A vehicle speed prediction method based on error synchronous prediction

Abstract

The present application relates to the technical field of vehicle speed prediction, and particularly relates to a vehicle speed prediction method based on error synchronous prediction. The present application firstly establishes a vehicle speed prediction model based on multiple standard working conditions; then, by using the prediction error obtained by the vehicle speed prediction model, error prediction models based on RBF neural network and Markov chain model are respectively established to synchronously predict the vehicle speed prediction error, and then correct the vehicle speed prediction result. The vehicle speed prediction method of the present application considers the prediction error and is more accurate. The present application uses RBF neural network and Markov chain model to synchronously predict the prediction error and correct the vehicle speed prediction result, thereby effectively improving the vehicle speed prediction accuracy. By synchronously predicting the vehicle speed prediction error and correcting the vehicle speed prediction result, the vehicle speed prediction accuracy can be effectively improved, and the fuel economy of the vehicle is improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle speed prediction technology, and in particular to a vehicle speed prediction method based on error synchronization prediction. Background Technology

[0002] Accuracy in vehicle speed prediction plays a crucial role in the energy-saving effect of predictive energy management strategies for hybrid electric vehicles.

[0003] Among current vehicle speed prediction methods, data-driven neural network prediction models have shown certain advantages in prediction accuracy. Relying on the generalization ability of neural network models to predict vehicle speeds in the short term has achieved some predictive results. However, prediction errors are inevitable in vehicle speed prediction, directly leading to a decrease in prediction accuracy. Accurately predicting these errors would be beneficial for obtaining better vehicle speed prediction results.

[0004] In order to predict errors and correct the accuracy of predicted vehicle speed to obtain better vehicle speed prediction results, we propose a vehicle speed prediction method based on synchronous error prediction according to the present invention. Summary of the Invention

[0005] The purpose of this invention is to provide a vehicle speed prediction method based on error synchronization prediction, which addresses the problem that current technologies inevitably generate certain prediction errors when predicting vehicle speed, leading to a decrease in prediction accuracy.

[0006] This method predicts both vehicle speed and prediction error simultaneously, feeding the error prediction results back to the vehicle speed prediction results, thereby improving the accuracy of vehicle speed prediction.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A vehicle speed prediction method based on error synchronization prediction, comprising the following steps:

[0009] S1. Collect historical vehicle speed data and establish the original vehicle speed sequence based on the time axis;

[0010] S2. According to different standard driving conditions, the original vehicle speed sequence is divided into dataset A, dataset B and dataset C;

[0011] S3. Based on dataset A, a single-step vehicle speed prediction model is established using an RBF neural network;

[0012] S4. Use a single-step rolling strategy to perform multi-step vehicle speed prediction on dataset B, and calculate the multi-step prediction error of dataset B based on the multi-step vehicle speed prediction results of dataset B.

[0013] S5. Based on the one-step lead error data, establish a one-step lead RBF neural network prediction error model;

[0014] S6. Based on the entire error dataset, establish a Markov chain prediction error model from the first two steps to the last step;

[0015] S7. Subtract the error correction result from the multi-step prediction results to obtain the final vehicle speed prediction result.

[0016] Furthermore, in S3, the RBF neural network is defined as follows:

[0017]

[0018] Among them, y kj The output corresponds to the k-th input; i = 1, 2, ..., m, where m is the number of hidden layer nodes; w ij x represents the weights from the hidden layer to the output layer. k c is the k-th input to the neural network. i σ is the center of the radial basis functions; i Let |x| be the standard deviation of the Gaussian function; k -c i || represents the Gaussian norm;

[0019] The input to the single-step vehicle speed prediction model established by the RBF neural network is the vehicle speed over a past period of time. Specifically, it represents the vehicle speed at the current moment and the vehicle speed at the past 5 sampling moments. Its output is the vehicle speed at the next sampling point, and its sampling period is 1 second.

[0020] Furthermore, in S3, an RBF neural network is used to establish a single-step vehicle speed prediction model. When training on dataset A, the vehicle speeds in dataset A over a past period are defined as the input to the RBF neural network, and the future vehicle speed is defined as the desired output. The training objective is the mean square error between the actual vehicle speed and the desired future vehicle speed, calculated as follows.

[0021]

[0022] In the formula, MSE is the mean square error for the entire operating condition; Predict the vehicle speed at time j; v j Let j be the actual vehicle speed at time j; n be the total duration of dataset A.

[0023] Training stops when the MSE value reaches the preset training target or the maximum number of training iterations, resulting in a single-step vehicle speed prediction model.

[0024] Furthermore, in S4, the single-step rolling prediction strategy is expressed as follows:

[0025]

[0026] Where t is the current time, is the predicted value at time t+h; h is the prediction step size, with a maximum value of 4.

[0027] The vehicle speed prediction result in the prediction time domain at the current time t can be obtained by using a single-step rolling prediction strategy.

[0028]

[0029] Multi-step prediction of vehicle speed in dataset B The result can be expressed as:

[0030]

[0031] In the predicted time domain p, the actual vehicle speed can be expressed as:

[0032]

[0033] Where p is the length of the vehicle speed prediction time domain, p = 4; m is the length of dataset B;

[0034] The prediction error e of dataset B B It can be represented as:

[0035]

[0036] Furthermore, in S5, the RBF neural network prediction error model with a one-step lead is denoted as f. e1 The training set is the error dataset [e] 11 ,e 21 ,…,e m1 The input is the vehicle speed error at the current time and the past 5 sampling times, denoted as [e]. t ,e t-1 ,e t-2 ,…,e t-5 The output is the vehicle speed prediction error for the next moment, denoted as . The expression is as follows:

[0037]

[0038] Furthermore, in S6, there are three Markov chain prediction error models that are two steps ahead to the last step:

[0039] Error prediction model f from 1 step ahead to 2 steps ahead e12 Error prediction model f from 2 steps ahead to 3 steps ahead e23 And the error prediction model f from 3 steps ahead to 4 steps ahead e34 ;

[0040] The error prediction process for each error model mainly includes the following steps: state space partitioning, calculation of the state transition probability matrix, and state transition.

[0041] Furthermore, f e12 The specific process of state space partitioning is as follows:

[0042] Let the error at the step ahead be the current state, the error at the step ahead of 2 steps be the state at the next time step, and the error at the step ahead of 1 step be the prediction error e of dataset B. B [e] 11 ,e 21 ,…,e m1 The error two steps ahead is the prediction error e of dataset B. B [e] 12 ,e 22 ,…,e m2 ];

[0043] Calculate the maximum and minimum values ​​of the 1-step and 2-step lead error sequences and round them down. These four values ​​are denoted as e. 1_max e 2_max e 1_min and e 2_min The calculation expression is as follows:

[0044]

[0045]

[0046]

[0047]

[0048] Using a step size of 0.5 km / h, the state space is divided into two steps ahead, and the error state set S is obtained. e1 and S e2 as follows:

[0049]

[0050] Let set S e1 and S e2 The number of states are N and M respectively;

[0051] f e12 The specific process for calculating the state transition probability is as follows:

[0052] For each error value that is 1 step ahead or 2 steps ahead, the corresponding state is determined by its relationship with the state values ​​in the state space, and this state is denoted as... and statistics From the i-th state to The frequency of the j-th state, divided by From the i-th state to The frequency of transitions in all states is obtained. The state transition probability P of the i-th state i,j The calculation expression is as follows:

[0053]

[0054] Where i∈[1,N], j∈[1,M];

[0055] Traversal For each state in the matrix, calculate the state transition probability for each state, and finally obtain the state transition probability matrix, as follows:

[0056]

[0057] f e12 The specific process of state transition is as follows:

[0058] Assuming the error one step ahead is e1, then the state space partitioning and state transition probability matrix calculation can determine the current state. a certain state Then find The state value corresponding to the highest transition probability. and The next state value The transition value corresponding to e1 for:

[0059]

[0060] Furthermore, the error e1 of the one-step advance comes from the output of the one-step advance RBF neural network prediction error model. The state transition assumption assumes an error of e1 one step ahead, which is equivalent to...

[0061] Furthermore, by replacing the error data, f can be established. e23 and f e34 Error prediction model and error prediction for 3-step and 4-step lead time and Make predictions;

[0062] Based on the established error prediction model, the error in the prediction time domain at time t can be predicted, and its expression is as follows:

[0063]

[0064] Finally, the prediction error in the prediction time domain at time t is obtained.

[0065] Furthermore, in S7, the final predicted vehicle speed in the time domain at the current moment is expressed as follows:

[0066]

[0067] The present invention has at least the following beneficial effects:

[0068] 1. This invention uses RBF neural network and Markov chain model to simultaneously predict prediction errors, corrects vehicle speed prediction results, and effectively improves vehicle speed prediction accuracy.

[0069] 2. This invention is mainly applied to energy management strategies based on model predictive control. By synchronously predicting vehicle speed prediction errors and correcting the vehicle speed prediction results, the accuracy of vehicle speed prediction can be effectively improved, thereby improving the fuel economy of the vehicle. Attached Figure Description

[0070] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a flowchart of the vehicle speed prediction method of the present invention.

[0072] Figure 2 This is a schematic diagram of the operating conditions of datasets A, B, and C in this invention.

[0073] Figure 3 This is a schematic diagram of the single-step vehicle speed prediction using the RBF neural network of the present invention.

[0074] Figure 4 This is a schematic diagram of the single-step rolling prediction strategy of the present invention.

[0075] Figure 5 This is a schematic diagram of single-step error prediction using the RBF neural network of the present invention.

[0076] Figure 6 This is the Markov state transition probability matrix from 1-step lead error to 2-step lead error in this invention.

[0077] Figure 7 This is the Markov state transition probability matrix from a 2-step lead error to a 3-step lead error in this invention.

[0078] Figure 8 This is the Markov state transition probability matrix for the present invention, which represents the error range from 3 steps ahead to 4 steps ahead.

[0079] Figure 9 This diagram illustrates the prediction results of two vehicle speed prediction methods with and without error compensation according to the present invention. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0081] See Figure 1 The present invention provides a vehicle speed prediction method based on error synchronization prediction, as follows:

[0082] 1) Collect historical vehicle speed data and establish the original vehicle speed sequence based on the time axis.

[0083] To facilitate data acquisition, the historical vehicle speed data described in this invention can be directly extracted from existing databases without affecting data acquisition quality. Alternatively, historical speed data for each vehicle can also be collected in person.

[0084] The entire historical driving data is divided into three parts: dataset A, dataset B, and dataset C. Dataset A includes five standard driving conditions: China_urban, HWFET, LA92, UDDS, and WLTC. Dataset B includes US06 and CTLCP. Dataset C includes NEDC driving conditions.

[0085] Please refer to the details. Figure 2 This is a time series diagram showing the original vehicle speed data divided into three parts, A, B, and C, in a specific embodiment of the present invention.

[0086] 2) Using dataset A as the training set, establish an RBF neural network single-step vehicle speed prediction model. The RBF neural network calculation is defined as follows:

[0087]

[0088] Among them, y kj The output corresponds to the k-th input; i = 1, 2, ..., m, where m is the number of hidden layer nodes; w ij x represents the weights from the hidden layer to the output layer. k c is the k-th input to the neural network. i σ is the center of the radial basis functions; i Let |x| be the standard deviation of the Gaussian function; k -c i || represents the Gaussian norm.

[0089] The input to the RBF neural network vehicle speed prediction model is the vehicle speed over a past period, specifically represented by the vehicle speed at the current moment and the speed at the past 5 sampling moments, for a total of 6 sampling points, denoted as [v]. t ,v t-1 ,…,v t-5 Its output is the vehicle speed at the next sampling point, denoted as . The sampling period is 1 second.

[0090] Figure 3 This is a schematic diagram of the RBF neural network prediction model in section 2.

[0091] 3) For RBF neural networks, the RBF Neural Network Toolbox in MATLAB is used, and its calling format is as follows:

[0092] net=newrb(P,T,goal,spread,MN,DF)

[0093] Where P is the input matrix; T is the target matrix; goal is the mean squared error; spread is the spread coefficient of the basis function; MN is the maximum number of neurons; and DF is the number of neurons added between two displays during training.

[0094] When training dataset A using an RBF neural network, this invention defines the vehicle speeds over a past period in dataset A as the input to the RBF neural network, the future vehicle speeds as the desired output, and the training objective as the mean square error between the actual vehicle speed and the desired output speed. The calculation formula is as follows:

[0095]

[0096] In the formula, MSE is the mean square error for the entire operating condition; Predict the vehicle speed at time j; vj is the actual vehicle speed at time j; n is the total duration of dataset A;

[0097] When the MSE value reaches the preset training target or the maximum number of training iterations, training stops, and a single-step vehicle speed prediction model is obtained.

[0098] 4) Use a single-step rolling strategy to perform multi-step predictions on dataset B. The expression is as follows:

[0099]

[0100] Where t is the current time, Let be the predicted value at time t+h; h is the prediction step size, with a maximum value of 4. This means that a single-step rolling prediction strategy is used to predict the vehicle speed at the next four sampling times. The single-step rolling prediction strategy can then be used to obtain the vehicle speed prediction result in the prediction time domain at the current time t.

[0101]

[0102] Figure 4 This is a schematic diagram of a single-step scrolling strategy;

[0103] 5) Let the true value of dataset B in the time domain be 4 steps ahead, and the prediction result 4 steps ahead be... The prediction error is e B They can be represented as follows:

[0104]

[0105]

[0106] Where p is the length of the vehicle speed prediction time domain, p = 4; m is the length of dataset B.

[0107]

[0108] 6) Based on the prediction error dataset from the one-step-ahead prediction [e] 11 ,e 21 ,…,e m1 Establish an RBF neural network one-step lead error prediction model f e1 The input is the vehicle speed error at the current time and the past 5 sampling times, denoted as [e]. t ,e t-1 ,…,e t-5 The output is the vehicle speed prediction error for the next moment, denoted as . The relationship is as follows:

[0109]

[0110] Figure 5 This is a schematic diagram of single-step error prediction using the RBF neural network of the present invention;

[0111] 7) For error predictions with 2 steps, 3 steps, and 4 steps ahead, a Markov chain model is used for prediction, denoted as f respectively. e12 f e23 and f e34 The implementation process is the same, mainly including three steps: state space partitioning, state transition probability calculation, and state transition, with f e12 The following explanation uses a prediction model as an example.

[0112] 8) State space partitioning process: using the error [e] of the lead-one step as the basis... 11 ,e 21 ,…,e m1 [e] represents the current state, with an error two steps ahead. 12 ,e22 ,…,e m2 Let ] be the state at the next moment. Calculate the maximum and minimum values ​​under the two state sets, denoted as e. 1_max e 2_max e 1_min and e 2_min Its calculation expression is as follows)

[0113]

[0114]

[0115]

[0116]

[0117] Then, with a step size of 0.5 km / h, the state space of the 1-step and 2-step advance steps is divided, and the error state set S is obtained. e1 and S e2 as follows)

[0118]

[0119] Let set S e1 and S e2 The number of states are N and M respectively;

[0120] 9) Calculation process of state transition probability: For [e 11 ,e 21 ,…,e m1 ] and [e 12 ,e 22 ,…,e m2 Each value of ] is obtained by interacting with S. e1 and S e2 The relationship between the magnitudes of the state values ​​determines the state to which the state belongs, and this relationship is denoted as... and statistics From the i-th state to The frequency of the j-th state, divided by From the i-th state to The frequency of transitions in all states is obtained. The state transition probability P of the i-th state i,j The calculation expression is as follows:

[0121]

[0122] Where i∈[1,N], j∈[1,M];

[0123] 10) Traversal For each state in the matrix, calculate the state transition probability for each state, and finally obtain the state transition probability matrix, as follows:

[0124]

[0125] 11) Specific process of state transition: Based on 8-10, the state transition probability matrices for 3-step and 4-step advance can be obtained. Figure 6 This is the error state probability transition matrix for the ahead of 2, 3, and 4 steps in this invention;

[0126] 12) Using the error prediction value one step ahead For the two-step error prediction model f e12 The input, the prediction error value two steps ahead. Perform calculations, through With S e1 The state values ​​in the data are compared to determine the value. a certain state Then find Downward S e2 The state value corresponding to the highest transition probability in the state transition. and The next state value The corresponding prediction error two steps ahead for:

[0127]

[0128] 13) Same as 12), we can obtain the prediction errors for 3 steps and 4 steps ahead. and

[0129] Figure 6-8 These are the result diagrams of the error transfer probability matrix from 1 step ahead to 2 steps ahead, from 2 steps ahead to 3 steps ahead, and from 3 steps ahead to 4 steps ahead, respectively.

[0130] 14) Based on the established 1-step lead RBF error prediction model and 2-4 step lead Markov chain prediction model, the prediction error in the prediction time domain at time t can be predicted, and its expression is as follows:

[0131]

[0132] Finally, the prediction error in the prediction time domain is obtained.

[0133] 15) The final predicted vehicle speed in the time domain at the current moment. It can be represented as:

[0134] 16)Reference Figure 2 In this embodiment, dataset C is used as the test condition and MATLAB is used as the simulation platform to verify the effectiveness of the method.

[0135] Figure 9 The diagram shows the vehicle speed prediction results with and without error compensation according to the present invention.

[0136] To quantitatively evaluate the effectiveness of the vehicle speed prediction method proposed in this invention, the root mean square error (RMSE) is used for quantitative evaluation, and its calculation formula is as follows:

[0137]

[0138] Table 1: Data table of vehicle speed prediction results of traditional method and method of the present invention.

[0139]

[0140]

[0141] pass Figure 9 As can be seen from Table 1, based on the given test conditions, the error synchronization prediction method proposed in this invention improves the RMSE of the vehicle speed prediction result by 5.84% over the entire test conditions compared with the single-step rolling prediction method without error compensation.

[0142] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A vehicle speed prediction method based on error synchronization prediction, characterized in that: The steps are as follows: S1. Collect historical vehicle speed data and establish the original vehicle speed sequence based on the time axis; S2. According to different standard driving conditions, the original vehicle speed sequence is divided into dataset A, dataset B and dataset C; S3. Based on dataset A, a single-step vehicle speed prediction model is established using an RBF neural network; S4. Use a single-step rolling prediction strategy to perform multi-step vehicle speed prediction on dataset B, and calculate the multi-step prediction error of dataset B based on the multi-step vehicle speed prediction results of dataset B. S5. Based on the one-step lead error data, establish a one-step lead RBF neural network prediction error model; S6. Based on the entire error dataset, establish a Markov chain prediction error model from the first two steps to the last step; S7. Subtract the error correction result from the multi-step prediction results to obtain the final vehicle speed prediction result; In S4, the single-step rolling prediction strategy is expressed as follows: in, For the current moment, For the first The predicted value at that time; The maximum value for the prediction step size is 4; The current moment can be obtained using a single-step rolling prediction strategy. Vehicle speed prediction results in the next prediction time domain : Multi-step prediction of vehicle speed in dataset B The result is expressed as follows: Prediction Time Domain Inside, the actual vehicle speed is expressed as: in, The length of the time domain for vehicle speed prediction. ; The length of dataset B; Prediction error of dataset B Represented as: ； In S6, there are three Markov chain prediction error models that are two steps ahead to the last step: Error prediction model from 1 step ahead to 2 steps ahead Error prediction models that are 2 to 3 steps ahead Error prediction models that are 3 to 4 steps ahead ; The error prediction process for each error model mainly includes the following steps: state space partitioning, state transition probability matrix calculation, and state transition. The specific process of state space partitioning is as follows: The current state is defined as the error at the step ahead of time, the next state is defined as the error at the step ahead of time (2 steps ahead), and the prediction error of dataset B is defined as the error at the step ahead of time (1 step ahead). In The error two steps ahead is the prediction error of dataset B. In ; Calculate the maximum and minimum values ​​of the 1-step and 2-step lead error sequences and round them down. These four values ​​are denoted as follows: , , and The calculation expression is as follows: Using a step size of 0.5 km / h, the state space is divided into two steps ahead, and the error state set is determined. and as follows: Set and The number of states are N and M respectively; The specific process for calculating the state transition probability is as follows: For each error value that is 1 step ahead or 2 steps ahead, the corresponding state is determined by its relationship with the state values ​​in the state space, and this state is denoted as... and ,statistics The Middle The state to The Middle The frequency of each state, divided by The Middle The state to The frequency of transitions in all states is obtained. The Middle The state transition probability of each state The calculation expression is as follows: in, , ; Traversal For each state in the matrix, calculate the state transition probability for each state, and finally obtain the state transition probability matrix, as follows: The specific process of state transition is as follows: Assuming the error is one step ahead. Then, by partitioning the state space and calculating the state transition probability matrix, we can determine the current state. a certain state And then find The state value corresponding to the highest transition probability. and The next state value , The corresponding transition value for: 。 2. The vehicle speed prediction method based on error synchronization prediction according to claim 1, characterized in that: In S3, the RBF neural network is defined as follows: in, For the first The output corresponding to each input; , This represents the number of hidden layer nodes. The weights from the hidden layer to the output layer; For the neural network One input; Center of radial basis functions; The standard deviation of the Gaussian function; It is the Gaussian norm; The input to the single-step vehicle speed prediction model established by the RBF neural network is the vehicle speed over a past period of time. Specifically, it represents the vehicle speed at the current moment and the vehicle speed at the past 5 sampling moments. Its output is the vehicle speed at the next sampling point, and its sampling period is 1 second.

3. The vehicle speed prediction method based on error synchronization prediction according to claim 1, characterized in that: In S3, an RBF neural network is used to establish a single-step vehicle speed prediction model. When training on dataset A, the vehicle speeds over a past period in dataset A are defined as the input to the RBF neural network, and the future vehicle speed is defined as the desired output. The training objective is the mean square error between the actual vehicle speed and the desired output future vehicle speed, calculated using the following formula. In the formula, The mean square error for the entire operating condition; for Predict vehicle speed in real time; for The actual speed at any given moment; The total duration of dataset A; when Training stops when the value reaches the preset training target or the maximum number of training iterations, thus obtaining the single-step vehicle speed prediction model.

4. The vehicle speed prediction method based on error synchronization prediction according to claim 1, characterized in that: In S5, the RBF neural network prediction error model with a one-step lead time is denoted as... The training set is the error dataset. Its input is the vehicle speed error at the current time and the past 5 sampling times, denoted as ; The output is the vehicle speed prediction error for the next time step, denoted as . The expression is as follows: 。 5. The vehicle speed prediction method based on error synchronization prediction according to claim 1, characterized in that: The error of being one step ahead is Output from the RBF neural network prediction error model (one step ahead) That is, the error of the state transition assumption with one step ahead is , equivalent to .

6. The vehicle speed prediction method based on error synchronization prediction according to claim 1, characterized in that: Simply replace the error data to establish and Error prediction model and error prediction for 3-step and 4-step lead time and Make predictions; Based on the established error prediction model, the error in the prediction time domain at time t can be predicted, and its expression is as follows: Finally obtained At time t, the prediction error in the prediction time domain .

7. The vehicle speed prediction method based on error synchronization prediction according to claim 1, characterized in that: In S7, the final prediction result of the vehicle speed in the time domain at the current moment. The expression is as follows: 。

Images