A vehicle speed multi-time scale prediction method, system, device, medium and terminal

Abstract

The application belongs to the technical field of vehicle speed prediction, and discloses a vehicle speed multi-time scale prediction method, system, device, medium and terminal, the vehicle speed multi-time scale prediction method comprises the following steps: preprocessing historical driving data, and establishing a sub-database; offline training of a long short-term memory (LSTM) neural network, and establishing a speed prediction sub-model for different working conditions; online multi-step speed prediction, real-time identification of a driving mode through fuzzy C-means clustering (FCM), and selection of a pre-trained LSTM speed prediction sub-model for prediction. The traditional speed prediction method cannot realize identification and self-adaptation of different driving modes, and the prediction accuracy still has room for improvement, and the application can realize accurate estimation of the vehicle motion state under complex driving conditions. The speed prediction method provided by the application meets online implementation conditions, and can lay a foundation for real-time energy management of a power system, further reduction of the whole vehicle operation cost and further optimization of the service life.

Description

Technical Field

[0001] This invention belongs to the field of vehicle speed prediction technology, and particularly relates to a method, system, device, medium and terminal for predicting vehicle speed over multiple time scales. Background Technology

[0002] Currently, energy management of vehicle powertrain systems is crucial for ensuring the safe and efficient operation of hybrid electric vehicles, and speed prediction is a necessary means to achieve real-time energy management. Speed, as an instantaneous indicator of a vehicle's motion state, determines the instantaneous load demand of a multi-source powertrain system, providing a reference for optimally allocating load power demand to each power source. A lack of high-precision prediction of driving conditions across multiple time scales increases the decision-making risk of energy optimization and control under unknown operating disturbances, thus affecting the optimal utilization of the vehicle's energy. However, accurately predicting the future speed of a vehicle is extremely challenging due to the combined influence of multiple factors such as traffic conditions, weather, driving style, and preferences.

[0003] Neural networks are a highly representative data-driven vehicle speed prediction algorithm. J. Hou et al. proposed a vehicle speed prediction algorithm based on an online adaptive radial basis function neural network in the paper "J.Hou, D. Yao, F. Wu, J. Shen and X. Chao, Online Vehicle Velocity Prediction Using an Adaptive Radial Basis Function Neural Network, IEEE Transactions on Vehicular Technology [J], 2021, 70(4): 3113-3122". The algorithm determines the optimal network structure parameters using the Akek information criterion and periodically retrains the network using real-time collected samples to enhance the adaptability of speed prediction to changes in the driving environment. The patent CN109118787B, "A Vehicle Speed ​​Prediction Method Based on Deep Neural Networks", establishes an information interaction model in an intelligent connected transportation system, then uses the autoregressive moving average method to predict acceleration, and uses deep neural networks to predict speed, providing a reference for drivers to make judgments and reducing the incidence of traffic accidents. However, it cannot achieve the recognition and adaptation of different driving modes.

[0004] Vehicle driving modes (e.g., city, suburban, highway modes, etc.) are a comprehensive description of the road traffic environment and vehicle driving status over a longer time scale. In order to achieve reliable prediction of power system load mode and aging status, it is necessary to accurately identify vehicle driving modes.

[0005] Based on the above analysis, the problems and defects of the existing technology are as follows: most technologies cannot recognize and adapt to different driving modes, and have the limitation of not being able to adjust to changing driving modes; the laws of future vehicle speed distribution under different driving modes and the online adjustment strategy of the prediction model have not been fully studied, which weakens the adaptability of the vehicle speed prediction model under different modes and increases the difficulty of high-precision vehicle speed prediction and vehicle energy optimization control. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method, system, device, medium, and terminal for predicting vehicle speed over multiple time scales.

[0007] This invention is implemented as follows: a multi-timescale vehicle speed prediction method. The method preprocesses historical driving data to establish a sub-database; trains a neural network offline to build speed prediction sub-models for different driving conditions; and performs online multi-step speed prediction by using fuzzy C-means clustering (FCM) to identify driving patterns in real time and selecting the corresponding pre-trained speed prediction sub-model for prediction. The innovation of this invention lies in achieving data-driven speed prediction based on driving pattern recognition, enabling speed prediction for different driving conditions and improving prediction accuracy.

[0008] Furthermore, the multi-timescale vehicle speed prediction method establishes a sub-database, specifically including:

[0009] (1) Extract the original vehicle dataset from the mail delivery route database collected by GPS to construct the driving profile of the Long Short-Term Memory (LSTM) neural network prediction model;

[0010] (2) Data preprocessing. The raw velocity and acceleration data are divided into many groups. dimensional vector, This represents the historical sampling range, meaning each vector contains historical driving information within a fixed sampling range. The feature parameters for each vector set are calculated, including: average speed. Average acceleration Speed ​​standard deviation To form a three-dimensional vector Normalize all three-dimensional vectors to obtain the feature vectors. ,in , , They are respectively The normalized results of the average velocity, average acceleration, and standard deviation of velocity calculated at each instant;

[0011] (3) Establishment based on FCM Multiple sub-databases are used to build multiple speed prediction sub-models, and driving data is divided into different clusters based on driving characteristics; given the number of cluster centers... and dataset FCM output membership matrix and Cluster centers Then, based on the maximum membership degree, all vectors are divided into 4 independent clusters; 10,000 sets of velocity data are selected from each cluster as the training set for step two, and then the velocity vectors are normalized to have unit variance and zero mean, resulting in... Sub-database.

[0012] Furthermore, the offline training neural network for the vehicle speed multi-timescale prediction method specifically includes:

[0013] Speed ​​prediction sub-models corresponding to different driving modes are trained. The LSTM neural network is trained in a regression manner for speed prediction; the training sequence at time k is a normalized sequence. dimensional velocity vector ,in and They are The normalized velocity at time k and time k, the response is the future. Predict the velocity vector in the time domain , and They are Time and The normalized velocity at time intervals Represents the historical sampling range. The LSTM network represents the prediction time domain length, in seconds; LSTM networks learn to predict the future. Predict velocity sequence values ​​in the time domain; generate after training. A neural network sub-model.

[0014] Furthermore, the online multi-step speed prediction of the vehicle speed multi-timescale prediction method specifically includes:

[0015] (1) Data sampling: At time k, the velocity and acceleration are sampled in real time by the sensor. Based on the data preprocessing in the established sub-database, the corresponding three-dimensional feature vector is calculated. ,in , , The past calculated at time k respectively The results of normalizing the average velocity, average acceleration, and standard deviation of velocity over each sampling time period;

[0016] (2) Real-time driving mode recognition, through calculation distance Cluster centers Distance to determine membership degree ,in and They are respectively Corresponding to the 1st and the 2nd The membership degree of each cluster center is used to select the highest membership degree and label it as . Complete driving mode recognition;

[0017] (3) Real-time velocity prediction: using the same normalized data as during offline training, the velocity data is normalized to... ,in and These are collected in real time. The normalized result of the velocity at time k. Input to the corresponding neural network. A prediction is made, resulting in a set of velocity data; this velocity data is then inversely normalized to obtain the future velocity data at time k. Speed ​​prediction results in seconds ,in, and These are collected in real time. The normalized velocity results at time k and time k. and The predicted number of Second and the The speed in seconds.

[0018] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the vehicle speed multi-timescale prediction method.

[0019] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the vehicle speed multi-timescale prediction method.

[0020] Another objective of this invention is to provide an information data processing terminal for implementing the vehicle speed multi-timescale prediction method.

[0021] Another objective of this invention is to provide a vehicle speed multi-timescale prediction system based on the aforementioned vehicle speed multi-timescale prediction method, the vehicle speed multi-timescale prediction system comprising:

[0022] The data preprocessing module is used to preprocess historical driving data and establish sub-databases;

[0023] The LSTM prediction sub-model building module is used to train neural networks offline and build speed prediction sub-models for different working conditions.

[0024] The speed prediction module is used for online multi-step speed prediction. It identifies the driving mode in real time through FCM and selects the corresponding pre-trained speed prediction sub-model for prediction.

[0025] Another object of the present invention is to provide a vehicle powertrain energy management terminal, wherein the vehicle powertrain energy management terminal includes the aforementioned vehicle speed multi-timescale prediction system.

[0026] Another object of the present invention is to provide a hybrid electric vehicle, the hybrid electric vehicle including the aforementioned vehicle powertrain energy management terminal.

[0027] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0028] First, addressing the technical problems and difficulties in solving the aforementioned existing technologies, and closely combining the technical solution to be protected by this invention with the results and data from the research and development process, this invention provides a detailed and in-depth analysis of how the technical solution of this invention solves the technical problems and the creative technical effects brought about after solving the problems. Specifically, this invention proposes an improved speed prediction method based on LSTM neural networks and FCM. FCM is used to identify driving modes, improving the adaptability of the vehicle speed prediction model under different modes. This invention innovatively proposes a method for speed prediction based on different driving modes, thereby overcoming the limitation of insufficient prediction accuracy in existing speed prediction methods.

[0029] Secondly, considering the technical solution as a whole or from a product perspective, the technical effects and advantages of the technical solution protected by this invention are specifically described as follows: This invention can achieve accurate prediction of vehicle motion state under complex driving conditions, and can cope with different driving modes. Compared with the traditional LSTM neural network speed prediction method without driving mode recognition, it can significantly improve the speed prediction accuracy, laying the foundation for efficient and accurate energy management; the speed prediction method proposed in this invention meets the conditions for online implementation, has high computational efficiency, and has practical application value; this invention can lay the foundation for real-time energy management of the power system, further reduction of vehicle operating costs, and further optimization of service life; this invention uses FCM for driving mode clustering, which is simple and reliable.

[0030] Third, as supplementary evidence of the inventive step of the claims of this invention, it is also reflected in the following important aspects:

[0031] (1) The expected benefits and commercial value of the technical solution of this invention after transformation are as follows:

[0032] This invention, focusing on high-precision identification of vehicle driving modes and improved accuracy of vehicle speed prediction, achieves accurate prediction of vehicle motion states and powertrain load under complex driving conditions. This enhances the accuracy of speed prediction and increases the reliability of overall vehicle energy optimization control, laying the foundation for powertrain energy management. Energy management technology is a core technological bottleneck that urgently needs to be overcome to clear the obstacles to the commercialization of fuel cell vehicles. Furthermore, this invention provides solid theoretical support for reducing vehicle operating costs and extending the service life of the powertrain, and has significant application value and importance for accelerating the commercialization of hydrogen fuel cell vehicles.

[0033] (2) The technical solution of this invention fills a technical gap in the industry both domestically and internationally:

[0034] Existing speed prediction technologies can predict speed, but their accuracy can only be improved because they cannot identify and predict speeds based on different driving modes. The technical solution of this invention fills the technological gap in speed prediction based on driving mode identification both domestically and internationally. It studies the patterns of future vehicle speed distribution under different driving modes and the online adjustment strategy of the prediction model, increasing the adaptability of the vehicle speed prediction model under different modes and achieving high-precision vehicle speed prediction. Attached Figure Description

[0035] Figure 1 This is a flowchart of the multi-timescale vehicle speed prediction method provided in the embodiments of the present invention;

[0036] Figure 2 This is a schematic diagram of the multi-timescale vehicle speed prediction method provided in this embodiment of the invention;

[0037] Figure 3 This is a comparison chart of the results provided by the embodiments of the present invention with those of the traditional speed prediction method. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] like Figure 1 As shown, the vehicle speed multi-timescale prediction method provided in this embodiment of the invention includes the following steps:

[0040] S101: Preprocess historical driving data and establish a sub-database;

[0041] S102: Offline training of neural networks to establish speed prediction sub-models for different working conditions;

[0042] S103: Online multi-step speed prediction, which uses FCM to identify driving modes in real time and selects the corresponding pre-trained speed prediction sub-model for prediction.

[0043] The vehicle speed multi-timescale prediction method provided in this invention includes three main steps: establishing a sub-database, offline training of a neural network, and online multi-step speed prediction; specifically including:

[0044] Step 1: Create a sub-database

[0045] (1a) Extract the raw vehicle dataset (speed and acceleration data) from the mail delivery route database collected by GPS to construct the driving profile of the LSTM prediction model.

[0046] (1b) Data preprocessing. The raw velocity and acceleration data are divided into many groups. A set of 3D vectors, each containing historical driving information within a fixed sampling range. Calculate the feature parameters for each vector set, including: average speed. Average acceleration Speed ​​standard deviation These vectors form a three-dimensional vector. Normalizing all three-dimensional vectors yields the eigenvectors, represented as... .in They are respectively The normalized results of the average velocity, average acceleration, and standard deviation of velocity calculated at each moment.

[0047] (1c) Based on FCM A sub-database. To build multiple speed prediction sub-models, the driving data needs to be divided into different clusters based on driving characteristics. The dataset consists of N three-dimensional feature vectors, given the number of cluster centers. and dataset FCM will output a membership matrix. and Cluster centers Then, based on the maximum membership degree, all vectors are divided into 4 independent clusters. In each cluster, 10,000 velocity data points are selected as the training set for step two. The velocity vectors are then normalized to have unit variance and zero mean, resulting in... Sub-database.

[0048] Step 2: Train the neural network offline

[0049] Speed ​​prediction sub-models corresponding to different driving modes are trained. The LSTM neural network is trained in a regression manner for speed prediction. The training sequence at time k is a normalized sequence. dimensional velocity vector ,in and They are The normalized velocity at time k and time k, the response is the future. velocity vector within seconds , and They are Time and The speed after time-normalization. In other words, the LSTM neural network learns to predict the future. Velocity sequence values ​​within a time step. Generated after training. A neural network sub-model.

[0050] Step 3, Online Multi-Step Velocity Prediction

[0051] (3a) Data sampling. At time k, the velocity and acceleration are sampled in real time by the sensor, and the corresponding three-dimensional feature vector is calculated according to step (1b) of step one. ,in These are the normalized results of the average velocity, average acceleration, and standard deviation of velocity calculated at time k.

[0052] (3b) Real-time driving mode recognition. This is achieved through calculation... distance Cluster centers Distance to determine membership degree , in and They are respectively Regarding the first and the second The membership degree of each cluster center is selected, and the highest membership degree is labeled as 'a' to complete the driving mode recognition.

[0053] (3c) Real-time velocity prediction. Using the same normalized data as during offline training. and Normalize the speed data to ,in and These are collected in real time. The normalized velocity results at time k and time k. Input to the corresponding neural network. A prediction is made, resulting in a set of velocity data; this velocity data is then inversely normalized to obtain the future velocity data at time k. Speed ​​prediction results in seconds ,in and The predicted number of Second and the The speed in seconds.

[0054] The vehicle speed multi-timescale prediction system provided in this embodiment of the invention includes:

[0055] The preprocessing module is used to preprocess historical driving data and establish sub-databases;

[0056] The prediction sub-model building module is used to train neural networks offline and build speed prediction sub-models for different working conditions.

[0057] The online speed prediction module is used for online multi-step speed prediction. It identifies the driving mode in real time through FCM and selects the corresponding pre-trained speed prediction sub-model for prediction.

[0058] To demonstrate the inventiveness and technical value of the technical solution of this invention, this section provides specific product or related technology application examples of the technical solution claimed.

[0059] This invention is applied to a fuel cell postal delivery vehicle, which has multiple driving modes. The number of FCM clusters is [not specified]. When set to 4, the driving mode can be divided into four types: parking delivery mode, acceleration mode, deceleration mode and high-speed driving mode. This invention can perform high-precision speed prediction.

[0060] This invention can also be applied to fuel cell heavy-duty vehicles with relatively fixed routes, such as fuel cell buses and fuel cell heavy-duty trucks. The commercialization of fuel cell buses is driven by national policy; my country's independently developed hydrogen fuel cell buses can start and operate normally in extremely cold climates down to -35°C. This invention can provide high-precision speed prediction for heavy-duty vehicles such as fuel cell buses and fuel cell heavy-duty trucks, offering a reference for advanced energy management, thereby reducing vehicle operation and maintenance costs. It is of significant value and importance in promoting the large-scale commercialization of fuel cell heavy-duty trucks.

[0061] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0062] A schematic diagram of the velocity prediction method of this invention is provided. Figure 2 The following is combined with Figure 2 The three main steps are described in detail: establishing a sub-database, training an LSTM neural network offline, and predicting speed online in multiple steps.

[0063] Step 1: Create a sub-database

[0064] (1a) Extract the raw vehicle dataset (speed and acceleration data) from the mail delivery route database collected by GPS to construct the driving profile of the LSTM prediction model, with a speed curve length of N.

[0065] (1b) Divide the original velocity and acceleration data into many... dimensional vector, the first The velocity and acceleration vectors are respectively represented as follows: , ,in, It is the first The speed of time It is the first The acceleration at each moment, each vector containing historical driving information within a fixed sampling range. Calculate the feature parameters of each vector set, including: average velocity. Average acceleration Speed ​​standard deviation These are used to form a three-dimensional vector. The three-dimensional vector is then normalized to obtain the eigenvectors, with the eigenvalue as the first eigenvector. Taking time as an example, the feature vector is represented as: ,in, , , The maximum value of the average velocity. The maximum value of the velocity variance The maximum value of the average acceleration Minimum value of average acceleration Therefore, driving data can be represented by many three-dimensional feature vectors, and the set of all three-dimensional vectors is represented as... .

[0066] (1c) Define driving modes. Given the number of cluster centers. and processed datasets After FCM, the membership matrix can be obtained. and Cluster centers ,and ,in Indicates sample Regarding the membership degree of category i. In this example, driving modes are divided into four types: cruise mode, acceleration mode, deceleration mode, and stop mode.

[0067] (1d) Construction A sub-database. The method for determining the maximum membership degree is as follows: If , Indicates sample Regarding the membership degree of category j, then Classify it into category j. Following this method, all velocity data... They are divided into four independent clusters according to driving modes, among which The first one established in (1b) The, the The and the first 10,000 velocity vectors are selected from each cluster as the training set for step two. Then, all velocity vectors are normalized to... This gives them unit variance and zero mean, resulting in four sub-databases. It is the first The result after normalizing the velocity vector , , yes The normalized value of the velocity at time t. yes The speed of time It is the average of all the raw velocity data. is the variance of all raw speed data, and N represents the total length of the driving curve.

[0068] Step two: Train the neural network offline, training speed prediction sub-models for different modes. In this example, the LSTM neural network has 4 layers, including 1 hidden layer with 200 units. It is trained in a regression manner, with each speed vector... The corresponding response is the future. Predict the vector of velocities in the time domain , in, , , and They are the first Time, k-th time, k-th time Time, Number The time-normalized speed. Each sub-training set generates a neural network sub-model after training. This example yields four neural network sub-models, represented as follows: .

[0069] Step 3, Online multi-step velocity prediction

[0070] (3a) In this example, the sampling interval is 1s. Calculate the three-dimensional feature vector at this moment according to step (1b) of step one. .

[0071] (3b) Real-time driving mode recognition. This is achieved through calculation... Distance from 4 cluster centers Euclidean distance, to determine membership degree ,in , They are respectively Regarding the first cluster center and the 4th cluster center The membership degree. The largest membership degree is selected and labeled as 'a', i.e. Complete driving mode recognition.

[0072] (3c) Real-time velocity prediction. Use the same normalized data as in step (1d) of step one. and Normalize the speed data to Input the corresponding neural network Make predictions and output a set of future... The velocity vector within a second. By inversely normalizing the output, we obtain the velocity vector at time k for the future. Speed ​​prediction results in seconds ,in and The predicted number of Second and the The speed in seconds.

[0073] This invention uses root mean square error The accuracy of velocity prediction at time k is determined, which is the positive effect of this invention. The average root mean square error is defined. The speed prediction accuracy for the entire driving cycle is calculated as follows:

[0074] (1);

[0075] (2);

[0076] Where N is the length of the entire driving cycle. To predict the length of the time domain, For the predicted first The speed of seconds It is the first The actual speed at any given moment. A smaller RMSE indicates less deviation between the predicted speed and the actual speed, resulting in higher prediction accuracy. A comparison of the performance of an example described in this invention with that of a traditional LSTM method without driving pattern recognition is provided for reference. Figure 3In a 1000-second driving cycle, the root mean square error of the predicted speed was reduced by 39%, demonstrating that the method effectively improves the accuracy of speed prediction.

[0077] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting vehicle speed across multiple time scales, characterized in that, The multi-timescale vehicle speed prediction method preprocesses historical driving data to establish a sub-database; trains a neural network offline to establish a speed prediction sub-model for different operating conditions; and performs online multi-step speed prediction by real-time identification of driving modes through FCM and selecting the corresponding pre-trained speed prediction sub-model for prediction. The multi-timescale vehicle speed prediction method establishes a sub-database, specifically including: (1) Extract the original vehicle dataset from the mail delivery route database collected by GPS to construct the driving profile of the LSTM prediction model; (2) Data preprocessing: the raw velocity and acceleration data are divided into many groups. A set of 3D vectors, each containing historical driving information within a fixed sampling range; calculate the feature parameters for each set of vectors, including: average speed. Average acceleration Speed ​​standard deviation To form a three-dimensional vector Normalize all three-dimensional vectors to obtain the feature vectors. ,in They are respectively The normalized results of the average velocity, average acceleration, and standard deviation of velocity calculated at each instant; (3) Establishment based on FCM Multiple sub-databases are used to build multiple speed prediction sub-models, and driving data is divided into different clusters based on driving characteristics; given the number of cluster centers... and dataset , The eigenvectors at time k; FCM outputs the membership matrix U and Cluster centers Then, based on the maximum membership degree, all vectors are divided into...

1. Select 10,000 velocity data points from each independent cluster as the training set for step two. Then, normalize the velocity vectors to have unit variance and zero mean, resulting in... Sub-databases; The offline training neural network for the vehicle speed multi-timescale prediction method specifically includes: Speed ​​prediction sub-models corresponding to different driving modes are trained. The LSTM neural network is trained in a regression manner for speed prediction; the training sequence at time k is a normalized sequence. dimensional velocity vector , in and They are The normalized velocity at time k and time k, the response is the future. Predict the velocity vector in the time domain , and They are Time and The normalized velocity at time intervals Represents the historical sampling range. Represents the length of the predicted time domain; LSTM neural networks learn to predict the future. Predict velocity sequence values ​​in the time domain; generate after training. One neural network sub-model; The online multi-step speed prediction of the vehicle speed multi-timescale prediction method specifically includes: (1) Data sampling: At time k, the velocity and acceleration are sampled in real time by the sensor. Based on the data preprocessing in the established sub-database, the corresponding three-dimensional feature vector is calculated. ,in They are respectively The normalized results of the average velocity, average acceleration, and standard deviation of velocity calculated at each instant; (2) Real-time driving mode recognition, through calculation distance Cluster centers Distance to determine membership degree ,in and The first and the The membership degree of each cluster center is used to select the highest membership degree and label it as . Complete driving mode recognition; (3) Real-time velocity prediction, using the same normalized data as during offline training. and Normalize the speed data to Input the corresponding neural network A prediction is made, resulting in a set of velocity data; this velocity data is then inversely normalized to obtain the future velocity data at time k. Speed ​​prediction results in seconds ; in, and These are collected in real time. The normalized velocity results at time k and time k. and The predicted number of Second and the The speed in seconds.

2. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the multi-timescale vehicle speed prediction method of claim 1.

3. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the multi-timescale vehicle speed prediction method of claim 1.

4. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the multi-timescale vehicle speed prediction method of claim 1.

5. A vehicle speed multi-timescale prediction system based on the vehicle speed multi-timescale prediction method of claim 1, characterized in that, The vehicle speed multi-timescale prediction system includes: The data preprocessing module is used to preprocess historical driving data and establish sub-databases; The prediction sub-model building module is used to train neural networks offline and build speed prediction sub-models for different working conditions. The real-time speed prediction module is used for online multi-step speed prediction. It identifies the driving mode in real time through FCM and selects the corresponding pre-trained speed prediction sub-model for prediction.

6. A vehicle powertrain energy management terminal, characterized in that, The vehicle powertrain energy management terminal includes the vehicle speed multi-timescale prediction system as described in claim 5.

7. A hybrid electric vehicle, characterized in that, The hybrid electric vehicle includes the vehicle powertrain energy management terminal as described in claim 6.

Images