Current identification model training method, current identification method, device and equipment

Abstract

The invention relates to a current identification model training method and device, a current identification method and device and equipment. The current identification model training method comprises the steps of obtaining a sample current signal of a traction network power supply system and a current type of the sample current signal; performing variational mode decomposition on the sample current signal to obtain a plurality of mode components of the sample current signal; extracting current characteristics of the sample current signal according to the plurality of modal components; and training a preset classification model by taking the current characteristics as input and the current categories as labels to obtain a current identification model of the traction network power supply system. The method can improve the current identification accuracy of the traction network power supply system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traction power supply system, in particular to a current identification model training method, a current identification method, a device and equipment. BACKGROUND

[0002] The traction power supply system provides guarantee for the safe and reliable operation of electric locomotives as the power source of electric locomotives. With the development of high-speed railways and electric locomotives, the safe and reliable operation of the traction power supply system is facing challenges. With the development of electric locomotives, the power of locomotives gradually increases, the load current of the traction power supply system rises, and may approach the short-circuit current when the traction power supply system fails, resulting in increased difficulty in identifying the short-circuit current and the load current. Therefore, how to improve the current identification accuracy of the traction power supply system has become a problem to be solved. SUMMARY

[0003] Therefore, it is necessary to provide a current identification model training method, a current identification method, a device and equipment to improve the current identification accuracy of the traction power supply system.

[0004] In a first aspect, the present application provides a current identification model training method, comprising:

[0005] obtaining a sample current signal of a traction power supply system and a current category of the sample current signal;

[0006] performing variational mode decomposition on the sample current signal to obtain a plurality of modal components of the sample current signal;

[0007] extracting current features of the sample current signal according to the plurality of modal components;

[0008] training a preset classification model with the current features as input and the current category as label to obtain a current identification model of the traction power supply system.

[0009] In one of the embodiments, performing variational mode decomposition on the sample current signal to obtain a plurality of modal components of the sample current signal comprises:

[0010] optimizing the decomposition layer number and the penalty factor of the variational mode decomposition algorithm with the minimum spatial correlation recursive sample entropy value of the plurality of modal components of the sample current signal as the optimization target to obtain an optimized variational mode decomposition algorithm; and performing noise reduction decomposition on the sample current signal by using the optimized variational mode decomposition algorithm to obtain the plurality of modal components of the sample current signal.

[0011] In one of the embodiments, extracting current features of the sample current signal according to the plurality of modal components comprises:

[0012] Screening effective components from the plurality of modal components; reconstructing the effective components to obtain a reconstructed signal, and performing spatial correlation recursive sample entropy extraction on the reconstructed signal to obtain a current feature of the sample current signal.

[0013] In one of the embodiments, the screening of the effective components from the plurality of modal components comprises:

[0014] Determining an index value of a screening index of each of the plurality of modal components, and screening a modal component with an index value satisfying a screening threshold from the plurality of modal components as an effective component; wherein the screening index comprises at least one of a complexity index, an energy index, a correlation index and a frequency index.

[0015] In one of the embodiments, the obtaining of the sample current signal of the traction power supply system and the current category of the sample current signal comprises:

[0016] Using a simulation model of the traction power supply system to simulate the traction network short-circuit fault to obtain a plurality of abnormal current signals as sample current signals with a short-circuit current category; and using the traction network simulation model to simulate a normal operation state of the powered train to obtain a plurality of normal current signals as sample current signals with a load current category.

[0017] In a second aspect, the application further provides a current identification method, comprising:

[0018] Obtaining a current feature of a current signal of a traction power supply system for powering a train;

[0019] Inputting the current feature of the current signal into a current identification model of the traction power supply system to obtain a current category of the current signal; wherein the current identification model is obtained by each method embodiment of the first aspect.

[0020] In a third aspect, the application further provides a current identification model training device, comprising:

[0021] A sample obtaining module is configured to obtain a sample current signal of a traction power supply system and a current category of the sample current signal;

[0022] A sample decomposition module is configured to perform variational modal decomposition on the sample current signal to obtain a plurality of modal components of the sample current signal;

[0023] A feature extraction module is configured to extract a current feature of the sample current signal according to the plurality of modal components;

[0024] A model training module is configured to train a preset classification model by taking the current feature as input and taking the current category as label to obtain a current identification model of the traction power supply system.

[0025] In a fourth aspect, the present application provides a current identification device, comprising:

[0026] a current acquisition module, configured to acquire a current feature of a current signal of a traction power supply system for powering a train;

[0027] a current identification module, configured to input the current feature of the current signal into a current identification model of the traction power supply system to obtain a current category of the current signal; wherein the current identification model is obtained by the method embodiments of the first aspect.

[0028] In a fifth aspect, the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps in the method embodiments of the first aspect and / or the second aspect when executing the computer program.

[0029] In a sixth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps in the method embodiments of the first aspect and / or the second aspect.

[0030] In a seventh aspect, the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the steps in the method embodiments of the first aspect and / or the second aspect.

[0031] The above-mentioned current identification model training, current identification method, device and equipment can obtain the sample current signal of the traction network power supply system and the current category of the sample current signal, and perform variational modal decomposition on the sample current signal to obtain multiple modal components of the sample current signal; in this way, the current characteristics of the sample current can be extracted based on the above-mentioned multiple modal components, and then the preset classification model can be trained with the current characteristics as input and the current category as a label to obtain the current identification model of the traction network power supply system. In this way, performing variational modal decomposition on the sample current signal can effectively suppress the modal aliasing between the obtained modal components, improve the anti-noise ability to reduce the interference of noise in the sample current signal on the modal components, and improve the stability of the modal components. Therefore, the effectiveness and accuracy of the current characteristics of the sample current signal extracted according to multiple modal components can be improved, and then the recognition accuracy of the current identification model trained according to the current characteristics of the sample current signal can be improved, so as to improve the recognition accuracy of the trained current identification model in identifying the current of the traction network power supply system, effectively reduce the probability of misjudgment and missed judgment, and provide a reliable basis for fault diagnosis of the traction network power supply system, so as to be able to timely identify the short-circuit current of the traction network power supply system, shorten the fault diagnosis time, help to quickly take corresponding fault handling measures, reduce the impact of the fault on the traction power supply system and the normal operation of rail transit, and reduce the economic losses and safety risks caused by the fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 A flowchart of a current identification model training method provided in some embodiments of the present application;

[0034] Figure 2 A schematic diagram of a process for determining multiple modal components provided in some embodiments of the present application;

[0035] Figure 3A Iterative graph of the optimization process using HHO provided in some embodiments of the present application;

[0036] Figure 3B This is a schematic diagram of the change process of the optimization target during the optimization process provided in some embodiments of the present application.

[0037] Figure 3CA schematic diagram of the change of the decomposition layer number of the variational modal decomposition algorithm with the iteration number in the optimization process provided by some embodiments of the present application;

[0038] Figure 3D A schematic diagram of the change of the penalty factor of the variational modal decomposition algorithm with the iteration number in the optimization process provided by some embodiments of the present application;

[0039] Figure 3E A schematic diagram of the plurality of modal components of the sample current signal decomposed by the optimized variational modal decomposition algorithm provided by some embodiments of the present application;

[0040] Figure 4 A schematic diagram of the process of extracting current features provided by some embodiments of the present application;

[0041] Figure 5 A schematic diagram of the process of training the current recognition model provided by some embodiments of the present application;

[0042] Figure 6 A schematic diagram of the process of the current recognition method provided by some embodiments of the present application;

[0043] Figure 7 A block diagram of the structure of the current recognition model training device provided by some embodiments of the present application;

[0044] Figure 8 A block diagram of the structure of the current recognition device provided by some embodiments of the present application;

[0045] Figure 9 An internal structure diagram of the computer device provided by some embodiments of the present application. DETAILED DESCRIPTION

[0046] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0047] The traction power supply system provides power for the safe and reliable operation of the electric locomotive. With the development of rail transit train technology, the safe and reliable operation of the traction power supply system is facing challenges. With the development of rail transit train technology, the train power gradually increases, the load current of the traction power supply system rises, and may approach the short-circuit current when the traction power supply system fails, resulting in an increase in the difficulty of identifying the short-circuit current and the load current. Therefore, how to improve the current recognition accuracy of the traction power supply system has become a problem to be solved.

[0048] Based on this, in order to solve the above technical problems, in one exemplary embodiment, a current recognition model training method is provided, which can be applied to a computer device, which can be a server or a terminal. Wherein, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services; the terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, etc. The following takes the method applied to the server as an example for illustration. As shown in Figure 1 The method can include the following steps:

[0049] S101, obtaining sample current signals of a traction network power supply system and current categories of the sample current signals.

[0050] Wherein, the traction network power supply system is a special power system for providing power energy for rail transit (such as high-speed railway, intercity railway, subway, tram, etc.) trains, and its core function is to efficiently and stably transmit the power energy of the power system to the moving trains and form a complete current loop. Generally, the traction network power supply system includes a power grid, a substation, a traction network, an AT (Auto-transformer Substation, Auto-transformer Substation) and a partition station.

[0051] In order to train a current recognition model for current recognition of the traction network power supply system, the sample current signals of the traction network power supply system and the current categories of the sample current signals can be obtained first. Wherein, the number of sample current signals is multiple.

[0052] Optionally, an experimental platform can be built to simulate different working conditions (normal, short circuit fault, etc.) of the traction network power supply system, and collect current signals under the simulated working conditions as sample current signals, and determine the current categories of the sample current signals according to the simulated working conditions.

[0053] Optionally, a current detection device can be installed in the traction network power supply system in actual operation to collect current signals under real working conditions as sample current signals, and determine the current categories of the sample current signals one by one.

[0054] Optionally, the sample current signals and the current categories of the sample current signals can be obtained through a public current signal data set.

[0055] In an optional embodiment, S101 can include: simulating the traction power supply system short-circuit fault by using a simulation model of the traction power supply system to obtain a plurality of abnormal current signals as sample current signals of the current category of short-circuit current; and simulating the normal operation state of the powered train by using the traction power supply system simulation model to obtain a plurality of normal current signals as sample current signals of the current category of load current.

[0056] In this embodiment, a simulation model of the traction power supply system can be built in advance, and current signals can be generated by simulation software to simulate the current signals of the traction power supply system in the short-circuit fault state and in the normal power supply state of the powered train to make the train run normally. In this way, a plurality of abnormal current signals in the case of simulating the traction power supply system short-circuit fault can be obtained as sample current signals of the current category of short-circuit current, and a plurality of normal current signals in the case of simulating the normal operation state of the powered train can be obtained as sample current signals of the current category of load current. Thus, sample current signals including short-circuit current and load current are obtained.

[0057] Optionally, Simulink and MATLAB (Matrix Laboratory) can be used to simulate and build the traction power supply system to obtain a simulation model of the traction power supply system.

[0058] Simulink is a visual simulation and modeling tool mainly used for modeling, simulation, analysis and code generation of dynamic systems (such as control systems, power systems, communication systems, mechanical systems, etc.). MATLAB is deeply integrated, supporting graphical drag-and-drop operations. MATLAB is a high-level technical computing language and interactive environment mainly used for algorithm development, data visualization, data analysis and numerical calculation, which takes matrix operations as the core and integrates a large number of professional toolboxes.

[0059] S102, variational mode decomposition is performed on the sample current signal to obtain a plurality of modal components of the sample current signal.

[0060] Variational Mode Decomposition (VMD) is an adaptive signal decomposition method for decomposing complex multi-component signals into a series of intrinsic mode function (IMF) components with clear physical meaning. In VMD, the IMF component can be referred to as a modal component, and different IMF components are independent of each other. Each IMF component has a finite bandwidth and a center frequency, and the superposition of all IMF components is equal to the original signal.

[0061] Variational mode decomposition (VMD) has the following characteristics: the number of decomposition levels (k), the penalty factor (α), and the convergence tolerance (τ). The number of decomposition levels (also known as the modal number) is the number of IMF components to be decomposed; the penalty factor controls the bandwidth of the decomposed IMF components; a larger α results in a narrower bandwidth; and the convergence tolerance (also known as the convergence threshold) is the terminating condition for the iteration (the threshold for the difference between the objective function values ​​of two iterations).

[0062] In the variational modal decomposition process of the sample current signal, the minimum bandwidth and value of the multiple modal components of the obtained sample current are taken as the objective function, and the final multiple modal components are obtained through multiple iterations.

[0063] Specifically, the decomposition level k, penalty factor α and convergence tolerance τ are set; based on the decomposition level k and penalty factor α, the sample current signal is subjected to variational modal decomposition to obtain k modal components, and each modal component u k The initial center frequency ω k ; For each modal component u k , solve the optimal solution (minimize bandwidth) by Fourier transform in the frequency domain; according to each modal component u k The energy center of each modal component u is updated k ω k ; Calculate the difference in modal components between two adjacent iterations. If it is less than the convergence tolerance τ, stop the iteration and output k modal components.

[0064] For example, assume that a sample current signal contains: 50Hz fundamental wave, 150Hz (3rd harmonic), 250Hz (5th harmonic) and Gaussian white noise (SNR 20dB), the sampling frequency is 1000Hz, and the signal length is 2048 points. Then the decomposition level k = 3, and the penalty factor α = (1000 / 2) 2 =250000, convergence tolerance τ = 10 -7 The three IMF components obtained by decomposition are as follows:

[0065] IMF1: The time domain waveform is a 50Hz sine wave, and the spectrum peak is concentrated at 50Hz (fundamental component);

[0066] IMF2: The time domain waveform is a 150Hz sine wave, and the spectrum peak is concentrated at 150Hz (3rd harmonic);

[0067] IMF3: The time domain waveform is a 250Hz sine wave, and the spectrum peak is concentrated at 250Hz (5th harmonic).

[0068] S103: extracting current characteristics of the sample current signal according to the multiple modal components.

[0069] As mentioned above, each IMF component after VMD decomposition corresponds to a "sub-signal" of a certain frequency range in the original current, and the time domain, frequency domain and energy characteristics of each IMF component can be combined to comprehensively reflect the physical nature of the original current signal (such as amplitude fluctuation, frequency distribution, energy concentration, etc.). Therefore, after obtaining the plurality of modal components of the sample current signal, the current features of the sample current signal can be extracted according to the plurality of modal components.

[0070] Optionally, the energy of each IMF component is calculated, and the energy weight of each IMF component is determined according to the proportion of the energy of each IMF component in the total energy of all IMF components, so as to highlight the components that have important contribution to the current features. Then, the energy entropy (Energy Entropy) and sample entropy (Sample Entropy, SampEn) of each IMF component are calculated. The energy entropy reflects the energy distribution of the IMF component, and the sample entropy measures the complexity and irregularity of the IMF component. The entropy values of the above-mentioned energy entropy and sample entropy can effectively represent the features of the IMF component, and provide an important basis for current recognition. The energy weight, energy entropy and sample entropy of each IMF component are combined to form a feature vector that can comprehensively represent the signal features of the sample current signal, as the current features of the sample current signal.

[0071] S104, training a preset classification model by taking the current features as input and taking the current categories as labels to obtain the current recognition model of the traction power supply system.

[0072] After obtaining the current features of the sample current signal, the preset classification model can be trained by taking the current features as input and taking the current categories as labels. In the training process of the above-mentioned preset model, the preset model learns the relationship between the current features and the current categories, establishes an identification mode for identifying the current categories based on the current features, and adjusts the parameters to gradually approach the current categories of the sample current signal until the loss value between them meets the preset loss value condition, for example, the loss value between them is less than the preset threshold, to obtain the current recognition model of the traction power supply system.

[0073] Among them, the above-mentioned preset classification models can be various types of decision trees (Decision Tree), support vector machines (SVM), K-nearest neighbors (KNN), random forests (RandomForest), gradient boosting trees (GBDT, Gradient Boosting Decision Tree), convolutional neural networks (CNN), recurrent neural networks (RNN, Recurrent Neural Network) and other models, without specific limitation.

[0074] Optionally, the preset classification model can be MKRVM (Multiple-Kernel Relevance Vector Machine). MKRVM is a machine learning model based on the Relevance Vector Machine (RVM). By introducing multiple kernel learning (MKL) technology, the model's ability to express data features and classification prediction performance is improved. RVM is considered a probability-based SVM that uses Bayesian reasoning to solve classification and regression problems.

[0075] Specifically:

[0076] For a given input vector X={x1,x2,…,x i ,…,x N} (current characteristics of the sample current signal) and the corresponding output target category t={t1,t2,…,t i ,…,t N}(current category of sample current signal), then the regression model of sample (current feature of sample current signal) is as follows:

[0077]

[0078] Where N is the total number of sample current signals, ε is the 2 ) The sample deviation of the normal distribution, ω i is the weight of the i-th related vector, K(x,x i ) is the kernel function for solving the model, and x is the related vector.

[0079] Assuming that the output categories t are independently distributed, the likelihood function of the training sample is:

[0080]

[0081] Where ω is the weight vector; Φ is an N×(N+1) matrix, as shown below:

[0082]

[0083] Then each ω presents a normal distribution with a mean of 0, and introduces the hyperparameter α=(α1,α2,…,α N ) T , transforming the problem of solving ω into the problem of solving α, the expression is as follows:

[0084]

[0085] Among them, α is the N+1-dimensional hyperparameter vector corresponding to ω.

[0086] According to the Bayesian empirical formula, the posterior probability of the unknown parameter is as follows:

[0087] P(ω,α,σ 2 |t)=p(ω|α,σ 2 ,t)p(α,σ 2 |t)

[0088] According to the Markov property (no aftereffect), the mathematical expression for probability prediction of target category t and input vector x is as follows:

[0089] P(t * |t)=∫p(t * |ω,α,σ 2 )p(ω,α,σ 2 |t)dωdαdσ 2

[0090] Among them, t * is the estimated value of t, P(·) is the probability function, and p(·) is the conditional probability function.

[0091] The posterior probability distribution calculation formula of the final weight vector is as follows:

[0092]

[0093] Where ∑ is the covariance of the posterior weighted probability, and μ is the mean of the posterior weighted probability. The calculation formulas for ∑ and μ are as follows:

[0094]

[0095] Where A is a diagonal matrix, A=diag(α1,α2,…,α N ).

[0096] To unify the hyperparameters, define P(t|a, s 2 ) as follows:

[0097] P(t|a, s 2 ) = (2p) -N / 2 |s 2 I + Fa -1 F T | -1 / 2 exp[-0.5t T (s 2 I + Fa -1 F T ) -1 t]

[0098] Where I is the unit matrix, I =∑ -1 ∑.

[0099] Through the concept of maximum likelihood boundary, a and s 2 are iteratively estimated, and the iteration formula is as follows:

[0100]

[0101] Where∑ i is the ith diagonal element of the covariance∑ of the posterior weight probability, andμ i is the ith diagonal element of the meanμ of the posterior weight probability. And (s 2 ) new are the optimal iteration values of a and s 2 .

[0102] Optionally, a Gaussian kernel function (Gaussian kernel function) with strong globality and wide convergence domain, and a Polynomial kernel function (Polynomial kernel function) with strong locality and good nonlinear approximation ability are used to construct the kernel function of the MKRVM model through weighted mixing. Further, the mathematical expression of the mixed kernel function is as follows:

[0103] K mix (x, x') = λK gauss (x, x') + (1-λ)K ploy (x, x')

[0104] Where K mix (x, x') is the mixed kernel function, K gauss (x, x') is the Gaussian kernel function, K ploy (x, x') is the Polynomial kernel function, and λ is the kernel function mixing weight coefficient.

[0105] Where the calculation formula of the Gaussian kernel function is as follows:

[0106]

[0107] wherein γ is the bandwidth of the Gaussian kernel, x' is the center of the kernel function, and ||x-x' || is the Euclidean distance between the vector x and the vector x'. 2 is the two-norm of the vector x and the vector x'.

[0108] The polynomial kernel function is calculated according to the following formula:

[0109] K ploy (x,x') = (βx T x' + c) d

[0110] wherein β is the scaling factor of the polynomial kernel, c is the offset of the polynomial kernel, and d is the order of the polynomial kernel.

[0111] Therefore, the use of the MKRVM can improve the robustness of the obtained current identification model to noise, interference and other factors, so that the short-circuit fault current can be stably and accurately identified under the complex and variable traction network operation environment, and the reliability and stability of the system can be improved.

[0112] In the training of the current identification model, by obtaining the sample current signal of the traction network power supply system and the current category of the sample current signal, the sample current signal can be subjected to variational modal decomposition to obtain multiple modal components of the sample current signal. In this way, the current features of the sample current can be extracted according to the multiple modal components, and then the pre-set classification model can be trained with the current features as the input and the current category as the label to obtain the current identification model of the traction network power supply system. In this way, the variational modal decomposition of the sample current signal can effectively suppress the modal aliasing between the obtained modal components, improve the anti-noise capability to reduce the interference of noise in the sample current signal on the modal components, and improve the stability of the modal components, so that the effectiveness and accuracy of the current features of the sample current signal extracted according to the multiple modal components can be improved, and then the identification accuracy of the current identification model trained according to the current features of the sample current signal can be improved, so as to improve the identification accuracy of the current identification model trained to identify the current of the traction network power supply system, effectively reduce the probability of misjudgment and omission, provide a reliable basis for fault diagnosis of the traction network power supply system, so as to identify the short-circuit current of the traction network power supply system in time, shorten the fault diagnosis time, help to take corresponding fault handling measures quickly, reduce the impact of the fault on the normal operation of the traction power supply system and the rail transit, and reduce the economic loss and safety risk caused by the fault.

[0113] On the basis of the above embodiments, in an exemplary embodiment, the determination of the plurality of modal components is further refined, which can optionally include the following steps: Figure 2 As shown, can include the following steps:

[0114] S201, with the minimum value of the spatial correlation recursive sample entropy sum of the plurality of modal components of the sample current signal as the optimization target, optimizing the decomposition layer number and the penalty factor of the variational modal decomposition algorithm to obtain an optimized variational modal decomposition algorithm.

[0115] The so-called spatial correlation recursive sample entropy (Spatial Correlation Recursive Sample Entropy) is a composite index combining spatial correlation, recursive analysis and sample entropy, mainly used to quantify the internal regularity and mutual independence of multiple signal components (such as the plurality of modal components obtained by variational modal decomposition). Among them, the smaller the value of the spatial correlation recursive sample entropy of the modal component, the stronger the component regularity of the modal component, so that the noise is smaller, and the independence between different modal components is higher, and there is no information redundancy. Therefore, with the minimum value of the spatial correlation recursive sample entropy sum of the plurality of modal components of the sample current signal as the optimization target, the decomposition layer number and the penalty factor of the variational modal decomposition algorithm can be optimized, so that the recursive complexity of the plurality of modal components obtained is low (i.e. strong self regularity) and mutually independent (i.e. no redundant information), so that the decomposition effect of the optimized variational modal decomposition algorithm on the sample current signal is optimal.

[0116] Optionally, the Harris Hawk Optimization (Harris Hawks Optimization, HHO) can be used to optimize the decomposition layer number and the penalty factor of the variational modal decomposition algorithm with the minimum value of the spatial correlation recursive sample entropy sum of the plurality of modal components of the sample current signal as the optimization target.

[0117] HHO is a bionic intelligent optimization algorithm inspired by the hunting behavior of Harris's Hawks in nature. By simulating the strategy of Harris's Hawk colony cooperative hunting, the algorithm realizes efficient solution to complex optimization problems, has the characteristics of strong optimization ability, few parameters, wide adaptability, etc., and is widely used in function optimization, engineering design, machine learning, etc.

[0118] Among them, HHO adopts the mechanism of exploration and development, and the hunting stage of the Harris's Hawk colony is mainly divided into three parts: search stage, search and development conversion stage and development stage. Assuming that the population size of the Harris's Hawk colony is M, and the spatial dimension of solving a certain problem is D, i.e. the position (parameter combination of decomposition layer number and penalty factor) of the i-th hawk in the dimension space is i = 1, 2, 3, …, M. These hawks are candidate solutions, and the global optimal solution is the position of the prey (the optimal Harris hawk, i.e., the optimal combination of the number of decomposition layers and the penalty factor) In particular:

[0119] 1) The search phase includes: the individual of the Harris hawk colony can stay anywhere and can capture the prey by two identical probability methods. The following is the detailed mathematical model:

[0120]

[0121] Where X(t), X(t+1) are the current and next iteration positions of the individual; t is the iteration number; X rand (t) is a random individual position; X rabbit (t) is the prey position; r1, r2, r3, r, q are random numbers between 0 and 1; q is used to randomly select the strategy to be adopted; X m (t) is the average position of the individual.

[0122] 2) The conversion phase between the search phase and the development phase includes: according to the escape energy E of the prey, the conversion between the search phase and the development behavior of different development nodes. When |E|≥1, enter the search phase; when |E|<1, enter the development phase. Wherein, the formula of escape energy is:

[0123]

[0124] Where E0 is the initial energy of the prey, and is a random number between -1 and 1, which is updated automatically each iteration; T max is the maximum number of iterations.

[0125] 3) The development phase includes: in this phase, assume that r is the prey's chance to escape before the attack, when r<0.5, it is a successful escape; when r≥0.5, it is an unsuccessful escape.

[0126] When r≥0.5 and |E|≥0.5, a soft siege strategy is adopted for position update, and the formula is:

[0127] X(t+1) = ΔX(t) - E|JX rabbit (t) - X(t)

[0128] ΔX(t) = X rabbit (t) - X(t)

[0129] Where ΔX(t) represents the difference between the prey position and the current position of the individual; J is a random number between 0 and 2;

[0130] When r≥0.5 and |E|<0.5, a soft siege strategy is adopted for position updating, and the formula is:

[0131] X(t+1)=X rabbit (t)-E|ΔX(t)|

[0132] When r<0.5 and |E|≥0.5, the following two strategies are implemented. When the first strategy is invalid, the second strategy is executed. Wherein, D is the dimension, S is a D-dimensional random vector, and LF is the Levy flight function; the position updating strategy is as follows:

[0133]

[0134] When r<0.5 and |E|<0.5, the position updating formula is as follows:

[0135]

[0136] Further, on the basis of the above-mentioned optimization of the number of decomposition layers and the penalty factor of the variational mode decomposition algorithm using HHO, in an optional embodiment, a Sine chaotic system can be introduced into HHO to replace the random initialization of the population, and an energy periodicity decreasing regulation method can be introduced to control the local and global search ability of the algorithm, and then a simulated annealing algorithm can be introduced for the position of the prey in each iteration.

[0137] HHO is a meta-heuristic algorithm, and its performance is also affected by the initial population. If the initial population is randomly irregularly distributed, the convergence speed of HHO will be affected to a certain extent. Therefore, in order to ensure the efficiency of HHO, it is required that the distribution of the eagle population in the search space is relatively uniform, so as to have higher ergodicity and diversity. Therefore, in order to improve the performance of the swarm intelligence algorithm, a chaotic mapping is used to initialize the population.

[0138] Optionally, a Sine mapping (Sine map) is introduced to initialize the population. The Sine mapping is a unimodal mapping, and the value range is [0, 1], and the formula is as follows:

[0139] x k+1 =μ(sinπx k )

[0140] Wherein, x k+1 is an iteration sequence value, k is a non-negative integer; μ is a system parameter, μ∈[0, 1], however, when μ∈[0.87, 0.93] and [0.95, 1], chaotic phenomenon will occur.

[0141] As mentioned above, the phase transition between the search phase and the development phase of HHO is determined by the escape energy E. When the escape energy |E|≥1, the Harris Hawks search different areas to further explore the position of the prey, which corresponds to the global search phase; when |E|<1, the Harris Hawks perform local exploration on the adjacent solutions, thus corresponding to the local development phase.

[0142] Further, in order to accurately describe the phenomenon that the Harris Hawk population cooperates to capture the prey, a nonlinear cosine function formula can be introduced to replace the original linear escape energy formula:

[0143]

[0144] where k is the number of decreasing periods of the prey energy E, and k can be any positive integer. However, the larger k is, the more the formula focuses on local search, and the more frequent the search and development phase transition is; on the contrary, the smaller k is, the more it focuses on global search. Therefore, the size of the period number k should be reasonably selected, for example, k=8.

[0145] In HHO, the position of the optimal Harris Hawk generated in each iteration, i.e. the position of the prey, does not exchange information with other individuals in the population, so it is easy to fall into local optimum. Therefore, in each iteration process, the prey position and the suboptimal position should be selected at the same time, and then the simulated annealing algorithm is introduced to exchange information between the prey position and the suboptimal position. The core is to update and mutate the position of the suboptimal solution obtained in each iteration by using the simulated annealing algorithm. If the fitness value of the mutated suboptimal position is better than that of the optimal position, the optimal position is updated to the global optimal position, and it is regarded as the prey position; otherwise, the simulated annealing algorithm accepts the suboptimal solution as the global optimal solution with a certain probability to prevent HHO algorithm from falling into local optimum, and the probability formula is:

[0146]

[0147] where f(i) is the fitness value of the i-th Harris Hawk; T0 is the initial temperature.

[0148] Exemplarily, Figure 3A is an iterative curve diagram of the optimization process using HHO, where the fitness value is the spatial correlation recursive sample entropy sum value of multiple modal components, the iteration curve is the iteration change curve of the fitness value, and the iteration number is the iteration number of the optimization process using HHO; for example, Figure 3B is a variation process diagram of the optimization target (spatial correlation recursive sample entropy sum value of multiple modal components) in the optimization process, where the objective function value is the spatial correlation recursive sample entropy sum value of multiple modal components; Figure 3C is a diagram showing the change of the number of decomposition layers of the variational modal decomposition algorithm with the iteration number in the optimization process; Figure 3DFor optimization, the penalty factor of the variational modal decomposition algorithm changes with the number of iterations. Figure 3E For the schematic diagram of the plurality of modal components of the sample current signal decomposed by the optimized variational modal decomposition algorithm, wherein the original data is the sample current signal.

[0149] S202, using the optimized variational modal decomposition algorithm, the sample current signal is decomposed to obtain a plurality of modal components of the sample current signal.

[0150] Among them, the sample current signal usually contains effective components (such as fundamental wave, characteristic harmonic, fault transient signal) and noise components (such as electromagnetic interference, measurement noise), and the optimal decomposition layer and penalty factor in the optimized variational modal decomposition algorithm can ensure that the effective components in the sample current signal are decomposed into independent low complexity modal (recursive sample entropy is small), and the noise components in the sample current signal are assigned to specific high frequency modal components due to strong randomness (high recursive sample entropy), and noise reduction can be achieved by threshold method (such as removing high entropy modal).

[0151] In this embodiment, the complexity and correlation of the modal components are used to define the optimization target, and the VMD parameters (decomposition layer and penalty factor) suitable for the current signal are obtained; then the optimized algorithm is used to realize accurate splitting and noise reduction of the signal, and finally a reliable modal component basis is provided for the type identification of the current signal.

[0152] On the basis of the above embodiment, in an exemplary embodiment, the extraction of current characteristics is further refined, which can include the following steps: Figure 4 As shown, the steps can include:

[0153] S401, screening the effective components in the plurality of modal components.

[0154] As described above, the sample current signal contains effective components and noise components, so in the plurality of modal components of the obtained sample current signal, there can be modal components that contribute less to the current characteristics, so the plurality of modal components of the obtained sample current signal can be screened to obtain the effective components in the plurality of modal components.

[0155] In an optional embodiment, the way to screen the effective components in the plurality of modal components can be to determine the index value of the screening index of each modal component in the plurality of modal components; wherein the screening index includes at least one of the complexity index, the energy index, the correlation index and the frequency index; and the modal components in the plurality of modal components whose index values meet the screening threshold are screened as effective components.

[0156] The complexity index can include sample entropy (SampEn), spatial correlation recursive sample entropy (SDRSEn), approximate entropy (ApEn), etc. The effective component usually has moderate complexity, and the non-effective component (such as a noise component) has high or low complexity. The energy index represents the energy proportion of the modal component. The energy of the effective component is usually high, and the energy of the non-effective component (such as a noise component) is low. The correlation index can include a correlation coefficient with the original signal. The effective component has stronger correlation with the sample current signal. The frequency domain index can include the energy proportion of the effective frequency band in the power spectrum. The sample current signal is usually concentrated in a specific frequency band, and the energy of the effective component in the specific frequency band is higher.

[0157] Further, when the number of screening indexes is one, a single index threshold can be set as a screening threshold. For example, the modal component with a SDRSEn value in the interval [0.3, 0.8] is screened as an effective component, and the modal component with an energy proportion greater than 5% is screened as an effective component. When the number of screening indexes is multiple, a combined index threshold can be set as a screening threshold. For example, the modal component that satisfies both “energy proportion > 3%” and “correlation coefficient with the sample current signal > 0.6” is screened as an effective component.

[0158] S402, reconstructing the effective component to obtain a reconstructed signal, and performing spatial correlation recursive sample entropy extraction on the reconstructed signal to obtain the current feature of the sample current signal.

[0159] Optionally, the extraction result of the spatial correlation recursive sample entropy extraction can include at least one of the SDRSEn value itself, multi-scale SDRSEn, time evolution of SDRSEn, and deviation of SDRSEn from a threshold. The SDRSEn value itself is used to reflect the overall complexity of the reconstructed signal (the larger the value, the more irregular the reconstructed signal); the multi-scale SDRSEn is calculated at different time scales to capture the multi-scale characteristics (such as interference of different frequency components) of the reconstructed signal; the time evolution of SDRSEn is used to analyze the trend of SDRSEn over time (such as sudden change of SDRSEn when a short-circuit fault occurs in the traction power supply system); and the deviation of SDRSEn from a threshold: set the SDRSEn threshold of the normal state, calculate the deviation degree of the current signal, and use it to mark the fault.

[0160] In an optional embodiment, the effective components obtained by screening can be linearly superimposed to restore the main current feature of the sample current signal, to obtain a reconstructed signal, and then the spatial correlation recursive sample entropy extraction is performed on the reconstructed signal to obtain the current feature of the sample current signal.

[0161] In another alternative embodiment, the plurality of effective components can constitute a time series signal X = {X(l), X(2),..., X(N)} with a time length of N. The effective component X(i) can be constructed into an m-dimensional vector in the reconstruction space as a reconstructed signal by means of coordinate delay. The expression of the reconstructed vector is as follows:

[0162] X(i) = {x(i), x(i+τ),..., x[i+τ(m-1)]}, (1≤i≤N-τ(m-1))

[0163] wherein τ is a delay factor, m is the reconstruction dimension, and x(i) is the i-th state of the effective component X(i) in the m-dimensional space.

[0164] wherein the recurrence graph can be constructed as an N x N matrix, wherein X i and X j are the i-th and j-th effective components of the image I, respectively.

[0165]

[0166] wherein H(·) is the Heaviside function, d(X i , X j ) is the maximum distance between X i and X j , and r is a spatial similarity threshold.

[0167] Further, a binary co-occurrence matrix (BCM) of the image I is constructed according to the above-mentioned recurrence graph, wherein the image I is a binary image. Each pixel in the image (located at (x, y)) can only take the value of 0 or 1. Generally, in image processing, a binary image is obtained by thresholding the original image, wherein 0 can represent the background (such as a non-target region), and 1 represents the target (such as an edge, a texture region, or a specific structure). The size of the obtained binary co-occurrence matrix is N x M. The construction of the BCM is mainly based on the concept of the gray level co-occurrence matrix, and the mathematical definition is as follows:

[0168]

[0169] wherein δ x and δ x are the offsets of the pixel positions in the x-axis and y-axis directions, respectively, I(x, y) is the pixel located at (x, y) in the image I, I(x+δ x , y+δy ) in the image I, p and q are 0 or 1, and the symbol ∧ represents a logical AND operation. x y

[0170] According to the recursive matrix of the recursive graph and the binary co-occurrence matrix of the image I, the m-dimensional vector reconstructed from each effective component can be converted into a BCM. In order to determine the similarity of different effective components in the reconstruction space, the probability of the total number of relevant recursions is defined as follows, under the condition that the Euclidean distance between the m-dimensional vectors reconstructed from two effective components is less than or equal to r, and the offset is δ:

[0171]

[0172] where B δ (1,1) is the number of pixel pairs (the number of component pairs formed by two effective components) that meet the distance requirement and the offset requirement, and N(δ) is the total number of pixel pairs (the number of component pairs formed by all effective components) that meet the offset requirement.

[0173] Further, the reconstruction dimension m is increased by 1, and the above process is repeated. When the difference between the total number of relevant recursions obtained in two consecutive times is less than a preset threshold, the reconstruction dimension m of the previous time is taken as the final reconstruction dimension. Furthermore, in combination with the concept of information entropy, the calculation formula of SDRSEn is as follows:

[0174]

[0175] For example, the calculation formula of SDRSEn is m = 4, r = 0.2, and δ = [0, 1].

[0176] Based on the above embodiments, an exemplary embodiment can include the following steps, as shown in Figure 5

[0177] S501, using a simulation model of a traction network power supply system to simulate a traction network short-circuit fault, to obtain a plurality of abnormal current signals as sample current signals of the current category of short-circuit current; and using the traction network simulation model to simulate the normal operating state of the supplied train, to obtain a plurality of normal current signals as sample current signals of the current category of load current.

[0178] S502, taking the minimum value of the spatially related recursive sample entropy of the plurality of modal components of the sample current signal as the optimization objective, optimizing the decomposition layer number and the penalty factor of the variational modal decomposition algorithm, to obtain an optimized variational modal decomposition algorithm.

[0179] ​​​S503, performing denoising decomposition on the sample current signal by using an optimized variational modal decomposition algorithm to obtain multiple modal components of the sample current signal.

[0180] S504, screening effective components from the multiple modal components.

[0181] S505, reconstructing the effective components to obtain a reconstructed signal, and performing spatial correlation recursive sample entropy extraction on the reconstructed signal to obtain a current feature of the sample current signal.

[0182] S506, training a preset classification model by taking the current feature as input and taking a current category as a label to obtain a current recognition model of the traction power supply system.

[0183] The specific implementation manners of S501-S506 are the same as those in the method embodiments, and will not be described here.

[0184] Based on the same inventive concept, the embodiments of the present application also provide a current recognition method based on the current recognition model training method described above. The method can be applied to a computer device, which can be a server or a terminal. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services; the terminal can be, but is not limited to, various personal computers, notebook computers, smartphones, tablet computers, etc.

[0185] Next, the current recognition method will be described by taking the application of the method to a server as an example. As shown in Figure 6 the method can include the following steps:

[0186] S601, obtaining a current feature of a current current signal of a traction power supply system for supplying power to a train.

[0187] When it is necessary to recognize the current current signal of the traction power supply system for supplying power to a train, the current feature of the current current signal can be obtained.

[0188] The current feature of the current current signal is obtained in the same way as the current feature of the sample current signal, that is, the current current signal is subjected to variational modal decomposition to obtain multiple modal components of the current current signal, and the current feature of the current current signal is extracted according to the multiple modal components of the current current signal.

[0189] S602, inputting the current feature of the current current signal into the current recognition model of the traction power supply system to obtain a current category of the current current signal.

[0190] The current recognition model is obtained based on the current recognition model training method.

[0191] The power recognition model trained by the power recognition model training method can improve the recognition accuracy of the current of the traction power supply system, effectively reduce the probability of misjudgment and missed judgment, provide a reliable basis for fault diagnosis of the traction power supply system, and enable the short-circuit current of the traction power supply system to be recognized in a timely manner, shorten the fault diagnosis time, help to quickly take corresponding fault handling measures, reduce the impact of the fault on the traction power supply system and the normal operation of the rail transit, and reduce economic losses and safety risks caused by the fault.

[0192] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or steps or stages in other steps. It can be understood that the steps in different embodiments can be freely combined as needed, and various non-contradictory schemes formed by the combination are within the scope of protection of the present application.

[0193] Based on the same inventive concept, the present application also provides a current recognition model training device for implementing the above-mentioned current recognition model training method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more current recognition model training device embodiments provided below can refer to the limitations of the current recognition model training method in the foregoing, which will not be repeated here.

[0194] In one exemplary embodiment, as shown in Figure 7 a current recognition model training device is provided, comprising: a sample acquisition module 710, a sample decomposition module 720, a feature extraction module 730, and a model training module 740, wherein:

[0195] The sample acquisition module 710 is configured to acquire a sample current signal of the traction power supply system and a current category of the sample current signal.

[0196] The sample decomposition module 720 is configured to perform variational modal decomposition on the sample current signal to obtain a plurality of modal components of the sample current signal.

[0197] The feature extraction module 730 is configured to extract a current feature of the sample current signal according to the plurality of modal components.

[0198] The model training module 740 is configured to train a preset classification model by taking the current feature as input and taking the current category as label to obtain a current recognition model of the traction power supply system.

[0199] In an exemplary embodiment, the sample decomposition module 720 is specifically configured to:

[0200] The decomposition layer number and the penalty factor of the variational modal decomposition algorithm are optimized with the minimum value of the spatial correlation recursive sample entropy of the plurality of modal components of the sample current signal as an optimization target to obtain an optimized variational modal decomposition algorithm, and the sample current signal is decomposed by using the optimized variational modal decomposition algorithm to obtain the plurality of modal components of the sample current signal.

[0201] In an exemplary embodiment, the feature extraction module 730 includes:

[0202] The component screening unit is configured to screen effective components from the plurality of modal components.

[0203] The feature extraction unit is configured to reconstruct the effective components to obtain a reconstructed signal, and perform spatial correlation recursive sample entropy extraction on the reconstructed signal to obtain the current feature of the sample current signal.

[0204] In an exemplary embodiment, the component screening unit is specifically configured to:

[0205] The component screening unit is configured to determine an index value of a screening index of each modal component from the plurality of modal components, and screen a modal component with an index value satisfying a screening threshold from the plurality of modal components as an effective component, wherein the screening index includes at least one of a complexity index, an energy index, a correlation index, and a frequency index.

[0206] In an exemplary embodiment, the sample acquisition module 710 is specifically configured to:

[0207] The simulation model of the traction power supply system is used to simulate the traction network short-circuit fault to obtain a plurality of abnormal current signals as sample current signals with the current category being short-circuit current, and the traction network simulation model is used to simulate the normal operation state of the powered train to obtain a plurality of normal current signals as sample current signals with the current category being load current.

[0208] Based on the same inventive concept, the embodiments of the present application also provide a current recognition device for implementing the above-mentioned current recognition method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more current recognition device embodiments provided below can refer to the limitations of the current recognition method described above, which will not be repeated here.

[0209] In an exemplary embodiment, as shown in Figure 8 a current recognition device is provided, comprising: a current acquisition module 810 and a current recognition module 820, wherein:

[0210] The current acquisition module 810 is configured to acquire a current feature of a current signal of a traction network power supply system for supplying power to a train;

[0211] The current recognition module 820 is configured to input the current feature of the current signal into a current recognition model of the traction network power supply system to obtain a current category of the current signal; wherein the current recognition model is obtained based on the above-mentioned current recognition model training method.

[0212] Each module in the above-mentioned current recognition model training device and current recognition device can be realized by software, hardware, and combinations thereof, in whole or in part. Each module can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0213] In an exemplary embodiment, a computer device is provided, which can be a server, and its internal structure diagram can be as shown in Figure 9 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store sample current signals, current signals, and other data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a current recognition model training method and / or a current recognition method.

[0214] Those skilled in the art can understand, Figure 9 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0215] In an exemplary embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the current identification model training method and / or the current identification method when executing the computer program.

[0216] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implements the steps in the current identification model training method and / or the current identification method when executed by a processor.

[0217] In an embodiment, a computer program product is provided, comprising a computer program, and the computer program implements the steps in the current identification model training method and / or the current identification method when executed by a processor.

[0218] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0219] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0220] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A current identification model training method, characterized in that: The method comprises: Acquiring a sample current signal of a traction network power supply system and a current category of the sample current signal; performing variational modal decomposition on the sample current signal to obtain multiple modal components of the sample current signal; extracting a current feature of the sample current signal according to the multiple modal components; The current characteristics are used as input and the current categories are used as labels to train a preset classification model to obtain a current recognition model for the traction network power supply system.

2. The method according to claim 1, characterized in that The performing variational modal decomposition on the sample current signal to obtain multiple modal components of the sample current signal includes: Taking the minimum sum of spatially correlated recursive sample entropies of multiple modal components of the sample current signal as the optimization goal, the number of decomposition layers and the penalty factor of the variational modal decomposition algorithm are optimized to obtain an optimized variational modal decomposition algorithm; The optimized variational modal decomposition algorithm is used to perform noise reduction decomposition on the sample current signal to obtain multiple modal components of the sample current signal.

3. The method according to claim 1, characterized in that The extracting the current feature of the sample current signal according to the multiple modal components includes: screening effective components among the multiple modal components; The effective component is reconstructed to obtain a reconstructed signal, and spatial correlation recursive sample entropy extraction is performed on the reconstructed signal to obtain a current feature of the sample current signal.

4. The method according to claim 3, characterized in that The screening of valid components among the multiple modal components includes: Determine an index value of a screening index for each of the multiple modal components, and screen the modal components whose index values ​​meet a screening threshold among the multiple modal components as valid components; wherein the screening index includes at least one of a complexity index, an energy index, a correlation index, and a frequency index.

5. The method according to any one of claims 1 to 4, characterized in that The obtaining of a sample current signal of the traction network power supply system and a current category of the sample current signal includes: Using a simulation model of the traction network power supply system, a traction network short-circuit fault is simulated to obtain a plurality of abnormal current signals as sample current signals of a short-circuit current type; and The traction network simulation model is used to simulate the normal operating state of the powered train, and a plurality of normal current signals are obtained as sample current signals whose current category is load current.

6. A current identification method, characterized in that: The method comprises: obtaining a current characteristic of a current signal of a traction network power supply system for powering a train; The current characteristics of the current current signal are input into the current identification model of the traction network power supply system to obtain the current category of the current current signal; wherein, the current identification model is trained based on the current identification model training method described in any one of claims 1-5.

7. A current identification model training device, characterized in that: The device comprises: A sample acquisition module, configured to acquire a sample current signal of a traction network power supply system and a current category of the sample current signal; a sample decomposition module, configured to perform variational modal decomposition on the sample current signal to obtain multiple modal components of the sample current signal; a feature extraction module, configured to extract current features of the sample current signal based on the multiple modal components; The model training module is used to train a preset classification model using the current characteristics as input and the current category as a label to obtain a current identification model of the traction network power supply system.

8. A current identification device, characterized in that: The device comprises: a current acquisition module, for acquiring current characteristics of a current signal of a traction network power supply system for powering the train; A current identification module is used to input the current characteristics of the current current signal into the current identification model of the traction network power supply system to obtain the current category of the current current signal; wherein, the current identification model is trained based on the current identification model training method described in any one of claims 1-5.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images