A traction power supply system power flow calculation method based on residual neural network

Abstract

The application discloses a traction power supply system power flow calculation method based on a residual neural network, and belongs to the traction power supply system simulation field.The method comprises the following steps: step 1, a traction power supply system steady-state model is built; step 2, a time-varying electric network equivalent model of the traction power supply system is established; step 3, direct current power flow calculation iteration is solved, and a power flow training data set is acquired; step 4, appropriate features are selected as input and output data of a neural network model, and normalization processing is performed; step 5, a residual neural network model is designed; step 6, the neural network is trained, and model effectiveness verification is performed, so that a power flow calculation model is acquired; and step 7, input parameters are set, residual network forward propagation calculation is performed, and a power flow calculation result is acquired.The application meets the power flow calculation precision, improves the power flow calculation solving efficiency, realizes online monitoring and state analysis of the traction power supply system, and provides a solution strategy for enhancing system power supply reliability and safety.

Description

Technical Field

[0001] This invention relates to the field of traction power supply system simulation, and specifically to a power flow calculation method for traction power supply systems based on residual neural networks. Background Technology

[0002] A typical traction power supply system consists of trains, traction substations, running rails, and overhead contact lines. By equating the traction substation with an ideal voltage source and internal resistance in series, the train with a power source and braking resistor in parallel, and the overhead contact lines and running rails with lines having internal resistance and leakage resistance, an equivalent network topology for the DC traction power supply system is established. The node admittance matrix is ​​then established using the nodal voltage method to solve for the power flow distribution at each node, thereby meeting the requirements of high power supply quality and high safety for urban rail transit, strengthening the operation and management of the power supply system, and improving the energy conservation and emission reduction level of the power supply system itself.

[0003] As urban rail transit lines continue to extend, the number of stations along these lines also increases, leading to a rise in unknown parameters in the traction power supply system. Traditional Newton-Raphson iterative solutions for DC power flow calculations become more complex and sometimes unsolvable. Mathematical models for DC traction power supply systems are even more difficult to establish and solve, and the system is nearing its limits in terms of high-speed real-time calculations, making it difficult to meet the current needs for online status analysis and ultra-real-time status simulation of rail transit traction power supply systems.

[0004] This invention introduces residual neural networks to participate in power flow calculation, which improves the efficiency of power flow calculation while meeting the accuracy requirements, realizes online monitoring and status analysis of traction power supply system, and provides a solution strategy to enhance the reliability and safety of system power supply. Summary of the Invention

[0005] To address the shortcomings in the aforementioned technical field, the present invention aims to provide a power flow calculation method for traction power supply systems based on residual neural networks, thereby meeting the needs of line state analysis and ultra-real-time state extrapolation for traction power supply systems.

[0006] The technical solution for implementing the present invention includes:

[0007] A power flow calculation method for a traction power supply system based on residual neural networks includes:

[0008] Step 1: Based on the actual urban rail line conditions, build a steady-state model of the traction power supply system;

[0009] Step 2: Perform circuit abstraction and equivalent transformation on each device in the steady-state model of the traction power supply system to establish an equivalent model of the time-varying power network of the traction power supply system;

[0010] Step 3: Based on the equivalent model of the time-varying power network, establish the node admittance matrix; based on the node admittance matrix and the train traction power data measured by the data acquisition terminal, perform DC power flow calculation and iterative solution to obtain the voltage data of each section of the train and the voltage and power of each traction substation, and save them as the power flow training dataset.

[0011] Step 4: Select appropriate features from the power flow training dataset based on the power flow calculation equation as input and output data for the residual neural network model, and normalize the selected feature data.

[0012] Step 5: Design the residual neural network model structure and set the model parameters;

[0013] Step 6: Train the residual neural network based on the selected feature data, verify the effectiveness of the model, and thus obtain the residual neural network power flow calculation model of the DC traction power supply system.

[0014] Step 7: After completing the residual neural network modeling, set the input parameters of the residual network model, perform forward propagation calculation of the residual network, and obtain the power flow calculation results under the corresponding state.

[0015] Furthermore, in the equivalent model of the time-varying network of the traction power supply system established in step 2, the circuit abstract equivalents of each device include:

[0016] Traction substation model: The rectifier unit converts the AC power from the grid into DC power required by the traction system. Its circuit abstract equivalent model is a voltage source with series internal resistance. Train equivalent model: The train has traction and braking states. It is equivalent to a power source in parallel with a resistor. The operation state is changed by switching the resistor. DC traction network model: It includes the contact network, running rails and related electrical equipment. It is regarded as a π-type equivalent model of the rail.

[0017] Furthermore, in step 3, based on the equivalent model of the time-varying electric network, the nodal admittance matrix is ​​established, including:

[0018] DC traction power supply system node admittance matrix:

[0019]

[0020] In the formula, Y dij Let i be the mutual admittance of node i and node j in the DC node admittance matrix, where i,j = 1, 2, 3...n, and n is the total number of nodes in the system.

[0021] Furthermore, in step 3, the DC power flow calculation and iterative solution yields the voltage data for each section of the train and the voltage and power of each traction substation, including:

[0022] DC power flow equations solved based on Newton-Raphson iteration:

[0023]

[0024] The vector form is:

[0025]

[0026] In the formula, ΔP d =[ΔP d1 ,…,ΔP dn ] T U is the power deviation vector; d =[U d1 ,...,U dn ] T ΔU is the node voltage vector; d =[ΔU d1 ,...,ΔU dn ] T J is the node voltage deviation vector; d The Jacobian matrix for DC power flow;

[0027] The DC power flow calculation method is as follows:

[0028] Step 31: Input traction power supply system data, including train timetable, platform location, substation location, line parameters, simulation time, etc.

[0029] Step 32: Determine the simulation step size Δt, calculate the train's position, online status, running status, and power meter data at the current simulation time, and store them in the node array;

[0030] Step 33: Sort the train nodes and substation nodes by location and assign node numbers. Calculate the mutual admittance and self-admittance values ​​for each node and establish the node admittance matrix Y. d ;

[0031] Step 34, set the iteration count k = 0;

[0032] Step 35, solve for the power deviation ΔP at each node. di ;

[0033] Step 36: Determine whether the maximum power deviation of the node is within the allowable range. If it is, proceed to step 311; otherwise, proceed to the next step.

[0034] Step 37: Solve for the Jacobian matrix in the power flow equation correction equation;

[0035] Step 38: Solve the corrected equation to obtain ΔU di Calculate U dc Node voltage;

[0036] Step 39, let k = k + 1, return to step 35;

[0037] Step 310, calculate the branch current;

[0038] Step 311: Using the obtained node voltages, solve for the power distribution of the DC system and output the results.

[0039] Furthermore, step 4 involves selecting appropriate features as input and output data for the residual neural network model, including:

[0040] Referring to the DC power flow equations, the input features can be extracted as follows:

[0041]

[0042] In the formula, m and n represent the number of sections and substations in the traction power supply system, respectively; P u1 ,P u2 ,…,P um For the power of train nodes in each section, the power of nodes in sections where the train is online is taken as the traction power, and the power of nodes in sections where the train is offline is taken as 0; P d1 ,P d2 ,…,P dN The power of each train node in the downlink section is determined using the same method as in the uplink case; Y L1 ,Y L2 ,…,Y Lm Y represents the impedance element at each node of the train. Q1 ,Y Q2 ,…,Y Qn These are the impedance elements of each node in the substation;

[0043] Referring to the DC power flow equations, the output characteristics can be extracted as follows:

[0044]

[0045] In the formula, U u1 U u2 ,...,U um U represents the voltage at each train node in the upstream section. d1 U d2 ,...,U dm P represents the voltage at each train node in the downlink section. Q1 ,P Q2 ,…,P Qn For the node power of each traction substation; U Q1 U Q2 ,…,U Qn This refers to the node voltage of each traction substation.

[0046] Furthermore, step 5 involves designing the residual neural network model structure, including:

[0047] The residual block structure is designed to consist of two parts: an LBR module and a DC module. The LBR module has one hidden layer and one BN layer, and the output value is nonlinearly activated using the ReLU activation function. The DC module contains one hidden layer and one BN layer. Jump connections are used between residual blocks to sum the input values ​​and activate them using the ReLU function. The output value serves as the input value for the next residual block.

[0048] The residual neural network model structure consists of four residual blocks and two fully connected layers. The LBR module and DC module of each residual block have 128 hidden nodes. The two fully connected layers serve as input and output layers, respectively. The input layer has M nodes, and the output layer has N nodes, where M is a multiple of P. u P d Y z The sum of the numbers, N is U U U D ,P Q U Q The sum of the numbers;

[0049] Model parameter settings: Select ReLU as the activation function, Adam as the parameter update method, and MSE as the loss function.

[0050] Furthermore, step 6, model validity verification, includes:

[0051] RMSE is selected as the error evaluation metric:

[0052]

[0053] Where n is the number of samples, y i It is the actual value. That is the corresponding predicted value.

[0054] Model validity verification: Set a threshold for model accuracy, and generate comparison results by judging whether the RMSE is greater than the preset threshold.

[0055] Furthermore, in step 6, the residual neural network is first trained based on the selected feature data. The model training process is as follows:

[0056] Step 61: Initialize the weights and thresholds of the residual neural network model;

[0057] Step 62: Generate offline simulation data through power flow calculation of the traction power supply system, and construct a power flow training sample set;

[0058] Step 63: Train the residual network using training set samples;

[0059] Step 64: Use validation set samples to test model accuracy;

[0060] Step 65: Determine whether the error exceeds the preset error threshold. If it does, update the residual neural network weights and threshold, and jump to step 63 to retrain the neural network. Otherwise, save the residual neural network model parameters and complete the training.

[0061] Among them, RMSE was selected as the error evaluation index:

[0062]

[0063] In the formula, n is the number of samples, y i It is the actual value. That is the corresponding predicted value.

[0064] Furthermore, in step 7, the forward propagation calculation of the residual network is performed to obtain the power flow calculation results, including: loading the parameters of the trained residual neural network model, inputting the power flow input feature data, and outputting the power flow calculation results.

[0065] According to the above technical solution, the beneficial effects of the present invention are as follows:

[0066] 1. This invention mainly relies on simulation data generated by the equivalent model of the time-varying network of the traction power supply system as the input and output of the residual neural network. It can obtain a large amount of residual neural network training dataset in a short time, thus solving the problem of difficulty in obtaining actual measured data of the traction power supply system.

[0067] 2. This invention mainly relies on selecting a residual neural network to map the traction power supply system to a model, which solves the problems of gradient vanishing and mesh degradation that exist when the BP neural network learns complex models, thereby improving the accuracy of the deep learning model of the traction power supply system.

[0068] 3. This invention performs residual network regression mapping on the traction power supply system, and the output power flow calculation results only require a series of matrix calculations. Compared with the power flow calculation model based on the Newton-Raphson algorithm iteration, the calculation efficiency is significantly improved and there is no unsolvable problem. Attached Figure Description

[0069] Figure 1 This is a flowchart of a power flow calculation method for a traction power supply system based on a residual neural network according to the present invention;

[0070] Figure 2 This invention relates to a steady-state model of a DC traction power supply system in a traction power supply system power flow calculation method based on residual neural networks.

[0071] Figure 3This is an equivalent model diagram of a traction substation in a traction power supply system power flow calculation method based on residual neural networks according to the present invention.

[0072] Figure 4 This is a diagram of the equivalent train model in the power flow calculation method for a traction power supply system based on residual neural networks of the present invention.

[0073] Figure 5 This is an equivalent model diagram of the DC traction network in the power flow calculation method of the traction power supply system based on residual neural network of the present invention;

[0074] Figure 6 This invention provides an equivalent model of the time-varying network of a DC traction power supply system in a power flow calculation method for a traction power supply system based on a residual neural network.

[0075] Figure 7 This is a flowchart of the DC traction power supply system power flow calculation method in the traction power supply system power flow calculation method based on residual neural network of the present invention;

[0076] Figure 8 This is a residual block structure diagram in the power flow calculation method for a traction power supply system based on a residual neural network of the present invention;

[0077] Figure 9 This is a residual neural network structure diagram in the power flow calculation method for a traction power supply system based on a residual neural network of the present invention;

[0078] Figure 10 This is a flowchart illustrating the specific training process of the residual neural network in the power flow calculation method for a traction power supply system based on a residual neural network, as described in this invention. Detailed Implementation

[0079] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the following examples provide a more detailed description of the invention. It should be noted that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.

[0080] Embodiments of the present invention provide a method for accelerating the simulation calculation of a traction power supply system based on a residual neural network, such as... Figure 1 As shown, it includes the following steps:

[0081] Step 1: The subway company provides accurate line-related data. This data includes basic line data, the locations of all traction substations and stations along the line, external characteristic parameters of the traction substations, original timetables, and power meter data. A steady-state model of the traction power supply system is then built. Figure 2 As shown;

[0082] Step 2: The circuits of each device in the steady-state model of the traction power supply system are abstracted and equivalent. Specifically, the circuit abstraction and equivalent model of the traction substation model is a voltage source with series internal resistance; the equivalent model of the train is equivalent to a power source with parallel resistance; and the DC traction network model is equivalent to a π-type resistor. The circuit structure diagram is shown below. Figure 3 , 4 As shown in Figure 5, the equivalent circuit diagrams of each device are integrated to establish an equivalent model of the time-varying electrical network of the traction power supply system, as shown in Figure 5. Figure 6 As shown.

[0083] Step 3: First, based on the equivalent model of the time-varying electrical network, establish the node admittance matrix of the DC traction power supply system:

[0084]

[0085] In the formula, Y dij Let i be the mutual admittance of node i and node j in the DC node admittance matrix, where i,j = 1, 2, 3...n, and n is the total number of nodes in the system.

[0086] Secondly, DC power flow calculations are performed iteratively based on the nodal admittance matrix and the measured train traction power data from the data acquisition terminal. This yields voltage data for each section of the train and the voltage and power of each traction substation, which are then saved as a power flow training dataset. The DC power flow equations are derived using the Newton-Raphson iterative solution:

[0087]

[0088] The vector form is:

[0089]

[0090] In the formula, ΔP d =[ΔP d1 ,...,ΔP dn ] T U is the power deviation vector; d =[U d1 ,...,U dn ] T ΔU is the node voltage vector; d =[ΔU d1 ,...,ΔU dn ] T J is the node voltage deviation vector; d The Jacobian matrix for DC power flow;

[0091] The DC power flow calculation method is as follows, and the flowchart is shown below. Figure 7 As shown:

[0092] Step 31: Input traction power supply system data, including train timetable, platform location, substation location, line parameters, simulation time, etc.

[0093] Step 32: Determine the simulation step size Δt, calculate the train's position, online status, running status, and power meter data at the current simulation time, and store them in the node array;

[0094] Step 33: Sort the train nodes and substation nodes by location and assign node numbers. Calculate the mutual admittance and self-admittance values ​​for each node and establish the node admittance matrix Y. d ;

[0095] Step 34, set the iteration count k = 0;

[0096] Step 35, solve for the power deviation ΔP at each node. di ;

[0097] Step 36: Determine whether the maximum power deviation of the node is within the allowable range. If it is, proceed to step 311; otherwise, proceed to the next step.

[0098] Step 37: Solve for the Jacobian matrix in the power flow equation correction equation;

[0099] Step 38: Solve the corrected equation to obtain ΔU di Calculate U dc Node voltage;

[0100] Step 39, let k = k + 1, return to step 35;

[0101] Step 310, calculate the branch current;

[0102] Step 311: Using the obtained node voltages, solve for the power distribution of the DC system and output the results.

[0103] Step 4: First, select appropriate features from the power flow training dataset based on the power flow calculation equation as the input and output data of the residual neural network model.

[0104] Among them, the extractable input features are:

[0105]

[0106] In the formula, m and n represent the number of sections and substations in the traction power supply system, respectively; P u1 ,P u2 ,…,P um For the power of train nodes in each section, the power of nodes in sections where the train is online is taken as the traction power, and the power of nodes in sections where the train is offline is taken as 0; P d1 ,P d2 ,…,P dNThe power of each train node in the downlink section is determined using the same method as in the uplink case; Y L1 ,Y L2 ,…,Y Lm Y represents the impedance element at each node of the train. Q1 ,Y Q2 ,…,Y Qn These are the impedance elements of each node in the substation;

[0107] The extractable output features are:

[0108]

[0109] In the formula, U u1 U u2 ,...,U um U represents the voltage at each train node in the upstream section. d1 U d2 ,...,U dm P represents the voltage at each train node in the downlink section. Q1 ,P Q2 ,…,P Qn For the node power of each traction substation; U Q1 U Q2 ,…,U Qn This refers to the node voltage of each traction substation.

[0110] Secondly, the residual neural network feature data of the traction power supply system is normalized using min-max normalization, and the formula is as follows:

[0111]

[0112] Where, x i Let x be the original data of the i-th dimension of a certain input variable. max Therefore, input the maximum value of the variable, x. min Therefore, input the minimum value of the variable y, where y is x. i The result after normalization.

[0113] Step 5, firstly, design the residual block structure, such as... Figure 8 As shown, it consists of two parts: an LBR module and a DC module. The LBR module has one hidden layer and one BN layer, and the output value is non-linearly activated using the ReLU activation function. The DC module contains one hidden layer and one BN layer. Skip connections are used between residual blocks to sum the input values ​​and activate them using the ReLU function. The output value serves as the input value for the next residual block. Next, the residual neural network model structure is designed, such as... Figure 9As shown, it consists of 4 residual blocks and 2 fully connected layers. The LBR module of the residual blocks has 128 hidden layers, and the DC module has 128 hidden layers. The two fully connected layers serve as input and output layers, with the input layer having M nodes and the output layer having N nodes, where M is the P... u P d Y z The sum of the numbers, N is U U U D ,P Q U Q The sum of the number of parameters; Model parameter settings: Select ReLU as the activation function, Adam as the parameter update method, and MSE as the loss function.

[0114] Step 6: First, train the residual neural network based on the selected feature data. The model training flowchart is as follows. Figure 10 As shown, the specific method is as follows:

[0115] Step 61: Initialize the weights and thresholds of the residual neural network model;

[0116] Step 62: Generate offline simulation data through power flow calculation of the traction power supply system, and construct a power flow training sample set;

[0117] Step 63: Train the residual network using training set samples;

[0118] Step 64: Use validation set samples to test model accuracy;

[0119] Step 65: Determine whether the error exceeds the preset error threshold. If it does, update the residual neural network weights and threshold, and jump to step 63 to retrain the neural network. Otherwise, save the residual neural network model parameters and complete the training.

[0120] Among them, RMSE was selected as the error evaluation index:

[0121]

[0122] In the formula, n is the number of samples, y i It is the actual value. That is the corresponding predicted value.

[0123] Step 7: Load the trained residual neural network model parameters, input the power flow input feature data, and output the power flow calculation results.

[0124] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for calculating power flow in a traction power supply system based on a residual neural network, characterized in that, The method is as follows: Step 1: Based on the actual urban rail line conditions, build a steady-state model of the traction power supply system; Step 2: Perform circuit abstraction and equivalent transformation on each device in the steady-state model of the traction power supply system to establish an equivalent model of the time-varying power network of the traction power supply system; Step 3: Based on the equivalent model of the time-varying power network, establish the node admittance matrix; based on the node admittance matrix and the train traction power data measured by the data acquisition terminal, perform DC power flow calculation and iterative solution to obtain the voltage data of each section of the train and the voltage and power of each traction substation, and save them as the power flow training dataset. Step 4: Select appropriate features from the power flow training dataset based on the power flow calculation equation as input and output data for the residual neural network model, and normalize the selected feature data. Step 5: Design the residual neural network model structure and set the model parameters; Step 6: Train the residual neural network based on the selected feature data, verify the effectiveness of the model, and thus obtain the residual neural network power flow calculation model of the DC traction power supply system. Step 7: After completing the residual neural network modeling, set the input parameters of the residual network model, perform forward propagation calculation of the residual network, and obtain the power flow calculation results under the corresponding state. In the equivalent model of the time-varying network of the traction power supply system established in step 2, the circuit abstract equivalents of each device include: Traction substation model: The rectifier unit converts the AC power from the grid into DC power required by the traction system. Its circuit abstract equivalent model is a voltage source with series internal resistance. Train equivalent model: The train's operating state has traction and braking, which can be equivalent to a power source connected in parallel with a resistor. The operating state is switched by switching the resistor. DC traction network model: This includes the overhead contact line, running rails, and related electrical equipment, and is considered as such. The equivalent model of the rail is performed.

2. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, Step 3, which involves establishing the node admittance matrix based on the equivalent model of the time-varying electric network, includes: DC traction power supply system node admittance matrix: （1.1） In the formula, For the nodes in the DC node admittance matrix i and nodes j The mutual admittance, of which, , n This represents the total number of system nodes.

3. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, In step 3, the DC power flow calculation and iterative solution yields the voltage data for each section of the train and the voltage and power of each traction substation, including: DC power flow equations solved based on Newton-Raphson iteration: （1.2） The vector form is: （1.3） In the formula, This is the power deviation vector; The node voltage vector; The node voltage deviation vector; It is the Jacobian matrix for DC power flow.

4. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, The DC power flow calculation method in step 3 is as follows: Step 31: Input traction power supply system data, including train timetable, platform location, substation location, line parameters, and simulation time data; Step 32, Determine the simulation step size Calculate the train's position, online status, running status, and power meter data at the current simulation time, and store them in the node array; Step 33: Sort the train nodes and substation nodes by location and assign node numbers. Calculate the mutual admittance and self-admittance values ​​for each node and establish a node admittance matrix. ; Step 34, Set the number of iterations ; Step 35, solve for the power deviation at each node. ; Step 36: Determine whether the maximum power deviation of the node is within the allowable range. If it is, proceed to step 311; otherwise, proceed to the next step. Step 37: Solve for the Jacobian matrix in the power flow equation correction equation; Step 38, solve the corrected equation to obtain ,calculate Node voltage; Step 39, let Return to step 35; Step 310, calculate the branch current; Step 311: Using the obtained node voltages, solve for the power distribution of the DC system and output the results.

5. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, Step 4, selecting appropriate features as input and output data for the residual neural network model, includes: Referring to the DC power flow equations, the input features can be extracted as follows: （1.4） In the formula, These represent the number of sections and substations in the traction power supply system, respectively. The power of train nodes in each section is taken as the traction power for trains in online sections and 0 for trains in offline sections. The power of train nodes in each down section is determined using the same method as for the up section. These are the impedance elements at each node of the train. Y Q1 , Y Q2 , Y Qn These are the impedance elements of each node in the substation; Referring to the DC power flow equations, the output characteristics can be extracted as follows: （1.5） In the formula, The voltage at each train node in the upstream section; This refers to the voltage at each train node in the downstream section; P Q1 , P Q2 , , P Qn The power of each traction substation node; U Q1 , U Q2 , , U Qn This refers to the node voltage of each traction substation.

6. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, Step 5, which involves designing the residual neural network model structure, includes: The residual block structure is designed as follows: It consists of two parts, an LBR module and a DC module. The LBR module has one hidden layer and one BN layer. The output value is nonlinearly activated using the ReLU activation function. The DC module contains one hidden layer and one BN layer. The residual blocks are connected by jump connections to sum the input values ​​and activate them with the ReLU function. The output value is used as the input value of the next residual block. The residual neural network model structure consists of four residual blocks and two fully connected layers. The LBR module and DC module of each residual block have 128 hidden nodes. The two fully connected layers serve as input and output layers, respectively. The input layer has M nodes, and the output layer has N nodes, where M is the number of nodes in the input layer. The sum of the numbers, N is The sum of the numbers; Model parameter settings: Select ReLU as the activation function, Adam as the parameter update method, and MSE as the loss function.

7. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, Step 6, model validity verification, includes: RMSE is selected as the error evaluation metric: （1.6） in, It is the sample size. It is the actual value. This is the corresponding predicted value; Model validity verification: Set a threshold for model accuracy, and generate comparison results by judging whether the RMSE is greater than the preset threshold.

8. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, In step 6, the residual neural network is first trained based on the selected feature data. The model training process is as follows: Step 61: Initialize the weights and thresholds of the residual neural network model; Step 62: Generate offline simulation data through power flow calculation of the traction power supply system, and construct a power flow training sample set; Step 63: Train the residual network using training set samples; Step 64: Use validation set samples to test model accuracy; Step 65: Determine whether the error exceeds the preset error threshold. If it does, update the residual neural network weights and threshold, and jump to step 63 to retrain the neural network. Otherwise, save the residual neural network model parameters and complete the training. Among them, RMSE was selected as the error evaluation index: （1.7） in, It is the sample size. It is the actual value. That is the corresponding predicted value.

9. The power flow calculation method for a traction power supply system based on a residual neural network according to claim 1, characterized in that, Step 7 involves performing forward propagation calculations in the residual network to obtain power flow calculation results, including: loading the parameters of the trained residual neural network model, inputting power flow input feature data, and outputting power flow calculation results.

Images