An abnormality recognition method based on multi-sensor fusion for high-speed railway traction and braking system

Abstract

The present invention proposes an abnormality recognition method for multi-sensor fusion of a high-speed railway traction and braking system, comprising: acquiring multi-parameter perception data of the high-speed railway traction and braking system, dividing the multi-parameter perception data into operating parameters and measurement parameters and performing preprocessing; inputting the preprocessed operating parameters and measurement parameters into a de-confusion model to obtain measurement parameter adjustment values; performing reverse prediction on the operating parameters using the measurement parameter adjustment values ​​according to an anti-causal multivariate regression model to obtain operating parameter prediction values; establishing degradation values ​​of each operating parameter through the operating parameters and the operating parameter prediction values, using the complexity of the historical time series of each operating parameter as a weight, and establishing a normalized health indicator model, thereby evaluating the health status of the high-speed railway traction and braking system.

Description

Technical Field

[0001] The invention belongs to the field of fault diagnosis. Background Art

[0002] The high-speed rail traction and braking system is one of the key systems in high-speed rail trains. Obtaining its health status is crucial to ensuring the safety of passengers and trains, evaluating system performance, and establishing system early warning. However, for such complex electromechanical systems, it is not feasible to measure their precise overall health status or even degradation trends. To address this problem, there are two main methods for constructing traction and braking system health indicators, namely physical health indicators and virtual health indicators. Physical health indicators are usually obtained by extracting high-frequency sampling signals from single mechanical component sensors (such as traction and braking force), or statistical features extracted from signals, and expressing the system health status in the form of numerical values ​​or curves. Such health indicators often cannot accurately evaluate the overall health status of the system. Virtual health indicators are usually one-dimensional dimensionless indicators extracted from multidimensional sensor readings or overall physical characteristics through supervised or unsupervised learning methods using fusion and dimensionality reduction techniques. For supervised learning, a large amount of degradation data is required for training, and features are extracted through deep convolutional neural networks or other nonlinear models and mapped to target values ​​representing the health status. These degradation data are often obtained through simulation experiments or field collection. However, due to the complexity of some electromechanical systems and the diversity of working environments, it is difficult to obtain high-quality comprehensive degradation state data of mechanical modules or systems of actual electromechanical systems. Relatively speaking, unsupervised learning methods do not require any degradation data and only use health state data to learn prediction algorithms. The problem is that the model trained directly with health data has learned the potential pattern of specific operating parameters, and deviations from such operating parameters will be mistakenly captured as system degradation. This leads to the problem that the early warning system established based on unsupervised learning virtual health indicators often has a high false alarm rate. On the other hand, both supervised and unsupervised methods are powerful tools for data-driven mining of the correlation between measured parameters and the health state of traction and braking systems. However, due to the influence of confounding factors, correlation does not mean causality. Controlled intervention is the only way to verify causality, but such intervention is often not feasible in practice. On this basis, misjudgment of causality will also seriously affect the robustness and generalization ability of health indicators. The unsupervised method based on causal inference proposed in the patent can overcome this problem and obtain high-speed rail traction and braking system health indicators with high robustness and generalization ability. Summary of the invention

[0003] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0004] To this end, the purpose of the present invention is to propose an abnormality identification method of multi-sensor fusion of high-speed railway traction and braking system, which is used to establish the overall health index of the electromechanical system.

[0005] To achieve the above-mentioned purpose, the first embodiment of the present invention proposes a method for identifying abnormalities of a high-speed railway traction and braking system by multi-sensor fusion, comprising:

[0006] Acquire multi-parameter sensing data of the high-speed rail traction braking system, divide the multi-parameter sensing data into operating parameters and measurement parameters, and perform preprocessing;

[0007] Inputting the preprocessed operation parameters and measurement parameters into the de-obfuscation model to obtain the measurement parameter adjustment value;

[0008] Perform reverse prediction of the operating parameter using the measured parameter adjustment value according to the reverse causal multivariate regression model to obtain a predicted operating parameter value;

[0009] The degradation value of each operating parameter is established through the operating parameter and the predicted value of the operating parameter, and the complexity of the historical time series of each operating parameter is used as a weight to establish a normalized health indicator model, so as to evaluate the health status of the high-speed railway traction and braking system.

[0010] In addition, the method for identifying abnormalities of a high-speed railway traction and braking system by multi-sensor fusion according to the above embodiment of the present invention may also have the following additional technical features:

[0011] Furthermore, in one embodiment of the present invention, the acquiring of multi-parameter sensing data of the high-speed railway traction braking system, dividing the multi-parameter sensing data into operating parameters and measurement parameters and performing pre-processing, includes:

[0012] Collect operating parameters, including traction / traction level, braking force / braking force level, acceleration / deceleration setting value, vehicle speed setting value, traction / braking mode;

[0013] Collect measurement parameters, including train speed, traction, braking force, acceleration / deceleration, brake temperature, wheel-track force, and braking distance;

[0014] The operation data and the measurement parameters are subjected to data cleaning, missing value processing, and outlier detection and removal.

[0015] Further, in one embodiment of the present invention, the step of inputting the preprocessed operating parameters and measurement parameters into a de-obfuscation model to obtain a measurement parameter adjustment value includes:

[0016] Inferring latent variables through historical multi-source time series;

[0017] Establish multiple regression models of operating parameters and measurement parameters;

[0018] The multivariate regression model is adjusted according to the latent variables to obtain adjusted values ​​of measurement parameters.

[0019] Further, in one embodiment of the present invention, the reverse prediction of the operating parameter using the measured parameter adjustment value according to the reverse causal multivariate regression model to obtain the operating parameter prediction value includes:

[0020] Establish a structural causal model between the adjusted values ​​of measurement parameters and the operational parameters;

[0021] Through the structural causal model, an anti-causal multiple regression model is established in which the measured parameter adjustment value is an independent variable and the operating parameter is a dependent variable.

[0022] To achieve the above-mentioned purpose, the second embodiment of the present invention proposes a high-speed railway traction and braking system multi-sensor fusion abnormality recognition device, comprising the following modules:

[0023] An acquisition module, used to acquire multi-parameter sensing data of the high-speed rail traction braking system, divide the multi-parameter sensing data into operating parameters and measurement parameters, and perform preprocessing;

[0024] An adjustment module, used for inputting the preprocessed operation parameters and measurement parameters into the de-obfuscation model to obtain the measurement parameter adjustment value;

[0025] A prediction module, configured to reversely predict the operating parameters using the measured parameter adjustment values ​​according to an inverse causal multivariate regression model to obtain predicted operating parameter values;

[0026] The identification module is used to establish the degradation value of each operating parameter through the operating parameter and the predicted value of the operating parameter, and use the complexity of the historical time series of each operating parameter as a weight to establish a normalized health indicator model, so as to evaluate the health status of the high-speed rail traction and braking system.

[0027] Furthermore, in one embodiment of the present invention, the acquisition module is further used to:

[0028] Collect operating parameters, including traction / traction level, braking force / braking force level, acceleration / deceleration setting value, vehicle speed setting value, traction / braking mode;

[0029] Collect measurement parameters, including train speed, traction, braking force, acceleration / deceleration, brake temperature, wheel-track force, and braking distance;

[0030] The operation data and the measurement parameters are subjected to data cleaning, missing value processing, and outlier detection and removal.

[0031] Furthermore, in one embodiment of the present invention, the adjustment module is further used to:

[0032] Inferring latent variables through historical multi-source time series;

[0033] Establish multiple regression models of operating parameters and measurement parameters;

[0034] The multivariate regression model is adjusted according to the latent variables to obtain adjusted values ​​of measurement parameters.

[0035] Furthermore, in one embodiment of the present invention, the prediction module is further used for:

[0036] Establish a structural causal model between the adjusted values ​​of measurement parameters and the operational parameters;

[0037] Through the structural causal model, an anti-causal multiple regression model is established in which the measured parameter adjustment value is an independent variable and the operating parameter is a dependent variable.

[0038] To achieve the above-mentioned purpose, the third aspect of the present invention proposes a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, it implements the above-mentioned method for abnormality identification of multi-sensor fusion of high-speed railway traction and braking system.

[0039] To achieve the above-mentioned purpose, the fourth aspect of the present invention proposes a computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, it implements the above-mentioned method for identifying abnormalities in multi-sensor fusion of a high-speed railway traction and braking system.

[0040] The abnormality identification method of multi-sensor fusion of high-speed railway traction and braking system proposed in the embodiment of the present invention breaks through the problems of poor robustness and insufficient generalization ability of virtual health indicators established by traditional unsupervised learning. The core idea is to extract the causal mechanism of the stability of the traction and braking system from the time series data and eliminate the influence of hidden confusion factors, and then predict the operating parameters that characterize the overall health status from the measured parameters through anti-causal learning, and then establish the overall health indicators of the electromechanical system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0042] Figure 1 A schematic flow chart of an abnormality identification method for multi-sensor fusion of a high-speed railway traction and braking system provided in an embodiment of the present invention.

[0043] Figure 2A schematic diagram of a method for constructing a latent variable to replace the influence of a confounding factor provided by an embodiment of the present invention.

[0044] Figure 3 A schematic diagram of a method for constructing a causal structure model and an anti-causal structure provided by an embodiment of the present invention.

[0045] Figure 4 A schematic diagram of a process for constructing health indicators of a high-speed railway traction and braking system provided in an embodiment of the present invention.

[0046] Figure 5 A schematic diagram of an abnormality identification device for multi-sensor fusion of a high-speed railway traction and braking system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0048] The following describes the abnormality identification method of multi-sensor fusion of high-speed railway traction and braking system according to an embodiment of the present invention with reference to the accompanying drawings.

[0049] Figure 1 A schematic flow chart of an abnormality identification method for multi-sensor fusion of a high-speed railway traction and braking system provided in an embodiment of the present invention.

[0050] like Figure 1 As shown, the high-speed railway traction brake system multi-sensor fusion abnormality recognition method includes the following steps:

[0051] S101: Acquire multi-parameter sensing data of a high-speed rail traction braking system, divide the multi-parameter sensing data into operating parameters and measurement parameters, and perform preprocessing;

[0052] Furthermore, in one embodiment of the present invention, multi-parameter sensing data of a high-speed railway traction braking system is obtained, and the multi-parameter sensing data is divided into operating parameters and measurement parameters and pre-processed, including:

[0053] Collect operating parameters, including traction / traction level, braking force / braking force level, acceleration / deceleration setting value, vehicle speed setting value, traction / braking mode;

[0054] Collect measurement parameters, including train speed, traction, braking force, acceleration / deceleration, brake temperature, wheel-track force, and braking distance;

[0055] Data cleaning, missing value processing, outlier detection and removal are performed on operational data and measurement parameters.

[0056] S102: inputting the preprocessed operating parameters and measurement parameters into a de-obfuscation model to obtain measurement parameter adjustment values;

[0057] Further, in one embodiment of the present invention, the pre-processed operating parameters and measurement parameters are input into the de-obfuscation model to obtain the measurement parameter adjustment value, including:

[0058] Inferring latent variables through historical multi-source time series;

[0059] Establish multiple regression models of operating parameters and measurement parameters;

[0060] The multivariate regression model was adjusted according to the latent variables to obtain the adjusted values ​​of the measurement parameters.

[0061] like Figure 2 As shown, the construction method of latent variables is shown:

[0062]

[0063] The above formula represents the construction principle of latent variables, where z t is the latent variable at time t, Represents all operating parameters in a period of time before time t and latent variables The latent variables include the influence of potential confounding variables, making the operational variables conditionally independent under the given latent variable conditions.

[0064]

[0065] Use recurrent neural network (RNN) as the basic architecture of the model and utilize historical data Fit the model and estimate the values ​​of the latent variables.

[0066]

[0067] Where L is the random initial parameter of RNN. The model is trained using gradient descent, and the model output is the latent variable Z t .

[0068] Considering the confounding factor term, a multiple linear regression model of each measurement parameter is established, which can be expressed as:

[0069]

[0070] Among them, β0, β j , γ jare the intercept and regression coefficient of the regression model respectively, and ε is the error term. First, the estimated regression coefficient of the regression model is obtained by the least squares method. According to the given operating parameters x and confusion factor z i The value of is substituted into the regression model to calculate the adjusted measurement parameter y′. According to the formula of the regression model, the following adjustment formula can be obtained, which is recorded as the de-obfuscation model:

[0071]

[0072] The above formula represents a method for calculating the measurement parameter value after removing the influence of confusion. The above method is used to calculate the adjustment value of each measurement parameter to obtain Y′, which is regarded as the theoretical value of the measurement parameter after removing the influence of the confusion factor.

[0073] S103: reversely predicting the operating parameters using the measured parameter adjustment values ​​according to the reverse causal multivariate regression model to obtain predicted values ​​of the operating parameters;

[0074] Further, in one embodiment of the present invention, reverse prediction of the operating parameters using the measured parameter adjustment values ​​according to the anti-causal multivariate regression model to obtain the predicted values ​​of the operating parameters includes:

[0075] Establish a structural causal model between the adjusted values ​​of measurement parameters and the operational parameters;

[0076] Through the structural causal model, an anti-causal multiple regression model is established with the measured parameter adjustment value as the independent variable and the operating parameter as the dependent variable.

[0077] like Figure 3 As shown in the figure, a structural causal model is constructed, and a certain threshold is set. If the regression coefficient is less than the threshold, it is regarded as the operating parameter x i There is no causal relationship between the measured parameter y. The structural causal model is a systematic stable causal structure extracted from unstable data.

[0078] According to the structural causal model, the arrows are reversed and regarded as a reverse structural causal model. The multiple linear regression model of each operating parameter is established, which can be expressed as:

[0079] x i =f i (Y′)+δ i ,

[0080] The aforementioned method is used to train a multivariate linear regression model with historical data.

[0081] S104: Establish degradation values ​​of various operating parameters through operating parameters and predicted values ​​of operating parameters, adopt the complexity of historical time series of various operating parameters as weights, and establish a normalized health indicator model to evaluate the health status of the high-speed rail traction and braking system.

[0082] The online measurement data is input into the de-obfuscation model to obtain the adjusted value of the measurement parameter, and then the operation parameter prediction value obtained by inputting the anti-causal multivariate linear regression model is x i ′, then the degradation value of the operating parameter is recorded as:

[0083]

[0084] After obtaining the degradation value of each operating parameter, it is necessary to set the weight of each operating parameter, and use the information entropy H(x i ) is used as the weight, reflecting the overall complexity of the operation parameters.

[0085] Normalization is performed to establish health indicators for high-speed rail traction and braking systems.

[0086]

[0087] The abnormality identification method of multi-sensor fusion of high-speed railway traction and braking system proposed in the embodiment of the present invention breaks through the problems of poor robustness and insufficient generalization ability of virtual health indicators established by traditional unsupervised learning. The core idea is to extract the causal mechanism of the stability of the traction and braking system from the time series data and eliminate the influence of hidden confusion factors, and then predict the operating parameters that characterize the overall health status from the measured parameters through anti-causal learning, and then establish the overall health indicators of the electromechanical system.

[0088] Figure 4 Construct a flow diagram for the health indicators of high-speed rail traction and braking systems.

[0089] In order to implement the above-mentioned embodiment, the present invention also proposes an abnormality identification device of a high-speed railway traction and braking system with multi-sensor fusion.

[0090] Figure 5 A schematic structural diagram of a multi-sensor fusion anomaly recognition device for a high-speed railway traction and braking system provided in an embodiment of the present invention.

[0091] like Figure 5 As shown, the high-speed railway traction brake system multi-sensor fusion abnormality recognition device includes: an acquisition module 100, an adjustment module 200, a prediction module 300, and an identification module 400, wherein:

[0092] An acquisition module is used to acquire multi-parameter sensing data of the high-speed rail traction braking system, divide the multi-parameter sensing data into operating parameters and measurement parameters, and perform pre-processing;

[0093] An adjustment module, used for inputting the preprocessed operation parameters and measurement parameters into a de-obfuscation model to obtain an adjustment value of the measurement parameter;

[0094] A prediction module, used for reversely predicting the operating parameters using the measured parameter adjustment values ​​according to the reverse causal multivariate regression model to obtain the predicted values ​​of the operating parameters;

[0095] The identification module is used to establish the degradation value of each operating parameter through the operating parameters and the predicted values ​​of the operating parameters, and to establish a normalized health indicator model using the complexity of the historical time series of each operating parameter as a weight, so as to evaluate the health status of the high-speed rail traction and braking system.

[0096] Furthermore, in one embodiment of the present invention, the acquisition module is further used to:

[0097] Collect operating parameters, including traction / traction level, braking force / braking force level, acceleration / deceleration setting value, vehicle speed setting value, traction / braking mode;

[0098] Collect measurement parameters, including train speed, traction, braking force, acceleration / deceleration, brake temperature, wheel-track force, and braking distance;

[0099] Data cleaning, missing value processing, outlier detection and removal are performed on operational data and measurement parameters.

[0100] Furthermore, in one embodiment of the present invention, the adjustment module is further used to:

[0101] Inferring latent variables through historical multi-source time series;

[0102] Establish multiple regression models of operating parameters and measurement parameters;

[0103] The multivariate regression model was adjusted according to the latent variables to obtain the adjusted values ​​of the measurement parameters.

[0104] Furthermore, in one embodiment of the present invention, the prediction module is also used to:

[0105] Establish a structural causal model between the adjusted values ​​of measurement parameters and the operational parameters;

[0106] Through the structural causal model, an anti-causal multiple regression model is established with the measured parameter adjustment value as the independent variable and the operating parameter as the dependent variable.

[0107] To achieve the above-mentioned purpose, the third aspect of the present invention proposes a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, it implements the above-mentioned high-speed rail traction and braking system multi-sensor fusion abnormality identification method.

[0108] To achieve the above-mentioned purpose, the fourth aspect of the present invention proposes a computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the above-mentioned high-speed railway traction and braking system multi-sensor fusion abnormality identification method is implemented.

[0109] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0110] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0111] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for identifying abnormalities of a high-speed railway traction and braking system by multi-sensor fusion, characterized in that: The following steps are involved: Acquire multi-parameter sensing data of the high-speed rail traction braking system, divide the multi-parameter sensing data into operating parameters and measurement parameters, and perform preprocessing; Inputting the preprocessed operation parameters and measurement parameters into the de-obfuscation model to obtain the measurement parameter adjustment value; Perform reverse prediction of the operating parameter using the measured parameter adjustment value according to the reverse causal multivariate regression model to obtain a predicted operating parameter value; Establishing degradation values ​​of each operating parameter through the operating parameters and the predicted values ​​of the operating parameters, using the complexity of the historical time series of each operating parameter as a weight, and establishing a normalized health indicator model, thereby evaluating the health status of the high-speed rail traction brake system; The step of inputting the preprocessed operating parameters and measurement parameters into the de-obfuscation model to obtain the measurement parameter adjustment value includes: Inferring latent variables through historical multi-source time series; Establish multiple regression models of operating parameters and measurement parameters; Adjust the multivariate regression model according to the latent variables to obtain the adjusted values ​​of the measurement parameters; The reverse prediction of the operating parameter using the measured parameter adjustment value according to the reverse causal multivariate regression model to obtain the operating parameter prediction value includes: Establish a structural causal model between the adjusted values ​​of measurement parameters and the operational parameters; Through the structural causal model, an anti-causal multiple regression model is established in which the measured parameter adjustment value is an independent variable and the operating parameter is a dependent variable.

2. The method according to claim 1, characterized in that The step of acquiring multi-parameter sensing data of the high-speed railway traction braking system, dividing the multi-parameter sensing data into operating parameters and measurement parameters and performing pre-processing includes: Collect operating parameters, including traction / traction level, braking force / braking force level, acceleration / deceleration setting value, vehicle speed setting value, traction / braking mode; Collect measurement parameters, including train speed, traction, braking force, acceleration / deceleration, brake temperature, wheel-track force, and braking distance; Data cleaning, missing value processing, and outlier detection and removal are performed on the operating parameters and the measurement parameters.

3. A high-speed railway traction and braking system multi-sensor fusion abnormality recognition device, characterized in that: Includes the following modules: An acquisition module, used to acquire multi-parameter sensing data of the high-speed rail traction braking system, divide the multi-parameter sensing data into operating parameters and measurement parameters, and perform preprocessing; An adjustment module, used for inputting the preprocessed operation parameters and measurement parameters into a de-obfuscation model to obtain an adjustment value of the measurement parameter; A prediction module, configured to reversely predict the operating parameters using the measured parameter adjustment values ​​according to an inverse causal multivariate regression model to obtain predicted operating parameter values; an identification module, used to establish degradation values ​​of each operating parameter through the operating parameters and the predicted values ​​of the operating parameters, and to establish a normalized health indicator model using the complexity of the historical time series of each operating parameter as a weight, so as to evaluate the health status of the high-speed rail traction brake system; Wherein, the adjustment module is also used for: Inferring latent variables through historical multi-source time series; Establish multiple regression models of operating parameters and measurement parameters; Adjust the multivariate regression model according to the latent variables to obtain adjusted values ​​of measurement parameters; The prediction module is further used for: Establish a structural causal model between the adjusted values ​​of measurement parameters and operational parameters; Through the structural causal model, an anti-causal multiple regression model is established in which the measured parameter adjustment value is an independent variable and the operating parameter is a dependent variable.

4. The device according to claim 3, characterized in that The acquisition module is further used for: Collect operating parameters, including traction / traction level, braking force / braking force level, acceleration / deceleration setting value, vehicle speed setting value, traction / braking mode; Collect measurement parameters, including train speed, traction, braking force, acceleration / deceleration, brake temperature, wheel-track force, and braking distance; Data cleaning, missing value processing, and outlier detection and removal are performed on the operating parameters and the measurement parameters.

5. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the abnormality identification method of multi-sensor fusion of high-speed railway traction and braking system as described in any one of claims 1-2.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the method for abnormality identification of multi-sensor fusion of high-speed railway traction and braking system as described in any one of claims 1-2 is implemented.

Images