security risks, system overshoot or delay of the boiler feed water treatment system caused by the running stability problems of the deaerator system.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, and in particular to a proportional-integral-derivative (PID) parameter tuning method based on a deaerator system. Background Technology

[0002] In the boiler feedwater system of a thermal power plant, the deaerator system heats the boiler feedwater before unit startup by activating the boiler steam supply equipment or temporary steam supply equipment. According to Dalton's law of partial pressures, Henry's law, and the principles of heat and mass transfer equations, this not only removes dissolved oxygen from the boiler feedwater but also removes corrosive gases such as free CO2, NH3, and H2S, preventing corrosion of thermal equipment. Therefore, ensuring the safe and stable operation of the deaerator system plays an extremely important role in the boiler feedwater system.

[0003] In existing technologies, deaerator systems suffer from numerous operational stability issues during unit startup, such as deaerator vibration, steam-water embrittlement, and water level fluctuations. Among these, deaerator vibration is the most common. Deaerator vibration not only poses safety hazards to the boiler feedwater treatment system but also significantly reduces the reliability of the deaerator system during commissioning and prolongs unit startup time, thus failing to achieve the overall safety and economic efficiency of unit startup.

[0004] Therefore, for the technical problems of safety hazards, system overshoot, or delay in boiler feedwater treatment systems caused by the operational stability issues of deaerator systems in existing technologies, there are no effective solutions for the technical problems in related fields. Summary of the Invention

[0005] This invention provides a proportional-integral-derivative (PID) parameter tuning method based on a deaerator system to solve the defects in the existing technology caused by the operational stability problem of the deaerator system, which leads to safety hazards, system overshoot, and delay in the boiler feedwater treatment system. This method achieves operational stability of the deaerator system and avoids safety hazards, system overshoot, or delay in the boiler feedwater treatment system.

[0006] This invention provides a proportional-integral-derivative (PID) parameter tuning method based on a deaerator system, comprising: constructing an action network and an evaluation network based on a neural network and a PID controller corresponding to the deaerator system, wherein the action network is used to simulate the PID controller and the evaluation network is used to evaluate the performance of the PID controller; in one loop task, acquiring the current weights of the action network and the evaluation network, acquiring the operating status data of the deaerator system and the output data of the action network, and iteratively updating the current weights of the evaluation network and the action network based on the operating status data and the output data; repeating the above loop task until the current network convergence accuracy of the action network reaches a preset network convergence accuracy, ending the loop task, and acquiring the current weights of the action network after the loop task ends as target weights, and acquiring the target PID parameters of the tuned PID controller based on the target weights.

[0007] According to the present invention, a proportional-integral-derivative (PID) parameter tuning method for a deaerator system includes iteratively updating the current weights of the evaluation network and the action network based on the operating state data and the output data. The method comprises: inputting the operating state data and the output data into the evaluation network to obtain the current performance index function of the PID controller; iteratively updating the current weights of the evaluation network based on the current performance index function to obtain the optimal performance index function of the PID controller; iteratively updating the current weights of the action network based on the optimal performance index function to obtain the current control law of the PID controller; and updating the output data of the action network based on the current control law.

[0008] According to the present invention, a proportional-integral-derivative (PID) parameter tuning method for a deaerator system includes the following steps: inputting the operating state data and the output data into the evaluation network to obtain the current performance index function of the PID controller; inputting the operating state data and the output data into the evaluation network to obtain a first input quantity of the evaluation network at the current time; obtaining a first weight matrix from the input layer to the hidden layer of the evaluation network, and based on the first input quantity and the first weight matrix, obtaining a first expression result from the input layer to the hidden layer of the evaluation network at the current time; obtaining a second weight matrix from the hidden layer to the output layer of the evaluation network, and based on the first expression result and the second weight matrix, obtaining a second expression result from the hidden layer to the output layer of the evaluation network at the current time; and using the second expression result as the current performance index function of the PID controller at the current time.

[0009] According to the present invention, a proportional-integral-derivative (PID) parameter tuning method for a deaerator system includes an iterative update of the current weights of an evaluation network based on the current performance index function to obtain the optimal performance index function of the PID controller. The method comprises: obtaining a first error function of the evaluation network at the current moment and using the first error function as a first objective function; iteratively updating the current weights of the evaluation network along the gradient descent direction of the first objective function to minimize the current performance index function; stopping the iterative update when the difference between the current weights of the evaluation network reaches a preset weight difference threshold or the current iteration count of the evaluation network reaches a preset iteration count; the current weight difference represents the difference between the current weight of the evaluation network and its corresponding previous weight; and obtaining the optimal performance index function of the PID controller at the current moment based on the iteratively updated evaluation network.

[0010] According to the present invention, a proportional-integral-derivative (PID) parameter tuning method for a deaerator system includes the following steps: iteratively updating the current weights of the action network based on the optimal performance index function to obtain the current control law of the PID controller; obtaining a second error function of the action network at the current moment and using the second error function as a second objective function; iteratively updating the current weights of the action network along the gradient descent direction of the second objective function to minimize the optimal performance index function; stopping the iterative update when the current weight difference of the action network reaches a preset weight difference threshold or the current iteration number of the action network reaches a preset iteration number; the current weight difference represents the difference between the current weight of the action network and its corresponding previous weight; and obtaining the current control law of the PID controller based on the iteratively updated action network.

[0011] According to the present invention, a proportional-integral-derivative (PID) parameter tuning method for a deaerator system includes the following steps: obtaining the current weights of the action network after the completion of a cyclic task as target weights, and obtaining the target PID parameters of the tuned PID controller based on the target weights. The method comprises: obtaining the current weights from the input layer to the hidden layer of the action network after the completion of the cyclic task as a first target weight, and obtaining the current weights from the hidden layer to the output layer of the action network as a second target weight; performing amplitude limiting processing on the first target weight and the second target weight based on a preset excitation function to obtain amplitude-limited first target weight and second target weight; and inputting the amplitude-limited first target weight and second target weight into a neural network to calculate the target PID parameters of the tuned PID controller.

[0012] This invention also provides a proportional-integral-derivative (PID) parameter tuning device based on a deaerator system, comprising: a network construction module, used to construct an action network and an evaluation network based on a neural network and a PID controller corresponding to the deaerator system, wherein the action network is used to simulate the PID controller and the evaluation network is used to evaluate the performance of the PID controller; a loop iteration module, used to obtain the current weights of the action network and the evaluation network, as well as the operating status data of the deaerator system and the output data of the action network, in a loop task, and to iteratively update the current weights of the evaluation network and the action network based on the operating status data and the output data; and a PID tuning module, used to repeatedly execute the above loop task until the current network convergence accuracy of the action network reaches a preset network convergence accuracy, end the loop task, and obtain the current weights of the action network after the loop task ends as target weights, and obtain the target PID parameters of the tuned PID controller based on the target weights.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system as described above.

[0016] The present invention provides a proportional-integral-derivative (PID) parameter tuning method, apparatus, and electronic device for deaerator systems. It simulates a PID controller by constructing an action network based on a neural network and evaluates the performance of the PID controller by constructing an evaluation network. This leverages the self-learning and adaptive characteristics of neural networks to iteratively optimize the action and evaluation networks. Combined with the operating status data of the deaerator system, it can adaptively optimize the action and evaluation networks to address specific operational stability issues in real-world application scenarios. This optimizes the control performance of the PID controller simulated by the action network, thereby controlling the deaerator system to operate stably. This eliminates safety hazards, system overshoot, and delays in boiler feedwater treatment systems caused by operational instability issues in the deaerator system, significantly improving the stability and reliability of the deaerator system during operation. It solves the technical problems of safety hazards, system overshoot, or delays in boiler feedwater treatment systems caused by operational instability issues in existing technologies. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the proportional-integral-derivative (PID) parameter tuning method for a deaerator system provided by the present invention.

[0019] Figure 2 This is the second flowchart of the proportional-integral-derivative PID parameter tuning method based on a deaerator system provided by the present invention;

[0020] Figure 3 This is the third flowchart of the proportional-integral-derivative PID parameter tuning method based on a deaerator system provided by the present invention;

[0021] Figure 4 This is the fourth flowchart of the proportional-integral-derivative PID parameter tuning method based on a deaerator system provided by the present invention;

[0022] Figure 5 This is the fifth flowchart of the proportional-integral-derivative PID parameter tuning method based on a deaerator system provided by the present invention;

[0023] Figure 6 This is the sixth flowchart of the proportional-integral-derivative PID parameter tuning method based on a deaerator system provided by the present invention;

[0024] Figure 7 This is a schematic diagram of the proportional-integral-derivative (PID) parameter tuning device based on a deaerator system provided by the present invention.

[0025] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0026] Figure label:

[0027] 100: Proportional-Integral-Derivative (PID) parameter tuning device based on deaerator system; 10: Network construction module; 20: Cyclic iteration module; 30: PID tuning module; 810: Processor; 820: Communication interface; 830: Memory; 840: Communication bus. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0029] The following is combined Figures 1-6 This invention describes the proportional-integral-derivative (PID) parameter tuning method for a deaerator system.

[0030] like Figure 1 As shown, the present invention provides a proportional-integral-derivative (PID) parameter tuning method based on a deaerator system, including...

[0031] Step S1: Construct an action network and an evaluation network based on the neural network and the PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller.

[0032] Among them, the PID controller is a proportional-integral-differential controller, or simply a PID controller.

[0033] Step S2: In one loop task, obtain the current weights of the action network and the evaluation network, as well as the operating status data of the deaerator system and the output data of the action network, and iteratively update the current weights of the evaluation network and the action network based on the operating status data and the output data.

[0034] The operational status data is used to measure the operational stability of the deaerator system. It should be noted that by iteratively updating the current weights of the evaluation network and action network using the operational status data, the evaluation network and action network corresponding to the PID controller can be optimized to address specific stability issues of the deaerator system. This optimizes the control performance of the PID controller, enabling it to control the deaerator system to operate normally and stably.

[0035] Step S3: Repeat the above loop task until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, end the loop task, and obtain the current weight of the action network after the loop task ends as the target weight. Based on the target weight, obtain the target PID parameters of the tuned PID controller.

[0036] Steps S1 to S4 above involve constructing an action network based on a neural network to simulate a PID controller and constructing an evaluation network to evaluate the performance of the PID controller. This allows for the optimization and iteration of the action network and evaluation network by leveraging the self-learning and adaptive characteristics of neural networks. Combined with the operating status data of the deaerator system, the action network and evaluation network can be adaptively optimized to address specific operational stability issues in actual application scenarios. This aims to optimize the control performance of the PID controller simulated by the action network, thereby controlling the deaerator system to operate in a stable state based on the optimized PID controller. This eliminates the safety hazards, system overshoot, and delays in the boiler feedwater treatment system caused by the operational stability issues of the deaerator system, significantly improving the stability and reliability of the deaerator system during operation. It also solves the technical problems in the prior art where the operational stability issues of the deaerator system lead to safety hazards, system overshoot, or delays in the boiler feedwater treatment system.

[0037] In addition, this invention terminates the loop task when the current network convergence accuracy of the action network reaches the preset network convergence accuracy, that is, stops the optimization when the control performance of the PID controller simulated by the action network is optimized to the best. This can realize the automatic judgment of the performance optimization process of the PID controller without the need for manual intervention by technicians, thereby improving the automation level and ease of operation of the PID controller performance optimization system.

[0038] In one embodiment, before obtaining the current weights of the action network and the evaluation network, the proportional-integral-derivative (PID) parameter tuning method for a deaerator system provided by the present invention further includes: assigning weights to the action network and the evaluation network to obtain the current weights of the action network and the evaluation network after weight assignment. Further, the weight assignment process for the action network and the evaluation network can be based on a random assignment strategy, or other weight assignment strategies can be used; the present invention does not impose specific limitations.

[0039] In one embodiment, such as Figure 2 As shown, step S3 specifically includes steps S31 to S33, wherein...

[0040] Step S31: Input the running status data and output data into the evaluation network to obtain the current performance index function of the PID controller.

[0041] The current performance index function is a function used to evaluate the network performance of the action network, that is, to evaluate the control performance of the PID controller simulated by the action network.

[0042] Step S32: Iteratively update the current weights of the evaluation network based on the current performance index function to obtain the optimal performance index function of the PID controller.

[0043] Step S33: Iteratively update the current weights of the action network based on the optimal performance index function to obtain the current control law of the PID controller, and update the output data of the action network based on the current control law.

[0044] It should be noted that by updating the output data of the action network based on the current control law in the current loop task, iterative updates can be performed in the next loop task based on the updated output data of the current loop task, thereby improving the iterative update effect of the evaluation network and the action network.

[0045] Steps S31 to S33 above provide an accurate iterative direction for the weight iteration update of the evaluation network by using the change of the current performance index function as a reference, thereby improving the weight iteration efficiency and effect of the evaluation network. Similarly, by using the change of the optimal performance index function as a reference, they provide an accurate iterative direction for the weight iteration update of the action network, thereby improving the weight iteration efficiency and effect of the action network.

[0046] In one embodiment, such as Figure 3 As shown, the above step S31 specifically includes steps S311 to S314, wherein...

[0047] Step S311: Input the running status data and output data into the evaluation network to obtain the first input of the evaluation network at the current moment.

[0048] The operational status data includes the first operational status quantity of the deaerator system at the current moment and the second operational status quantity at the previous moment. The output data includes the first action output quantity of the action network at the current moment and the second action output quantity at the previous moment.

[0049] Furthermore, the input function of the evaluation network is obtained, and the running state data and output data are input into the input function of the evaluation network to obtain the first input quantity of the evaluation network at the current time. The input function is shown below.

[0050] x c (k)=[s(k),s(k-1),u(k),u(k-1)]

[0051] Where, x c (k) represents the first input quantity, s(k) represents the first running state quantity, s(k-1) represents the second running state quantity, u(k) represents the first action output quantity, and u(k-1) represents the second action output quantity.

[0052] Step S312: Obtain the first weight matrix from the input layer to the hidden layer of the evaluation network, and based on the first input quantity and the first weight matrix, obtain the first expression result from the input layer to the hidden layer of the evaluation network at the current time.

[0053] Furthermore, the first expression function from the input layer to the hidden layer of the predefined evaluation network is obtained, and the first input quantity and the first weight matrix are substituted into the first expression function to obtain the first expression result from the input layer to the hidden layer of the evaluation network at the current time. The first expression function is as follows.

[0054] F1(k)=tansig(w c1 *x c (k))

[0055] Where F1(k) represents the first expression result, x c (k) represents the first input, w c1 This represents the first weight matrix.

[0056] Step S313: Obtain the second weight matrix from the hidden layer to the output layer of the evaluation network, and based on the first expression result and the second weight matrix, obtain the second expression result from the hidden layer to the output layer of the evaluation network at the current time.

[0057] Furthermore, the second expression function from the hidden layer to the output layer of the predefined evaluation network is obtained, and the first expression result and the second weight matrix are substituted into the second expression function to obtain the second expression result from the hidden layer to the output layer of the evaluation network at the current time. The second expression function is as follows.

[0058]

[0059] in, w represents the second expression result or the current performance metric function. c2 Let F1(k) represent the second weight matrix, and F1(k) represent the first expression result.

[0060] Step S314: Use the second expression result as the current performance index function of the PID controller at the current moment.

[0061] In one embodiment, such as Figure 4 As shown, step S32 specifically includes steps S321 to S324, wherein...

[0062] Step S321: Obtain the first error function of the evaluation network at the current time, and use the first error function as the first objective function.

[0063] Furthermore, the prediction error of the evaluation network at the current moment is obtained, and a first error function for the current moment is constructed based on the prediction error and a preset error function definition. The prediction error of the evaluation network is the difference between the expected output value and the actual output value of the evaluation network. The first error function is shown below.

[0064]

[0065] Among them, E c (t) represents the first error function, e c (t) represents the evaluation network prediction error.

[0066] Step S322: Iteratively update the current weights of the evaluation network along the gradient descent direction of the first objective function to minimize the current performance index function.

[0067] The weight update formula for the evaluation network can be expressed as follows.

[0068]

[0069] in, This represents the weights used to evaluate the network at the current moment. λ represents the weights used to evaluate the network at the next time step, and λ represents the learning rate or step size used to evaluate the network. This represents the partial derivative of the first objective function with respect to the evaluation network weights. This indicates the direction of gradient descent for the first objective function.

[0070] Furthermore, the formula for chain-like partial derivatives of the first objective function with respect to the weights of the evaluation network is shown below.

[0071]

[0072] in, This represents the partial derivative of the first objective function with respect to the evaluation network weights at the current moment, and F1 represents the first expression result. S represents the weights used to evaluate the network at the current moment. cLet represent the activation function for evaluating the network. Further, the activation function for evaluating the network is as follows.

[0073] S c =(1-e t ) / (1+e t )

[0074] Among them, S c Let e ​​represent the activation function of the evaluation network, and let e represent the natural constant. t Let t represent the natural constant raised to the power of t, which makes the activation function smooth and differentiable.

[0075] Step S323: Stop iterative updates when the current weight difference of the evaluation network reaches a preset weight difference threshold or the current iteration number of the evaluation network reaches a preset iteration number; the current weight difference represents the difference between the current weight of the evaluation network and its corresponding previous weight.

[0076] Furthermore, the preset weight difference threshold represents the absolute difference between the current weight of the evaluation network at the current time and the previous weight of the evaluation network at the previous time. The judgment conditions corresponding to the preset weight difference threshold are as follows.

[0077]

[0078] in, This represents the current weights corresponding to the network evaluation at the current moment. ξ represents the previous weight corresponding to the network evaluation at the previous time step, and ξ represents the preset weight difference threshold.

[0079] Step S324: Based on the iteratively updated evaluation network, obtain the optimal performance index function of the PID controller at the current moment.

[0080] Steps S321 to S324 above, by iteratively updating the current weights of the evaluation network along the direction of gradient descent of the first objective function, can complete the update of the evaluation network weights at the fastest speed and minimize the current performance index function, thereby further improving the iterative update speed of the evaluation network.

[0081] In one embodiment, such as Figure 5 As shown, step S33 specifically includes steps S331 to S334, wherein...

[0082] Step S331: Obtain the second error function of the action network at the current time and use the second error function as the second objective function.

[0083] Furthermore, the action network error at the current moment is obtained, and a second error function at the current moment is constructed based on the action network error and the preset error function definition, as shown below.

[0084]

[0085] Among them, E a (t) represents the second error function, e a (t) represents the action network error.

[0086] Step S332: Iteratively update the current weights of the action network along the gradient descent direction of the second objective function to minimize the optimal performance index function.

[0087] The weight update formula for the action network can be expressed as follows.

[0088]

[0089] in, This represents the weights of the action network at the current moment. Let λ represent the weights of the action network at the next time step, and let λ represent the learning rate or step size of the action network. This represents the partial derivative of the second objective function with respect to the action network weights. This indicates the direction of gradient descent for the second objective function.

[0090] Furthermore, the formula for chain-like partial derivatives of the weights of the action network with respect to the second objective function is shown below.

[0091]

[0092] in, J represents the partial derivative of the second objective function with respect to the action network weights at the current moment. * (k) represents the optimal performance index function, K represents the weights from the hidden layer to the output layer of the action network, and F1 represents the first expression result. S represents the weights of the action network at the current moment. a Let represent the activation function of the action network. Further, the activation function of the action network is as follows.

[0093] S a =(1-e t ) / (1+e t )

[0094] Among them, S a Let e ​​represent the activation function of the action network, and let e represent the natural constant. t Let t represent the natural constant raised to the power of t, which makes the activation function smooth and differentiable.

[0095] It should be noted that K also represents the output gain coefficient of the action network. Its selection is guided by the performance index function value of the network output. The performance index function represents the operation of the deaerator system. The faster the performance index function converges, the better the control effect of the PID controller on the deaerator system. The ultimate goal of the PID controller is to select the most suitable output gain coefficient K to achieve the optimal operating state of the entire deaerator system. Therefore, the selection of the K value aims to minimize the derivative of the performance index function, promoting its rapid convergence. Its calculation formula is as follows:

[0096]

[0097] Where J represents the performance index function, K represents the output gain coefficient, F1 represents the first expression result, and S a Let x represent the activation function of the action network. a represents the input to the action network, and u represents the output of the action network or the PID control rate.

[0098] Step S333: Stop iterative update when the current weight difference of the action network reaches the preset weight difference threshold or the current iteration number of the action network reaches the preset iteration number; the current weight difference represents the difference between the current weight of the action network and its corresponding previous weight.

[0099] Furthermore, the weight difference threshold represents the absolute difference between the current weight of the action network at the current moment and the previous weight of the action network at the previous moment. The judgment conditions corresponding to the weight difference threshold are as follows.

[0100]

[0101] in, This represents the current weights of the action network at the current moment. ξ represents the previous weight corresponding to the action network at the previous moment, and ξ represents the preset weight difference threshold.

[0102] Step S334: Based on the iteratively updated action network, obtain the current control law of the PID controller.

[0103] Steps S331 to S334 above, by iteratively updating the current weights of the action network along the direction of gradient descent of the second objective function, can complete the update of the action network weights at the fastest speed and minimize the optimal performance index function, thereby further improving the iterative update speed of the action network.

[0104] In one embodiment, the optimal performance index function can be minimized by iteratively updating the current weights of the action network along the direction of gradient descent of the second objective function. The minimization formula of the optimal performance index function is as follows:

[0105]

[0106] Among them, J * (t) represents the optimal performance index function after minimization, U(t) represents the efficiency function, γ represents the discount factor, and J * (t+Δt) represents the optimal performance index function before minimization. This is achieved by minimizing the PID parameter k. p and k i and k d This minimizes the optimal performance index function.

[0107] Obtain the tracking error of the deaerator system at the current moment, and obtain the utility function at the current moment based on the tracking error and the preset utility function formula, as shown below.

[0108]

[0109] Where U(t) represents the utility function at the current time, j represents the maximum number of iterations in the global loop, and k represents the current iteration number or the current sampling number (equivalent to the current time t). α represents the discount factor, which is used to prevent the utility function from diverging, where 0 < α < 1. x(t) represents the input of the evaluation network at the current time t, i.e. e(t) represents the tracking error of the deaerator system, Q and R are both positive definite matrices, and u(t) represents the PID control rate of the deaerator system at the current time t, which is also the output of the motion network at the current time t.

[0110] Furthermore, the tracking error of the deaerator system is the difference between the preset water level response signal and the actual water level response signal of the deaerator system, as shown below.

[0111] e(t) = r(t) - y(t)

[0112] Where e(t) represents the tracking error of the deaerator system, r(t) represents the preset water level response signal of the deaerator system set at the user end, and y(t) represents the actual water level response signal fed back to the user end by the deaerator system.

[0113] In one embodiment, such as Figure 6 As shown, step S5 above includes steps S51 to S53, wherein...

[0114] Step S51: Based on the action network after the cyclic task ends, obtain the current weights from the input layer to the hidden layer of the action network as the first target weights, and obtain the current weights from the hidden layer to the output layer of the action network as the second target weights.

[0115] Step S52: Based on the preset excitation function, the first target weight and the second target weight are subjected to amplitude limiting processing to obtain the first target weight and the second target weight after amplitude limiting processing.

[0116] The preset activation function is a smooth function that is differentiable everywhere, which is used to prevent the first target weight and the second target weight from being too large, thus causing the weights to not converge.

[0117] Furthermore, the preset excitation function can be in the form shown in the following formula, or it can be other forms of excitation function. This embodiment does not impose specific limitations.

[0118]

[0119] Where e represents the natural constant, W represents the first or second target weight after amplitude limiting. p (x) represents the first objective weight or the second objective weight.

[0120] Step S53: Input the first target weight and the second target weight after the amplitude limiting process into the neural network to calculate the target PID parameters of the tuned PID controller.

[0121] Furthermore, the PID controller can be represented in error form, as follows:

[0122]

[0123] Where, k p and k i and k d Represents the PID parameters, k p k represents the proportional gain of the PID controller. i k represents the integral time of the PID controller. d The derivative time of the PID controller is represented by , and e represents the tracking error of the deaerator system.

[0124] It should be noted that by inputting the first and second target weights after the amplitude limiting process into the neural network, the PID parameters in the form of error can be calculated, which are the target PID parameters of the tuned PID controller. The PID parameters obtained in the form of error are the general form of PID parameters.

[0125] Steps S51 to S53 above involve inputting the first and second target weights after amplitude limiting into a neural network to calculate the target PID parameters of the tuned PID controller. These target PID parameters can then be applied to control the stable operation of the deaerator system, thereby reducing the overshoot or delayed response of the corresponding PID control system, improving the stability and anti-interference capability of the corresponding PID controller, and thus assisting or replacing manual tuning of PID parameters. This reduces labor and time costs, improves the work efficiency and service capabilities of technical personnel, alleviates the problem of insufficient technical manpower and limited technical capabilities in enterprises, and promotes the intelligentization, digitalization, and automation of thermal power plants.

[0126] Two specific embodiments are provided below to further illustrate the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system provided by the present invention.

[0127] In a specific embodiment, the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system provided by the present invention includes the following steps.

[0128] Step 1: Construct an action network and an evaluation network based on the neural network and the corresponding PID controller of the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller.

[0129] Step 2: In one loop task, obtain the current weights of the action network and evaluation network, as well as the operating status data of the deaerator system and the output data of the action network. Input the operating status data and output data into the evaluation network to obtain the current performance index function of the PID controller. Iteratively update the current weights of the evaluation network based on the current performance index function to obtain the optimal performance index function of the PID controller. Iteratively update the current weights of the action network based on the optimal performance index function to obtain the current control law of the PID controller, and update the output data of the action network based on the current control law.

[0130] Step 3: Repeat the above loop task until the current convergence accuracy of the action network reaches the preset convergence accuracy, then end the loop task. Based on the action network after the loop task ends, obtain the current weights from the input layer to the hidden layer of the action network as the first target weights, and obtain the current weights from the hidden layer to the output layer of the action network as the second target weights. Perform amplitude limiting processing on the first target weights and the second target weights based on the preset activation function to obtain the amplitude-limited first target weights and the second target weights. Input the amplitude-limited first target weights and the second target weights into the neural network to calculate the target PID parameters of the tuned PID controller.

[0131] In a specific embodiment two, the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system provided by the present invention includes the following steps.

[0132] Step (1): Construct an action network and an evaluation network based on the neural network and the PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller.

[0133] Step (2): Start the global loop task. In the current global loop, assign weights to the action network and the evaluation network to obtain the current weights of the action network and the evaluation network, and jump to start the local loop task corresponding to the current global loop.

[0134] Step (3): In the current local loop, obtain the operating status data of the deaerator system and the output data of the action network, input the operating status data and output data into the evaluation network to obtain the first input quantity of the evaluation network at the current moment; obtain the first weight matrix from the input layer to the hidden layer of the evaluation network, and based on the first input quantity and the first weight matrix, obtain the first expression result from the input layer to the hidden layer of the evaluation network at the current moment; obtain the second weight matrix from the hidden layer to the output layer of the evaluation network, and based on the first expression result and the second weight matrix, obtain the second expression result from the hidden layer to the output layer of the evaluation network at the current moment; use the second expression result as the current performance index function of the PID controller at the current moment.

[0135] Step (4): Obtain the first error function of the evaluation network at the current moment and use the first error function as the first objective function; iteratively update the current weights of the evaluation network along the gradient descent direction of the first objective function to minimize the current performance index function; stop iterative updating when the current weight difference of the evaluation network reaches the preset weight difference threshold or the current iteration number of the evaluation network reaches the preset iteration number; the current weight difference represents the difference between the current weight of the evaluation network and its corresponding previous weight; based on the iteratively updated evaluation network, obtain the optimal performance index function of the PID controller at the current moment.

[0136] Step (5): Obtain the second error function of the action network at the current time and use the second error function as the second objective function; iteratively update the current weights of the action network along the gradient descent direction of the second objective function to minimize the optimal performance index function; stop iterative updating when the current weight difference of the action network reaches the preset weight difference threshold or the current iteration number of the action network reaches the preset iteration number, where the current weight difference represents the difference between the current weight of the action network and its corresponding previous weight; based on the iteratively updated action network, obtain the current control law of the PID controller and update the output data of the action network based on the current control law.

[0137] Step (6): Exit the current global loop and jump to the next global loop until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, end the global loop task, and based on the action network after the global loop task ends, obtain the current weights from the input layer to the hidden layer of the action network as the first target weights, and obtain the current weights from the hidden layer to the output layer of the action network as the second target weights; perform amplitude limiting processing on the first target weights and the second target weights based on the preset activation function to obtain the amplitude-limited first target weights and the second target weights; input the amplitude-limited first target weights and the second target weights into the neural network to calculate the target PID parameters of the tuned PID controller.

[0138] It should be noted that the global loop is the outer loop, and the local loop is the inner loop. In the current global loop, after assigning weights to the action network and evaluation network, it automatically jumps to the corresponding local loop task to iteratively update the weights of the action network and evaluation network. If the current local loop count reaches the preset local loop count, the local loop task corresponding to the current global loop ends, and it checks whether the current convergence accuracy of the action network has reached the preset network convergence accuracy. If the current convergence accuracy of the action network has not reached the preset network convergence accuracy, it jumps to the next global loop and executes the weight assignment task, and so on, until the current convergence accuracy of the action network reaches the preset network convergence accuracy, at which point the global loop task ends.

[0139] The proportional-integral-derivative (PID) parameter tuning device for a deaerator system provided by the present invention is described below. The PID parameter tuning device for a deaerator system described below can be referred to in correspondence with the PID parameter tuning method for a deaerator system described above.

[0140] like Figure 7As shown, the present invention provides a proportional-integral-derivative (PID) parameter tuning device based on a deaerator system. The PID parameter tuning device 100 based on a deaerator system includes a network construction module 10, a loop iteration module 20, and a PID tuning module 30, wherein...

[0141] The network construction module 10 is used to construct an action network and an evaluation network based on the neural network and the PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller.

[0142] The iterative loop module 20 is used to acquire the current weights of the action network and the evaluation network, as well as the operating status data of the deaerator system and the output data of the action network in a single loop task, and to iteratively update the current weights of the evaluation network and the action network based on the operating status data and the output data.

[0143] The PID tuning module 30 is used to repeatedly execute the above loop task until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, then end the loop task, and obtain the current weight of the action network after the loop task ends as the target weight, and obtain the target PID parameters of the tuned PID controller based on the target weight.

[0144] In one embodiment, the loop iteration module 20 includes a performance function acquisition unit, an evaluation network update unit, and an action network update unit, wherein...

[0145] The performance function acquisition unit is used to input the operating status data and output data into the evaluation network to obtain the current performance index function of the PID controller.

[0146] The evaluation network update unit is used to iteratively update the current weights of the evaluation network based on the current performance index function to obtain the optimal performance index function of the PID controller.

[0147] The action network update unit is used to iteratively update the current weights of the action network based on the optimal performance index function, obtain the current control law of the PID controller, and update the output data of the action network based on the current control law.

[0148] In one embodiment, the performance function acquisition unit includes an input acquisition subunit, a first result acquisition subunit, a second result acquisition subunit, and a performance function acquisition subunit, wherein...

[0149] The input acquisition subunit is used to input the running status data and output data into the evaluation network to obtain the first input of the evaluation network at the current moment.

[0150] The first result acquisition subunit is used to acquire the first weight matrix from the input layer to the hidden layer of the evaluation network, and based on the first input quantity and the first weight matrix, acquire the first expression result from the input layer to the hidden layer of the evaluation network at the current time.

[0151] The second result acquisition subunit is used to acquire the second weight matrix from the hidden layer to the output layer of the evaluation network, and based on the first expression result and the second weight matrix, acquire the second expression result from the hidden layer to the output layer of the evaluation network at the current time.

[0152] The performance function acquisition subunit is used to take the second expression result as the current performance index function of the PID controller at the current moment.

[0153] In one embodiment, the network update evaluation unit includes a first function acquisition subunit, a network update evaluation subunit, a first iterative judgment subunit, and an optimal function acquisition subunit, wherein...

[0154] The first function acquisition sub-unit is used to obtain the first error function of the network at the current time and use the first error function as the first objective function.

[0155] The evaluation network update subunit is used to iteratively update the current weights of the evaluation network along the gradient descent direction of the first objective function, so as to minimize the current performance index function.

[0156] The first iteration judgment subunit is used to stop iterative updates when the current weight difference of the evaluation network reaches a preset weight difference threshold or the current iteration number of the evaluation network reaches a preset iteration number; the current weight difference represents the difference between the current weight of the evaluation network and its corresponding previous weight.

[0157] The optimal function acquisition sub-unit is used to obtain the optimal performance index function of the PID controller at the current moment based on the iteratively updated evaluation network.

[0158] In one embodiment, the action network update unit includes a second function acquisition subunit, an action network update subunit, a second iterative judgment subunit, and a control law acquisition subunit, wherein...

[0159] The second function is a sub-unit used to obtain the second error function of the action network at the current time, and to use the second error function as the second objective function.

[0160] The action network update subunit is used to iteratively update the current weights of the action network along the direction of gradient descent of the second objective function, so as to minimize the optimal performance index function.

[0161] The second iteration judgment subunit is used to stop iterative updates when the current weight difference of the action network reaches a preset weight difference threshold or the current iteration number of the action network reaches a preset iteration number; the current weight difference represents the difference between the current weight of the action network and its corresponding previous weight.

[0162] The control law acquisition subunit is used to obtain the current control law of the PID controller based on the iteratively updated action network.

[0163] In one embodiment, the PID tuning module 30 includes a weight acquisition unit, a limiting processing unit, and a parameter acquisition unit, wherein...

[0164] The weight acquisition unit is used to acquire the current weights from the input layer to the hidden layer of the action network as the first target weights, and to acquire the current weights from the hidden layer to the output layer of the action network as the second target weights, based on the action network after the completion of the loop task.

[0165] A limiting processing unit is used to perform limiting processing on the first target weight and the second target weight based on a preset excitation function, so as to obtain the first target weight and the second target weight after limiting processing.

[0166] The parameter acquisition unit is used to input the first target weight and the second target weight after the amplitude limiting process into the neural network to calculate the target PID parameters of the tuned PID controller.

[0167] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logic instructions in the memory 830 to execute a proportional-integral-derivative (PID) parameter tuning method based on the deaerator system. This method includes: constructing an action network and an evaluation network based on a neural network and the corresponding PID controller of the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller. In one loop task, the current weights of the action network and the evaluation network are obtained, as well as the operating status data of the deaerator system and the output data of the action network. The current weights of the evaluation network and the action network are iteratively updated based on the operating status data and the output data. The above loop task is repeated until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, the loop task ends, and the current weights of the action network after the loop task ends are obtained as target weights. The target PID parameters of the tuned PID controller are obtained based on the target weights.

[0168] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0169] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the proportional-integral-derivative (PID) parameter tuning method based on the deaerator system provided by the above methods. The method includes: constructing an action network and an evaluation network based on a neural network and a PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller. In one loop task, the current weights of the action network and the evaluation network are obtained, as well as the operating status data of the deaerator system and the output data of the action network are obtained, and the current weights of the evaluation network and the action network are iteratively updated based on the operating status data and the output data. The above loop task is repeated until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, the loop task is terminated, and the current weights of the action network after the loop task is terminated are obtained as the target weights. The target PID parameters of the tuned PID controller are obtained based on the target weights.

[0170] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a proportional-integral-derivative (PID) parameter tuning method for a deaerator system provided by the methods described above. The method includes: constructing an action network and an evaluation network based on a neural network and a PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller. In one loop task, the current weights of the action network and the evaluation network are obtained, as well as the operating status data of the deaerator system and the output data of the action network are obtained. The current weights of the evaluation network and the action network are iteratively updated based on the operating status data and the output data. The loop task is repeated until the current network convergence accuracy of the action network reaches a preset network convergence accuracy, the loop task is terminated, and the current weights of the action network after the loop task is terminated are obtained as target weights. The target PID parameters of the tuned PID controller are obtained based on the target weights.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0172] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for tuning proportional-integral-derivative (PID) parameters based on a deaerator system, characterized in that, include: An action network and an evaluation network are constructed based on a neural network and a PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller. In one cyclic task, the current weights of the action network and the evaluation network are obtained, as well as the operating status data of the deaerator system and the output data of the action network are obtained, and the current weights of the evaluation network and the action network are iteratively updated based on the operating status data and the output data. Repeat the above loop task until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, then end the loop task, and obtain the current weight of the action network after the loop task ends as the target weight. Based on the target weight, obtain the target PID parameters of the PID controller after tuning. The iterative update of the current weights of the evaluation network and the action network based on the operating status data and the output data includes: The operating status data and the output data are input into the evaluation network to obtain the current performance index function of the PID controller; The current weights of the evaluation network are iteratively updated based on the current performance index function to obtain the optimal performance index function of the PID controller. The current weights of the action network are iteratively updated based on the optimal performance index function to obtain the current control law of the PID controller, and the output data of the action network is updated based on the current control law. The step of obtaining the current weights of the action network after the completion of the cyclic task as the target weights, and obtaining the target PID parameters of the tuned PID controller based on the target weights, includes: Based on the action network after the cyclic task ends, the current weights from the input layer to the hidden layer of the action network are obtained as the first target weights, and the current weights from the hidden layer to the output layer of the action network are obtained as the second target weights. Based on a preset activation function, the first target weight and the second target weight are subjected to amplitude limiting processing to obtain the first target weight and the second target weight after amplitude limiting processing; The first and second target weights after the amplitude limiting process are input into the neural network to calculate the target PID parameters of the PID controller after tuning. The calculated and tuned PID controller is determined by the following expression, with error as the variable: Where, k p and k i and k d Represents the PID parameters, k p k represents the proportional gain of the PID controller. i k represents the integral time of the PID controller. d δ represents the derivative time of the PID controller. e This indicates the tracking error of the deaerator system.

2. The proportional-integral-derivative (PID) parameter tuning method based on a deaerator system according to claim 1, characterized in that, The step of inputting the operating status data and the output data into the evaluation network to obtain the current performance index function of the PID controller includes: The operating status data and the output data are input into the evaluation network to obtain the first input quantity of the evaluation network at the current moment; Obtain the first weight matrix from the input layer to the hidden layer of the evaluation network, and based on the first input quantity and the first weight matrix, obtain the first expression result from the input layer to the hidden layer of the evaluation network at the current time. Obtain the second weight matrix from the hidden layer to the output layer of the evaluation network, and based on the first expression result and the second weight matrix, obtain the second expression result from the hidden layer to the output layer of the evaluation network at the current time. The second expression result is used as the current performance index function of the PID controller at the current moment.

3. The proportional-integral-derivative (PID) parameter tuning method based on a deaerator system according to claim 1, characterized in that, The step of iteratively updating the current weights of the evaluation network based on the current performance index function to obtain the optimal performance index function of the PID controller includes: Obtain the first error function of the evaluation network at the current moment, and use the first error function as the first objective function; The current weights of the evaluation network are iteratively updated along the direction of gradient descent of the first objective function to minimize the current performance index function; If the current weight difference of the evaluation network reaches a preset weight difference threshold or the current iteration number of the evaluation network reaches a preset iteration number, the iteration update shall be stopped; the current weight difference represents the difference between the current weight of the evaluation network and its corresponding previous weight. Based on the iteratively updated evaluation network, the optimal performance index function of the PID controller at the current moment is obtained.

4. The proportional-integral-derivative (PID) parameter tuning method based on a deaerator system according to claim 1, characterized in that, The iterative update of the current weights of the action network based on the optimal performance index function to obtain the current control law of the PID controller includes: Obtain the second error function of the action network at the current moment, and use the second error function as the second objective function; The current weights of the action network are iteratively updated along the direction of gradient descent of the second objective function to minimize the optimal performance index function; If the current weight difference of the action network reaches a preset weight difference threshold or the current iteration number of the action network reaches a preset iteration number, the iteration update stops; the current weight difference represents the difference between the current weight of the action network and its corresponding previous weight. Based on the iteratively updated action network, the current control law of the PID controller is obtained.

5. A proportional-integral-derivative (PID) parameter tuning device based on a deaerator system, characterized in that, include: A network construction module is used to construct an action network and an evaluation network based on a neural network and a PID controller corresponding to the deaerator system. The action network is used to simulate the PID controller, and the evaluation network is used to evaluate the performance of the PID controller. The iterative loop module is used to obtain the current weights of the action network and the evaluation network, as well as the operating status data of the deaerator system and the output data of the action network in a single loop task, and to iteratively update the current weights of the evaluation network and the action network based on the operating status data and the output data. The PID tuning module is used to repeatedly execute the above loop task until the current network convergence accuracy of the action network reaches the preset network convergence accuracy, then end the loop task, and obtain the current weight of the action network after the loop task ends as the target weight, and obtain the target PID parameters of the tuned PID controller based on the target weight. The iterative update of the current weights of the evaluation network and the action network based on the operating status data and the output data includes: The operating status data and the output data are input into the evaluation network to obtain the current performance index function of the PID controller; The current weights of the evaluation network are iteratively updated based on the current performance index function to obtain the optimal performance index function of the PID controller. The current weights of the action network are iteratively updated based on the optimal performance index function to obtain the current control law of the PID controller, and the output data of the action network is updated based on the current control law. The PID tuning module is specifically used for: Based on the action network after the cyclic task ends, the current weights from the input layer to the hidden layer of the action network are obtained as the first target weights, and the current weights from the hidden layer to the output layer of the action network are obtained as the second target weights. Based on a preset activation function, the first target weight and the second target weight are subjected to amplitude limiting processing to obtain the first target weight and the second target weight after amplitude limiting processing; The first and second target weights after the amplitude limiting process are input into the neural network to calculate the target PID parameters of the PID controller after tuning. The calculated and tuned PID controller is determined by the following expression, with error as the variable: Where, k p and k i and k d Represents the PID parameters, k p k represents the proportional gain of the PID controller. i k represents the integral time of the PID controller. d δ represents the derivative time of the PID controller. e This indicates the tracking error of the deaerator system.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the proportional-integral-derivative PID parameter tuning method based on a deaerator system as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the proportional-integral-derivative (PID) parameter tuning method based on a deaerator system as described in any one of claims 1 to 4.

Images