Vehicle braking performance degradation prediction method considering operating condition density

Abstract

The invention belongs to the technical field of vehicle braking performance degradation prediction, and particularly relates to a vehicle braking performance degradation prediction method considering operating condition density. According to the prediction method disclosed by the invention, the probability distribution characteristics of the micro-stroke working condition are incorporated into brake performance degradation modeling, and the degradation of the brake performance is accurately predicted based on statistical characteristics and probability analysis of the actual working condition. In the prediction process, the micro-stroke category is converted into a state sequence, and the probability of continuous occurrence of extreme working conditions in a short time is calculated by using the transition probability matrix, so that the short-term heat fade risk is better predicted.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle braking performance degradation prediction, and particularly relates to a method for predicting vehicle braking performance degradation considering the density of operating conditions. Background Art

[0002] With the wide application of intelligent networked and new energy vehicles, the operating conditions of vehicles have gradually become diversified and complex. As a key component to ensure driving safety, the performance of brakes will gradually degrade with the service time and operating conditions. Traditional maintenance strategies often rely on fixed periods or mileage, and cannot fully consider the long-term wear conditions and short-term thermal degradation risks under the actual operating conditions of vehicles. Therefore, a braking performance degradation prediction method based on actual operating conditions is needed, which can capture both the long-term wear trend and predict the short-term thermal degradation probability, so as to provide accurate decision-making support for maintenance and replacement strategies. Summary of the Invention

[0003] The present invention proposes a method for predicting vehicle braking performance degradation considering the density of operating conditions. The prediction method in the present invention accurately predicts the degradation of braking performance according to the actual operating conditions, including the risks of long-term wear and short-term thermal degradation. Compared with the existing maintenance strategies (such as fixed periods or mileage), the present invention can provide a more accurate and personalized maintenance plan, and solves the above problems.

[0004] The technical solution of the present invention is realized as follows:

[0005] A method for predicting vehicle braking performance degradation considering the density of operating conditions, comprising the following steps:

[0006] S1. Collect data on the operating conditions of the vehicle;

[0007] S2. Detect outliers using the single-dimensional local outlier factor (LOF);

[0008] S3. Perform single-dimensional Gaussian weighted moving average filtering;

[0009] S4. Divide micro-trips;

[0010] S5. Construct feature vectors;

[0011] S6. Extract the depth features of operating conditions using autoencoder for dimensionality reduction;

[0012] S7. Cluster in the dimensionality-reduced space;

[0013] S8. Model and classify the long-term wear degradation test;

[0014] The long-term wear degradation model is:

[0015]

[0016] Among them, D long is the long-term wear recession degree, N is the total number of brake applications, E j is the energy consumed during the j-th brake application, α1 represents the basic wear generated by each brake application, and α2 represents the influence degree of the energy consumed during each brake application on wear;

[0017] Classify the long-term wear:

[0018]

[0019] Among them, d L2 , d L3 , d L4 are the long-term wear thresholds determined based on experience and tests;

[0020] S9, short-term thermal recession risk test modeling, probability calculation, classification;

[0021] Obtain the micro-trip transition probability matrix P according to the vehicle operating condition data:

[0022] P ij =P(C t+1 =j|C t =i), i, j = 1,..., K;

[0023] Obtain the extreme brake class set according to the test results and calculate the probability of at least r occurrences of extreme classes in the next h micro-trips:

[0024]

[0025] Among them, h is the short-term time window length, also known as the number of micro-trips, r is the minimum number of occurrences of extreme classes required within the short-term window, is the indicator function;

[0026] Establish a short-term thermal recession model and calculate the short-term thermal recession degree D temp :

[0027] D temp =β1(T brake -T0)+β2(T brake -T0) 2 ;

[0028] Among them, T brake =T0+γP extreme , T brake is the brake temperature, T0 is the reference ambient temperature, and β1, β2, γ are bench test fitting parameters;

[0029]

[0030] Among them, d T2 , d T3 , d T4 are the short-term heat recession grading thresholds, and ρ3, ρ4 are the thresholds matching the extreme probabilities.

[0031] Through the above technical solution, the method can calculate the long-term wear recession degree based on the basic wear generated by each braking and the energy consumption during the braking process. By using the transition probability matrix of the micro-stroke and the probability of extreme braking events, the method can quickly evaluate the heat recession risk of the vehicle under the current driving conditions according to the real-time working condition data.

[0032] Optionally, in step S1, long-term collection of working condition data gives

[0033] X(t) = [x1(t), x2(t), …, x i (t)], t = 1, …, T;

[0034] Among them, x i (t) represents the value of the i-th dimension of the working condition parameter collected at time t.

[0035] Optionally, in step S2, LOF detection is performed on each dimension x i (t) respectively, and the specific content is as follows:

[0036] (2) Define the distance function:

[0037] d(x i (t), x i (t′)) = |x i (t) - x i (t′)|;

[0038] Among them, x i (t′) represents the value of the i-th working condition parameter at time t′;

[0039] (2) Define the k-distance function:

[0040] k-dist(x i (t)) = min{r∣|{x i (t′): d(x i (t), x i (t′)) ≤ r}| ≥ k};

[0041] The meaning of k-dist(x i (t)) is to start from the point x i (t) and find a distance r such that there are at least k data points within this distance;

[0042] (3) Define the reachable distance function:

[0043] reach-dist k (x i (t), x i (t′)) = max{k-dist(x i (t′)), d(x i (t), x i (t′))};

[0044] (4) Define the local reachability density function;

[0045]

[0046] where N k (x i (t)) is the set of k-nearest neighbors of x i (t);

[0047] (5) Define the local outlier factor LOF function:

[0048]

[0049] If LOF k (x i (t)) > γ i , then x i (t) is an outlier in this dimension, and γ i is a user-defined threshold.

[0050] Through the above technical solution, by using the LOF method, outliers under specific working conditions can be effectively identified.

[0051] Optionally, in step S3, replace the outlier points with a filter. Let the window length be W, Define the weight:

[0052]

[0053] where σ is the standard deviation of Gaussian filtering, w(τ) represents the weight of the data point at distance τ, and M represents the half-width of the window;

[0054] Filter output:

[0055]

[0056] Use to replace the outlier which represents the data of the data point at the current moment and the data at the previous and next moments. τ is the offset, indicating the data point at the moment offset by τ before and after the current moment.

[0057] Through the above technical solution, the Gaussian weighted moving average filtering method provides reliable and accurate input data for subsequent analysis and prediction by smoothing and removing outliers, and is applicable to noise processing and data cleaning in vehicle operating condition information.

[0058] Optionally, in step S4, according to the longitudinal v(t) = x1(t), where: v(t) is the vehicle speed (m / s), the operating condition data is divided into micro-trips: a period of time when the speed increases from 0 and finally returns to 0 again, and the sequence is one micro-trip S i :

[0059] S = {S1, S2, …, S M};

[0060] where, S i is the i-th micro-trip data segment, M is the total number of micro-trips, and S represents the set of all micro-trips.

[0061] Through the above technical solution, the above formula ensures that each micro-trip represents a complete driving operation cycle, which is used to more finely capture the dynamic characteristics within each small cycle.

[0062] Optionally, in step S5, for each micro-trip S i , a set of statistical features is extracted from the original time series data to form a feature vector F i ; Let the micro-trip S i contain the time series where: v(t) is the vehicle speed (m / s), a(t) is the vehicle acceleration (m / s 2 ), m(t) is the load (kg), and θ(t) is the slope (°);

[0063] Define the micro-trip length as |S i |, that is, the micro-trip S i contains |S i | discrete time points. The following feature calculations are all for the data within this set;

[0064] (1) Average speed:

[0065]

[0066] (2) Maximum speed:

[0067]

[0068] (3) Minimum speed:

[0069]

[0070] (4) Speed variance:

[0071]

[0072] (5) Average acceleration:

[0073]

[0074] (6) Acceleration variance:

[0075]

[0076] (7) Maximum acceleration:

[0077]

[0078] (8) Minimum acceleration:

[0079]

[0080] (9) Proportion of rapid deceleration:

[0081] Define rapid deceleration as the time proportion when the acceleration a(t) < -3m / s 2 ;

[0082]

[0083] where is an indicator function, which is 1 when the condition is satisfied and 0 otherwise.

[0084] (10) Average gradient:

[0085]

[0086] (11) Average load:

[0087]

[0088] Combine these features into a vector:

[0089]

[0090] F i contains the statistical indicators of various important data of the vehicle in the i-th micro-trip.

[0091] Through the above technical solution, by calculating the statistical indicators, the statistical characteristics of the vehicle in different micro-trips can be reflected, providing an accurate statistical data basis for subsequent performance prediction and driving behavior analysis.

[0092] Optionally, in step S6, the feature vector d = 11, which is the number of feature dimensions, and an autoencoder can be introduced for {F i}Perform dimensionality reduction and deep feature extraction. The autoencoder represents high-dimensional input data through a low-dimensional latent variable space and is trained under the reconstruction error.

[0093] Let the encoder mapping be:

[0094] z i = f enc (F i ; θ enc );

[0095] where and l < d, θ enc is the encoder parameter;

[0096] The decoder mapping is:

[0097]

[0098] where is the reconstructed feature vector, θ dec is the decoder parameter;

[0099] The training objective function of the autoencoder can be defined as minimizing the mean square error:

[0100]

[0101] After training, the output z i of the encoder is the deep feature representation after dimensionality reduction, capturing the more essential latent structure in the original features.

[0102] Through the above technical solution, the data can be mapped from the high-dimensional space to the low-dimensional latent space without losing key information, and the most representative part of the original features is extracted. Compared with the linear dimensionality reduction method, the autoencoder method is more suitable for multi-dimensional, complex, and non-linear data sets such as the operating condition data of vehicles.

[0103] After adopting the above technical solution, the beneficial effects of the present invention are:

[0104] 1. The prediction method of the present invention incorporates the probability distribution characteristics of the micro-trip conditions into the braking performance degradation modeling. Based on the statistical characteristics and probability analysis of the actual conditions, it accurately predicts the degradation of the braking performance. During the prediction process, the micro-trip categories are converted into state sequences, and the transition probability matrix is used to calculate the probability of consecutive extreme conditions occurring in a short time, so as to better predict the short-term heat fade risk.

[0105] 2. The present invention extracts statistical characteristics covering multiple angles such as speed, acceleration, slope, load, and sudden deceleration ratio, fully characterizing the nature of the micro-trip. The autoencoder is used for non-linear dimensionality reduction and deep feature extraction of the features, enhancing the clustering discrimination ability.

[0106] 3. The present invention simultaneously predicts two types of degradations with different time scales, namely long-term wear and short-term heat fade, and calibrates parameters using bench tests to more accurately characterize the changes in braking performance, providing a basis for the maintenance strategy of vehicle brakes, thereby improving vehicle safety and economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0108] Figure 1 It is a comparison between the degradation prediction result of the present invention and the result of manual detection. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0109] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0110] The embodiments of the present application disclose a method for predicting the degradation of vehicle braking performance considering the operating condition density.

[0111] Embodiment

[0112] According to Figure 1 As shown, a method for predicting the degradation of vehicle braking performance considering the operating condition density includes the following steps:

[0113] S1. Collect data on the operating conditions of the vehicle;

[0114] Long-term collection of operating condition data results in

[0115] X(t) = [x1(t), x2(t), …, x i (t)], t = 1, …, T;

[0116] where, x i (t) represents the i-th dimensional value of the operating condition parameter collected at time t. X(t) represents the operating condition data vector collected at time t.

[0117] S2. Single - dimensional Local Outlier Factor (LOF) outlier detection; ensure the accuracy of subsequent data processing by detecting whether there are outliers in the operating condition data. If outliers are not processed, it will affect the prediction effect. Perform LOF detection on each dimension x i (t) respectively, and the specific content is as follows:

[0118] (1) Define the distance function (single - dimension):

[0119] d(x i (t),x i (t′))=|x i (t)-x i (t′)|;

[0120] x i (t′) represents the value of the i - th operating condition parameter at time t′.

[0121] (2) Define the k - distance function:

[0122] k - dist(x i (t))=min{r∣|{x i (t′):d(x i (t),x i (t′))≤r}|≥k};

[0123] k - dist(x i (t)) means starting from the point x i (t), find a distance r such that there are at least k data points within this distance.

[0124] (3) Define the reach - distance function:

[0125] reach - dist k (x i (t),x i (t′))=max{k - dist(x i (t′)),d(x i (t),x i (t′))};

[0126] The function helps to measure the difference in local density between two points, so as to more accurately judge whether a point is an outlier.

[0127] (4) Define the local reach - density function (lrd);

[0128]

[0129] Among them, N k (x i (t)) is xi The k-nearest neighbor set of (t). This function is used to measure the density difference between a data point and other points in its neighborhood.

[0130] (5) Define the Local Outlier Factor (LOF) function:

[0131]

[0132] If LOF k (x i (t)) > γ i , then x i (t) is an outlier in this dimension. γ i is a user-defined threshold. This function is an indicator for evaluating whether a data point is an outlier by measuring the density difference of the data point relative to its neighborhood. The larger the LOF value, the more likely the point is an outlier.

[0133] S3, One-dimensional Gaussian weighted moving average filtering; Gaussian weighted moving average filtering is a technique for smoothing data, usually used to remove outliers or noise. It achieves smoothing by performing a weighted average of each point in the data with the points in its neighborhood, where nearby points are given higher weights and distant points are given lower weights.

[0134] Replace the outlier points with a filter. Let the window length be W, Define the weights:

[0135]

[0136] where σ is the standard deviation of Gaussian filtering. w(τ) represents the weight of the data point at distance τ. M represents the half-width of the window.

[0137] Filter output:

[0138]

[0139] Use to replace the outlier w(τ) represents the weight of the data point at distance τ. represents the data of the data point at the current moment and the data at the previous and next moments. τ is the offset, indicating the data point at an offset of τ moments before and after the current moment.

[0140] Step S4, Divide the micro-trip; By dividing the operating condition data into multiple micro-trips, the performance of the vehicle under different operating conditions can be analyzed and modeled more precisely. Each micro-trip can be regarded as an independent operating cycle with similar vehicle behaviors and performances. By dividing the micro-trips, the operating condition cycles with different characteristics can be analyzed separately, and then meaningful features can be extracted for subsequent degradation prediction.

[0141] According to the longitudinal direction \(v(t)=x1(t)\), where: \(v(t)\) is the vehicle speed (m / s), divide the operating condition data into micro-trips (Micro-trip division is a technique for segmenting operating condition data based on the speed change of vehicle operation. Specifically, in this step, the operating condition data is divided by the longitudinal speed of the vehicle, and each micro-trip represents a cycle in which the vehicle speed increases from 0 to a certain maximum value and then returns to 0): When the speed increases from 0 and finally returns to 0 again during a period of time, the sequence is a micro-trip \(S\) i :

[0142] \(S = \{S_1, S_2, \ldots, S\) M \(\}\);

[0143] where \(S\) i is the \(i\)-th micro-trip data segment, \(M\) is the total number of micro-trips, and \(S\) represents the set of all micro-trips;

[0144] Step S5, Construct the feature vector; The construction of the feature vector is to extract a set of representative statistical features from each micro-trip and form these features into a vector. These features describe the main dynamic characteristics of the micro-trip, and these features can fully describe the performance and behavior of the vehicle in a specific operating stage. Forming these features into a vector can be used as the input for subsequent prediction.

[0145] For each micro-trip \(S\) i , extract a set of statistical features from the original time series data to form the feature vector \(F\) i .

[0146] Let the micro-trip \(S\) i contain the time series where: \(v(t)\) is the vehicle speed (m / s), \(a(t)\) is the vehicle acceleration (m / s 2 ²), \(m(t)\) is the load (kg), and \(\theta(t)\) is the slope (°).

[0147] Define the length of the micro-trip as \(|S\) i |, that is, the micro-trip \(S\) i contains \(|S\) i | discrete time points. The following feature calculations are all for the data within this set.

[0148] (1) Average speed:

[0149]

[0150] (2) Maximum speed:

[0151]

[0152] (3) Minimum speed:

[0153]

[0154] (4) Speed variance:

[0155]

[0156] (5) Average acceleration:

[0157]

[0158] (6) Acceleration variance:

[0159]

[0160] (7) Maximum acceleration:

[0161]

[0162] (8) Minimum acceleration:

[0163]

[0164] (9) Proportion of rapid deceleration:

[0165] Define rapid deceleration as the time proportion when the acceleration a(t) < -3 m / s 2 ².

[0166]

[0167] Wherein, is an indicator function, which is 1 when the condition is satisfied and 0 otherwise.

[0168] (10) Average slope:

[0169]

[0170] (11) Average load:

[0171]

[0172] Combine these features into a vector:

[0173] F iContains the statistical indicators of various important data of the vehicle in the i-th micro-trip, which provides a multi-dimensional statistical description of the micro-trip. In subsequent steps (such as using an autoencoder), these feature vectors can be used as input for dimensionality reduction to better capture the core information of the data.

[0174] S6. Extract deep features of working conditions using autoencoder dimensionality reduction; Autoencoder is an unsupervised learning algorithm widely used for data dimensionality reduction and feature learning. Through dimensionality reduction, redundant information in the original features is removed and the core information in the data is retained, thereby improving the efficiency and accuracy of the model and facilitating more accurate predictions. Moreover, the dimension of the data is reduced after dimensionality reduction, which reduces the consumption of computing resources.

[0175] Eigenvector (Here d = 11 is the number of feature dimensions.) In order to better capture the nonlinear relationship and potential structure between features, an autoencoder pair {F i}Perform dimensionality reduction and deep feature extraction.

[0176] The autoencoder consists of an encoder and a decoder, whose goal is to represent high-dimensional input data through a low-dimensional latent variable space and be trained under reconstruction error.

[0177] Assume the encoder mapping is:

[0178] z i =f enc (F i θ enc );

[0179] in, And l <d,θ enc are encoder parameters.

[0180] The decoder mapping is:

[0181]

[0182] in, is the reconstructed feature vector, θ dec Decoder parameters.

[0183] The training objective function of the autoencoder can be defined as minimizing the mean square error (MSE):

[0184]

[0185] After training, the encoder output z i That is, the deep feature representation after dimensionality reduction, which captures the more essential potential structure in the original features.

[0186] S7. Cluster in the dimensionality-reduced space; the purpose of performing K-Means clustering in the dimensionality-reduced space is to classify micro-strokes with similar features into the same category, helping to identify micro-strokes with similar working conditions.

[0187] Obtain the low-dimensional latent variable representation After that, in the dimensionality-reduced space perform K-Means clustering on {z i} to discover the categories of similar micro-strokes. Given the number of clusters K, solve by minimizing the within-class sum of squared errors. After completing the clustering, obtain the category label c i ∈{1,…,K}:

[0188]

[0189] where μ k is the centroid of the k-th class.

[0190] S8. Long-term wear degradation test modeling and grading;

[0191] On the brake test bench, conduct simulation tests on various types of micro-strokes, measure the brake energy consumption and braking force change of the brake under this type of working condition, in order to fit the long-term wear model parameters α1, α2. The long-term wear degradation model is:

[0192]

[0193] where D long is the long-term wear degradation degree. N is the total number of braking times. E j is the energy consumed in the j-th braking (obtained from the test working condition). α1 represents the basic wear generated by each braking (related to the number of braking times). α2 represents the influence degree of the energy consumed during each braking on the wear. α1, α2 are obtained by fitting experimental data.

[0194] Then grade the long-term wear:

[0195]

[0196] where d L2 , d L3 , d L4 are the long-term wear thresholds determined according to experience and tests.

[0197] S9. Short-term thermal degradation risk test modeling, probability calculation, and grading;

[0198] On the brake test bench, simulate experiments for various micro-strokes, measure the temperature rise of the brake under such working conditions, and fit the short-term fade parameters. According to the rapid continuous extreme braking test, fit the short-term thermal fade parameters β1, β2 and the temperature mapping parameter γ. And obtain the micro-stroke transition probability matrix P based on the vehicle working condition data:

[0199] P ij = P(C t+1 = j|C t = i), i, j = 1, …, K;

[0200] Obtain the extreme braking class set according to the test results And calculate the probability that at least r extreme classes occur in the next h micro-strokes:

[0201]

[0202] where h is the short-term time window length (number of micro-strokes). r is the minimum number of times the extreme class appears within the short-term window. is the indicator function.

[0203] Establish a short-term thermal fade model and calculate the short-term thermal fade degree D temp :

[0204] D temp = β1(T brake - T0)+β2(T brake - T0) 2 ;

[0205] where T brake = T0+γP extreme . T brake is the brake temperature. T0 is the reference ambient temperature. β1, β2, γ are the bench test fitting parameters.

[0206] Finally, obtain the short-term thermal fade risk level:

[0207]

[0208] d T2 、d T3 、d T4 are the short-term thermal fade classification thresholds, and ρ3, ρ4 are the thresholds matching the extreme probability. The long-term wear threshold and the short-term thermal fade threshold are different evaluation indicators. When judging whether the brake needs maintenance, the situations of long-term wear and short-term thermal fade should be considered simultaneously and evaluated according to different indicators respectively.

[0209] S10, maintenance and replacement strategies;

[0210] Comprehensive judgment based on long-term wear level and short-term heat fade level:

[0211] If the long-term wear level is relatively high (severe wear, extreme wear), it is recommended to immediately replace or deeply overhaul the brake. If the short-term heat fade risk level is high or extremely high, it is recommended to replace the brake with better high-temperature resistance and strengthen the maintenance of the brake cooling system.

[0212] If both the long-term wear level and the short-term heat fade risk level are normal, the maintenance and replacement cycle can be appropriately extended to reduce the use cost.

[0213] According to the actual use situation of the vehicle, 20 possible driving conditions are simulated. 20 groups of brand-new brakes are used to conduct experiments under 20 groups of working conditions with different intensities. After each group of brakes completes the preset random working condition test, the researchers disassemble, inspect and measure them, and mark the wear condition of the brakes in combination with the test data. Among them, the marking of high-wear brakes is based on the results of manual inspection, including the observation and measurement of the actual wear degree of components such as friction plates and brake discs. Then, high-wear brakes are marked based on manual inspection, and the consistency between the brake fade prediction effect based on the method of the present invention and the results of manual inspection marking is compared. As Figure 1 shown, most of the normal brakes represented by dots are below the threshold d L3 and to the left of the threshold d T3 , and all the high-wear brakes marked manually represented by triangles are above the threshold d L3 . Specifically analyzed, 100% of the high-wear brakes marked manually are predicted by the method of the present invention as "long-term severe wear" or "long-term severe wear + high short-term heat fade risk", indicating that the detection rate of high-wear brakes by this method is 100%. Among the brakes predicted by this method as "long-term severe wear" or "long-term severe wear + high short-term heat fade risk", 78% are marked as high-wear brakes by manual inspection, indicating that the misjudgment rate of misjudging normal brakes as high-wear brakes can be controlled to 22%. The experiment shows that the prediction result of the method of the present invention has good consistency with the manual inspection result, the method has a high accuracy in predicting high-wear brakes, and it avoids missing the detection of high-wear brakes.

[0214] Funding statement: This patent is funded by the Shandong Natural Science Foundation (ZR2023QE208), the China Postdoctoral Science Foundation (2024M751577) and the Key Research and Development Plan of Shandong Province (2022CXGC020302).

[0215] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the technical solution of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting vehicle braking performance degradation considering operating condition density, characterized in that: The following steps are involved: S1. Collect data on the operating conditions of the vehicle; S2, single-dimensional local outlier factor LOF outlier detection; S3, single-dimensional Gaussian weighted moving average filtering; S4, dividing micro-strokes; S5, construct feature vector; S6, deep feature extraction of working conditions using autoencoder dimensionality reduction; S7, clustering in reduced dimensionality space; S8, long-term wear decay test modeling and grading; The long-term wear and tear model is: Among them, D long is the degree of long-term wear and tear, N is the total number of braking times, E j is the energy consumed in the jth braking, α1 represents the basic wear caused by each braking, and α2 represents the influence of the energy consumed in each braking process on the wear; Rating of long-term wear: Among them, d L2 d L3 d L4 is the long-term wear threshold determined based on experience and testing; S9, short-term thermal decay risk test modeling, probability calculation, and grading; According to the vehicle operating condition data, the micro-trip transfer probability matrix P is obtained: P ij =P(C t+1 =j∣C t =i),i,j=1,…,K; According to the test results, the extreme braking class set is obtained And calculate the probability that the extreme class appears at least r times in the next h micro-trips: Among them, h is the length of the short-term time window, also known as the number of micro-trips, and r is the minimum number of times that extreme classes appear in the short-term window. is the indicator function; Establish a short-term thermal decay model and calculate the short-term thermal decay degree D temp : D temp =β1(T brake -T0)+β2(T brake -T0) 2 ; Among them, T brake =T0+γP extreme , T brake is the brake temperature, T0 is the reference ambient temperature, β1, β2, γ are bench test fitting parameters; Among them, d T2 d T3 d T4 is the short-term thermal recession classification threshold, and ρ3 and ρ4 are thresholds that match the extreme probabilities.

2. The method for predicting vehicle braking performance degradation considering operating condition density according to claim 1, characterized in that: In step S1, long-term collection of operating data is performed to obtain X(t)=[x1(t),x2(t),…,x i (t)],t=1,…,T; Among them, x i (t) represents the i-th dimension value of the operating condition parameter collected at time t.

3. The method for predicting vehicle braking performance degradation considering operating condition density according to claim 2, characterized in that: In step S2, for each dimension x i (t) LOF detection is performed separately, the specific contents are as follows: (1) Define the distance function: d(x i (t),x i (t′))=|x i (t)-x i (t′)|; Among them, x i (t′) represents the value of the i-th operating condition parameter at time t′; (2) Define the k-distance function: k-dist(x i (t))=min{r∣|{x i (t′):d(x i (t),x i (t′))≤r}|≥k}; k-dist(x i (t)) means that from x i Starting from point (t), find a distance r such that this distance contains at least k data points; (3) Define the reachable distance function: reach-dist k (x i (t),x i (t′))=max{k-dist(x i (t′)),d(x i (t),x i (t′))}; (4) Define the local reachability density function; Among them, N k (x i (t)) is x i The set of k nearest neighbors of (t); (5) Define the local outlier factor LOF function: If LOF k (x i (t))>γ i , then x i (t) is the outlier value of this dimension, γ i This is an artificially set threshold.

4. The method for predicting vehicle braking performance degradation considering operating condition density according to claim 3, characterized in that: In step S3, the outlier points Replace with a filter, set the window length W, Define the weights: Among them, σ is the standard deviation of Gaussian filtering, w(τ) represents the weight of the data point at distance τ, and M represents the half-width of the window; Filter output: use Replace outliers Represents the data point at the current moment The data before and after the moment, τ is the offset, which means the data point offset by τ before and after the current moment.

5. The method for predicting vehicle braking performance degradation considering operating condition density according to claim 4, characterized in that: In step S4, according to the longitudinal v(t)=x1(t), where v(t) is the vehicle speed (m / s), the operating condition data is divided into micro-trips: when the speed increases from 0 and finally returns to 0 again, the sequence is a micro-trip S i : S={S1,S2,…,S M }; Among them, S i is the i-th micro-trip data fragment, M is the total number of micro-trips, and S represents the set of all micro-trips.

6. The method for predicting vehicle braking performance degradation considering operating condition density according to claim 5, characterized in that: In step S5, for each micro-stroke S i , extract a set of statistical features from the original time series data to form a feature vector F i ; Set micro stroke S i Contains time series Where: v(t) is the vehicle speed (m / s), a(t) is the vehicle acceleration (m / s 2 ), m(t) is the load (kg), θ(t) is the slope (°); Define the micro stroke length as |S i |, i.e. micro stroke S i Contains|S i | discrete time points, the following features are calculated for the data in this set; (1) Average speed: (2) Maximum speed: (3) Minimum speed: (4) Speed ​​variance: (5) Average acceleration: (6) Acceleration variance: (7) Maximum acceleration: (8)Minimum acceleration: (9) Percentage of rapid deceleration: Define rapid deceleration as acceleration a(t) < -3m / s 2 The proportion of time in, is an indicator function, which is 1 when the condition is met and 0 otherwise; (10) Average slope: (11) Average load: Combine these features into a vector: F i Contains the statistical indicators of various important data of the vehicle in the i-th micro-trip.

7. The method for predicting vehicle braking performance degradation considering operating condition density according to claim 6, characterized in that: In step S6, the feature vector d = 11, which is the number of feature dimensions, can introduce the autoencoder pair {F i } Perform dimensionality reduction and deep feature extraction. The autoencoder represents high-dimensional input data through a low-dimensional latent variable space and is trained under reconstruction error; Assume the encoder mapping is: z i =f enc (F i ;θ enc ); in, And l <d,θ enc is the encoder parameter; The decoder mapping is: in, is the reconstructed feature vector, θ dec is the decoder parameter; The training objective function of the autoencoder can be defined as minimizing the mean square error: After training, the encoder output z i That is, the deep feature representation after dimensionality reduction, which captures the more essential potential structure in the original features.

Images