Thermoelectric coordinated heat supply pipeline safe operation optimization system and method

Abstract

The invention provides a thermoelectric coordinated heat supply pipeline safe operation optimization system and method. The technical purpose is to achieve self-adaption of the heat supply steam extraction flow of the unit within the full flow range and improve the heat supply flow adjusting precision. A thermoelectric coordination controller is added in a thermal load control loop, and user electrical load deviation and differences between pressure set values and actual values at a CV / AGV valve and an EV / AEV valve serve as input and output valve opening signals of the thermoelectric coordination controller. When a heat load increasing instruction is given, the thermoelectric coordination controller starts to act, and the opening degrees of the CV valve, the AGV valve, the EV valve and the AEV valve are changed in a coordinated mode to rapidly meet the requirement. Parameters of the thermoelectric coordination controller are optimized through the genetic algorithm and the BP neural network, and a good effect is achieved.

Description

Technical Field

[0001] The present application relates to the field of new energy technology, and in particular to a system and method for optimizing the safe operation of a heat-electricity coordinated heating pipeline. Background Art

[0002] To achieve the "dual carbon" goals, it is imperative to reduce fossil fuel use and improve energy efficiency. Therefore, large condensing steam turbine units are being retrofitted for heating, transforming them into combined heat and power (CHP) units. To assist with electrical load regulation, extraction steam flow also varies when the load fluctuates. Due to the need for deep peak shaving, the load fluctuates over a wide range, necessitating full-range control of extraction steam flow. When the extraction steam flow is low, the throttling butterfly valves in the heating piping system have a very small opening and high sensitivity. When heat user demand fluctuates within a small range, the butterfly valves struggle to accurately adjust the flow in small increments, failing to meet operating conditions. Therefore, a control method for precise full-range flow regulation is needed to meet regulation requirements under low flow conditions. In heating piping systems, CV valves typically control the intermediate discharge pressure, while EV valves control the extraction steam flow. These two valves work together to regulate heating supply. The CV valves flexibly distribute steam flow, facilitate heat and power coordination, protect the low-pressure cylinders, and maintain stable main steam pressure. The EV valves on the heating extraction steam main directly control the extraction steam flow, minimizing time delay and improving safety. When the heating extraction steam flow is high, the steam flow entering the low-pressure cylinder is low, making the CV valve inaccurate. The AGV valve on the auxiliary connecting pipe is used instead. When the heating extraction steam flow is low, the flow in the extraction main pipe is low, making the EV valve inaccurate. The AEV valve on the auxiliary extraction pipe is used instead. To coordinate the opening of these four valves, this application proposes a system and method for optimizing the safe operation of a heating pipeline with coordinated heat and power.

[0003] Genetic Algorithm (GA) is a global optimization method that simulates the natural evolution process and is suitable for solving complex, nonlinear, multi-peak and discrete problems. It maintains population diversity through mechanisms such as selection, crossover and mutation, so that it has strong global search capabilities and is not easily trapped in local optimality. Genetic algorithms do not rely on gradient information, have strong versatility and scalability, and are applicable to various optimization problems. Based on the high-quality initial point found by GA, the BP neural network uses the BP algorithm to further reduce the error and improve the control accuracy. Therefore, this application uses genetic algorithms and BP neural networks to optimize the new heating pipeline. Summary of the Invention

[0004] This application proposes a heat and power coordinated heating pipeline safe operation optimization system and method. Its technical purpose is to achieve self-adaptation of the unit's heating extraction steam flow under different flow conditions, especially low flow conditions, and improve the accuracy of heating flow regulation.

[0005] The specific plan is as follows:

[0006] A heat-electricity coordinated heating pipeline safe operation optimization system utilizes a boiler-turbine coordinated control strategy. The main steam pressure setpoint and unit load setpoint are input into the boiler and turbine master controllers, respectively, to achieve variable load control. Specifically, a heat-electricity coordinated controller is incorporated into the heat load control loop to output opening signals for the CV, AGV, EV, and AEV valves, coordinating the heat and electrical loads. After the heating system retrofit, the heating supply needs to be regulated. In practice, since the heating supply cannot be directly measured, it is regulated by decomposing the heating supply into the extraction steam enthalpy and the extraction steam flow rate. The EV valve directly acts on the extraction steam flow rate, throttling it and thereby regulating the heating supply. The CV valve regulates the intermediate exhaust pressure, essentially controlling the flow rate. When the heating extraction steam main flow rate is very low, the AEV valve replaces the EV valve; when the flow rate in the intermediate and low-pressure connecting pipes is very low, the AGV valve replaces the CV valve.

[0007] Furthermore, when the total load of the power plant remains unchanged, the electrical load will also fluctuate when the thermal network's thermal energy consumption fluctuates. The user's electrical load deviation and the difference between the set and actual pressure values ​​at the CV / AGV valves and EV / AEV valves serve as inputs to the thermal power coordination controller, which outputs valve opening signals. The parameters of the thermal power coordination controller are optimized using the genetic algorithm and BP neural network described later. Its operating process is as follows: when the thermal load changes, the thermal power coordination controller receives the difference between the set and actual pressure values ​​at the CV / AGV valves and EV / AEV valves. Changes in thermal load will also cause changes in the electrical load. The electrical load deviation is also introduced into the thermal power coordination controller for auxiliary regulation. After real-time optimization, it outputs four valve opening signals to control the extraction steam flow to track the thermal load.

[0008] Among them, the thermal load control loop includes a connected heating steam extraction main pipe and a medium and low pressure connecting pipe, an EV valve is installed on the heating steam extraction main pipe, the two ends of the medium and low pressure connecting pipe are respectively connected to the medium pressure cylinder and the low pressure cylinder, and a CV valve is installed on the medium and low pressure connecting pipe, the EV valve and the CV valve jointly regulate the heating steam extraction flow of the thermal load control loop; an auxiliary steam extraction pipe as a bypass is connected in parallel to the side of the heating steam extraction main pipe, and an AEV valve for controlling the bypass flow is installed on the auxiliary steam extraction pipe, and an auxiliary connecting pipe as a bypass is connected in parallel to the side of the medium and low pressure connecting pipe, and an AGV valve for controlling the middle row pressure is installed on the auxiliary connecting pipe, which is used to indirectly control the flow entering the low pressure cylinder, and the two ends of the auxiliary connecting pipe are respectively connected to the medium pressure cylinder and the low pressure cylinder.

[0009] When the heat user's consumption is very small, the heating butterfly valve CV opens widely, and the auxiliary connecting pipe stop valve AGV is fully closed; the heating butterfly valve EV is fully closed, the heating extraction steam is diverted to the bypass auxiliary extraction pipe, and the auxiliary extraction pipe stop valve AEV is almost fully closed. When the heat user's consumption is large enough, the auxiliary extraction pipe stop valve AEV is fully closed, and the heating butterfly valve EV is activated; at this time, the flow entering the low-pressure cylinder is very small, and the heating butterfly valve CV is fully closed. The flow is adjusted by the auxiliary connecting pipe stop valve AGV on the auxiliary connecting pipe, which is almost fully closed. When the heat user's consumption fluctuates in the middle range, the steam flow entering the low-pressure cylinder is not very small, and the extraction steam flow is not very small either; at this time, the AGV valve and the AEV valve are fully closed, and the heating extraction steam flow is controlled by the heating butterfly valve CV and the heating butterfly valve EV. In the above process, the opening of the CV valve, AGV valve, EV valve, and AEV valve are all controlled by the heat and power coordination controller to form a coordinated system.

[0010] To some extent, thermal load and electrical load conflict. To ensure the thermal load requirements of heat users, CHP units adopt a heat-electricity coordination model. Electricity load deviation signals are also incorporated into thermal load regulation, serving only as auxiliary adjustments and constraints to ensure unit stability and prevent overloads.

[0011] Furthermore, this application uses genetic algorithms and BP neural networks to control and optimize the heating pipeline system:

[0012] For the genetic algorithm encoding, this application uses real number encoding, which is suitable for continuous variables. The output is the opening of the CV valve, AGV valve, EV valve, and AEV valve, i.e., the individuals in the genetic algorithm. xi∈[0,1], the opening vector x=(x1,x2,x3,x4), where x1 is the CV valve opening, x2 is the AGV valve opening, x3 is the EV valve opening, and x4 is the AEV valve opening.

[0013] For the fitness function of the genetic algorithm, two indicators need to be examined: overshoot M P and adjustment time t s , that is, the instantaneous peak value of the heat load must not exceed 10% of the steady-state value, and the heat load must stabilize to within ±2% of the steady-state value within 15 seconds. The constraints set in this application allow for minor violations to avoid premature convergence of the algorithm, so a gradual penalty is used. The penalty function β is:

[0014] β=k1×[max(0,M P -0.1)] 2 +k2×[max(0,t s -15)] 2

[0015] k1 and k2 are penalty coefficients, which are adjusted dynamically. They are set to 50 for the first 50 generations to encourage the genetic algorithm to explore extensively and find the global optimal solution as much as possible; after that, the penalty coefficients are set to 80 to ensure rapid convergence. The total fitness is the sum of the original performance value γ and the penalty function value β. The original performance value is the weighted sum of the overshoot and the adjustment time, which have different dimensions. Therefore, normalization adjustment must be performed first, namely:

[0016]

[0017] γ=0.4×a+0.6×b

[0018] f=γ+β

[0019] This is a minimization problem, and the smaller the fitness f, the better.

[0020] The beneficial effects of this application are:

[0021] (1) A heat-electric coordination controller is added to the heat load control loop. The difference between the pressure set value and the actual value at the CV / AGV valve and EV / AEV valve, and the user's electric load deviation are used as inputs of the heat-electric coordination controller, which outputs the opening signal of the CV valve and EV valve to coordinate the heat load and the electric load.

[0022] (2) The thermoelectric coordination controller enables the CV valve, AGV valve, EV valve, and AEV valve to cooperate and control in a coordinated manner, optimizing the valve opening in real time to form a whole;

[0023] (3) The genetic algorithm is used to optimize the valve opening adjustment, which does not rely on gradient information, has a strong global search capability, and is not prone to falling into local optimality;

[0024] (4) After the genetic algorithm is optimized, a differentiable proxy system is obtained, which is then further optimized using the BP neural network. Parameter A is obtained after the genetic algorithm is optimized, and parameter B is obtained after the BP neural network is optimized. After comparing the fitness of the two, a trade-off is made to avoid fitness degradation in certain situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is the control logic diagram used in this application.

[0026] Figure 2 This is the basic flow chart of the optimization method used in this application.

[0027] Figure 3 This is a framework diagram of the heat load control loop in this application. DETAILED DESCRIPTION

[0028] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0029] like Figure 1 As shown, the present invention provides a heat-electric coordinated heating pipeline safe operation optimization system, adds a heat-electric coordination controller to the heat load control loop, outputs the opening signals of the CV valve, AGV valve, EV valve, and AEV valve, and coordinates the heat load and the electric load; wherein, the difference between the pressure setting value and the actual value at the CV / AGV valve and the EV / AEV valve, and the user's electric load deviation are all used as inputs of the heat-electric coordination controller, and the output is the valve opening signal, and the parameters of the heat-electric coordination controller are optimized by genetic algorithm and BP neural network; its working process is that when the heat load changes, the heat-electric coordination controller receives the difference signal between the pressure setting value and the actual value at the CV / AGV valve and the EV / AEV valve, and the change in heat load will cause the electric load to change as well. The electric load deviation is also introduced into the heat-electric coordination controller for auxiliary regulation, and after real-time optimization, the four valve opening signals are output to control the steam extraction flow to track the heat load.

[0030] In this embodiment, the specific principle of the system is as follows: when the thermal load changes, the heat and power coordination controller receives the difference signal between the set value and the actual value of the pressure at the CV / AGV valve and the EV / AEV valve. The change in thermal load will also cause the electrical load to change. The electrical load deviation is also introduced into the heat and power coordination controller for auxiliary regulation. After real-time optimization, it outputs four valve opening signals to control the extraction steam flow to track the thermal load. e After the change occurs (assuming N e The difference between N0 and N1 changes, ΔN increases, and the difference signal is sent to the PID controller to adjust the combustion rate μ B Increase, compensate for part of the electrical load, so that N e Keep up with N0; ΔN is also limited and constrained to be different from P0, pulling down the set value. And if N0 changes, after PD is accelerated and strengthened, it can speed up μ B Adjustment and increase the adjustment action. B Increase leads to P T Increases, ΔP increases. Due to the aforementioned effect of lowering the set value, ΔP increases more. After the PI controller acts, the main steam valve opening μ T Increase, so that P T Decrease and track P0. The above process can achieve the purpose of thermal and electrical coordination.

[0031] like Figure 3As shown, the thermal load control loop includes a connected heating extraction steam main pipe and a medium and low pressure connecting pipe, wherein an EV valve is installed on the heating extraction steam main pipe, and the two ends of the medium and low pressure connecting pipes are respectively connected to the medium pressure cylinder and the low pressure cylinder, and a CV valve is installed on the medium and low pressure connecting pipes, and the EV valve and the CV valve jointly regulate the heating extraction steam flow of the thermal load control loop; an auxiliary steam extraction pipe as a bypass is connected in parallel to the side of the heating extraction steam main pipe, and an AEV valve for controlling the bypass flow is installed on the auxiliary steam extraction pipe, and an auxiliary connecting pipe as a bypass is connected in parallel to the side of the medium and low pressure connecting pipe, and an AGV valve for controlling the middle row pressure is installed on the auxiliary connecting pipe, which is used to indirectly control the flow entering the low pressure cylinder, and the two ends of the auxiliary connecting pipe are respectively connected to the medium pressure cylinder and the low pressure cylinder.

[0032] When the heat user's consumption is very small, the CV valve is opened widely and the AGV valve is fully closed; the EV valve is fully closed, the heating extraction steam is diverted to the bypass auxiliary extraction pipe, and the AEV valve is almost fully closed; when the heat user's consumption is large enough, the AEV valve is fully closed and the EV valve is actuated; at this time, the flow entering the low-pressure cylinder is very small, the CV valve is fully closed, and the flow is adjusted by the AGV valve on the auxiliary connecting pipe, which is almost fully closed; when the heat user's consumption fluctuates in the middle range, the steam flow entering the low-pressure cylinder is not very small, and the extraction steam flow is not very small either; at this time, the AGV valve and the AEV valve are fully closed, and the heating extraction steam flow is controlled by the CV valve and the EV valve; in the above process, the opening of the CV valve, AGV valve, EV valve and AEV valve are all controlled by the thermal power coordination controller to form a coordination.

[0033] This application adopts a thermal-electric coordination model, and adds the electric load deviation as feedback to the thermal load regulation, which only serves as an auxiliary regulation and constraint to ensure the stability of the unit and prevent over-limit.

[0034] In this embodiment, the encoding of the genetic algorithm uses real number encoding, which is suitable for continuous variables; the output is the opening of the CV valve, AGV valve, EV valve, and AEV valve, that is, the individuals in the genetic algorithm; xi∈[0,1], the opening vector x=(x1,x2,x3,x4), x1 is the CV valve opening, x2 is the AGV valve opening, x3 is the EV valve opening, and x4 is the AEV valve opening;

[0035] The fitness function of the genetic algorithm examines two indicators: overshoot M P and adjustment time t s , that is, the instantaneous peak value of the heat load shall not exceed 10% of the steady-state value, and the heat load shall stabilize to within ±2% of the steady-state value within 15 seconds, and a progressive penalty shall be adopted; the penalty function β is:

[0036] β=k1×[max(0,M P -0.1)] 2 +k2×[max(0, t s -15)] 2

[0037] Among them, k1 and k2 are penalty coefficients, which are adjusted dynamically;

[0038] In the first 50 generations, it is set to 50 to encourage the genetic algorithm to explore widely and find the global optimal solution as much as possible; and the penalty coefficient is set to 80 thereafter to make the algorithm converge as quickly as possible; the total fitness f is the sum of the original performance value γ and the penalty function value β; the original performance value is the weighted sum of the overshoot and the adjustment time. Since the two have different dimensions, they are first normalized, that is:

[0039]

[0040] γ=0.4×a+0.6×b

[0041] f=γ+β

[0042] This is a minimization problem, and the smaller the fitness f, the better.

[0043] like Figure 2 As shown, the present invention also provides a method for optimizing safe operation of a heat supply pipeline with heat and electricity coordination, which is characterized by comprising:

[0044] S1, encoding: using real number encoding, the output is the opening of the CV valve, AGV valve, EV valve, and AEV valve, that is, the individuals in the genetic algorithm; xi∈[0,1], the opening vector x=(x1,x2,x3,x4), x1 is the CV valve opening, x2 is the AGV valve opening, x3 is the EV valve opening, and x4 is the AEV valve opening;

[0045] S2, initialization population: randomly generate 100 opening vectors as the initial population;

[0046] S3, fitness evaluation: the penalty function is

[0047] β=k1×[max(0,M P -0.1)] 2 +k2×[max(0,t s -15)] 2

[0048] Among them, k1 and k2 are penalty coefficients, which are adjusted dynamically. They are both set to 50 in the first 50 generations to encourage the genetic algorithm to explore widely and find the global optimal solution as much as possible. After that, the penalty coefficient is set to 80 to make the algorithm converge as quickly as possible. The total fitness f is the sum of the original performance value γ and the penalty function value β. The original performance value is the weighted sum of the overshoot and the adjustment time. Since the two have different dimensions, they are first normalized, that is:

[0049]

[0050] γ=0.4×a+0.6×b

[0051] f=γ+β

[0052] This is a minimization problem, and the smaller the fitness, the better;

[0053] S4, selection operation: adopt the selection strategy of "roulette wheel selection", that is, the probability is proportional to the fitness;

[0054] S5, Crossover operation: arithmetic crossover is used; the offspring is generated by the weighted average of the two parents, x′ = α·x1 + (1-α)·x2, where α is the crossover factor, which is randomly generated by the system; the crossover rate is selected as 0.8 to encourage crossover between individuals;

[0055] S6, mutation operation: Use real number perturbation, that is, add a random small value to the original value; if the magnitude of the opening value is 0.1, the magnitude of the small value is 0.001, and it is normally distributed, while avoiding the value exceeding 1; the mutation probability is set to 0.05 to increase diversity and avoid falling into local optimality;

[0056] S7, generate a new generation: select the elite strategy, that is, retain the best individuals of the previous generation to prevent the loss of high-quality solutions;

[0057] S8, determine the termination condition: if the fitness of the optimal individual reaches the expected target, then output the optimal individual, otherwise return to step S3, evaluate the fitness, and continue iteration;

[0058] S9, taking the optimal individual outputted in the genetic algorithm optimization phase GA, i.e., steps S1-S8, as the initial parameter of the BP neural network, denoted as parameter A;

[0059] S10, parameter A is used as the initial parameter, MSE is selected as the loss function, the learning rate is 0.001, Adam is selected as the optimizer, which is stable and converges quickly, and the number of training rounds is 100;

[0060] S11. After 100 rounds of training, parameter B is obtained; after evaluating the fitness of parameter A and parameter B respectively, they are compared; if the fitness of parameter A is less than that of parameter B, parameter A is retained, otherwise parameter B is retained.

[0061] The above is a detailed introduction to the method and specific implementation of the present invention. Of course, in addition to the above examples, the present invention can also have other implementations, and any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection of the present invention.

Claims

1. A heat and power coordinated heating pipeline safe operation optimization system, characterized by: A thermoelectric coordination controller is added to the thermal load control loop to output the opening signals of the CV valve, AGV valve, EV valve, and AEV valve to coordinate the thermal load and the electrical load. The difference between the pressure set value and the actual value at the CV / AGV valve and EV / AEV valve, as well as the user's electrical load deviation, are used as inputs to the thermoelectric coordination controller, and the output is the valve opening signal. The parameters of the thermoelectric coordination controller are optimized by genetic algorithms and BP neural networks. The working process is as follows: when the thermal load changes, the thermoelectric coordination controller receives the difference signal between the pressure set value and the actual value at the CV / AGV valve and EV / AEV valve. The change in thermal load will cause the electrical load to change as well. The electrical load deviation is also introduced into the thermoelectric coordination controller for auxiliary regulation. After real-time optimization, the four valve opening signals are output to control the extraction steam flow to track the thermal load.

2. A heat and power coordinated heating pipeline safe operation optimization system according to claim 1, characterized in that: The thermal load control loop includes a connected heating steam extraction main pipe and a medium- and low-pressure connecting pipe, wherein an EV valve is installed on the heating steam extraction main pipe, and the two ends of the medium- and low-pressure connecting pipes are respectively connected to the medium-pressure cylinder and the low-pressure cylinder, and a CV valve is installed on the medium- and low-pressure connecting pipes, and the EV valve and the CV valve jointly regulate the heating steam extraction flow of the thermal load control loop; an auxiliary steam extraction pipe serving as a bypass is connected in parallel to the side of the heating steam extraction main pipe, and an AEV valve for controlling the bypass flow is installed on the auxiliary steam extraction pipe, and an auxiliary connecting pipe serving as a bypass is connected in parallel to the side of the medium- and low-pressure connecting pipe, and an AGV valve for controlling the middle exhaust pressure is installed on the auxiliary connecting pipe, which is used to indirectly control the flow entering the low-pressure cylinder, and the two ends of the auxiliary connecting pipe are respectively connected to the medium- and low-pressure cylinder.

3. A heat and power coordinated heating pipeline safe operation optimization system according to claim 2, characterized in that: When the heat user's consumption is very small, the CV valve is opened widely and the AGV valve is fully closed; the EV valve is fully closed, the heating extraction steam is diverted to the bypass auxiliary extraction pipe, and the AEV valve is almost fully closed; when the heat user's consumption is large enough, the AEV valve is fully closed and the EV valve is actuated; at this time, the flow entering the low-pressure cylinder is very small, the CV valve is fully closed, and the flow is adjusted by the AGV valve on the auxiliary connecting pipe, which is almost fully closed; when the heat user's consumption fluctuates in the middle range, the steam flow entering the low-pressure cylinder is not very small, and the extraction steam flow is not very small either; at this time, the AGV valve and the AEV valve are fully closed, and the heating extraction steam flow is controlled by the CV valve and the EV valve; in the above process, the opening of the CV valve, AGV valve, EV valve and AEV valve are all controlled by the thermal power coordination controller to form a coordination.

4. The heat and power coordinated heating pipeline safe operation optimization system according to claim 1 is characterized in that: By adopting the thermal-electric coordination mode, the electric load deviation signal is added to the thermal load regulation, which only serves as an auxiliary regulation and constraint to ensure the stability of the unit and prevent over-limit.

5. The heat and power coordinated heating pipeline safe operation optimization system according to claim 2 is characterized in that: The encoding of the genetic algorithm adopts real number encoding, which is suitable for continuous variables; the output is the opening of the CV valve, AGV valve, EV valve, and AEV valve, that is, the individuals in the genetic algorithm; xi∈[0,1], the opening vector x=(x1,x2,x3,x4), x1 is the CV valve opening, x2 is the AGV valve opening, x3 is the EV valve opening, and x4 is the AEV valve opening; The fitness function of the genetic algorithm examines two indicators: overshoot M P and adjustment time t s , that is, the instantaneous peak value of the heat load shall not exceed 10% of the steady-state value, and the heat load shall stabilize to within ±2% of the steady-state value within 15 seconds, and a progressive penalty shall be adopted; the penalty function β is: β=k1×[max(0,M P -0.1)] 2 +k2×[max(0,t s -15)] 2 Among them, k1 and k2 are penalty coefficients, which are adjusted dynamically; In the first 50 generations, it is set to 50 to encourage the genetic algorithm to explore widely and find the global optimal solution as much as possible; and the penalty coefficient is set to 80 thereafter to make the algorithm converge as quickly as possible; the total fitness f is the sum of the original performance value γ and the penalty function value β; the original performance value is the weighted sum of the overshoot and the adjustment time. Since the two have different dimensions, they are first normalized, that is: γ=0.4×a+0.6×b f=γ+β This is a minimization problem, and the smaller the fitness f, the better.

6. The method for optimizing safe operation of a heat supply pipeline with heat and power coordination according to claim 1, characterized in that: A system according to any one of claims 1 to 5, comprising: S1, encoding: using real number encoding, the output is the opening of the CV valve, AGV valve, EV valve, and AEV valve, that is, the individuals in the genetic algorithm; xi∈[0,1], the opening vector x=(x1,x2,x3,x4), x1 is the CV valve opening, x2 is the AGV valve opening, x3 is the EV valve opening, and x4 is the AEV valve opening; S2, initialization population: randomly generate 100 opening vectors as the initial population; S3, fitness evaluation: the penalty function is β=k1×[max(0,M P -0.1)] 2 +k2×[max(0,t s -15)] 2 Among them, k1 and k2 are penalty coefficients, which are adjusted dynamically. They are both set to 50 in the first 50 generations to encourage the genetic algorithm to explore widely and find the global optimal solution as much as possible. After that, the penalty coefficient is set to 80 to make the algorithm converge as quickly as possible. The total fitness f is the sum of the original performance value γ and the penalty function value β. The original performance value is the weighted sum of the overshoot and the adjustment time. Since the two have different dimensions, they are first normalized, that is: γ=0.4×a+0.6×b f=γ+β This is a minimization problem, and the smaller the fitness, the better; S4, selection operation: adopt the "roulette wheel selection" selection strategy, that is, the probability is proportional to the fitness; S5, Crossover operation: arithmetic crossover is used; the offspring is generated by the weighted average of the two parents, x′ = α·x1 + (1-α)·x2, where α is the crossover factor, which is randomly generated by the system; the crossover rate is selected as 0.8 to encourage crossover between individuals; S6, mutation operation: Use real number perturbation, that is, add a random small value to the original value; if the magnitude of the opening value is 0.1, the magnitude of the small value is 0.001, and it is normally distributed, while avoiding the value exceeding 1; the mutation probability is set to 0.05 to increase diversity and avoid falling into local optimality; S7, generate a new generation: select the elite strategy, that is, retain the best individuals of the previous generation to prevent the loss of high-quality solutions; S8, determine the termination condition: if the fitness of the optimal individual reaches the expected target, then output the optimal individual, otherwise return to step S3, evaluate the fitness, and continue iteration; S9, taking the optimal individual outputted in the genetic algorithm optimization phase GA, i.e., steps S1-S8, as the initial parameter of the BP neural network, denoted as parameter A; S10, parameter A is used as the initial parameter, MSE is selected as the loss function, the learning rate is 0.001, Adam is selected as the optimizer, which is stable and converges quickly, and the number of training rounds is 100; S11. After 100 rounds of training, parameter B is obtained; after evaluating the fitness of parameter A and parameter B respectively, they are compared; if the fitness of parameter A is less than that of parameter B, parameter A is retained, otherwise parameter B is retained.

Images