Online vehicle speed prediction method and system based on multi-source information fusion

Abstract

The invention relates to the technical field of vehicle speed prediction, in particular to an online vehicle speed prediction method and system based on multi-source information fusion, and the method comprises the steps: obtaining a multi-source vehicle data set, and carrying out the preprocessing of the multi-source vehicle data set; constructing a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extracting correlation features according to the correlation coefficient matrix; calculating the contribution rate and the cumulative contribution rate of the principal component, and carrying out dimension reduction optimization on the multi-source vehicle data set; and according to the multi-source vehicle data set after dimension reduction optimization, vehicle speed prediction is carried out based on a regression neural network. The objective of the invention is to solve the technical problems of insufficient input information dimension reduction processing capability of a vehicle speed prediction model and weak generalization capability of the vehicle speed prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle speed prediction, and in particular to an online vehicle speed prediction method and system based on multi-source information fusion. Background Art

[0002] As a key technology, vehicle speed prediction aims to estimate the vehicle speed at a specific time in the future based on the vehicle's current environment and its own status. In the development of intelligent vehicle control strategies, vehicle speed prediction has attracted much attention. On the one hand, accurate vehicle speed prediction is like a sharp early warning system that can timely understand potential dangerous conditions and build the first line of defense for safe vehicle driving. On the other hand, it plays a key role in the gear decision technology of automatic transmissions. By accurately predicting the vehicle speed, it can realize intelligent gear switching, thereby greatly improving economy and power. In the safety assisted driving system, vehicle speed prediction provides key data support for collision warning, adaptive cruise control and other functions, significantly enhancing the safety of the vehicle. In the predictive control strategy of electric vehicles, vehicle speed prediction helps reduce the dimension and optimize battery management and energy distribution, further improving the overall performance of the vehicle. It can be seen that vehicle speed prediction has become an indispensable key element in the intelligent control system of automobiles.

[0003] At present, the existing vehicle speed prediction technology lacks an effective dimensionality reduction processing strategy for the massive multi-source information obtained by vehicles. Since the information collected by vehicles is often complex in dimensions and contains a large amount of redundant and weakly correlated data, this not only makes the constructed vehicle speed prediction model structure extremely complex, increasing the difficulty of model training and prediction, but also greatly increases the amount of calculation, reduces the calculation efficiency, and is difficult to meet the application scenarios with high real-time requirements. On the other hand, most vehicle speed prediction models generally adopt the traditional mode of offline training and online use. During the actual vehicle driving process, the working conditions are affected by multiple factors such as driver behavior, dynamic changes in road conditions, and changing weather conditions, and are in continuous and complex changes. The offline training model cannot perceive and adapt to these changes in time, resulting in its weak generalization ability, which limits the accuracy of vehicle speed prediction. Summary of the invention

[0004] In order to solve the technical problems of insufficient dimensionality reduction processing capability of the vehicle speed prediction model input information and weak generalization capability of the vehicle speed prediction model, the present invention provides an online vehicle speed prediction method and system based on multi-source information fusion, and the technical solutions adopted are as follows: The technical solution of the first aspect of the present invention provides an online vehicle speed prediction method based on multi-source information fusion, the method comprising: Obtain multi-source vehicle datasets and perform preprocessing; Constructing a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extracting relevant features according to the correlation coefficient matrix; Calculate the contribution rate and cumulative contribution rate of the principal components, and perform dimensionality reduction optimization on multi-source vehicle data sets; According to the multi-source vehicle dataset after dimensionality reduction and optimization, vehicle speed prediction is performed based on regression neural network.

[0005] Furthermore, obtaining a multi-source vehicle dataset and preprocessing it includes: Using vehicle-mounted sensors and V2X communication modules to obtain a multi-source vehicle data set and construct a sample matrix, the multi-source vehicle data set includes at least vehicle information, road environment information, and traffic status information; The sample matrix is ​​preprocessed by zero-mean normalization.

[0006] Furthermore, a correlation coefficient matrix is ​​constructed based on the preprocessed multi-source vehicle data set, and relevant features are extracted based on the correlation coefficient matrix, including: Based on the preprocessed multi-source vehicle dataset sample matrix, calculate the correlation coefficient between data of different dimensions; A correlation coefficient matrix is ​​constructed according to the correlation coefficients between data of different dimensions, and the characteristic equation of the correlation coefficient matrix is ​​solved to obtain non-negative eigenvalues ​​and eigenvectors.

[0007] Furthermore, the contribution rate and cumulative contribution rate of the principal components are calculated, and the dimensionality reduction optimization of the multi-source vehicle data set is performed, including: The principal components are sorted according to their contribution rates, and the cumulative contribution rates of the sorted principal components are calculated. The main principal components are selected according to the preset cumulative contribution rate threshold, and then the information with the largest contribution in each main principal component is selected to form a multi-source vehicle data set after dimensionality reduction as the input of the vehicle speed prediction model.

[0008] Furthermore, the expression for calculating the principal component contribution rate is:

[0009] In the formula, Indicates The contribution rate of each principal component reflects the contribution of each principal component to the overall information of the data. The first character equation of the correlation coefficient matrix is ​​represented by eigenvalues; The first character equation of the correlation coefficient matrix is ​​represented by eigenvalues, is the index variable; represents the number of eigenvalues; The expression for calculating the cumulative contribution rate of the principal component is:

[0010] In the formula, Before The cumulative contribution rate of the principal components; The first character equation of the correlation coefficient matrix is ​​represented by feature values.

[0011] Furthermore, according to the multi-source vehicle data set after dimensionality reduction optimization, vehicle speed prediction based on regression neural network includes: Construct a vehicle speed prediction model based on regression neural network; Use the sliding time window method to update the multi-source vehicle dataset after dimensionality reduction optimization; Extracting a training sample input vector and a training sample output vector based on the updated multi-source vehicle data set, and updating a vehicle speed prediction model training sample matrix; Train the vehicle speed prediction model based on the updated training sample matrix; The prediction sample input vector is input into the trained vehicle speed prediction model, and the vehicle speed prediction result is output.

[0012] Furthermore, the training sample matrix for updating the vehicle speed prediction model includes: The extracted training sample input vector and training sample output vector are placed in the last column of the training sample input matrix and output matrix of the previous moment respectively, so as to update the training sample matrix.

[0013] Furthermore, according to the multi-source vehicle data set after dimensionality reduction optimization, vehicle speed prediction based on regression neural network includes: If the current time is less than the sum of the preset training time domain length and the prediction time domain length, the vehicle speed is predicted for one time step in the future, including: Obtain the multi-source vehicle data set after dimensionality reduction processing at the current moment and the previous moment, and construct a training sample matrix; Update the training sample matrix and train the vehicle speed prediction model based on the sliding time window method; The multi-source vehicle data at the current moment is used as the prediction sample input vector, and the vehicle speed at the next moment is predicted based on the trained vehicle speed prediction model.

[0014] Furthermore, according to the multi-source vehicle data set after dimensionality reduction optimization, vehicle speed prediction based on regression neural network includes: If the current time is not less than the sum of the preset training time domain length and the prediction time domain length, the vehicle speed for multiple time steps in the future is predicted, including: According to the preset training time domain length and prediction time domain length, the historical multi-source vehicle data set after dimensionality reduction is obtained and a training sample matrix is ​​constructed; Based on the sliding time window method, the historical multi-source vehicle dataset and training sample matrix are updated, and the vehicle speed prediction model is trained; The historical multi-source vehicle data is used as the prediction sample input vector, and the vehicle speed of the prediction time domain length is predicted based on the trained vehicle speed prediction model.

[0015] The technical solution of the second aspect of the present invention provides an online vehicle speed prediction system based on multi-source information fusion, using the online vehicle speed prediction method based on multi-source information fusion described in the technical solution of the first aspect of the present invention, and the system includes: A data acquisition and preprocessing module configured to acquire multi-source vehicle data sets and perform preprocessing; A feature extraction module is configured to construct a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extract relevant features according to the correlation coefficient matrix; A data dimension reduction module, configured to calculate the contribution rate and cumulative contribution rate of the principal components and perform dimension reduction optimization on the multi-source vehicle data set; The vehicle speed prediction module is configured to perform vehicle speed prediction based on a regression neural network according to the multi-source vehicle data set after dimensionality reduction optimization.

[0016] The present invention has the following beneficial effects: The online vehicle speed prediction method based on multi-source information fusion provided by the present invention removes redundant information, reduces the dimension of data, improves data processing efficiency, and retains information that is important for vehicle speed prediction by fusing and reducing the dimensionality of multi-source information. In the prediction process, a regression neural network algorithm is adopted and training samples are updated online, so that the vehicle speed prediction model can adapt to changing traffic conditions in real time, improves the accuracy and generalization ability of the model, can predict the vehicle speed more accurately, and provides reliable vehicle speed data support for online intelligent decision-making of vehicles. On the one hand, the present invention uses the PCA method to construct a correlation coefficient matrix and extract relevant features. By calculating the principal component contribution rate and the cumulative contribution rate, the information with large contribution is accurately screened out, and the redundant parts in the data can be effectively removed. The high-dimensional multi-source vehicle data set is reduced in dimension, and only the most critical information is retained as the input of the vehicle speed prediction model, avoiding the problems of high model complexity and low calculation efficiency caused by the input information dimension being too high and containing too much irrelevant information; on the other hand, the present invention aims at the problem of weak generalization ability of the vehicle speed prediction model, and adopts a recurrent neural network algorithm in the prediction stage. The recurrent neural network has a strong nonlinear mapping ability and can adapt to the complex and changeable relationship between the vehicle speed and multi-source data information. During the training process, the training samples are updated online by using a sliding time window, so that the model can obtain new data in real time, and timely adjust itself according to the dynamic changes of the working conditions during the vehicle driving process, so as to learn the speed change law under different working conditions. This online learning and updating mechanism enables the model to work effectively in various complex traffic scenes, greatly enhances the generalization ability of the model, and enables it to more accurately predict the vehicle speed under different working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A method flow chart of an online vehicle speed prediction method based on multi-source information fusion provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an online vehicle speed prediction system based on multi-source information fusion provided by an embodiment of the present invention; Figure 3 A schematic diagram of online vehicle speed prediction provided by an embodiment of the present invention; Figure 4 A schematic diagram of online vehicle speed prediction results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the online vehicle speed prediction method and system based on multi-source information fusion proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0020] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0021] The specific scheme of an online vehicle speed prediction method and system based on multi-source information fusion provided by the present invention is described in detail below with reference to the accompanying drawings.

[0022] See also Figure 1 , which shows a method flow chart of an online vehicle speed prediction method based on multi-source information fusion provided by an embodiment of the present invention, the method comprising: Step S100: Acquire a multi-source vehicle data set and perform preprocessing; Step S100 specifically includes: Step S110: using the vehicle-mounted sensor and the V2X communication module to obtain a multi-source vehicle data set and construct a sample matrix, where the multi-source vehicle data set includes at least vehicle information, road environment information, and traffic status information; wherein the sample matrix can be expressed as:

[0023] In the formula, represents the sample matrix; represents the number of samples, that is, the number of samples of multi-source information collected from vehicle-mounted sensors and V2X communication modules; The dimension of each sample corresponds to different information dimensions, such as different feature dimensions of the vehicle information, road environment information, traffic status information, etc. Indicates The sample in Dimensions of data.

[0024] Step S120: Perform zero-mean normalization preprocessing on the training sample matrix; specifically, in order to make data of different dimensions comparable, for the elements in the sample matrix The conversion can be expressed as:

[0025]

[0026]

[0027] In the formula, After standardization, The sample in Data of dimensions; Indicates The mean of the data in each dimension; Indicates The standard deviation of the data in each dimension.

[0028] Step S200: constructing a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extracting relevant features according to the correlation coefficient matrix; Step S200 specifically includes: Step S210: Based on the preprocessed multi-source vehicle data set sample matrix, calculate the correlation coefficient between data of different dimensions; wherein the correlation coefficient can be expressed as:

[0029] In the formula, Indicates Dimensions and The correlation coefficient between the dimensions measures the linear correlation between the data in two dimensions.

[0030] Step S220: construct a correlation coefficient matrix according to the correlation coefficients between data of different dimensions, and solve the characteristic equation of the correlation coefficient matrix to obtain non-negative eigenvalues ​​and eigenvectors; specifically, the correlation coefficient matrix can be expressed as:

[0031] In the formula, the correlation coefficient matrix It describes the linear correlation between data of different dimensions, reflects the degree of association between data, and helps to find redundant information and potential main information directions in the data; its value range is between [-1,1], close to 1 indicates positive correlation, close to -1 indicates negative correlation, and close to 0 indicates weak correlation. The correlation coefficient matrix helps to identify redundant information in the data and the inherent structure of the data, providing a basis for subsequent principal component analysis.

[0032] Solving the correlation matrix The characteristic equation of is: ,available non-negative eigenvalues And the corresponding Eigenvectors:

[0033] Among them, the non-negative eigenvalue is the eigenvalue of the correlation coefficient matrix, which reflects the variance of the data in the direction of the corresponding eigenvector. The larger the eigenvalue, the greater the variance of the data in the direction of the eigenvector, that is, the greater the amount of information contained; the eigenvector represents the direction of data change, which is a linear combination of the original data in different dimensions. They constitute a set of bases of the data, and different eigenvectors represent different main directions of data change.

[0034] Step S300: Calculate the contribution rate and cumulative contribution rate of the principal component, and perform dimensionality reduction optimization on the multi-source vehicle data set; The expression for calculating the principal component contribution rate is:

[0035] In the formula, Indicates The contribution rate of each principal component reflects the contribution of each principal component to the overall information of the data. The first character equation of the correlation coefficient matrix is ​​represented by eigenvalues; The first character equation of the correlation coefficient matrix is ​​represented by eigenvalues, is the index variable; represents the number of eigenvalues; The expression for calculating the cumulative contribution rate of the principal component is:

[0036] In the formula, Before The cumulative contribution rate of the principal components; The first character equation of the correlation coefficient matrix is ​​represented by feature values.

[0037] Step S300 specifically includes: Step S310: Sort the principal components according to the principal component contribution rate, calculate the cumulative contribution rate of the sorted principal components, select the main principal components according to the preset cumulative contribution rate threshold, and then select the information with the largest contribution in each main principal component to form a multi-source vehicle data set after dimensionality reduction as the input of the vehicle speed prediction model; specifically, sort the principal components from large to small according to the contribution rate, and select the principal components with a cumulative contribution rate of 85% or more as the main principal components. This means that the principal components that can contain most of the data information are selected, and the dimensionality reduction of the data is achieved while ensuring that the information loss is small. Then, the information with the largest contribution is selected from these main principal components as the input of the online vehicle speed prediction model. This is to reduce the input dimension by screening the more important information dimensions in the main principal components while retaining the key information, thereby providing more concise and representative data for the subsequent vehicle speed prediction model, reducing the complexity and calculation amount of the model, and improving the performance and efficiency of the model.

[0038] Step S400: predicting vehicle speed based on a regression neural network according to the multi-source vehicle data set after dimensionality reduction optimization; Step S400 specifically includes: Step S410: constructing a vehicle speed prediction model based on a recurrent neural network; specifically, the vehicle speed prediction module in this embodiment is preferably a generalized recurrent neural network (General Regression Neural Network), including an input layer, a pattern layer, a summation layer and an output layer; in this embodiment, the input vector of the GRNN model is a historical multi-source vehicle data set, and the output vector is the predicted vehicle speed of the main vehicle in the future; a training sample of the GRNN model is composed of a training sample input vector and a training sample output vector, and the training sample input vector and output vector are extracted from the historical vehicle speed according to the model input and output vector according to the sliding time window, and a number of training samples constitute a training sample matrix; Step S410 specifically includes: Step S411: configure the number of input layer neurons according to the dimension of the training sample input vector; the number of neurons in the input layer is equal to the dimension of the training sample input vector, and the neurons in the input layer serve as simple distribution units, whose function is to pass the input vector directly to the pattern layer.

[0039] Step S412: configure the number of neurons in the pattern layer according to the number of training samples, and use the transfer function to extract the output of the pattern layer; specifically, the number of neurons in the pattern layer is equal to the number of training samples, and each neuron corresponds to a different training sample. This implementation uses the transfer function to calculate the similarity relationship between the input vector and each training sample, and obtains the output of the pattern layer by calculating the distance between the input vector and the corresponding training sample and performing mathematical transformation. The transfer function can be expressed as:

[0040] In the formula, represents the input vector of the GRNN model, Presentation model layer The training samples corresponding to neurons are Represents the smoothing factor; this transfer function can reflect the similarity between the input vector and the training sample, providing a basis for subsequent summation layer calculations.

[0041] Step S413: configure the number of neurons in the summation layer and the output layer according to the prediction duration, the summation layer sums the output of the pattern layer, and the output layer outputs the predicted vehicle speed data; specifically, the number of neurons in the summation layer is the prediction duration plus 1; in this embodiment, the summation layer performs two summation operations, one is to perform arithmetic summation on the outputs of all neurons in the pattern layer, and the connection weight between the pattern layer and the summation neuron is 1; the other is to perform weighted summation on the neurons in the pattern layer, and the weight is determined by the output vector elements of the training sample; finally, the output layer calculates the output value according to the two summation results of the summation layer, and for each neuron in the output layer, the weighted summation result is divided by the arithmetic summation result to obtain the predicted vehicle speed data, thereby completing the construction and prediction process of the vehicle speed prediction model.

[0042] Step S420: using a sliding time window method to update the multi-source vehicle dataset after dimensionality reduction optimization.

[0043] Step S430: extract the training sample input vector and the training sample output vector based on the updated multi-source vehicle data set, and update the vehicle speed prediction model training sample matrix; specifically, place the extracted training sample input vector and training sample output vector in the last column of the training sample input matrix and output matrix of the previous moment, respectively, to update the training sample matrix.

[0044] Step S440: training the vehicle speed prediction model based on the updated training sample matrix.

[0045] Step S450: input the prediction sample input vector into the trained vehicle speed prediction model, and output the vehicle speed prediction result.

[0046] The following describes the vehicle speed prediction process of the present invention by taking the prediction of the vehicle speed of a future time step as the first embodiment: S1: Single-step vehicle speed prediction S11: Obtain the multi-source vehicle data set after dimensionality reduction processing at the current moment and the previous moment, and construct a training sample matrix; specifically, set the training time domain length and the predicted time domain length 1, which means that the data of the previous 1 second of the current moment is considered during the training process, and the model will predict the vehicle speed in the next 1 second; S12: Initialize multi-source vehicle dataset And the training sample matrix , which can be expressed as:

[0047] In the formula, is a multi-source vehicle dataset after dimensionality reduction. for dimensional matrix, storing Moment and Multi-source information after dimensionality reduction at all times; Indicates the current moment, with a discrete interval of 1s; It indicates the dimension of multi-source information after dimensionality reduction, that is, the number of dimensions retained by multi-source information after dimensionality reduction operations such as principal component analysis; express After dimensionality reduction processing Multiple source information.

[0048]

[0049] In the formula, Represents the training sample input matrix, which stores the data from the initial moment to The information after dimensionality reduction at each moment is stored in columns, reflecting the input information of the time series. It aims to provide historical information for the model and help the model learn the relationship between multi-source information and vehicle speed at different moments.

[0050]

[0051] In the formula, Represents the training sample output matrix; stores the data from time 1 to The partial information is used as the expected output part of the training sample to provide the target value for model training, so that the model can learn how to map the input information to the output information; S13: Update the training sample matrix based on the sliding time window method and train the vehicle speed prediction model; specifically, collect The multi-source vehicle dataset after dimensionality reduction can be expressed as: , ... ; Use the sliding time window method to update the multi-source vehicle dataset. Multi-source information data acquired at all times is added to the dataset The last row and delete the first row of the dataset, the updated dataset It is composed of multi-source information data at the current moment and multi-source information data from the previous second, realizing rolling updates of multi-source vehicle data sets, ensuring that the data always reflects the latest time range information, and enabling the model to adapt to new data. It can be expressed as:

[0052] In the formula, Represents the updated training sample matrix.

[0053] The updated training sample matrix is ​​obtained from the updated multi-source vehicle dataset Extract new training sample input , output vector , and place them in the training sample input matrix of the previous moment , output matrix The last column, the number of training samples , which can be expressed as:

[0054]

[0055]

[0056]

[0057] In the formula, Represents the training sample input vector, from the updated multi-source vehicle dataset Extracted from The dimension reduction information at the moment is used as a new input sample element to update the training sample input matrix, providing the model with the latest input sample data, reflecting the latest information status; Represents the training sample output vector, extracted An information element at the moment is used as a new output sample element to update the training sample output matrix, providing the latest expected output for model training, so that the model can adjust its prediction ability according to the latest data; Represents the training sample input matrix, that is, the newly extracted training sample input vector Placed in the last column of the matrix while maintaining the dimension of the matrix, the updated matrix contains the latest input sample information, so that the training samples are continuously updated, allowing the model to learn the latest input features, reflecting the dynamic evolution of the data, and helping the model adapt to new data patterns; Represents the training sample output matrix, which outputs the newly extracted training sample vector Placed in the last column of the matrix, the updated matrix reflects the latest expected output information, provides the latest training target for the model, and enables the model to continuously optimize the prediction performance during the training process.

[0058] S14: Use the multi-source vehicle data at the current moment as the prediction sample input vector, and predict the vehicle speed at the next moment based on the trained vehicle speed prediction model; specifically, use the updated training sample input matrix , output matrix Through the multi-source information provided by the input sample matrix and the expected results provided by the output sample matrix, the model learns the relationship between input and output and adjusts the parameters of the model to adapt to the characteristics of the data; preferably, the performance of the model can be evaluated by the root mean square error to obtain the optimal vehicle speed prediction model.

[0059] The prediction sample input vector can be expressed as:

[0060] In the formula, Represents the prediction sample input vector; the prediction sample input vector is input into the updated GRNN model to predict the vehicle speed; is a dimensional vector, containing The dimensionality reduction information at the moment is used as the latest input, and the relationship learned by the model during training is used to predict the speed in the next 1 second; S15: Delay 1 second, ; S16: Determine whether to stop the vehicle speed prediction. If the vehicle stops running, stop the vehicle speed prediction. If not, proceed to the next step. S17: Judgment t Is it less than the preset time? If so, return to S13; if not, proceed to the next stage; The following is an explanation of the vehicle speed prediction process of the present invention by predicting the vehicle speed at multiple time steps in the future: S2: Multi-step vehicle speed prediction S21: According to the preset training time domain length and prediction time domain length, obtain the historical multi-source vehicle data set after dimensionality reduction and construct a training sample matrix; specifically, in the multi-step vehicle speed prediction stage, the preset , By setting these two parameters, the historical data range and the future duration to be predicted required for multi-step prediction are determined. Compared with single-step prediction, its time range is no longer limited to the current moment and the previous second, but is based on the specific and The time range is expanded to meet the needs of multi-step prediction.

[0061] S22: Update the historical multi-source vehicle dataset and training sample matrix based on the sliding time window method, and train the vehicle speed prediction model; S23: Use historical multi-source vehicle data as a prediction sample input vector and predict the vehicle speed of the prediction time domain length based on the trained vehicle speed prediction model.

[0062] S24: Delay 1 second, ; S25: Determine whether to stop the vehicle speed prediction. If the vehicle stops running, stop the vehicle speed prediction. If not, return to S22; It should be noted that the data processing process of multi-step speed prediction is similar to that of single-step speed prediction, so it will not be described in detail. The dimensions and data range of the input matrix and output matrix of the training samples in multi-step speed prediction are also increased accordingly, and the time span considered is larger, providing the model with richer historical information for learning the long-term speed change law. The sliding time window method is also used to update the multi-source vehicle data set, but the size of the window depends on the setting and value, which can retain data in a longer time range and is more conducive to learning long-term trends and periodicity in the data; the main differences between single-step speed prediction and multi-step speed prediction lie in the time range, data processing complexity and requirements for model performance, which are suitable for different application scenarios and prediction needs respectively. Single-step prediction is suitable for rapid estimation of short-term speed changes, while multi-step prediction is more suitable for situations where speed trends need to be considered over a longer time range.

[0063] See also Figure 3 As shown, Figure 3The method of realizing the functions of the present invention in the actual driving process of a car is illustrated, wherein the main vehicle V2X communication module communicates with the roadside unit V2X communication module and the nearest preceding vehicle V2X communication module to obtain multi-source information such as road speed limit, traffic light distance, remaining time of traffic light, vehicle density, preceding vehicle distance, and preceding vehicle speed. The main vehicle information acquisition module acquires the main vehicle speed and acceleration information through the vehicle-mounted sensor. The main vehicle V2X communication module sends the acquired multi-source information (including road speed limit, traffic light distance, remaining time of traffic light, vehicle density, preceding vehicle distance, preceding vehicle speed, main vehicle speed, and acceleration) to the main vehicle speed prediction module. The main vehicle speed prediction module executes an online speed prediction algorithm based on multi-source information fusion of a generalized regression neural network to predict the future speed of the main vehicle. The multi-source vehicle data set collected in this embodiment is shown in Table 1. Table 1 records the multi-source vehicle data including vehicle speed, road speed limit, traffic light distance, traffic light remaining time, vehicle density, front vehicle distance, acceleration, and front vehicle speed; Table 1: Multi-source vehicle data table

[0064] According to step S100 of the present invention, the multi-source vehicle data after the collected multi-source vehicle data set is standardized is shown in Table 2: Table 2: Preprocessed multi-source vehicle dataset

[0065] In step S200 of the present invention, a correlation coefficient matrix is ​​constructed based on the preprocessed multi-source vehicle data set as shown in Table 3: Table 3: Correlation coefficient matrix

[0066] In step S200 of the present invention, 8 non-negative eigenvalues ​​are extracted according to the correlation coefficient matrix, and the 8 non-negative eigenvalues ​​are: : 2.83380, : 1.28350, :1.05370, :0.93580, :0.77320, :0.61680, :0.42580, : 0.07730; The eigenvector matrix corresponding to the above 8 non-negative eigenvalues ​​is shown in Table 4: Table 4: Eigenvector matrix corresponding to non-negative eigenvalues

[0067] In step S300 of the present invention, the contribution rate and cumulative contribution rate of the main component are calculated. The calculation results are shown in Table 5. In the table, a1 to a8 represent the main components. The vector composed of the corresponding values ​​is the eigenvector of the principal component: Table 5: Statistics of principal component analysis results

[0068] In step S300, select the information with the largest contribution from the main principal components; sort the main components in order from large to small according to their contribution rates, and then select the main components with a cumulative contribution rate of 85% or more as the main principal components. As can be seen from Table 5, the cumulative contribution rate of the first 5 principal components has reached 86%, so the first 5 principal components are selected as the main principal components. First, determine the number of indicators that need to be selected for each main principal component according to the size of the contribution rate, and then select specific indicators according to the size of the absolute value of the coefficient in the eigenvector; the contribution rates of the first 5 main principal components decrease successively, so the number of indicators that need to be selected for each principal component also decreases successively and is at least 1. The number of indicators selected for the first 5 main principal components are 3, 2, 1, 1, and 1 respectively; then, combined with the indicators selected for each main principal component, finally select , , , , , , Instead of the original data, the vehicle speed, road speed limit, traffic light distance, traffic light remaining time, front vehicle distance, acceleration and front vehicle speed are selected as the input of the online vehicle speed prediction model; the present invention selects multi-source vehicle data information of vehicles running in a certain section of road (with a total length of about 45km) during a certain period of time, and uses the method described in this embodiment to predict the vehicle speed. The comparison between the predicted vehicle speed and the actual vehicle speed is shown in the figure. Figure 4 As shown, according to Figure 4 It can be understood that the predicted vehicle speed is highly consistent with the actual vehicle speed; the total time of the vehicle speed prediction condition is 4463s, the average root mean square error is 3.3974km / h, and the average time taken for a single prediction is 0.0199s; it can be understood that the online vehicle speed prediction method of multi-source information fusion provided in this application has good prediction accuracy and real-time performance; In summary, the online vehicle speed prediction method based on multi-source information fusion provided by the present invention removes redundant information, reduces the dimension of data, improves data processing efficiency, and retains information that is important for vehicle speed prediction by fusing and reducing the multi-source information; in the prediction process, a regression neural network algorithm is adopted and the training samples are updated online, so that the vehicle speed prediction model can adapt to the changing traffic conditions in real time, improves the accuracy and generalization ability of the model, can predict the vehicle speed more accurately, and provides reliable vehicle speed data support for the vehicle's online intelligent decision-making. On the one hand, the present invention uses the PCA method to construct a correlation coefficient matrix and extract relevant features. By calculating the principal component contribution rate and the cumulative contribution rate, the information with large contribution is accurately screened out, and the redundant parts in the data can be effectively removed. The high-dimensional multi-source vehicle data set is reduced in dimension, and only the most critical information is retained as the input of the vehicle speed prediction model, avoiding the problems of high model complexity and low calculation efficiency caused by the input information dimension being too high and containing too much irrelevant information; on the other hand, the present invention aims at the problem of weak generalization ability of the vehicle speed prediction model, and adopts a recurrent neural network algorithm in the prediction stage. The recurrent neural network has a strong nonlinear mapping ability and can adapt to the complex and changeable relationship between the vehicle speed and multi-source data information. During the training process, the training samples are updated online by using a sliding time window, so that the model can obtain new data in real time, and timely adjust itself according to the dynamic changes of the working conditions during the vehicle driving process, so as to learn the speed change law under different working conditions. This online learning and updating mechanism enables the model to work effectively in various complex traffic scenes, greatly enhances the generalization ability of the model, and enables it to more accurately predict the vehicle speed under different working conditions.

[0069] See also Figure 2 , which shows a schematic diagram of the structure of an online vehicle speed prediction system based on multi-source information fusion provided by an embodiment of the present invention, the system comprising: A data acquisition and preprocessing module configured to acquire multi-source vehicle data sets and perform preprocessing; A feature extraction module is configured to construct a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extract relevant features according to the correlation coefficient matrix; A data dimension reduction module, configured to calculate the contribution rate and cumulative contribution rate of the principal components and perform dimension reduction optimization on the multi-source vehicle data set; The vehicle speed prediction module is configured to perform vehicle speed prediction based on a regression neural network according to the multi-source vehicle data set after dimensionality reduction optimization.

[0070] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0071] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An online vehicle speed prediction method based on multi-source information fusion is characterized by: The method comprises: Obtain multi-source vehicle datasets and perform preprocessing; Constructing a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extracting relevant features according to the correlation coefficient matrix; Calculate the contribution rate and cumulative contribution rate of the principal components, and perform dimensionality reduction optimization on multi-source vehicle data sets; According to the multi-source vehicle dataset after dimensionality reduction and optimization, vehicle speed prediction is performed based on regression neural network.

2. The online vehicle speed prediction method based on multi-source information fusion according to claim 1, characterized in that: Acquiring multi-source vehicle datasets and preprocessing them includes: Using vehicle-mounted sensors and V2X communication modules to obtain a multi-source vehicle data set and construct a sample matrix, the multi-source vehicle data set includes at least vehicle information, road environment information, and traffic status information; The sample matrix is ​​preprocessed by zero-mean normalization.

3. The online vehicle speed prediction method based on multi-source information fusion as claimed in claim 2, characterized in that: A correlation coefficient matrix is ​​constructed based on the preprocessed multi-source vehicle data set, and relevant features are extracted based on the correlation coefficient matrix, including: Based on the preprocessed multi-source vehicle dataset sample matrix, calculate the correlation coefficient between data of different dimensions; A correlation coefficient matrix is ​​constructed according to the correlation coefficients between data of different dimensions, and the characteristic equation of the correlation coefficient matrix is ​​solved to obtain non-negative eigenvalues ​​and eigenvectors.

4. The online vehicle speed prediction method based on multi-source information fusion as claimed in claim 3, characterized in that: Calculate the contribution rate and cumulative contribution rate of the principal components and perform dimensionality reduction optimization on the multi-source vehicle data set, including: The principal components are sorted according to their contribution rates, and the cumulative contribution rates of the sorted principal components are calculated. The main principal components are selected according to the preset cumulative contribution rate threshold, and then the information with the largest contribution in each main principal component is selected to form a multi-source vehicle data set after dimensionality reduction as the input of the vehicle speed prediction model.

5. The online vehicle speed prediction method based on multi-source information fusion as claimed in claim 4, characterized in that: The expression for calculating the principal component contribution rate is: In the formula, Indicates The contribution rate of each principal component reflects the contribution of each principal component to the overall information of the data. The first character equation of the correlation coefficient matrix eigenvalues; The first character equation of the correlation coefficient matrix is ​​represented by eigenvalues, is the index variable; represents the number of eigenvalues; The expression for calculating the cumulative contribution rate of the principal component is: In the formula, Before The cumulative contribution rate of the principal components; The first character equation of the correlation coefficient matrix is ​​represented by feature values.

6. The online vehicle speed prediction method based on multi-source information fusion according to any one of claims 1 to 5, characterized in that: According to the multi-source vehicle dataset after dimensionality reduction optimization, the vehicle speed prediction based on the regression neural network includes: Construct a vehicle speed prediction model based on regression neural network; Use the sliding time window method to update the multi-source vehicle dataset after dimensionality reduction optimization; Extracting a training sample input vector and a training sample output vector based on the updated multi-source vehicle data set, and updating a vehicle speed prediction model training sample matrix; Train the vehicle speed prediction model based on the updated training sample matrix; The prediction sample input vector is input into the trained vehicle speed prediction model, and the vehicle speed prediction result is output.

7. The online vehicle speed prediction method based on multi-source information fusion as claimed in claim 6, characterized in that: The training sample matrix for updating the vehicle speed prediction model includes: The extracted training sample input vector and training sample output vector are placed in the last column of the training sample input matrix and output matrix of the previous moment respectively, so as to update the training sample matrix.

8. The online vehicle speed prediction method based on multi-source information fusion as claimed in claim 6, characterized in that: According to the multi-source vehicle dataset after dimensionality reduction optimization, the vehicle speed prediction based on the regression neural network includes: If the current time is less than the sum of the preset training time domain length and the prediction time domain length, the vehicle speed is predicted for one time step in the future, including: Obtain the multi-source vehicle data set after dimensionality reduction processing at the current moment and the previous moment, and construct a training sample matrix; Update the training sample matrix and train the vehicle speed prediction model based on the sliding time window method; The multi-source vehicle data at the current moment is used as the prediction sample input vector, and the vehicle speed at the next moment is predicted based on the trained vehicle speed prediction model.

9. The online vehicle speed prediction method based on multi-source information fusion as claimed in claim 6, characterized in that: According to the multi-source vehicle dataset after dimensionality reduction optimization, the vehicle speed prediction based on the regression neural network includes: If the current time is not less than the sum of the preset training time domain length and the prediction time domain length, the vehicle speed for multiple time steps in the future is predicted, including: According to the preset training time domain length and prediction time domain length, the historical multi-source vehicle data set after dimensionality reduction is obtained and a training sample matrix is ​​constructed; Based on the sliding time window method, the historical multi-source vehicle dataset and training sample matrix are updated, and the vehicle speed prediction model is trained; The historical multi-source vehicle data is used as the prediction sample input vector, and the vehicle speed of the prediction time domain length is predicted based on the trained vehicle speed prediction model.

10. The online vehicle speed prediction system based on multi-source information fusion is characterized by: The online vehicle speed prediction method based on multi-source information fusion according to any one of claims 1 to 9 is adopted, and the system comprises: A data acquisition and preprocessing module configured to acquire multi-source vehicle data sets and perform preprocessing; A feature extraction module is configured to construct a correlation coefficient matrix according to the preprocessed multi-source vehicle data set, and extract relevant features according to the correlation coefficient matrix; A data dimension reduction module, configured to calculate the contribution rate and cumulative contribution rate of the principal components and perform dimension reduction optimization on the multi-source vehicle data set; The vehicle speed prediction module is configured to perform vehicle speed prediction based on a regression neural network according to the multi-source vehicle data set after dimensionality reduction optimization.

Images