A neural network-based hydroxylamine synergizes with fe 2+ Process parameter optimization method for activated persulfate degradation of orange ii

Abstract

The application discloses a kind of hydroxylamine based on neural network cooperates Fe 2+ Process parameter optimization method for activated persulfate degrading orange yellow II, including on the basis of reaction of hydroxylamine cooperates Fe 2+ Activated persulfate, artificial neural network is used to hydroxylamine cooperates Fe 2+ Degradation of AO7 process is modeled by activated persulfate, and Garson algorithm and PaD2 algorithm are used for sensitivity analysis of neural network, and neural network-genetic algorithm coupling intelligent algorithm is constructed, and genetic algorithm is embedded in neural network for extremum optimization.The optimization result obtained by the genetic algorithm of the application is: Fe 2+ The concentration is 35.33 μmol·L ‑1 , the concentration of HA is 0.46 mmol·L ‑1 , the concentration of PS is 0.93 mmol·L ‑1 , and the degradation effect of AO7 is 95.7% within 5 minutes, which is only 0.5% different from the predicted value 96.2% of the model.

Description

Technical Field

[0001] This invention relates to the field of degradation technology of Orange II, specifically to a hydroxylamine-co-Fe degradation method based on neural networks. 2+ Optimization method for process parameters of activated persulfate degradation of Orange II. Background Technology

[0002] Dye wastewater, exemplified by the azo dye Orange II (AO7), is discharged in large quantities into environmental water bodies, posing a serious threat to human health by causing adverse effects such as skin allergies, respiratory difficulties, and carcinogenicity. Based on research into traditional wastewater treatment methods, scholars both domestically and internationally have continuously developed various processes for degrading and mineralizing AO7. Among these, advanced oxidation technologies based on persulfate (PS) have gradually gained widespread attention. The oxidizing power of sulfate free radicals (E0 = 2.5-3.1V) is significantly higher than that of single persulfate (S2O82-E0 = 2.1V, HSO5-E0 = 1.82V). Compared with other activation types, Fe... 2+ With its abundant reserves and low price, hydroxylamine (HA) has become the most commonly used transition metal activator. In recent years, hydroxylamine (HA) has been used in conjunction with Fe... 2+ The activation process of persulfate has attracted widespread attention, and hydroxylamine can accelerate the process of Fe... 2+ The formation of this slows down the accumulation of iron sludge.

[0003] Current information about Fe 2+ Research on activated persulfate mainly focuses on the reduction, recovery, reaction mechanism, and degradation efficiency of Fe3+, but there are some problems.

[0004] 1. Fe 2+ While existing methods for activating persulfate have addressed issues related to Fe3+ reduction, recovery, reaction mechanisms, and degradation efficiency to some extent, it remains difficult to find hydroxylamine-assisted Fe3+ reduction methods under different conditions. 2+ The research approach of activating persulfate to degrade AO7 has significant limitations;

[0005] 2. In existing technologies, persulfate can generate sulfate radicals (SO4·–) through activation reactions under heat, alkali, ultraviolet light, and transition metals, but in Fe... 2+ During the activation of persulfate, due to Fe 2+ The regeneration rate is slow, and iron sludge is generated during the reaction, which affects the generation of sulfate free radicals, resulting in a low degradation rate.

[0006] Therefore, this invention proposes a hydroxylamine-Fe synergistic method based on neural networks. 2+ Optimization method for process parameters of activated persulfate degradation of Orange II. Summary of the Invention

[0007] This invention addresses the problem of overly simplistic solutions in existing technologies by providing a significantly different approach. Specifically, the objective of this invention is to provide a neural network-based hydroxylamine-synergistic Fe... 2+ An optimization method for the process parameters of activated persulfate degradation of Orange II is proposed to address the aforementioned background technical issues. 2+ While existing methods for activating persulfate have addressed issues related to Fe3+ reduction, recovery, reaction mechanisms, and degradation efficiency to some extent, it remains difficult to find hydroxylamine-assisted Fe3+ reduction methods under different conditions. 2+ The research approach of activating persulfate to degrade AO7 has significant limitations.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a neural network-based hydroxylamine-synergistic Fe... 2+ Optimization method for process parameters of activated persulfate degradation of Orange II, with hydroxylamine synergistic Fe 2+ Based on the reaction of activating persulfate, an artificial neural network is used to synergistically activate hydroxylamine and Fe. 2+ The degradation process of AO7 by activated persulfate was modeled, and the sensitivity analysis of the neural network was performed using the Garson algorithm and the PaD2 algorithm. A coupled intelligent algorithm of neural network and genetic algorithm was constructed, and the genetic algorithm was embedded into the neural network for extreme value optimization, resulting in the hydroxylamine-Fe... 2+ Optimization method for optimal process parameters of activated persulfate degradation AO7 system.

[0009] The above-mentioned process parameter optimization method specifically includes the following steps:

[0010] Step 1: Establish a BP neural network model: A typical three-layer BP neural network is used, with HA concentration, Fe... 2+ The concentrations of Fe and PS were used as input variables, and the degradation effect of AO7 was used as the output variable. Based on single-factor experiments, Fe was obtained. 2+ The experimental variables were Fe, HA, and PS, and their coding levels were determined. The Box-Behnken design (BBD) method was used to study Fe at different concentrations. 2+ The predicted values ​​of AO7 degradation effect were obtained under the three variables HA and PS. The Premnmx function in Matlab was used to normalize the data. The influence of different numbers of hidden layer nodes, training functions and activation functions on the BP neural network was investigated by trial and error method to find the optimal BP neural network topology of 3 input layers, 11 hidden layers and 1 output layer.

[0011] Step 2: Sensitivity analysis of the established BP neural network model: Using the Garson algorithm, the influence of individual factors on the model response value is calculated through connection weights. This allows for local sensitivity analysis of the BP neural network connection weights, and the results show that the influence of each factor on the AO7 degradation effect is: HA concentration > Fe. 2+ Concentration > PS concentration; Using the PaD2 algorithm, the influence of the interaction between the two factors on the model response value was analyzed, and the influence of the interaction between the two factors on the AO7 degradation effect was determined as follows: Fe 2+ Concentration and PS concentration > HA concentration and PS concentration > Fe 2+ Concentration and HA concentration;

[0012] Step 3: Embed the established BP neural network into the genetic algorithm to form a hybrid intelligent algorithm, and derive Fe... 2 The optimal results for HA and PS.

[0013] Preferably, the formula for the Garson algorithm is:

[0014]

[0015] In the formula, Garson ik is the sensitivity coefficient of the i-th input variable to the k-th output variable; M is the number of neurons in the input layer; N is the number of neurons in the hidden layer; L is the number of neurons in the output layer; w ij The weights for the connection between the i-th neuron in the input layer and the j-th neuron in the hidden layer; v jk The weights are the connection weights between the j-th neuron in the hidden layer and the k-th neuron in the output layer.

[0016] Preferably, the formula for the PaD2 algorithm is:

[0017]

[0018] In the formula, d ik t For the factor x of the t-th sample i With x k Sensitivity coefficient to response value y; N is the number of neurons in the hidden layer; w ij The weights for the connection between the i-th neuron in the input layer and the j-th neuron in the hidden layer; w kj The weights for the connection between the k-th neuron in the input layer and the j-th neuron in the hidden layer; v j f′(net) represents the weights of the connection between the j-th neuron in the hidden layer and the neuron in the output layer. t f′(net) is the first-order partial derivative of the activation function of the output layer neuron; j tf″(net) represents the first-order partial derivative of the activation function of the hidden layer neurons; j t ) represents the second-order partial derivative of the activation function of the hidden layer neurons;

[0019] Then x i With x k The overall sensitivity coefficient to the response value y is:

[0020]

[0021] In the formula, Sd ik For x i With x k The overall sensitivity coefficient to the response value y; m is the total number of samples.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] 1) This invention is based on existing Fe 2+ Based on the process of activating persulfate, the Garson algorithm and PaD2 algorithm were used to perform sensitivity analysis on the neural network. The analysis results and the artificial neural network's response to hydroxylamine-induced Fe... 2+ The process of AO7 degradation by activated persulfate was modeled, and a neural network was constructed using a genetic algorithm coupled with an intelligent algorithm to understand the influence of various reaction conditions and their interactions on the model response value. Furthermore, the genetic algorithm was embedded into the neural network for extreme value optimization, yielding results showing the hydroxylamine-Fe... 2+ Optimal process conditions for the activation of persulfate degradation of AO7 system.

[0024] 2) Using the Garson algorithm based on neural network weights, the influence of each factor on the degradation effect of AO7 was determined to be: HA concentration > Fe 2+ Concentration > PS concentration. The PaD2 algorithm determined the degree of influence of the interaction between the two factors on the AO7 degradation effect as follows: Fe 2+ Concentration and PS concentration > HA concentration and PS concentration > Fe 2+ Concentration and HA concentration, thereby seeking Fe 2+ The optimal conditions for activating persulfate were determined to minimize the impact on sulfate radical generation, thereby achieving Fe... 2+ Optimization of process parameters for activated persulfate degradation of Orange II. Attached Figure Description

[0025] Figure 1a and Figure 1b This diagram illustrates the impact of the number of hidden layer neurons, activation function, and training function on the performance of a BP neural network.

[0026] Figure 2This is a diagram of the BP neural network structure.

[0027] Figure 3 The performance curve of the BP neural network;

[0028] Figure 4 The graph shows the linear fit of the BP neural network.

[0029] Figure 5a and Figure 5b This is a schematic diagram of the training results of the BP neural network model;

[0030] Figure 6a , 6b 6c represents the concentration of HA and Fe. 2+ Concentration interaction, HA concentration and PS concentration interaction, and PS concentration and Fe 2+ Schematic diagram of the effect of concentration interaction on AO7 degradation. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Comparative Example

[0033] The reaction was carried out at a constant stirring rate at a temperature of (20±1)℃, with a certain concentration of AO7 and Fe. 2+ The HA solution was thoroughly mixed, and a certain amount of PS was added to initiate the reaction. AO7 and Fe 2+ The concentrations of HA and PS can be found in Tables 1 and 2. The initial volume of the mixed solution was kept at 50 mL. After the reaction started, 1 mL of the reaction solution was extracted at 5 min and quenched with 1 mL of ethanol. The absorbance was measured using a UV-Vis spectrophotometer. All experiments were conducted under the background pH of the mixed solution. The AO7 degradation experiment was repeated 3 times, and the average and standard deviation of the 3 experiments were calculated. In addition, the absorbance of AO7 was measured at a wavelength of 485 nm using a UV-Vis spectrophotometer, and the concentration was converted using a standard curve.

[0034] Example

[0035] Based on the comparative examples, this example uses the method of the present invention to synergistically target hydroxylamine with Fe. 2+ The optimal process parameters for the activated persulfate degradation AO7 system were optimized, including the following steps:

[0036] Step 1

[0037] Establishing a BP neural network model: A BP neural network model was established using Matlab R2016b software. A BP neural network (Back Propagation Neural Network) is a multi-layer feedforward neural network and one of the most widely used models in artificial neural networks. It consists of one input layer, one or more hidden layers, and one output layer. This experiment uses a typical three-layer BP neural network, with HA concentration and Fe... 2+ Concentration and PS concentration were used as input variables, and AO7 degradation effect was used as the output variable. Based on the single-factor experiments, the experimental variables and their coding levels were obtained (Table 1). The Box-Behnken design (BBD) method was used to design the experiments, and the experimental design scheme is shown in Table 2. To avoid the influence of data dimensions, the data were dimensionless. The Premnmx function in Matlab was used to normalize the data. The data were randomly allocated according to the ratio of 70% training set, 15% validation set, and 15% test set. The influence of different numbers of hidden layer nodes, training functions, and activation functions on the BP neural network was investigated using a trial-and-error method to determine the optimal BP neural network structure. The root mean square error (RMSE) was used as the performance evaluation index of the neural network; a lower RMSE value indicates better neural network performance.

[0038] Table 1 Experimental variables and coding level table

[0039]

[0040] Table 2 Experimental scheme and BP neural network prediction values

[0041] Table 2Experimental scheme and predicted value of BP neural

[0042] network

[0043]

[0044]

[0045] The structure of the BP neural network is determined as follows:

[0046] The number of hidden layer neurons affects the convergence performance of the error function, and thus the accuracy of the BP neural network. The number of hidden layer neurons is estimated to be 3-12 using the empirical equation (Equation 4). The number of hidden layer neurons is determined through trial and error, and the results are as follows: Figure 1aAs shown.

[0047]

[0048] In the formula, N is the number of hidden layer neurons; A is the number of input layer neurons; B is the number of output layer neurons; and C is a positive integer between 1 and 10.

[0049] In backpropagation (BP) neural networks, the activation functions of the hidden layers are generally sigmoid functions, which are divided into logarithmic (logsig) and hyperbolic tangent (tansig) functions. The activation function of the output layer is generally a linear function (purelin). This experiment examines the impact of two activation function combinations, logsig+purelin and tansig+purelin, on the performance of BP neural networks. The results are as follows: Figure 1a As shown. This invention compares the impact of 11 training functions, including gradient descent (trainingd), elastic backpropagation (trainrp), and Levenberg-Marquardt (trainlm), on the performance of backpropagation neural networks. The results are as follows. Figure 1b As shown. Through analysis and comparison, the final BP neural network topology adopted in this invention is 3-11-1, with the activation functions for the hidden layer and output layer being the tansig function and the purelin function, respectively, and the training function being trainlm. The optimal BP neural network structure is as follows. Figure 2 As shown.

[0050] The performance evaluation of the BP neural network is as follows:

[0051] Figure 3 The graph shows the performance curves of the BP neural network. A smaller MSE value indicates higher accuracy in predicting data. The graph shows that the error on the validation set decreases as the error on the training set decreases, without any increase. The validation set error reaches its minimum of 0.0025177 after the fifth iteration of the BP neural network, indicating that the BP neural network did not overfit during training. The error on the test set remains stable and lower than the validation set error after three iterations, demonstrating that the established BP neural network model has good generalization ability.

[0052] Figure 4This is a linear fit graph of the BP neural network. A larger R² value indicates a higher fit. The R² value for the training set is 0.99985, meaning the model can explain 99.985% of the response value variations. The R² values ​​for the validation set are 0.99628, the test set is 0.99660, and the all-set set is 0.99852. The data distribution is near the straight line y = x, indicating a small error between the measured and predicted values. The BP neural network exhibits excellent predictive ability and a non-linear mapping relationship.

[0053] The final predicted values ​​of the trained BP neural network are shown in Table 2. The model training results are as follows: Figure 5a , 5b As shown. From Figure 5a It can be seen that the error between the predicted and measured values ​​of the BP neural network is very small. From... Figure 5b It can be seen that the error is mainly concentrated near the zero point. The results show that the predicted values ​​of the BP neural network model are in excellent agreement with the measured values, verifying the reliability and accuracy of the selected BP neural network model. This indicates that the BP neural network can model the hydroxylamine-assisted Fe2+ activation of persulfate process.

[0054] Step Two

[0055] Sensitivity analysis was performed on the established BP neural network model: Sensitivity analysis, i.e., assuming the model is as follows:

[0056] y = f(x1, x2, ..., x n );

[0057] In the formula (xi is the value of the i-th factor in the model), the influence of each factor on the model response value when it varies within a possible range is studied. The magnitude of the influence on the factor is called the sensitivity coefficient. The larger the sensitivity coefficient, the greater the influence of the factor on the model response value. Based on the object of analysis, sensitivity analysis is divided into local sensitivity analysis and global sensitivity analysis. Local sensitivity analysis studies the influence of a single factor on the model response value, while global sensitivity analysis studies the influence of multiple factors on the model response value simultaneously, and analyzes the influence of the interaction between factors on the model response value. The Garson algorithm is a local sensitivity analysis method based on the connection weights of a neural network, calculating the influence of a single factor on the model response value through the connection weights. The PaD2 algorithm analyzes the influence of the interaction between two factors on the model response value. Assuming the BP neural network topology is MN-1, the network output is as follows:

[0058] y = f(x1, x2, ..., x n );

[0059] The influence of the interaction between the two factors on the model response value is analyzed by calculating the second-order partial derivatives of the equation. This invention uses the Garson algorithm (Equation 1) to perform local sensitivity analysis on the model and the PaD2 algorithm (Equation 2) to perform global sensitivity analysis on the model.

[0060]

[0061] In the formula, Garson ik is the sensitivity coefficient of the i-th input variable to the k-th output variable; M is the number of neurons in the input layer; N is the number of neurons in the hidden layer; L is the number of neurons in the output layer; w ij The weights for the connection between the i-th neuron in the input layer and the j-th neuron in the hidden layer; v jk The weights are the connection weights between the j-th neuron in the hidden layer and the k-th neuron in the output layer.

[0062]

[0063] The above formula is "(Equation 2)".

[0064] In the formula, d ik t For the factor x of the t-th sample i With x k Sensitivity coefficient to response value y; N is the number of neurons in the hidden layer; w ij The weights for the connection between the i-th neuron in the input layer and the j-th neuron in the hidden layer; w kj The weights for the connection between the k-th neuron in the input layer and the j-th neuron in the hidden layer; v j f′(net) represents the weights of the connection between the j-th neuron in the hidden layer and the neuron in the output layer. t f′(net) is the first-order partial derivative of the activation function of the output layer neuron; j t f″(net) represents the first-order partial derivative of the activation function of the hidden layer neurons; j t ) represents the second-order partial derivative of the activation function of the hidden layer neurons.

[0065] Then x i With x k The overall sensitivity coefficient (Equation 3) to the response value y is:

[0066]

[0067] In the formula, Sd ik For x i With x k The overall sensitivity coefficient to the response value y; m is the total number of samples;

[0068] Step 3

[0069] A hybrid intelligent algorithm is formed by embedding a BP neural network into a genetic algorithm.

[0070] The sensitivity analysis of the BP neural network is shown below:

[0071] Table 3 lists the weights and thresholds of the BP neural network model. Using the Garson algorithm based on the neural network weights, the influence of each factor on the AO7 degradation effect was determined to be: HA concentration (39.6%) > Fe. 2+ Concentration (32.4%) > PS concentration (28%). The PaD2 algorithm determined the degree of influence of the interaction between the two factors on the AO7 degradation effect as follows: Fe 2+ The concentrations of Fe2+ and PS (7.54) > HA and PS (5.97) > Fe2+ and HA (2.21).

[0072] Table 3 shows the weights and thresholds of the BP neural network.

[0073]

[0074] Among them, Fe 2+ The effects of concentrations of HA, PS, and HA on the degradation efficiency of AO7 are as follows:

[0075] Fe was obtained based on the BP neural network model. 2+ The nonlinear mapping relationship between concentrations of HA, PS, and AO7 degradation efficiency was plotted using Origin 2018 software to create three-dimensional surface plots, as shown in Figure 6. Each plot only shows the effect of the interaction of two factors on the model response value, while other factors remain at the central level.

[0076] Depend on Figure 6a and Figure 6b It is known that increasing the HA concentration improves the degradation effect of AO7. Sufficient HA can quickly degrade Fe... 3+ Reduced to Fe 2+ This ensures that the system has sufficient Fe. 2+ Activation of PS produces SO4 ·_ Degradation of AO7. (By...) Figure 6a and Figure 6c It can be seen that increasing Fe 2+ Concentration will enhance the degradation effect of AO7. Fe 2+ Increasing the concentration can activate PS to produce more SO4. ·_ This also improves the degradation effect of AO7.

[0077] Depend on Figure 6b It is known that when HA is at a low concentration, increasing PS concentration will reduce the degradation effect of AO7. Excess PS in the reaction system will compete with the target pollutant for SO4. ·_ (Equation 5) leads to a decrease in the degradation effect of AO7. When HA is at a high concentration, increasing the PS concentration will improve the degradation effect of AO7. This further reveals the interaction between HA concentration and PS concentration.

[0078] HSO5 - +SO4 ·- →SO4 2- +SO5 ·- +H + (Equation 5)

[0079] Depend on Figure 6c It can be seen that when Fe 2+ At low concentrations, increasing the PS concentration reduces the degradation effect of AO7. When Fe... 2+ At higher concentrations, increasing the PS concentration enhances the degradation of AO7. This further reveals the role of Fe... 2+ The interaction between concentration and PS concentration is significant.

[0080] Step 3

[0081] A hybrid intelligent algorithm is formed by embedding a BP neural network into a genetic algorithm, and Fe is derived. 2 The optimal results for HA and PS are as follows: Genetic Algorithm (GA) is a process of seeking optimal solutions by simulating the biological evolution theory in nature. It is widely used in solving complex global optimization problems and has high robustness. Currently, the combination of neural networks fitting uncertain nonlinear functions and embedding them into genetic algorithms to form hybrid intelligent algorithms has been successfully applied in environmental optimization. The global optimization of the genetic algorithm is performed with the goal of achieving the optimal degradation effect of AO7. The optimization result obtained is: Fe 2+ The concentration was 35.33 μmol·L⁻¹ 1 The HA concentration was 0.46 mmol·L⁻¹. 1 The PS concentration was 0.93 mmol·L⁻¹. 1 Experiments verified that the degradation rate of AO7 within 5 minutes was 95.7%, only 0.5% different from the model's predicted value of 96.2%. This indicates that the HA concentration and Fe concentration obtained by combining a BP neural network with a genetic algorithm are... 2+ Ideal AO7 degradation can be achieved under both high and low PS concentrations, with relatively small errors. Therefore, a backpropagation neural network combined with a genetic algorithm can be used for HA / Fe 2+ Parameter optimization for AO7 degradation in the / PS system.

[0082] Conclusions: 1) The final BP neural network topology adopted in this invention is 3-11-1, with the activation functions of the hidden layer and output layer being the tansig function and the purelin function, respectively, and the training function being trainlm. The established BP neural network model has an R² of 0.99852, and the data distribution is near the straight line y = x. The error between the predicted and measured values ​​is very small. The results show that the predicted values ​​based on the BP neural network model agree well with the measured values, and also demonstrate that the BP neural network can effectively target hydroxylamine and Fe... 2+ Modeling the activation process of persulfate.

[0083] 2) Using the Garson algorithm based on neural network weights, the influence of each factor on the degradation effect of AO7 was determined to be: HA concentration > Fe 2+ Concentration > PS concentration. The PaD2 algorithm determined the degree of influence of the interaction between the two factors on the AO7 degradation effect as follows: Fe 2+ Concentration and PS concentration > HA concentration and PS concentration > Fe 2+ Concentration and HA concentration.

[0084] 3) Optimization results obtained through genetic algorithm: Fe 2+ The concentration was 35.33 μmol·L. -1 The HA concentration was 0.46 mmol·L⁻¹. -1 The PS concentration was 0.93 mmol·L⁻¹. -1 Experiments verified that the degradation rate of AO7 within 5 minutes was 95.7%, which was only 0.5% different from the model's predicted value of 96.2%.

[0085] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing process parameters of hydroxylamine-assisted Fe2+ activation persulfate degradation of Orange II based on neural networks, characterized in that: Based on the reaction of hydroxylamine synergistic Fe2+ activation of persulfate, an artificial neural network was used to model the degradation process of AO7 by hydroxylamine synergistic Fe2+ activation of persulfate. The sensitivity analysis of the neural network was carried out using the Garson algorithm and the PaD2 algorithm. A neural network-genetic algorithm coupled intelligent algorithm was constructed, and the genetic algorithm was embedded in the neural network to perform extreme value optimization. The optimal process parameter optimization method of the hydroxylamine synergistic Fe2+ activation of persulfate degradation of AO7 system was obtained. The process parameter optimization method specifically includes the following steps: Step 1: Establish a BP neural network model: A typical three-layer BP neural network is used, with HA concentration, Fe... 2+ The concentrations of Fe and PS were used as input variables, and the degradation effect of AO7 was used as the output variable. Based on single-factor experiments, Fe was obtained. 2+ Three experimental variables, HA, and PS, and their coding levels; the Box-Behnken design method was used to study Fe at different concentrations. 2+ The predicted values ​​of AO7 degradation effect were obtained under the three variables HA and PS. The Premnmx function in Matlab was used to normalize the data. The influence of different numbers of hidden layer nodes, training functions and activation functions on the BP neural network was investigated by trial and error method to find the optimal BP neural network topology of 3 input layers, 11 hidden layers and 1 output layer. Step 2: Sensitivity analysis of the established BP neural network model: Using the Garson algorithm, the influence of individual factors on the model response value is calculated through connection weights. This allows for local sensitivity analysis of the BP neural network connection weights, and the results show that the influence of each factor on the AO7 degradation effect is: HA concentration > Fe. 2+ Concentration > PS concentration; Using the PaD2 algorithm, the influence of the interaction between the two factors on the model response value was analyzed, and the influence of the interaction between the two factors on the AO7 degradation effect was determined as follows: Fe 2+ Concentration and PS concentration > HA concentration and PS concentration > Fe 2+ Concentration and HA concentration; Step 3: Embed the established BP neural network into the genetic algorithm to form a hybrid intelligent algorithm, and derive Fe... 2 The optimal results for HA and PS.

2. The method for optimizing process parameters of hydroxylamine-assisted Fe2+ activation persulfate degradation of Orange II based on neural networks according to claim 1, characterized in that: The formula for the Garson algorithm is: ； In the formula, Garson ik Let be the sensitivity coefficient of the i-th input variable to the k-th output variable; M is the number of neurons in the input layer; N is the number of neurons in the hidden layer; w ij The weights for the connection between the i-th neuron in the input layer and the j-th neuron in the hidden layer; v jk The weights are the connection weights between the j-th neuron in the hidden layer and the k-th neuron in the output layer.

3. The method for optimizing process parameters of hydroxylamine-assisted Fe2+ activation persulfate degradation of Orange II based on neural networks according to claim 1, characterized in that: The formula for the PaD2 algorithm is as follows: ； In the formula, d ik t For the factor x of the t-th sample i With x k Sensitivity coefficient to response value y; N is the number of neurons in the hidden layer; w ij The weights for the connection between the i-th neuron in the input layer and the j-th neuron in the hidden layer; w kj The weights for the connection between the k-th neuron in the input layer and the j-th neuron in the hidden layer; v j f'(net) represents the weights of the connection between the j-th neuron in the hidden layer and the neuron in the output layer. t f'(net) is the first-order partial derivative of the activation function of the output layer neuron; j t ) represents the first-order partial derivative of the activation function of the hidden layer neurons; f ″(net j t ) represents the second-order partial derivative of the activation function of the hidden layer neurons; Then x i With x k The overall sensitivity coefficient to the response value y is: ； In the formula, Sd ik For x i With x k The overall sensitivity coefficient to the response value y; m is the total number of samples.

Images