Vehicle wheel edge feature extraction method based on big data clustering

Abstract

The invention relates to a vehicle wheel edge feature extraction method based on big data clustering, and the method comprises the following steps: Internet of Vehicles data collection and acquisition: employing a vehicle-mounted sensor to collect GPS altitude, vehicle speed, rotation speed, torque, main braking, gear and dead weight signal data, and obtaining the needed signal data through an Internet of Vehicles data platform; taking a single vehicle as a sample of the obtained signal data, aligning GPS altitude, vehicle speed, rotating speed, torque, main braking, gears and vehicle and cargo total weight data according to time through a polynomial interpolation method, and smoothing the GPS altitude signal data through an abnormal value and mean value filtering method; calculating wheel edge data; extracting static features and dynamic features; and feature clustering is carried out by a K-MEANS method.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy commercial vehicle control, and particularly to a method for extracting vehicle wheel side features based on big data clustering. Background Art

[0002] Hybrid vehicle models usually use the backward simulation software ADVISOR to establish a vehicle model based on SIMULINK, and analyze and calculate along the opposite direction of energy transfer starting from the simulation working condition requirements; calculate the power demand for vehicle components according to the vehicle dynamics equation, and then each component makes a passive output response according to the demand. Among them, the simulation working conditions generally use general working conditions or self-constructed working conditions to simulate user working conditions. The general working conditions are generally chtc, c-wtvc, etc.; the self-constructed working conditions generally use principal component analysis, Markov chain, neural network algorithm to construct the vehicle cycle working condition (time-speed spectrum of 1800s) under specific scenarios;

[0003] However, the power demand based on backward simulation is calculated based on multiple vehicle parameters along the opposite direction of energy transfer from the simulation working condition. The simulation working condition is only a time-speed spectrum, and the characteristics of altitude and slope are generally reflected in the speed spectrum in the form of acceleration conversion, lacking information such as slope, and the calculation of power demand is not accurate enough; the cycle working condition road spectrum is usually synthesized based on the recombination of motion segments. The selection and recombination of motion segments will inevitably cause deviations between the constructed working condition and the original working condition characteristics. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for extracting vehicle wheel side features based on big data clustering in view of the deficiencies of the prior art.

[0005] The present invention is implemented by adopting the following technical solutions:

[0006] A method for extracting vehicle wheel side features based on big data clustering includes the following steps:

[0007] S1, Vehicle networking data collection and acquisition: Use on-vehicle sensors to collect signal data of GPS altitude, vehicle speed, rotation speed, torque, main brake, gear position and self-weight, and obtain the required signal data through the vehicle networking data platform;

[0008] S2, Vehicle networking data preprocessing: Take the acquired signal data as a single vehicle sample, align the GPS altitude, vehicle speed, rotation speed, torque, main brake, gear position, and total vehicle and cargo weight data according to time through polynomial interpolation method, and smooth the GPS altitude signal data through the outlier plus mean filtering method;

[0009] S3, Wheel side data calculation;

[0010] S4, Static feature and dynamic feature extraction;

[0011] In S5, the K - MEANS method is used for feature clustering.

[0012] Preferably, in S3, it is judged whether the current moment is a driving condition or a braking condition according to the vehicle speed, the engine output torque, and the braking signal. The judgment conditions are:

[0013] When the vehicle speed > 0 km / h and the engine output torque > 0 Nm, it is a driving condition;

[0014] When the vehicle speed > 0 km / h, the engine output torque = 0 Nm, and the main brake > 0 or the auxiliary brake > 0, it is a braking condition.

[0015] Preferably, the driving condition power and the braking condition braking power are:

[0016]

[0017] In the formula, P 制动 (t) represents that the moment t is a braking condition, satisfying that the vehicle speed v(t) > 0, v(t) represents the vehicle speed at the moment t, and the braking signal main brake > 0 or the auxiliary brake > 0. m is the vehicle mass, g is the acceleration due to gravity, θ is the longitudinal slope, ρair is the air density, C D is the air resistance coefficient, A is the frontal area, ν is the vehicle speed, δ is the rotating mass conversion coefficient, and α is the driving acceleration;

[0018] Among them, the longitudinal slope θ is calculated from the GPS altitudes at the moments t + 1 and t - 1;

[0019] P 驱动 (t) represents that the moment t is a driving condition, satisfying that the vehicle speed v(t) > 0 and the engine output torque T(t) > 0;

[0020] P 发动机净输出 (t) is calculated from the engine speed n(t) and the engine output torque T(t);

[0021]

[0022] P 风扇 (t) is obtained by looking up a table from the current fan speed fan_rpm(t);

[0023] P 打气泵 (t) is the power consumed by the current air pump;

[0024] P 发电机 (t) is the power consumed by the current generator;

[0025] η1 is the overall efficiency of the vehicle driveline at the current moment;

[0026] Obtain the wheel-end data Bi of a single sample = {(p i , v i , t i ) | i ∈ N}.

[0027] Preferably, for static feature extraction in S4: The wheel-end static feature is a two-dimensional distribution matrix of wheel-end power - vehicle speed. The vehicle speed range is 0 - 120 km / h, with an interval of 5 km / h. The wheel-end power range is -600 - 400 Kw, with an interval of 50 Kw, to obtain a vehicle speed - power matrix.

[0028] Preferably, for dynamic feature extraction in S4: The wheel-end dynamic feature is a state transition probability matrix of wheel-end features. First, segment the operation fragments according to continuous driving and continuous braking conditions. For each operation fragment, extract the average power, duration, and power standard deviation as features. The average power range is -600 - 400 Kw, with an interval of 100 Kw. The duration range is 0 - 300 S, with an interval of 60 S. The power standard deviation range is 0 - 200, with an interval of 50. A total of 200 states are obtained to get the state transition probability matrix.

[0029] Preferably, for feature clustering using the K - MEANS method in S5: First, apply the K - means method to cluster the static feature matrices of multiple samples, and use the silhouette coefficient evaluation method to select the appropriate number of clusters to obtain K1 classes. Then, apply the K - means method to cluster the dynamic features of each class of samples, and use the silhouette coefficient evaluation method to select the appropriate number of clusters. After two - layer clustering, obtain K2 classes and the cluster centers of each class;

[0030] There are multiple similar samples in each class. Select the sample B' = {(p i , v i , t i ) | i ∈ N} with the highest similarity to the cluster center to represent the features of this class.

[0031] Compared with the prior art, the present invention has the following beneficial technical effects:

[0032] The wheel-end working condition features based on user big data clustering obtained by the present invention are time - continuous and have a long mileage, representing the real cycle working conditions of customer operations, and solving the problem of deviation between the constructed working conditions and the original working condition features caused by the selection and recombination of motion fragments in the cycle working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0034] To enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0035] As Figure 1 shown, a method for extracting vehicle wheel-end features based on big data clustering includes the following steps:

[0036] S1, Vehicle networking data collection and acquisition: Use on-vehicle sensors to collect signal data such as GPS altitude, vehicle speed, rotational speed, torque, main brake, gear position, and self-weight, and obtain the required signal data through the vehicle networking data platform;

[0037] S2, Vehicle networking data preprocessing: Take the acquired signal data with a single vehicle as a sample, align the GPS altitude, vehicle speed, rotational speed, torque, main brake, gear position, and total vehicle and cargo weight data according to time through polynomial interpolation, and smooth the GPS altitude signal data through the outlier plus mean filtering method;

[0038] Among them, the outlier plus mean filtering method for smoothing the GPS altitude signal data is to remove outliers from the altitude data through the maximum and minimum values. The minimum value is 0, and the maximum value is 5000. The mean filtering method is that the altitude of the current point is replaced by the average value of the altitudes of the previous and next 5 points.

[0039] S3, Wheel-end data calculation;

[0040] The wheel-end data refers to the wheel-end vehicle speed - power characteristics, representing the vehicle's demand.

[0041] S4, Static feature and dynamic feature extraction;

[0042] S5, Feature clustering using the K-MEANS method.

[0043] S3 is to judge whether the current moment is a driving condition or a braking condition according to the vehicle speed, engine output torque, and braking signal. The judgment conditions are:

[0044] Satisfying vehicle speed > 0 km / h and engine output torque > 0 Nm is the driving condition;

[0045] It means that if the current vehicle speed is greater than 0 km / h and the engine output torque is greater than 0 NM, it indicates that the vehicle is currently in a driving state;

[0046] When the vehicle speed > 0 km / h, the engine output torque = 0 Nm, and the main brake > 0 or the auxiliary brake > 0, it is a braking condition.

[0047] It means that if the current vehicle speed is greater than 00 km / h, the engine has no torque output, and the main brake / auxiliary brake is greater than 0, indicating that the current vehicle is in a braking state.

[0048] The driving condition power and the braking condition braking power are as follows:

[0049]

[0050] In the formula, P 制动 (t) indicates that the t-th moment is a braking condition, satisfying the vehicle speed v(t) > 0, v(t) represents the vehicle speed at the t-th moment, and the braking signal main brake > 0 or auxiliary brake > 0, m is the vehicle mass, g is the acceleration due to gravity, θ is the longitudinal slope, ρair is the air density, C D is the air resistance coefficient, A is the frontal area, ν is the vehicle speed, δ is the rotating mass conversion coefficient, α is the driving acceleration;

[0051] Among them, the longitudinal slope θ is calculated from the GPS altitudes at the t + 1 moment and the t - 1 moment;

[0052] P 驱动 (t) indicates that the t-th moment is a driving condition, satisfying the vehicle speed v(t) > 0 and the engine output torque T(t) > 0;

[0053] P 发动机净输出 (t) is calculated from the engine speed n(t) and the engine output torque T(t);

[0054]

[0055] P 风扇 (t) is obtained by looking up a table from the current fan speed fan_rpm(t);

[0056] P 打气泵 (t) is the power consumed by the current air pump;

[0057] P 发电机 (t) is the power consumed by the current generator;

[0058] η1 is the overall efficiency of the vehicle driveline at the current moment;

[0059] The wheel-end data Bi of a single sample is obtained as Bi = {(p i , v i , t i ) | i ∈ N).

[0060] Static feature extraction in S4: The static feature of the wheel end is a two-dimensional distribution matrix of wheel end power - vehicle speed. The vehicle speed range is 0 - 120 km / h, with an interval of 5 km / h. The wheel end power range is -600 - 400 Kw, with an interval of 50 Kw. The vehicle speed - power matrix is obtained as follows:

[0061] <![CDATA[v1]]> <![CDATA[v2]]> … <![CDATA[v m > <![CDATA[p1]]> <![CDATA[q 11 > <![CDATA[q 21 > <![CDATA[q m1 > <![CDATA[p2]]> <![CDATA[q 12 > <![CDATA[q 22 > <![CDATA[q m2 > … <![CDATA[p n > <![CDATA[q 1n > <![CDATA[q 2n > <![CDATA[q nm >

[0062] Among them, v1 represents the vehicle speed range between 0 and 5 Km / h, and P1 represents the wheel end power range between -600 and -550 Kw, where q ij represents the frequency proportion of the vehicle speed range from v i to v i+1 and the power range from p i to p i+1 , and it satisfies

[0063]

[0064] Dynamic feature extraction in S4: The dynamic feature of the wheel end is the state transition probability matrix of the wheel end features. First, the operation segments are segmented according to the continuous driving and continuous braking conditions. For each operation segment, the average power, duration, and power standard deviation are extracted as features. The average power range is -600 - 400 Kw, with an interval of 100 Kw. The duration range is 0 - 300 S, with an interval of 60 S. The power standard deviation range is 0 - 200, with an interval of 50. A total of 200 states are obtained, and the state transition probability matrix is obtained as follows:

[0065] <![CDATA[S1]]> <![CDATA[S2]]> … <![CDATA[S m > <![CDATA[S1]]> <![CDATA[p 11 > <![CDATA[p 21 > <![CDATA[p m1 > <![CDATA[S2]]> <![CDATA[p 12 > <![CDATA[p 22 > <![CDATA[p m2 > … <![CDATA[S m > <![CDATA[p 1m > <![CDATA[p 2m > <![CDATA[p mm >

[0066] Among them, S i represents the state (p i , t i , s i ), and S1 represents the state (p1, t1, s1), which specifically represents that the average power of this segment is between -600 kW and -500 Kw, the duration is between 0 and 60 seconds, and the power standard deviation is 0 - 50. p ij represents the probability of transitioning from state S i to S j , and it satisfies

[0067]

[0068] Feature clustering using the K-Means method in S5: First, perform K-means clustering on the static feature matrices of multiple samples, and use the silhouette coefficient evaluation method to select the appropriate number of clusters. The silhouette coefficient is an index used to evaluate the effect of clustering algorithms, which combines the tightness within clusters and the separation between clusters. The larger the silhouette coefficient, the farther the distance between clusters and the better the clustering effect. Select the number of clusters according to the silhouette coefficient to obtain m classes, namely L1, L2, L3, … Lm, that is, class 1, class 2, class 3, …, and each class contains N i samples. Next, cluster the dynamic features of multiple samples in each class. Similarly, apply the silhouette coefficient evaluation method to obtain n classes. Taking L1 as an example, the clustering results are L11, L12, L13, that is, class 11, class 12, class 13, …, and each class contains M samples. And it satisfies that the total number of samples corresponding to L11, L12, and L13 is equal to the number of samples in L1.

[0069] During clustering, output the rim data B′ = {(p i , v i , t i )|i ∈ N} corresponding to the samples at the cluster center to represent the features of this class. The features of all cluster centers constitute the representative working condition library.

Claims

1. A vehicle wheel edge feature extraction method based on big data clustering, characterized in that: The following steps are involved: S1, Internet of Vehicles data collection and acquisition: Use vehicle-mounted sensors to collect GPS altitude, vehicle speed, rotation speed, torque, main brake, gear position and deadweight signal data, and obtain the required signal data through the Internet of Vehicles data platform; S2, IoV data preprocessing: The acquired signal data is sampled from a single vehicle, and the GPS altitude, vehicle speed, rotation speed, torque, main brake, gear position, and vehicle-cargo gross weight data are aligned according to time through polynomial interpolation method, and the GPS altitude signal data is smoothed through outlier plus mean filtering method; S3, wheel-side data calculation; S4, static feature and dynamic feature extraction; S5, K-MEANS method is used for feature clustering.

2. The vehicle wheel edge feature extraction method based on big data clustering according to claim 1 is characterized in that: S3 is to judge whether the current moment is a driving condition or a braking condition according to the vehicle speed, engine output torque and brake signal. The judgment conditions are: The driving condition is when the vehicle speed is greater than 0km / h and the engine output torque is greater than 0Nm; The braking condition is when the vehicle speed is greater than 0 km / h, the engine output torque is 0 Nm, and the main brake is greater than 0 or the auxiliary brake is greater than 0.

3. The vehicle wheel edge feature extraction method based on big data clustering according to claim 2 is characterized in that: The driving power and braking power are: Where P 制动 (t) indicates that the braking condition is at time t, and the vehicle speed v(t)>0, v(t) indicates the vehicle speed at time t, and the braking signal main braking>0 or auxiliary braking>0, m is the vehicle mass, g is the acceleration of gravity, θ is the longitudinal slope, ρair is the air density, C D is the air resistance coefficient, A is the frontal area, ν is the vehicle speed, δ is the rotation mass conversion coefficient, and α is the driving acceleration; The longitudinal slope θ is calculated from the GPS altitude at time t+1 and time t-1; P 驱动 (t) indicates that the time t is the driving condition, satisfying the vehicle speed v(t)>0 and the engine output torque T(t)>0; P 发动机净输出 (t) is calculated by the engine speed n(t) and the engine output torque T(t); P 风扇 (t) is obtained by looking up the table of the current fan speed fan_rpm(t); P 打气泵 (t) The power currently consumed by the air pump; P 发电机 (t) the power currently consumed by the generator; η1 is the comprehensive efficiency of the vehicle powertrain at the current moment; Get the wheel edge data of a single sample Bi = {(p i , v i , t i )|i∈N}.

4. The vehicle wheel edge feature extraction method based on big data clustering according to claim 1 is characterized in that: Static feature extraction in S4: The wheel-side static feature is a two-dimensional distribution matrix of wheel-side power-vehicle speed. The vehicle speed range is 0 to 120 km / h, with an interval of 5 km / h. The wheel-side power range is -600 to 400 kW, with an interval of 50 kW, and the vehicle speed-power matrix is ​​obtained.

5. The vehicle wheel edge feature extraction method based on big data clustering according to claim 1, characterized in that: Dynamic feature extraction in S4: The wheel-side dynamic feature is the state transfer probability matrix of the wheel-side feature. First, the operation segments are divided according to continuous driving and continuous braking conditions. The average power, duration, and power standard deviation are extracted as features for each operation segment. The average power range is -600~400Kw, with an interval of 100Kw. The duration range is 0~300S, with an interval of 60S. The power standard deviation range is 0~200, with an interval of 50. A total of 200 states are obtained, and the state transfer probability matrix is ​​obtained.

6. The vehicle wheel edge feature extraction method based on big data clustering according to claim 1 is characterized in that: The K-MEANS method in S5 performs feature clustering as follows: first, the static feature matrix of multiple samples is clustered using the K-means method, and the silhouette coefficient evaluation method is used to select the appropriate number of clusters to obtain K1 classes; then, the dynamic features of each class of samples are clustered using the K-means method, and the silhouette coefficient evaluation method is used to select the appropriate number of clusters. After two layers of clustering, K2 classes and the cluster center of each class are obtained; There are multiple similar samples in each class, and the sample with the highest similarity to the cluster center is selected. i , v i , t i )|i∈N} features to represent the characteristics of this class.

Images