CN120686146A

CN120686146A - A method and device for diagnosing grounding faults in electric locomotive converters based on MLP

Info

Publication number: CN120686146A
Application number: CN202510771207.7A
Authority: CN
Inventors: 徐文婧; 刘佳伟; 侯欣延; 汪盼盼; 王雪迪
Original assignee: CRRC Dalian R&D Co Ltd
Current assignee: CRRC Dalian R&D Co Ltd
Priority date: 2025-06-10
Filing date: 2025-06-10
Publication date: 2025-09-23

Abstract

The invention discloses an electric locomotive converter ground fault diagnosis method and device based on MLP. Relates to the technical field of operation and maintenance of electric locomotives. The method mainly comprises the steps of obtaining historical operation data of the electric locomotive converter, including operation state data and operation condition data corresponding to the operation state data, constructing an electric locomotive converter ground fault diagnosis model based on a multi-layer perceptron model, taking the operation state characteristic data as input of the model, taking the corresponding operation condition data as output of the model to train the electric locomotive converter ground fault diagnosis model, inputting the operation state characteristic data to be diagnosed into the trained electric locomotive converter ground fault diagnosis model, and obtaining a diagnosis result output by the electric locomotive converter ground fault diagnosis model. According to the invention, the MLP model is trained by collecting historical data, a small number of features are screened and applied to online fault diagnosis, so that the ground fault occurrence point is positioned more accurately.

Description

MLP-based electric locomotive converter ground fault diagnosis method and device

Technical Field

The invention relates to the technical field of operation and maintenance of electric locomotives, in particular to a method and a device for diagnosing a grounding fault of an electric locomotive converter based on MLP.

Background

The electric locomotive converter is one of the core components for ensuring efficient and stable operation of the locomotive. However, ground faults, one of the common types of electrical faults, severely affect the reliability and safety of the current transformer. The traditional ground fault diagnosis method mainly depends on empirical rules and simple threshold judgment, and can be positioned to a large range of four-quadrant side, middle direct current loop positive end, middle direct current loop negative end, inversion side and the like. However, with the development of power electronics technology, the requirements on fault diagnosis precision are increasingly increased, and the conventional diagnosis method is difficult to meet the actual requirements. Particularly, under the complex working condition, the fault point is accurately positioned, and the method has important significance for reducing maintenance time and reducing cost of manpower and material resources.

The fault diagnosis method for the electric locomotive converter, which is commonly used at present, mainly comprises the following steps:

Methods based on current or voltage rate of change. Such methods rely primarily on monitoring the rate of change of current and voltage at critical nodes in the current transformer. By setting certain thresholds, the system will trigger an alarm when the detected change exceeds those thresholds, indicating that a ground fault has occurred. However, this method can only determine that the fault occurs in a large area such as the four-quadrant side, the positive side of the intermediate dc loop, the negative side of the intermediate dc loop, the inverter side, etc., and cannot accurately locate a specific circuit element. And it can only be determined that the fault occurs approximately in a large area of the four-quadrant side, the positive end of the intermediate dc loop, the negative end of the intermediate dc loop, the inversion side, etc., and the specific circuit element cannot be precisely located. Meanwhile, the method has poor adaptability, needs to readjust the threshold value for different types of electric locomotives or different running conditions, and is not timely enough for rapid change fault response due to the fact that the method depends on fixed threshold value judgment.

Spectral analysis. The spectrum analysis method is used for identifying abnormal frequency components by carrying out frequency domain analysis on signals in the converter circuit so as to judge whether the ground fault occurs. But this approach requires high performance data acquisition and processing equipment to perform the spectral analysis, increasing the cost and complexity of the system. In addition, for aperiodic signal variation, the spectrum analysis has poor effect, and can lead to missed detection or misjudgment. In addition, the method has high demand for computing resources for processing a large amount of data in real time, and the popularity of the method in practical application is affected.

Model Predictive Control (MPC) based fault detection and isolation. MPC is an advanced control strategy that predicts potential failure modes by building a mathematical model of the converter system and updating model parameters based on real-time data. However, MPC involves complex mathematical modeling and algorithm design, and has great difficulty in implementation, especially in dynamic environment, difficult adjustment of model parameters, and long debugging period. Meanwhile, MPC involves complex mathematical modeling and algorithm design, so that the implementation difficulty is high, and especially the adjustment of model parameters in a dynamic environment is very difficult. In addition, because the working environment of the electric locomotive is changeable, the model is difficult to reflect the actual situation completely and accurately, and the prediction result may be inaccurate.

Disclosure of Invention

In view of the defects of the prior art, the invention provides an electric locomotive converter ground fault diagnosis method and device based on MLP. According to the invention, the MLP model is trained by collecting historical data, a small number of features are screened and applied to online fault diagnosis, so that the ground fault occurrence point is positioned more accurately.

The invention adopts the following technical means:

an electric locomotive converter ground fault diagnosis method based on MLP comprises the following steps:

Acquiring historical operation data of an electric locomotive converter, wherein the historical operation data of the electric locomotive converter comprises operation state data and operation condition data corresponding to the operation state data;

Respectively carrying out time domain and frequency domain analysis on the running state data, and calculating running state characteristic data based on analysis results;

constructing an electric locomotive converter grounding fault diagnosis model based on the multi-layer perceptron model, taking the running state characteristic data as the input of the model, and taking the corresponding running condition data as the output of the model to train the electric locomotive converter grounding fault diagnosis model;

acquiring running state data to be diagnosed in real time, and calculating running state characteristic data to be diagnosed according to the running state data to be diagnosed;

And inputting the running state characteristic data to be diagnosed into a trained ground fault diagnosis model of the electric locomotive converter, and obtaining a diagnosis result output by the ground fault diagnosis model of the electric locomotive converter.

Further, performing time domain and frequency domain analysis on the operation state data respectively, and screening operation state characteristic data based on analysis results, including:

Calculating a characteristic value for each state quantity in the operating state data,

And sequencing the characteristic values by adopting a replacement importance method, and determining the running state characteristic data according to the sequencing result.

Further, the running state characteristic data comprise frequency variance of the grounding voltage, frequency root mean square value of the grounding voltage, phase difference of the grounding voltage and four-quadrant homonymous terminal current and phase difference of the grounding voltage and U-phase current.

Further, the operation state characteristic data is used as the input of a model, the corresponding operation condition data is used as the output of the model to train the ground fault diagnosis model of the electric locomotive converter, and the operation condition data is converted into a vector form through independent heat coding.

Further, the operation state data comprise a converter grounding voltage, a four-quadrant homonymous terminal current and a U-phase current, and the operation condition data comprise a normal operation state, a four-quadrant homonymous terminal grounding state, a four-quadrant abnormal terminal grounding state, a middle loop positive terminal grounding state, a middle loop negative terminal grounding state, an inverter U grounding state, an inverter V grounding state and an inverter W grounding state.

The invention also discloses an electric locomotive converter grounding fault diagnosis device based on the MLP, which is used for executing the fault diagnosis method and comprises the following steps:

The historical data acquisition unit is used for acquiring historical operation data of the electric locomotive converter, wherein the historical operation data of the electric locomotive converter comprises operation state data and operation condition data corresponding to the operation state data;

A historical data analysis unit for performing time domain and frequency domain analysis on the operation state data respectively, and calculating operation state characteristic data based on an analysis result;

The model building unit is used for building an electric locomotive converter ground fault diagnosis model based on the multi-layer perceptron model, taking the running state characteristic data as the input of the model and taking the corresponding running working condition data as the output of the model to train the electric locomotive converter ground fault diagnosis model;

the real-time data acquisition unit is used for acquiring the running state data to be diagnosed in real time and calculating the running state characteristic data to be diagnosed according to the running state data to be diagnosed;

The fault diagnosis unit is used for inputting the running state characteristic data to be diagnosed into a trained ground fault diagnosis model of the electric locomotive converter and obtaining a diagnosis result output by the ground fault diagnosis model of the electric locomotive converter.

The invention also discloses an electronic device, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor runs and executes the electric locomotive converter ground fault diagnosis method based on the MLP through the computer program.

Compared with the prior art, the invention has the following advantages:

By introducing the MLP model, the invention not only improves the accuracy of the ground fault diagnosis, but also reduces the false alarm rate. In addition, due to the adoption of a machine learning algorithm, the system has self-learning capability, and can continuously improve the diagnosis precision along with the time. The safety and reliability of the operation of the electric locomotive are greatly improved, and meanwhile, the maintenance cost is effectively saved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.

Fig.1 is a flowchart of a method for diagnosing a ground fault of an electric locomotive converter based on MLP according to the present invention.

Fig. 2 is a schematic diagram of a traction converter main circuit in an embodiment.

Fig. 3 is a flowchart of an MLP-based fault diagnosis scheme in an embodiment.

FIG. 4 is a specific execution flow of fault diagnosis by using the method of the present invention in an embodiment.

FIG. 5 is a ranking chart of importance of alternative features in an embodiment.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

As shown in fig. 1-4, the embodiment of the application discloses a ground fault diagnosis method for an electric locomotive converter based on MLP, which comprises the following steps:

S1, acquiring historical operation data of an electric locomotive converter, wherein the historical operation data of the electric locomotive converter comprises operation state data and operation condition data corresponding to the operation state data. In the embodiment of the application, the running state data comprise the grounding voltage of the converter, the current of the four-quadrant homonymous terminal and the current of the U phase. The operating condition data comprise a normal operating state, a four-quadrant homonymous grounding state, a four-quadrant abnormal grounding state, a middle loop positive grounding state, a middle loop negative grounding state, an inverter U grounding state, an inverter V grounding state and an inverter W grounding state.

Specifically, as shown in fig. 2, a schematic diagram of a main circuit of the traction converter of the electric locomotive is shown. The invention positions the position of the ground fault to the four-quadrant homonymous ground, the four-quadrant heteronymous ground, the positive end of the middle loop to the ground, the negative end of the middle loop to the ground, the inverter U to the ground, the inverter V to the ground and the inverter W to the ground. And collecting the grounding voltage U _G, the four-quadrant homonymous terminal current i _A and the U-phase current i _U as data sources when the traction converter main circuit is in a normal working state and each point is in a grounding fault (hereinafter referred to as each fault working condition). Thereby constructing historical operation data of the electric locomotive converter.

Further, each failure condition is defined as a failure signature due to the need to train and test the neural network model. In order to eliminate misleading of the increasing relation of the fault labels and ensure mutual exclusion equality among the categories, the category labels are converted into vector forms through single-hot coding, as shown in the table 1, so that loss function calculation can be simplified, and the multi-category classification problem can be effectively processed. After removing the non-numeric data, the historical operating dataset is divided into a training set, a testing set, and a validation set.

TABLE 1 failure tag definition and one-time thermal encoding

S2, performing time domain and frequency domain analysis on the operation state data respectively, and calculating operation state characteristic data based on analysis results.

Specifically, after the category label of the historical operation data is obtained, the operation state data is analyzed and processed. The application preferably divides the data in 1 second after the fault occurrence into 50 time windows according to every 0.02 second, and calculates the characteristic values under each fault working condition in each time window, including the maximum value, the mean value, the root mean square value, the skewness value, the pulse index and the kurtosis value in the time domain, the center of gravity frequency, the frequency root mean square, the average frequency, the frequency variance, the phase difference between U _G and i _A and the phase difference between U _G and i _U in the frequency domain.

The time domain features may provide information about the behavior and characteristics of the signal over time, such as the overall level of the signal and the sharpness of the distribution, etc., and the frequency domain features may provide information about the frequency content of the signal, such as the degree of dispersion of the main frequency content and distribution of the signal, etc. In order to more fully analyze the characteristics of the signal, feature engineering analyzes both the time domain features and the frequency domain features, thereby constructing alternative features.

And (3) carrying out normalization processing on the characteristic values because of obvious numerical differences among the characteristics, and taking negative numbers in the characteristic values into consideration, and adopting a maximum normalization method to attribute the characteristic values to [ -1,1]. In order to further complete screening of the constructed alternative features, a method of 'replacement importance (Permutation Importance)' is adopted to complete feature importance sorting, wherein the replacement importance method is that in a built multi-class classification ECOC model, for each alternative feature, the numerical sequence of the feature is randomly disturbed in sequence, and if the model accuracy rate is remarkably reduced, the feature is very important for model prediction. The importance ranking of the obtained alternative features is shown in fig. 5, so that the contribution degree of each feature to model prediction can be effectively identified, the feature quantity of the top four ranks is selected as the input of the MLP, and the frequency variance of U _G, the phase difference of U _G and i _A, the root mean square value of U _G and the phase difference of U _G and i _U are selected as the input quantities.

And S3, constructing an electric locomotive converter grounding fault diagnosis model based on the multi-layer perceptron model, taking the operation state characteristic data as the input of the model, and taking the corresponding operation condition data as the output of the model to train the electric locomotive converter grounding fault diagnosis model.

The type of the neural network is preferably a multi-layer perceptron (Multilayer Perceptron, MLP), which is a feedforward artificial neural network model, the complex nonlinear relation in input data can be captured through multi-layer nonlinear transformation, progressive relation among fault labels considered by the neural network model is avoided, and the classification labels are converted into vector forms through independent thermal coding, so that the multi-class classification problem can be effectively treated.

Further, during model training, regularization techniques are employed to optimize the superparameter and model performance is monitored in real time by early-stop (Early Stopping) to prevent overfitting. Specifically, regularization is used as a key strategy to limit model complexity and avoid overfitting to training data. By imposing constraints on the model parameters, regularization not only helps to reduce the risk of overfitting, but also improves the generalization ability of the model. In the application, an implicit random regularization constraint is applied to the weight parameters of the full-connection layer by introducing a Dropout mechanism in the neural network training process. This mechanism randomly "turns off" part of the neurons with a certain probability p (typically set to 0.5) in each training iteration, forcing the network to learn a more robust and generalized feature representation. The mathematical expression is as follows:

let the output of the layer i neurons be:

a^(l)＝f(W^(l)a^(l-1)+b^(l))

after Dropout is introduced, the output becomes:

wherein ζ _i -Bernoulli (p)

Where ζ is a Bernoulli random vector of the same dimension as a (l) for simulating random inactivation of neurons. This mechanism is equivalent to applying a structured regularization to the weight matrix W, avoiding its excessive adaptation to noise or local patterns in the training data, thus improving the generalization performance of the model.

In the invention, the neural network uses a Cross entropy loss function (Cross-Entropy Loss) which is suitable for multi-classification tasks. The multi-class cross entropy loss function is defined as follows:

let the true label be the one-hot coding vector y epsilon {0,1} ^C, the probability distribution of model output be The cross entropy loss is:

wherein:

And C, total category number.

Y _c whether the real tag belongs to class c.

Model predictive class c probability.

This is the default penalty function used by MATLAB, which applies to classificationLayer, and therefore does not require additional definition.

In the invention, although L1 or L2 regular terms are not explicitly added in the loss function, the indirect control of model complexity is realized by the following modes:

(1) And an Adam optimizer is used for parameter updating, and the self-adaptive learning rate characteristic is utilized to prevent overlarge updating amplitude of certain parameters.

(2) The maximum training round number (epochs) was set to 200, preventing over-fitting of the model to the training data.

(3) The verification set monitoring mechanism is introduced, training is stopped when the performance of the verification set is not improved any more, early stopping is achieved, in MATLAB, the mechanism is achieved by setting a net.trainParam.max_fail parameter, the parameter is set to be 10, and training is automatically stopped if errors on the verification set are not reduced by 10 epochs continuously.

(4) Dropout layer (discarding rate is 0.5) is introduced into the neural network, so that the hidden layer neurons are randomly shielded, and the generalization capability of the model is enhanced.

(5) The original data is divided into a training set and a testing set according to the proportion of 7:3, so that the model is ensured to have enough training samples and reliable verification data is reserved.

The method ensures that the model can obtain good performance on a training set, can keep stable prediction performance on unseen data, and enhances the generalization capability of the model.

Furthermore, the application compiles the trained neural network model into c language by utilizing the MBD model and using s-function, and writes the header file and the source file into the chip to complete the deployment of the model. Through the feasible model deployment scheme, the MLP can realize the real-time diagnosis function in the converter control system

S4, acquiring running state data to be diagnosed in real time, and calculating running state characteristic data to be diagnosed according to the running state data to be diagnosed.

S5, inputting the running state characteristic data to be diagnosed into a trained ground fault diagnosis model of the electric locomotive converter, and obtaining a diagnosis result output by the ground fault diagnosis model of the electric locomotive converter.

The application utilizes historical data to construct and screens the characteristics through an importance sorting method to construct the MLP model suitable for the diagnosis of the grounding faults of the converter. The fault accurate positioning from the macro area to the specific port level is realized, namely the ground fault occurrence point diagnosis is extended to the four-quadrant side, the middle direct current loop positive end, the middle direct current loop negative end and the inversion side, and the ground fault occurrence point diagnosis can be realized to the four-quadrant homonymous end, the four-quadrant heteronymous end, the middle direct current loop positive end, the middle direct current loop negative end, the inversion side U phase, the inversion side V phase and the inversion side W phase.

The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.

In the foregoing embodiments of the present invention, the descriptions of the embodiments are emphasized, and for a portion of this disclosure that is not described in detail in this embodiment, reference is made to the related descriptions of other embodiments.

In the several embodiments provided in the present application, it should be understood that the disclosed technology may be implemented in other manners. The above-described embodiments of the apparatus are merely exemplary, and the division of the units, for example, may be a logic function division, and may be implemented in another manner, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interfaces, units or modules, or may be in electrical or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes a U disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, etc. which can store the program code.

It should be noted that the above embodiments are merely for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solution described in the above embodiments may be modified or some or all of the technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the scope of the technical solution of the embodiments of the present invention.

Claims

1. The MLP-based grounding fault diagnosis method for the electric locomotive converter is characterized by comprising the following steps of:

2. The MLP-based ground fault diagnosis method of an electric locomotive converter according to claim 1, wherein time domain and frequency domain analysis are performed on the operation state data, respectively, and operation state feature data are screened based on the analysis result, comprising:

3. The method for diagnosing the grounding fault of the electric locomotive converter based on the MLP of claim 2, wherein the operation state characteristic data comprise a frequency variance of a grounding voltage, a frequency root mean square value of the grounding voltage, a phase difference between the grounding voltage and a four-quadrant homonymous terminal current and a phase difference between the grounding voltage and a U-phase current.

4. The MLP-based electrical locomotive converter ground fault diagnosis method of claim 1, wherein training an electrical locomotive converter ground fault diagnosis model with the operating condition characteristic data as an input of the model and the corresponding operating condition data as an output of the model comprises converting the operating condition data into a vector form through independent heat encoding.

5. The method for diagnosing a grounding fault of an electric locomotive converter based on MLP according to claim 1, wherein the operation state data includes a converter grounding voltage, a four-quadrant homonymous terminal current and a U-phase current, and the operation condition data includes a normal operation state and a four-quadrant homonymous terminal grounding state, a four-quadrant abnormal terminal grounding state, an intermediate circuit positive terminal grounding state, an intermediate circuit negative terminal grounding state, an inverter U grounding state, an inverter V grounding state and an inverter W grounding state.

6. An MLP-based ground fault diagnosis apparatus for an electric locomotive converter for performing the fault diagnosis method according to claim 1, comprising:

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor is operative to perform a MLP-based electrical locomotive converter ground fault diagnosis method as claimed in any one of claims 1 to 5 by means of the computer program.