Improved PSO optimization-based fuzzy neural network PID photovoltaic series welding temperature control method

Abstract

The invention discloses a fuzzy neural network PID photovoltaic series welding temperature control method based on improved PSO optimization. The method comprises the steps that a target photovoltaic series welding equipment heating transfer function model is acquired; setting an initial PID parameter; adjusting a PID increment parameter of the PID control module in real time according to the temperature error, the temperature error change rate and a preset fuzzy rule base; the neural network is combined with fuzzy control, and parameters and rules of fuzzy control are automatically modified through training data; according to the photovoltaic series welding temperature condition, the fitness function of the PSO algorithm is improved, the adjustment mode of the inertia weight is improved, and the improved PSO algorithm is used for optimizing the initial PID parameters of fuzzy neural network PID photovoltaic series welding temperature control. Self-adaptive precise control and robust control of the welding temperature control system are achieved, the temperature fluctuation phenomenon in the photovoltaic series welding process is effectively improved, and the robustness and control reliability of the system are enhanced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of photovoltaic string welding temperature control, and particularly relates to a fuzzy neural network PID photovoltaic string welding temperature control method based on improved PSO optimization. BACKGROUND

[0002] The photovoltaic string welding equipment plays a key role in realizing efficient and reliable interconnection of battery pieces, and the temperature field regulation of the welding area has a decisive influence on the final welding quality. However, the inherent heat radiation, heat convection and other multi-physical field coupling characteristics of the welding process make the controlled system present strong nonlinearity and time-varying dynamics, resulting in great challenges for temperature control.

[0003] The current industrial experience setting method commonly relied on by the industry is difficult to realize accurate, adaptive and robust control of temperature, which limits the improvement of welding quality and production efficiency. SUMMARY

[0004] The application aims to solve the problems of non-adaptive control, inaccurate temperature control and susceptibility to external interference of the photovoltaic string welding machine during the welding process, and proposes a fuzzy neural network PID photovoltaic string welding temperature control method based on improved PSO optimization.

[0005] In order to achieve the purpose of the application, the application provides a fuzzy neural network PID photovoltaic string welding temperature control method based on improved PSO optimization, which comprises the following steps:

[0006] Step 1, obtaining a target photovoltaic string welding equipment heating transfer function model;

[0007] Step 2, setting initial PID parameters using Ziegler-Nichols tuning method, and establishing a basic PID closed-loop control;

[0008] Step 3, adjusting the PID increment parameters of the PID control module in real time according to the temperature error and the temperature error change rate, and according to the preset fuzzy rule base, to establish a fuzzy PID closed-loop control system;

[0009] Step 4, combining the neural network with the fuzzy control, automatically modifying the parameters and rules of the fuzzy control through training data, and establishing a fuzzy neural network PID closed-loop control system;

[0010] Step 5, improving the fitness function of the PSO algorithm according to the photovoltaic string welding temperature conditions, improving the adjustment mode of the inertia weight, and optimizing the initial PID parameters of the fuzzy neural network PID photovoltaic string welding temperature control using the improved PSO algorithm.

[0011] Compared with the prior art, the significant progress of the application is that the application combines fuzzy control with neural network to adaptively control PID increment, improves the fitness function of PSO algorithm according to photovoltaic temperature conditions, improves the adjustment mode of inertia weight, uses the improved PSO algorithm to optimize the initial PID parameters of photovoltaic string welding temperature control, realizes adaptive accurate control and robust control of the welding temperature control system, effectively improves the temperature fluctuation phenomenon in the photovoltaic string welding process, and enhances the robustness and control reliability of the system. The application has adaptive control capability and strong robustness.

[0012] To more clearly illustrate the functional properties and structural parameters of the application, the following further describes the application in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 is a fuzzy neural network PID photovoltaic string welding temperature control structure diagram based on improved PSO optimization;

[0014] Figure 2 is one of the simulation comparison diagrams of different control schemes;

[0015] Figure 3 is the second simulation comparison diagram of different control schemes;

[0016] Figure 4 is a temperature curve schematic diagram;

[0017] Figure 5 is a closed-loop control system structure diagram;

[0018] Figure 6 is a fuzzy PID control structure diagram;

[0019] Figure 7 is a network structure diagram;

[0020] Figure 8 is a control structure diagram of using fuzzy neural network PID for photovoltaic string welding temperature control;

[0021] Figure 9 is an improved PSO optimization fuzzy neural network PID control structure diagram. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0023] Step 1, obtain the target photovoltaic string welding equipment heating transfer function model.

[0024] The dynamic characteristics of the temperature control system of the stringer are complex, with nonlinearity, time delay and large inertia. A first-order process with time delay can better represent the characteristics of the temperature control system of the stringer, obtain the temperature step-up curve of the welding area, and determine the temperature rise transfer function of the welding area by step response method. The temperature control system transfer function model is as follows:

[0025] ;

[0026] In the formula K is the static gain; τ is the time constant; is the pure delay time; is the temperature error, is the Laplace variable, is the Laplace transform of the system input; is the Laplace transform of the system output, is the temperature transfer function.

[0027] Small power temperature rise experiments were carried out at the standby temperature of the welding area of the stringer. The temperature data of 16 sampling points in the welding area were recorded every 1s and the average value was obtained. The obtained temperature curve is shown in Figure 4 .

[0028] The step response method was used to identify the parameters, and the temperature transfer function mathematical model was obtained as follows:

[0029] ;

[0030] Step 2, use Ziegler-Nichols tuning method to set initial PID parameters and establish PID closed-loop control system. Using PID control algorithm to control the welding temperature of photovoltaic stringer is the most common temperature regulation method in industrial field at present. PID controller adjusts the system temperature error through three items of proportion, integral and differential. The proportional term is used to provide fast response according to the current error, the integral term is used to eliminate the steady-state error of the system, and the differential term is used to predict the error trend, so as to improve the response speed and stability of the system. The output of the PID controller is expressed as:

[0031] ;

[0032] Among them, is the output of the PID controller; is the error between the system reference input and the actual output; t is the time; , , proportional, integral and derivative gain coefficients, respectively.

[0033] The structure block diagram of the closed-loop control system is shown in Figure 5 .

[0034] The initial parameters of the PID module are determined by the Ziegler-Nichols tuning method as = 0.24, = 0.019, = 1.5;

[0035] Step 3, according to the temperature error and the temperature error rate of change, and according to the pre-set fuzzy rule base, the PID incremental parameters of the PID control module are adjusted in real time, and a fuzzy PID closed-loop control system is established.

[0036] The actual input temperature error and the error rate of change are converted into fuzzy language variables by fuzzification. The input variables are fuzzily divided by using a Gaussian membership function. According to the requirements of control accuracy and response speed, the error and the error rate of change are divided into seven language variables: NB (negative big), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), and PB (positive big). The value range is normalized according to the specific performance index of the system, each language variable corresponds to a fuzzy subset, and the fuzzy degree is described by a membership function.

[0037] The fuzzy inference deduces the adjustment strategy for the controller output according to the input variables after fuzzification by using the pre-set fuzzy control rules. The Mamdani type fuzzy inference system is used, the current error and the error rate of change are input language variables, and the PID parameter correction amount , , are proportional, integral and differential gain coefficients corresponding to the output language variables. The design of fuzzy rules is based on control experience and system response characteristics to construct a rule table, a total of 7x7=49 fuzzy rules are formed to cover all possible input combinations. , , The fuzzy rules of the PID module are shown in the following table.

[0038] Table 1 fuzzy rule table

[0039]

[0040] Table 2 fuzzy rule table

[0041]

[0042] Table 3 fuzzy rule table

[0043]

[0044] Defuzzification converts the fuzzy set obtained by fuzzy inference into a definite numerical control variable. The centroid method is used for defuzzification, and a representative and smooth output value is obtained by weighted average of the membership functions of the output fuzzy set:

[0045] ;

[0046] wherein, is the membership degree corresponding to the i-th fuzzy subset, is the centroid of the i-th fuzzy subset, is the defuzzification result, and i is an index variable.

[0047] The three linguistic variable values of the output variable are defuzzified, and the three parameters , , of the PID controller are output. The linguistic variable values of the output variable are "negative big" (NB), "negative medium" (NM), "negative small" (NS), "zero" (Z), "positive small" (PS), "positive medium" (PM), and "positive big" (PB), and the membership degree functions are also Gaussian functions.

[0048] The block diagram of the fuzzy PID control structure is shown in Figure 6 .

[0049] Step 4: Combine the neural network with the fuzzy control, automatically modify the parameters and rules of the fuzzy control through training data, and establish a fuzzy neural network PID closed-loop control system.

[0050] The fuzzy neural network PID control algorithm is used for photovoltaic string welding temperature control, aiming to solve the problems of strong nonlinearity, large inertia, and time delay inherent in the process, so as to improve the dynamic response performance and temperature control accuracy of the system. Through the input temperature error and temperature error change rate , the three parameters , , of the PID controller are output through the fuzzy neural network structure to realize adaptive adjustment of the PID, and the fuzzy neural network structure contains five layers, corresponding to the input layer, the membership function layer, the rule layer, the normalization layer, and the output layer, respectively. The network structure diagram is shown in Figure 7 .

[0051] The function of each layer in a fuzzy neural network is as follows:

[0052] The first input layer is used to input data. Passed to each node:

[0053] ;

[0054] In the formula, This is the i-th output of the first layer; It is the i-th input; i represents the index variable.

[0055] The second membership function layer calculates the membership degree of each node to its respective fuzzy set.

[0056] ;

[0057] In the formula, This is the i-th output of the second layer; It is a learnable membership function including Gaussian and triangle functions; the parameters, including center and width, are adjusted through training. Indicates input In fuzzy sets The membership value under; i represents the index variable.

[0058] If the Gaussian membership function is used, the membership degree is expressed as:

[0059] ;

[0060] In the formula, Indicates input In fuzzy sets The membership value is denoted by ; x is the input parameter. and These are the center and width parameters of the membership function, respectively, i.e., the antecedent parameters; i represents the index variable.

[0061] The third layer is the rule layer. Each node in the rule layer represents a rule. The formula for calculating the activation strength of a rule is:

[0062] ;

[0063] In the formula, This is the i-th output of the third layer; Let be the activation strength of the i-th rule; x be the first input variable; y be the second input variable; i be the index variable; the activation strength formula shows that the activation strength of a rule is the product of the membership degrees of the input variables;

[0064] The fourth layer is a normalization layer, each node of which normalizes the activation strength of the rule:

[0065]

[0066] wherein, is the i-th output of the fourth layer; is the normalized strength of the i-th rule; is the total activation strength of all rules; i, j represent index variables;

[0067] The fifth layer is an output layer, which is used to calculate the output of each rule and to perform a weighted summation of the outputs of all rules to obtain the final system output:

[0068]

[0069] wherein, is the i-th output of the fifth layer; is the linear consequent function of the i-th rule; , is a linear function of the input variables x, y; is a bias term of the consequent function; , , are three consequent parameters, which are obtained by least square learning; i represents an index variable.

[0070] In order to realize efficient training of the fuzzy neural network model, a Hybrid training algorithm is usually used to iteratively optimize the parameters in the network. This method combines the advantages of gradient descent method and least square method, which are used to adjust the antecedent parameters and consequent parameters respectively, so as to balance the convergence speed and calculation accuracy.

[0071] The first stage: forward propagation and least square estimation optimization of the consequent parameters, in this stage, the antecedent parameters are kept unchanged, the shapes of all membership functions are fixed, and the rule activation strength and the normalized value are calculated by forward propagation, and a linear equation set is constructed:

[0072]

[0073] wherein, is the target output of the k-th sample; , are two inputs; M is the number of rules; is the normalized strength of the i-th rule in the k-th sample; , , ​​​Respectively, three consequent parameters to be solved; i represents the index variable; k represents the kth sample.

[0074] Second stage: back propagation and gradient descent optimization of antecedent parameters, fixing the consequent parameters obtained in the first stage, deriving the error function of the model with respect to the antecedent parameters, and adjusting the antecedent parameters by the gradient descent method. The error function is:

[0075] ;

[0076] In the formula, is the expected output; is the actual output; N is the number of samples; k represents the index variable.

[0077] The control structure block diagram of the fuzzy neural network PID for photovoltaic string welding temperature control is shown in Figure 8 .

[0078] Step 5, using the improved PSO algorithm to optimize the initial PID parameters of the fuzzy neural network PID photovoltaic string welding temperature control.

[0079] PSO gradually approaches the optimal solution of the problem through information sharing and collaborative search between individual particles in the group, and has the advantages of simple structure, high computational efficiency, and few parameters. The basic execution process is as follows:

[0080] Step 51, initialization, initializing the size of the particle group , the dimension of the search space , and the maximum number of iterations , randomly initializing the position and speed of each particle, initializing the individual optimal position = , and finding the current global optimal position ;

[0081] Step 52, fitness evaluation, calculating the fitness of the current position of each particle according to the fitness function defined by the optimization problem ; if the current fitness of the particle is better than the historical optimal , update the individual optimal; if the individual optimal is better than the current global optimal , update the global optimal;

[0082] Step 53, speed and position update, the position and speed of the particle are iteratively updated according to the following method:

[0083] ;

[0084] ;

[0085] where, is the inertia weight, which controls the search range and convergence speed of the particles; , are two learning factors, respectively representing the influence degree of individual experience and group experience; , are two random numbers in the interval [0, 1], which are used to increase the randomness of the search; is the historical optimal position of particle i; is the global optimal position of the entire particle swarm; is the current position of particle i; is the next position of particle i; is the current velocity of particle i; is the next velocity of particle i; i represents the i-th particle, and k represents the iteration number.

[0086] Step 54, iteration judgment, if the maximum iteration number is reached or the optimal stop condition is met, the algorithm terminates; otherwise, go back to step 52 and continue the next round of iteration.

[0087] Step 55, output the optimal solution, after the algorithm ends, output the position of the current optimal particle as the optimal solution of the problem.

[0088] Since the PSO algorithm often falls into local optimal solution, for the photovoltaic string welding which requires fast response, high precision temperature control and small overshoot, a suitable dynamic inertia weight adjustment method is designed for PID parameter optimization. A nonlinear decreasing inertia weight adjustment method based on S-shaped function is used; compared with the common linear decreasing strategy, this nonlinear method can better balance the global and local search ability.

[0089] A reverse sigmoid function is used to control the change of inertia weight; the formula is as follows:

[0090] ;

[0091] where, is the inertia weight at the t-th iteration; t is the current iteration number; is the maximum iteration number; is the upper limit value of inertia weight, set to 0.9; is the lower limit value of inertia weight, set to 0.4; k is an adjustment factor used to control the steepness of the curve, set to 7;

[0092] A typical welding cycle contains the following actions and requirements:

[0093] ​Ramp-up phase: fast heating from ambient temperature to welding set point, requiring fast speed, but strictly prohibited overshoot;

[0094] Stable phase: after reaching the set point, the temperature will fluctuate slightly, and it is required to enter a very small error band as soon as possible, and the time to enter the stable state should be short;

[0095] Welding phase: the battery string is welded on the hot table, and the temperature must be stable, and the smaller the fluctuation, the better;

[0096] Disturbance phase: a new string of battery pieces enters the hot table, which will instantly take away a lot of heat, causing the temperature to drop, and the temperature drop should be small and quickly recover to the set temperature to ensure the quality of the subsequent welding points;

[0097] According to the photovoltaic temperature condition, the fitness function of the PSO algorithm is improved, and a multi-stage comprehensive fitness function is determined:

[0098] ;

[0099] Wherein, is the fitness function; is the overshoot penalty; is the adjustment time penalty; is the welding start time; is the welding end time; is the error at time t; t is the time; is the maximum temperature drop; is the recovery time.

[0100] The improved PSO optimization fuzzy neural network PID control structure block diagram is shown in Figure 9 .

[0101] The application will be further described in detail below in conjunction with the embodiments.

[0102] Embodiment

[0103] In this embodiment, 250℃ is selected as the step response signal for simulation comparison. The simulation conditions are that the initial parameters of the PID module are determined by the Ziegler-Nichols tuning method =0.24, =0.019, =1.5; the input variable domain of fuzzy PID is [-6, 6], the output variable domain is [-3, 3], the quantization factors are 0.01 and 0.012 respectively, and the proportional factors are 0.08, 0.04 and 0.2 respectively; the fuzzy neural network PID is trained through the anfis module in MATLAB, and the training number is set to 1000 times; the optimal PID initial value obtained by the improved PSO algorithm is = 0.0623, = 0.0202, = 1.523; The four control schemes were compared and simulated with a step signal of 250 as input and a running time of 800 s. The dynamic performance of the system obtained by simulation is shown in the following table.

[0104]

[0105] Through Simulink simulation of different controller structures, it can be observed that the traditional PID controller has a significant overshoot in the response process, with an overshoot of 19.92% and a relatively long adjustment time of 134.67 s, which is difficult to meet the demand for fast response and high-precision control. In contrast, the fuzzy PID controller with fuzzy control rules reduces the overshoot to a certain extent, with an overshoot of 7.98% and a shorter adjustment time of 92.79 s. The system response is more stable and exhibits certain self-adaptive ability. Further use of the fuzzy neural network PID controller significantly improves the system performance, reducing the overshoot to 3.36% and speeding up the adjustment to 68.19 s. On this basis, the use of the particle swarm optimization algorithm to optimize the fuzzy neural network parameters results in the most ideal system response, with essentially no overshoot and the shortest adjustment time of 18.53 s. The dynamic and steady-state performances are better than those of the previous three methods.

[0106] In actual production, temperature will be affected by various external conditions. Firstly, it will be disturbed by cooling air to ensure uniform temperature. Secondly, the battery piece is periodically stepped throughout the welding process. To verify the robustness of the control system, a fluctuation with an amplitude of -15 was added to the entire stringer temperature control system at 200 s of system operation to simulate the wind disturbance of the cooling air. At 400 s of system operation, a periodic disturbance with an amplitude of -5, a period of 1 s, and a duty cycle of 90 was superimposed to simulate the temperature disturbance caused by the periodic stepping of the battery piece in actual working conditions.

[0107] When the system is disturbed by external disturbance, different control algorithms show obvious differences. When disturbed by wind, the traditional PID controller is most sensitive to disturbance, showing obvious output fluctuation and oscillation phenomenon, poor system stability, response delay of 24s, and temperature deviation of 23.5℃. In comparison, the fuzzy PID control effect is optimized, but its response to disturbance is still relatively slow, there is a certain degree of hysteresis, and the temperature fluctuation is still large, the response delay is 19s, and the temperature deviation is 17.9℃. The fuzzy neural network PID control can effectively suppress system oscillation, and the response is relatively rapid, the response is more stable, the response delay is 16s, and the temperature deviation is 14.3℃. The fuzzy neural network PID controller improved by PSO optimization performs best under disturbance, with a response delay of only 9s and a temperature deviation of only 11.4℃, not only the system response speed is significantly improved, and the deviation caused by disturbance is the smallest, showing stronger robustness and dynamic adaptability. Under the influence of battery piece periodic stepping, the performance is similar to that under wind disturbance, the response delay of the traditional PID controller is 23s, and the temperature deviation is 7℃. The response delay of fuzzy PID control is 18s, and the temperature deviation is 5℃. The response delay of fuzzy neural network PID control is 15s, and the temperature deviation is 4℃. The response delay of fuzzy neural network PID control based on improved PSO optimization is 8s, and the temperature deviation is 3.8℃.

[0108] The method of the improved PSO optimization based fuzzy neural network PID photovoltaic string welding temperature control method combines fuzzy control with neural network to adaptively control PID increment, improves the fitness function of PSO algorithm according to the photovoltaic temperature condition, improves the adjustment mode of the inertia weight, optimizes the initial PID parameters of the photovoltaic string welding temperature control by using the improved PSO algorithm, and simulation verification shows that the method can realize adaptive control of accurate temperature control, and can quickly respond when disturbed, has small disturbance error, and has strong robustness.

[0109] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A PID photovoltaic string bonding temperature control method based on an improved PSO-optimized fuzzy neural network, characterized in that, The method includes: Step 1: Obtain the heating transfer function model of the target photovoltaic string welding equipment; Step 2: Use the Ziegler-Nichols tuning method to set the initial PID parameters and establish basic PID closed-loop control; Step 3: Based on the temperature error and the rate of change of temperature error, and according to the preset fuzzy rule library, adjust the PID incremental parameters of the PID control module in real time to establish a fuzzy PID closed-loop control system. Step 4: Combine neural networks with fuzzy control, and automatically modify the parameters and rules of fuzzy control through training data to establish a fuzzy neural network PID closed-loop control system. Step 5: Improve the fitness function of the PSO algorithm based on the photovoltaic string bonding temperature conditions, improve the adjustment method of the inertia weight, and use the improved PSO algorithm to optimize the initial PID parameters of the fuzzy neural network PID photovoltaic string bonding temperature control.

2. The method according to claim 1, characterized in that, Obtain the heating transfer function model of the target photovoltaic string bonding equipment, including: Obtain the temperature step rise curve of the welding zone, and determine the temperature transfer function of the welding zone using the step response method; Transfer function model of the welding zone: ； In the formula This is the static gain; It is a time constant; This is the pure time delay; For temperature error, For Laplace variables, Laplace transform of the system input; The Laplace transform of the system output. This is the temperature transfer function.

3. The method according to claim 1, characterized in that, The PID closed-loop control is shown below: ； in, For PID controller output; The input is the error between the system reference input and the actual output; t is time. , , These are the proportional, integral, and differential gain coefficients, respectively.

4. The method according to claim 1, characterized in that, Based on the temperature error and its rate of change, and according to a preset fuzzy rule base, the PID incremental parameters of the PID control module are adjusted in real time to establish a fuzzy PID closed-loop control system, including: The actual input temperature error is represented by fuzzification. and error change rate Convert to fuzzy linguistic variables; use Gaussian membership functions to fuzzily partition the input; based on the requirements of control accuracy and response speed, the error... and error change rate They are divided into seven linguistic variables: NB negative large, NM negative medium, NS negative small, ZO zero, PS positive small, PM positive medium, and PB positive large. The value range is normalized according to the specific performance indicators of the system. Each linguistic variable corresponds to a fuzzy subset, and the degree of fuzziness is described by a membership function. Fuzzy inference, based on the fuzzified input variables and through pre-defined fuzzy control rules, derives an adjustment strategy for the controller output; a Mamdani-type fuzzy inference system is employed, with the current error... and error change rate For input language variables, PID parameter correction amount , , These are the output linguistic variables corresponding to the proportional, integral, and derivative gain coefficients, respectively; the fuzzy rules are constructed into a rule table based on control experience and system response characteristics, forming a total of 7×7=49 fuzzy rules to cover all possible input combinations; , , The fuzzy rules are shown in Tables 1, 2, and 3, respectively: Table 1 Fuzzy rule table ； Table 2 Fuzzy rule table; ； Table 3 Fuzzy rule table; ； Defuzzification converts the fuzzy sets obtained from fuzzy inference into explicit numerical control quantities. The centroid method is used for defuzzification, and the output value is obtained by calculating a weighted average of the membership functions of the output fuzzy sets. ； In the formula, in the formula, To represent the membership degree corresponding to the i-th fuzzy subset, Denotes the centroid of the i-th fuzzy subset. For the defuzzing result, i represents an index variable.

5. The method according to claim 1, characterized in that, By combining neural networks with fuzzy control, and automatically modifying the parameters and rules of fuzzy control through training data, a fuzzy neural network PID closed-loop control system is established, including: The fuzzy neural network structure consists of five layers, corresponding to the input layer, membership function layer, rule layer, normalization layer, and output layer, respectively. The function of each layer in a fuzzy neural network is as follows: The first input layer is used to input data. Passed to each node: ； In the formula, This is the i-th output of the first layer; It is the i-th input; i represents the index variable; The second membership function layer calculates the membership degree of each node to its respective fuzzy set. ； In the formula, This is the i-th output of the second layer; It is a learnable membership function including Gaussian and triangle functions; the parameters, including center and width, are adjusted through training. Indicates input In fuzzy sets The membership value under; i represents the index variable; If the Gaussian membership function is used, the membership degree is expressed as: ； In the formula, Indicates input In fuzzy sets The membership value is denoted by ; x is the input parameter; and These are the center and width parameters of the membership function, respectively, i.e., the antecedent parameters; i represents the index variable; The third layer is the rule layer. Each node in the rule layer represents a rule. The formula for calculating the activation strength of a rule is: ； In the formula, This is the i-th output of the third layer; Let be the activation strength of the i-th rule; x be the first input variable; y be the second input variable; i be the index variable; the activation strength formula shows that the activation strength of a rule is the product of the membership degrees of the input variables; The fourth normalization layer normalizes the activation strength of each rule at each node. ； In the formula, This is the i-th output of the fourth layer; Let be the normalized strength of the i-th rule; The total activation strength of all rules; i and j represent index variables; The fifth layer is the output layer, which calculates the output of each rule and performs a weighted sum of the outputs of all rules to obtain the final system output. ； In the formula, This is the i-th output of the fifth layer; Let be the linear consequent function of the i-th rule; , It is a linear function of input variables x and y; This is the bias term for the consequent function; , , There are three consequent parameters, which are learned through the least squares method; i represents the index variable.

6. The method according to claim 1, characterized in that, The fitness function of the improved PSO algorithm is modified based on the photovoltaic string bonding temperature conditions, and the adjustment method of the inertia weight is improved. The initial PID parameters of the fuzzy neural network PID photovoltaic string bonding temperature control are optimized using the improved PSO algorithm, including: Step 51, Initialization: Initialize and set the size of the particle swarm. Search space dimension and the maximum number of iterations Randomly initialize the position for each particle. and speed Initialize the individual's optimal position = And find the current global optimal position. ; Step 52, Fitness Evaluation, based on the fitness function defined in the optimization problem. Calculate the current fitness of each particle; if a particle's current fitness is better than its historical best... If the individual optimal value is better than the current global optimal value, then update the individual optimal value; Then update the global optimum; Step 53, velocity and position update: The particle's position and velocity are iteratively updated according to the following method: ； ； In the formula, Inertial weights control the search range and convergence speed of particles; , These are two learning factors, representing the degree of influence of individual experience and group experience, respectively; , These are two random numbers within the interval [0,1], used to increase the randomness of the search; For particles The historical best position; The global optimal position of the entire particle swarm; Let i be the current position of particle i. Let i be the position of particle i at the next moment; Let i be the velocity of particle i at the current moment; Let be the velocity of particle i at the next moment; i represents the i-th particle, and k represents the iteration number. Step 54: Iterative judgment. If the maximum number of iterations is reached or the set optimal stopping condition is met, the algorithm terminates; otherwise, return to step 52 and continue a new round of iteration. Step 55: Output the optimal solution. After the algorithm ends, output the position of the current best particle as the optimal solution to the problem. Since the PSO algorithm often gets stuck in local optima, a nonlinear decreasing inertia weight adjustment method based on a sigmoid function is used. The inverse sigmoid function is used to control the change in inertia weights; the formula is as follows: ； in, It represents the inertia weight at the t-th iteration; t is the current iteration number. It is the maximum number of iterations; This is the upper limit of the inertia weight, set to 0.9; is the lower limit of the inertia weight, set to 0.4; k is an adjustment factor used to control the steepness of the curve, set to 7; A typical welding cycle includes the following actions and requirements: Heating phase: Rapidly heat from ambient temperature to the welding set point, requiring high speed, but overshoot is strictly prohibited; Stable phase: After reaching the set point, the temperature will fluctuate slightly and needs to enter a very small error band as soon as possible. The time to enter a stable state should be short. Welding stage: The battery strings are welded on a hot plate, and the temperature must be stable, with the smallest fluctuations being better; Disturbance phase: When a new string of solar cells enters the heating stage, it will instantly carry away a large amount of heat, causing a temperature drop. The temperature drop should be small and can quickly recover to the set temperature. Based on the improved fitness function of the PSO algorithm under photovoltaic temperature conditions, the multi-stage integrated fitness function is determined: ； in, The fitness function; As a penalty for overshooting; To adjust time penalties; This refers to the welding start time; This refers to the welding completion time; Let be the error at time t; t is time. For the maximum temperature drop; This refers to the recovery time.

Images