A method for determining the transient stability limit transmission power based on CNN

Abstract

A method for determining the transient stability limit transmission power based on CNN. First, a simulation model is constructed through historical data, and then simulation experiments are carried out through Monte Carlo fault simulation to obtain data such as system voltage before and after the accident, fault impact power of the tie line, and transient stability limit transmission power. Then, a convolutional neural network is used to study the non-linear mapping relationship between the system input characteristic quantities of the power grid and the transient stability limit transmission power of the transmission channel, and a convolutional network model is constructed through multiple trainings. Finally, this non-linear mapping relationship is used to realize the rapid calculation of the transient stability limit power. The convolutional neural network proposed by the present invention for determining the transient stability limit power has a faster calculation speed, higher accuracy and stronger model adaptability compared with the direct method and the time-domain simulation method.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and particularly to a method for determining the transient stability limit transmission power based on CNN. Background Art

[0002] The transient stability transmission power limit is an important indicator for measuring the stability of a power system. In the transient stability analysis and security control analysis of a power system, it is often necessary to calculate this critical parameter of the system. The estimation and judgment of the transient stability limit transmission power are an important basis for arranging the operation mode of the power system and grasping the transient stability ability of the system. Obtaining it is beneficial to provide valuable information for the safe and economic dispatching and preventive control of the system. At present, there are mainly two types of traditional methods for obtaining the transient stability transmission power limit of a tie line: the time-domain simulation method and the direct method based on Lyapunov stability theory. With the development of artificial intelligence, the realization of obtaining the transient stability limit power by neuron technology has also been achieved.

[0003] The time-domain simulation method forms a full-system model by connecting the models of all components in the whole system through the power network, and then uses the integration method to obtain the curves of the state variables and algebraic variables of the whole system changing with time. The system stability is judged by the relative swing curve between the rotors of synchronous generators. The time-domain simulation method is a process of repeated trial and error. The above process is repeated until the above interval is small enough. At this time, the midpoint of this interval is the limit value. Generally, multiple integrations are required to obtain a limit value, and its computational workload is obvious. However, the estimation result of the trial-and-error method is the most accurate among all current methods. The patent application with the patent application number 201010512896.3 and the name "A Determination Method for Rapidly Obtaining the Transient Stability Limit Transmission Power of Tie Lines" is calculated based on time-domain simulation software. The time-domain simulation method can adopt accurate mathematical models of each component and can conveniently take into account the effects of various regulators on the transient stability of the system. Therefore, it has high accuracy and good numerical stability. The disadvantage is that in the analysis of the stability limit, it is necessary to approach the stability limit through repeated simulations, so the calculation is time-consuming and the stability index analysis cannot be carried out. The direct method usually adopts the sensitivity analysis method of transient stability margin. The direct method uses the energy margin as the transient stability margin and can quantitatively analyze the stability of the system. The energy margin is an estimate of the system stability from the perspective of energy. It was initially obtained from the transient stability analysis of the direct method, and the magnitude of its value reflects the degree to which the system maintains stability. If the energy margin at the fault clearing time is greater than zero, the system tends to be stable after the fault; on the contrary, if the energy margin is less than zero, the system is unstable after the fault. Therefore, the fault clearing time corresponding to the energy margin equal to zero is the critical clearing time, and the output of the generator corresponding to it is the critical output of the generator under this fault and this clearing time. With its fast calculation speed, the direct method has good prospects in the online application of transient security analysis. However, it still cannot get rid of the inherent defects of the direct method itself - the conservativeness of the direct method and the limitations of the model. Using the direct method for transient stability limit analysis is restricted by the development status of the direct method, and the corresponding calculation accuracy and speed are closely related to the calculation accuracy and speed of the direct method.

[0004] The traditional neural network is used to realize the nonlinear mapping relationship among the transient energy margin, the generator output power, and the fault clearing time, and to realize the rapid and accurate solution of the transient stability limit power. Its input only considers the influence of the fault clearing time on the transient stability limit power, and it is difficult to realize the transient stability limit power of the system under the influence of multiple operating modes and multiple system faults. Summary of the Invention

[0005] To overcome the deficiencies of the above-mentioned existing calculations, the object of the present invention is to provide a method for determining the transient stability limit transmission power based on CNN. By using data such as system voltages before and after accidents, fault impact power of tie lines, and transient stability limit transmission power obtained from historical data or simulation experiments, a deep learning convolutional neural network is used to study the non-linear mapping relationship between the system input characteristic quantities of the power grid and the transient stability limit transmission power of the transmission channel, and the precise estimation of the transient stability limit power is obtained by using this non-linear mapping relationship; using a convolutional neural network to determine the transient stability limit power has a faster calculation speed, higher accuracy and stronger model adaptability compared with the direct method and the time-domain simulation method.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A method for determining the transient stability limit transmission power based on CNN, comprising the following steps:

[0008] (1), Random fault simulation is carried out through PSAT simulation software and Monte Carlo simulation method, and transient stability limit analysis is carried out. The power system analysis software is used to simulate the state of the system when a transient stability fault occurs, and the characteristic values related to the transient stability limit of the transmission line in the system are exported from the simulation software, including: system power generation output, line impedance, node load, node voltage and voltage phase angle; then through equivalent parameter calculation, the static stability limit P of the system is obtained s 、The static stability limit transmission power P of the transmission line steady 、The maximum power P of the transmission line max And the maximum impact power ΔP;

[0009] (2), The characteristic values of the transient stability limit of the transmission line, the fault clearing time, and the fault type are processed as the input part of constructing the training data set, and the output part is the transient stability limit transmission power obtained by the above formula for each line;

[0010] (3), Establish a convolutional neural network model structure;

[0011] (4), Input the constructed training set into the convolutional network model, and train it in the way of grouped convolution and multi-channel, learn and optimize the weight coefficients of the model, and obtain the result of the transient stability limit of the transmission line through training.

[0012] The specific content of the step (1) is as follows:

[0013] (1.1) Use the power system analysis software PSAT for transient stability analysis. First, set the system under typical operating conditions, and then conduct fault simulations for transient stability limits. Based on the initial power flow, by adjusting the generator power and load values, at 75%-125% of the base load, with a step of 5%, set different load growth patterns for each load level, and set various different power generations for different load growth patterns. Under each power flow pattern, use the Monte Carlo method to select faults, and use the power system analysis software to simulate the state of the system when a transient stability fault occurs. Export the eigenvalues related to the transient stability limit of the transmission line in the system from the simulation software, including: system power generation, line impedance, node load, node voltage, and voltage phase angle;

[0014] (1.2) Through equivalent parameter calculation, based on the transient equivalent model of the system, calculate the static stability limit P of the transmission line transmission power s ;

[0015]

[0016] In formula (1), U 1 and U 2 are the equivalent voltages on both sides of the transmission line, and ∑Z is the total impedance of the system;

[0017] Use the initial power P steady read from the system simulation, and obtain the maximum power P max of the system transmission line after the accident from the software simulation. At this time, measure the maximum impact power ΔP through the Monte Carlo simulation method:

[0018] ΔP = P max -P steady (2)

[0019] Calculate the transient stability limit transmission power of the transmission line based on the following formula (3), which is derived based on the relationship between the static stability limit and the transient stability limit of the transmission line transmission power:

[0020] P t = P s -ΔP (3)

[0021] In formula (3), P t is the transient stability limit transmission power, P s is the static stability limit of the transmission line transmission power, and ΔP is the maximum impact power;

[0022] According to the relationship between the static stability limit and the transient stability limit of the transmission line transmission power, use the static stability limit power of the transmission line minus the impact power ΔP to preliminarily calculate the transient stability limit power of the line as the input part of the training set:

[0023] ΔP = P s - P t (5)

[0024] Obtain the static stability limit P of the transmission power of the transmission line s , the static stability limit transmission power P of the transmission line steady , the maximum power P of the transmission line max and the maximum impact power ΔP;

[0025] Study the non - linear mapping relationship between the system input characteristic quantity of the power grid and the transient stability limit transmission power of the transmission channel, and use a deep learning model to quickly obtain the transient stability limit power of the transmission line:

[0026] P t = F(V m , θ m , P n , Q n ), (V m , θ m , P, Q n ∈R d ) (4)

[0027] In formula (4), P t is the transient stability limit transmission power, m is the bus node number, n is the transmission line number, d is the number of sampling points, V is the node voltage amplitude, θ is the node voltage phase angle, P is the active power of the line, Q is the reactive power of the line, R d is a two - dimensional matrix.

[0028] The construction of the training data set in step (2) is specifically as follows:

[0029] (2.1), Delete the error data from the simulation data in the input part and supplement the missing data; After standardizing the sorted data, put it into the data set;

[0030] (2.2), Each fault condition under each operation mode is divided into a training sample, and the ratio of the learning set to the test set is 8:2;

[0031] (2.3) Expand the dimension of the data sample and convert it into a two - dimensional data format to keep the sample data and the sample label synchronized.

[0032] In step (3), establish a convolutional neural network model structure; This model structure includes a convolutional layer, a pooling layer, a fully - connected layer, and an identity function output layer, specifically:

[0033] (3.1) The convolutional layer performs feature non - linear mapping. For the input data, the first convolutional layer uses convolutional kernels of multiple sizes, and the padding method is used during the convolution process;

[0034] (3.2) Perform max pooling operation to reduce the scale of data, improve the system robustness. The fully connected layer and the output layer are used for regression fitting;

[0035] (3.3) The third convolutional layer further extracts data features;

[0036] (3.4) The fourth convolutional layer with multiple convolutional kernels and the activation function ReLU;

[0037] (3.5) The fifth pooling layer continues the max pooling operation to remove redundant data;

[0038] (3.6) Perform Fatten operation on the data to stretch the high-dimensional data into one-dimensional;

[0039] (3.7) The stretched data is input into the fully connected layer, and the output result is the transient stability limit power value of the transmission line. Verify the calculated result data.

[0040] The method for training the convolutional neural network model in step (4) specifically includes the following steps:

[0041] (4.1) The training parameters of the convolutional neural network include the weights ω and biases b between the connection layers. The random gradient descent method is used in the training process, and the backpropagation algorithm is used for parameter optimization;

[0042] (4.2) Initialize the weights ω and biases b by using the method of random initialization;

[0043] (4.3) Import the data set into the convolutional neural network, and the data features are propagated forward through each layer, where X is the input part of the training set and Y is the output part of the training set;

[0044] (4.4) Calculate the error ε between the test set output and the actual output;

[0045] (4.5) Take minimizing the training error as the optimization goal of the network weight parameters;

[0046] (4.6) Gradually train to improve the model accuracy.

[0047] Advantages of the present invention:

[0048] The present invention adopts the method of deep learning, constructs a transient stability limit calculation model based on convolutional neural network. Through the combination of traditional simulation and neural network, and based on the data regression of one-dimensional convolutional neural network, it fully excavates the non-linear mapping relationship between the system input feature quantities of the power grid and the transient stability limit transmission power of the transmission channel. By training the model, it completes the rapid calculation of the transient stability limit transmission power of the transmission line, and realizes the rapid calculation of the transient stability limit of the transmission channel under the influence of faults. Description of the Drawings

[0049] Figure 1 This is the flow chart for calculating the transient stability limit transmission power of the present invention.

[0050] Figure 2 This is the hierarchical diagram for constructing the convolutional neural network model of the present invention.

[0051] Figure 3 This is the prediction error graph of the present invention under 150 epochs. Detailed implementation manners

[0052] The present invention will be further described below in conjunction with the accompanying drawings and examples.

[0053] Figure 1 A method for determining the transient stability limit transmission power based on CNN of the present invention includes the following steps.

[0054] Step 1: Perform random fault simulation through the PSAT simulation software and the Monte Carlo simulation method, and perform simulation operation on the system to be calculated under typical operating modes. On the basis of the initial power flow, by adjusting the generators and loads, at 75%-125% (with a step of 5%) of the base load, set several different load growth modes for each load level, and set six different power generations for different load growth modes. After obtaining the initial data, organize it, and calculate the static stability limit of the transmission power of the transmission line on the basis of the system's transient equivalent model through equivalent parameter calculation.

[0055] The specific content of the said step (1) is as follows:

[0056] (1.1) Use the power system analysis software PSAT to perform transient stability analysis. First, set the system under typical operating modes, and then perform fault simulation of the transient stability limit. On the basis of the initial power flow, by adjusting the generator power and load value, at 75%-125% of the base load, with a step of 5%, set different load growth modes for each load level, and set various different power generations for different load growth modes; under each power flow mode, use the Monte Carlo method to select faults, and select the faults to be simulated specifically in combination with the fault probability under the actual operating conditions of the system; in this way, the randomness and authenticity of the fault simulation are realized. Use the power system analysis software to simulate the state of the system when a transient stability fault occurs, and export the characteristic values related to the transient stability limit of the transmission line in the system from the simulation software, including: system power generation, line impedance, node load, node voltage, voltage phase angle;

[0057] (1.2) Calculate the static stability limit P of the transmission power of the transmission line on the basis of the system's transient equivalent model through equivalent parameter calculation. s ;

[0058]

[0059] In formula (1), U 1 and U 2 are the equivalent voltages on both sides of the transmission line, and ∑Z is the total impedance of the system;

[0060] Using the initial power P read by system simulation steady , the maximum power P of the system transmission line after the accident is obtained from software simulation max . At this time, the maximum impact power ΔP is measured through fault simulation by the Monte Carlo simulation method:

[0061] ΔP = P max - P steady (2)

[0062] Based on the following formula (3), calculate the transient stability limit transmission power of the transmission line. This formula is derived based on the relationship between the static stability limit and the transient stability limit of the transmission line transmission power:

[0063] P t = P s - ΔP (3)

[0064] In formula (3), P t is the transient stability limit transmission power, P s is the static stability limit of the transmission line transmission power, and ΔP is the maximum impact power.

[0065] According to the relationship between the static stability limit and the transient stability limit of the transmission line transmission power, use the static stability limit power of the transmission line minus the impact power ΔP to preliminarily calculate the transient stability limit power of the line, which is used as the input part of the training set:

[0066] ΔP = P s - P t (5)

[0067] Obtain the static stability limit P of the transmission line transmission power s , the static stability limit transmission power P of the transmission line steady , the maximum power P of the transmission line max and the maximum impact power ΔP;

[0068] The above formula is based on the connection between the transient stability limit power and the static stability power. In order to ensure that the transient stability limit transmission power has good applicability under different operating conditions, the most severe fault is generally selected when choosing fault simulation.

[0069] Study the non - linear mapping relationship between the system input characteristic quantities of the power grid and the transient stability limit transmission power of the transmission channel, and use a deep learning model to quickly calculate the transient stability limit power of the transmission line:

[0070] P t = F(V m , θ m , P n , Q n )(V m , θ m , P n , Q n ∈R d )(4)

[0071] In formula (4), P t is the transient stability limit transmission power, m is the bus node number, n is the transmission line number, d is the number of sampling points, V is the node voltage amplitude, θ is the node voltage phase angle, P is the line active power, Q is the line reactive power, and the input feature X corresponding to a single sample is a two - dimensional matrix in the form of R d . Construct the input vector of the neural network from the data obtained from the simulation system and solve it using a convolutional neural network.

[0072] (2), Process the eigenvalue of the transient stability limit of the transmission line, the fault clearing time, and the fault type as the input part of constructing the training data set, and the output part is the transient stability limit transmission power of each line obtained from the above formula.

[0073] The construction of the training data set in step (2) is specifically as follows:

[0074] (2.1), Delete the error data from the simulation data in the input part and supplement the missing data; standardize the sorted data and put it into the data set;

[0075] (2.2), Divide each fault condition under each operation mode into a training sample, and the ratio of the learning set to the test set is 8:2;

[0076] (2.3) Expand the dimension of the data sample and convert it into a two - dimensional data format to keep the sample data and the sample label synchronized;

[0077] (3), Establish the convolutional neural network model structure; use Keras as the modeling environment to model the convolutional neural network; among them, perform feature non - linear mapping in the convolutional layer, use the pooling layer to reduce the scale of the data, improve the system robustness, and use the fully - connected layer and the output layer for regression fitting; use convolutional kernels of multiple sizes instead of single - size convolutional kernels in the convolutional structure, and use grouped convolution instead of conventional convolution.

[0078] Figure 2It is the hierarchical diagram for constructing the convolutional neural network model of the present invention. In step (3), a convolutional neural network model structure is established; this model structure includes a convolutional layer, a pooling layer, a fully connected layer, and an identity function output layer, specifically as follows:

[0079] (3.1) For a data input of 20×8, the first convolutional layer uses 32 4×4 convolutional kernels, and the data features after convolution are 32×20×8; the convolutional layer performs feature non-linear mapping. For the data input, the first convolutional layer uses convolutional kernels of multiple sizes, and the Padding method is used during the convolution process.

[0080] (3.2) The second pooling layer performs a 2×2 max pooling operation, and the features after pooling are 32×10×4; a max pooling operation is performed to reduce the scale of the data, improve the robustness of the system, and the fully connected layer and the output layer are used for regression fitting.

[0081] (3.3) The third convolutional layer uses a 4×4 convolutional kernel to further extract the data features; the data features after convolution are 32×10×4.

[0082] (3.4) The fourth convolutional layer uses 3×3 convolutional kernels, multiple convolutional kernels, and the activation function is ReLU; the activation function is ReLU.

[0083] (3.5) The fifth pooling layer continues with a 2×2 max pooling operation; redundant data is removed.

[0084] (3.6) A Fatten operation is performed on the data to stretch the high-dimensional data into one-dimensional data.

[0085] (3.7) The stretched data is fed into the fully connected layer, and the output result is the transient stability limit power value of the transmission line, and the calculated result data is verified.

[0086] (4) Input the constructed training set into the convolutional network model, and train it in the way of grouped convolution with multiple channels. The weight coefficients of this model are learned and optimized, and the result of the transient stability limit of the transmission line is obtained through training.

[0087] Specifically: Standardize the data set, convert the format of the initial data, record and store all the data of the fault simulation, and on this basis, perform data processing to delete incorrect data and supplement missing data to complete the establishment of the training set; Use MinMaxScaler to transform the training set to be between [-1, 1] as the training set. Input the training set constructed in step 4 into the convolutional neural network model in step 3, input the training samples into the model, and learn and optimize the parameters of the convolutional neural network model; The transient stability limit calculation model can be obtained through repeated training; Among them, the neural network weights ω and biases b and the activation function, the activation function uses the ReLU activation function, and the model optimization method uses the method of stochastic gradient descent to gradually optimize the parameters of the network. After 150 rounds of training, the model will output the calculation result, and perform fitting verification on the settlement result. The sampling result is as Figure 3 shown. Specifically, it includes the following steps:

[0088] (4.1) The training parameters of the convolutional neural network include the weights ω and biases b between the connection layers. The training process uses the stochastic gradient descent method and uses the backpropagation algorithm to optimize the parameters;

[0089] (4.2) Initialize the weights ω and biases b using the method of random initialization;

[0090] (4.3) Import the data set into the convolutional neural network, and the data features are propagated forward through each layer, where X is the input part of the training set and Y is the output part of the training set;

[0091] (4.4) Calculate the error ε between the test set output and the actual output;

[0092] (4.5) Minimize the training error as the optimization goal of the network weight parameters;

[0093] (4.6) Gradually train to improve the model accuracy.

Claims

1. A method for determining the transient stability limit transmission power based on CNN, characterized in that, it includes the following steps: (1) Random fault simulation is carried out through PSAT simulation software and Monte Carlo simulation method, and transient stability limit analysis is performed. The power system analysis software is used to simulate the state of the system when a transient stability fault occurs. The characteristic values related to the transient stability limit of the transmission line in the system are exported from the simulation software, including: system power generation output, line impedance, node load, node voltage, and voltage phase angle; then through equivalent parameter calculation, the static stability limit P of the system is obtained s 、The static stability limit transmission power P of the transmission line steady 、The maximum power P of the transmission line max and the maximum impact power ΔP; (2), Process the characteristic values of the transient stability limit of the transmission line, the fault clearing time, and the fault type, and use them as the input part of constructing the training data set. The output part is the transient stability limit transmission power obtained by each line from the above formula; (3), Establish the structure of the convolutional neural network model; (4), Input the constructed training set into the convolutional network model, and train it in the way of grouped convolution multi-channels. Learn and optimize the weight coefficients of the model, and obtain the result of the transient stability limit of the transmission line through training; The specific step (1) is as follows: (1.1) Use the power system analysis software PSAT for transient stability analysis. First, set the system in the typical operating mode, and then perform the fault simulation of the transient stability limit; on the basis of the initial power flow, by adjusting the generator power and load value, at 75%-125% of the base load, with a step of 5%, set different load growth methods for each load level, and set a variety of different power generations for different load growth methods; in each power flow mode, use the Monte Carlo method to select faults, and use the power system analysis software to simulate the state of the system when a transient stability fault occurs, and export the characteristic values related to the transient stability limit of the transmission line in the system from the simulation software, including: system power generation, line impedance, node load, node voltage, voltage phase angle; (1.2) Through equivalent parameter calculation, based on the system transient equivalent model, calculate the static stability limit P of the transmission power of the transmission line s ; In formula (1), U 1 and U 2 are the equivalent voltages on both sides of the transmission line, and ∑Z is the total impedance of the system; The initial power P read by system simulation steady , the maximum power P of the system transmission line after the accident obtained from software simulation max , at this time, the maximum impact power ΔP is measured through fault simulation by the Monte Carlo simulation method: ΔP = P max -P steady (2) Calculate the transient stability limit transmission power of the transmission line based on the following formula (3), which is deduced based on the relationship between the static stability limit and the transient stability limit of the transmission line transmission power: P t = P s - ΔP(3) In formula (3), P t is the transient stability limit transmission power, P s is the steady-state stability limit of the transmission power of the transmission line, and ΔP is the maximum impact power; According to the relationship between the static stability limit and the transient stability limit of the transmission line transmission power, use the static stability limit power of the transmission line minus the impact power ΔP to preliminarily calculate the transient stability limit power of the line, and use it as the input part of the training set: ΔP = P s - P t (5) Obtain the static stability limit P of the transmission power of the transmission line s 、the static stability limit transmission power P of the transmission line steady 、the maximum power P of the transmission line max and the maximum impact power ΔP; Study the non-linear mapping relationship between the system input characteristic quantity of the power grid and the transient stability limit transmission power of the transmission channel, and use the deep learning model to quickly obtain the transient stability limit power of the transmission line; P t = F(V m , θ m , P n , Q n ), (V m , θ m , P, Q n ∈ R d ) (4) In formula (4), P t is the transient stability limit transmission power, m is the bus node number, n is the transmission line number, d is the number of sampling points, V is the node voltage amplitude, θ is the node voltage phase angle, P is the active power of the line, Q is the reactive power of the line, and R d is a two-dimensional matrix.

2. A method for determining the transient stability limit transmission power based on CNN according to claim 1, characterized in that, The specific construction of the training data set in step (2) is as follows: (2.1), Delete the error data from the simulation data of the input part, and supplement the missing data; standardize the sorted data and put it into the data set; (2.2), Divide each fault condition under each operating mode into a training sample, and the ratio of the learning set to the test set is 8:2; (2.3) Expand the dimension of the data sample and convert it into a two-dimensional data format to keep the sample data and the sample label synchronized.

3. A method for determining the transient stability limit transmission power based on CNN according to claim 1, characterized in that, Establish the structure of the convolutional neural network model in step (3); this model structure includes a convolutional layer, a pooling layer, a fully connected layer, and an identity function output layer, specifically: (3.1) The convolutional layer performs feature non - linear mapping. For the input data, the first convolutional layer uses convolutional kernels of multiple sizes, and the Padding method is used during the convolution process; (3.2) Max - pooling operation is carried out to reduce the scale of the data, improve the robustness of the system. The fully - connected layer and the output layer are used for regression fitting; (3.3) The third convolutional layer further extracts the data features; (3.4) The fourth convolutional layer uses multiple convolutional kernels, and the activation function is ReLU; (3.5) The fifth pooling layer continues the max - pooling operation to remove redundant data; (3.6) The Fatten operation is performed on the data to stretch the high - dimensional data into one - dimensional data; (3.7) The stretched data is fed into the fully - connected layer, and the output result is the transient stability limit power value of the transmission line, and the calculated result data is verified.

4. A method for determining the transient stability limit transmission power based on CNN according to claim 1, characterized in that, the method for training the convolutional neural network model in step (4) specifically includes the following steps: (4.1) The training parameters of the convolutional neural network include the weights ω and biases b between the connection layers. The random gradient descent method is used during the training process, and the backpropagation algorithm is used for parameter optimization; (4.2) The weights ω and biases b are initialized using the random initialization method; (4.3) The data set is imported into the convolutional neural network, and the data features are propagated forward through each layer, where X is the input part of the training set and Y is the output part of the training set; (4.4) Calculate the error ε between the test set output and the actual output; (4.5) Minimize the training error as the optimization goal of the network weight parameters; (4.6) Gradually train to improve the model accuracy.

Images