Driving intention identifying and sharing method and system applied to non-networked automobile
By deploying RSU and MEC nodes on non-connected vehicles and using 5G networks and bidirectional LSTM algorithms to identify the driving intentions of non-connected vehicles, the problem of identifying and sharing driving intentions of non-connected vehicles has been solved, achieving efficient and accurate identification and sharing of driving intentions, and improving the safety and management efficiency of the traffic system.
Patent Information
- Application Number
- CN202410988156.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-23
AI Technical Summary
Existing intelligent connected vehicle systems struggle to effectively identify and share the driving intentions of non-connected vehicles, resulting in limited driving decisions. Existing methods suffer from issues such as low accuracy, poor real-time performance, or excessive cost.
Driving data from non-connected vehicles is collected by roadside units (RSUs), transmitted to MEC nodes via 5G communication networks, and driving intentions are identified using a bidirectional LSTM network and extreme gradient boosting algorithm. The data is then broadcast to connected intelligent vehicles via the 5G network, enabling real-time data processing and sharing through mobile edge computing and 5G technology.
It improves the accuracy and response speed of recognizing driving intentions in non-connected vehicles, reduces reliance on central cloud resources, optimizes resource utilization efficiency, and enhances the intelligence and safety of traffic management.
Smart Images

Figure CN119028128A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent transportation systems, and in particular relates to a driving intention recognition sharing method and system applied to non-networked vehicles. Background Art
[0002] With the rapid development of intelligent transportation systems (ITS), the intelligence and networking of vehicles have become an important development direction of the modern automobile industry. Intelligent connected vehicles use advanced sensors, controllers, actuators and other equipment to exchange and share information with external nodes, thereby improving driving safety and road traffic efficiency. However, most existing intelligent connected vehicle systems rely on the vehicle's own communication capabilities. For non-connected vehicles, that is, those that are not equipped with or fail to effectively use communication systems, the recognition and sharing of driving intentions remains a technical challenge.
[0003] Existing intelligent connected multi-vehicle collaborative driving decisions mostly rely on their own communication capabilities, and the recognition and sharing of driving intentions of non-connected vehicles remains a technical challenge. Since non-connected vehicles lack the ability to communicate directly with other vehicles or infrastructure, even intelligent connected vehicles equipped with advanced communication systems are subject to certain constraints in their driving decisions.
[0004] Some studies focus on modeling and predicting vehicle behavior, trying to infer the vehicle's driving intentions through its historical driving data and environmental information. Other studies are committed to developing more advanced sensors and algorithms to improve the perception of the driving status of non-connected vehicles. However, these methods often have problems such as low accuracy, poor real-time performance, or high cost.
[0005] Mobile edge computing (MEC), as an emerging computing architecture, provides a new approach to solving the above problems by sinking computing power to the edge of the network. The rapid development of 5G communication technology, with its high bandwidth, low latency and high reliability, provides strong support for high-speed data transmission and real-time communication in the Internet of Vehicles. As a key facility in the Internet of Vehicles, the roadside unit (RSU) can be deployed on both sides of the road to collect vehicle driving data and realize real-time monitoring and analysis of vehicle driving status through collaboration with MEC and 5G networks. Summary of the invention
[0006] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a driving intention recognition and sharing method applied to non-networked vehicles, the method comprising: collecting driving data of non-networked vehicles through RSU equipment, and transmitting the driving data to MEC nodes in real time through the 5G communication network; the MEC node pre-processes the input data, and adopts a bidirectional LSTM network and an extreme gradient boosting algorithm to obtain the driving intention of the non-networked vehicle; broadcasting the driving intention of the non-networked vehicle to the OBU of the intelligent networked vehicle through the 5G network; the intelligent networked vehicle makes corresponding driving decisions according to the received driving intention of the non-networked vehicle.
[0007] A driving intention recognition and sharing system applied to non-networked vehicles, the system comprising: a real-time driving data acquisition unit, an intelligent driving intention analysis module, a vehicle networking intention sharing service module, and a collaborative driving behavior execution unit;
[0008] The real-time driving data collection unit is composed of a roadside RSU, and the driving data of non-networked vehicles is collected through the RSU;
[0009] The intelligent driving intention analysis module uses the Bi-LSTM algorithm model on the MEC node to identify the driving intention of the non-connected vehicle based on the collected driving data;
[0010] The vehicle networking intention sharing service module broadcasts the identified driving intention to the intelligent connected vehicles within the relevant lane range through the 5G MEC node;
[0011] The collaborative driving behavior execution unit enables the connected vehicle to make decisions using the decision support system of the MEC node and execute corresponding driving behaviors based on the received driving intention information of the non-connected vehicle.
[0012] Beneficial effects of the present invention:
[0013] The present invention recognizes and shares the driving intentions of non-connected vehicles in real time, so that intelligent connected vehicles can respond in a timely manner, thereby reducing the occurrence of traffic accidents and improving the overall safety of road use; by utilizing MEC nodes for data processing and analysis, the need for data transmission to remote servers is reduced, latency is reduced, and response speed is improved; mobile edge computing reduces dependence on central cloud resources and optimizes resource utilization efficiency through distributed computing; the real-time data and analysis results provided by the system can assist traffic management departments in optimizing traffic signal control, conducting traffic flow analysis and prediction, and achieving more intelligent traffic management; the present invention uses the Adam optimization algorithm to optimize the loss function of the Bi-LSTM driving intention feature extraction model, thereby improving the training efficiency and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1A flow chart of the method for identifying and sharing driving intentions of non-connected vehicles of the present invention;
[0015] Figure 2 This is a schematic diagram of the structure of the non-networked vehicle driving intention recognition sharing system of the present invention;
[0016] Figure 3 The technical roadmap of the non-connected vehicle driving intention recognition sharing system of the present invention;
[0017] Figure 4 This is an overall schematic diagram of the non-networked vehicle driving intention recognition and sharing system of the present invention. DETAILED DESCRIPTION
[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0019] A driving intention recognition and sharing method applied to non-networked vehicles, the method comprising: collecting driving data of non-networked vehicles through RSU equipment, and transmitting the driving data to MEC nodes in real time through a 5G communication network; the MEC nodes preprocess the input data, and use a bidirectional LSTM network and an extreme gradient boosting algorithm to obtain the driving intention of the non-networked vehicle; broadcasting the driving intention of the non-networked vehicle to the OBU of an intelligent networked vehicle through a 5G network; and the intelligent networked vehicle makes corresponding driving decisions according to the received driving intention of the non-networked vehicle.
[0020] Among them, the Extreme Gradient Boosting algorithm (XGBoost for short) is an ensemble learning algorithm based on decision trees, which is widely used in classification and regression tasks. It is known for its efficient computing and powerful performance. The Extreme Gradient Boosting algorithm of this application is an improved and optimized Extreme Gradient Boosting algorithm.
[0021] Specifically, Figure 1 and Figure 2 As shown, a non-connected vehicle driving intention recognition and sharing method based on 5G MEC and RSU includes the following steps:
[0022] S1. The non-connected vehicle target vehicle enters the information collection range of the RSU unit, and the sensors deployed by the RSU collect data from its target vehicle.
[0023] Specifically, the non-connected vehicle target vehicle enters the information collection range of the RSU unit. The RSU is deployed at a key road location and is equipped with a high-definition camera and a high-precision lidar sensor. The sensor data acquisition unit of the RSU synchronizes the timestamp to ensure the timing accuracy of the data. The camera and lidar monitor and record the driving status of the non-connected vehicle in real time, including the driving characteristics of the target vehicle and the road environment characteristics. The target vehicle information includes the steering wheel angle, steering wheel steering torque, steering wheel speed, speed, acceleration, lateral and longitudinal relative displacement, and position coordinates. The surrounding environment information includes the speed of the cooperative vehicle in the current and adjacent lanes, and the distance between the cooperative vehicle and the target vehicle in the current and adjacent lanes. Let F be the feature matrix, where each row f i Represents at a specific time point t i The state vector of the target vehicle collected, each column represents a specific feature, example feature vector f i It can be expressed as follows:
[0024] f i =[θ i Δy i M i ω i v i a i Δx i V c,i V a,i D c,i D a,i x i y i ] T
[0025] Among them, θ i is the vehicle at time t i Steering wheel angle; M i is the vehicle at time t i Steering wheel torque; ω i is the vehicle at time t i Steering wheel speed; v i is the instantaneous speed of the vehicle at a point in time; a i is the vehicle at time t i The instantaneous acceleration of i is the vehicle at time t i The lateral distance from the centerline of its path; Δy i is the vehicle at time t i The longitudinal distance to the vehicle in front; V c,i is the cooperative vehicle in the current lane at time point t i Speed; V a,i is the cooperative vehicle in the adjacent lane at time t i Speed; Dc,i is the time between the cooperative vehicle and the target vehicle in the current lane at time t i Distance; D a,i is the time between the cooperative vehicle and the target vehicle in the adjacent lane at time point t i The distance (x i ,y i ) is the target vehicle at time t i The location coordinates of .
[0026] S2. The data processing unit of the MEC node includes a data preprocessor and a feature extractor, which are used to improve data quality and extract effective features. Specifically, the data preprocessor is responsible for receiving the raw data from the RSU; data cleaning is performed to remove invalid or erroneous data points; wavelet transform denoising technology is used to reduce the impact of sensor noise; data interpolation or filling should also be included to handle missing values. The feature extractor extracts key features that are helpful for driving intention recognition from the preprocessed data; and uses time domain analysis, frequency domain analysis, and time-frequency analysis methods to extract the characteristics of the vehicle's driving mode.
[0027] S3. Bidirectional long short-term memory network (Bi-LSTM) is used as the core model for driving intention feature extraction. This model consists of two independent long short-term memory (LSTM) sub-networks and is specially designed to extract key features from the driving data of non-connected vehicles in order to more accurately identify driving intentions.
[0028] Specifically, the training of the Bi-LSTM driving intention feature model includes: obtaining a training data set, in which the data in the training data set are vehicle driving data; inputting the training data into the Bi-LSTM model, and each LSTM sub-network independently extracts features, respectively, from the beginning to the end and from the end to the beginning of the time series; splicing the feature vectors extracted by the two sub-networks to form a comprehensive feature vector, which integrates the comprehensive information of the vehicle driving data; further inputting the spliced feature vector into the deep structure of the model to perform deeper feature extraction and enhance the model's ability to understand complex driving behaviors; using the improved Adam optimization algorithm to minimize the loss function, and dynamically adjusting the learning rate during the training process to accelerate the convergence speed and improve the model performance; when the loss function reaches a predetermined minimum value, or the model performance is no longer significantly improved, the training process is stopped, and the model parameters at this time are considered to be optimal, and the model training is completed.
[0029] In this embodiment, the LSTM model processes data including:
[0030] Initialize and update the hidden state h t and cell state C t ;
[0031] The forget gate f is calculated by the following formula t ;
[0032] f t =σ(W f ·[h t-1 ,x t ]+b f )
[0033] The input gate i is calculated by the following formula t and candidate cell states
[0034] i t =σ(W i ·[h t-1 ,x t ]+b i )
[0035]
[0036] Update cell status C t ;
[0037]
[0038] Calculate the output gate o t ;
[0039] o t =σ(W o ·[h t-1 ,x t ]+b o )
[0040] Generate a new hidden state h t ;
[0041] h t =o t *tanh(C t )
[0042] Among them, f t is the output of the forget gate at time step t, W f is the weight matrix of the forget gate, [h t-1 ,x t ] is the hidden state h of the previous time step t-1 and the input data x at the current time step t Combination of f is the bias term of the forget gate, σ is the sigmoid activation function, which maps the input to between 0 and 1, indicating the degree of forgetting; the input gate consists of a sigmoid layer and a tanh layer, i t is the output of the input gate at time step t, Wi and W C are the weight matrices of the input gate and candidate cell states, respectively, and b i and b C are the bias terms of the input gate and candidate cell state, respectively. Tanh is the hyperbolic tangent activation function, which is used to introduce nonlinearity and generate new candidate cell states. C t is the cell state at the current time step, C t-1 is the cell state at the previous time step, f t With C t-1 Multiplication means forgetting the past cell state information, i t and Multiplication means adding new candidate cell state information; output gate o t is the output of the output gate at time step t, W o is the weight matrix of the output gate, b o is the bias term of the output gate; the hidden state h t is the hidden state of the current time step, tanh(C t ) is a nonlinear transformation of the current cell state; o t Determines how much information the hidden state should obtain from the cell state.
[0043] Since LSTM can only use past information and ignore future information, in order to simultaneously use the past and future information of time series data, a Bi-LSTM model is established. The model contains two independent LSTM networks. The concatenation of feature vectors is to combine the feature vectors extracted by the forward LSTM and the reverse LSTM to obtain more comprehensive sequence information; the forward LSTM processes the sequence from x1 to x t , and generates a hidden state at each time step t The reverse LSTM processes the sequence from x t to x1, and generates a hidden state at each time step t For each time step t, the feature vectors extracted by the forward and backward LSTMs are and are concatenated together to form a single feature vector h t .
[0044] Using weighted sum to merge feature vectors can be expressed as:
[0045]
[0046] Among them, α is a weight parameter (usually between 0 and 1) that is used to control the contribution of the forward and reverse feature vectors when merging.
[0047] These feature vectors are input into a hidden layer for deeper feature extraction; the expression is:
[0048] z=W·h t +b
[0049] a=σ(z)
[0050] Where W is the weight matrix; b is the bias term; σ is the activation function Sigmoid.
[0051] In order to improve the efficiency of model training and reduce the impact of erroneous output on subsequent network input, a proportional planned sampling method is adopted. This method directly inputs labeled data into the Bi-LSTM network to optimize the training process and accelerate convergence. In order to determine the appropriate ratio, the Adam optimization algorithm is used to optimize the Bi-LSTM loss function. The ratio corresponding to the minimum loss function is the optimal solution.
[0052] For the system in the scenario where the implementation scheme needs to process a large amount of real-time data and respond quickly, it is necessary to ensure the computational efficiency of the optimization algorithm in order to achieve fast processing in the edge computing environment. Therefore, the Adam optimization algorithm is used, which has the advantage that it automatically adjusts the learning rate of each parameter, adjusts the update step size through historical gradient information, helps the model converge faster, and is insensitive to the initial value of the learning rate.
[0053] The Adam optimization algorithm is used to optimize the parameters in the model, including:
[0054] Step 1: Initialize model parameters θ0; initialize first-order moment m0 = 0 and second-order moment v0 = 0; set learning rate η, first-order moment decay rate β1, second-order moment decay rate β2, loss function J(θ t ) and the numerical stability parameter ∈.
[0055] Step 2: For each training batch, calculate the gradient of the model at the current moment according to the loss function
[0056] Step 3: Update the first-order moment and second-order moment according to the model gradient:
[0057]
[0058] Step 4: Calculate the updated unbiased estimates of the first-order moment and the second-order moment:
[0059]
[0060] Step 5: Update the model parameters based on unbiased estimation:
[0061]
[0062] Step 6: Repeat steps 1 to 5 until the performance of the model on the validation set no longer improves or the preset number of iterations is reached.
[0063] Where m t is the first-order moment estimate of the tth step; v t is the second-order moment estimate at step t; is the corrected first-order moment estimate; is the modified second-order moment estimate; θ t is the parameter of step t.
[0064] Based on the driving intention feature extraction model, the MEC node uses the XGBoost algorithm to optimize the driving intention recognition model. By iteratively training weak classifiers (decision trees), each weak classifier is trained on the residual of the previous classifier. The prediction results of all weak classifiers are weighted and summed to form a strong classifier to improve the prediction accuracy of the overall model. The objective function of the driving intention recognition model includes the loss function and the regularization part:
[0065]
[0066] In each iteration, a new weak predictor is added To update the current model
[0067]
[0068] For new weak predictors The contribution of the second-order Taylor expansion is:
[0069]
[0070] The second-order Taylor expansion of the new weak predictor provides information about the curvature of the loss function, which helps to select the best features and thresholds when splitting nodes in the tree.
[0071] For each leaf node j, the optimal score This can be obtained by solving the following optimization:
[0072]
[0073] Among them, l is the loss function; y i is the actual driving intention of the i-th sample; is the driving intention predicted by the model; f is a log function; N is the number of training samples; Ω reg (f) is the regularization term; and I jis the index set of samples on leaf node j, γ and λ are regularization parameters; w is the weight used to adjust the influence of the residual of each sample.
[0074] The loss function of the model is expressed as:
[0075]
[0076] The present invention uses the Adam optimization algorithm to accurately adjust the data sampling input ratio, and then optimizes the Bi-LSTM feature extraction model and the XGBoost driving intention recognition model. In this way, the model constructed by the present invention shows higher accuracy in identifying the driving intention of non-connected vehicles. Compared with traditional or single deep learning models, the Bi-LSTM model of the present invention is trained by optimizing the data sampling ratio, which enhances the ability to extract key features from driving data; the strong classifier constructed by the XGBoost algorithm is combined with the features extracted by Bi-LSTM to improve the accuracy of driving intention recognition; the recognition results are shared in real time with intelligent connected vehicles through the 5G network, so that the cooperative vehicles can respond to the driving behavior of non-connected vehicles in a timely manner; by accurately identifying and sharing driving intentions, the model of the present invention helps intelligent connected vehicles make safer driving decisions and improve the traffic efficiency of road traffic.
[0077] S4. After the non-connected vehicle target vehicle completes the driving intention recognition, the system on the MEC node formats the recognition results to adapt to the broadcast requirements of the 5G network, ensures the stability of the connection between the MEC node and the 5G network, configures the broadcast parameters, and prepares to broadcast the driving intention information to the intelligent connected vehicle. Using the low latency characteristics of the 5G network, the driving intention information is broadcast to the intelligent connected vehicle in real time. After receiving the driving intention information, the intelligent connected vehicle synchronizes it with its own driving data to achieve real-time response to the traffic environment.
[0078] S5. The OBU carried by the intelligent connected car receives the driving intention information of the non-connected car broadcasted by the MEC node through the 5G network, and integrates the received driving intention information with the vehicle's own perception data to obtain a comprehensive understanding of the traffic environment, parse the acquired driving intention information, and determine the possible driving behavior of the non-connected car: turn left, keep going straight or turn right. The decision results are executed through the vehicle's control system, such as adjusting the speed, sending lane change signals or taking avoidance measures. When executing driving behaviors, the surrounding environment and vehicle status are monitored in real time to ensure safety and adjust decisions as needed. After the non-connected car target vehicle leaves the RSU collection range, the intention information sharing stops.
[0079] In this embodiment, the OBU parses the received information and extracts key driving intention parameters, such as lane change, acceleration or deceleration intention. The on-board system of the intelligent connected vehicle evaluates the current driving status of the vehicle, including speed, position, surrounding traffic environment, etc. Based on the received driving intention information and the vehicle's own status, the on-board decision-making system formulates a corresponding driving strategy. After confirming that the driving intention information is correct and the risk is controllable, the intelligent connected vehicle executes the corresponding driving behavior, such as adjusting the vehicle speed, changing lanes, etc.
[0080] In this embodiment, a corresponding driving strategy and a feedback mechanism are formulated, and the intelligent connected vehicle (OBU) sends feedback to the MEC node based on the received driving intention information and actual driving conditions for continuous learning and optimization of the model.
[0081] A driving intention recognition and sharing system applied to non-networked vehicles, the system comprising: a real-time driving data acquisition unit, an intelligent driving intention analysis module, a vehicle networking intention sharing service module, and a collaborative driving behavior execution unit;
[0082] The real-time driving data collection unit is composed of a roadside RSU, and the driving data of non-networked vehicles is collected through the RSU;
[0083] The intelligent driving intention analysis module uses the Bi-LSTM algorithm model on the MEC node to identify the driving intention of the non-connected vehicle based on the collected driving data;
[0084] The vehicle networking intention sharing service module broadcasts the identified driving intention to the intelligent connected vehicles within the relevant lane range through the 5G MEC node;
[0085] The collaborative driving behavior execution unit enables the connected vehicle to make decisions using the decision support system of the MEC node and execute corresponding driving behaviors based on the received driving intention information of the non-connected vehicle.
[0086] In this embodiment, the system includes at least: an RSU for collecting driving data of non-connected vehicles; a 5G communication network for realizing high-speed data transmission between the RSU and the MEC node; an MEC node, including a data processing unit and a driving intention recognition unit, for processing the data transmitted by the RSU and identifying the driving intention; an OBU installed on the intelligent connected vehicle, for receiving and responding to the driving intention information broadcast by the MEC node; and a central monitoring unit for monitoring the operating status of the entire system and performing manual intervention when necessary.
[0087] In a specific embodiment, Figure 3As shown in the technical roadmap of the present invention, RSU uses cameras and lidar to collect real-time driving data of non-connected vehicles; quickly transmits data to MEC nodes through 5G networks; MEC nodes preprocess and extract features of the received data, use Bi-LSTM models to analyze feature data, and identify driving intentions; broadcast driving intentions to smart connected vehicles through 5G networks; the OBU of smart connected vehicles receives driving intention information, and smart connected vehicles execute driving decisions and behaviors based on the received information. The overall schematic diagram is as follows Figure 4 shown.
[0088] In this embodiment, the three driving intentions of turning left, keeping straight and turning right are mainly identified. In this embodiment, the simulation parameters shown in Table 1 and Table 2 are used:
[0089] Table 1 Bi-LSTM parameter settings
[0090]
[0091] Table 2 XGBoost parameter settings
[0092]
[0093] Based on the model parameters, the recognition results shown in Table 3 can be obtained:
[0094] Table 3 Driving intention recognition results
[0095]
[0096]
[0097] In Table 3, the accuracy of the three driving intentions of turning left, keeping straight, and turning right is over 97%, among which the accuracy of the intention to change lanes to the right is as high as nearly 98%. This high accuracy may be attributed to the fact that the characteristics of the vehicle are more obvious when changing lanes to the right, making it easier to be captured and extracted by the model. In actual traffic scenarios, accurate recognition of lane change intentions plays an important role in intention sharing and response requests of collaborative vehicles. High-accuracy lane change intention recognition can provide timely driving information for intelligent connected vehicles and help them make safer and more reasonable driving decisions. The recognition results of the embodiments of this patent not only conform to the actual situation, but also improve the training efficiency and prediction accuracy of the model through optimized data sampling and algorithm improvement, providing strong technical support for intelligent transportation systems.
[0098] The system implementation of the present invention is the same as the method implementation.
[0099] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium, which can include: ROM, RAM, disk or CD, etc.
[0100] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A driving intention recognition and sharing method applied to non-networked vehicles, characterized in that: include: The driving data of non-networked vehicles is collected through RSU devices, and transmitted to MEC nodes in real time through the 5G communication network; the MEC nodes pre-process the input data and use a bidirectional LSTM network and extreme gradient boosting algorithm to obtain the driving intention of non-networked vehicles; the driving intention of non-networked vehicles is broadcast to the OBU of smart networked vehicles through the 5G network; the smart networked vehicles make corresponding driving decisions based on the received driving intention of non-networked vehicles.
2. The driving intention recognition and sharing method for non-networked vehicles according to claim 1, characterized in that: The driving data of non-connected vehicles includes the driving characteristics of the target vehicle and the road environment characteristics; The target vehicle information includes steering wheel angle, steering wheel steering torque, steering wheel speed, speed, acceleration, lateral and longitudinal relative displacement and position coordinates; the surrounding environment information includes the speed of the cooperative vehicles in the current and adjacent lanes, and the distance between the cooperative vehicles and the target vehicle in the current and adjacent lanes.
3. The driving intention recognition and sharing method for non-networked vehicles according to claim 1, characterized in that: The MEC node performs driving intention recognition on input data, including: using the Bi-LSTM network intention recognition model to process time series data to obtain driving intention recognition results; the Bi-LSTM network intention recognition model consists of two independent LSTM network structures and an optimized XGBoost model; using the trained Bi-LSTM network intention recognition model to extract features from the preprocessed data to obtain data features; using the optimized XGBoost model to perform driving intention recognition on the data features to obtain recognition results.
4. The driving intention recognition and sharing method for non-networked vehicles according to claim 3 is characterized in that: Training the Bi-LSTM network includes: Step 1: Obtain historical driving data of vehicle users; Step 2: Input the historical driving data of the vehicle user into the forward LSTM network and the reverse LSTM network respectively to extract hidden features; Step 3: Fuse the feature vectors extracted by the forward LSTM and the backward LSTM to obtain comprehensive sequence information; Step 4: Input the comprehensive sequence information into the hidden layer for deep feature extraction; Step 5: Calculate the loss function of the model based on the deep feature map; Step 6: Use the Adam optimization algorithm to optimize the parameters in the model, and complete the model training when the loss function converges.
5. The driving intention recognition and sharing method for non-networked vehicles according to claim 4, characterized in that: The fusion of feature vectors extracted by forward LSTM and backward LSTM includes: in, is the hidden feature output by the forward LSTM network, is the hidden feature output by the reverse LSTM network, and α is a weight parameter.
6. The driving intention recognition and sharing method for non-networked vehicles according to claim 4, characterized in that: The Adam optimization algorithm is used to optimize the parameters in the model, including: Step 1: Initialize model parameters θ0, first-order moment m0 and second-order moment v0; set learning rate η, first-order moment decay rate β1, second-order moment decay rate β2, loss function J(θ t ) and numerical stability parameter ∈; Step 2: Calculate the gradient of the model at the current moment based on the loss function Step 3: Update the first-order moment and the second-order moment according to the model gradient; Step 4: Calculate the updated unbiased estimates of the first-order moment and the second-order moment; Step 5: Update the model parameters based on unbiased estimation; Step 6: Repeat steps 1 to 5 until the model performance no longer improves.
7. The driving intention recognition and sharing method for non-networked vehicles according to claim 3 is characterized in that: The optimized XGBoost algorithm is used to identify driving intentions based on data features. The optimized XGBoost algorithm is composed of multiple weak classifiers, where the weak classifier is a decision tree. The prediction results of all weak classifiers are weighted and summed to form a strong classifier. The specific process includes setting the objective function of the driving intention recognition model, which includes a loss function and a regularization part. In the iterative training of the driving intention recognition model, a new weak predictor is added in each iterative training process. Update the current model Calculate the optimal score for each leaf node j of the decision tree in the current model All weak classifiers are assigned weights according to the optimal scores, and all prediction results are weighted fused according to the assigned weights to obtain the user's driving intention recognition result.
8. The driving intention recognition and sharing method for non-networked vehicles according to claim 7, characterized in that: The objective function expression is: Among them, Ω is the objective function, f is the log function, N is the number of training samples, l is the loss function, y i is the actual driving intention of the i-th sample, is the driving intention predicted by the model, Ω reg (f) is the regularization term.
9. The driving intention recognition and sharing method for non-networked vehicles according to claim 7, characterized in that: Best score The expression is: Among them, I j is the index set of samples on leaf node j, γ and λ are regularization parameters, The loss function is The first-order derivative at , w is the weight, which is used to adjust the influence of each sample residual; The loss function is The second derivative of .
10. A driving intention recognition sharing system applied to a non-networked vehicle, the system being used to execute a driving intention recognition sharing method applied to a non-networked vehicle as claimed in any one of claims 1 to 9, characterized in that: include: Real-time driving data collection unit, intelligent driving intention analysis module, Internet of Vehicles intention sharing service module and collaborative driving behavior execution unit; The real-time driving data collection unit is composed of a roadside RSU, and the driving data of non-networked vehicles is collected through the RSU; The intelligent driving intention analysis module uses the Bi-LSTM algorithm model on the MEC node to identify the driving intention of the non-connected vehicle based on the collected driving data; The vehicle networking intention sharing service module broadcasts the identified driving intention to the intelligent connected vehicles within the relevant lane range through the 5G MEC node; The collaborative driving behavior execution unit enables the connected vehicle to make decisions using the decision support system of the MEC node and execute corresponding driving behaviors based on the received driving intention information of the non-connected vehicle.
Citation Information
Patent Citations
Continuous predicating method and system for steering intention of driver
CN111976733A
Bus driving risk assessment and early warning system and method based on intelligent network connection
CN117746632A
Method of predicting the future accident risk rate of the drivers using artificial intelligence and its device
US20230267549A1
Cited By
CAV lane changing method and system based on macro and micro traffic flow information
CN119920104A
