Wastewater treatment control strategy based on fuzzy neural network

Abstract

The invention discloses a wastewater treatment control strategy based on a fuzzy neural network, and the strategy comprises the steps: S1, collecting sample data of a wastewater treatment system, and dividing the sample data into a training set and a test set according to a fixed proportion; determining an input parameter and a pre-measurement parameter; s2, performing data normalization processing on the sample data; s3, constructing a PSO-BP-based wastewater treatment prediction model; S4, training the wastewater treatment prediction model through the training set sample data after data preprocessing; s5, performing performance evaluation on the wastewater treatment prediction model through test set sample data; s6, performing water quality prediction on the wastewater treatment system through the trained wastewater treatment prediction model; the problems that the cost is high and the real-time performance is poor when parameters of effluent chemical oxygen demand and inflow suspended solids of a sewage treatment system are monitored are solved, and the method has the advantages that the key parameters of water quality can be predicted through an improved neural network model, so that high-precision prediction data are obtained, and the real-time performance is high.

Description

Technical Field

[0001] The invention relates to the technical field of wastewater treatment prediction, and in particular to a wastewater treatment control strategy based on fuzzy neural network. Background Art

[0002] In wastewater treatment, it is necessary to measure the chemical oxygen demand (COD) of the system’s effluent. eff and influent solid suspended matter SS eff The two water quality parameters are monitored in real time; conventional wastewater quality detection methods include online detection and manual detection, but the installation and maintenance costs of some online detection instruments are relatively expensive, which increases the detection cost; and the manual detection method is time-consuming and the detection data is relatively lagging; compared with these conventional detection methods, soft measurement technology can significantly solve the above problems, with the advantages of low cost, short time consumption and relative accuracy. In recent years, it has been widely used in online monitoring of key water quality in wastewater treatment. Artificial neural network is a very widely used soft measurement modeling method, which has the advantages of nonlinearity, parallel distribution, fault tolerance and adaptability; there are methods of using BP neural network for prediction in the prior art, but due to the defects of BP neural network such as slow convergence speed, long training time, network sensitivity to initial value, easy to fall into local optimum, over-reliance on input samples and different network structure selection, the prediction error of BP neural network is too large, which affects the precision and accuracy of prediction; therefore, it is necessary to introduce a new algorithm to improve the BP neural network structure, so as to design a wastewater treatment control strategy based on fuzzy neural network to solve the above problems. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a wastewater treatment control strategy based on fuzzy neural network to solve the problems of high cost and poor real-time performance when monitoring the effluent chemical oxygen demand and influent solid suspended matter parameters of the sewage treatment system. The strategy has the characteristics of being able to predict key water quality parameters through an improved neural network model, thereby obtaining high-precision prediction data and having strong real-time performance.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: The wastewater treatment control strategy based on fuzzy neural network includes the following steps: S1, Data Collection: Collect sample data from the wastewater treatment system and divide the sample data into a training set and a test set according to a fixed ratio; determine input quantity parameters and predicted quantity parameters; S2, data preprocessing: normalize the sample data; S3, constructing a wastewater treatment prediction model based on BP neural network, and optimizing the BP neural network structure of the wastewater treatment prediction model through PSO algorithm; wherein the PSO algorithm is improved; S4, training the wastewater treatment prediction model through the training set sample data after data preprocessing; S5, evaluating the performance of the wastewater treatment prediction model through the test set sample data; S6, predicting the water quality of the wastewater treatment system through the trained wastewater treatment prediction model.

[0005] Preferably, in step S1, determining the input quantity parameters and the predicted quantity parameters includes: The flow rate Q, pH, temperature T, dissolved oxygen DO, and influent chemical oxygen demand COD inf and influent solid suspended matter SS inf As input quantity parameter; The chemical oxygen demand (COD) of the effluent from the wastewater treatment system eff and influent solid suspended matter SS eff as a predictor parameter.

[0006] Preferably, in step S2, performing data normalization processing on the sample data includes: The sample data is normalized, and the normalization formula is as follows: ; In the formula, x max is the maximum value in the sample data, x min is the minimum value in the sample data; x is any value in the sample data, and X is the normalized value.

[0007] Preferably, in step S3, constructing a wastewater treatment prediction model based on a BP neural network includes: A three-layer BP neural network model is constructed, including an input layer, a hidden layer, and an output layer; the nodes in the previous layer and the next layer are connected by corresponding weights.

[0008] Furthermore, the BP neural network model includes signal forward transmission and error reverse propagation. Signal forward transmission is to transmit the input signal to the hidden layer and the output layer to obtain the predicted output, and then obtain the error signal from the predicted output and the expected output. The reverse propagation of the error is to feed the error signal back to the previous neural network, and then the neural network adjusts the corresponding weights and values ​​according to the size of the error. This process is carried out according to the error gradient descent method; The BP neural network first trains the frontmost layer of the network, fixes its weights and thresholds after the training is completed, and then uses the output of its hidden unit as the input vector of the next layer of the network to continue training; after all the networks are trained, the back propagation algorithm is used to repeatedly adjust the weights and thresholds until the error signal between the predicted output and the expected output meets the target requirements and the operation is terminated.

[0009] Preferably, in step S3, the wastewater treatment prediction model is optimized by an improved PSO algorithm, and the specific method is as follows: S501, calculating the total number of network weights and thresholds, i.e., the length of particles, according to the number of nodes in each layer of the BP neural network; S502, encoding the weights and thresholds of the BP neural network into corresponding particles in a binary encoding manner; S503, initialize the particle swarm, including the population size N, the maximum number of iterations G, the position x of each particle i，j and speed v i，j , i represents the number of particles, j represents the dimension of particles; set the relevant parameters of particle swarm algorithm; S504, calculating the fitness of each particle in the initial particle swarm; S505, recording the individual extreme value and the global extreme value of the particle group respectively; S506, respectively update the velocity v of each particle in the current particle group i and position x i ; ; ; In the formula, V represents the updated speed of the particle, X represents the updated position of the particle, and V i，j and X i，j Assign to x i，j and v i，j , update the position and velocity of the particle; gbest is the individual optimal value, Bestx is the global optimal value, c 1 、c 2 is the learning factor, w is the adaptive inertia weight, is the maximum value of the adaptive inertia weight, is the minimum value of the adaptive inertia weight; the adaptive variable c 1Determine the optimization process to be biased towards local optimization, adaptive variables c2 Determine the optimization process to be biased towards the global optimum, adaptive inertia weight w Prefer global search; Adaptive variables c 1 andc 2The calculation method is as follows: ; ; In the formula, G is the maximum number of iterations, and e is the remaining number of iterations; S507, calculating the fitness of each particle in the new particle group; S508, updating the individual extreme value of each particle; S509, updating the global extreme value of the particle swarm; S5010, if the end condition is met or the number of iterations reaches the maximum number of loop iterations, exit the loop, otherwise return to S506; S5011, output the optimal solution of the problem, and assign the optimal solution to the BP neural network for prediction.

[0010] Preferably, in step S3, the PSO algorithm is improved, including using a linear weight reduction method and introducing a mutation operator to improve the PSO algorithm, and the specific method is: Linearly reduce the weight coefficient: ; In the formula, is the maximum value of the adaptive inertia weight, is the minimum value of the inertia weight, t is the current number of iterations, and G is the maximum number of iterations; Introducing mutation operators: In the iterative optimization process, mutation operators are introduced, specifically: pos = unidrnd(numsum); if rand>0.95 pop(j,pos)=3*rands(1,1); Among them, pos is the position where the population mutates, unidrnd is a function that generates random integers, and numsum is the particle length.

[0011] Preferably, in step S5, when the performance of the wastewater treatment prediction model is evaluated by the test set sample data, the experimental root mean square error RMSE and the determination coefficient R 2 To evaluate the model performance, specifically: ; ; Among them, among them, is the predicted value of the test set sample data, is the true value of the test set sample data, is the average value of the test set sample data, and n is the number of test set sample data; Model optimization degree R m is defined as: ; in, is the model correlation coefficient before optimization, is the correlation coefficient of the optimized model.

[0012] Preferably, in step S6, predicting the water quality of the wastewater treatment system using the trained wastewater treatment prediction model includes: S601, BP neural network model parameter setting: The input layer of the BP neural network sets 6 input nodes, corresponding to 6 wastewater quality parameter variables Q, pH, T, DO, COD inf and SS inf ; The output layer of the BP neural network sets one node to predict COD eff and SS eff ; The number of hidden layer nodes of the BP neural network is set by the empirical value formula: ; Where q is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, and a∈{1,2,…,10} is the adjustment constant; S602, wastewater treatment prediction model parameter setting: Based on the structure of BP neural network, the relevant parameters of the improved PSO algorithm are set as follows: Fitness function setting: The fitness value is defined as the absolute value of the total error of the optimized BP neural network structure. The formula is: ; In the formula, a is the predicted output of the BP neural network, and o is the true value of the sample output.

[0013] The beneficial effects of a wastewater treatment control strategy based on fuzzy neural network provided by the present invention are as follows: 1. The present invention optimizes the BP neural network structure through the PSO algorithm, thereby overcoming the defects of the traditional BP neural network model in predicting the effluent water quality of water treatment, such as slow convergence speed, long training time, high sensitivity of the network to initial values, easy to fall into local optimum, over-reliance on input samples and different network structure selection, and improving the prediction accuracy; 2. This method also applies adaptive inertia weight and adaptive learning factor to increase the optimization strength of the early stage of the identification algorithm to avoid falling into the local optimum, while reducing the optimization range of the late stage of the identification algorithm and improving the parameter identification accuracy; 3. This method improves the PSO algorithm by linearly reducing the weight coefficient and introducing the mutation operator, so that the algorithm has a stronger global optimization ability in the early stage and a stronger local development ability in the late stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a schematic diagram of the process flow of the invention; Figure 2 It is a schematic diagram of the BP neural network structure in the present invention; Figure 3 It is a schematic diagram of the PSO algorithm flow in the present invention; Figure 4 is a fitness function curve diagram of the PSO-BP wastewater treatment prediction model in an embodiment of the present invention; Figure 5 is the COD of the PSO-BP wastewater treatment prediction model in the embodiment of the present invention. eff Schematic diagram of prediction results; Figure 6 is the SS of the PSO-BP wastewater treatment prediction model in the embodiment of the present invention. eff Schematic diagram of prediction results; Figure 7 is the COD of the BP model in the embodiment of the present invention eff Schematic diagram of prediction results; Figure 8 is the SS of the BP model in the embodiment of the present invention eff Schematic diagram of prediction results; Fig. 9 is the COD of the GA-BP model in the embodiment of the present invention eff Schematic diagram of prediction results; Fig.10 is the SS of the GA-BP model in the embodiment of the present invention. eff Schematic diagram of prediction results. DETAILED DESCRIPTION

[0015] Embodiment 1: like Figure 1 As shown, the wastewater treatment control strategy based on fuzzy neural network includes the following steps: S1, Data Collection: Collect sample data from the wastewater treatment system and divide the sample data into a training set and a test set according to a fixed ratio; determine input quantity parameters and predicted quantity parameters; S2, data preprocessing: normalize the sample data; S3, constructing a wastewater treatment prediction model based on BP neural network, and optimizing the BP neural network structure of the wastewater treatment prediction model through PSO algorithm; wherein the PSO algorithm is improved; S4, training the wastewater treatment prediction model through the training set sample data after data preprocessing; S5, evaluating the performance of the wastewater treatment prediction model through the test set sample data; S6, predicting the water quality of the wastewater treatment system through the trained wastewater treatment prediction model.

[0016] Preferably, in step S1, determining the input quantity parameters and the predicted quantity parameters includes: The flow rate Q, pH, temperature T, dissolved oxygen DO, and influent chemical oxygen demand COD inf and influent solid suspended matter SS inf As input quantity parameter; The chemical oxygen demand (COD) of the effluent from the wastewater treatment system eff and influent solid suspended matter SS eff as a predictor parameter.

[0017] Preferably, in step S2, performing data normalization processing on the sample data includes: The sample data is normalized, and the normalization formula is as follows: ; In the formula, x max is the maximum value in the sample data, x min is the minimum value in the sample data; x is any value in the sample data, and X is the normalized value.

[0018] like Figure 2 As shown, preferably, in step S3, constructing a wastewater treatment prediction model based on a BP neural network includes: A three-layer BP neural network model is constructed, including an input layer, a hidden layer, and an output layer; the nodes in the previous layer and the next layer are connected by corresponding weights.

[0019] Furthermore, the BP neural network model includes signal forward transmission and error reverse propagation. Signal forward transmission is to transmit the input signal to the hidden layer and the output layer to obtain the predicted output, and then obtain the error signal from the predicted output and the expected output. The reverse propagation of the error is to feed the error signal back to the previous neural network, and then the neural network adjusts the corresponding weights and values ​​according to the size of the error. This process is carried out according to the error gradient descent method; The BP neural network first trains the frontmost layer of the network, fixes its weights and thresholds after the training is completed, and then uses the output of its hidden unit as the input vector of the next layer of the network to continue training; after all the networks are trained, the back propagation algorithm is used to repeatedly adjust the weights and thresholds until the error signal between the predicted output and the expected output meets the target requirements and the operation is terminated.

[0020] like Figure 3 As shown, preferably, in step S3, the wastewater treatment prediction model is optimized by an improved PSO algorithm, and the specific method is: S501, calculating the total number of network weights and thresholds, i.e., the length of particles, according to the number of nodes in each layer of the BP neural network; S502, encoding the weights and thresholds of the BP neural network into corresponding particles in a binary encoding manner; S503, initialize the particle swarm, including the population size N, the maximum number of iterations G, the position x of each particle i，j and speed v i，j , i represents the number of particles, j represents the dimension of particles; set the relevant parameters of particle swarm algorithm; S504, calculating the fitness of each particle in the initial particle swarm; S505, recording the individual extreme value and the global extreme value of the particle group respectively; S506, respectively update the velocity v of each particle in the current particle group i and position x i ; ; ; In the formula, V represents the updated speed of the particle, X represents the updated position of the particle, and V i，j and X i，j Assign to x i，j and v i，j , update the position and velocity of the particle; gbest is the individual optimal value, Bestx is the global optimal value, c 1 、c 2 is the learning factor, w is the adaptive inertia weight, is the maximum value of the adaptive inertia weight, is the minimum value of the adaptive inertia weight; the adaptive variable c 1Determine the optimization process to be biased towards local optimization, adaptive variables c2 Determine the optimization process to be biased towards the global optimum, adaptive inertia weight w Prefer global search; Adaptive variablesc 1 and c 2The calculation method is as follows: ; ; In the formula, G is the maximum number of iterations, and e is the remaining number of iterations; S507, calculating the fitness of each particle in the new particle group; S508, updating the individual extreme value of each particle; S509, updating the global extreme value of the particle swarm; S5010, if the end condition is met or the number of iterations reaches the maximum number of loop iterations, exit the loop, otherwise return to S506; S5011, output the optimal solution of the problem, and assign the optimal solution to the BP neural network for prediction.

[0021] Preferably, in step S3, the PSO algorithm is improved, including using a linear weight reduction method and introducing a mutation operator to improve the PSO algorithm, and the specific method is: Linearly reduce the weight coefficient: ; In the formula, is the maximum value of the adaptive inertia weight, is the minimum value of the inertia weight, t is the current number of iterations, and G is the maximum number of iterations; Introducing mutation operators: In the iterative optimization process, mutation operators are introduced, specifically: pos = unidrnd(numsum); if rand>0.95 pop(j,pos)=3*rands(1,1); Among them, pos is the position where the population mutates, unidrnd is a function that generates random integers, and numsum is the particle length.

[0022] Preferably, in step S5, when the performance of the wastewater treatment prediction model is evaluated by the test set sample data, the experimental root mean square error RMSE and the determination coefficient R 2 To evaluate the model performance, specifically: ; ; Among them, among them, is the predicted value of the test set sample data, is the true value of the test set sample data, is the average value of the test set sample data, and n is the number of test set sample data; Model optimization degree R m is defined as: ; in, is the model correlation coefficient before optimization, is the correlation coefficient of the optimized model.

[0023] Preferably, in step S6, predicting the water quality of the wastewater treatment system using the trained wastewater treatment prediction model includes: S601, BP neural network model parameter setting: The input layer of the BP neural network sets 6 input nodes, corresponding to 6 wastewater quality parameter variables Q, pH, T, DO, COD inf and SS inf ; The output layer of the BP neural network sets one node to predict COD eff and SS eff ; The number of hidden layer nodes of the BP neural network is set by the empirical value formula: ; Where q is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, and a∈{1,2,…,10} is the adjustment constant; In this experiment, the number of nodes in the input layer is 6, and the number of nodes in the output layer is 1. According to the experiment, when the number of nodes in the hidden layer of the network is set to 5, the performance of the network is better in all aspects, and the model structure is 6-5-1 structure. S602, wastewater treatment prediction model parameter setting: Based on the structure of BP neural network, the relevant parameters of the improved PSO algorithm are set as follows: Fitness function setting: The fitness value is defined as the absolute value of the total error of the optimized BP neural network structure. The formula is: ; In the formula, a is the predicted output of the BP neural network, and o is the true value of the sample output.

[0024] Embodiment 2: like Figure 4As shown in the figure, the results use the trained PSO-BP model to test the wastewater treatment system, and the fitness curve and simulation prediction results of the PSO-optimized BP neural network can be obtained. From the fitness curve of the model, it can be seen that PSO can effectively optimize the BP neural network, and the convergence speed of the network is significantly accelerated. When the evolutionary generation is 54 generations, the fitness has reached the optimal level. like Figure 5~Figure 6 It can be seen that the prediction results of the PSO-BP model are very close to the actual data, and the output curve of the model has a high degree of fit with the actual output curve. The PSO-BP model is used to predict COD eff and SS eff When , RMSE are 3.9955 and 0.6503 respectively, R 2 They are 0.6401 and 0.6811 respectively. It can be judged that the PSO-BP model has good generalization ability and can effectively predict the effluent quality of the wastewater treatment system.

[0025] Embodiment three: like Figure 7~Figure 10 As shown in the figure, in order to compare the prediction performance of the PSO-BP model, the GA-BP model and the BP model were used to calculate COD eff and SS eff Soft sensor modeling is carried out, and the network structure settings of the BP model and the GA-BP model are the same as the PSO-BP model. From the prediction simulation results of the BP, GA-BP, and PSO-BP models in Table 1, we can see that:

[0026] In predicting the effluent quality of wastewater treatment systems, PSO is more effective than GA in optimizing BP neural networks.

[0027] The performance of the PSO-BP model is the best: Compared with the BP model, the PSO-BP model predicts COD eff and SS eff R 2 The PSO-BP model improved by 4.49% and 40.11% respectively, and the RMSE decreased by 8.55% and 8.38% respectively. Compared with the GA-BP model, the PSO-BP model predicted COD eff and SS eff R 2 An increase of 0.44% and 24.77% respectively.

Claims

1. The wastewater treatment control strategy based on fuzzy neural network is characterized by: The following steps are involved: S1, Data Collection: Collect sample data from the wastewater treatment system and divide the sample data into a training set and a test set according to a fixed ratio; determine input quantity parameters and predicted quantity parameters; S2, data preprocessing: normalize the sample data; S3, constructing a wastewater treatment prediction model based on BP neural network, and optimizing the BP neural network structure of the wastewater treatment prediction model through PSO algorithm; wherein the PSO algorithm is improved; S4, training the wastewater treatment prediction model through the training set sample data after data preprocessing; S5, evaluating the performance of the wastewater treatment prediction model through the test set sample data; S6, predicting the water quality of the wastewater treatment system through the trained wastewater treatment prediction model.

2. The wastewater treatment control strategy based on fuzzy neural network according to claim 1 is characterized in that: In step S1, determining the input quantity parameters and the predicted quantity parameters includes: The flow rate Q, pH, temperature T, dissolved oxygen DO, and influent chemical oxygen demand COD inf and influent solid suspended matter SS inf As input quantity parameter; The chemical oxygen demand (COD) of the effluent from the wastewater treatment system eff and influent solid suspended matter SS eff as a predictor parameter.

3. The wastewater treatment control strategy based on fuzzy neural network according to claim 2 is characterized in that: In step S2, performing data normalization processing on the sample data includes: The sample data is normalized, and the normalization formula is as follows: ； In the formula, x max is the maximum value in the sample data, x min is the minimum value in the sample data; x is any value in the sample data, and X is the normalized value.

4. The wastewater treatment control strategy based on fuzzy neural network according to claim 3 is characterized in that: In step S3, constructing a wastewater treatment prediction model based on BP neural network includes: A three-layer BP neural network model is constructed, including an input layer, a hidden layer, and an output layer; the nodes in the previous layer and the next layer are connected by corresponding weights.

5. The wastewater treatment control strategy based on fuzzy neural network according to claim 4 is characterized in that: In step S3, the wastewater treatment prediction model is optimized by using the improved PSO algorithm. The specific method is as follows: S501, calculating the total number of network weights and thresholds, i.e., the length of particles, according to the number of nodes in each layer of the BP neural network; S502, encoding the weights and thresholds of the BP neural network into corresponding particles in a binary encoding manner; S503, initialize the particle swarm, including the population size N, the maximum number of iterations G, the position x of each particle i，j and speed v i，j , i represents the number of particles, j represents the dimension of particles; set the relevant parameters of particle swarm algorithm; S504, calculating the fitness of each particle in the initial particle swarm; S505, recording the individual extreme value and the global extreme value of the particle group respectively; S506, respectively update the velocity v of each particle in the current particle group i and position x i ; ； ； In the formula, V represents the updated speed of the particle, X represents the updated position of the particle, and V i，j and X i，j Assign to x i，j and v i，j , update the position and velocity of the particle; gbest is the individual optimal value, Bestx is the global optimal value, c 1 、c 2 is the learning factor, w is the adaptive inertia weight, is the maximum value of the adaptive inertia weight, is the minimum value of the adaptive inertia weight; Adaptive variables c 1Determine the optimization process to be biased towards local optimization, adaptive variables c2 Determine the optimization process to be biased towards the global optimum, adaptive inertia weight w Prefer global search; Adaptive variables c 1 and c 2The calculation method is as follows: ； ； In the formula, G is the maximum number of iterations, and e is the remaining number of iterations; S507, calculating the fitness of each particle in the new particle group; S508, updating the individual extreme value of each particle; S509, updating the global extreme value of the particle swarm; S5010, if the end condition is met or the number of iterations reaches the maximum number of loop iterations, exit the loop, otherwise return to S506; S5011, output the optimal solution of the problem, and assign the optimal solution to the BP neural network for prediction.

6. The wastewater treatment control strategy based on fuzzy neural network according to claim 5 is characterized in that: In step S3, the PSO algorithm is improved, including using a linear weight reduction method and introducing a mutation operator to improve the PSO algorithm. The specific method is: Linearly reduce the weight coefficient: ； In the formula, is the maximum value of the adaptive inertia weight, is the minimum value of the inertia weight, t is the current number of iterations, and G is the maximum number of iterations; Introducing mutation operators: In the iterative optimization process, mutation operators are introduced, specifically: pos = unidrnd(numsum); if rand>0.95 pop(j,pos)=3*rands(1,1); Among them, pos is the position where the population mutates, unidrnd is a function that generates random integers, and numsum is the particle length.

7. The wastewater treatment control strategy based on fuzzy neural network according to claim 1 is characterized in that: In step S5, when the performance of the wastewater treatment prediction model is evaluated by the test set sample data, the experimental root mean square error RMSE and the determination coefficient R 2 To evaluate the model performance, specifically: ； ； Among them, among them, is the predicted value of the test set sample data, is the true value of the test set sample data, is the average value of the test set sample data, and n is the number of test set sample data; Model optimization degree R m is defined as: ； in, is the model correlation coefficient before optimization, is the correlation coefficient of the optimized model.

8. The wastewater treatment control strategy based on fuzzy neural network according to claim 6 is characterized in that: In step S6, water quality prediction of the wastewater treatment system using the trained wastewater treatment prediction model includes: S601, BP neural network model parameter setting: The input layer of the BP neural network sets 6 input nodes, corresponding to 6 wastewater quality parameter variables Q, pH, T, DO, COD inf and SS inf ; The output layer of the BP neural network sets one node to predict COD eff and SS eff ; The number of hidden layer nodes of the BP neural network is set by the empirical value formula: ； Where q is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, and a∈{1,2,…,10} is the adjustment constant; S602, PSO algorithm parameter setting: Based on the structure of BP neural network, the relevant parameters of the improved PSO algorithm are set as follows: Fitness function setting: The fitness value is defined as the absolute value of the total error of the optimized BP neural network structure. The formula is: ； In the formula, a is the predicted output of the BP neural network, and o is the true value of the sample output.

Images