An IMU-based method and device for monitoring the operating state of an engineering vehicle

Abstract

The application discloses an IMU-based engineering vehicle running state monitoring method and device, and the method comprises the following steps: firstly, data is preprocessed, and then a plurality of characteristic values with high distinguishability are extracted for various vehicle types; characteristic value analysis is performed on the plurality of characteristic values, the characteristic values suitable for the vehicle types are automatically found, and machine learning classification is performed, so that the various vehicle types are recognized, and the robustness to environmental changes is achieved.The device comprises a data acquisition module, a man-machine interaction module, an algorithm module and a network transmission module.The method and the device can effectively solve the problems of few applicable vehicle types and low recognition rate.The application can count the working hours of engineering operation vehicles, so as to accurately perform wage settlement, can monitor driving time in real time, can judge whether a driver drives for a long time, and can guarantee transportation safety.The vehicle state provided by the application can assist in correcting vehicle positioning.

Description

Technical Field

[0001] This invention belongs to the field of vehicle networking, specifically relating to a method and device for monitoring the operating status of engineering vehicles based on IMU, which is applicable to positioning assistance, vehicle working hours statistics, and driver fatigue reminders. Background Technology

[0002] As the application of vehicle-to-everything (V2X) technology continues to deepen, cities are increasingly demanding intelligent management of engineering vehicles, which has raised new requirements for monitoring the operating status, routes, and working hours of these vehicles.

[0003] Engineering vehicle operating status identification can be achieved through two methods: acquiring vehicle speed via the CAN bus or acquiring vehicle vibration information via an IMU (Integrated Mutor Unit). Compared to the vehicle speed acquisition method, the IMU sensor method is unaffected by the vehicle terminal and can independently provide vehicle operating status information, while also offering advantages such as high real-time performance, high accuracy, and low cost. Currently, existing IMU-based engineering vehicle status monitoring methods identify status by extracting vibration signals and classifying time-domain and frequency-domain feature values. However, this method has poor adaptability to different vehicle types and environments. When vehicle types change or environments become more complex, the set feature values ​​often lose their discriminative power, leading to a decrease in recognition rate. Currently, there is no IMU-based vehicle status identification method or device that is robust to different vehicle types and environments, effectively addressing the problems of limited vehicle type recognition and low recognition rate in existing devices. Summary of the Invention

[0004] To address the issues of limited applicability and low recognition rate of engineering vehicle devices, this invention provides an IMU-based method and device for monitoring the operating status of engineering vehicles.

[0005] This invention first preprocesses the data, then extracts multiple highly distinguishable feature values ​​for various vehicle models. Feature value analysis is performed on these multiple feature values ​​to automatically find feature values ​​suitable for the specific vehicle model. Machine learning classification is then applied to achieve the goal of recognizing multiple vehicle models and ensuring robustness to environmental changes. The device includes a data acquisition module, a human-computer interaction module, an algorithm module, and a network transmission module. This method and device effectively solve problems such as a limited number of applicable vehicle models and low recognition rates. The device can also calculate the working hours of engineering vehicles for accurate wage settlement; monitor driving time in real time to determine if drivers are driving for extended periods, ensuring transportation safety; and the vehicle status information provided by the device can assist in correcting vehicle positioning.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A method for monitoring the operating status of engineering vehicles based on IMU includes the following steps:

[0008] Step 1) Build an SVM classifier model and train it using historical data. Use time-series cross-validation to verify the recognition effect. If the accuracy and precision of the model do not meet the set requirements after validation, continue training the model until they are met.

[0009] Step 2) Read the IMU data and perform preprocessing such as interpolation and filtering on the x, y, and z axis acceleration data collected in real time by the IMU.

[0010] Step 3) Determine if the current state is stationary based on the variance of the three-axis acceleration. If so, perform installation angle calibration; otherwise, wait for the stationary state to appear. Then, after calculating the installation angle, transform the equipment coordinate system to the vehicle coordinate system and store the calibrated installation angle.

[0011] Step 4) Extract multiple eigenvalues ​​from the z-axis acceleration data after coordinate system transformation in both the time and frequency domains to form a feature vector. The extracted feature vectors are then normalized using the Z-score normalization method, and principal component analysis is performed on the normalized feature vectors to extract suitable three-dimensional vehicle feature vectors.

[0012] Step 5) Use the trained SVM classifier to perform real-time vehicle state recognition and apply logical filtering to the real-time recognition results. Then, perform statistical analysis on the final recognition results and calculate the vehicle running time.

[0013] Step 6) Upload the status identification and statistical results to the remote terminal, and at the same time display the current total running time and continuous running time of the vehicle.

[0014] Execute steps 1)–6) to achieve IMU-based monitoring of the operating status of engineering vehicles, identify the vehicle's stopped, idle, and working states, and calculate the vehicle's working time.

[0015] Furthermore, in step 1), the specific steps for the time-cross sequence validation are as follows: First, the dataset should be divided into k consecutive subsets according to time order. To maintain the time order of the data, the data in each subset should be arranged consecutively. Then, the first k-1 subsets are used as the training set to train the model. The k-th subset is used as the test set to evaluate the model performance, and metrics such as accuracy and precision are used for evaluation.

[0016] In step 2), the sampling frequency of the IMU should be at least twice the vibration frequency of the vehicle body generated by the fastest engine speed of the vehicle being monitored. Simultaneously, the order of the low-pass filter used for data preprocessing needs to be determined based on the actual situation.

[0017] In step 3), the calibration only requires calculating the roll and pitch angles; the heading angle does not need to be calibrated. After calibration, the equipment needs to record the calibration results. Restarting the equipment does not require recalibration. During the calibration process, the vehicle must be on a level surface and stationary. If this requirement is not met, recalibration is required. In addition, if the equipment is installed in a different location, recalibration is required.

[0018] In step 4), the Z-score normalization method converts the feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1. Feature value normalization eliminates differences in dimensions and numerical ranges between different features, and principal component analysis is used to extract feature vectors with good classification performance and low correlation as input to the classifier.

[0019] An IMU-based monitoring device for the operating status of engineering vehicles, comprising the following modules:

[0020] The data acquisition module is used to collect vibration information of the vehicle; the human-machine interaction module is used to provide initial calibration notification and display recognition results when the device is first installed; the algorithm module performs preprocessing, time-domain and frequency-domain multi-feature extraction, principal component analysis, support vector machine (SVM) classification, and post-processing statistics of recognition status on the collected data to identify the vehicle status; the network transmission module is used to upload the vehicle status recognition information.

[0021] The main features of this invention are as follows: it provides a method and device for identifying the operating status of engineering vehicles. Compared with traditional monitoring methods that rely on vehicle system data, this invention uses an IMU to collect vehicle vibration information for monitoring, solving the problem of high dependence of existing equipment on vehicle CAN data. Through multi-feature extraction in the time and frequency domains and machine learning classification, it achieves accurate monitoring of the vehicle's operating status, including its standby state. By performing principal component analysis on multiple feature values ​​of the data, it extracts feature values ​​that have good classification effects for different vehicles and low correlation between feature values, and then establishes a classification model to achieve monitoring of multiple vehicle types. This method solves the problems of limited vehicle type recognition and low recognition rate of monitoring equipment, greatly improving the accuracy and practicality of vehicle operating status identification. Therefore, this invention provides an efficient, accurate, and convenient solution for real-time monitoring and efficient management of the operating status of engineering vehicles. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention.

[0023] Figure 2 This is a physical diagram of the device of the present invention.

[0024] Figure 3 This is an overall framework diagram of the device of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0026] refer to Figure 1 This example provides a flowchart of an IMU-based method for monitoring the operating status of engineering vehicles. The method includes the following steps:

[0027] Step 1) Build an SVM classifier model. Train the SVM model using historical data, collect vehicle state data for different states, and simultaneously label the data and input it into the classifier for training. Evaluate the classifier's prediction results in real time. Evaluation metrics include accuracy, precision, and recall. These metrics are used to determine whether the model needs further training and to finalize the model parameters.

[0028] Step 2) Read the raw IMU data and preprocess the raw data to remove noise, missing values ​​and outliers.

[0029] Step 3) Read the data stored in the device to determine if calibration is required. If so, perform angle calibration; otherwise, read the calibration angle from the data stored in the device. Transform the collected three-axis data from the device coordinate system to the vehicle coordinate system.

[0030] Step 4) Extract feature values ​​from each window of the z-axis accelerometer data after coordinate system transformation. These features include time-domain features and frequency-domain features, where the time-domain features include gradient F′ and gradient variance σ. 2 In addition to correlation coefficients such as r, frequency domain features include fundamental frequency f0, spectrum signal-to-noise ratio S / N, and high-low frequency energy ratio H / L. All feature values ​​are normalized, and then principal component analysis is performed on the feature vector composed of multiple feature values ​​to extract feature values ​​suitable for the vehicle.

[0031] Step 5) Use the trained SVM classifier to classify the new data, and perform logical filtering on the recognition results. Then, perform recognition status statistics and calculate the total vehicle running time and continuous running duration.

[0032] Execute steps 1)–5) to achieve IMU-based monitoring of the operating status of engineering vehicles, identify the vehicle's stopped, idle, and working states, and calculate the vehicle's running time.

[0033] The overall process flow chart for the equipment is explained in detail below:

[0034] Step 1): Train the SVM classifier model. An initial n×n kernel matrix K is built based on historical samples. Then, a new sample z is added to the kernel matrix to calculate the new kernel matrix. The following are the SVM training and update steps:

[0035] First, calculate the Gaussian RBF kernel function values ​​between the new sample z and the original n samples to obtain a 1×n vector K. z :

[0036]

[0037] Among them, K z (i) represents the Gaussian RBF kernel function value between z and the i-th sample, ||.|| is the second norm of the vector, γ is the Gaussian kernel function parameter, and x i Let z be the i-th sample. Then calculate the kernel function value of the new sample z with respect to itself, and obtain a scalar K. zz :

[0038]

[0039] Second, K z and K zz Adding this to the original kernel matrix K results in an updated kernel matrix K′ of (n+1)×(n+1). K represents z The transpose of .

[0040]

[0041] After calculating the new kernel matrix, update the Lagrange multipliers using the following formula:

[0042]

[0043] α i (t) represents the value of the i-th Lagrange multiplier in the t-th iteration. η is the learning rate, which controls the update magnitude of the Lagrange multipliers in each iteration, η > 0. i and y j These are the labels of the i-th and j-th samples, respectively. K(x) i ,x j ) is the kernel function value between the i-th sample and the j-th sample. n is the number of training samples.

[0044] Based on the updated Lagrange multipliers, new support vectors can be determined, where the support vectors are the data points in the training samples that are closest to the classification boundary. Let the selected support vector be x. s Its corresponding category label is y s The formula for calculating the intercept term b is as follows:

[0045]

[0046] Where SV is the sample set, α i For Lagrange multipliers, xi For the i-th sample, y i It is the label of the i-th sample, (x i ·x s ) represents x i and x s The inner product of the two. The final decision function of the SVM is:

[0047] f(x)=∑α i ×y i ×K(x i ,x)-b (6)

[0048] Among them, y i Here, x represents the sample class label, K is the selected kernel function, and x represents the sample class label. i Here, x is the training sample, b is the new unlabeled sample, and b is the intercept term of the decision function. The model's recognition rate is calculated using the test set, and training ends when the requirements are met. The current model parameters are then stored.

[0049] Step 2): The acquisition module collects the raw acceleration data a along the X, Y, and Z axes. x a y a z The data is fed into the data processing module, which includes missing value handling, outlier handling, and noise processing. Missing value handling uses interpolation to estimate the missing value using nearby known data points; outlier handling uses the interquartile range method; and noise processing uses low-pass filtering, with a sampling frequency of f. s The cutoff frequency is f c The input is x i The output is y i The output at the previous time step was y. i-1 The filtering formula is as follows:

[0050]

[0051] Where α is the low-pass filter coefficient and π is pi.

[0052] Step 3): Use the calibration module to calibrate the installation angle, and use the processed triaxial acceleration values ​​to calculate the roll angle (θ). pitch ) and pitch angle (θ) roll The calculation formula is as follows:

[0053] θ pitch =arctan(a x / a z (8)

[0054] θ roll =arctan(-a y / a z (9)

[0055] Among them, the heading angle (θ) yaw The angle is set to the default value of 0°. A state transition matrix is ​​generated based on the installation angle. The device coordinate system is projected onto the vehicle coordinate system. The three-axis acceleration values ​​in the vehicle coordinate system are set to... The triaxial acceleration value in the device coordinate system is a x a y a z The conversion formula is as follows:

[0056]

[0057] Step 4): Convert the... As vehicle vibration data, a certain window of data is selected, and the vibration data feature values ​​of that window are extracted. Let... Then the sample at the current time is F n The sampled value at the previous time step was F. n-1 Calculate the temporal feature gradient F′ and gradient variance σ of the data. 2 :

[0058] F′=F n -F n-1 (11)

[0059]

[0060] Where N is the window length. The data within the window is divided into t parts. Using the first part F1 as reference data, the values ​​of the first part F1 and other parts F1 are compared. i correlation coefficient r i Find the maximum value as the correlation coefficient r of the data in that window. Let the window data be F(n) and the sub-window data be Fi. i (N / t), where N and t are chosen based on the actual situation, and the formula is as follows:

[0061]

[0062]

[0063] After calculating the time-domain eigenvalues, the frequency-domain eigenvalues ​​are calculated. First, the time-domain signal is converted to a frequency-domain signal using a Fast Fourier Transform (FFT). Then, the spectrum is converted to an energy spectrum, and the maximum value of the spectrum is the fundamental frequency f0. Next, the spectral signal-to-noise ratio is determined. Let the spectral function be S(f), where f is the frequency and the sampling frequency is f0. s If the signal-to-noise ratio is S / N, the calculation formula is as follows:

[0064]

[0065] Let the high-frequency energy ratio be H / L, and the calculation formula is as follows:

[0066]

[0067] Where S(f) is the spectrum function, f is the frequency, f0 is the fundamental frequency, and f s The sampling frequency.

[0068] After all eigenvalues ​​are calculated, they are normalized using the Z-score normalization method. Let x be the original eigenvector, σ be the standard deviation of the eigenvector, and n be the number of eigenvalues. The calculation formula is as follows:

[0069]

[0070] Principal component analysis is performed on the normalized feature vectors to calculate the covariance matrix composed of each feature vector. The three feature vectors with the largest eigenvalues ​​in the covariance matrix are then used as the input feature vectors for the classifier.

[0071] Step 5): The vehicle status recognition module uses the extracted feature vectors and the trained model to recognize the vehicle status. Then, it performs post-processing on the recognized status, calculates vehicle runtime based on the identified vehicle status, and performs work hour statistics. Simultaneously, the vehicle position should remain unchanged when not in motion, and assisted vehicle positioning is performed based on the current vehicle status. Furthermore, based on the vehicle's continuous operating time, it determines whether the vehicle has been running continuously for an extended period and provides driver fatigue alerts.

[0072] Accordingly, this example also provides an IMU-based engineering vehicle operation status monitoring device for executing the above methods. (Reference) Figure 2 , Figure 3 The physical diagrams and various modules of the device in this example are described below:

[0073] Please refer to Figure 2 , Figure 2 This is a picture of the actual device. The device is small in size, connects to the vehicle body wirelessly, is easy to use, and is convenient to install and remove.

[0074] Please refer to Figure 3 , Figure 3 This is a schematic diagram of the device modules. As shown in the figure, the engineering vehicle operation status monitoring device includes: a data acquisition module for collecting vehicle vibration data; a human-machine interaction module for equipment calibration and equipment operation status prompts; an algorithm module for processing raw data and identifying vehicle status; a network transmission module for reporting statistical results and raw data; and a computer network receiving module for observing and statistically analyzing real-time data and real-time status.

[0075] Specifically, this device needs to be fixed to the vehicle body during use. For the initial installation, a calibration notification must be sent using the human-machine interface module. During calibration, the vehicle must be stationary on a level surface. The completion of calibration can be determined by the indicator lights in the human-machine interface module.

[0076] For example, the working method of the engineering vehicle operation status monitoring device is as follows: First, the device collects vehicle vibration information through the data acquisition module; second, the algorithm module performs calibration and status identification based on the collected data, and performs statistical analysis of the identification results; then, the raw data and identification results are sent to the network platform through the network transmission module; finally, the vehicle status and running time can be observed in real time using the computer network receiving module.

[0077] By executing the above methods and steps 1)–5), the operational status monitoring of engineering vehicles can be achieved.

Claims

1. A method for monitoring the operating status of engineering vehicles based on an IMU, comprising the following steps: Step 1) Build an SVM classifier model and train it using historical data; use time series cross-validation to verify the recognition effect. If the accuracy and precision of the model do not meet the set requirements after verification, continue training the model until they are met. Step 2) Read the IMU data and perform interpolation and filtering preprocessing on the x, y, and z axis acceleration data acquired in real time by the IMU; Step 3) Determine whether the current state is stationary based on the variance of the three-axis acceleration. If so, perform installation angle calibration; otherwise, wait for the stationary state to appear. Then, after calculating the installation angle, transform the equipment coordinate system to the vehicle coordinate system and store the calibrated installation angle. Step 4) Extract multiple eigenvalues ​​from the z-axis acceleration data after coordinate system transformation in both the time and frequency domains to form a feature vector; normalize the extracted feature vectors using the Z-score normalization method, and perform principal component analysis on the normalized feature vectors to extract the vehicle's three-dimensional feature vectors; specifically, this includes: transforming the z-axis acceleration data... As vehicle vibration data, a certain window of data is selected, and the vibration data feature values ​​of that window are extracted. Let... Then the current sampling time is The previous sampling time was Calculate the time-domain feature gradient of the data and gradient variance : (11) (12) in, N Define the window length; divide the data within the window into... t Use the first set of data. Using this as reference data, calculate the first set of data. Other data correlation coefficient The maximum value is taken as the correlation coefficient value of the data in that window. Let the window data be... The data in the sub-window is ,in N and t The value is determined based on the actual situation, and the formula is as follows: (13) (14) After calculating the time-domain eigenvalues, the frequency-domain eigenvalues ​​are calculated. First, the time-domain signal is converted into a frequency-domain signal using a Fast Fourier Transform (FFT). Then, the spectrum is converted into an energy spectrum, and the maximum value of the spectrum is the fundamental frequency. Next, the spectral signal-to-noise ratio is calculated, assuming the spectral function is... , The frequency is [frequency], and the sampling frequency is [frequency]. The signal-to-noise ratio is The calculation formula is as follows: (15) Let the energy ratio of high frequency to low frequency be . The calculation formula is as follows: (16) in For the spectrum function, For frequency, For the fundamental frequency, The sampling frequency; After all eigenvalues ​​have been calculated, they are normalized using the Z-score normalization method. The original feature vector, It is the standard deviation of the eigenvectors. The number of eigenvalues ​​in the eigenvector is given by the following formula: (17) Principal component analysis is performed on the normalized feature vectors to calculate the covariance matrix composed of each feature vector. The three feature vectors with the largest eigenvalues ​​of the covariance matrix are then used as the input feature vectors of the classifier. Step 5) Use the trained SVM classifier to perform real-time vehicle status recognition, and perform logical filtering on the real-time recognition results; then, perform statistics on the final recognition results and calculate the vehicle running time. Step 6) Upload the status identification and statistical results to the remote terminal, and at the same time display the current total running time and continuous running time of the vehicle.

2. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 1) The time series cross-validation method includes: first, dividing the dataset into k consecutive subsets according to time order; the data in each subset should be arranged consecutively; then, using the first k-1 subsets as the training set to train the model; using the kth subset as the test set to evaluate the model performance, and simultaneously using accuracy and precision metrics for evaluation.

3. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 1) Training the SVM classifier model includes: building a model based on historical samples. Initialization kernel matrix Furthermore, using a new sample This is added to the kernel matrix to calculate a new kernel matrix; specifically, this includes: First, calculate the new sample. Compared to the original The Gaussian RBF kernel function values ​​between samples yield a vector : (1) in, for With the i Gaussian RBF kernel function values ​​between samples Let be the second norm of the vector. For the parameters of the Gaussian kernel function, For the first A sample is generated; then the kernel function value of the new sample z with itself is calculated to obtain a scalar. : (2) Second, and Add to the original kernel matrix In the middle, we get one Updated kernel matrix , express Transpose of; (3) After calculating the new kernel matrix, update the Lagrange multipliers using the following formula: (4) Indicates the first The Lagrange multiplier in the th The value at the next iteration; It is the learning rate, which controls the update magnitude of the Lagrange multipliers in each iteration. ; and They are the first and the The labels of each sample; It is the first The first sample and the first Kernel function values ​​between samples; It is the number of training samples; Based on the updated Lagrange multipliers, new support vectors can be determined, where the support vectors are the data points in the training samples that are closest to the classification boundary; let the selected support vectors be... Its corresponding category label is Intercept term The calculation formula is as follows: (5) in, For the sample set, For Lagrange multipliers, For the first One sample, It is the first The label of each sample, express and The inner product of the two; the final decision function of the SVM is: (6) in, These are sample category labels. It is the selected kernel function. These are training samples. It is a new, unlabeled sample. It is the intercept term of the decision function; the recognition rate of the model is calculated through the test set, and the model training ends when the requirements are met, and the current model parameters are stored.

4. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 2) The sampling frequency of the IMU is more than twice the vibration frequency of the vehicle body generated by the fastest speed of the engine of the vehicle to be monitored; at the same time, the order of the low-pass filter used for data preprocessing needs to be determined according to the actual situation.

5. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 2) includes: processing the collected data... X , Y and Z Raw acceleration data of the shaft , , Missing value processing, outlier processing, and noise reduction are performed. Missing value processing uses interpolation, estimating the missing value using nearby known data points. Outlier processing uses the interquartile range method. Noise reduction is achieved using low-pass filtering, with a sampling frequency of [value missing]. The cutoff frequency is The input is The output is The output at the previous time step was The filtering formula is as follows: (7) in These are the low-pass filter coefficients. Pi is the mathematical constant of a circle.

6. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 3) describes the calibration process, which calculates the roll and pitch angles but does not calibrate the heading angle. After calibration, the equipment records the calibration results. Restarting the equipment does not require recalibration. During the calibration process, the vehicle must be on a level surface and stationary. If this requirement is not met, recalibration is required. If the equipment is installed in a different location, recalibration is required.

7. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 3) specifically includes: using the processed triaxial acceleration values ​​to determine the roll angle during installation angle calibration. and pitch angle The calculation formula is as follows: (8) (9) Among them, heading angle Set as default value The state transition matrix is ​​generated by the installation angle, and the equipment coordinate system is projected onto the vehicle coordinate system. Let the three-axis acceleration values ​​in the vehicle coordinate system be... , , The three-axis acceleration values ​​in the equipment coordinate system are , , The conversion formula is as follows: (10)。 8. The method for monitoring the operating status of engineering vehicles based on IMU as described in claim 1, characterized in that: Step 4) describes the Z-score normalization method, which converts the feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1. By normalizing the feature values, the differences in the dimensions and numerical ranges between different features are eliminated. Principal component analysis is used to extract feature vectors with good classification performance and low correlation as input to the classifier.

9. A monitoring device for the operating status of engineering vehicles based on an IMU, characterized in that: include: The data acquisition module is used to collect vibration information from the vehicle. The human-computer interaction module is used to provide initial calibration notifications and display recognition results during the initial installation of the device. The algorithm module performs preprocessing, time-domain and frequency-domain multi-feature extraction, principal component analysis, support vector machine (SVM) classification, and post-processing statistics on the recognition results of the collected data to identify the vehicle status; specifically, it includes: converting the data... As vehicle vibration data, a certain window of data is selected, and the vibration data feature values ​​of that window are extracted. Let... Then the current sampling time is The previous sampling time was Calculate the time-domain feature gradient of the data and gradient variance : (11) (12) in, N Define the window length; divide the data within the window into... t Use the first set of data. Using this as reference data, calculate the first set of data. Other data correlation coefficient The maximum value is taken as the correlation coefficient value of the data in that window. Let the window data be... The data in the sub-window is ,in N and t The value is determined based on the actual situation, and the formula is as follows: (13) (14) After calculating the time-domain eigenvalues, the frequency-domain eigenvalues ​​are calculated. First, the time-domain signal is converted into a frequency-domain signal using a Fast Fourier Transform (FFT). Then, the spectrum is converted into an energy spectrum, and the maximum value of the spectrum is the fundamental frequency. Next, the spectral signal-to-noise ratio is calculated, assuming the spectral function is... , The frequency is [frequency], and the sampling frequency is [frequency]. The signal-to-noise ratio is The calculation formula is as follows: (15) Let the energy ratio of high frequency to low frequency be . The calculation formula is as follows: (16) in For the spectrum function, For frequency, For the fundamental frequency, The sampling frequency; After all eigenvalues ​​have been calculated, they are normalized using the Z-score normalization method. The original feature vector, It is the standard deviation of the eigenvectors. The number of eigenvalues ​​in the eigenvector is given by the following formula: (17) Principal component analysis is performed on the normalized feature vectors to calculate the covariance matrix composed of each feature vector. The three feature vectors with the largest eigenvalues ​​of the covariance matrix are then used as the input feature vectors of the classifier. The network transmission module is used to upload vehicle status identification information.

Images