A vehicle speed prediction method based on K-means clustering and wavelet decomposition

Abstract

The invention discloses a vehicle speed prediction method based on K-means clustering and wavelet decomposition, which collects historical vehicle driving data and establishes the original vehicle speed sequence; divides the original vehicle speed sequence into several working condition blocks of equal time length, and classifies each working condition block based on the K-means clustering algorithm; uses wavelet decomposition to decompose all working condition blocks into Y subsequences; uses RBF neural network to train each subsequence to obtain a subsequence prediction model; uses the K-means clustering algorithm to judge the category to which the current working condition block belongs and determines the corresponding vehicle speed prediction model; decomposes the current working condition block, and uses the corresponding subsequence vehicle speed prediction model to predict the subsequence vehicle speed obtained by decomposing the current working condition block; accumulates the prediction results to obtain the final vehicle speed prediction result. The method can reduce the influence of the non-stationarity of vehicle speed and the reduction of the generalization ability of a single prediction model on the vehicle speed prediction accuracy, thereby improving the vehicle speed prediction accuracy.

Description

Technical Field

[0001] The invention relates to an improvement of vehicle speed prediction technology, and in particular to a vehicle speed prediction method based on K-means clustering and wavelet decomposition, and belongs to the technical field of vehicle speed prediction and neural network learning. Background Art

[0002] The speed prediction accuracy plays an important role in the energy saving effect of the predictive energy management strategy of hybrid electric vehicles. Among the current speed prediction methods, the data-driven neural network prediction model has shown certain advantages in prediction accuracy.

[0003] At present, vehicle speed prediction mainly uses a single prediction model to train the collected vehicle speed data, relying on the generalization ability of the neural network model to predict the changes in vehicle speed in the short-term time domain in the future. Although the single prediction model has achieved certain prediction results, there is still room for improvement. In addition, the single prediction model also ignores the non-stationarity of vehicle speed, which further reduces the prediction accuracy. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to propose a vehicle speed prediction method based on K-means clustering and wavelet decomposition, which can reduce the impact of the non-stationarity of vehicle speed and the reduced generalization ability of a single prediction model on the vehicle speed prediction accuracy, thereby improving the vehicle speed prediction accuracy.

[0005] The technical solution of the present invention is achieved in this way:

[0006] A vehicle speed prediction method based on K-means clustering and wavelet decomposition, the steps are as follows:

[0007] 1) Collect historical vehicle speed data and establish the original vehicle speed sequence based on the time axis;

[0008] 2) Divide the original vehicle speed sequence into several operating condition blocks of equal length, and cluster all operating condition blocks into X categories based on the K-means clustering algorithm;

[0009] 3) Using wavelet decomposition to decompose the working condition blocks under each category, and decompose each working condition block into Y subsequences;

[0010] 4) The Y subsequences obtained by decomposing each working condition block under the same category are constructed according to the corresponding relationship of the subsequences to obtain Y data sets, and X independent categories have a total of X*Y data sets. The Y data sets constructed in each category are trained using the RBF neural network to obtain the vehicle speed prediction model corresponding to each subsequence under each category;

[0011] 5) constructing the current working condition according to the working condition block duration rule of step 2) to obtain the current working condition block; using the K-means clustering algorithm to determine the category to which the current working condition block belongs, thereby determining all subsequence vehicle speed prediction models under the corresponding category;

[0012] 6) Decomposing the current operating condition block by using the wavelet decomposition to obtain Y subsequences under the current operating condition block; predicting the subsequence speed obtained by decomposing the current operating condition block by using the subsequence speed prediction model corresponding to the subsequence determined in step 5) to obtain Y subsequence speed prediction results corresponding to the number of subsequences;

[0013] 7) Accumulate the Y subsequence vehicle speed prediction results obtained in step 6) to obtain the final vehicle speed prediction result.

[0014] Among them, the division of the operating condition blocks of equal duration in step 2) is to push the moment corresponding to each vehicle speed in the original vehicle speed sequence forward by the same duration, and all the vehicle speed data within this duration constitutes a operating condition block.

[0015] Preferably, the historical vehicle speed data is directly extracted from an existing database, and the operating conditions corresponding to the historical vehicle speed data include five standard driving conditions: China_urban, HWFET, LA92, UDDS and WLTC.

[0016] The calculation of the RBF neural network in step 4 is defined as follows:

[0017]

[0018] Among them, y kj is the output corresponding to the kth input; i = 1, 2, ..., m, m is the number of hidden layer nodes; w ij is the weight from the hidden layer to the output layer; x k is the kth input of the neural network; c i is the center of the radial basis function; σ i is the standard deviation of the Gaussian function; ||x k -c i || is the Gaussian norm.

[0019] When the step 4) uses the RBF neural network to train the Y data sets constructed in each category,

[0020] Define the subsequence speed in the data set as the RBF neural network input, the future subsequence speed of the vehicle as the expected output, and the training target as the mean square error between the actual subsequence speed of the vehicle and the expected output speed. The calculation formula is as follows:

[0021]

[0022] Where, MSE is the mean square error of the entire working condition; p is the prediction time; v r (j+i) is the predicted vehicle speed at time j+i in the prediction time domain at time j; v(j+i) is the actual vehicle speed at time j+i in the prediction time domain at time j; n is the total length of the working condition;

[0023] When the MSE value reaches a preset training target or reaches a maximum number of training times, the training is stopped to obtain a vehicle speed prediction model corresponding to each subsequence under each category.

[0024] Specifically, the operating condition block includes vehicle speeds at 5 sampling points, wherein the sampling period is 1 s.

[0025] In step 2) of the present invention, it is assumed that there are three cluster centers determined by the K-means clustering algorithm, namely c1=[c 11 ,c 12 ,…,c 1m ],c2=[c 21 ,c 22 ,…,c 2m ],c3=[c 31 ,c 32 ,…,c 3m ]; the corresponding categories are three, that is, X=3.

[0026] In step 5), the category of the current working condition block is determined as follows:

[0027] During the actual driving process of the vehicle, at the current sampling time, the operating condition characteristic parameters [x1, x2, …, x5] of the past 5 seconds are calculated to form the current operating condition block, and the distance from the current operating condition block to each cluster center is calculated according to the following formula:

[0028]

[0029] Among them, j = 1, 2, 3, corresponding to three cluster centers, that is, three independent categories; the current operating condition block has the smallest center distance with which cluster center, and the current operating condition block belongs to which category, and the subsequence vehicle speed prediction model under the corresponding category is selected to predict the vehicle speed.

[0030] In step 3), all the operating condition blocks in each category are decomposed into three layers by wavelet decomposition, and each operating condition block obtains four subsequences, that is, Y=4.

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

[0032] 1. The present invention adopts K-means clustering algorithm to classify working conditions, which reduces the mapping range of a single neural network, improves the prediction ability of a single neural network, and effectively improves the vehicle speed prediction accuracy.

[0033] 2. The present invention further uses wavelet decomposition to decompose the historical operating conditions into a plurality of relatively stable sub-vehicle speed sequences, thereby reducing the impact of the non-stationarity of the vehicle speed on the vehicle speed prediction result and further improving the vehicle speed prediction accuracy.

[0034] 3. The present invention is mainly used in energy management strategies based on model predictive control. It uses K-means clustering and wavelet decomposition to predict vehicle speed in real time, effectively improving the speed prediction accuracy and thus improving the fuel economy of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The figure is a flow chart of the vehicle speed prediction method of the present invention.

[0036] Figure 2 It is a schematic diagram of historical operating condition data composed of five standard operating conditions of the present invention.

[0037] Figure 3 This is the classification result of the original vehicle speed sequence after K-means clustering.

[0038] Figure 4 This is a diagram of each vehicle speed subsequence after the original vehicle speed sequence is decomposed by wavelet.

[0039] Figure 5 This is a schematic diagram of RBF neural network prediction of the present invention.

[0040] Figure 6 Schematic diagram of the test working condition of the present invention.

[0041] Figure 7 Schematic diagram of vehicle speed prediction results of two prediction methods of the present invention. DETAILED DESCRIPTION

[0042] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] See also Figure 1 The present invention provides a vehicle speed prediction method based on K-means clustering and wavelet decomposition, the steps are as follows:

[0044] 1) Collect the historical speed data of the vehicle and establish the original speed sequence of the vehicle based on the time axis. To facilitate data acquisition, the historical speed data of the present invention can be directly extracted from the existing database, which does not affect the quality of data acquisition. Of course, the historical speed data of each vehicle can also be actually collected. The operating conditions corresponding to the historical speed data include five standard driving conditions: China_urban, HWFET, LA92, UDDS and WLTC. Figure 2 It is a time series diagram of the original vehicle speed in an embodiment of the present invention.

[0045] 2) Divide the original vehicle speed sequence into several operating condition blocks of equal duration, and cluster all operating condition blocks into X categories based on the K-means clustering algorithm. Step 2) The equal duration operating condition block division is to push the moment corresponding to each vehicle speed in the original vehicle speed sequence forward by the same duration, and all vehicle speed data within this duration is a operating condition block. In the embodiment, a period of vehicle operation between the current moment and the past 4s is defined as a operating condition block, that is, the operating condition block includes the vehicle speed at 5 sampling points, where the sampling period is 1s.

[0046] For the K-means clustering algorithm, the K-means function provided by MATLAB is used. Its calling format is as follows:

[0047] center = kmeans(x,k)

[0048] Among them, center is the cluster center; x is the input quantity, which is represented by the dimension of the feature parameter; k is the number of clusters, and k is set to 3.

[0049] Assume that the number of cluster centers determined by the K-means clustering algorithm is three, namely c1=[c 11 ,c 12 ,…,c 1m ],c2=[c 21 ,c 22 ,…,c 2m ],c3=[c 31 ,c 32 ,…,c 3m ]; the corresponding categories are three, that is, X=3.

[0050] Figure 3 That is to Figure 2 The classification result of the original vehicle speed time series.

[0051] 3) Wavelet decomposition is used to decompose the working condition blocks under each category, and each working condition block is decomposed into Y subsequences; in the embodiment, the number of wavelet decomposition layers is 3, and each working condition block obtains four subsequences, which are set as D1, D2, D3, and A3, that is, Y=4.

[0052] Figure 4 That is to Figure 3 The result of wavelet decomposition of the working condition in .

[0053] 4) Y subsequences obtained by decomposing each working condition block under the same category are constructed according to the corresponding relationship of the subsequences to obtain Y data sets, that is, the subsequences D1 of all working condition blocks under category I constitute one data set, and the subsequences D2 of all working condition blocks under category I constitute one data set. Thus, category I has a total of four data sets, and the three categories in the embodiment have a total of 3*4=12 data sets. The RBF neural network is used to train the four data sets constructed in each category respectively, and the four subsequence vehicle speed prediction models corresponding to each subsequence under each category are obtained;

[0054] The calculation definition of the RBF neural network in step 4 is as follows:

[0055]

[0056] Among them, y kj is the output corresponding to the kth input; i = 1, 2, ..., m, m is the number of hidden layer nodes; w ij is the weight from the hidden layer to the output layer; x k is the kth input of the neural network; c i is the center of the radial basis function; σ i is the standard deviation of the Gaussian function; ||x k -c i || is the Gaussian norm.

[0057] Figure 5 This is the schematic diagram of the RBF neural network prediction model in step 4).

[0058] For RBF neural network, MATLAB's RBF neural network toolbox is used, and its calling format is as follows:

[0059] net=newrb(P,T,goal,spread,MN,DF)

[0060] Among them, P is the input matrix; T is the target matrix; goal is the mean square error; spread is the expansion coefficient of the basis function; MN is the maximum number of neurons; DF is the number of neurons added between two displays during training.

[0061] In step 4 of the present invention, when the Y data sets constructed in each category are trained using the RBF neural network, the subsequence speed in the data set is defined as the RBF neural network input, and the future subsequence speed of the vehicle is defined as the expected output. The training target is the mean square error between the actual subsequence speed of the vehicle and the expected output speed, and the calculation formula is as follows:

[0062]

[0063] Where, MSE is the mean square error of the entire working condition; p is the prediction time; v r(j+i) is the predicted vehicle speed at time j+i in the prediction time domain at time j; v(j+i) is the actual vehicle speed at time j+i in the prediction time domain at time j; n is the total length of the working condition;

[0064] When the MSE value reaches a preset training target or reaches a maximum number of training times, the training is stopped to obtain Y subsequence vehicle speed prediction models corresponding to each subsequence under each category.

[0065] Reference Figure 6 ,In this embodiment, CLTCP standard working condition is used as the ,test condition, and MATLAB is used as the simulation platform to ,verify the effectiveness of this method.

[0066] 5) constructing the current working condition according to the working condition block duration rule of step 2) to obtain the current working condition block; using the K-means clustering algorithm to determine the category to which the current working condition block belongs, once the category is determined, the corresponding subsequence vehicle speed prediction model is determined;

[0067] In step 5), the category of the current working condition block is determined as follows:

[0068] During the actual driving process of the vehicle, at the current sampling time, the operating condition characteristic parameters [x1, x2, …, x5] of the past 5 seconds are calculated to form the current operating condition block, and the distance from the current operating condition block to each cluster center is calculated according to the following formula:

[0069]

[0070] Among them, j = 1, 2, 3, corresponding to three cluster centers, that is, three categories; the current operating condition block has the smallest center distance with which cluster center, which category the current operating condition block belongs to, and the subsequence vehicle speed prediction model under the corresponding category is selected to predict the vehicle speed.

[0071] 6) Decomposing the current operating condition block by using the wavelet decomposition to obtain Y (i.e., four) subsequences under the current operating condition block; each subsequence uses the subsequence vehicle speed prediction model corresponding to the subsequence determined in step 5) to predict the subsequence vehicle speed obtained by decomposing the current operating condition block, and obtains Y subsequence vehicle speed prediction results corresponding to the number of subsequences;

[0072] 7) Accumulate the Y subsequence vehicle speed prediction results obtained in step 6) to obtain the final vehicle speed prediction result.

[0073] Figure 7 The figure shows the vehicle speed prediction effect of the present invention and the vehicle speed prediction result of the traditional single RBF prediction model.

[0074] In order to quantitatively evaluate the effect of the vehicle speed prediction method proposed in the present invention, the root mean square error RMSE is used for quantitative evaluation, and its calculation formula is as follows:

[0075]

[0076] Table 1: Vehicle speed prediction result data table of the traditional method and the method of the present invention.

[0077]

[0078] pass Figure 7 It can be seen from Table 1 that, based on the given test conditions, the RMSE of the vehicle speed prediction result of the present invention is improved by 59.05% over the entire test conditions compared with the single RBF prediction method.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the applicant has described the present invention in detail with reference to the preferred embodiments, ordinary technicians in the field should understand that modifications or equivalent substitutions of the technical solution of the present invention without departing from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A vehicle speed prediction method based on K-means clustering and wavelet decomposition, characterized by: Here are the steps: 1) Collect historical vehicle speed data and establish the original vehicle speed sequence based on the time axis; 2) Divide the original vehicle speed sequence into several operating condition blocks of equal length, and cluster all operating condition blocks into X categories based on the K-means clustering algorithm; 3) Using wavelet decomposition to decompose the working condition blocks under each category, and decompose each working condition block into Y subsequences; 4) The Y subsequences obtained by decomposing each working condition block under the same category are constructed according to the corresponding relationship of the subsequences to obtain Y data sets, and X independent categories have a total of X*Y data sets. The Y data sets constructed in each category are trained using the RBF neural network to obtain the vehicle speed prediction model corresponding to each subsequence under each category; 5) constructing the current working condition according to the working condition block duration rule of step 2) to obtain the current working condition block; using the K-means clustering algorithm to determine the category to which the current working condition block belongs, thereby determining all subsequence vehicle speed prediction models under the corresponding category; 6) Decomposing the current operating condition block by using the wavelet decomposition to obtain Y subsequences under the current operating condition block; predicting the subsequence speed obtained by decomposing the current operating condition block by using the subsequence speed prediction model corresponding to the subsequence determined in step 5) to obtain Y subsequence speed prediction results corresponding to the number of subsequences; 7) Accumulate the Y subsequence vehicle speed prediction results obtained in step 6) to obtain the final vehicle speed prediction result.

2. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1 is characterized in that: Step 2) The division of the operating condition blocks of equal duration is to push the moment corresponding to each vehicle speed in the original vehicle speed sequence forward by the same duration, and all the vehicle speed data within the duration constitutes a operating condition block.

3. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1 is characterized in that: The historical driving speed data is directly extracted from an existing database, and the operating conditions corresponding to the historical driving speed data include five standard driving conditions: China_urban, HWFET, LA92, UDDS and WLTC.

4. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1 is characterized in that: The calculation definition of the RBF neural network in step 4 is as follows: Among them, y kj is the output corresponding to the kth input; i = 1, 2, ..., m, m is the number of hidden layer nodes; w ij is the weight from the hidden layer to the output layer; x k is the kth input of the neural network; c i is the center of the radial basis function; σ i is the standard deviation of the Gaussian function; ||x k -c i || is the Gaussian norm.

5. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1 is characterized in that: When the step 4) uses the RBF neural network to train the Y data sets constructed in each category, Define the subsequence speed in the data set as the RBF neural network input, the future subsequence speed of the vehicle as the expected output, and the training target as the mean square error between the actual subsequence speed of the vehicle and the expected output speed. The calculation formula is as follows: Where, MSE is the mean square error of the entire working condition; p is the prediction time; v r (j+i) is the predicted vehicle speed at time j+i in the prediction time domain at time j; v(j+i) is the actual vehicle speed at time j+i in the prediction time domain at time j; n is the total length of the working condition; When the MSE value reaches a preset training target or reaches a maximum number of training times, the training is stopped to obtain a vehicle speed prediction model corresponding to each subsequence under each category.

6. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1 is characterized in that: The operating condition block includes vehicle speeds at 5 sampling points, wherein the sampling period is 1 s.

7. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1 is characterized in that: In step 2), assume that the number of cluster centers determined by the K-means clustering algorithm is three, namely c1 = [c 11 ,c 12 ,…,c 1m ],c2=[c 21 ,c 22 ,…,c 2m ],c3=[c 31 ,c 32 ,…,c 3m ]; the corresponding categories are three, that is, X=3.

8. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 7 is characterized in that: In step 5), the category of the current working condition block is determined as follows: During the actual driving process of the vehicle, at the current sampling time, the operating condition characteristic parameters [x1, x2, …, x5] of the past 5 seconds are calculated to form the current operating condition block, and the distance from the current operating condition block to each cluster center is calculated according to the following formula: Among them, j = 1, 2, 3, corresponding to three cluster centers, that is, three independent categories; the current operating condition block has the smallest center distance with which cluster center, and the current operating condition block belongs to which category, and the subsequence vehicle speed prediction model under the corresponding category is selected to predict the vehicle speed.

9. The vehicle speed prediction method based on K-means clustering and wavelet decomposition according to claim 1, characterized in that: In step 3), all the operating condition blocks in each category are decomposed into three layers by wavelet decomposition, and each operating condition block obtains four subsequences, that is, Y=4.

Images