Motor train unit dynamic load modeling method based on measured data

Abstract

The invention discloses a motor train unit dynamic load modeling method based on measured data, which comprises the following steps: acquiring load-related data in the interval operation process of a motor train unit according to a time sequence, preprocessing, extracting load characteristics, and forming a load characteristic sequence by all load characteristics under each time sequence; according to the load extraction method based on data window state discrimination, load state classification is carried out on all the load feature sequences, the load feature sequences in the on-load state are input into a multi-classification SVM based on a kernel principal component analysis method to carry out main load feature recognition, and the recognized main load features are input into a BP neural network. Predicting the load of the motor train unit in the on-load state; and for the load characteristic sequence in the no-load state, constructing the load of the motor train unit in the no-load state by combining the load active power in the no-load state with the no-load loss. According to the method, the dynamic load condition of the motor train unit in the interval operation process is obtained by collecting the actual operation data of the motor train unit.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic load modeling of electric train sets, and in particular to a dynamic load modeling method of electric train sets based on measured data. Background Art

[0002] In order to ensure the safe and stable operation of EMUs, it is essential to monitor the dynamic load during operation in real time. In recent years, some scholars have begun to try to use the measured data such as voltage and current measured during the operation of EMUs to establish a dynamic load model that is closer to reality. These dynamic load models can better capture the dynamic characteristics of EMU loads by statistically analyzing the measured data and establishing mathematical relationships.

[0003] However, since all EMUs in the traction power supply system are powered by the same traction substation, it is difficult to monitor the normally operating EMUs in real time. Summary of the invention

[0004] In response to the problems existing in the prior art, the present invention provides a dynamic load modeling method for EMUs based on measured data, which aims to obtain the dynamic load conditions of the EMUs during section operation by collecting the actual operation data of the EMUs and performing analysis and modeling based on these data, thereby providing important technical support for ensuring the safe and stable operation of the EMUs, providing decision-making basis for operation managers, and improving the safety, stability and operation efficiency of the EMUs.

[0005] To achieve the above technical objectives, the present invention adopts the following technical solution: a dynamic load modeling method for EMU based on measured data, comprising the following steps:

[0006] Step 1: Collect load-related data during the EMU interval operation in time sequence and perform preprocessing;

[0007] Step 2: Extract load characteristics from the pre-processed load-related data, and all load characteristics in each time series constitute a load characteristic sequence;

[0008] Step 3: Based on the load extraction method of data window state discrimination, all load characteristic sequences are classified into load characteristic sequences of loaded state and load characteristic sequences of unloaded state;

[0009] Step 4: For the load feature sequence in the loaded state, input it into the multi-classification SVM based on the kernel principal component analysis method to identify the main load features, input the identified main load features into the BP neural network, and predict the load of the EMU in the loaded state;

[0010] Step 5: For the load characteristic sequence in the no-load state, the load of the EMU in the no-load state is constructed by combining the load active power in the no-load state with the no-load loss.

[0011] Furthermore, each piece of load-related data collected in step 1 includes: a current signal, a voltage signal, a load active power, and a load reactive power.

[0012] Furthermore, the load characteristics extracted in step 2 are divided into load time domain characteristics and load frequency domain characteristics. The process of extracting the load time domain characteristics includes: extracting statistical characteristics of current signals and voltage signals, extracting waveform characteristics of current signals and voltage signals, and extracting average values ​​and fluctuation amplitudes of load active power and load reactive power; the process of extracting the load frequency domain characteristics includes: extracting spectral characteristics of current signals and voltage signals, analyzing spectral amplitudes and frequency component proportions, and calculating harmonic content and total harmonic distortion of current signals and voltage signals.

[0013] Furthermore, step 3 includes the following sub-steps:

[0014] Step 301: sort all load feature sequences in chronological order, divide the load feature sequences equally into data segments with the same length, and calculate the average value of the load active power of all load feature sequences in each data segment;

[0015] Step 302: If the average value of the load active power of all load characteristic sequences in the data segment is less than the load threshold, all load characteristic sequences in the data segment are preliminarily determined to be in an unloaded state; otherwise, all load characteristic sequences in the data segment are in a loaded state;

[0016] Step 303: for the load characteristic sequence initially determined to be in the no-load state, determine whether the no-load time lasts for more than 5 minutes. If so, the load characteristic sequence is in the no-load state; otherwise, the load characteristic sequence is in the loaded state.

[0017] Step 304: sort all load characteristic sequences in the no-load state in chronological order, and sort all load characteristic sequences in the loaded state in chronological order.

[0018] Furthermore, step 4 includes the following sub-steps:

[0019] Step 401, reducing the dimension of the load feature sequence in the loaded state by using the kernel principal component analysis method;

[0020] Step 402: Determine the number of binary classification SVMs according to the number of main load features to be identified;

[0021] Step 403: input the reduced-dimensional load feature sequence into each binary SVM in turn to identify the main load features;

[0022] Step 404: Input all main load characteristics into the BP neural network to predict the load of the train set under loaded state.

[0023] Furthermore, step 401 includes the following sub-steps:

[0024] Step 4011: construct all load characteristic sequences in the loaded state into a load characteristic matrix and perform z-score standardization processing;

[0025] Step 4012: determine a kernel matrix by using a Gaussian kernel function, where the elements in the kernel matrix are expressed as:

[0026]

[0027] Among them, t i ,t j They represent the elements of the i-th and j-th rows in the load characteristic matrix, respectively, and δ represents the parameters in the Gaussian kernel function;

[0028] Step 4013, after the kernel matrix is ​​decentralized, the eigenvalues ​​and the corresponding eigenvectors are calculated, and the eigenvectors are arranged in descending order of the eigenvalues ​​to form an eigenvector matrix, and the eigenvector matrix is ​​subjected to unit orthogonal processing using the Schmidt orthogonal method to obtain an orthogonal matrix;

[0029] Step 4014: Select the part of the eigenvalues ​​whose total contribution rate reaches more than 90% from the orthogonal matrix to form a contribution matrix, and combine it with the decentralized kernel matrix to obtain a reduced-dimensional eigenvector matrix Z=Kp·Qs, where Kp represents the centralized kernel matrix and Qs represents the contribution matrix.

[0030] Furthermore, the number S of the two-class SVM is:

[0031] S=M(M-1) / 2

[0032] Where M represents the number of main load features to be identified.

[0033] Furthermore, the binary SVM performs the following process to identify the main load features:

[0034]

[0035] in, represents the classification result of the main load characteristics identified, n represents the total number of load characteristic sequences in the loaded state after dimensionality reduction, i represents the index of n, and α i represents the i-th Lagrange multiplier, y i represents the category value of the load feature sequence of the i-th loaded state after dimension reduction, k <t i,t> represents the Gaussian kernel function between the feature vector of the i-th load feature sequence in the loaded state after dimensionality reduction and the feature vector of the identified load feature sequence in the loaded state after dimensionality reduction.

[0036] Furthermore, the construction process of the load P0 in the no-load state of the EMU is:

[0037] P0=P1+P n +P 空

[0038] Among them, P1 represents the line loss, P n represents the measurement noise, P 空 Indicates the load active power in no-load state.

[0039] Compared with the prior art, the present invention has the following beneficial effects: the dynamic load modeling method of the EMU based on measured data of the present invention first proposes a load extraction method based on data window state discrimination for the differences in load characteristics of the substation during the interval operation, and realizes the extraction of dynamic load during the interval operation of the EMU; secondly, based on the multi-classification SVM of the kernel principal component analysis method, all extracted load feature sequences under the loaded state are classified and identified, and the kernel principal component analysis method is used to effectively reduce the load feature dimension while retaining the main load features, which helps to reduce the computational complexity and storage requirements, thereby realizing the load modeling of the EMU under the loaded state; and for the load feature sequence in the unloaded state, the load in the unloaded state of the EMU is constructed by combining the load active power in the unloaded state with the unloaded loss. The present invention can more accurately capture the load characteristics under different working conditions and improve the prediction accuracy of the load model by modeling the dynamic load of the EMU in the loaded state and the unloaded state respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in describing the specific implementation methods. Obviously, the drawings described below are only some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 It is a flow chart of the dynamic load modeling method of the EMU based on measured data of the present invention;

[0042] Figure 2 A flow chart of load state classification in the present invention;

[0043] Figure 3 A load curve diagram for a single EMU calculated using the EMU dynamic load modeling method based on measured data of the present invention;

[0044] Figure 4 The load curve diagram is calculated by the dynamic load modeling method of the EMU based on measured data of the present invention for multiple EMU interactions. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0046] like Figure 1 The flowchart of the dynamic load modeling method of the EMU based on measured data of the present invention comprises the following steps:

[0047] Step 1: collect load-related data during the interval operation of the EMU in time sequence and perform preprocessing; each load-related data collected by the present invention includes: current signal, voltage signal, load active power and load reactive power. In order to reduce the measurement error caused by the inconsistency of the measuring device or the lack of data when collecting the load-related data, the load-related data needs to be preprocessed, including: signal denoising, filling missing values, smoothing processing and normalization processing, wherein the normalization processing process is specifically as follows:

[0048] i. Find the maximum and minimum values ​​of each type of data in the load-related data;

[0049] ii. Normalize each data according to the maximum and minimum values ​​of the corresponding class data: Normalized value = (data value - minimum value) / (maximum value - minimum value), so as to map the original data to the normalized value within the specified range.

[0050] Step 2, extract the load characteristics of the pre-processed load-related data, and all the load characteristics under each time sequence constitute a load characteristic sequence; the load characteristics extracted in the present invention are divided into load time domain characteristics and load frequency domain characteristics, and the process of extracting the load time domain characteristics includes: extracting the statistical characteristics of the current signal and the voltage signal, including: mean value, variance, skewness and kurtosis, extracting the waveform characteristics of the current signal and the voltage signal such as amplitude, peak value and effective value, extracting the average value and fluctuation amplitude of the load active power and the load reactive power; the process of extracting the load frequency domain characteristics includes: using Fourier transform or wavelet transform to extract the spectrum characteristics of the current signal and the voltage signal, analyzing the spectrum amplitude and the proportion of frequency components, and calculating the harmonic content and total harmonic distortion of the current signal and the voltage signal. The load time domain characteristics reflect the dynamic changes of the load over time, and the frequency domain load characteristics reflect the composition and energy distribution of the load in the frequency domain. The present invention can obtain multi-dimensional load information by extracting the time domain and frequency domain characteristics of the load.

[0051] Step 3, the load extraction method based on data window state discrimination classifies the load state of all load characteristic sequences into load characteristic sequences of loaded state and load characteristic sequences of unloaded state. The traction power supply arm can be divided into unloaded state and loaded state according to whether there is an EMU running in the power supply section. Therefore, by judging the loaded part of the load data and merging the continuous loaded state data segments, the dynamic load extraction of the loaded state during the EMU section operation process can be realized; in addition, the idling condition is also a kind of loaded state, during which only the auxiliary power supply system works, but its power is small and easy to be confused with the unloaded load. The present invention analyzes the measured load data of the power supply section during the window time, and according to the distribution law of its load size, it is obtained that the load size of the power supply arm when unloaded is mainly concentrated at 0.05MW, and the fluctuation range does not exceed 0.05MW; while under the idling condition, its load size distribution is mainly concentrated at 0.5MW, and the fluctuation range does not exceed 0.1MW. Therefore, an appropriate load threshold is set to distinguish between no-load and idling states. In the measured data, noise caused by grid voltage fluctuations or measurement errors is inevitable, which makes the load power greater than the threshold when no-load. However, this situation lasts for a short time, so these false judgments can be eliminated by judging the length of the no-load state. Compared with the traditional event detection method based on sliding windows, it eliminates the need to judge the time of the process start event, saves calculations, and the accuracy of the extraction results meets the requirements. Figure 2 , specifically including the following sub-steps:

[0052] Step 301: sort all load feature sequences in chronological order, divide the load feature sequences equally into data segments with the same length, and calculate the average value of the load active power of all load feature sequences in each data segment;

[0053] Step 302: If the average value of the load active power of all load characteristic sequences in the data segment is less than the load threshold, all load characteristic sequences in the data segment are preliminarily determined to be in an unloaded state; otherwise, all load characteristic sequences in the data segment are in a loaded state;

[0054] Step 303: For the load characteristic sequence initially determined to be in no-load state, its load active power will inevitably have noise caused by grid voltage fluctuation or measurement error, so that the load active power is greater than the power threshold when no-loaded. However, this situation lasts for a very short time, so these erroneous judgments can be eliminated by judging the length of time of the no-load state. Therefore, the present invention judges whether the no-load time of the load characteristic sequence initially determined to be in no-load state lasts for more than 5 minutes. If so, the load characteristic sequence is in no-load state, otherwise, the load characteristic sequence is in loaded state;

[0055] Step 304: sort all load characteristic sequences in the no-load state in chronological order, and sort all load characteristic sequences in the loaded state in chronological order.

[0056] Step 4: For the load feature sequence in the loaded state, input it into the multi-classification SVM based on the kernel principal component analysis method to identify the main load features, input the identified main load features into the BP neural network, and predict the load of the EMU in the loaded state. Since the load feature sequence contains load features of many dimensions, directly predicting the load based on the high-dimensional load features will cause the calculation complexity to increase sharply and cause overfitting. The present invention can effectively reduce the load feature dimensions through the kernel principal component analysis method, map the high-dimensional load features to the low-dimensional space, thereby reducing the calculation amount of the multi-classification SVM, and identify the main load features, greatly improving the recognition accuracy of the SVM. The BP neural network uses its nonlinear modeling ability and multi-layer structure to accurately predict the load in the loaded state by identifying the main load features, thereby enhancing the discrimination ability and adaptability of the BP neural network. Specifically including the following sub-steps:

[0057] Step 401, reducing the dimension of the load feature sequence in the loaded state by using the kernel principal component analysis method; specifically includes the following sub-steps:

[0058] Step 4011: construct all load characteristic sequences in the loaded state into a load characteristic matrix and perform z-score standardization processing;

[0059] Step 4012: determine the kernel matrix by using the Gaussian kernel function. The elements in the kernel matrix are expressed as:

[0060]

[0061] Among them, t i ,t j They represent the elements of the i-th and j-th rows in the load characteristic matrix, respectively, and δ represents the parameters in the Gaussian kernel function;

[0062] Step 4013, after the kernel matrix is ​​decentralized, the eigenvalues ​​and the corresponding eigenvectors are calculated, and the eigenvectors are arranged in descending order of the eigenvalues ​​to form an eigenvector matrix, and the eigenvector matrix is ​​subjected to unit orthogonal processing using the Schmidt orthogonal method to obtain an orthogonal matrix;

[0063] Step 4014: Select the part of the eigenvalues ​​whose total contribution rate reaches more than 90% from the orthogonal matrix to form a contribution matrix, and combine it with the decentralized kernel matrix to obtain a reduced-dimensional eigenvector matrix Z=Kp·Qs, where Kp represents the centralized kernel matrix and Qs represents the contribution matrix.

[0064] Step 402: Determine the number of binary SVMs S=M(M-1) / 2 according to the number of main load features to be identified, where M represents the number of main load features to be identified.

[0065] Step 403: Input the reduced-dimensional load feature sequence into each binary SVM in turn to identify the main load features:

[0066]

[0067] in, represents the classification result of the main load characteristics identified, n represents the total number of load characteristic sequences in the loaded state after dimensionality reduction, i represents the index of n, and α i represents the i-th Lagrange multiplier, y i represents the category value of the load feature sequence of the i-th loaded state after dimension reduction, k <t i ,t> represents the Gaussian kernel function between the feature vector of the load feature sequence in the loaded state after dimensionality reduction of the i-th load feature sequence and the feature vector of the identified load feature sequence in the loaded state after dimensionality reduction, and b represents the intercept of the classification hyperplane curve.

[0068] Step 404: All main load characteristics are input into the BP neural network to predict the load of the EMU under loaded state. The BP neural network in the present invention includes: an input layer, a hidden layer and an output layer. The function of neurons in each layer is different. The information transmission path is "input layer-hidden layer-output layer". If the predicted result calculated by the output layer deviates from the actual result, the deviation is transmitted in reverse, and the weight parameters of the BP neural network are adjusted by the gradient descent method to gradually reduce the error between the predicted output and the actual result.

[0069] The load prediction process of the EMU under the loaded state in the present invention is as follows:

[0070] Y=f(WX+B)

[0071] Wherein, Y represents the load prediction result of the EMU under the loaded state, X represents all the main load characteristics, W represents the weight matrix, B represents the deviation matrix, f() represents the transfer function, and the hidden layer function in the present invention is the Tan-Sigmoid function, and the output layer function is the Purlin linear function.

[0072] Step 5: For the load characteristic sequence in the no-load state, the load in the no-load state of the EMU is constructed by combining the load active power in the no-load state with the no-load loss:

[0073] P0=P1+P n +P 空

[0074] Where P1 represents the line loss. For a certain traction substation, if it is assumed that the external power grid state remains unchanged, P1 remains unchanged during the no-load process; P n represents the measurement noise, P 空Indicates the load active power in no-load state.

[0075] The present invention can more accurately capture the load characteristics under different working conditions and improve the prediction accuracy of the load model by modeling the dynamic load of the EMU in a loaded state and an unloaded state respectively.

[0076] like Figure 3 The load curve diagram is a single EMU calculated by the EMU dynamic load modeling method based on measured data of the present invention. It can be seen that the load during the operation of the EMU predicted by the load modeling method of the present invention is basically consistent with the curve of the actual load, indicating that the EMU dynamic load modeling method of the present invention is feasible for load prediction during the operation of a single EMU, and the prediction accuracy is relatively high.

[0077] like Figure 4 The load curve diagram is calculated by the dynamic load modeling method of the EMU based on the measured data for the interaction of multiple EMUs. It can be seen that the load during the operation of the EMU predicted by the load modeling method of the present invention is basically consistent with the curve of the actual load, indicating that the dynamic load modeling method of the EMU of the present invention is feasible for load prediction during the interactive operation of multiple EMUs, and the prediction accuracy is relatively high.

[0078] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0079] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0080] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A dynamic load modeling method for EMU based on measured data, characterized in that: The steps include: Step 1: Collect load-related data during the EMU interval operation in time sequence and perform preprocessing; Step 2: Extract load characteristics from the pre-processed load-related data, and all load characteristics in each time series constitute a load characteristic sequence; Step 3: Based on the load extraction method of data window state discrimination, all load characteristic sequences are classified into load characteristic sequences of loaded state and load characteristic sequences of unloaded state; Step 4: For the load feature sequence in the loaded state, input it into the multi-classification SVM based on the kernel principal component analysis method to identify the main load features, input the identified main load features into the BP neural network, and predict the load of the EMU in the loaded state; Step 5: For the load characteristic sequence in the no-load state, the load of the EMU in the no-load state is constructed by combining the load active power in the no-load state with the no-load loss.

2. The method for dynamic load modeling of a train set based on measured data according to claim 1 is characterized in that: Each piece of load-related data collected in step 1 includes: current signal, voltage signal, load active power and load reactive power.

3. The method for dynamic load modeling of a train set based on measured data according to claim 2 is characterized in that: The load characteristics extracted in step 2 are divided into load time domain characteristics and load frequency domain characteristics. The process of extracting the load time domain characteristics includes: extracting the statistical characteristics of the current signal and the voltage signal, extracting the waveform characteristics of the current signal and the voltage signal, and extracting the average value and fluctuation amplitude of the load active power and the load reactive power; the process of extracting the load frequency domain characteristics includes: extracting the spectrum characteristics of the current signal and the voltage signal, analyzing the spectrum amplitude and the frequency component ratio, and calculating the harmonic content and total harmonic distortion of the current signal and the voltage signal.

4. The method for dynamic load modeling of a train set based on measured data according to claim 3 is characterized in that: Step 3 includes the following sub-steps: Step 301: sort all load feature sequences in chronological order, divide the load feature sequences equally into data segments with the same length, and calculate the average value of the load active power of all load feature sequences in each data segment; Step 302: If the average value of the load active power of all load characteristic sequences in the data segment is less than the load threshold, all load characteristic sequences in the data segment are preliminarily determined to be in an unloaded state; otherwise, all load characteristic sequences in the data segment are in a loaded state; Step 303: for the load characteristic sequence initially determined to be in the no-load state, determine whether the no-load time lasts for more than 5 minutes. If so, the load characteristic sequence is in the no-load state; otherwise, the load characteristic sequence is in the loaded state. Step 304: sort all load characteristic sequences in the no-load state in chronological order, and sort all load characteristic sequences in the loaded state in chronological order.

5. The method for dynamic load modeling of a train set based on measured data according to claim 4 is characterized in that: Step 4 includes the following sub-steps: Step 401, reducing the dimension of the load feature sequence in the loaded state by using the kernel principal component analysis method; Step 402: Determine the number of binary classification SVMs according to the number of main load features to be identified; Step 403: input the reduced-dimensional load feature sequence into each binary SVM in turn to identify the main load features; Step 404: Input all main load characteristics into the BP neural network to predict the load of the train set under loaded state.

6. The method for dynamic load modeling of a train set based on measured data according to claim 5 is characterized in that: Step 401 includes the following sub-steps: Step 4011: construct all load characteristic sequences in the loaded state into a load characteristic matrix and perform z-score standardization processing; Step 4012: determine a kernel matrix by using a Gaussian kernel function, where the elements in the kernel matrix are expressed as: Among them, t i ,t j They represent the elements of the i-th and j-th rows in the load characteristic matrix, respectively, and δ represents the parameters in the Gaussian kernel function; Step 4013, after the kernel matrix is ​​decentralized, the eigenvalues ​​and the corresponding eigenvectors are calculated, and the eigenvectors are arranged in descending order of the eigenvalues ​​to form an eigenvector matrix, and the eigenvector matrix is ​​subjected to unit orthogonal processing using the Schmidt orthogonal method to obtain an orthogonal matrix; Step 4014: Select the part of the eigenvalues ​​whose total contribution rate reaches more than 90% from the orthogonal matrix to form a contribution matrix, and combine it with the decentralized kernel matrix to obtain a reduced-dimensional eigenvector matrix Z=Kp·Qs, where Kp represents the centralized kernel matrix and Qs represents the contribution matrix.

7. The method for dynamic load modeling of a train set based on measured data according to claim 5 is characterized in that: The number S of the two-class SVM is: S=M(M-1) / 2 Where M represents the number of main load features to be identified.

8. The method for dynamic load modeling of a train set based on measured data according to claim 5 is characterized in that: The binary SVM performs the following process to identify the main load features: in, represents the classification result of the main load characteristics identified, n represents the total number of load characteristic sequences in the loaded state after dimensionality reduction, i represents the index of n, and α i represents the i-th Lagrange multiplier, y i represents the category value of the load feature sequence of the i-th loaded state after dimension reduction, k <t i ,t> represents the Gaussian kernel function between the feature vector of the i-th load feature sequence in the loaded state after dimensionality reduction and the feature vector of the identified load feature sequence in the loaded state after dimensionality reduction.

9. The method for dynamic load modeling of a train set based on measured data according to claim 4 is characterized in that: The construction process of the load P0 in the no-load state of the EMU is: P0=P1+P n +P 空 Among them, P1 represents the line loss, P n represents the measurement noise, P 空 Indicates the load active power in no-load state.

Images