Automobile driving condition multi-scale identification method

Abstract

The invention relates to an automobile driving condition multi-scale identification method. The method comprises the following steps: S1, acquiring and preprocessing driving condition data; s2, short-term working condition sample division and long-term working condition sample division are carried out, and long-term working condition categories are divided according to long-term working condition samples; s3, performing characteristic parameter calculation on the divided working condition data to obtain characteristic parameter data of each section of working condition sample; s4, performing principal component analysis and dimension reduction; s5, performing short-term working condition clustering to obtain a short-term working condition category; and S6, performing model training based on the feature parameter data after dimension reduction, the short-term working condition category and the long-term working condition category to obtain a multi-time-scale working condition recognition model, acquiring actual working condition data by the multi-time-scale working condition recognition model, and outputting a recognition result. Compared with the prior art, the method has the advantages of improving the accuracy of identifying the vehicle driving working condition, especially the sudden change working condition, and the like.

Description

Technical Field

[0001] The invention relates to the field of automobile technology, and in particular to a multi-scale recognition method for automobile driving conditions. Background Art

[0002] Hybrid vehicles have attracted widespread attention due to their high fuel efficiency and low emissions. In order to further improve the energy efficiency and driving experience of hybrid vehicles, accurate identification and classification of vehicle driving conditions are crucial. Existing hybrid vehicle operating condition identification methods are mostly based on a single operating condition classification strategy, which often cannot effectively cope with complex and changeable actual driving conditions. In a dynamically changing driving environment, the vehicle needs to adjust the working mode of its power system according to different operating conditions to achieve optimal power performance and fuel economy. To solve this problem, some researchers have proposed some operating condition classification methods. However, these methods still have limitations in dealing with sudden changes in operating conditions and complex driving modes, making it difficult to achieve accurate operating condition identification. In addition, existing operating condition identification often lacks flexibility in dealing with sudden changes in operating conditions. Summary of the invention

[0003] The purpose of the present invention is to provide a multi-scale identification method for vehicle driving conditions in order to improve the accuracy of identifying vehicle driving conditions, especially sudden change conditions.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] The present invention proposes a multi-scale identification method for vehicle driving conditions, the method comprising the following steps:

[0006] S1. Acquire and preprocess driving condition data to obtain preprocessed condition data and long-term condition categories;

[0007] S2. The pre-processed working condition data is divided into short-term working condition samples and long-term working condition samples, and the long-term working condition categories are divided according to the long-term working condition samples, which correspond one to one with the long-term working condition samples;

[0008] S3, calculating characteristic parameters of the divided working condition data to obtain characteristic parameter data of each working condition sample;

[0009] S4. Principal component analysis and dimensionality reduction are performed on the characteristic parameter data of the working condition samples;

[0010] S5, clustering the characteristic parameter data after the principal component dimension reduction into short-term operating condition categories to obtain short-term operating condition categories;

[0011] S6. Perform model training based on the feature parameter data after dimensionality reduction, the short-term operating condition category and the long-term operating condition category to obtain a multi-time scale operating condition recognition model, wherein the multi-time scale operating condition recognition model obtains actual operating condition data and outputs a recognition result.

[0012] Furthermore, the preprocessing is specifically as follows:

[0013] A1. Check whether the amount of driving condition data reaches a threshold. If not, expand the driving condition data based on loop reconstruction or data enhancement technology and return to A1. Otherwise, execute A2.

[0014] A2. Check whether the driving condition data obtained in A1 has missing values ​​and abnormal values. If yes, perform missing value processing and abnormal value processing, and then execute A3. Otherwise, directly execute A3.

[0015] A3, identifying the long-term operating condition category of the driving operating condition data obtained in A2, and whether there are continuous zero values ​​above the quantity threshold. If so, remove the continuous zero values ​​and mark the long-term parking long-term operating condition in the driving operating condition data to obtain the pre-processed operating condition data, and then execute A4. Otherwise, obtain the pre-processed operating condition data and directly execute A4;

[0016] A4. Calculate the acceleration from the pre-processed working condition data.

[0017] Furthermore, the specific steps of dividing the short-term working condition samples and the long-term working condition samples in S2 are:

[0018] B1. Select the 1st to Tth data points in the preprocessed working condition data to form the first section of working condition samples;

[0019] B2, with an interval of ΔT data points, select the (1+ΔT)th to (1+ΔT+T)th data points of the preprocessed operating condition data to form the second section of the operating condition sample;

[0020] B3. Repeat the above steps;

[0021] B4. Select the (1+kΔT)th to (1+kΔT+T)th data points of the preprocessed operating condition data to form the last section of the operating condition sample (k∈Z, 1+kΔT+T≤N<1+kΔT+ΔT+T), and repeat the above steps with different operating condition sample lengths T to obtain short-term operating condition sample division and long-term operating condition sample division.

[0022] Furthermore, the specific steps of S3 are:

[0023] D1. Calculate the maximum speed from the divided working condition data where v i Indicates the speed of each working condition sample;

[0024] D2. Calculate the average speed from the divided working condition data

[0025] D3. Calculate the maximum acceleration from the divided working condition data Among them, a i Indicates the acceleration of each working condition sample;

[0026] D4. Calculate the maximum deceleration from the divided working condition data

[0027] D5. Calculate the average acceleration from the divided working condition data

[0028] D6. Calculate the average deceleration from the divided working condition data

[0029] D7. Calculate the acceleration time ratio based on the divided working condition data in

[0030] D8. Calculate the deceleration time ratio based on the divided working condition data

[0031] D9. Calculate the uniform speed time ratio from the divided working condition data

[0032] D10. Calculate the idle time ratio based on the divided working condition data The above parameters constitute the characteristic parameter data.

[0033] Furthermore, the specific steps of S4 are:

[0034] The following operations are performed on the characteristic parameter data corresponding to the short-term working condition samples and the long-term working condition samples:

[0035] E1. The characteristic parameter data are organized into a corresponding characteristic parameter matrix X, where the number of rows of X is the number of working condition samples and the number of columns is the number of features;

[0036] E2. Standardize the feature parameter matrix X, subtract the mean of each feature parameter, and then divide it by its standard deviation to obtain the standardized feature parameter matrix X std ;

[0037] E3, after the standardized feature parameter matrix X std Calculate its covariance matrix;

[0038] E4. Solve the eigenvalues ​​and eigenvectors of the covariance matrix;

[0039] E5. Calculate the principal component contribution rate and cumulative contribution rate;

[0040] E6. Select the eigenvector a corresponding to the first k largest eigenvalues ​​whose cumulative contribution rate is greater than 95%. 1 ,a 2 ,…,a k as the principal component;

[0041] E7, the principal component a 1 ,a 2 ,…,a k The composite projection matrix P = [a 1 ,a 2 ,…,a k ];

[0042] E8. Based on the projection matrix, the feature parameter matrix X std Projected into the new space defined by the principal component, the data Y=X after the principal component dimension reduction is obtained std P.

[0043] Furthermore, the covariance matrix is:

[0044]

[0045] Among them, ∑ is the covariance matrix and n is the number of working condition samples.

[0046] Furthermore, the eigenvalues ​​and eigenvectors of the covariance matrix are solved as follows:

[0047] det(∑-λI)=0

[0048] (∑-intoI)a=0

[0049] Where λ is the eigenvalue, I is the identity matrix, and a is the eigenvector.

[0050] Furthermore, the principal component contribution rate and the cumulative contribution rate are:

[0051]

[0052] Among them, d i is the main component contribution rate, r i is the cumulative contribution rate.

[0053] Furthermore, the specific steps of S5 are:

[0054] K-MEANS clustering is performed on the characteristic parameter data after principal component dimensionality reduction corresponding to the short-term operating condition samples to obtain the short-term operating condition category.

[0055] Furthermore, the steps of S6 are:

[0056] G1. Use the reduced dimension feature parameter data as features, short-term working condition category and long-term working condition category as labels, and divide them into validation set, training set and test set;

[0057] G2, select the multi-layer perceptron neural network as the main framework of the multi-time scale working condition recognition model, select the cross entropy loss function as the loss function, select the Adam optimizer as the optimizer, use the data of the training set to train the model, and observe the loss and accuracy of the model on the validation set at the same time. The training process ends when the loss of two consecutive validation sets is lower than the threshold;

[0058] G3. Use the test set data to test the performance of the trained multi-time scale working condition recognition model;

[0059] G4. Adjust the training parameters and repeat the above steps until the final multi-time scale working condition recognition model is obtained. Based on the final multi-time scale working condition recognition model, the actual working condition data is obtained and the recognition result is output.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The present invention introduces multi-time scale analysis and recognition, which can simultaneously identify short-term and long-term operating conditions. It can identify six short-term operating conditions, namely idling, constant speed, acceleration, deceleration, emergency stop and emergency start, as well as three long-term operating conditions, namely urban driving conditions, suburban driving conditions and high-speed driving conditions. This helps to improve the recognition accuracy of vehicle driving conditions and the robustness of condition recognition, and thus helps to predict future vehicle failures and abnormalities, optimize energy distribution and improve vehicle energy utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a flow chart of the present invention;

[0063] Figure 2 is a working condition curve of a specific embodiment of the present invention;

[0064] Figure 3 Schematic diagram of the division of working condition samples;

[0065] Figure 4 is a working condition category curve of a specific application scenario of the present invention;

[0066] Figure 5 A loss graph of a model training process for a specific application scenario of the present invention;

[0067] Figure 6 This is a graph of the accuracy of the model training process for a specific application scenario of the present invention. DETAILED DESCRIPTION

[0068] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0069] The present invention proposes a multi-scale identification method for vehicle driving conditions, the flow chart is as follows: Figure 1 As shown, the method comprises the following steps:

[0070] S1. Acquire and preprocess driving condition data to obtain preprocessed condition data and long-term condition categories;

[0071] S2. The pre-processed working condition data is divided into short-term working condition samples and long-term working condition samples, and the long-term working condition categories are divided according to the long-term working condition samples, which correspond one to one with the long-term working condition samples;

[0072] S3, calculating characteristic parameters of the divided working condition data to obtain characteristic parameter data of each working condition sample;

[0073] S4. Principal component analysis and dimensionality reduction are performed on the characteristic parameter data of the working condition samples;

[0074] S5, clustering the characteristic parameter data after the principal component dimension reduction into short-term operating condition categories to obtain short-term operating condition categories;

[0075] S6. Perform model training based on the feature parameter data after dimensionality reduction, the short-term operating condition category and the long-term operating condition category to obtain a multi-time scale operating condition recognition model, wherein the multi-time scale operating condition recognition model obtains actual operating condition data and outputs a recognition result.

[0076] The pre-processing process described in S1 is as follows:

[0077] A1. Check whether the working condition data samples are sufficient. If not, consider using cyclic reconstruction or data enhancement technology to expand the working condition samples;

[0078] A2. Check whether there are missing values ​​or abnormal values ​​in the working condition data. If there are, process the missing values ​​and abnormal values ​​of the working condition data;

[0079] A3. Check whether there are more than 200 consecutive zero values ​​in the operating condition data. If so, remove the consecutive zero values ​​and mark them as long-term parking and long-term operating conditions in the original sample;

[0080] A4. Calculate the acceleration from the working condition sample data, where the acceleration calculation formula is as follows:

[0081]

[0082] In the formula, a iis the acceleration in meters per second (m / s), i is the sampling time in seconds (s), v is the vehicle speed in kilometers per hour (km / h), and T is the working condition sample length in seconds (s).

[0083] The process of dividing the working condition samples with N data points described in S2 is:

[0084] B1. Select the 1st to Tth data points of the working condition data to form the first section of working condition samples;

[0085] B2, with an interval of ΔT data points, select the (1+ΔT)th to (1+ΔT+T)th data points of the operating condition data to form the second section of the operating condition sample;

[0086] B3. Repeat the above steps;

[0087] B4. Select the (1+kΔT)th to (1+kΔT+T)th data points of the operating condition data to form the last section of the operating condition sample (k∈Z, 1+kΔT+T≤N<1+kΔT+ΔT+T).

[0088] In the process of classifying long-term working conditions as described in S2, working condition switching may occur. At this time, the long-term working condition category is determined according to the working condition time ratio, and the working condition with a working condition time ratio greater than or equal to 50% is taken as the long-term working condition corresponding to the long-term working condition sample. The specific process is:

[0089] C1. Classify long-term working conditions according to the steps of working condition sample classification;

[0090] C2. Calculate the average of each sample and round it off. The resulting value is the long-term operating condition corresponding to the sample.

[0091] The process of calculating characteristic parameters of the working condition sample data divided by S2 described in S3 is:

[0092] D1. Calculate the maximum speed from the working condition sample data

[0093] D2. Calculate the average speed from the working condition sample data

[0094] D3. Calculate the maximum acceleration from the working condition sample data

[0095] D4. Calculate the maximum deceleration from the working condition sample data

[0096] D5. Calculate the average acceleration from the working condition sample data

[0097] D6. Calculate the average deceleration from the working condition sample data

[0098] D7. Acceleration time ratio calculated from working condition sample data in

[0099] D8. Calculate the deceleration time ratio based on the working condition sample data

[0100] D9. Calculate the uniform speed time ratio from the working condition sample data

[0101] D10, Calculate the idle time ratio based on the working condition sample data

[0102] The process of performing principal component analysis and dimensionality reduction on the characteristic parameter data of the working condition samples calculated by S3 described in S4 is:

[0103] E1. The characteristic parameter data of each working condition sample calculated by S3 is combined into a corresponding characteristic parameter matrix The number of rows of X is the number of working condition samples, and the number of columns is the number of features;

[0104] E2. Standardize the feature parameter matrix X, subtract the mean of each feature, and then divide it by its standard deviation to avoid the impact of different dimensions between different features on model training, and obtain X std , the standardized formula is as follows:

[0105]

[0106] Where, X std represents the characteristic parameter matrix after standardization, X represents the characteristic parameter matrix before processing, μ represents the characteristic mean, and σ represents the characteristic standard deviation;

[0107] E3, after the standardized feature parameter matrix X std Calculate its covariance matrix, the calculation formula is as follows:

[0108]

[0109] Where, Σ is the covariance matrix, n is the number of working condition samples;

[0110] E4. Solve the eigenvalues ​​and eigenvectors of the covariance matrix. The solution formula is as follows:

[0111] det(∑-λI)=0

[0112] (∑—λI)a=0

[0113] Where λ is the eigenvalue, I is the identity matrix, and a is the eigenvector.

[0114] The eigenvalue λ 1 ≥λ 2 ≥…λ 10 ≥0, the eigenvector is a 1 ,a 2 ,…,a 10 ;

[0115] E5. Calculate the principal component contribution rate And the cumulative contribution rate

[0116] E6. Select the eigenvector a corresponding to the first k largest eigenvalues ​​whose cumulative contribution rate is greater than 95%. 1 ,a 2 ,…,a k as the principal component;

[0117] E7, the principal component a 1 ,a 2 ,…,a k The composite projection matrix P = [a 1 ,a 2 ,…,a k ];

[0118] E8. Based on the projection matrix, the feature parameter matrix is ​​projected into the new space defined by the principal component to obtain the data Y=X after the principal component dimension reduction. std P.

[0119] The process of clustering the characteristic parameter data after the principal component dimension reduction in S4 for short-term working conditions described in S5 is:

[0120] F1, at this time, the sample length T of the divided working condition is 30, and the sample interval ΔT of the divided working condition is 5;

[0121] F2. Perform K-MEANS clustering on the characteristic parameter data after dimensionality reduction of the principal component of S4, and the number of clusters is optimized based on the result of the silhouette coefficient.

[0122] The process of S6 is:

[0123] G1, take the characteristic parameter data after dimensionality reduction of the principal component of S4 as features, the short-term working condition category obtained by S5 and the long-term working condition category obtained by S2 as labels, randomly select 20% of the data as the test set, and then randomly select 20% of the data from the remaining data as the validation set, and the remaining data as the training set;

[0124] G2. Select the multi-layer perceptron neural network as the main framework of the multi-time scale working condition recognition model, select the cross entropy loss function as the loss function, select the Adam optimizer as the optimizer, use the data of the training set to train the model, and observe the loss and accuracy of the model on the validation set at the same time. End the training process when the loss of the validation set does not decrease significantly;

[0125] G3. Use the test set data to test the performance of the trained multi-time scale working condition recognition model;

[0126] G4. Adjust the training parameters and repeat the above steps until good model performance is obtained and save the model parameters.

[0127] The multi-scale identification method of automobile driving conditions is used to assist in the development of automobile energy management strategies, automatic driving algorithms, etc.

[0128] The present invention introduces multi-time scale analysis and recognition, which can simultaneously identify short-term and long-term operating conditions. It can identify six short-term operating conditions, namely idling, constant speed, acceleration, deceleration, emergency stop and emergency start, as well as three long-term operating conditions, namely urban driving conditions, suburban driving conditions and high-speed driving conditions. This helps to improve the recognition accuracy of vehicle driving conditions and the robustness of condition recognition, and thus helps to predict future vehicle failures and abnormalities, optimize energy distribution and improve vehicle energy utilization.

[0129] The following is a specific application scenario of the present invention.

[0130] A multi-time scale operating condition identification method is developed for heavy trucks. The specific steps are as follows:

[0131] Step 1: Referring to GB / T 38146.2-2019, for Chinese trucks with a gross vehicle weight exceeding 5500kg, the driving condition data adopts CHTC-HT. Since the driving time of the original data is short, the original data conditions are cyclically reconstructed to achieve the purpose of expanding the sample length;

[0132] Step 2: Preprocess the reconstructed operating conditions, including missing value processing and outlier processing. Then remove the continuous zero values ​​in the operating conditions and mark them as long-term parking conditions in the original samples. The final operating condition curve is as follows: Figure 2 As shown, the long-term operating condition category 0 represents a long-term parking condition, 1 represents an urban condition, 2 represents a suburban condition, and 3 represents a high-speed condition;

[0133] Step 3: Divide the working condition data obtained in step 2 into short-term working condition samples with a working condition sample length T of 60 and a working condition sample interval ΔT of 5 and long-term working condition samples with a working condition sample length T of 300 and a working condition sample interval ΔT of 5. The working condition sample division example is as follows: Figure 3 As shown;

[0134] Step 4: Divide the long-term operating condition samples into categories according to the long-term operating condition categories in step 3, and correspond them one by one with the long-term operating condition samples obtained in step 3;

[0135] Step 5: Calculate characteristic parameters of the short-term working condition sample data and the long-term working condition sample data divided in step 2 to obtain characteristic parameter data. Some characteristic parameter data are shown in Tables 1 and 2.

[0136] Table 1 Some short-term characteristic parameter data

[0137]

[0138] Table 2 Some long-term characteristic parameter data

[0139]

[0140]

[0141] Table 3 Characteristic parameter data corresponding to short-term working conditions

[0142]

[0143]

[0144] Step 6: Perform principal component dimensionality reduction analysis on the characteristic parameter data calculated in step 5. The analysis shows that the number of principal components selected for short-term working conditions is 5, and the cumulative contribution rate is 0.973. The number of principal components selected for long-term working conditions is 6, and the cumulative contribution rate is 0.982.

[0145] Step 7: Perform short-term operating condition clustering on the short-term characteristic parameter data after principal component dimension reduction in step 6. The number of clusters in the optimization result is 6, and the silhouette coefficient is 0.372. The clustering results are as follows: Figure 4 Indicates that the corresponding features are shown in Table 3;

[0146] Step 8: Combine the feature parameter data after the principal component dimension reduction in step 6 with the short-term working condition category obtained in step 7 and the long-term working condition category obtained in step 4 to obtain a multi-time scale working condition recognition model. The loss and accuracy during the model training process are as follows: Figure 5 , Figure 6 As shown, the accuracy on the final test set reached 99.29%.

[0147] The present invention introduces multi-time scale analysis and recognition, which can simultaneously identify short-term and long-term operating conditions. It can identify six short-term operating conditions and three long-term operating conditions: urban driving conditions, suburban driving conditions, and high-speed driving conditions. This helps to improve the recognition accuracy of vehicle driving conditions and the robustness of condition recognition, and thus helps to predict future vehicle failures and anomalies, optimize energy distribution, and improve vehicle energy utilization.

[0148] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A multi-scale identification method for vehicle driving conditions, characterized in that: The method comprises the following steps: S1. Acquire and preprocess driving condition data to obtain preprocessed condition data and long-term condition categories; S2. The pre-processed working condition data is divided into short-term working condition samples and long-term working condition samples, and the long-term working condition categories are divided according to the long-term working condition samples, which correspond one to one with the long-term working condition samples; S3, calculating characteristic parameters of the divided working condition data to obtain characteristic parameter data of each working condition sample; S4. Principal component analysis and dimensionality reduction are performed on the characteristic parameter data of the working condition samples; S5, clustering the characteristic parameter data after the principal component dimension reduction into short-term operating condition categories to obtain short-term operating condition categories; S6. Perform model training based on the feature parameter data after dimensionality reduction, the short-term operating condition category and the long-term operating condition category to obtain a multi-time scale operating condition recognition model, wherein the multi-time scale operating condition recognition model obtains actual operating condition data and outputs a recognition result.

2. The multi-scale identification method for vehicle driving conditions according to claim 1, characterized in that: The pre-processing is specifically as follows: A1. Check whether the amount of driving condition data reaches a threshold. If not, expand the driving condition data based on loop reconstruction or data enhancement technology and return to A1. Otherwise, execute A2. A2. Check whether the driving condition data obtained in A1 has missing values ​​and abnormal values. If yes, perform missing value processing and abnormal value processing, and then execute A3. Otherwise, directly execute A3. A3, identifying the long-term operating condition category of the driving operating condition data obtained in A2, and whether there are continuous zero values ​​above the quantity threshold. If so, remove the continuous zero values ​​and mark the long-term parking long-term operating condition in the driving operating condition data to obtain the pre-processed operating condition data, and then execute A4. Otherwise, obtain the pre-processed operating condition data and directly execute A4; A4. Calculate the acceleration from the pre-processed working condition data.

3. The multi-scale identification method for vehicle driving conditions according to claim 1, characterized in that: The specific steps for dividing the short-term working condition samples and the long-term working condition samples in S2 are: B1. Select the 1st to Tth data points in the preprocessed working condition data to form the first section of working condition samples; B2, with an interval of ΔT data points, select the (1+ΔT)th to (1+ΔT+T)th data points of the preprocessed operating condition data to form the second section of the operating condition sample; B3. Repeat the above steps; B4. Select the (1+kΔT)th to (1+kΔT+T)th data points of the preprocessed operating condition data to form the last section of the operating condition sample (k∈Z, 1+kΔT+T≤N<1+kΔT+ΔT+T), and repeat the above steps with different operating condition sample lengths T to obtain short-term operating condition sample division and long-term operating condition sample division.

4. The multi-scale identification method for vehicle driving conditions according to claim 1, characterized in that: The specific steps of S3 are: D1. Calculate the maximum speed from the divided working condition data where v i Indicates the speed of each working condition sample; D2. Calculate the average speed from the divided working condition data D3. Calculate the maximum acceleration from the divided working condition data Among them, a i Indicates the acceleration of each working condition sample; D4. Calculate the maximum deceleration from the divided working condition data D5. Calculate the average acceleration from the divided working condition data D6. Calculate the average deceleration from the divided working condition data D7. Calculate the acceleration time ratio based on the divided working condition data in D8. Calculate the deceleration time ratio based on the divided working condition data D9. Calculate the uniform speed time ratio from the divided working condition data D10. Calculate the idle time ratio based on the divided working condition data The above parameters constitute the characteristic parameter data.

5. The multi-scale identification method for vehicle driving conditions according to claim 1, characterized in that: The specific steps of S4 are: The following operations are performed on the characteristic parameter data corresponding to the short-term working condition samples and the long-term working condition samples: E1. The characteristic parameter data are organized into a corresponding characteristic parameter matrix X, where the number of rows of X is the number of working condition samples and the number of columns is the number of features; E2. Standardize the feature parameter matrix X, subtract the mean of each feature parameter, and then divide it by its standard deviation to obtain the standardized feature parameter matrix X std ; E3, after the standardized feature parameter matrix X std Calculate its covariance matrix; E4. Solve the eigenvalues ​​and eigenvectors of the covariance matrix; E5. Calculate the principal component contribution rate and cumulative contribution rate; E6. Select the eigenvectors a1, a2, ..., a corresponding to the first k largest eigenvalues ​​whose cumulative contribution rate is greater than 95%. k as the principal component; E7, the principal components a1, a2, ..., a k The composite projection matrix P = [a1, a2, ..., a k ]; E8. Based on the projection matrix, the feature parameter matrix X std Projected into the new space defined by the principal component, the data Y=X after the principal component dimension reduction is obtained std P.

6. The multi-scale identification method for vehicle driving conditions according to claim 5, characterized in that: The covariance matrix is: Among them, ∑ is the covariance matrix and n is the number of working condition samples.

7. The multi-scale identification method for vehicle driving conditions according to claim 6, characterized in that: The specific solution for the eigenvalues ​​and eigenvectors of the covariance matrix is: det(Σ-λI)=0 (Σ-λI)a=0 Where λ is the eigenvalue, I is the identity matrix, and a is the eigenvector.

8. The multi-scale identification method for vehicle driving conditions according to claim 7, characterized in that: The principal component contribution rate and cumulative contribution rate are: Among them, d i is the main component contribution rate, r i is the cumulative contribution rate.

9. The multi-scale identification method for vehicle driving conditions according to claim 1, characterized in that: The specific steps of S5 are: K-MEANS clustering is performed on the characteristic parameter data after principal component dimensionality reduction corresponding to the short-term operating condition samples to obtain the short-term operating condition category.

10. The multi-scale identification method for vehicle driving conditions according to claim 1, characterized in that: The steps of S6 are: G1. Use the reduced dimension feature parameter data as features, short-term working condition category and long-term working condition category as labels, and divide them into validation set, training set and test set; G2. Select the multi-layer perceptron neural network as the main framework of the multi-time scale working condition recognition model, select the cross entropy loss function as the loss function, select the Adam optimizer as the optimizer, use the data of the training set to train the model, and observe the loss and accuracy of the model on the validation set at the same time. The training process ends when the loss of two consecutive validation sets is lower than the threshold; G3. Use the test set data to test the performance of the trained multi-time scale working condition recognition model; G4. Adjust the training parameters and repeat the above steps until the final multi-time scale working condition recognition model is obtained. Based on the final multi-time scale working condition recognition model, the actual working condition data is obtained and the recognition result is output.

Images