Vehicle control method and device, vehicle and storage medium

Abstract

The invention relates to the technical field of intelligent driving, and discloses a vehicle control method and device, a vehicle and a storage medium, the method comprises the following steps: collecting current multi-source driving data, and fusing the multi-source driving data to obtain fused driving data; performing coupling processing on road section information of different dimensions in the fused driving data to obtain corresponding driving scene data under the current road condition; inputting the driving scene data and the fused driving data into a preset neural network model, and predicting an occurrence probability and a risk level of a preset risk driving scene in a future road section; and based on the occurrence probability and the risk level, generating a corresponding vehicle control strategy to control the vehicle. According to the technical scheme, the slope and curve coupling scene can be effectively handled, and the vehicle control precision and safety are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, specifically to a vehicle control method, device, vehicle, and storage medium. Background Technology

[0002] In intelligent driving systems, forward-looking planning solutions generally focus on the independent processing of single road elements, such as optimizing slope control or cornering strategies separately. However, in real-world road environments, especially in mountainous areas, slope and curves often coexist in complex coupled forms, creating composite scenarios such as long downhill continuous sharp curves and uphill gentle curves. These coupled characteristics make traditional methods difficult to effectively address. Summary of the Invention

[0003] This application provides a vehicle control method, device, vehicle, and storage medium that can effectively address scenarios involving slope and curve coupling, and improve vehicle control accuracy and safety.

[0004] Firstly, this application provides a vehicle control method, which includes: collecting current multi-source driving data and fusing the multi-source driving data to obtain fused driving data; coupling road segment information of different dimensions in the fused driving data to obtain corresponding driving scenario data under the current road conditions; inputting the driving scenario data and the fused driving data into a preset neural network model to predict the probability and risk level of the occurrence of preset risk driving scenarios in future road segments; and generating a corresponding vehicle control strategy to control the vehicle based on the probability and risk level.

[0005] Based on the above technical means, the technical solution in this implementation can effectively avoid the inappropriate control strategy caused by traditional single road element planning, and achieve a balance between dynamics and comfort in complex coupled scenarios. In one optional implementation, the road segment information of different dimensions includes slope nodes and curve nodes. The road segment information of different dimensions in the fused driving data is coupled to obtain the corresponding driving scenario data under the current road conditions. This includes: coupling the slope nodes and the corresponding curve nodes to obtain coupled node data; and processing the coupled node data based on a graph neural network to obtain the probability of occurrence and risk level of the corresponding driving scenario under the current road conditions.

[0006] Based on the above-mentioned technical means, not only is the ability to perceive and understand complex road conditions improved, but a reliable basis is also provided for generating more accurate and safer vehicle control strategies, thereby improving the safety and driving experience of intelligent driving vehicles in complex road environments. In one optional implementation, the slope node and the corresponding curve node are coupled to obtain coupled node data, including: transmitting the slope information and slope change rate information of the slope node to the adjacent curve node through the coupling model to obtain coupled curve node; and transmitting the curvature information and curvature change rate information of the coupled curve node to the adjacent slope node again through the coupling model to obtain coupled node data.

[0007] Based on the above technical means, the problem of insufficient coupling caused by one-way information transmission is effectively avoided, ensuring that the obtained coupling node data can comprehensively and accurately reflect the real characteristics and interaction of the slope and curve composite scenario.

[0008] In one optional implementation, the preset risk driving scenarios include long slope continuous curves, long slope sharp curves, and gentle slope curves. Based on a graph neural network, the coupled node data is processed to obtain the probability of occurrence and risk level of the corresponding driving scenarios under the current road conditions. This includes: processing the coupled node data based on a fully connected layer of the graph neural network to obtain the probability of occurrence and corresponding risk level of long slope continuous curves, long slope sharp curves, and gentle slope curves under the current road conditions.

[0009] Based on the aforementioned technical means, detailed scenario classification and risk quantification enable subsequent vehicle control strategies to coordinate and control the power system, steering system, and braking system in a targeted manner.

[0010] In one optional implementation, driving scenario data and fused driving data are input into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments. This includes: inputting the probability and corresponding risk level of driving scenarios, current vehicle driving segment data, and corresponding high-definition map data into the preset neural network model to predict the probability and corresponding risk level of long slope continuous curves, long slope sharp curves, and gentle slope curves in future road segments.

[0011] Based on the aforementioned technical means, a forward-looking risk prediction mechanism enables the vehicle control system to obtain risk information about future road sections in advance, thus allowing sufficient time for prediction and adjustment.

[0012] In one optional implementation, a corresponding vehicle control strategy is generated to control the vehicle based on the probability of occurrence and the risk level, including: coordinating the control of the vehicle's power system, steering system and braking system based on the current driving mode and in combination with the predicted probability of occurrence and risk level of preset risk driving scenarios in future road segments.

[0013] Based on the aforementioned technical means, coordinated control of the vehicle's powertrain, steering, and braking systems has been achieved. This coordinated control mechanism can effectively solve the problem of simplistic or uncoordinated control strategies in complex and risky driving scenarios.

[0014] In one optional implementation, based on the current driving mode and combined with the predicted probability and risk level of pre-set risky driving scenarios in future road segments, the vehicle's powertrain, steering system, and braking system are coordinated and controlled, including: based on the current driving mode, controlling the vehicle's powertrain according to the following expression: ,in, For the limited torque, For maximum torque, R represents the risk sensitivity coefficient and the risk level. Based on the current driving mode, the vehicle's steering system is controlled according to the following expression: ,in, To enhance the electric power steering force, Where m represents the maximum electric power steering effort, R represents the power steering enhancement coefficient, and m represents the risk level. Based on the current driving mode, the vehicle's braking system is controlled according to the following expression: ,in, Braking pressure, Based on braking pressure, R represents the pressure adjustment factor and the risk level.

[0015] Based on the aforementioned technical means, by adjusting quantitative parameters, the collaborative control becomes more precise and adaptive, effectively improving the operational safety and control efficiency of vehicles in complex and risky scenarios.

[0016] In one optional implementation, the multi-source driving data includes at least one of the following: vehicle visual data, vehicle radar data, vehicle cloud communication interface data, vehicle status data, driving behavior data, and high-definition map data; collecting the current multi-source driving data includes: sequentially downloading high-definition map data to the local vehicle according to different road segments traveled by the vehicle.

[0017] Based on the above technical means, not only is the efficiency of data acquisition improved, but the quality and real-time performance of the fused driving data are also ensured, thus providing a reliable foundation for subsequent risk prediction and vehicle control strategy generation. This avoids inaccurate control strategies caused by missing or delayed map data, and improves the safety, accuracy and robustness of vehicle control.

[0018] In one optional implementation, high-definition map data is downloaded to the local vehicle sequentially according to different road segments traveled by the vehicle, including: When the local high-definition map data for the current route is insufficient, a task to download high-definition map data for the next route is triggered. The high-definition map data is compressed using a preset algorithm to reduce the amount of data transmitted.

[0019] Based on the above technical means, it is ensured that the vehicle can always obtain timely, complete and efficient high-definition map data support in complex driving scenarios, providing a reliable foundation for the fusion of multi-source driving data and vehicle control, thereby improving the safety, real-time performance and economy of the intelligent driving system.

[0020] In one optional implementation, the high-definition map data is compressed using a preset algorithm, including: decomposing the high-definition map data into low-frequency high-definition map data and high-frequency high-definition map data according to the frequency of occurrence of road segment information in different dimensions in the high-definition map data; retaining the low-frequency high-definition map data and retaining the high-frequency high-definition map data according to a preset threshold; and reconstructing the retained high-frequency high-definition map data and low-frequency high-definition map data according to the inverse transformation of the preset algorithm to obtain compressed high-definition map data.

[0021] Based on the above-mentioned technical means, not only is the download efficiency and timeliness of high-definition map data improved, especially when the map data of the next road segment needs to be dynamically downloaded during vehicle operation, but it also provides a more accurate and reliable data foundation for subsequent risk scenario prediction, thereby improving the effectiveness and safety of vehicle control strategies.

[0022] In one optional implementation, when the local high-definition map data for the current travel segment of the vehicle is insufficient, a download task for the high-definition map data for the next travel segment is triggered, including: when the local high-definition map data for the current travel segment of the vehicle is insufficient, determining the time point for triggering the download task based on the historical average delay time and redundancy time; and triggering the download task for the high-definition map data for the next travel segment based on the time point for triggering the download task.

[0023] Based on the aforementioned technical means, risk prediction and control strategy generation can be performed more accurately, thereby significantly improving the decision-making quality and driving safety of intelligent driving vehicles.

[0024] In one optional implementation, after triggering the download task of high-definition map data for the next driving segment, the method further includes: discarding the high-definition map data of the next driving segment that the vehicle has already passed; if the data transmission through the vehicle-to-cloud communication interface conflicts with the download of high-definition map data, the data transmitted through the vehicle-to-cloud communication interface shall be transmitted first.

[0025] Based on the above technical means, key instructions and real-time information from the cloud can be received and responded to in a timely manner, avoiding control delays or interruptions caused by map download tasks occupying communication bandwidth, and improving the timeliness and reliability of vehicle control strategies.

[0026] Secondly, this application provides a vehicle control device, the device comprising: a fusion module, used to collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data; The coupling module is used to couple road segment information from different dimensions in the fused driving data to obtain the corresponding driving scenario data under the current road conditions; the prediction module is used to input the driving scenario data and the fused driving data into a preset neural network model to predict the probability and risk level of the occurrence of preset risk driving scenarios in future road segments; the control module is used to generate corresponding vehicle control strategies to control the vehicle based on the probability and risk level.

[0027] Thirdly, this application provides a vehicle, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the vehicle control method of the first aspect or any corresponding embodiment described above.

[0028] Fourthly, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the vehicle control method of the first aspect or any corresponding embodiment described above.

[0029] Fifthly, this application provides a computer program product, including computer instructions for causing a computer to execute the vehicle control method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0031] Figure 1 This is a schematic flowchart of a first embodiment of a vehicle control method according to this application. Figure 2 This is a second flowchart illustrating a vehicle control method according to an embodiment of this application; Figure 3 This is a schematic diagram of a third process of a vehicle control method according to an embodiment of this application; Figure 4 This is a schematic flowchart of the fourth vehicle control method according to an embodiment of this application; Figure 5 This is a fifth flowchart illustrating the vehicle control method according to an embodiment of this application; Figure 6 This is a structural block diagram of a vehicle control device according to an embodiment of this application; Figure 7 This is a schematic diagram of the hardware structure of the vehicle according to an embodiment of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] It is understood that before using the technical solutions disclosed in the various embodiments of this application, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this application in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0034] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0035] In related technologies, the forward-looking planning of intelligent driving vehicles often focuses on single road elements. However, in actual mountainous roads, slope and curves are often coupled, forming complex composite scenarios. This can lead to inappropriate control strategies. For example, if control is based solely on slope information at the end of a long downhill sharp curve, it will pose a safety hazard. Simultaneously, in uphill curves, power requirements and lateral stability requirements are coupled, making it difficult to balance power and comfort with a single slope-based planning approach. Related technologies lack efficient methods for processing two-dimensional coupled data of slope and curves and for adaptive scenario classification and decision-making.

[0036] In response, this application proposes a vehicle control method, such as... Figure 1 As shown, the method includes: Step S101: Collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data.

[0037] Multi-source driving data refers to a collection of data about vehicle operation and the environment acquired from multiple sources, including internal vehicle sensors (such as vision sensors and LiDAR), vehicle status sensors, driving behavior records, and external communication interfaces (such as vehicle-to-cloud communication interfaces). This data provides comprehensive information input for vehicle decision-making.

[0038] Fusion driving data refers to a consistent, complete, and reliable dataset formed by integrating, calibrating, and deduplicating multi-source driving data from different sources using specific algorithms. This dataset is used for subsequent road condition analysis and scene recognition.

[0039] Step S102: Couple the road segment information of different dimensions in the fused driving data to obtain the corresponding driving scenario data under the current road conditions.

[0040] Among these, road segment information of different dimensions refers to various parameters describing road characteristics, such as longitudinal slope, lateral curvature, road surface friction coefficient, lane line information, etc. These dimensions collectively characterize the geometric shape and physical properties of the road.

[0041] Coupling processing refers to the correlation analysis and integration of road segment information from different dimensions to reveal their mutual influence and synergistic effects. Through coupling processing, the characteristics of complex road conditions can be understood more comprehensively.

[0042] Step S103: Input the driving scenario data and fused driving data into the preset neural network model to predict the probability and risk level of the occurrence of preset risk driving scenarios in future road segments.

[0043] Driving scenario data refers to data that describes specific driving situations on current or future road sections, obtained by coupling and processing road segment information from different dimensions of fused driving data. This data reflects the specific environment in which the vehicle is located and the potential driving challenges.

[0044] A pre-trained neural network model refers to an artificial intelligence model containing multiple layers of neurons and connection weights, capable of learning and recognizing complex patterns in data. This model is used to analyze driving scenario data and fused driving data to predict risks.

[0045] Preset risk driving scenarios refer to specific road conditions that may be encountered during vehicle operation, which may be potentially dangerous or have special requirements for driving performance, such as continuous curves or steep slopes.

[0046] The probability of occurrence refers to the likelihood of a pre-defined risk driving scenario occurring on a specific road segment in the future, usually expressed as a value between 0 and 1. The risk level refers to the degree of harm or challenge to vehicle control that a pre-defined risk driving scenario may cause once it occurs, usually expressed as a grade or a continuous numerical value.

[0047] Step S104: Based on the probability of occurrence and the risk level, generate a corresponding vehicle control strategy to control the vehicle.

[0048] Among them, vehicle control strategy refers to the set of operating instructions formulated for the vehicle's power system, steering system, braking system and other actuators based on predicted risk information, with the aim of optimizing the vehicle's driving safety, power and comfort.

[0049] Specifically, vehicles can acquire data from multiple sources during operation. For example, they can obtain visual image information through onboard cameras, point cloud data through LiDAR, vehicle status data such as speed, acceleration, and steering wheel angle through their own sensors, and driving behavior data through driver input. This raw data may contain noise, be incomplete, or have inconsistent formats. To obtain a unified and reliable data foundation, this multi-source data can be fused. For example, methods such as weighted averaging, Kalman filtering, or simple rule matching can be used to synchronize and spatially register data from different sensors, removing redundant information to generate comprehensive fused driving data. This fused driving data provides comprehensive input for subsequent road condition analysis and risk prediction.

[0050] Secondly, after obtaining the fused driving data, it is necessary to extract and process road segment information from different dimensions. For example, longitudinal slope and lateral curvature information can be identified from the fused driving data. These different dimensions of information do not exist in isolation but influence each other. To more accurately describe the actual road conditions, this information can be coupled. For example, a rule-based logical model can be established, identifying a specific composite road segment when both the slope and curvature values ​​reach a certain threshold. Alternatively, a simple mathematical model can be used to linearly combine or nonlinearly map the slope and curvature information to generate a comprehensive road segment feature value. Through this coupling process, driving scenario data that reflects the complexity of the current road conditions can be obtained, such as identifying "slope curves" or "continuous curves."

[0051] Furthermore, to predict the risks of future road sections, the driving scenario data obtained above, along with the original fused driving data, can be fed into a pre-trained neural network model. This neural network model can be a multilayer perceptron or a recurrent neural network, which establishes a mapping relationship between input data and risk prediction by learning from a large amount of historical driving data and corresponding road condition risk labels. For example, the model can receive information such as the slope, curvature, and vehicle speed of the current road section, and output the probability value of risk scenarios (such as "sharp bends" and "steep slopes") that may occur within a certain distance in the future, along with the corresponding risk level (such as "low," "medium," or "high"). This prediction process enables the vehicle to perceive potential dangers in advance, providing forward-looking information for subsequent control decisions.

[0052] Finally, after obtaining the probability and risk level of the predicted risk driving scenarios in future road sections, the vehicle control system can generate corresponding control strategies based on this information. For example, when a high-risk sharp bend is predicted in the future road section, the system can generate instructions to decelerate and adjust steering effort. The generation of control strategies can be based on a pre-set lookup table, that is, a set of corresponding control instructions is pre-stored for different risk levels and scenario types. Alternatively, a fuzzy logic controller can be used to output corresponding throttle, braking, and steering instructions based on the input of risk probability and level. These control instructions are sent to the vehicle's actuators, such as the engine, brakes, and electric power steering system, to achieve effective control of the vehicle, thereby improving driving safety, stability, and comfort.

[0053] It is understood that this application, by fusing multi-source driving data and coupling road segment information from different dimensions such as slope and curves, can accurately identify complex and composite scenarios on mountain roads. By using a pre-set neural network model to predict the probability and level of risks occurring in future road segments, vehicles can perceive potential dangers in advance. Therefore, an adaptive vehicle control strategy can be generated based on the prediction results, effectively avoiding the control strategy errors caused by traditional single-road-element planning, and achieving a balance between power and comfort in complex coupled scenarios.

[0054] In some embodiments, this application further proposes a vehicle control method, where road segment information of different dimensions includes slope nodes and curve nodes, such as... Figure 2 As shown, the method includes: Step S201: Collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data. See details... Figure 1 Step S101 in the embodiment will not be described again here.

[0055] Step S202 involves coupling road segment information from different dimensions in the fused driving data to obtain corresponding driving scenario data under the current road conditions, including: Step S2021: Couple the slope node and the corresponding curve node to obtain the coupled node data.

[0056] Slope nodes refer to points or areas on a road that exhibit specific slope characteristics. These nodes typically contain information such as slope value and slope change rate, used to describe the longitudinal undulation of the road. For example, significant changes in slope can be identified using vehicle inertial measurement unit (IMU) data, high-precision map data, or LiDAR point cloud data, serving as slope nodes. Another approach is to divide the road into segments of equal length or distance and calculate the average slope of each segment, using the center point or endpoint of each segment as slope nodes. Curve nodes refer to points or areas on a road that exhibit specific curvature characteristics. These nodes typically contain information such as curvature, rate of curvature change, and curve radius, used to describe the lateral curvature of the road. For example, predefined curve geometry information from high-precision map data, or key points of curves identified by vehicle GPS trajectory data or visual sensors, can be extracted as curve nodes. Another approach is to sample the road centerline and calculate the curvature of the sampled points, using points with curvature exceeding a preset threshold or with significant curvature changes as curve nodes.

[0057] To correlate longitudinal (slope) and lateral (curve) information of roads and reflect the complex interaction between slope and curves in actual roads, this application couples slope nodes and corresponding curve nodes to obtain coupled node data. Coupling processing can more comprehensively describe road segment characteristics and provide more accurate input for subsequent risk assessment. For example, coupling can be achieved through spatial proximity. For each slope node, all relevant curve nodes are searched in its geographical vicinity (e.g., within a preset distance range), and the information of these slope and curve nodes is merged or spliced ​​to form a composite data structure containing multiple dimensions of information such as slope and curvature—that is, coupled node data. Alternatively, coupling can be achieved through time-series correlation. When a vehicle travels through a road segment, the slope and curve information of that segment are recorded, and this information is matched and correlated according to the vehicle's travel sequence or timestamp, thereby generating coupled node data reflecting the coupling characteristics of the slope and curves of that road segment.

[0058] Step S2022: Based on the graph neural network, the coupled node data is processed to obtain the probability of occurrence and risk level of the corresponding driving scenario under the current road conditions.

[0059] Graph Neural Networks (GNNs) are neural network models used to process graph-structured data. They effectively capture the relationships between nodes and the topological structure of the graph, learning node feature representations by propagating information across nodes and edges. In this application, a GNN is used to process associated data of road elements. It effectively captures non-Euclidean structure data (such as the topological relationship between slope nodes and curve nodes in a road network), and mines the coupling features of slope and curves through node information propagation and aggregation. Specifically, the GNN first receives node data coupled with slope and curves, extracts high-dimensional features through multi-layer processing, and then outputs the probability and risk level of scenarios such as long slopes with continuous curves and long slopes with sharp curves under the current road conditions through a fully connected layer, providing accurate basis for subsequent vehicle control strategy generation. For example, a graph can be constructed where each coupled node data is a node, and the edges between nodes represent their connection relationship or spatial proximity relationship on the road. Then, the coupled node data (including features such as slope and curvature) is input into the GNN as the initial features of the nodes. Graph neural networks learn deep representations of each node through multi-layer information propagation and aggregation, and ultimately predict the probability of a specific driving scenario and its corresponding risk level under current road conditions through the output layer. Another approach is to treat different types of coupled nodes (e.g., slope nodes and curve nodes) as different types of nodes in the graph and construct a heterogeneous graph. Graph neural networks can handle this heterogeneous graph structure, using different aggregation functions or attention mechanisms to process the information interaction between different types of nodes, thereby capturing the complex relationship between slope and curve more precisely.

[0060] Step S203 involves inputting the driving scenario data and fused driving data into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments. (See details...) Figure 1 Step S103 in the embodiment will not be described again here.

[0061] Step S204: Based on the probability of occurrence and the risk level, generate a corresponding vehicle control strategy to control the vehicle. See details... Figure 1 Step S104 in the embodiment will not be described again here.

[0062] It is understood that the technical solution in this embodiment clearly defines road segment information of different dimensions as slope nodes and curve nodes. This application can directly model complex scenarios where slope and curve are coupled in complex road conditions such as mountain roads. By coupling the slope nodes and curve nodes, coupled node data containing information about their mutual influence is generated, effectively solving the risk of inappropriate control strategies caused by processing road segment information in a single dimension in the prior art, as well as the problem of not being able to balance power and comfort. On this basis, processing the coupled node data based on graph neural networks can efficiently capture the topological relationship and spatial association between slope and curve, deeply explore the inherent characteristics of the complex scenario, and thus accurately predict the probability and risk level of the corresponding driving scenario under the current road conditions. The technical solution in this embodiment not only improves the perception and understanding of complex road conditions, but also provides a reliable basis for generating more accurate and safer vehicle control strategies, improving the safety and driving experience of intelligent driving vehicles in complex road environments.

[0063] In some embodiments, this application provides a vehicle control method, such as... Figure 3 As shown, the method includes: Step S301: Collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data. See details... Figure 1 Step S101 in the embodiment will not be described again here.

[0064] Step S302 involves coupling road segment information from different dimensions in the fused driving data to obtain corresponding driving scenario data under the current road conditions, including: Step S3021: Couple the slope nodes and the corresponding curve nodes to obtain coupled node data. This includes: Step S30211: The slope information and slope change rate information of the slope node are transmitted to the adjacent curve node through the coupling model to obtain the coupled curve node.

[0065] In road geometry data, a slope node represents a data point or region that signifies the slope characteristics of a specific road segment. The slope information it contains refers to the vertical inclination of the road segment, such as an uphill or downhill gradient expressed as a percentage or angle; the slope change rate information describes how quickly the slope changes along the road direction, such as the trend of a gentle slope rapidly becoming a steep slope. This information is crucial for vehicle longitudinal dynamics and energy management. A coupling model is an algorithm or mathematical structure used to process and convey the correlation between different types of data. For example, this coupling model could be a rule-based expert system that judges the degree of influence of slope information on curves based on pre-defined logic; or it could be a lightweight neural network layer that learns from historical data to establish a non-linear mapping relationship between slope and curves. Adjacent curve nodes refer to curve data points or regions that are spatially continuous or closely connected to the current slope node in the road topology. By transmitting slope information and slope change rate information to these adjacent curve nodes, the curve nodes can consider the influence of the slope when processing their own curvature information. For example, the requirements for vehicle control on a curve after a sharp downhill slope will be much higher than those on an equivalent curve on a flat road.

[0066] In step S30212, the curvature information and curvature change rate information of the coupled curve node are transmitted to the adjacent slope node again through the coupling model to obtain the coupled node data.

[0067] In this context, coupled curve nodes refer to curve nodes that have received and integrated information from slope nodes. The curvature information they contain refers to the degree of road curvature, expressed as a radius of curvature or a curvature value; the rate of change of curvature information describes how quickly the curvature changes along the road direction, such as a rapid transition from a large-radius curve to a small-radius curve. This information is crucial for vehicle lateral stability, steering control, and driving comfort. The role of the coupling model here is to receive the curvature and rate of change of curvature information from the coupled curve nodes and pass them back to adjacent slope nodes. For example, this coupling model could use a weighted average algorithm to incorporate curve information into the slope node data with specific weights; alternatively, it could be a fuzzy logic-based inference system that corrects the slope node data based on curve characteristics. Through this back-transmission mechanism, adjacent slope nodes can receive and integrate the curvature and rate of change of curvature information from the curve nodes. This allows slope nodes to consider the influence of curves before and after them when evaluating their own slope characteristics, resulting in more comprehensive and accurate coupled node data. For example, on an uphill section of road before a sharp bend, the strategy for controlling vehicle power output and speed will need to take into account the impending lateral constraints.

[0068] Step S3022: Based on a graph neural network, the coupled node data is processed to obtain the probability of occurrence and risk level of the corresponding driving scenario under the current road conditions. (See details...) Figure 2 Step S2022 in the embodiment will not be described again here.

[0069] Step S303 involves inputting the driving scenario data and fused driving data into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments. (See details...) Figure 1 Step S103 in the embodiment will not be described again here.

[0070] Step S304: Based on the probability of occurrence and the risk level, generate a corresponding vehicle control strategy to control the vehicle. See details... Figure 1 Step S104 in the embodiment will not be described again here.

[0071] It is understood that the technical solution in this embodiment achieves bidirectional information transmission and deep interactive coupling between slope nodes and curve nodes. Specifically, the slope information and slope change rate information of the slope node are first transmitted to the adjacent curve node, enabling the curve node to fully consider the longitudinal dynamic impact of the slope when processing its own curvature information. For example, in a downhill curve, the vehicle's center of gravity shift and braking demand will change due to the slope. Subsequently, the curvature information and curvature change rate information of the coupled curve node that has integrated slope information are transmitted back to the adjacent slope node. This allows the slope node to incorporate the lateral constraint impact of the curve when evaluating its own characteristics. For example, in an uphill curve, the vehicle's power output must not only overcome slope resistance but also take into account lateral stability. The bidirectional, iterative coupling mechanism in this embodiment effectively avoids the problem of insufficient coupling caused by unidirectional information transmission, ensuring that the obtained coupled node data can comprehensively and accurately reflect the real characteristics and interactions of the combined slope and curve scenario. Therefore, subsequent driving scenario identification and risk prediction based on the data from this coupled node will be more accurate, thus providing a more reliable basis for the generation of vehicle control strategies and significantly improving the driving safety, power and comfort of vehicles in complex mountainous roads and other complex scenarios.

[0072] In some embodiments, this application further proposes a vehicle control method, such as Figure 4 As shown, the method includes: Step S401: Collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data. See details... Figure 1 Step S101 in the embodiment will not be described again here.

[0073] Step S402 involves coupling road segment information from different dimensions in the fused driving data to obtain corresponding driving scenario data under the current road conditions, including: Step S4021: Couple the slope nodes and corresponding curve nodes to obtain coupled node data. This includes: Step S40211 involves transmitting the slope information and slope change rate information of the slope node to the adjacent curve node through the coupling model, thus obtaining the coupled curve node. (See details...) Figure 3 Step S30211 in the embodiment will not be repeated here.

[0074] Step S40212 involves transmitting the curvature information and rate of change of curvature of the coupled curve nodes back to the adjacent slope nodes via the coupled model, thus obtaining the coupled node data. (See details...) Figure 3 Step S30211 in the embodiment will not be repeated here.

[0075] Step S4022: Preset risk driving scenarios include long slopes with continuous curves, long slopes with sharp curves, and gentle slopes with curves. Based on a graph neural network, process the coupled node data to obtain the probability of occurrence and risk level of the corresponding driving scenario under the current road conditions, specifically including: Step S40221: Based on the fully connected layer of the graph neural network, the coupled node data is processed to obtain the probability and corresponding risk level of the occurrence of long slope continuous curves, long slope sharp curves and gentle slope curves under the current road conditions.

[0076] The pre-defined risk driving scenarios, namely long, continuous curves, long, sharp curves, and gentle curves, are based on the complex situations arising from the coupling of slope and curves in actual mountainous roads. "Long, continuous curves" refers to multiple curves appearing consecutively on a long slope, posing a continuous challenge to the vehicle's power, braking, and steering control. "Long, sharp curves" refers to sharp curves with a small radius of curvature suddenly appearing at the end or middle of a long slope, requiring the vehicle to significantly decelerate and adjust its steering within a short period. "Gentle curves" refers to road sections with minimal slope changes but containing curves; although the risk level may be lower than the previous two, lateral stability still needs to be considered. By clearly defining these specific scenario types, this application can decompose complex road conditions into identifiable and quantifiable risk units, laying the foundation for subsequent accurate prediction and control.

[0077] Graph Neural Networks (GNNs) can effectively handle non-Euclidean structured data, such as the node and edge relationships in road networks. A fully connected layer is a basic layer in a neural network, where each neuron is connected to all neurons in the previous layer. In this application, the role of the fully connected layer is to further nonlinearly transform and map the high-dimensional features extracted by the graph neural network from the coupled node data. Specifically, it can transform the abstract representation of the coupling relationship between slope nodes and curve nodes learned by the graph neural network into feature vectors related to specific risky driving scenarios. For example, the fully connected layer can identify combination patterns indicating "long slopes" and "continuous curves," or specific interaction patterns of "long slopes" and "sharp curves," in the coupled node data. Furthermore, the fully connected layer can also serve as the output layer of the graph neural network, directly mapping the node embeddings or graph embeddings after multi-layer graph convolution and pooling operations to the final prediction result, thereby achieving effective parsing and classification of complex coupled data. "Obtained separately" means that for each preset risky driving scenario, the system independently outputs its probability of occurrence and corresponding risk level. The probability of occurrence refers to the likelihood of a specific risk scenario occurring on a future road segment under current road conditions, typically represented as a value between 0 and 1. The risk level is a quantitative assessment of the potential harm or control difficulty of the risk scenario if it occurs; it can be a discrete level (e.g., low, medium, high) or a continuous value. This separate output approach allows the system to perform fine-grained identification and assessment of different types of risk scenarios, rather than providing a general overall risk value. In implementation, this can be achieved through multiple independent output neurons or branches, each responsible for predicting the probability and level of a specific risk scenario; alternatively, it can be implemented through a multi-task learning model, where a shared feature extractor (e.g., a GNN layer) is followed by multiple predictions for different risk scenarios, each outputting the probability and risk level of its corresponding scenario.

[0078] Step S403 involves inputting the driving scenario data and fused driving data into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments. See details... Figure 1 Step S103 in the embodiment will not be described again here.

[0079] Step S404: Based on the probability of occurrence and the risk level, generate a corresponding vehicle control strategy to control the vehicle. (See details...) Figure 1 Step S104 in the embodiment will not be described again here.

[0080] It is understood that the technical solution in this embodiment solves the problem of inaccurate prediction caused by the lack of clear distinction between different types of preset risk driving scenarios in the prior art by defining specific preset risk driving scenario types, namely long slope continuous curves, long slope sharp curves, and gentle slope curves. Clear scenario definition allows the system to focus on actual high-risk situations, avoiding fuzzy processing of complex road conditions. Based on this, a fully connected layer of a graph neural network is used to process the coupled node data, obtaining the probability of occurrence and corresponding risk level of each preset risk driving scenario under the current road conditions, further improving the accuracy and specificity of the prediction. The fully connected layer can efficiently learn and identify the characteristic patterns of different risk scenarios from the coupled node data, and the resulting mechanisms ensure independent evaluation of each scenario, rather than general prediction. Detailed scenario classification and risk quantification enable subsequent vehicle control strategies to coordinate and control the power system, steering system, and braking system in a targeted manner. For example, when predicting long slope sharp curves, more aggressive deceleration and steering assistance can be applied in advance, effectively avoiding the risk of inappropriate control strategies and better balancing vehicle power, comfort, and safety. Compared to methods that only deal with a single road element or make general risk predictions, this application can more accurately identify and quantify the risks of complex composite scenarios formed by the coupling of slope and curves in mountainous roads, providing a more reliable basis for forward-looking planning and control of intelligent driving vehicles.

[0081] In some embodiments, this application further proposes inputting driving scenario data and fused driving data into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments, specifically including: Step a1: Input the probability and corresponding risk level of the driving scenario, the current vehicle driving section data, and the corresponding high-definition map data into the preset neural network model to predict the probability and corresponding risk level of long slope continuous curves, long slope sharp curves, and gentle slope curves in the future road section.

[0082] The probability of a driving scenario occurring and its corresponding risk level are obtained by coupling gradient and curve nodes on the current road segment and analyzing them using a graph neural network. This quantifies the likelihood and potential danger of specific risky driving scenarios (such as long, continuous curves, long, sharp curves, and gentle curves) occurring under current road conditions. It can be represented as a set of numerical values; for example, the probability value ranges from 0 to 1, and the risk level can be a discrete level (e.g., low, medium, high) or a continuous value (e.g., 1 to 10). This data provides important current state references and contextual information for predicting the risks of future road segments.

[0083] Current vehicle travel segment data refers to the real-time dynamic information of the vehicle on the current travel segment, including but not limited to the vehicle's precise position, speed, acceleration, heading angle, yaw rate, lane departure, etc. This data can be collected by the vehicle's own sensors (such as GPS, IMU, wheel speed sensors, cameras, radar, etc.) and obtained through data fusion processing. Its purpose is to provide a pre-defined neural network model with the vehicle's current motion state and the specific road environment it is in, so that the model can combine the vehicle's real-time dynamics to make more accurate predictions of future risks.

[0084] The corresponding high-definition map data refers to detailed map information covering the current driving segment and future planned driving segments, with an accuracy far exceeding that of ordinary navigation maps. High-definition map data can include road geometry (such as lane lines, curbs, curvature, slope, and heading), road topology, traffic signs, traffic lights, obstacle information, elevation information, etc. This data is typically pre-downloaded to the local vehicle (e.g., via a vehicle-to-cloud communication interface) and updated and loaded as needed during vehicle operation. It provides static environmental information of future road segments for the pre-set neural network model, serving as a necessary foundation for identifying potentially risky driving scenarios (such as long slopes and curves).

[0085] The preset neural network model is a trained machine learning model that learns the complex relationships between driving scenarios, vehicle status, and map information in historical data, thereby enabling it to predict the risks of future road segments.

[0086] The output of a pre-defined neural network model predicts the probability and corresponding risk levels of long, steep, sharp, and gentle curves on future road sections. This indicates the likelihood of specific types of risky driving scenarios (i.e., long, steep, sharp, and gentle curves) occurring on the road section the vehicle is about to travel on, along with their corresponding risk levels. These predictions are crucial for proactively adjusting vehicle control strategies, enabling the vehicle to anticipate and respond to potential hazards. For example, the predictions may include the probability distribution and risk levels of these risk scenarios over different time windows or spatial areas within the next 500 meters, 1 kilometer, or longer distances.

[0087] It is understandable that the technical solution in this embodiment effectively solves the problem of insufficient foresight caused by focusing only on the current road condition assessment. Specifically, by using the probability and risk level of the current driving scenario as input, contextual information based on the current state is provided for future predictions, ensuring the continuity and accuracy of the predictions. Simultaneously, by combining real-time dynamic data of the current vehicle's driving segment, the prediction results closely reflect the actual operating state of the vehicle, enhancing its adaptability to future risk scenarios. Furthermore, the introduction of high-definition map data provides the model with detailed geometric and topological information of future road segments, greatly improving the model's ability to identify and predict complex coupled risk scenarios (such as long slopes with continuous curves, long slopes with sharp curves, and gentle slopes with curves). This forward-looking risk prediction mechanism allows the vehicle control system to obtain risk information of future road segments in advance, thus having sufficient time for prediction and adjustment, such as early deceleration, adjustment of power output, or steering assist, effectively avoiding safety hazards caused by reaction delays and significantly improving the safety, comfort, and control precision of intelligent driving vehicles in complex road conditions. In some embodiments, this application further proposes a vehicle control method, such as... Figure 5 As shown, it includes: Step S501: Collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data. See details... Figure 1 Step S101 in the embodiment will not be described again here.

[0088] Step S502 involves coupling road segment information from different dimensions in the fused driving data to obtain the corresponding driving scenario data under the current road conditions. See details... Figure 1 Step S102 in the embodiment will not be described again here.

[0089] Step S503 involves inputting the driving scenario data and fused driving data into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments. See details... Figure 1 Step S103 in the embodiment will not be described again here.

[0090] Step S504: Based on the probability of occurrence and the risk level, generate a corresponding vehicle control strategy to control the vehicle, including: Step S5041: Based on the current driving mode and combined with the predicted probability and risk level of the occurrence of preset risk driving scenarios in future road sections, the vehicle's power system, steering system and braking system are coordinated and controlled.

[0091] The current driving mode refers to the vehicle's current operating state or the driver's preferred setting. This mode can be actively selected by the driver through the human-machine interface (HMI), such as "Sport Mode," "Comfort Mode," "Eco Mode," or "Standard Mode," to express their preference for vehicle performance, comfort, or fuel economy. Alternatively, the current driving mode can also be automatically identified and switched by the vehicle's intelligent system based on the driver's driving style (e.g., aggressive or smooth) or real-time road conditions.

[0092] Coordinated control of a vehicle's powertrain, steering, and braking systems refers to the unified coordination and integrated management of multiple key subsystems (including the powertrain, steering, and braking systems) during vehicle operation, rather than independent control. Specifically, a central vehicle control unit (VCU) or domain controller can receive input information from driving mode selection and risk prediction, generate coordinated control commands based on this information, and then distribute these commands to the electronic control units (ECUs) of various subsystems, such as the engine / motor ECU, electric power steering (EPS) ECU, and anti-lock braking system (ABS) / electronic stability control (ESC) ECU. Another implementation approach is to use a hierarchical control architecture, where higher-level controllers determine the overall behavioral objectives of the vehicle and decompose them into coordinated sub-commands for each subsystem, which are then executed by lower-level controllers.

[0093] It is understood that the technical solution in this embodiment, based on the prediction of future road risks, further incorporates consideration of the current driving mode and achieves coordinated control of the vehicle's powertrain, steering, and braking systems. This coordinated control mechanism effectively solves the problem of singular or uncoordinated control strategies in complex and risky driving scenarios. Specifically, by combining the current driving mode, the vehicle control strategy can dynamically adapt to the driver's personalized preferences and real-time risk predictions. For example, a more aggressive power response is allowed in Sport mode, while a greater emphasis is placed on stability in Comfort mode, thereby improving the personalized responsiveness of the control strategy. Simultaneously, coordinated control of the powertrain, steering, and braking systems ensures that in complex scenarios such as long slopes and continuous curves, the systems can cooperate rather than operate independently. For instance, when facing high-risk curves, the system can simultaneously optimize power output to maintain vehicle speed, adjust steering assist to stabilize cornering, and apply braking force in a timely manner to effectively decelerate, thus avoiding the imbalance that may result from adjustments to a single system and significantly improving the overall safety and driving comfort of the vehicle in complex road conditions. In some embodiments, this application further proposes controlling the vehicle's powertrain based on the current driving mode, according to the following expression: ,in, For the limited torque, For maximum torque, R represents the risk sensitivity coefficient and the risk level.

[0094] The expression for controlling the vehicle's powertrain defines the torque limiting strategy when considering risk level R. This expression dynamically adjusts the vehicle's power output based on the predicted risk level, limiting torque in high-risk scenarios to prevent over-acceleration or loss of control, thereby improving driving safety. In practice, the vehicle's power control unit (PCU) or engine management system (EMS) receives the calculated... Once the value is set, the actual output torque can be limited to that value by adjusting the engine's throttle opening, fuel injection quantity, or the motor's output current. In hybrid or pure electric vehicles, the maximum output torque of the motor can be limited through the coordinated work of the battery management system (BMS) and the motor controller (MCU).

[0095] Based on the current driving mode, the vehicle's steering system is controlled according to the following expression: ,in, To enhance the electric power steering force, The maximum electric power steering effort is represented by m, which is the power steering enhancement coefficient, and R is the risk level.

[0096] The expression controlling the vehicle's steering system describes the electric power steering (EPS) system's assist adjustment strategy when considering the risk level R. This expression dynamically enhances steering assist based on the predicted risk level, providing stronger steering assistance in high-risk scenarios to help the driver control the vehicle more stably and precisely, especially during curves or emergency obstacle avoidance. In practice, the EPS control unit calculates... The steering effort can be increased by adjusting the steering motor's output torque through changes in the current or voltage. In more advanced steer-by-wire systems, the steering actuator can be directly controlled by the electronic control unit to achieve precise adjustment of steering effort.

[0097] Based on the current driving mode, the vehicle's braking system is controlled according to the following expression: ,in, Braking pressure, Based on braking pressure, R represents the pressure adjustment factor and the risk level.

[0098] The expression for controlling the vehicle's braking system defines the braking pressure adjustment strategy when considering risk level R. This expression dynamically adjusts the braking pressure based on the predicted risk level, proactively or preemptively increasing braking pressure in high-risk scenarios to shorten braking distance, improve braking response speed, and thus enhance vehicle braking safety. In practice, the vehicle's braking control unit (such as the ABS / ESC module) receives the calculated... After the value is set, the hydraulic pressure in the brake line can be precisely adjusted by controlling actuators such as the brake hydraulic pump and solenoid valve, thereby applying the corresponding braking pressure. In electronic parking brake (EPB) or brake-by-wire (BBW) systems, the clamping force of the brake caliper can be directly controlled by electronic signals to adjust the braking pressure.

[0099] It is understood that the technical solution in this embodiment introduces specific mathematical expressions to achieve precise coordinated control of the vehicle's powertrain, steering, and braking systems, thereby solving the problem of imprecise parameter adjustment in coordinated control. Firstly, based on the current driving mode and risk level, the adaptability of the control strategy to the driving situation is ensured, avoiding rigid responses in a single mode. Specifically, the powertrain is controlled to automatically reduce power output in high-risk scenarios, preventing safety hazards caused by excessive acceleration. The steering system is controlled to improve steering stability in high-risk areas such as curves, avoiding understeer or oversteer. The braking system is controlled to achieve timely deceleration response, ensuring vehicle safety performance in emergency situations. Overall, through quantitative parameter adjustment, the coordinated control becomes more refined and adaptive, effectively improving the vehicle's operational safety and control efficiency in complex and risky scenarios. In some embodiments, this application proposes a vehicle control method, wherein the multi-source driving data includes at least one of the following: vehicle visual data, vehicle radar data, vehicle-to-cloud communication interface data, vehicle status data, driving behavior data, and high-definition map data; the current multi-source driving data is collected, including: Step b1: Download high-definition map data to the local vehicle in sequence according to the different road sections the vehicle is traveling on.

[0100] Multi-source driving data refers to a comprehensive data set acquired from multiple sensors and information sources inside and outside the vehicle, used to describe the vehicle's driving status, environmental information, and driving intentions. It provides comprehensive, redundant, and complementary information to improve the vehicle's perception and understanding of complex driving scenarios. In addition to the types listed in this application, multi-source driving data can also include millimeter-wave radar data, ultrasonic radar data, Global Positioning System (GPS) data, Inertial Measurement Unit (IMU) data, etc. Vehicle-to-cloud communication interface data refers to data generated by the vehicle interacting with a cloud server through its onboard communication module. This data can include real-time traffic information, weather information, map updates, remote diagnostic data, vehicle platooning information, etc. It can be implemented through cellular networks (such as 4G / 5G) or V2X (Vehicle-to-Everything) communication technologies. Vehicle status data refers to real-time data on the vehicle's own operating status, including but not limited to vehicle speed, acceleration, steering angle, braking pressure, engine speed, gear position, and throttle opening. This data is typically acquired through the vehicle's Controller Area Network (CAN bus) and reflects the vehicle's dynamic performance and the driver's control intentions. Driving behavior data refers to data on a driver's operating habits and style during driving, including acceleration and deceleration habits, steering habits, following distance, and lane change frequency. This data can be obtained by analyzing vehicle status data, visual data (such as driver facial recognition and eye tracking), or specialized sensors (such as steering wheel torque sensors), and used for personalized driving assistance or risk assessment.

[0101] Specifically, to address the challenges of large data volumes and high real-time requirements for high-definition maps, this embodiment proposes downloading high-definition map data to the local vehicle sequentially according to different road segments traveled by the vehicle. This aims to optimize the acquisition and management of high-definition map data. Due to the massive volume of high-definition map data, downloading the entire journey or a large area's map data at once would consume significant bandwidth and storage resources, potentially leading to download delays. By downloading high-definition map data to the local vehicle in segments, on demand, according to the vehicle's current and upcoming road segments, the demand for communication bandwidth and local storage can be significantly reduced, while ensuring that the vehicle always has the high-precision map information needed for current and future short-distance travel. This can be achieved through a preset geofence trigger mechanism, automatically triggering the download of map data for the next road segment when the vehicle enters a certain area or approaches the boundary of a road segment; or by predicting the required map data in advance and pre-downloading it based on the vehicle's navigation path planning.

[0102] It is understood that the technical solution in this embodiment effectively solves the problems of large data volume, high real-time requirements, and bandwidth storage pressure and latency caused by full download of high-definition map data by optimizing the acquisition method. It avoids the huge consumption of network bandwidth and local storage resources in the traditional full download mode, significantly reduces data transmission latency, and ensures that high-definition map data can be provided to vehicles in a timely and accurate manner. By downloading high-definition map data sequentially by road segment, not only is data acquisition efficiency improved, but more importantly, the quality and real-time performance of the fused driving data are ensured. This provides a reliable foundation for subsequent risk prediction and vehicle control strategy generation, avoiding inaccurate control strategies due to missing or delayed map data, and improving the safety, accuracy, and robustness of vehicle control. In some embodiments, this application further proposes downloading high-definition map data to the local vehicle sequentially according to different road segments traveled by the vehicle, including: Step c1: When the local high-definition map data for the current driving segment is insufficient, a task to download high-definition map data for the next driving segment is triggered. The high-definition map data is compressed using a preset algorithm to reduce the amount of data transmitted.

[0103] Specifically, when the local high-definition map data for the vehicle's current route is insufficient, the system intelligently determines when to initiate the download of high-definition map data for subsequent routes. For example, a safe distance threshold can be set; when the vehicle is only 5 kilometers away from the end of the currently downloaded high-definition map data, the system determines that the data is insufficient. Alternatively, the system can monitor the amount of locally stored high-definition map data; when it falls below a preset minimum available data size (e.g., 100MB), the system considers the data insufficient. Furthermore, the system can combine the vehicle's real-time speed to predict the travel time that the remaining map data can cover; when the predicted time is less than a certain threshold (e.g., 30 seconds), downloading is triggered.

[0104] Once the above conditions are met, the system will trigger the download task of high-definition map data for the next driving segment. This ensures that map information for subsequent segments can be obtained in a timely manner when data is insufficient. Specifically, the vehicle's local control unit or in-vehicle infotainment system can send a download request to the cloud server to obtain high-definition map data for the next predetermined segment in the vehicle's current driving direction. Alternatively, the system can pre-set a download queue, and when the triggering conditions are met, add the map data download task for the next segment to the queue and execute it immediately. This task triggering can be handled by a dedicated map data management module, which continuously monitors the status of the local map data.

[0105] Compression is achieved through preset algorithms to optimize the transmission efficiency of high-definition map data. Specifically, the preset algorithm can be a lossy compression algorithm designed specifically for the characteristics of map data. For example, it simplifies or removes non-critical information (such as background textures and low-precision features) from the map, retaining only the geometric and semantic information crucial for vehicle control. Alternatively, the preset algorithm can be a lossless compression algorithm, such as using Huffman coding, Run-Length Encoding (RLE), or Lempel-Ziv-Welch (LZW) algorithms to encode the map data, reducing data volume without losing any information. Sparsity processing based on feature points or critical paths can also be used, transmitting only critical road nodes and lane line information, while compressing or transmitting other auxiliary information on demand. Through these compression processes, the ultimate goal is to reduce the amount of data transmitted, thereby reducing the demand for vehicle communication bandwidth, ensuring timely acquisition of map data with limited network resources, shortening data download time, reducing communication costs, and reducing vehicle battery energy consumption.

[0106] It is understood that the technical solution in this embodiment, when the local high-definition map data for the vehicle's current driving segment is insufficient, can intelligently trigger the download task of high-definition map data for the next driving segment, thereby effectively avoiding the problem of affecting the real-time performance of vehicle control due to data loss or delay. Simultaneously, by compressing the high-definition map data using a preset algorithm, the amount of data transmission is significantly reduced, lowering the network bandwidth consumption and alleviating communication burden and latency risks. This ensures that the vehicle can always obtain timely, complete, and efficient high-definition map data support in complex driving scenarios, providing a reliable foundation for the fusion of multi-source driving data and vehicle control, thereby improving the safety, real-time performance, and economy of the intelligent driving system. In some embodiments, this application further proposes compressing high-definition map data using a preset algorithm, including: Step d1: Based on the frequency of occurrence of road segment information in different dimensions in the high-definition map data, the high-definition map data is decomposed into low-frequency high-definition map data and high-frequency high-definition map data.

[0107] Frequency of occurrence refers to the rate of change or recurrence density of specific road segment information (such as slope, curvature, road width, etc.) in high-definition map data. Low-frequency high-definition map data typically represents the overall trend, macroscopic geometric features, or slowly changing information of the road, such as large-scale slope changes or long straight road sections. High-frequency high-definition map data represents the details, local features, or rapidly changing information of the road, such as sharp bends, abrupt slope changes, and road surface texture. This decomposition aims to distinguish parts of the data with different importance and changing characteristics. In practice, frequency domain analysis methods such as Fourier transform or wavelet transform from signal processing can be used. High-definition map data (e.g., treating road geometry information, elevation information, etc. as one-dimensional or multi-dimensional signals) is converted to the frequency domain. By setting a frequency cutoff point, the portion below the cutoff frequency is classified as low-frequency data, and the portion above the cutoff frequency is classified as high-frequency data. Alternatively, it can be based on statistical methods, such as calculating the rate of change or standard deviation of road segment information (such as slope, curvature) within a certain sampling interval. The portions with smaller rates of change or lower standard deviations are identified as low-frequency data, while the portions with larger rates of change or higher standard deviations are identified as high-frequency data.

[0108] Step d2: Retain low-frequency high-definition map data and retain high-frequency high-definition map data according to a preset threshold.

[0109] Retaining low-frequency high-definition map data ensures the map's basic framework and macroscopic information are not lost, as this information is typically essential for vehicle navigation and basic route planning. Retaining high-frequency high-definition map data according to preset thresholds means that not all high-frequency details are fully preserved; instead, they are selected based on their importance or impact on vehicle control. The preset threshold can be a quantitative standard used to determine whether high-frequency information is important enough to be retained. For example, only curve information with curvature changes exceeding a certain threshold is retained, or only slope information with a gradient change rate exceeding a certain threshold is retained. Low-frequency data can be directly lossless encoded or stored with high precision. For high-frequency data, an importance scoring mechanism can be set, for example, scoring its impact on risk scenarios (such as sharp curves and steep slopes), retaining only high-frequency feature points or data segments with scores higher than the preset threshold. Another approach is to use lossy compression techniques for high-frequency data, but control the compression ratio and information loss through preset thresholds. For example, high-frequency data with small variations can be quantified or discarded, while critical high-frequency data with large variations can be retained with higher precision.

[0110] Step d3 involves reconstructing the retained high-frequency high-definition map data and low-frequency high-definition map data using an inverse transformation based on a preset algorithm, resulting in compressed high-definition map data.

[0111] Inverse transformation refers to the reverse operation corresponding to the decomposition process, used to recombine the selectively retained low-frequency and high-frequency data to form a structurally complete but smaller compressed dataset. The reconstruction process does not completely restore the original data, but rather restores its key information and structure to meet the needs of vehicle control and risk prediction. The resulting compressed data is significantly smaller in size than the original high-definition map data, but still effectively represents the key features of the road. If the decomposition used Fourier transform or wavelet transform, the reconstruction will use the corresponding inverse Fourier transform or inverse wavelet transform. By performing an inverse transform on the retained frequency domain components, the compressed map data can be reconstructed in the time or spatial domain. If the decomposition is based on statistical features or feature extraction, the reconstruction can be achieved through interpolation, spline fitting, or model-based reconstruction methods. For example, using retained key feature points and low-frequency trends, mathematical models or machine learning models can be used to reconstruct the road's geometry and elevation information.

[0112] Specifically, the technical solution in this embodiment addresses the challenge of balancing efficiency and integrity in high-definition map data compression by proposing a compression mechanism based on frequency decomposition and selective retention. By decomposing high-definition map data into low-frequency and high-frequency components according to the frequency of occurrence of road segment information in different dimensions, the compression algorithm can differentiate the processing of information with different variation characteristics in the data, avoiding the inefficiency or loss of key information that may result from traditional uniform compression. Retaining low-frequency high-definition map data ensures the integrity and stability of road macro-features and basic information, while selectively retaining high-frequency high-definition map data according to a preset threshold effectively filters redundant details while ensuring that crucial local features (such as sharp bends and steep slopes) essential for vehicle control and risk prediction are preserved. The compressed data obtained through inverse transformation reconstruction significantly reduces data transmission volume while maximizing the integrity of key information in the high-definition map data. This not only improves the download efficiency and timeliness of high-definition map data, especially when dynamic download of the next road segment map data is required during vehicle operation, but also provides a more accurate and reliable data foundation for subsequent risk scenario prediction, thereby enhancing the effectiveness and safety of vehicle control strategies.

[0113] In some embodiments, this application further proposes a specific implementation method for triggering the download task of high-definition map data for the next driving segment when the local high-definition map data for the current driving segment is insufficient, which includes: Step e1: When the local high-definition map data for the current driving segment is insufficient, determine the time point to trigger the download task based on the historical average delay time and redundancy time; based on the time point to trigger the download task, trigger the download task of high-definition map data for the next driving segment.

[0114] Insufficient local high-definition map data for the current driving segment means that the amount of locally stored high-definition map data for the current driving segment is insufficient to meet the vehicle's needs for continued safe and efficient driving. This can be determined from multiple dimensions. For example, the coverage area of ​​the currently downloaded high-definition map data may be less than a preset safe distance threshold, or the data accuracy and completeness may not meet the requirements of the current or upcoming driving task. For instance, this can be determined by real-time monitoring of the distance between the vehicle's current location and the boundary of the downloaded high-definition map data. When this distance is less than a preset safe distance (e.g., the vehicle's current location is 5 kilometers from the end of the downloaded map data), the data is considered insufficient. Alternatively, it can be determined by assessing the match between the current driving task's (such as navigation, route planning, risk prediction, etc.) demand for high-definition map data and the amount of locally available data. If the current driving task requires detailed map information for the next 10 kilometers, but the local data only covers the next 3 kilometers, it is considered insufficient.

[0115] Historical average latency refers to the average time interval from the triggering of a high-definition map data download task to the complete download and availability of the required data. It is influenced by various factors, including network bandwidth, server response speed, data packet size, network congestion, and network signal strength at the vehicle's location. For example, the system can continuously record the start and end times of each high-definition map data download task, calculate the time difference, and then statistically average these time differences to obtain the historical average latency. Furthermore, historical average latency can be statistically calculated and stored separately for different network environments (such as 4G, 5G, Wi-Fi), geographical regions, or time periods to select more accurate reference values ​​in different scenarios.

[0116] Redundancy time refers to an extra buffer period reserved when determining the download task time to cope with unpredictable factors such as network fluctuations, temporary server failures, and data transmission interruptions. This period ensures that even under the most unfavorable circumstances, high-definition map data can be downloaded in time before vehicles actually need it, thereby improving the robustness and reliability of the system. For example, redundancy time can be a fixed preset value, such as 5 seconds, 10 seconds, or longer, set according to the system's requirements for data timeliness and its tolerance for risk. Alternatively, redundancy time can also be dynamically adjusted, for example, by dynamically calculating a suitable redundancy time based on real-time information such as current network conditions (e.g., signal strength, packet loss rate), weather conditions, and traffic flow through algorithms.

[0117] Determining the optimal time to trigger the download task involves calculating an optimal moment based on factors such as the vehicle's current location, speed, future road segment length, historical average latency, and redundancy time. This ensures that the data is downloaded promptly and completely before the vehicle reaches the next road segment. For example, the calculation can be done using the following formula: Trigger Time = Current Time + (Estimated Time to Reach Next Road Segment - Historical Average Latency - Redundancy Time). The estimated time to reach the next road segment can be estimated using the current vehicle speed and the road segment distance. Alternatively, a predictive model-based approach can be used, employing machine learning algorithms to comprehensively consider various dynamic factors such as vehicle driving patterns, network conditions, and historical download success rates to predict an optimal download trigger time.

[0118] Triggering the download of high-definition map data for the next road segment refers to the system sending a request to the cloud server at a predetermined time to begin downloading high-definition map data for the next road segment the vehicle will be traveling on. This typically involves steps such as establishing a data transmission protocol, receiving, verifying, and storing data packets. For example, when a preset trigger time is reached, the vehicle communication module sends a download request to the cloud map service through the vehicle-to-cloud communication interface and begins receiving and processing the transmitted high-definition map data. Alternatively, the triggering mechanism can be closely integrated with the vehicle's navigation system or route planning module; when the navigation system determines that the vehicle is about to enter a new road segment and the download conditions are met, it automatically initiates the download of high-definition map data for that segment.

[0119] It is understood that the technical solution in this embodiment optimizes the triggering timing of the high-definition map data download task, solves the problem of potential resource waste or delays during the download process, and ensures that the data is available in a timely manner to support subsequent vehicle control. When the local high-definition map data for the vehicle's current driving segment is insufficient, the system can clearly determine and trigger the download task, avoiding unnecessary network resource consumption. By introducing historical average latency and redundancy time to determine the timing of triggering the download task, the system can use historical data to predict potential download delays and reserve buffer time, thereby effectively reducing the risk of download delays caused by poor network conditions or large data volumes, ensuring that high-definition map data is ready before the vehicle actually needs it. Executing the download operation based on the calculated optimal timing ensures the high efficiency and reliability of the download process, providing a solid data foundation for the accuracy of subsequent multi-source driving data fusion and the execution efficiency of vehicle control strategies. This not only improves the overall system's responsiveness and safety, but also avoids potential driving safety hazards caused by missing or delayed map data. It enables vehicles to more accurately predict risks and generate control strategies in complex road conditions, such as long slopes with continuous curves, long slopes with sharp curves, and gentle slopes with curves, thus significantly improving the decision-making quality and driving safety of intelligent driving vehicles.

[0120] In some embodiments, this application further proposes that after triggering the download task of high-definition map data for the next travel segment, the method also includes: Step f1 discards the high-resolution map data of the next road segment that the vehicle has already traveled.

[0121] Specifically, high-resolution map data for the next route the vehicle has already traversed is discarded. This proactively removes unnecessary map data, effectively freeing up local storage resources and preventing unnecessary storage space occupation. For example, the system can be configured to automatically identify and delete the high-resolution map data corresponding to a preset geofence when the vehicle's actual driving trajectory exceeds that geofence. Alternatively, the system can periodically clean up map data that has not been accessed for a long time or has expired, based on the timestamps or usage frequency of downloaded map data. Furthermore, when the vehicle's local storage space usage reaches a preset threshold, the system can also activate a cleanup mechanism, prioritizing the discarding of high-resolution map data for already traversed routes to ensure the availability of critical data storage.

[0122] Step f2: If there is a conflict between data transmission through the vehicle-to-cloud communication interface and download of high-definition map data, the data transmitted through the vehicle-to-cloud communication interface shall be transmitted first.

[0123] Specifically, if data transmission via the vehicle-to-cloud communication interface conflicts with the download of high-definition map data, the data transmitted via the vehicle-to-cloud communication interface will be prioritized. This establishes a data transmission priority mechanism to ensure smooth operation of critical real-time communications. For example, the vehicle's communication management module can employ a Quality of Service (QoS) strategy, prioritizing real-time control commands, emergency alarm information, or dynamic traffic data transmitted via the vehicle-to-cloud communication interface, while prioritizing the download of high-definition map data. When communication bandwidth resources are limited or congestion occurs, the system will prioritize bandwidth allocation for high-priority data transmission and may even temporarily suspend or limit the download speed of high-definition map data to ensure that the real-time data required for vehicle control strategies can be transmitted in a timely and reliable manner.

[0124] It is understood that the technical solution in this embodiment solves the problems of wasted vehicle storage resources and communication conflicts. By promptly discarding high-definition map data of already traversed road sections, not only can local storage space be effectively freed up, avoiding system efficiency reduction due to data redundancy, but the effective utilization of storage resources is also ensured, providing sufficient space for subsequent downloads of new map data or system operation. Simultaneously, by prioritizing the transmission of data from the vehicle-to-cloud communication interface, this application ensures that the vehicle can promptly receive and respond to critical instructions and real-time information from the cloud in complex driving scenarios, avoiding control delays or interruptions caused by map download tasks occupying communication bandwidth, thus improving the timeliness and reliability of vehicle control strategies. For intelligent driving systems that rely on multi-source driving data fusion (such as vehicle visual data, radar data, vehicle-to-cloud communication interface data, etc.) for risk prediction and control strategy generation, this ensures the system's continuous and stable operation in dynamically changing environments, thereby improving overall driving safety and user experience.

[0125] In one example, a more specific case will be used to illustrate the above technical solution in greater detail: Suppose an autonomous vehicle is driving on a mountain road with complex gradient changes and continuous curves, such as a long downhill followed by a sharp turn, or a series of curves on an uphill section. Existing technologies, when handling such scenarios, often focus only on a single road element, such as using only gradient for energy recovery or speed control, while neglecting the impact of curves on vehicle dynamics. This can lead to inappropriate control strategies and safety hazards. Furthermore, in uphill curves where both power and comfort need to be balanced, a single gradient planning approach is insufficient to achieve optimal control. This method addresses these issues through the following steps: First, the vehicle continuously collects multi-source driving data, including vehicle vision data, vehicle radar data, vehicle-to-cloud communication interface data, vehicle status data, driving behavior data, and high-precision map data. Before data fusion, the system downloads high-precision map data sequentially according to the different road segments the vehicle is traveling on. For example, if the local high-precision map data storage for the current road segment is insufficient, the system determines and triggers the download task for the high-precision map data of the next road segment based on historical average latency and redundancy time. To reduce data transmission volume, the high-precision map data is compressed using a preset algorithm, decomposing the frequency of road segment information in different dimensions of the map data into low-frequency and high-frequency high-precision map data. Low-frequency data is retained, and high-frequency data is retained according to a preset threshold. Finally, the retained high-frequency and low-frequency data are reconstructed through the inverse transformation of a preset algorithm to obtain compressed data. After the download task is triggered, the high-precision map data for the road segments already traveled by the vehicle is discarded. If there is a conflict between data transmission through the vehicle-to-cloud communication interface and the download of high-precision map data, the system will prioritize transmitting data from the vehicle-to-cloud communication interface. All collected data is then fused to obtain fused driving data.

[0126] Next, the system couples road segment information from different dimensions within the fused driving data. For slope nodes and curve nodes common in mountainous roads, the system performs deep coupling. First, the slope information and slope change rate information of the slope node are passed to the adjacent curve node through the coupling model, resulting in coupled curve nodes. Subsequently, the curvature information and curvature change rate information of the coupled curve nodes are again passed to the adjacent slope nodes through the coupling model, thus obtaining coupled node data. This bidirectional transmission mechanism ensures a thorough modeling of the mutual influence between slope and curve, rather than a simple superposition.

[0127] Then, the obtained driving scenario data (i.e., coupled node data) and fused driving data are input into a pre-defined neural network model to predict the probability and risk level of pre-defined risk driving scenarios in future road segments. Pre-defined risk driving scenarios include long, continuous curves, long, sharp curves, and gentle curves. Specifically, based on the fully connected layer of a graph neural network, the coupled node data is processed to obtain the probability and corresponding risk level of long, continuous curves, long, sharp curves, and gentle curves under the current road conditions. Subsequently, the probability and corresponding risk level of these driving scenarios, the current vehicle driving segment data, and the corresponding high-precision map data are input into the pre-defined neural network model to predict the probability and corresponding risk level of these risk scenarios in future road segments. For example, the system predicts that there is a high-risk long, sharp curve 500 meters ahead.

[0128] Finally, based on the predicted probability of occurrence and risk level, the system generates a corresponding vehicle control strategy to control the vehicle. This strategy, combined with the current driving mode (e.g., Eco, Comfort, or Sport), coordinates the control of the vehicle's powertrain, steering, and braking systems. For example, when the system predicts a high-risk long incline and sharp curve ahead, it will control the vehicle's powertrain to limit torque output based on the current driving mode.

[0129] Through the aforementioned coordinated control, this method effectively avoids the risk of inappropriate control strategies caused by processing only a single road element in existing technologies. For example, in scenarios with sharp curves at the end of long downhill slopes, this method, by coupling slope and curve information and predicting the high-risk scenario of "long slope and sharp curves," can preemptively limit the torque of the power system, apply additional braking pressure to the braking system, and enhance steering assist. This allows the necessary deceleration and attitude adjustment to be completed before the vehicle enters the curve, significantly improving driving safety.

[0130] The technical solution in this application goes beyond simply recovering energy based on slope; it comprehensively considers the coupled effects of slope and curves, achieving more accurate and forward-looking risk prediction and control. In uphill curves where both power and comfort are required, the coordinated control of the powertrain and steering systems ensures sufficient power for climbing while providing a smoother and more comfortable steering experience, solving the problem of existing technologies struggling to balance power and comfort. This holistic solution, involving multi-dimensional information coupling, scenario-adaptive prediction, and coordinated control, ensures safe, efficient, and comfortable driving in complex road conditions.

[0131] This embodiment also provides a vehicle control device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0132] This embodiment provides a vehicle control device, such as... Figure 6 As shown, it includes: The fusion module 601 is used to collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data.

[0133] The coupling module 602 is used to couple road segment information of different dimensions in the fused driving data to obtain the corresponding driving scenario data under the current road conditions.

[0134] The prediction module 603 is used to input driving scenario data and fused driving data into a preset neural network model to predict the probability and risk level of the occurrence of preset risk driving scenarios in future road segments.

[0135] The control module 604 is used to generate corresponding vehicle control strategies to control the vehicle based on the probability of occurrence and the risk level.

[0136] The vehicle control device provided in this application can execute the vehicle control method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0137] Figure 7 This is a structural schematic diagram of a vehicle provided in an embodiment of this application.

[0138] The following is a detailed reference. Figure 7 The diagram illustrates a structural schematic suitable for implementing a vehicle according to embodiments of this application. The vehicle may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for vehicle operation. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0139] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows the vehicle to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 Vehicles with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0140] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 709, or installed from memory 708, or installed from ROM 702. When the computer program is executed by processor 701, it performs the functions defined in the vehicle control method of embodiments of this application.

[0141] Figure 7 The vehicle shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0142] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the vehicle control method shown in the above embodiments is implemented.

[0143] A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0144] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A vehicle control method, characterized in that, The method includes: Collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data; The road segment information of different dimensions in the fused driving data is coupled and processed to obtain the corresponding driving scenario data under the current road conditions; The driving scenario data and the fused driving data are input into a preset neural network model to predict the probability and risk level of the occurrence of preset risk driving scenarios in future road sections; Based on the occurrence probability and the risk level, a corresponding vehicle control strategy is generated to control the vehicle.

2. The method according to claim 1, characterized in that, The different dimensions of road segment information include slope nodes and curve nodes. The coupling processing of road segment information from different dimensions in the fused driving data to obtain corresponding driving scenario data under the current road conditions includes: The slope node and the corresponding curve node are coupled to obtain coupled node data; Based on a graph neural network, the data of the coupled nodes is processed to obtain the probability of occurrence and risk level of the corresponding driving scenario under the current road conditions.

3. The method according to claim 2, characterized in that, The coupling of the slope node and the corresponding curve node to obtain coupled node data includes: The slope information and slope change rate information of the slope node are transmitted to the adjacent curve node through the coupling model to obtain the coupled curve node; The curvature information and curvature change rate information of the coupled curve node are then transmitted to the adjacent slope node through the coupling model to obtain the coupled node data.

4. The method according to claim 3, characterized in that, The preset risk driving scenarios include long, continuous curves, long, sharp curves, and gentle curves. The graph neural network-based processing of the coupled node data yields the probability of occurrence and risk level of the corresponding driving scenario under the current road conditions, including: Based on the fully connected layer of the graph neural network, the coupled node data is processed to obtain the probability and corresponding risk level of the occurrence of the long slope continuous curve, the long slope sharp curve, and the gentle slope curve under the current road conditions.

5. The method according to claim 4, characterized in that, The step of inputting the driving scenario data and the fused driving data into a preset neural network model to predict the probability and risk level of preset risk driving scenarios in future road segments includes: The probability and corresponding risk level of the driving scenario, the current vehicle driving section data, and the corresponding high-definition map data are input into a preset neural network model to predict the probability and corresponding risk level of the occurrence of the long slope continuous curves, the long slope sharp curves, and the gentle slope curves in future road sections.

6. The method according to claim 4, characterized in that, The step of generating a corresponding vehicle control strategy based on the occurrence probability and the risk level to control the vehicle includes: Based on the current driving mode and combined with the predicted probability and risk level of the occurrence of preset risk driving scenarios in future road sections, the vehicle's power system, steering system and braking system are coordinated and controlled.

7. The method according to claim 6, characterized in that, The method of coordinating the control of the vehicle's powertrain, steering system, and braking system based on the current driving mode and the predicted probability and risk level of pre-set risky driving scenarios in future road sections includes: Based on the current driving mode, the vehicle's powertrain is controlled according to the following expression: ,in, For the limited torque, For maximum torque, R is the risk sensitivity coefficient, and R is the risk level. Based on the current driving mode, the vehicle's steering system is controlled according to the following expression: ,in, To enhance the electric power steering force, The maximum electric power steering effort is represented by m, the power steering enhancement coefficient is represented by R, and the risk level is represented by R. Based on the current driving mode, the vehicle's braking system is controlled according to the following expression: ,in, Braking pressure, Based on braking pressure, R represents the pressure adjustment factor and the risk level.

8. The method according to claim 1, characterized in that, The multi-source driving data includes at least one of the following: vehicle vision data, vehicle radar data, vehicle cloud communication interface data, vehicle status data, driving behavior data, and high-definition map data. The collection of current multi-source driving data includes: The high-definition map data is downloaded to the local vehicle in sequence according to the different road sections the vehicle is traveling on.

9. The method according to claim 8, characterized in that, The process of sequentially downloading the high-definition map data to the local vehicle according to different road segments traveled by the vehicle includes: When the local high-definition map data for the current route of the vehicle is insufficient, a download task for the high-definition map data for the next route is triggered. The high-definition map data is compressed using a preset algorithm to reduce the amount of data transmitted.

10. The method according to claim 9, characterized in that, The high-definition map data is compressed using a preset algorithm, including: Based on the frequency of occurrence of road segment information of different dimensions in the high-definition map data, the high-definition map data is decomposed into low-frequency high-definition map data and high-frequency high-definition map data. The low-frequency high-definition map data is retained, and the high-frequency high-definition map data is retained according to a preset threshold. The retained high-frequency high-definition map data and the low-frequency high-definition map data are reconstructed according to the inverse transformation of the preset algorithm to obtain compressed data of the high-definition map data.

11. The method according to claim 9, characterized in that, When the local high-definition map data for the current driving segment is insufficient, triggering the download task of the high-definition map data for the next driving segment includes: When the local high-definition map data for the current route of the vehicle is insufficient, the time point for triggering the download task is determined based on the historical average latency and redundancy time. Based on the time point at which the download task is triggered, the download task of the high-definition map data for the next driving segment is triggered.

12. The method according to claim 9, characterized in that, After triggering the download task of the high-definition map data for the next driving segment, the method further includes: The high-definition map data of the next driving segment that the vehicle has already passed is discarded; If the data transmission via the vehicle-to-cloud communication interface conflicts with the download of the high-definition map data, the data transmitted via the vehicle-to-cloud communication interface shall be transmitted first.

13. A vehicle control device, characterized in that, The device includes: The fusion module is used to collect current multi-source driving data and fuse the multi-source driving data to obtain fused driving data; The coupling module is used to couple road segment information of different dimensions in the fused driving data to obtain the corresponding driving scenario data under the current road conditions. The prediction module is used to input the driving scenario data and the fused driving data into a preset neural network model to predict the probability and risk level of the occurrence of preset risk driving scenarios in future road segments; The control module is used to generate a corresponding vehicle control strategy to control the vehicle based on the occurrence probability and the risk level.

14. A vehicle, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the vehicle control method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the vehicle control method according to any one of claims 1 to 12.

Images