New energy vehicle GPS antenna fault detection method and system

Abstract

The present invention discloses a new energy vehicle GPS antenna fault detection method and system, including S1: used for collecting antenna behavior data, and obtaining key behavior information by improving principal component analysis operation; S2: used for taking the key behavior information as an input vector of a support vector machine model, and outputting a first detection result through the support vector machine model; S3: used for collecting antenna status data, and taking the antenna status data as an input layer of a neural network model based on a multi-layer perceptron, and outputting a second detection result; S4: used for combining the first detection result and the second detection result, and outputting a final fault detection result. Through the collaborative work of the support vector machine model and the neural network model based on the multi-layer perceptron, these differences can be accurately captured and distinguished, thereby significantly improving the accuracy of short-circuit position judgment, and effectively solving the problem of being difficult to accurately locate the short-circuit fault position in the past.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile detection, and in particular to a method and system for detecting faults of a GPS antenna of a new energy automobile. Background Art

[0002] With the rapid development of new energy vehicles, their intelligence level is constantly improving, and their dependence on the global positioning system (GPS) is increasing. GPS plays a vital role in new energy vehicles. However, as a key component for receiving GPS signals, the GPS antenna will have a serious impact on the normal operation of new energy vehicles once it fails.

[0003] In the existing antenna system, faults may occur in different positions, such as inside the antenna feeder, between the antenna vibrator and the grounding point, etc. Faults in different positions have different effects on antenna performance. For example, a short circuit of the antenna feeder near the antenna end may lead to enhanced signal reflection, while a short circuit of the feeder near the receiver end may cause a sharp drop in signal strength. If the short circuit faults at these different positions are not classified and analyzed, when a fault occurs, the diagnostic system may not be able to accurately determine the specific location of the short circuit, and the positions need to be checked one by one, which affects the pertinence and efficiency of maintenance. Therefore, a new energy vehicle GPS antenna fault detection method and system are proposed. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art that the actual location of the short circuit cannot be directly obtained and needs to be checked in sequence, thereby affecting the pertinence and efficiency of maintenance, and to propose a new energy vehicle GPS antenna fault detection method and system.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] New energy vehicle GPS antenna fault detection method and system:

[0007] The above technical solution further includes: a method for detecting a fault of a GPS antenna of a new energy vehicle, comprising:

[0008] S1: used to collect antenna behavior data and obtain key behavior information by improving principal component analysis operations;

[0009] S2: used to use the key behavior information as an input vector of a support vector machine model, and output a first detection result through the support vector machine model;

[0010] S3: used to collect antenna status data, use the antenna status data as an input layer of a neural network model based on a multi-layer perceptron, and output a second detection result;

[0011] S4: used to combine the first detection result and the second detection result to output a final fault detection result.

[0012] Preferably, the antenna behavior data includes signal strength, carrier-to-noise ratio, phase noise, voltage and current at different positions of the antenna feeder. Assume that the number of collected data samples is , each sample type is .

[0013] Preferably, the steps of obtaining key behavior information by improving the principal component analysis operation are:

[0014] The antenna behavior data is formed into a raw data matrix , whose dimensions are , centering each element in the matrix: ,in, , , represents the element mean, Represents the centralized data matrix and introduces a series of weight vectors , calculate the weighted covariance matrix , ,in, for The transpose of is a diagonal weight matrix, the diagonal elements are composed of a series of weight vectors Composition, perform eigenvalue decomposition on the weighted covariance matrix to obtain the key eigenvalues and the corresponding eigenvector , based on the key eigenvalues Get key contribution value , select before feature vectors, The eigenvectors form a principal component, and the centralized data matrix Projection to the front feature vector In the principal component space composed of ,in, ;

[0015] is a Matrix, each row represents the coordinates of a sample in the principal component space. These coordinates are the extracted key behavior information, expressed as .

[0016] Preferably, based on key feature values Get key contribution value The steps are:

[0017] According to the key characteristic value Sort the eigenvectors from large to small and obtain the top The key contribution value formula is: , determine a contribution threshold, when When the corresponding contribution threshold is reached, the corresponding feature vector data is .

[0018] Furthermore, in reality, the working environment of the GPS antenna of new energy vehicles is complex, the amount of behavioral data is large and contains a variety of interference factors. Traditional principal component analysis cannot effectively highlight important information. By introducing weight vectors to calculate the weighted covariance matrix, weights can be assigned according to the importance of data features. For example, when the signal strength data is stable but the phase noise fluctuates greatly and is more critical to fault judgment, the weight of phase noise-related features can be increased, so that the extracted key behavioral information is more focused on features closely related to the fault, providing a more accurate basis for subsequent fault detection;

[0019] Preferably, the steps of obtaining the first detection result are:

[0020] Build a support vector machine model;

[0021] Preferably, the construction process of the support vector machine model is:

[0022] Select a kernel function and train the model;

[0023] The key feature value and the corresponding eigenvector As model training data, key feature values The corresponding eigenvector As a set of training data, and select a kernel function;

[0024] Furthermore, the purpose of the model selecting the kernel function and solving the optimization problem is to find the optimal classification hyperplane so that the model can accurately classify the data;

[0025] First, minimize a cost function. The cost function consists of two parts: one is the penalty for misclassified points, and the other is the complexity of the hyperplane. The cost function is used to solve the optimization problem of the model, so as to find the optimal classification hyperplane, so that the model can accurately classify the data, which is expressed as: ,in, is a regularization term used to control the complexity of the model. is the normal vector of the hyperplane, which determines the direction of the hyperplane, and b is the bias term, which determines the position of the hyperplane. Represents the slack variable of the i-th group of training data, which is used to deal with noise or outliers in the data;

[0026] Then a constraint is used to ensure that the sample is correctly classified or as close to the correct classification as possible. The constraint is: ,in, represents the transpose of the hyperplane normal vector, Indicates Set training data, Indicates that the input data is the key behavior data, Represents the input data after being mapped by the kernel function;

[0027] Obtain the optimal solution through optimization problem and get the optimal hyperplane parameters. After satisfying the constraints, a trained support vector machine model is obtained.

[0028] After the training is completed, the key behavior information is used as the input vector of the support vector machine and the discrimination result is output: ,like If it is greater than or equal to 0, it is initially judged as no fault; if If it is less than 0, it is initially judged as a fault, and the fault type is further determined.

[0029] Furthermore, the kernel function selected after the key behavior information obtained by principal component analysis is used as the input vector of the support vector machine model can be a polynomial kernel function, and the formula is: ,in, is the scaling factor, c is a constant term, is the degree of the polynomial, and the polynomial kernel function can be adjusted by adjusting the parameters (the degree of the polynomial) to control the complexity of the decision boundary. When , the polynomial kernel function degenerates into a linear kernel function, which is suitable for linearly separable data; when It can handle nonlinear data and as As the number of nodes increases, the decision boundary becomes more complex. In antenna fault detection, short-circuit faults in different locations may require decision boundaries of different complexities to distinguish. For example, for short-circuit faults of the antenna feeder close to the antenna end and close to the receiver end, decision boundaries of different shapes may be required for accurate classification. The polynomial kernel function can adapt to this requirement by adjusting parameters.

[0030] Preferably, the steps of acquiring the antenna status data are:

[0031] Preferably, the antenna status data includes antenna connection status data: by detecting whether the connection between the antenna and the receiver is firm, loose or disconnected, using a contact sensor or a resistance measuring device to detect the resistance of the connection point. If the resistance is abnormally high, there may be a problem of poor connection;

[0032] Antenna damage data: Check the appearance of the antenna to see if there is any physical damage, such as whether the antenna element is deformed or broken, or whether the antenna casing is damaged;

[0033] Antenna environment interference data: Use electromagnetic interference detection equipment to measure the electromagnetic environment around the antenna and obtain information such as the strength and frequency of the interference signal;

[0034] Signal strength value: Receive GPS satellite signals through the antenna and use a signal strength measurement device to obtain the signal strength value in decibels;

[0035] Phase noise data: Use phase noise measurement equipment to measure the phase noise of the signal received by the antenna. Phase noise will affect the demodulation and decoding process of the GPS signal, thereby affecting the positioning accuracy.

[0036] Preferably, the step of obtaining the second detection result is:

[0037] Build a neural network model based on multi-layer perceptron;

[0038] Preferably, the neural network model based on the multi-layer perceptron includes an output layer, two hidden layers and an output layer;

[0039] Furthermore, the input layer has neurons, To obtain the number of data points for state data, the first hidden layer has neurons, and the second layer has neurons, and the output layer has neurons;

[0040] Preferably, the first hidden layer is expressed as:

[0041] The antenna state data is input into the first hidden layer through the input layer. neurons, , the input process is expressed as: ,in, is the connection weight from the input layer to the first hidden layer, The first hidden layer The bias term of a neuron, For the The input data corresponding to each neuron is the antenna status data;

[0042] For the first hidden layer neurons, whose output is ,in, is the activation function;

[0043] Perform random dropout on the output of the first hidden layer;

[0044] The output of the neurons in the first hidden layer , with probability Randomly set its output to , for the neuron output Generate a Uniformly distributed random numbers ,when ,but ,in, Represents the output of the first hidden layer Output after random dropout operation;

[0045] Furthermore, by performing a random inactivation operation on each output of the first hidden layer, each neuron has a probability is set to ,have The probability remains the same, but is multiplied by , the purpose of this is to keep the expected value of the neuron output unchanged during training;

[0046] Preferably, the second hidden layer process is expressed as:

[0047] The neurons in the second hidden layer receive the output from the first hidden layer after random inactivation, and further extract features. neurons, , the output process is expressed as: ,in, is the connection weight from the input layer to the second hidden layer, The second hidden layer The bias term of a neuron, Represents the output of the second hidden layer;

[0048] For the neuron output of the second hidden layer , the same random dropout operation is performed to obtain , and at the same time, for the connection weights from the second hidden layer to the output layer , which becomes ,in, is the output layer neuron index, which is determined by the number of fault types;

[0049] Furthermore, in each hidden layer output process, the neuron output of the corresponding hidden layer is randomly deactivated, which means that when calculating the input from the first hidden layer to the second hidden layer, the neuron output of the first hidden layer after random deactivation is used as ; When calculating the input from the second hidden layer to the output layer, the output of the second hidden layer neurons after random inactivation is , because when calculating the gradient, it is necessary to consider that only some neurons participate in the forward propagation. Specifically, for the neurons that are randomly deactivated, their gradient is 0. Therefore, when calculating the weight gradient from the first hidden layer to the input layer, only the contribution of neurons that are not randomly deactivated in the current forward propagation is considered. By introducing the random deactivation operation technology in the neural network model of the multilayer perceptron, the output capacity of the model can be effectively improved, overfitting can be prevented, and more reliable results can be obtained in practical applications such as antenna fault detection.

[0050] Preferably, the output layer process is expressed as:

[0051] The output layer receives the output of the second hidden layer. The output formula of a neuron is: ,in The output layer The bias term of a neuron, ;

[0052] The number of neurons in the output layer indicates the type of output result. Here, the fault type is set to species, that is , no fault, short circuit of antenna feeder near antenna end, short circuit of antenna feeder near receiver end, then all output results of output layer are .

[0053] Furthermore, once the number of fault types is determined, the number of neurons in the output layer should match it. Three neurons are set in the output layer for three fault types. For the three fault types, the output values ​​of the three neurons are compared. The fault type corresponding to the neuron with the largest output value is the predicted fault type. The number of neurons in the output layer of the neural network can be flexibly adjusted according to actual needs to adapt to different numbers of fault types.

[0054] Preferably, the process of outputting the final fault detection result is:

[0055] Firstly, the first detection result is output, and when the first detection is detected as a fault, the second detection result is output;

[0056] Second test result setting state thresholds, ,when Greater than , Less than , Less than , it is judged as no fault. Less than , Greater than , Less than , it is judged that the antenna feeder is short-circuited near the antenna end. Less than , Less than , Greater than , it is judged that the antenna feed line is short-circuited near the receiver end.

[0057] Furthermore, when the first test result shows normal and the second test result also shows no fault, then the final result can be determined as normal; when the first test result shows a fault and the second test result shows a specific fault type (such as a short circuit in the antenna feed line near the antenna end), then the final result can be determined as the specific fault type shown by the second test result.

[0058] New energy vehicle GPS antenna fault detection system, including:

[0059] Information extraction module: used to collect antenna behavior data and obtain key behavior information by improving principal component analysis operations;

[0060] Behavior detection module: used to use the key behavior information as an input vector of a support vector machine model, and output a first detection result through the support vector machine model;

[0061] State detection module: used to collect antenna state data, use the antenna state data as an input layer of a neural network model based on a multi-layer perceptron, and output a second detection result;

[0062] Detection output module: used to combine the first detection result and the second detection result to output the final fault detection result.

[0063] The present invention has the following beneficial effects:

[0064] In the present invention, through multi-source data collection and analysis, the antenna status is comprehensively evaluated from multiple aspects such as signal strength and carrier-to-noise ratio, and multiple models are used for collaborative detection. In terms of fault detection, accurate fault location allows maintenance personnel to directly repair the problem area to avoid blind operation. The improved principal component analysis operation pre-processes the data and extracts key features, so that the extracted key behavior information is more focused on features closely related to the fault;

[0065] Based on the key behavior information after the principal component analysis operation, the information is input into the support vector machine model and the neural network model based on the multi-layer perceptron for collaborative detection. In the early stage of antenna fault detection, there are fewer fault samples, and the antenna fault data often presents nonlinear characteristics, such as the irregular changes in signal strength and carrier-to-noise ratio in the early stage of antenna feeder short circuit. By using the key behavior information extracted from the improved principal component analysis operation as the input vector of a support vector machine model, the fault situation can be quickly screened out, providing a preliminary direction for the subsequent direct acquisition of the fault location.

[0066] Then, the neural network model based on the multi-layer perceptron mines the complex nonlinear relationship in the state data. For short circuits at different positions of the antenna feeder (close to the antenna end or close to the receiver end), the multi-layer perceptron can learn the subtle differences in multiple data features of short circuits at different positions, thereby accurately distinguishing the fault type. By combining the two detection results, these differences can be accurately captured and distinguished, thereby significantly improving the accuracy of short circuit position judgment, avoiding the need to check one by one to obtain the specific location when a fault is found, and effectively solving the problem of accurately locating the short circuit fault location in the past. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a method step diagram of the new energy vehicle GPS antenna fault detection method proposed by the present invention.

[0068] Figure 2 This is a system block diagram of the new energy vehicle GPS antenna fault detection system proposed by the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] Embodiment 1, as Figure 1 As shown, the new energy vehicle GPS antenna fault detection method proposed by the present invention includes:

[0071] S1: used to collect antenna behavior data and obtain key behavior information by improving principal component analysis operations;

[0072] The steps to obtain key behavioral information by improving the principal component analysis operation are:

[0073] The antenna behavior data is formed into a raw data matrix , whose dimensions are , is the number of rows, Center each element in the matrix with the number of columns as the center: ,in, , , represents the element mean, Represents the centralized data matrix and introduces a series of weight vectors , calculate the weighted covariance matrix ,in, for The transpose of is a diagonal weight matrix, the diagonal elements are composed of a series of weight vectors Composition, perform eigenvalue decomposition on the weighted covariance matrix to obtain the key eigenvalues and the corresponding eigenvector , based on the key eigenvalues Get key contribution value , select before feature vectors, The eigenvectors form a principal component, and the centralized data matrix Projection to the front feature vector In the principal component space composed of ,in, ;

[0074] is a Matrix, each row represents the coordinates of a sample in the principal component space. These coordinates are the extracted key behavior information, expressed as .

[0075] Based on key feature values Get key contribution value The steps are:

[0076] According to the key characteristic value Sort the eigenvectors from large to small and obtain the top The key contribution value formula is: , determine a contribution threshold, when When the corresponding contribution threshold is reached, the corresponding feature vector data is .

[0077] Specifically, in reality, the working environment of the GPS antenna of new energy vehicles is complex, the amount of behavioral data is large and contains multiple interference factors. Traditional principal component analysis cannot effectively highlight important information. By introducing weight vectors to calculate the weighted covariance matrix, weights can be assigned according to the importance of data features. For example, when the signal strength data is stable but the phase noise fluctuates greatly and is more critical to fault judgment, the weight of the phase noise-related features can be increased, so that the extracted key behavioral information can be more focused on features closely related to the fault, providing a more accurate basis for subsequent fault detection.

[0078] S2: used to use the key behavior information as an input vector of a support vector machine model, and output a first detection result through the support vector machine model;

[0079] The steps for obtaining the first detection result are:

[0080] Build a support vector machine model. The construction process of the support vector machine model is:

[0081] Select a kernel function and train the model;

[0082] The key feature value and the corresponding eigenvector As model training data, key feature values The corresponding eigenvector As a set of training data, and select a kernel function;

[0083] The purpose of the model selecting the kernel function and solving the optimization problem is to find the optimal classification hyperplane so that the model can accurately classify the data;

[0084] First, minimize a cost function. The cost function consists of two parts: one is the penalty for misclassified points, and the other is the complexity of the hyperplane. The cost function is used to solve the optimization problem of the model, so as to find the optimal classification hyperplane, so that the model can accurately classify the data, which is expressed as: ,in, is a regularization term used to control the complexity of the model. is the normal vector of the hyperplane, which determines the direction of the hyperplane, and b is the bias term, which determines the position of the hyperplane. Represents the slack variable of the i-th group of training data, which is used to deal with noise or outliers in the data;

[0085] Then a constraint is used to ensure that the sample is correctly classified or as close to the correct classification as possible. The constraint is: ,in, represents the transpose of the hyperplane normal vector, Indicates Set training data, Indicates that the input data is the key behavior data, Represents the input data after being mapped by the kernel function;

[0086] Obtain the optimal solution through optimization problem and get the optimal hyperplane parameters. After satisfying the constraints, a trained support vector machine model is obtained.

[0087] After the training is completed, the key behavior information is used as the input vector of the support vector machine and the discrimination result is output: ,like If it is greater than or equal to 0, it is initially judged as no fault; if If it is less than 0, it is initially judged as a fault, and the fault type is further determined.

[0088] Among them, the kernel function selected after the key behavior information obtained by principal component analysis is used as the input vector of the support vector machine model can be a polynomial kernel function, and the formula is: ,in, is the scaling factor, c is a constant term, is the degree of the polynomial, and the polynomial kernel function can be adjusted by adjusting the parameters (the degree of the polynomial) to control the complexity of the decision boundary. When , the polynomial kernel function degenerates into a linear kernel function, which is suitable for linearly separable data; when It can handle nonlinear data and as As the number of nodes increases, the decision boundary becomes more complex. In antenna fault detection, short-circuit faults in different locations may require decision boundaries of different complexities to distinguish. For example, for short-circuit faults of the antenna feeder close to the antenna end and close to the receiver end, decision boundaries of different shapes may be required for accurate classification. The polynomial kernel function can adapt to this requirement by adjusting parameters.

[0089] S3: used to collect antenna status data, use the antenna status data as an input layer of a neural network model based on a multi-layer perceptron, and output a second detection result;

[0090] Antenna behavior data includes signal strength, carrier-to-noise ratio, phase noise, voltage and current at different positions of the antenna feeder. Suppose the number of collected data samples is , each sample type is .

[0091] Antenna status data includes antenna connection status data: by detecting whether the connection between the antenna and the receiver is firm, loose or disconnected, using a contact sensor or resistance measuring device to detect the resistance of the connection point. If the resistance is abnormally high, there may be a poor connection problem;

[0092] Antenna damage data: Check the appearance of the antenna to see if there is any physical damage, such as whether the antenna element is deformed or broken, or whether the antenna casing is damaged;

[0093] Antenna environment interference data: Use electromagnetic interference detection equipment to measure the electromagnetic environment around the antenna and obtain information such as the strength and frequency of the interference signal;

[0094] Signal strength value: Receive GPS satellite signals through the antenna and use a signal strength measurement device to obtain the signal strength value in decibels;

[0095] Phase noise data: Use phase noise measurement equipment to measure the phase noise of the signal received by the antenna. Phase noise will affect the demodulation and decoding process of the GPS signal, and thus affect the positioning accuracy. Suppose the antenna status data is .

[0096] The steps for obtaining the second test result are:

[0097] Build a neural network model based on multi-layer perceptron;

[0098] The neural network model based on multi-layer perceptron includes an output layer, two hidden layers and an output layer;

[0099] The input layer has neurons, To obtain the number of data points for state data, the first hidden layer has neurons, and the second layer has neurons, and the output layer has neurons;

[0100] The first hidden layer is represented as:

[0101] The antenna state data is input into the first hidden layer through the input layer. neurons, , the input process is expressed as: ,in, is the connection weight from the input layer to the first hidden layer, The first hidden layer The bias term of a neuron, For the The input data corresponding to each neuron is the corresponding antenna state data;

[0102] For the first hidden layer neurons, whose output is ,in, is the activation function;

[0103] Perform random dropout on the output of the first hidden layer;

[0104] The output of the neurons in the first hidden layer , with probability Randomly set its output to , for the neuron output Generate a Uniformly distributed random numbers ,when ,but ,in, Represents the output of the first hidden layer Output after random dropout operation;

[0105] Among them, by randomly deactivating each output of the first hidden layer, each neuron has a probability is set to ,have The probability remains the same, but is multiplied by , the purpose of this is to keep the expected value of the neuron output unchanged during training;

[0106] The second hidden layer process is expressed as:

[0107] The neurons in the second hidden layer receive the output from the first hidden layer after random inactivation, and further extract features. neurons, , the output process is expressed as: ,in, is the connection weight from the input layer to the second hidden layer, The second hidden layer The bias term of a neuron, Represents the output of the second hidden layer;

[0108] For the neuron output of the second hidden layer , the same random dropout operation is performed to obtain , and at the same time, for the connection weights from the second hidden layer to the output layer , which becomes ,in, is the output layer neuron index, which is determined by the number of fault types;

[0109] In each hidden layer output process, the neuron output of the corresponding hidden layer is randomly deactivated, which means that when calculating the input from the first hidden layer to the second hidden layer, the neuron output of the first hidden layer after random deactivation is used as ; When calculating the input from the second hidden layer to the output layer, the output of the second hidden layer neurons after random inactivation is , because when calculating the gradient, it is necessary to consider that only some neurons participate in the forward propagation. Specifically, for the neurons that are randomly deactivated, their gradient is 0. Therefore, when calculating the weight gradient from the first hidden layer to the input layer, only the contribution of neurons that are not randomly deactivated in the current forward propagation is considered. By introducing the random deactivation operation technology in the neural network model of the multilayer perceptron, the output capacity of the model can be effectively improved, overfitting can be prevented, and more reliable results can be obtained in practical applications such as antenna fault detection.

[0110] The output layer process is expressed as:

[0111] The output layer receives the output of the second hidden layer. The output formula of a neuron is: ,in The output layer The bias term of a neuron, ;

[0112] The number of neurons in the output layer indicates the type of output result. Here, the fault type is set to species, that is , no fault, short circuit of antenna feeder near antenna end, short circuit of antenna feeder near receiver end, then all output results of output layer are .

[0113] Among them, once the number of fault types is determined, the number of neurons in the output layer should match it. Three neurons are set in the output layer for three fault types. For the three fault types, the output values ​​of the three neurons are compared. The fault type corresponding to the neuron with the largest output value is the predicted fault type. The number of neurons in the output layer of the neural network can be flexibly adjusted according to actual needs to adapt to different numbers of fault types.

[0114] S4: used to combine the first detection result and the second detection result to output a final fault detection result.

[0115] The process of outputting the final fault detection results is:

[0116] Firstly, the first detection result is output, and when the first detection is detected as a fault, the second detection result is output;

[0117] Second test result setting state thresholds, ,when Greater than , Less than , Less than , it is judged as no fault. Less than , Greater than , Less than , it is judged that the antenna feeder is short-circuited near the antenna end. Less than , Less than , Greater than , it is judged that the antenna feed line is short-circuited near the receiver end.

[0118] Among them, when the first test result shows normal, and the second test result also shows no fault, then the final result can be determined as normal. When the first test result shows a fault, and the second test result shows a specific fault type (such as a short circuit of the antenna feeder near the antenna end), then the final result can be determined as the specific fault type shown by the second test result. The state threshold The threshold can be set based on the data from previous antenna fault detection;

[0119] Specifically, if it is found in long-term practice that, in the absence of faults, the average value (state threshold) of a parameter related to antenna performance is 0.8, then the average value of the parameter related to antenna performance can be set as , when the antenna feeder is short-circuited near the antenna end, the average value of the parameter related to the antenna performance becomes 1.2, then the average value of the parameter related to the antenna performance is set to , the average value of the parameters related to antenna performance when the antenna feeder is short-circuited near the receiver end is .

[0120] In this embodiment, the final detection result uses an improved principal component analysis to extract key behavior information, and then uses a support vector machine model to obtain a first detection result, and uses a multi-layer perceptron neural network model to obtain a second detection result, and the two are combined to determine the final fault. The support vector machine model has advantages in processing small samples and nonlinear data. The multi-layer perceptron neural network model can learn complex nonlinear relationships. The two methods complement each other. The support vector machine model is better at recognizing direct simple fault patterns. The support vector machine model is used to preliminarily determine whether there is a fault. The multi-layer perceptron neural network model is used to deal with more complex fault characteristics. The collaborative work can more accurately detect faults of different types and degrees, thereby improving the overall detection accuracy.

[0121] Embodiment 2, as Figure 2 As shown, based on the first embodiment, this embodiment proposes a new energy vehicle GPS antenna fault detection system, including:

[0122] Information extraction module: used to collect antenna behavior data and obtain key behavior information by improving principal component analysis operations;

[0123] Behavior detection module: used to use the key behavior information as an input vector of a support vector machine model, and output a first detection result through the support vector machine model;

[0124] State detection module: used to collect antenna state data, use the antenna state data as an input layer of a neural network model based on a multi-layer perceptron, and output a second detection result;

[0125] Detection output module: used to combine the first detection result and the second detection result to output the final fault detection result.

[0126] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting faults in a GPS antenna of a new energy vehicle, characterized in that: The steps include: S1: Collect antenna behavior data and obtain key behavior information by improving principal component analysis operations; The antenna behavior data includes signal strength, carrier-to-noise ratio, phase noise, voltage and current at different positions of the antenna feeder. Assume that the number of collected data samples is , each sample type is ; The steps of obtaining key behavior information by improving the principal component analysis operation are: The antenna behavior data is formed into a raw data matrix , whose dimensions are , centering each element in the matrix: ,in, , , represents the element mean, Represents the centralized data moment and introduces a series of weight vectors , calculate the weighted covariance matrix , ,in, for The transpose of is a diagonal weight matrix, the diagonal elements are composed of a series of weight vectors Composition, perform eigenvalue decomposition on the weighted covariance matrix to obtain the key eigenvalues and the corresponding eigenvector , based on key eigenvalues Get key contribution values , select before feature vectors, The eigenvectors form a principal component, and the centralized data matrix Projection to the front feature vector In the principal component space composed of ,in, ; is a Matrix, each row represents the coordinate of a sample in the principal component space, which is the extracted key behavior information, expressed as ; S2: using the key behavior information as an input vector of a support vector machine model, and outputting a first detection result through the support vector machine model; S3: Collect antenna status data, use the antenna status data as an input layer of a neural network model based on a multi-layer perceptron, and output a second detection result; The neural network model based on the multi-layer perceptron includes an output layer, two hidden layers and an output layer; S4: Combining the first detection result and the second detection result, outputting a final fault detection result; The number of neurons in the output layer indicates the type of output result. The fault type is set to species, that is , no fault, short circuit of antenna feeder near antenna end, short circuit of antenna feeder near receiver end, then all output results of output layer are ; The process of outputting the final fault detection result is: First, the first detection result is output, and when the first detection result is a fault, the second detection result is output as the final fault detection result; Second test result setting state thresholds, ,when Greater than , Less than and Less than , it is judged as no fault. Less than , Greater than and Less than , it is judged that the antenna feeder is short-circuited near the antenna end. Less than , Less than and Greater than , it is judged that the antenna feed line is short-circuited near the receiver end.

2. The method for detecting a fault in a new energy vehicle GPS antenna according to claim 1, characterized in that: Based on the key feature value Get key contribution values The steps are: According to the key characteristic value Sort the eigenvectors from large to small and obtain the top The key contribution value formula is: , determine a contribution threshold, when When the corresponding contribution threshold is reached, the corresponding feature vector data is .

3. The method for detecting faults of GPS antennas of new energy vehicles according to claim 1, characterized in that: The steps of obtaining the first detection result are: Build a support vector machine model; The construction process of the support vector machine model is: The key feature value and the corresponding eigenvector As model training data, key feature values The corresponding eigenvector As a set of training data, and select a kernel function; First, minimize a cost function, expressed as: ,in, is a regularization term used to control the complexity of the model. is the normal vector of the hyperplane, which determines the direction of the hyperplane, and b is the bias term, which determines the position of the hyperplane. Represents the slack variable of the i-th group of training data, which is used to deal with noise or outliers in the data; Then a constraint is passed to ensure that the sample is correctly classified: ,in, represents the transpose of the hyperplane normal vector, Indicates Set training data, Indicates that the input data is the key behavior data, Represents the input data after being mapped by the kernel function; Obtain the optimal solution through optimization problem and get the optimal hyperplane parameters. After satisfying the constraints, a trained support vector machine model is obtained. After the training is completed, the key behavior information is used as the input vector of the support vector machine and the discrimination result is output: ,like If it is greater than or equal to 0, it is initially judged as no fault; if If it is less than 0, it is initially judged as a fault, and the fault type is further determined.

4. The method for detecting a fault in a new energy vehicle GPS antenna according to claim 3, characterized in that: The kernel function selected in the support vector machine model training process is a polynomial kernel function, and the formula is: ,in, is the scaling factor, c is a constant term, is the degree of the polynomial.

5. The method for detecting faults of GPS antennas of new energy vehicles according to claim 1, characterized in that: The antenna status data includes antenna connection status data, antenna damage data, antenna environment interference data, signal strength value and phase noise data.

6. The method for detecting faults of a new energy vehicle GPS antenna according to claim 5, characterized in that: The steps of obtaining the second detection result are: Build a neural network model based on multi-layer perceptron; The input layer has neurons, To obtain the number of data points for state data, the first hidden layer has neurons, and the second hidden layer has neurons, and the output layer has neurons; The first hidden layer is expressed as: The antenna state data is input into the first hidden layer through the input layer. neurons, , the input process is expressed as: ,in, is the connection weight from the input layer to the first hidden layer, The first hidden layer The bias term of a neuron, For the The input data corresponding to each neuron is the antenna status data; For the first hidden layer neurons, whose output is ,in, is the activation function; Perform random dropout on the output of the first hidden layer; The output of each neuron in the first hidden layer , with probability Randomly set its output to , for each neuron output Generate a Uniformly distributed random numbers ,when ,but ,in, Represents the output of the first hidden layer Output after random dropout operation; The second hidden layer process is expressed as: The neurons in the second hidden layer receive the output from the first hidden layer after random inactivation, and further extract features. neurons, , the output process is expressed as: ,in, is the connection weight from the input layer to the second hidden layer, The second hidden layer The bias term of a neuron, Represents the output of the second hidden layer; right Perform random dropout to obtain For each neuron output of the second hidden layer, the random inactivation operation is also performed to obtain , and at the same time, for the connection weights from the second hidden layer to the output layer , which becomes ,in, is the output layer neuron index, which is determined by the number of fault types; The output layer process is expressed as: The output layer receives the output of the second hidden layer. The output formula of a neuron is: ,in The output layer The bias term of a neuron, .

7. A new energy vehicle GPS antenna fault detection system, which is implemented according to the new energy vehicle GPS antenna fault detection method according to any one of claims 1 to 6, characterized in that: include: Information extraction module: used to collect antenna behavior data and obtain key behavior information by improving principal component analysis operations; Behavior detection module: used to use the key behavior information as an input vector of a support vector machine model, and output a first detection result through the support vector machine model; State detection module: used to collect antenna state data, use the antenna state data as an input layer of a neural network model based on a multi-layer perceptron, and output a second detection result; Detection output module: used to combine the first detection result and the second detection result to output the final fault detection result.

Images