CN-121997276-A - Multi-mode data fusion type rail transit vehicle health state assessment and prediction method

CN121997276ACN 121997276 ACN121997276 ACN 121997276ACN-121997276-A

Abstract

The application relates to the technical field of state prediction, in particular to a method for evaluating and predicting the health state of a rail transit vehicle by multi-modal data fusion, which comprises the steps of collecting multi-modal data and working condition control signals at any moment in the running process of the rail transit vehicle; the method comprises the steps of constructing modal characteristics, identifying current working conditions of a vehicle, calculating information entropy of each modal characteristic, comparing the information entropy with reference information entropy under the current working conditions to obtain information value of each modal data under the current working conditions, calculating variation coefficients of the information value of each modal data, adjusting temperature parameters of an attention mechanism according to the variation coefficients, carrying out weighted aggregation on each modal characteristic by utilizing the temperature parameters and the information value to obtain multi-modal fusion characteristics, inputting the multi-modal fusion characteristics at each moment into an evaluation prediction model, and outputting the health state of the vehicle. According to the technical scheme, the accurate prediction of the vehicle health state can be realized.

Inventors

WU GAO
WU YUPING
DU YONGJUN
YU JIANHUA
LIU SHAOYUAN

Assignees

广东华能机电集团有限公司

Dates

Publication Date: 20260508
Application Date: 20260407

Claims (10)

1. The method for evaluating and predicting the health state of the rail transit vehicle by multi-mode data fusion is characterized by comprising the following steps of: acquiring multi-mode data and working condition control signals at any moment in the running process of the rail transit vehicle; Constructing a plurality of modal characteristics of the multi-modal data, and identifying the current working condition of the vehicle according to the working condition control signal; Calculating information entropy of each mode feature in a current time window, and comparing the information entropy with a reference information entropy under a current working condition to obtain information value of each mode data under the current working condition, wherein the information value is positively correlated with the absolute value of the difference value of the information entropy and the reference information entropy; calculating a variation coefficient of information value of each mode of data, and adaptively adjusting a temperature parameter of an attention mechanism according to the variation coefficient, wherein the temperature parameter and the variation coefficient are in negative correlation; And the multi-mode fusion characteristics at a plurality of moments are input into a preset evaluation prediction model to output the vehicle health state.
2. The method for evaluating and predicting the health status of a rail transit vehicle by using multi-modal data fusion according to claim 1, wherein after collecting the multi-modal data and the working condition control signal at any time during the running of the rail transit vehicle, the method further comprises: filling the missing values of multi-mode data with different sampling frequencies by using a linear interpolation method, wherein the multi-mode data comprises at least two of vibration data, temperature data, electric data and acoustic data; And taking the fixed duration as a time window, and performing time alignment on the filled data of each mode.
3. The method for assessing and predicting the health of a rail transit vehicle fused with multimodal data according to claim 2, wherein constructing the plurality of modality features of the multimodal data comprises: extracting time domain features and frequency domain features of the vibration signals to obtain vibration features; extracting the temperature rise rate and the temperature deviation characteristics of the temperature data to obtain temperature characteristics; Extracting load fluctuation and harmonic characteristics of electrical parameters to obtain electrical characteristics; Extracting spectrum envelope characteristics of the acoustic signals to obtain acoustic characteristics; and mapping each feature into the modal features with the same dimension through a preset full connection layer.
4. The method for evaluating and predicting the health status of the rail transit vehicle with the multi-mode data fusion according to claim 1, wherein the current working condition is any one of stop, start acceleration, constant-speed cruising and deceleration braking.
5. The method for evaluating and predicting the health state of the rail transit vehicle by the multi-mode data fusion according to claim 1 is characterized in that calculating the information entropy of each mode feature in a current time window comprises the steps of carrying out histogram statistics on each component value of the mode feature to obtain discrete probability distribution of the feature value falling into each interval, and calculating the information entropy of each mode feature according to the discrete probability distribution.
6. The method for evaluating and predicting the health state of the rail transit vehicle with the multi-mode data fusion according to claim 1 is characterized in that before the information value of each mode data under the current working condition is obtained, the method further comprises the steps of counting information entropy average values of each mode feature in historical operation data under different working conditions to obtain reference information entropy corresponding to each working condition, counting transition frequencies among different working conditions in historical operation records, and constructing a working condition transition probability matrix.
7. The method for evaluating and predicting the health status of a rail transit vehicle by fusion of multi-modal data according to claim 6, wherein obtaining the information value of each modal data under the current working condition comprises: Determining a pre-judging working condition according to the working condition transition probability matrix in response to the current time window being in a working condition transition period, and carrying out weighted average on a reference information entropy corresponding to the pre-judging working condition and a reference information entropy under the current working condition to obtain a calibration reference; And determining the information value according to the deviation degree of the information entropy relative to the calibration standard.
8. The method for evaluating and predicting the health status of a rail transit vehicle by using multi-modal data fusion according to claim 1, wherein the weighting and aggregating the modal features by using the temperature parameter and the information value comprises: calculating dot products between any two modal characteristics, adding bias items corresponding to the information values to the ratio of the dot products to the temperature parameters, and obtaining a cross-modal attention matrix, wherein the cross-modal attention moment matrix comprises attention scores among the modal characteristics; Weighting and summing the modal characteristics by using the cross-modal attention matrix to obtain enhanced characteristics; And splicing and dimension reduction processing are carried out on each enhancement feature, so as to obtain the multi-mode fusion feature.
9. The method for evaluating and predicting the health status of a rail transit vehicle with multi-modal data fusion according to claim 1, wherein the evaluation prediction model is a long-short-term memory network, the health status of the vehicle is a value from 0 to 1, 0 represents health, and 1 represents a fault.
10. The method for evaluating and predicting the health status of the rail transit vehicle by using the multi-modal data fusion according to claim 9 is characterized in that the step of evaluating and predicting the model for training comprises the steps of collecting a sequence consisting of multi-modal fusion characteristics at a plurality of historical moments as a training sample, taking the real status of the training sample at the next moment as a label, and training the evaluating and predicting model by using a cross entropy loss function until the cross entropy loss function is smaller than a preset value or the number of iterations reaches the maximum number.

Description

Multi-mode data fusion type rail transit vehicle health state assessment and prediction method Technical Field The application relates to the technical field of state prediction, in particular to a method for evaluating and predicting the health state of a rail transit vehicle by multi-mode data fusion. Background With the rapid development of urban rail transit, the long-term running safety of rail transit vehicles is particularly important. In daily operation, key components such as a running part, a traction system and the like of a vehicle are in a complex and changeable running environment, and the health state of the key components is directly related to the overall running safety. In order to realize intelligent operation and maintenance of rail transit vehicles, multi-mode sensing data such as vibration, temperature, electricity, acoustics and the like in the running process of the vehicles are generally collected to monitor states. Therefore, how to fully utilize the heterogeneous data containing the equipment state information to realize accurate prediction of the health state of the rail transit vehicle becomes a problem to be solved urgently. Vehicle health assessment based on single modality data or multimodal fusion is currently commonly employed in the industry. The time domain or frequency domain features of each monitoring data are generally extracted respectively, then the features are simply aggregated by utilizing a fixed-weight splicing mode or a static conventional attention mechanism, and finally the fused features are input into a long-short-time memory network and other machine learning models for training, so that the current health state or fault diagnosis result of the vehicle is output. However, the rail transit vehicle frequently undergoes working condition switching such as platform stopping, starting acceleration and the like in daily operation, and under different operation working conditions, the characterization capability of each mode data on the vehicle health state is huge, the traditional static fusion method ignores the influence of the working condition change on the information quantity of the multi-mode data, and the weight of each mode data in fusion can not be dynamically quantized and adjusted according to the specific working condition, so that the prediction result of the vehicle health state is inaccurate. Disclosure of Invention In order to solve the technical problem that the prediction result of the vehicle health state is inaccurate, the application provides the rail transit vehicle health state assessment and prediction method based on multi-mode data fusion, which can fully utilize the complementary information of each mode, thereby realizing accurate and stable vehicle health state prediction. The application provides a rail transit vehicle health state assessment and prediction method based on multi-modal data fusion, which comprises the steps of collecting multi-modal data and working condition control signals at any moment in the running process of a rail transit vehicle, constructing a plurality of modal characteristics of the multi-modal data, identifying the current working condition of the vehicle according to the working condition control signals, calculating information entropy of each modal characteristic in a current time window, comparing the information entropy with reference information entropy under the current working condition to obtain information value of each modal data under the current working condition, wherein the information value is positively correlated with the absolute value of the difference between the information entropy and the reference information entropy, calculating a variation coefficient of the information value of each modal data, adaptively adjusting temperature parameters of an attention mechanism according to the variation coefficient, wherein the temperature parameters are negatively correlated with the variation coefficient, weighting and aggregating each modal characteristic by utilizing the temperature parameters and the information value, obtaining multi-modal fusion characteristics, and inputting the multi-modal fusion characteristics at a plurality of moments into a preset assessment prediction model, and outputting the vehicle health state. The information value is defined by the deviation degree of the information entropy relative to the working condition reference, the attention temperature parameter is regulated by the variation coefficient of the information value, and the attention can be automatically concentrated in the mode with the most discriminant under different working conditions, so that the interference of the working condition change on the health condition assessment result is obviously reduced, and the accuracy of vehicle health condition assessment is improved. Preferably, after the multi-mode data and the working condition control signals at any time in the running process of the ra