Turnout fault diagnosis method and system based on novel neural network

Abstract

The invention provides a turnout fault diagnosis method and system based on a novel neural network, and belongs to the technical field of fault data analysis, and the method comprises the steps: determining turnout fault diagnosis features based on historical turnout operation state data; and training the initial RBF neural network based on an exponential distribution optimizer to obtain a center parameter and a width parameter of a hidden layer of the RBF neural network. And inputting the turnout fault diagnosis features into the trained RBF neural network to obtain an output weight matrix. And obtaining a turnout fault diagnosis model based on the center parameter, the width parameter and the output weight matrix, and inputting the target turnout operation state data into the turnout fault diagnosis model to obtain a turnout fault diagnosis result. The target turnout operation state data is turnout operation state data to be diagnosed. According to the turnout fault diagnosis method and system based on the novel neural network, the accuracy of turnout fault diagnosis can be improved.

Description

Technical Field

[0001] The present disclosure belongs to the technical field of fault data analysis, and more specifically, to a turnout fault diagnosis method and system based on a novel neural network. Background Art

[0002] As the core switching equipment of the railway transportation system, turnouts not only ensure the flexible dispatch and efficient operation of trains between different lines, but also form the basis of a well-connected railway network. They are also directly related to the overall safety and transportation efficiency of the railway. The maintenance and management of turnouts are extremely important. Any failure may cause serious safety accidents or operational delays. They are one of the key links to ensure the safe and efficient operation of the railway system.

[0003] The microcomputer monitoring system currently deployed by railway electrical departments can display the operating data of various electrical equipment in real time. At the same time, the microcomputer monitoring system also performs simple fault alarms through a series of manually set thresholds. For fault diagnosis of electrical equipment, most of the time, the electrical department relies on technicians to analyze the real-time data of the microcomputer monitoring system to determine the operating status of the equipment.

[0004] In summary, the existing methods for turnout fault diagnosis have low accuracy, and it is necessary to seek more intelligent and automated turnout accurate fault diagnosis methods and technologies. Summary of the invention

[0005] The purpose of the present disclosure is to provide a turnout fault diagnosis method and system based on a novel neural network to improve the accuracy of turnout fault diagnosis.

[0006] A first aspect of the embodiments of the present disclosure provides a turnout fault diagnosis method based on a novel neural network, comprising: Determining turnout fault diagnosis features based on historical turnout operation status data, wherein the historical turnout operation status data includes turnout starting current data, turnout action data, and weather data; Based on the exponential distribution optimizer, the initial RBF neural network is trained to obtain the center parameter and width parameter of the hidden layer of the RBF neural network; the turnout fault diagnosis feature is input into the trained RBF neural network to obtain the output weight matrix; A turnout fault diagnosis model is obtained based on the center parameter, the width parameter and the output weight matrix, and target turnout operation status data is input into the turnout fault diagnosis model to obtain a turnout fault diagnosis result; the target turnout operation status data is the turnout operation status data to be diagnosed.

[0007] A second aspect of the disclosed embodiment provides a turnout fault diagnosis system based on a novel neural network, comprising: A turnout feature extraction module, used to determine turnout fault diagnosis features based on historical turnout operation status data, wherein the historical turnout operation status data includes turnout starting current data, turnout action data and weather data; The network model training module is used to train the initial RBF neural network based on the exponential distribution optimizer to obtain the center parameter and width parameter of the hidden layer of the RBF neural network; input the turnout fault diagnosis feature into the trained RBF neural network to obtain the output weight matrix; A fault diagnosis module is used to obtain a turnout fault diagnosis model based on the center parameter, the width parameter and the output weight matrix, and input the target turnout operation status data into the turnout fault diagnosis model to obtain a turnout fault diagnosis result; the target turnout operation status data is the turnout operation status data to be diagnosed.

[0008] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned method for diagnosing turnout faults based on a novel neural network when executing the computer program.

[0009] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing turnout faults based on a novel neural network are implemented.

[0010] The beneficial effect of a turnout fault diagnosis method and system based on a new neural network provided by the embodiment of the present disclosure is that: the embodiment of the present disclosure integrates multi-source information such as turnout starting current, action data and weather data, which can more comprehensively reflect the operating status of the turnout, avoid the limitations of single data source diagnosis, and improve the accuracy and reliability of fault diagnosis. Using an exponential distribution optimizer to train the RBF neural network can more effectively adjust the center parameter and width parameter of the hidden layer according to the characteristics of the data, so that the neural network can better fit complex fault modes and enhance the generalization ability of the model, thereby accurately diagnosing turnout faults under different working conditions. This embodiment can quickly output fault diagnosis results, provide timely and accurate decision-making basis for railway maintenance personnel, help to reasonably arrange maintenance plans, reduce the impact of turnout failures on railway transportation, and improve the safety and operational efficiency of the railway system. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A schematic diagram of a flow chart of a turnout fault diagnosis method based on a novel neural network provided in one embodiment of the present disclosure; Figure 2 A schematic diagram of a flow chart of a turnout fault diagnosis method based on a novel neural network provided in another embodiment of the present disclosure; Figure 3 A structural block diagram of a turnout fault diagnosis system based on a novel neural network provided in one embodiment of the present disclosure; Figure 4 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0013] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.

[0014] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0015] Please refer to Figure 1 , Figure 1 A schematic flow chart of a method for diagnosing turnout faults based on a novel neural network is provided in accordance with an embodiment of the present disclosure. The method may include S101 to S103.

[0016] S101: Determine turnout fault diagnosis features based on historical turnout operation status data, where the historical turnout operation status data includes turnout starting current data, turnout action data, and weather data.

[0017] In this embodiment, the turnout fault diagnosis features are determined based on the historical turnout operation status data, including: The turnout starting current data characteristics, turnout action characteristics and weather characteristics are determined based on the historical turnout operation status data.

[0018] The turnout fault diagnosis characteristics are obtained based on the turnout starting current data characteristics, turnout action characteristics and weather characteristics.

[0019] In this embodiment, the historical turnout operation status data refers to the turnout operation status data within a certain period of time in the past. The turnout starting current data may include turnout starting current curve data, which is a one-dimensional real number array, and its specific meaning is the value of the current passing through the turnout machine when the turnout machine moves the turnout. For example: 0, 0.48, 0.48...1.56, 1.52...0.04, 0; the turnout action data may include turnout action duration, turnout action type and turnout representation. The switch action duration indicates the duration of the switch machine moving the switch, for example, 8.3 seconds. The switch action type is a Boolean type, which is used to mark whether the switch movement is from the fixed position to the reverse position, or from the reverse position to the fixed position. Example data: 1 (true) represents movement from the reverse position to the fixed position; 0 (false) represents movement from the fixed position to the reverse position. The switch representation is a Boolean type, which is used to identify the position of the switch point rail, for example: 1 (true) represents that the point rail is in the fixed position; 0 (false) represents that the point rail is in the reverse position.

[0020] Weather data can include temperature, humidity, rain and snow, etc., where temperature and humidity are both real number types, which means the temperature and humidity of the area where the turnout is located. Rain and snow are integer types, which means whether it is raining or snowing in the area where the turnout is located. For example: 0 means no rain or snow; 1 means raining; 2 means snowing. Considering that severe weather conditions such as low temperature, high humidity, rain and snow will affect the electrical performance of the turnout and the operating conditions of mechanical components, weather data needs to be collected. For example, low temperature may cause the viscosity of the lubricating oil to increase, affecting the turning flexibility of the turnout, which is then reflected in the changes in the starting current and action characteristics.

[0021] In this embodiment, the characteristics of the turnout starting current data may include characteristics of the change of the current magnitude over time extracted from the turnout starting current data. These characteristics are helpful in judging the turnout fault situation. For example, if the current peak value is far beyond the normal range, it may be that the switch machine has a jamming fault. The turnout action characteristics may include turnout action duration characteristics, turnout action type characteristics, and turnout representation characteristics. Abnormal changes in the turnout action duration characteristics mean that the working efficiency of the switch machine is reduced or there is a mechanical failure; the recording of the action type characteristics helps to analyze the performance differences of the turnout in different running directions; the turnout representation characteristics can reflect the actual position status of the turnout point rail, and when it does not match the expected position, it may indicate that there is a problem with the turnout position detection or conversion mechanism.

[0022] For example, the starting current data, action data and weather data of the turnout in the past period of time are continuously collected, and stored and sorted in the prescribed data format to ensure the accuracy and completeness of the data. For the starting current data, various characteristic parameters as mentioned above are extracted; for the turnout action data, logical judgment and feature quantification are performed based on the duration, type and representation information of the records; for the weather data, the potential impact of different weather conditions on the turnout operation is analyzed in combination with meteorological knowledge and railway equipment operation experience, and converted into feature vectors that can be used by the model.

[0023] The above-extracted turnout starting current data features, turnout action features and weather features are integrated, and a turnout fault diagnosis feature set is constructed through feature combination and weighting, which serves as the input of the subsequent fault diagnosis model.

[0024] S102: Train the initial RBF neural network based on the exponential distribution optimizer to obtain the center parameter and width parameter of the hidden layer of the RBF neural network. Input the turnout fault diagnosis features into the trained RBF neural network to obtain the output weight matrix.

[0025] In this embodiment, the exponential distribution optimizer is an optimization algorithm based on an exponential distribution probability model, which aims to achieve an optimal value of an objective function (such as a loss function of a neural network) by adjusting parameters. The exponential distribution optimizer may include a rate parameter of the exponential distribution, which controls the shape of the exponential distribution, thereby affecting the search step size and direction during the optimization process. In addition, parameters such as the number of iterations and the learning rate decay factor may also be included to control the convergence speed and accuracy of the optimization process.

[0026] RBF neural network refers to radial basis function neural network (RBF), which is a feedforward neural network with good approximation ability and generalization performance. The characteristics of RBF neural network are that it has a single hidden layer, and the activation function of the hidden layer neurons is the radial basis function. The radial basis function is a function with the distance between the input vector and the center vector as the independent variable, such as the Gaussian function.

[0027] In this embodiment, the center parameter of the hidden layer of the RBF neural network usually refers to a vector with the same dimension as the input data, which determines the center position of the radial basis function; the width parameter controls the width of the radial basis function, affecting its response range to the input data; the weight matrix of the output layer is used to connect the hidden layer and the output layer, thereby determining the final output result. The number of neurons in the input layer is usually consistent with the characteristic dimension of the input data, and the number of neurons in the output layer depends on the number of categories that need to be classified or predicted.

[0028] Exemplarily, an initial RBF neural network is constructed, the number of nodes in the input layer, hidden layer and output layer is determined, and the center parameter and width parameter of the hidden layer and the weight matrix of the output layer are randomly initialized.

[0029] Using the exponential distribution optimizer, the turnout fault diagnosis features are used as input data, and the center parameter and width parameter of the hidden layer are adjusted iteratively to minimize the error between the predicted output of the neural network and the actual fault label. In each iteration, new parameter update values ​​are generated according to the characteristics of the exponential distribution, gradually optimizing the network performance.

[0030] After the hidden layer parameters are determined, the turnout fault diagnosis features are input into the trained RBF neural network again, and the output weight matrix is ​​obtained by solving the linear equations, such as using the least squares method, so that the neural network can accurately diagnose and classify the turnout faults.

[0031] S103: A turnout fault diagnosis model is obtained based on the center parameter, the width parameter and the output weight matrix, and the target turnout operation state data is input into the turnout fault diagnosis model to obtain a turnout fault diagnosis result. The target turnout operation state data is the turnout operation state data to be diagnosed.

[0032] In this embodiment, for turnout fault diagnosis, if the input features include turnout starting current, action duration, etc., the center parameter can be a multidimensional vector, such as [starting current center value, action duration center value, ...], whose value is determined according to the distribution characteristics of historical data, for example, the starting current center value can be the typical value of the starting current during normal operation. The width parameter can be a vector with the same dimension as the center parameter, such as [starting current width value, action duration width value, ...]. The larger its value, the flatter the radial basis function and the wider the response range to the input; the smaller its value, the sharper the function and the narrower the response range.

[0033] Exemplarily, the structure and parameters of the RBF neural network are determined according to the center parameter, width parameter and output weight matrix obtained by training, and a turnout fault diagnosis model is constructed. The target turnout operation status data is preprocessed and feature extracted to make it consistent with the data format during model training, and then input into the turnout fault diagnosis model. The turnout fault diagnosis model calculates the distance between the input data and the hidden layer center parameter, obtains the hidden layer output through the radial basis function, and then multiplies it with the output weight matrix to obtain the final fault diagnosis result, which is output as a probability value or category label of normal or a certain fault type.

[0034] It can be concluded from the above that this embodiment integrates multi-source information such as turnout starting current, action data, and weather data, which can more comprehensively reflect the operating status of the turnout, avoid the limitations of single data source diagnosis, and improve the accuracy and reliability of fault diagnosis. Using the exponential distribution optimizer to train the RBF neural network can more effectively adjust the center parameter and width parameter of the hidden layer according to the data characteristics, so that the neural network can better fit complex fault modes and enhance the generalization ability of the model, so as to accurately diagnose turnout faults under different working conditions. This embodiment can quickly output fault diagnosis results, provide timely and accurate decision-making basis for railway maintenance personnel, help to reasonably arrange maintenance plans, reduce the impact of turnout failures on railway transportation, and improve the safety and operational efficiency of the railway system.

[0035] In one embodiment of the present disclosure, the characteristics of turnout starting current data, turnout action characteristics and weather characteristics are determined based on historical turnout operation status data, including: The historical turnout operation status data is divided into multiple fault data subsets based on the turnout fault type. Each fault data subset corresponds to a turnout fault type.

[0036] Data sampling is performed based on the data size of each fault data subset.

[0037] Based on the sampled data, the turnout starting current data characteristics, turnout action characteristics and weather characteristics are determined.

[0038] In this embodiment, the switch starting current data features include original features, integral features and statistical features corresponding to the switch starting current data.

[0039] Based on the characteristics of turnout starting current data, turnout action characteristics and weather characteristics, turnout fault diagnosis characteristics are obtained, including: The original features, integral features and statistical features of the turnout starting current data are extracted, the turnout action features of the turnout action data are extracted, and the weather features of the weather data are extracted.

[0040] A first turnout feature is generated based on the original feature, the integral feature and the statistical feature.

[0041] The turnout action feature and weather feature are embedded into the first turnout feature to obtain a turnout feature set.

[0042] The turnout feature set is extracted based on the path integral algorithm to obtain the turnout fault diagnosis features.

[0043] In this embodiment, the turnout fault data subset refers to the classification and division of the historical turnout operation status data according to different fault types. Each turnout fault data subset represents a specific fault mode, so that independent analysis and feature extraction can be performed for each fault type. The data volume and specific values ​​vary depending on the fault type and the actual number of occurrences. The first turnout feature is an intermediate feature set generated based on the relevant features of the turnout starting current data during the turnout fault diagnosis process.

[0044] In this embodiment, the path integration method can be used to obtain the integral characteristics of the turnout starting current data. The original characteristics are basic characteristics directly obtained from the starting current curve data, such as the maximum value of the current. The statistical characteristics refer to the characteristics obtained by statistical analysis of the starting current data, such as the mean, variance, etc.

[0045] For example, various sensors installed on the turnout, such as current sensors, position sensors, meteorological sensors, etc., can be used to continuously collect the turnout starting current, operation status, and surrounding weather data, and store them in time series. Combined with historical turnout fault records and maintenance reports, a large amount of historical turnout operation status data is classified according to known fault types to form multiple fault data subsets.

[0046] For example, in order to balance the data volume differences between different fault data subsets and avoid model training bias caused by data imbalance, data sampling is performed on each subset. By calculating the data scale of each subset, a benchmark data volume is determined, such as taking the mean as the benchmark data volume. For subsets with less data volume than the benchmark data volume, an upsampling method is used to increase the data volume, and for subsets with more data volume than the benchmark data volume, a downsampling method is used to reduce the data volume, so as to ensure that each subset has a relatively balanced influence in subsequent analysis.

[0047] Exemplarily, the original feature, integral feature and statistical feature of the starting current are combined to generate the first turnout feature, and then the turnout action feature and weather feature are embedded to form a turnout feature set, integrating multiple aspects of information. The turnout feature set is further processed using the path integral algorithm to mine hidden features and obtain turnout fault diagnosis features that can accurately reflect the turnout fault condition, providing high-quality input for subsequent fault diagnosis models.

[0048] This embodiment divides the data subsets by fault type, and can analyze different fault modes independently. By using methods such as path integral to extract features, the operation characteristics of the turnout can be fully obtained. Data sampling balances the amount of data in each subset, avoids model training bias, improves the model's ability to identify different faults and diagnostic accuracy, and provides strong support for the maintenance of railway turnouts.

[0049] In one embodiment of the present disclosure, data sampling is upsampling or downsampling. Data sampling is performed based on the data size of each fault data subset, including: The data size of each fault data subset is calculated to obtain the data size sequence corresponding to all fault data subsets.

[0050] A first data size is determined based on the data size sequence.

[0051] For each failure data subset: A first sampling value is determined based on a weight coefficient of the fault data subset and a first data scale.

[0052] In response to the data size of the fault data subset being smaller than the first sampling value, the fault data subset is up-sampled based on the first sampling value of the fault data subset.

[0053] In response to the data size of the fault data subset being larger than the first sampling value, the fault data subset is downsampled based on the first sampling value of the fault data subset.

[0054] In this embodiment, the data scale can be the number of samples in the fault data subset. After calculating the data scale of all fault data subsets, they are arranged in order according to the numerical value to obtain a data scale sequence. The weight coefficient of the fault data subset reflects the importance, occurrence probability or data scale of each fault data subset in the overall data, which can be obtained based on the statistical analysis of historical fault data. The first data scale is a reference value or statistical value selected from the data scale sequence, which is used to determine the number of samples for each fault data subset as a benchmark for balancing the data volume. The first sampling value refers to the number of samples, such as sampling 200 data.

[0055] Exemplarily, the first data scale can be a variety of statistical data such as the maximum value, minimum value, average value or median in the data scale sequence, and different parameters can be selected according to actual needs. If the maximum value in the data scale sequence is used as the first data scale, the fault data subset is upsampled based on the weight coefficient of the fault data subset and the first data scale, including: substituting the weight coefficient of the fault data subset and the first data scale into the sampling value calculation formula to obtain the first sampling value.

[0056] The sampling value calculation formula is: =M

[0057] in, represents the first sampling value of the i-th fault data subset, M represents the first data scale, Represents the weight coefficient of the i-th fault data subset.

[0058] Compare the data size of each fault data subset with the first sampling value. If the data size is smaller than the corresponding first sampling value, increase the data volume by copying, interpolating, etc., i.e. upsampling; if the data size is larger than the first sampling value, randomly extract part of the data to reduce the data volume, i.e. downsampling.

[0059] This embodiment determines the reference value by calculating the size of each fault data subset, and accurately determines the number of samples by combining the weight coefficient. Up-sampling and down-sampling are performed according to the sampling value, effectively balancing the data volume of different fault data subsets. This can avoid model training deviation caused by data imbalance, enable the model to learn various fault characteristics more comprehensively, improve the recognition and diagnosis capabilities of various turnout faults, and ensure the accuracy and reliability of railway turnout fault diagnosis.

[0060] In one embodiment of the present disclosure, the turnout fault diagnosis feature is input into the trained RBF neural network to obtain an output weight matrix, including: The turnout fault diagnosis features are input into the trained RBF neural network, the output matrix corresponding to the turnout fault diagnosis features is calculated based on the radial basis function of the hidden layer, and the pseudo-inverse matrix corresponding to the output matrix is ​​calculated.

[0061] The output weight matrix is ​​determined based on the pseudo-inverse matrix and the fault type labels corresponding to the turnout fault diagnosis features.

[0062] In this embodiment, the output matrix refers to the matrix obtained by the hidden layer radial basis function calculation after the turnout fault diagnosis features are input into the RBF neural network. The number of rows is related to the number of neurons in the hidden layer, the number of columns is related to the number of input samples, and the matrix elements represent the calculated output values. For a non-square output matrix, the pseudo-inverse matrix is ​​its generalized inverse matrix, which is used to solve the problem of solving a system of linear equations, and its dimension is related to the output matrix. The fault type label is a predefined mark indicating the type of turnout fault. Each label corresponds to a fault category, usually expressed in vector form, and is related to the determination of the output weight matrix.

[0063] Exemplarily, the turnout fault diagnosis features are input into the trained RBF neural network, and each input feature is calculated by the radial basis function of the hidden layer. For each input sample, each neuron in the hidden layer calculates an output value according to the calculation rule of the radial basis function, and these output values ​​form the output matrix. Since the output matrix may be a non-square matrix, it cannot be directly inverted to solve the linear equations. Therefore, singular value decomposition is used to calculate the pseudo-inverse matrix of the output matrix to prepare for the subsequent solution of the output weight matrix.

[0064] The output weight matrix is ​​solved by the least square method using the pseudo-inverse matrix and the corresponding fault type label. The output weight matrix establishes the connection between the hidden layer output and the fault type, so that the network can accurately output the fault diagnosis results according to the input turnout fault diagnosis characteristics. For example, when new turnout data is input, the probability of each fault type can be calculated through the output weight matrix, thereby determining the most likely fault type.

[0065] like Figure 2 As shown, the complete fault diagnosis process can be divided into four modules: data acquisition and analysis module, turnout feature extraction module, model parameter solution module and real-time fault diagnosis module.

[0066] The turnout feature extraction module extracts features from the turnout historical data from the data acquisition and analysis module. Then, it is sent to the model parameter solution module. The model solution module solves the various parameters of the RBF neural network (the center and width of the hidden layer nodes, the weight vector from the hidden layer nodes to the output layer nodes), and saves the parameters to the local file rbfParam.npy. The file rbfParam.npy is copied to the turnout fault diagnosis data monitoring center where the real-time fault diagnosis module is located.

[0067] The real-time fault diagnosis module reconstructs the network model EDO_RBFNN using the parameters of the RBF neural network saved in the file rbfParam.npy. The data acquisition and analysis module collects the turnout operation-related data from the microcomputer monitoring system in real time and sends it to the turnout feature extraction module. After the turnout feature extraction module extracts the turnout operation-related data just received in real time, it sends the extraction results to the real-time fault diagnosis module. The real-time fault diagnosis module uses the EDO_RBFNN model to perform fault diagnosis on the data from the turnout feature extraction module. The real-time fault diagnosis module sends the diagnosis results to the human-computer interaction interface. And decide whether to issue a sound alarm based on the results. If the diagnosis is a fault, a sound and text alarm will be issued to remind the staff to deal with it quickly; if the diagnosis result is wrong, the data and its correct type will be sent to the model parameter solver. The model parameter solver will start the model parameter correction process after the diagnostic accuracy of the fault diagnosis model is reduced to a certain extent, so as to further increase the coverage of the turnout fault diagnosis fault data of the real-time fault diagnosis module.

[0068] This embodiment uses RBF neural network to solve parameters and reconstruct models, and can accurately output fault diagnosis results. Feedback of diagnostic results is used for model parameter correction to improve diagnostic accuracy and fault data coverage. Sound and text alarms remind staff to handle faults in a timely manner, ensuring the safety of railway turnout operation and reducing fault risks and operational losses.

[0069] In one embodiment of the present disclosure, a turnout fault diagnosis method based on a novel neural network further includes: The initialization parameters of the input layer of the RBF neural network are determined based on the number of data dimensions of the historical turnout operation status data.

[0070] The initialization parameters of the output layer of the RBF neural network are determined based on the number of turnout fault types.

[0071] The initial RBF neural network is constructed based on the initialization parameters of the input layer, the output layer and the hidden layer.

[0072] In this embodiment, the number of data dimensions of the historical turnout operation status data refers to the number of different features extracted from the historical turnout operation status data for describing the turnout operation. The parameters include the number of turnout starting current features, the number of turnout action features, the number of weather features, etc. These features together constitute the dimension of the data.

[0073] Exemplarily, the turnout fault types may include normal, starting current disconnection, external line mixing, starting delay, motor inter-turn short circuit, failure to unlock, poor conversion, conversion process timeout, etc. Based on the configuration parameters of the RBF neural network: the number of input layer nodes is 40, the number of output layer nodes is 8, the number of hidden layer nodes is M (needs to be finally determined by multiple attempts according to the test results in the experiment), and the hidden layer node activation function is a Gaussian function.

[0074] Exemplarily, the historical turnout operation status data is analyzed and the number of data dimensions is counted to determine the number of neurons in the input layer of the RBF neural network to ensure that the input layer can receive and process all relevant feature information. According to the number of known turnout fault types, the number of neurons in the output layer is determined, and an output neuron is assigned to each fault type so that the network outputs prediction results for different fault types. The number of hidden layer neurons is set based on experience or debugging, and the center parameter and width parameter of the radial basis function are randomly initialized. According to the determined initialization parameters of the input layer, output layer, and hidden layer, the initial RBF neural network architecture is built to prepare for subsequent training and fault diagnosis.

[0075] This embodiment determines the input layer parameters according to the data dimension, and can fully receive the turnout operation characteristics. The output layer parameters are determined according to the fault type, which is conducive to the accurate output of fault prediction. A reasonable neural network architecture including a hidden layer is constructed to provide a data basis for subsequent training and diagnosis. The whole process makes the neural network closely fit the turnout fault diagnosis needs, effectively improves the diagnostic accuracy, and ensures the safe and stable operation of railway turnouts.

[0076] Corresponding to the above embodiment, a turnout fault diagnosis method based on a novel neural network is described. Figure 3This is a structural block diagram of a turnout fault diagnosis system based on a novel neural network provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Figure 3 The turnout fault diagnosis system 20 based on the novel neural network includes: a turnout feature extraction module 21, a network model training module 22 and a fault diagnosis module 23.

[0077] The turnout feature extraction module 21 is used to determine the turnout fault diagnosis features based on historical turnout operation status data, and the historical turnout operation status data includes turnout starting current data, turnout action data and weather data.

[0078] The network model training module 22 is used to train the initial RBF neural network based on the exponential distribution optimizer to obtain the center parameter and width parameter of the hidden layer of the RBF neural network. The turnout fault diagnosis features are input into the trained RBF neural network to obtain the output weight matrix.

[0079] The fault diagnosis module 23 is used to obtain a turnout fault diagnosis model based on the center parameter, the width parameter and the output weight matrix, and input the target turnout operation state data into the turnout fault diagnosis model to obtain a turnout fault diagnosis result. The target turnout operation state data is the turnout operation state data to be diagnosed.

[0080] In one embodiment of the present disclosure, the turnout feature extraction module 21 is specifically used to determine the turnout starting current data features, turnout action features and weather features based on historical turnout operation status data.

[0081] The turnout fault diagnosis characteristics are obtained based on the turnout starting current data characteristics, turnout action characteristics and weather characteristics.

[0082] In one embodiment of the present disclosure, the turnout feature extraction module 21 is further used to divide the historical turnout operation status data into a plurality of fault data subsets based on the turnout fault type. Each fault data subset corresponds to a turnout fault type.

[0083] Data sampling is performed based on the data size of each fault data subset.

[0084] Based on the sampled data, the turnout starting current data characteristics, turnout action characteristics and weather characteristics are determined.

[0085] In one embodiment of the present disclosure, data sampling is upsampling or downsampling. The turnout feature extraction module 21 is further used to calculate the data size of each fault data subset to obtain a data size sequence corresponding to all fault data subsets.

[0086] A first data size is determined based on the data size sequence.

[0087] For each failure data subset: A first sampling value is determined based on a weight coefficient of the fault data subset and a first data scale.

[0088] In response to the data size of the fault data subset being smaller than the first sampling value, the fault data subset is up-sampled based on the first sampling value of the fault data subset.

[0089] In response to the data size of the fault data subset being larger than the first sampling value, the fault data subset is downsampled based on the first sampling value of the fault data subset.

[0090] In one embodiment of the present disclosure, the switch start current data features include original features, integral features and statistical features corresponding to the switch start current data. The switch feature extraction module 21 is also specifically used to extract the original features, integral features and statistical features of the switch start current data, extract the switch action features of the switch action data, and extract the weather features of the weather data.

[0091] A first turnout feature is generated based on the original feature, the integral feature and the statistical feature.

[0092] The turnout action feature and weather feature are embedded into the first turnout feature to obtain a turnout feature set.

[0093] The turnout feature set is extracted based on the path integral algorithm to obtain the turnout fault diagnosis features.

[0094] In one embodiment of the present disclosure, the network model training module 22 is specifically used to input the turnout fault diagnosis features into the trained RBF neural network, calculate the output matrix corresponding to the turnout fault diagnosis features based on the radial basis function of the hidden layer, and calculate the pseudo-inverse matrix corresponding to the output matrix.

[0095] The output weight matrix is ​​determined based on the pseudo-inverse matrix and the fault type labels corresponding to the turnout fault diagnosis features.

[0096] In one embodiment of the present disclosure, a turnout fault diagnosis system 20 based on a novel neural network further includes: The model initialization module is used to determine the initialization parameters of the input layer of the RBF neural network based on the number of data dimensions of the historical turnout operation status data.

[0097] The initialization parameters of the output layer of the RBF neural network are determined based on the number of turnout fault types.

[0098] The initial RBF neural network is constructed based on the initialization parameters of the input layer, the output layer and the hidden layer.

[0099] See also Figure 4 , Figure 4 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 4 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303 and one or more memories 304. The processors 301, input devices 302, output devices 303 and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned system embodiments, such as Figure 3 The functions of modules 21 to 23 are shown.

[0100] It should be understood that in the embodiment of the present disclosure, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0101] The input device 302 may include a touch panel, a fingerprint collection sensor (for collecting the user's fingerprint information and fingerprint direction information), a microphone, etc., and the output device 303 may include a display (LCD, etc.), a speaker, etc.

[0102] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.

[0103] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present disclosure can execute the implementation methods described in the first and second embodiments of a turnout fault diagnosis method based on a new neural network provided in the embodiments of the present disclosure, and can also execute the implementation methods of the electronic device 300 described in the embodiments of the present disclosure, which will not be repeated here.

[0104] In another embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by the processor, all or part of the processes in the above-mentioned embodiment method are implemented, and the computer program can also be completed by instructing the relevant hardware through the computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0105] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0106] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] In the several embodiments provided in the present application, it should be understood that the disclosed electronic device and method can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or it can be an electrical, mechanical or other form of connection.

[0109] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present disclosure.

[0110] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0111] The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and these modifications or replacements should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A turnout fault diagnosis method based on a new neural network, characterized in that: include: Determining turnout fault diagnosis features based on historical turnout operation status data, wherein the historical turnout operation status data includes turnout starting current data, turnout action data, and weather data; Based on the exponential distribution optimizer, the initial RBF neural network is trained to obtain the center parameter and width parameter of the hidden layer of the RBF neural network; the turnout fault diagnosis feature is input into the trained RBF neural network to obtain the output weight matrix; A turnout fault diagnosis model is obtained based on the center parameter, the width parameter and the output weight matrix, and target turnout operation status data is input into the turnout fault diagnosis model to obtain a turnout fault diagnosis result; the target turnout operation status data is the turnout operation status data to be diagnosed.

2. A turnout fault diagnosis method based on a novel neural network as claimed in claim 1, characterized in that: The method of determining the turnout fault diagnosis characteristics based on historical turnout operation status data includes: Determine the turnout starting current data characteristics, turnout action characteristics and weather characteristics based on historical turnout operation status data; The turnout fault diagnosis characteristics are obtained based on the turnout starting current data characteristics, the turnout action characteristics and the weather characteristics.

3. A turnout fault diagnosis method based on a novel neural network as claimed in claim 2, characterized in that: The method of determining the turnout starting current data characteristics, turnout action characteristics and weather characteristics based on the historical turnout operation status data includes: The historical turnout operation status data is divided into multiple fault data subsets based on the turnout fault type; each fault data subset corresponds to a turnout fault type; Perform data sampling based on the data size of each fault data subset; Based on the sampled data, the turnout starting current data characteristics, turnout action characteristics and weather characteristics are determined.

4. A turnout fault diagnosis method based on a novel neural network as claimed in claim 3, characterized in that: The data sampling is upsampling or downsampling; The data sampling based on the data scale of each fault data subset includes: Calculate the data size of each fault data subset to obtain the data size sequence corresponding to all fault data subsets; Determining a first data size based on the data size sequence; For each failure data subset: A first sampling value is determined based on a weight coefficient of the fault data subset and the first data scale. In response to the data size of the fault data subset being smaller than the first sampling value, the fault data subset is upsampled based on the first sampling value of the fault data subset. In response to the data size of the fault data subset being larger than the first sampling value, the fault data subset is downsampled based on the first sampling value of the fault data subset.

5. A turnout fault diagnosis method based on a novel neural network as claimed in claim 2, characterized in that: The switch start current data features include original features, integral features and statistical features corresponding to the switch start current data; The step of obtaining the turnout fault diagnosis feature based on the turnout starting current data feature, the turnout action feature and the weather feature includes: Extract the original features, integral features and statistical features of the turnout starting current data, extract the turnout action features of the turnout action data, and extract the weather features of the weather data; Generate a first turnout feature based on the original feature, the integrated feature and the statistical feature; Embedding the turnout action feature and the weather feature into the first turnout feature to obtain a turnout feature set; The turnout feature set is extracted based on a path integral algorithm to obtain turnout fault diagnosis features.

6. A turnout fault diagnosis method based on a novel neural network as claimed in claim 1, characterized in that: The step of inputting the turnout fault diagnosis feature into the trained RBF neural network to obtain an output weight matrix includes: Inputting the turnout fault diagnosis feature into the trained RBF neural network, calculating the output matrix corresponding to the turnout fault diagnosis feature based on the radial basis function of the hidden layer, and calculating the pseudo-inverse matrix corresponding to the output matrix; An output weight matrix is ​​determined based on the pseudo-inverse matrix and the fault type labels corresponding to the turnout fault diagnosis features.

7. A turnout fault diagnosis method based on a novel neural network as claimed in claim 1, characterized in that: Also includes: Determine the initialization parameters of the input layer of the RBF neural network based on the number of data dimensions of the historical turnout operation status data; Determine the initialization parameters of the output layer of the RBF neural network based on the number of turnout fault types; An initial RBF neural network is constructed based on the initialization parameters of the input layer, the initialization parameters of the output layer and the initialization parameters of the hidden layer.

8. A turnout fault diagnosis system based on a new neural network, characterized in that: include: A turnout feature extraction module, used to determine turnout fault diagnosis features based on historical turnout operation status data, wherein the historical turnout operation status data includes turnout starting current data, turnout action data and weather data; The network model training module is used to train the initial RBF neural network based on the exponential distribution optimizer to obtain the center parameter and width parameter of the hidden layer of the RBF neural network; input the turnout fault diagnosis feature into the trained RBF neural network to obtain the output weight matrix; A fault diagnosis module is used to obtain a turnout fault diagnosis model based on the center parameter, the width parameter and the output weight matrix, and input the target turnout operation status data into the turnout fault diagnosis model to obtain a turnout fault diagnosis result; the target turnout operation status data is the turnout operation status data to be diagnosed.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, 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 7 are implemented.

Images