A multi-dimensional fine-grained intelligent driving system evaluation method

Abstract

The present invention belongs to the field of data analysis and processing, and discloses a multi-dimensional fine-grained intelligent driving system evaluation method; the method comprises: collecting road posture, environmental data and vehicle status parameters during the intelligent driving process, dynamically selecting three driving modes of high-speed defense, baseline cruise and environmental penetration speed based on a cognitive computing operation paradigm tree, and using a gradual state machine to smoothly switch the driving modes; switching the driving modes in a variety of test scenarios, and using a multi-layer perception evaluation system to evaluate each selected driving mode to obtain a score and a comprehensive score for the safety, efficiency and comfort of the intelligent driving system; wherein the multi-layer perception evaluation system comprises a three-dimensional association network, which forms a multi-level three-dimensional representation by granularly refining multiple dimensions, and forms a complete three-dimensional association network by node simulation and directed edge connection, thereby realizing a multi-dimensional fine-grained evaluation of the intelligent driving system.

Description

Technical Field

[0001] The present invention relates to the field of data analysis and processing, and more specifically, to a multi-dimensional fine-grained intelligent driving system evaluation method. Background Art

[0002] When evaluating intelligent driving systems, it is generally necessary to evaluate them from the perspectives of safety (whether the system can avoid accidents and respond to emergencies or adverse situations, such as sudden rain or snow, driving at night, or pedestrians suddenly crossing the road), efficiency (the system's performance in terms of time, energy consumption, path optimization, etc.), and comfort (user experience, including acceleration and deceleration smoothness, lane changing strategies, etc.). However, there are often trade-offs or conflicts between different evaluation dimensions.

[0003] These conflicts make it challenging to scientifically balance and comprehensively evaluate intelligent driving when evaluating system design. Furthermore, safety, comfort, and efficiency are relatively general and broad concepts in the evaluation process, making it a technical challenge to break down these evaluation dimensions into quantifiable metrics. For example, the performance of a perception module can be broken down into metrics such as object recognition rate, false detection rate, and missed detection rate, but these metrics still need to be defined and measured.

[0004] Therefore, design and innovation are needed to meet actual needs when conducting multi-dimensional and fine-grained evaluation of intelligent driving systems. Summary of the Invention

[0005] To overcome the above-mentioned shortcomings of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solutions: a multi-dimensional, fine-grained intelligent driving system evaluation method, comprising: collecting road posture, environmental data, and vehicle state parameters during the intelligent driving process; dynamically selecting three driving modes (high-speed defense, baseline cruise, and environmental penetration speed) based on a cognitive computing-based operational paradigm tree; and using a gradual state machine to smoothly switch between driving modes;

[0006] A pre-set multi-layer perception evaluation system is used to switch driving modes in various test scenarios. Based on each selected driving mode, the system is used to evaluate the safety, efficiency, and comfort scores of the intelligent driving system, as well as a comprehensive score, and to provide visual interaction.

[0007] The multi-layer perception evaluation system includes a stereo association network, which forms a multi-level stereo representation by refining the granularity of multiple dimensions and forms a complete stereo association network through node simulation and directed edge connection.

[0008] For any node in the three-dimensional correlation network, the influence evolution propagation mechanism is used to update it. The influence evolution propagation mechanism is designed based on the graph attention network, and the saturation effect of energy propagation, the gating mechanism, the score difference drive of diffusion propagation and the damping effect of energy propagation are introduced to realize the real diffusion of the influence interaction between evolution dimensions.

[0009] Preferably, the three driving modes of high-speed defense, baseline cruise and environmental penetration speed include:

[0010] In each driving mode, a weight value is used to quantify the priority of each key dimension, and the primary dimension has the largest weight value. Among them, the key dimensions are safety, efficiency and comfort, and the primary dimension is the most prioritized key dimension in each driving mode;

[0011] Hard constraints are set for key indicators for each driving mode to meet minimum performance requirements in all situations. For the high-speed defense mode, the safety constraints are set to emergency braking response time less than A1, target detection recall rate greater than B1, and braking distance less than C1; the efficiency constraint is set to path length optimization rate greater than D1, and the comfort constraint is set to acceleration change rate less than E1.

[0012] For the baseline cruise mode, the safety constraints are set as emergency braking response time less than A2, target detection recall rate greater than B2, and braking distance less than C2; the efficiency constraint is set as path length optimization rate greater than D2; and the comfort constraint is set as acceleration change rate less than E2.

[0013] For the environmental penetration speed mode, the safety constraints are set as emergency braking response time less than A3, target detection recall rate greater than B3, and braking distance less than C3; the efficiency constraint is set as path length optimization rate less than D3; and the comfort constraint is set as acceleration change rate less than E3.

[0014] Among them, A1, A2, A3, B1, B2, B3, C1, C2, C3, D1, D2, D3, E1, E2 and E3 are pre-set values, and the values meet the requirements of A1<A2<A3; B1>B2>B3; C1>C2>C3; D1>D2>D3; E1>E2>E3.

[0015] Preferably, the method of collecting road posture, environmental data, and vehicle status parameters during intelligent driving and dynamically selecting the three driving modes of high-speed defense, baseline cruise, and environmental penetration speed based on a cognitive computing operation paradigm tree includes:

[0016] Road posture data includes slope, curvature and road friction coefficient; environmental data includes weather data, visibility and traffic density; vehicle state parameters include speed and acceleration;

[0017] The cognitive computing operation paradigm tree is composed of a root node and intermediate nodes, and the input is road posture, environmental data and vehicle state parameters. The root node represents the driving mode selection, and the intermediate nodes are used for subtask decomposition.

[0018] At each intermediate node, a cognitive computing model is used to reason about the input data and output the priority and confidence of the subtask;

[0019] The driving modes include high-speed defense, baseline cruise, and environmental penetration speed. The high-speed defense mode prioritizes safety and is triggered by the following conditions: slope > q1, curvature > s1, and road friction coefficient < d1 in the road posture; weather data in the environmental data is severe weather, visibility < n1, and traffic density > g1; and vehicle speed > c1 and acceleration change rate > w1 in the vehicle status.

[0020] The baseline cruise mode prioritizes efficiency, and the triggering conditions are: slope ≤ q1, curvature ≤ s1, and road friction coefficient ≥ d1 in the road posture; weather data in the environmental data is moderate weather, visibility ≥ n1, and traffic density ∈ (g2, g1); vehicle speed ∈ (c2, c1), and acceleration change rate ≤ w1 in the vehicle state;

[0021] The environmental penetration speed mode prioritizes comfort, and the triggering conditions are: slope ≤ q2, curvature > s2 in road posture; visibility < n2, traffic density > g3 in environmental data; vehicle speed < c2, acceleration change rate ≤ w2 in vehicle status;

[0022] Wherein, q1>q2, s1<s2; d1; n1>n2; g1>g2>g3; c1>c2; w1>w2, and q1, q2, s1, s2, d1, n1, n2, g1, g2, g3, c1, c2, w1 and w2 are all preset values;

[0023] The vehicle's road posture, environmental data, and vehicle status parameters are collected in real time at a fixed frequency hb, and input into the cognitive computing operation paradigm tree to output the selected driving mode. If the currently output driving mode is the same as the previous output, there is no need to switch the driving mode. If the currently output driving mode is different from the previous output, a gradual state machine is used to smoothly switch the driving mode.

[0024] Preferably, the method for smoothly switching the driving mode using a gradual state machine includes:

[0025] Smooth switching includes switching the weight of driving modes and switching the control parameters of the vehicle;

[0026] For driving mode weight switching, the weights of safety, efficiency, and comfort in each driving mode are preset, and a switching time window is added. Within the window, each weight is gradually adjusted using linear interpolation until the preset weight value in the driving mode is reached;

[0027] For the vehicle control parameter switching, an intermediate state is introduced in the middle of the switching. The intermediate state is the intermediate value of the control parameters in the two modes. The control parameters include vehicle speed, acceleration and following distance. The RNN algorithm is used to evaluate the scene jump. The lengthening multiple of the switching time window is set according to the scene jump. The adjusted switching time window is obtained and used as the final switching time window and applied to the control parameter switching process.

[0028] Preferably, the design method of the multi-layer perception evaluation system includes:

[0029] Step L1: Construct a three-dimensional association network;

[0030] Taking safety, efficiency, and comfort as key dimensions, pre-set sub-dimensions under the key dimensions. The sub-dimensions are decomposed into sub-sub-dimensions. The key dimensions, sub-dimensions, and sub-sub-dimensions are distributed on the same level. This continues until all levels are filled. Each level has dimensions. All dimensions are simulated into nodes. Any two nodes are connected by directed edges, and attribute values are attached to each directed edge, forming a three-dimensional network structure.

[0031] Among them, directed edges represent the mutual influence relationship between dimensions, from the influencer to the influenced, and the attribute value is used to represent the intensity of the influence, including positive and negative influences;

[0032] Step L2: The dimensions of the last level in the stereo association network define quantifiable indicators, and these indicators are used as inputs of the stereo association network;

[0033] Initialize the attribute value of each directed edge and the energy value of each node, and use the influence evolution propagation mechanism to update the attribute value of the directed edge in the three-dimensional association network, thereby evolving the energy value of the node;

[0034] Obtain the energy value of the node corresponding to the key dimension in the three-dimensional association network as the final output value.

[0035] Preferably, the method for initializing the attribute value of each directed edge and the energy value of each node includes:

[0036] Pre-train machine learning models in different modes, and use historical data to input the model in the corresponding mode to obtain the attribute value of each directed edge;

[0037] The expert scoring method is used to obtain the initial score of each node and perform normalization.

[0038] Preferably, the method of using the influence evolution propagation mechanism to update the attribute values of directed edges in the three-dimensional association network, thereby evolving the energy values of the nodes, includes:

[0039] The indicators of the last level in the stereo association network are collected as input, the evaluation period is preset, and the indicators are discretized into time steps for dynamic iteration;

[0040] At the current time step, the update direction is passed from the lowest level to the highest level in sequence. For any two nodes connected by a directed edge, the influence evolution propagation mechanism is used to update until the entire three-dimensional association network is updated. The iteration is repeated at the next time step until the preset evaluation cycle time is exhausted. The iteration stops and the energy value of each node and the attribute value of the directed edge are updated.

[0041] Then the energy values of the nodes corresponding to safety, efficiency and comfort on the first level are output as the scores of the key dimensions;

[0042] Construct a comprehensive scoring function to evaluate the overall system performance. This involves normalizing the energy values of the nodes corresponding to safety, efficiency, and comfort, and performing a weighted summation based on the corresponding weights under the selected driving mode to obtain a comprehensive score.

[0043] The scores of key dimensions and the comprehensive scores are output as the evaluation scores of the intelligent driving system.

[0044] Preferably, the influence evolution propagation mechanism is designed based on a graph attention network, including:

[0045] For any two nodes in the three-dimensional association network, at each time step, the graph attention network is used to calculate the attention weights of the influencer and the influenced node according to the direction of the directed edge, which are used as the new attribute value of the directed edge. The energy propagation and saturation effect are added. The new energy value received by the influenced node is the weighted sum of the attention weights of all predecessor nodes, the initial attribute value of the directed edge, and the energy value.

[0046] In the process of influence transfer, a gating mechanism is introduced to control the fusion ratio of the new energy value and the current energy value. The specific method is as follows: the update gate is the Sigmoid function value calculated by the weighted sum of the current energy value and the new energy value; the candidate state is the hyperbolic tangent function value calculated by the weighted sum of the current energy value and the new energy value. The node energy value is updated using the weighted sum of the current energy value and the candidate state, and the weight is determined by the update gate;

[0047] Based on the node energy value update, a score difference drive for diffusion propagation is introduced. This involves obtaining the energy value difference between the influencer and the affected node, taking the weighted sum of the initial energy values of all predecessor nodes corresponding to the affected node and the energy value difference, then multiplying the result by the diffusion coefficient to obtain the diffusion term. The updated energy value is then added to the diffusion term to obtain the corrected energy value of the node.

[0048] On the basis of the correction of the node energy value, the damping effect of energy propagation is introduced, including the introduction of the damping coefficient. The absolute value is obtained by subtracting the damping coefficient from 1. For the current time step, the final state of the energy value at the node in the next time step is the product of the corrected energy value and the absolute value.

[0049] Preferably, the saturation effect of energy propagation includes calculating the Sigmoid function value of the energy value of the node corresponding to the influencer and mapping it to the interval (0, 1).

[0050] Preferably, the visual interaction method includes:

[0051] Use graph visualization tools to draw a three-dimensional association network structure. The nodes in each layer are the same size. Starting from the first layer, the node shapes gradually become smaller. The thickness of the directed edges is drawn according to the size of the attribute value. Directed edges are drawn using different colors, and positive and negative influences are set based on the color.

[0052] Use dynamic visualization tools to demonstrate the updating process of the stereo correlation network structure;

[0053] The entire evaluation process is arranged on a visual interface for interaction with users.

[0054] The technical effects and advantages of the multi-dimensional fine-grained intelligent driving system evaluation method of the present invention are as follows:

[0055] A multi-level three-dimensional association network is designed. Through the three-dimensional association network, the evaluation dimensions are decomposed into a multi-level structure, and the dynamic influence relationship between dimensions is represented by directed edges. This not only achieves fine-grained decomposition of the evaluation dimensions, but also can truly simulate the complex interactions between dimensions, thereby improving the scientificity and accuracy of the evaluation.

[0056] Based on the graph attention network, an influence evolution propagation mechanism is designed. In the update of the three-dimensional association network, the influence evolution propagation mechanism is introduced. Combined with the saturation effect (sigmoid mapping), gating mechanism (update gate and candidate state), score difference drive (diffusion propagation) and damping effect, the propagation and evolution of influence between dimensions are dynamically simulated.

[0057] Dynamic mode selection is achieved through a cognitive computing-based operational paradigm tree, integrating data from multiple sources (road posture, environmental data, and vehicle status parameters). This tree structure breaks down mode selection into subtasks, inferring priorities and confidence levels through cognitive computing models. This allows for flexible switching of driving strategies based on real-time scenarios to meet diverse driving needs.

[0058] When switching driving modes, a smooth transition mechanism using a gradual state machine is introduced. This mechanism implements weight switching through linear interpolation and control parameter switching through intermediate states. Furthermore, an RNN algorithm is used to evaluate scene hopping and dynamically adjust the switching time window. This significantly reduces risks during the switching process (such as safety hazards or reduced comfort), providing significant value in the field of intelligent driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Schematic diagram of the steps of a multi-dimensional fine-grained intelligent driving system evaluation method of the present invention;

[0060] Figure 2 This is a structural diagram of a multi-dimensional fine-grained intelligent driving system evaluation method of the present invention. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example 1

[0062] See also Figure 1 As shown, the multi-dimensional fine-grained intelligent driving system evaluation method described in this embodiment includes:

[0063] When evaluating intelligent driving systems, it is generally necessary to assess safety (whether the system can avoid accidents and respond to emergencies or adverse situations, such as sudden rain or snow, driving at night, or pedestrians suddenly crossing the road), efficiency (system performance in terms of time, energy consumption, and path optimization), and comfort (user experience, including acceleration and deceleration smoothness, lane change strategies, etc.). However, there are often trade-offs or conflicts between different evaluation dimensions. For example:

[0064] Safety vs. Efficiency: Overly conservative driving strategies (such as maintaining excessive distance between vehicles) may improve safety but reduce traffic efficiency.

[0065] Comfort vs. Efficiency: To improve comfort (e.g., smooth acceleration), the system may need to sacrifice efficiency (e.g., missing a green light).

[0066] Safety vs. Comfort: In certain emergency situations, emergency accident avoidance may result in reduced comfort (e.g., safety becomes difficult to ensure due to the impact of comfort, and accidents cannot be avoided).

[0067] These conflicts make it challenging to scientifically balance and comprehensively evaluate intelligent driving when evaluating system design. Furthermore, safety, comfort, and efficiency are relatively general and broad concepts in the evaluation process, making it a technical challenge to break down these evaluation dimensions into quantifiable metrics. For example, the performance of a perception module can be broken down into metrics such as object recognition rate, false detection rate, and missed detection rate, but these metrics still need to be defined and measured.

[0068] Therefore, design and innovation are needed to meet actual needs when conducting multi-dimensional and fine-grained evaluation of intelligent driving systems.

[0069] A multi-dimensional, fine-grained intelligent driving system evaluation method includes: collecting road posture and environmental data during the intelligent driving process, dynamically selecting three driving modes (high-speed defense, baseline cruise, and environmental penetration speed) based on a cognitive computing-based operational paradigm tree, and using a gradual state machine to smoothly switch between driving modes;

[0070] A pre-set multi-layer perception evaluation system is used to switch driving modes in various test scenarios. Based on each selected driving mode, the system is used to evaluate the safety, efficiency, and comfort of the intelligent driving system, as well as its overall score.

[0071] The multi-layer perception evaluation system includes a stereo association network, which forms a multi-level stereo representation by refining the granularity of multiple dimensions and forms a complete stereo association network through node simulation and directed edge connection.

[0072] For any node in the three-dimensional correlation network, the influence evolution propagation mechanism is used to update it. The influence evolution propagation mechanism is designed based on the graph attention network, and the saturation effect of energy propagation, the gating mechanism, the score difference drive of diffusion propagation and the damping effect of energy propagation are introduced to realize the real diffusion of the influence interaction between evolution dimensions.

[0073] Based on the problems in the background technology, the solution is analyzed in detail, and the specific contents are as follows:

[0074] 1. Traditional systems may select driving modes based on only a single dimension or fixed rules and cannot dynamically adapt to complex scenarios.

[0075] To address these issues, this solution defines three driving modes (High Speed Defense, Baseline Cruise, and Environmental Penetration Speed), prioritizing safety, efficiency, and comfort, respectively. This design ensures the system selects the most appropriate mode for different scenarios, rather than relying on a single mode for all scenarios. By incorporating a cognitive computing model, it dynamically infers and selects driving modes based on real-time data collected from road conditions (slope, curvature, road friction coefficient), environmental data (weather, visibility, traffic density), and vehicle state parameters (speed, acceleration). This overcomes the limitations of static rules and adapts to complex and changing driving scenarios. By defining clear trigger conditions for each driving mode (e.g., slope > q1, visibility < n1), and by using parameter thresholds (e.g., q1 > q2, n1 > n2), the system achieves fine-grained mode differentiation, ensuring that mode selection is closely aligned with scenario requirements.

[0076] 2. Traditional systems lack quantitative assessment and dynamic priority adjustment mechanisms for multiple dimensions such as safety, efficiency, and comfort.

[0077] To address these issues, this solution quantifies the priorities of safety, efficiency, and comfort within each driving mode using weighted values, ensuring that the primary dimension (such as safety in Highway Defense Mode) receives the highest weight. This quantification approach makes multi-dimensional trade-offs more scientific and controllable. Hard constraints are set for key metrics for each driving mode (such as emergency braking response time, object detection recall, and braking distance), and different thresholds are set based on the characteristics of the mode (e.g., A1 < A2 < A3, B1 > B2 > B3, etc.). This ensures that the system meets minimum performance requirements in all situations, avoiding performance shortcomings caused by inappropriate trade-offs. By using a cognitive computing model to infer the priorities and confidence levels of subtasks within the operational paradigm tree, the system can dynamically adjust the priorities of each dimension to further optimize the multi-dimensional trade-offs.

[0078] 3. Traditional systems may cause sudden changes in control parameters when switching driving modes, affecting safety, comfort and efficiency.

[0079] To address these issues, this solution introduces a gradual transition state machine to achieve smooth driving mode switching. This involves switching weights and control parameters, dynamically adjusting the length of the switching window based on the magnitude of the jump. This approach ensures a smooth transition of control parameters such as speed, acceleration, and following distance, minimizing the impact of mode switching on driving experience and safety.

[0080] Therefore, this solution addresses the issues of singleness and static mode selection through multimodal design and cognitive computing models; addresses the difficulty of multi-dimensional trade-offs through weight quantization and hard constraints; addresses the abruptness of mode switching through a gradual state machine and switching time windows; addresses the lack of adaptability in complex scenarios through multi-dimensional data input and parameter threshold design; and addresses the coarse-grained nature of evaluation methods through fine-grained indicator design and dynamic evaluation. This makes the evaluation and operation of intelligent driving systems more scientific, precise, and efficient, enabling a safe, efficient, and comfortable driving experience in a variety of complex scenarios.

[0081] The three driving modes of high-speed defense, baseline cruise and environmental penetration speed include:

[0082] In each driving mode, a weight value is used to quantify the priority of each key dimension, and the primary dimension has the largest weight value. Among them, the key dimensions are safety, efficiency and comfort, and the primary dimension is the most prioritized key dimension in each driving mode;

[0083] For example, in high-speed defense mode, safety is the primary goal, so the weight value of safety is assigned the highest, such as 0.8, and efficiency and comfort are second, which can be set to 0.1 and 0.1.

[0084] Hard constraints are set for key indicators for each driving mode to ensure that minimum performance requirements (such as safety indicators) are met under all circumstances. For the high-speed defense mode, the safety constraints are set as follows: the emergency braking response time is less than A1, for example, A1 = 300ms, the target detection recall rate (an important indicator for measuring the model's detection capability, indicating the model's ability to correctly identify all real targets) is greater than B1, for example, B1 = 98%, and the braking distance (calculated based on vehicle speed and road conditions) is less than C1, for example, C1 = 10m. The efficiency constraint is set as the path length optimization rate is greater than D1, for example, D1 = 80% (a certain degree of efficiency sacrifice is allowed). The comfort constraint is set as the acceleration change rate is less than E1, for example, E1 = 3.0 m / s³ (an appropriate reduction in comfort is allowed).

[0085] For the baseline cruise mode, safety constraints are set as emergency braking response time less than A2, for example, A2 = 500ms, target detection recall rate greater than B2, for example, B2 = 95%, and braking distance less than C2, for example, C2 = 15m. Efficiency constraints are set as path length optimization rate greater than D2, for example, D2 = 90%, and average travel time less than 1.1 times the baseline time. Comfort constraints are set as acceleration change rate less than E2, for example, E2 = 2.0 m / s³. Lateral acceleration can be introduced, for example, lateral acceleration less than 2.0 m / s².

[0086] For the environmental penetration speed mode, the safety constraints are set as emergency braking response time less than A3, for example, A3 = 700ms, target detection recall rate greater than B3, for example, B3 = 90%, and braking distance less than C3, for example, C3 = 5m. The efficiency constraint is set as path length optimization rate less than D3, for example, D3 = 70% (low efficiency requirement). The comfort constraint is set as acceleration change rate less than E3, for example, E3 = 1.5 m / s³. Lateral acceleration can be introduced, for example, lateral acceleration less than 1.0 m / s².

[0087] Among them, A1, A2, A3, B1, B2, B3, C1, C2, C3, D1, D2, D3, E1, E2 and E3 are pre-set values, and the values meet the requirements of A1 < A2 < A3; B1 > B2 > B3; C1 > C2 > C3; D1 > D2 > D3; E1 > E2 > E3;

[0088] The path length optimization rate is (theoretical optimal path length / actual driving path length) × 100%. The theoretical optimal path length refers to the shortest path length from the starting point to the end point calculated by a path planning algorithm (such as Dijkstra or A*) under ideal conditions (no obstacles, no traffic regulations, and no dynamic environmental interference). The actual driving path length refers to the total length of the vehicle's track recorded during actual driving (obtained through a high-precision odometer or GPS track integration).

[0089] In high-speed defense mode, it is allowed to sacrifice 20% path efficiency in exchange for safety (such as actively bypassing potential risk areas); in baseline cruise mode, safety and efficiency are balanced, and the path deviation from the optimal value does not exceed 10%. In environmental penetration speed mode, more than 30% path redundancy is accepted (such as multiple adjustments to posture in narrow areas and refined obstacle avoidance).

[0090] The method for collecting road posture, environmental data, and vehicle status parameters during intelligent driving and dynamically selecting three driving modes, namely, high-speed defense, baseline cruise, and environmental penetration speed, based on a cognitive computing-based operation paradigm tree includes the following:

[0091] Road posture data includes slope, curvature, and road friction coefficient; environmental data includes weather data, visibility, and traffic density; vehicle state parameters include speed and acceleration;

[0092] In the evaluation process of multi-dimensional, fine-grained intelligent driving systems, road posture and environmental data during intelligent driving are the basis for system decision-making and evaluation. The specific collection content and methods are as follows:

[0093] Road posture data is used to describe the geometric and physical characteristics of the vehicle's driving path, and mainly includes the following parameters:

[0094] Slope: This value, expressed as a percentage (%), indicates the longitudinal inclination of the road. This value is obtained using an onboard IMU (Inertial Measurement Unit) or high-precision map data, with a measurement accuracy of ±0.1%.

[0095] Curvature: Measured in 1 / m, this value represents the curvature of the road. Calculated using onboard cameras, LiDAR, or high-precision map data, curvature calculations must consider the continuity of the road's centerline and maintain an accuracy of ±0.01 / m.

[0096] Road friction coefficient: A dimensionless parameter that represents the friction characteristics between the road surface and the tire. It is estimated using onboard sensors (such as tire pressure sensors and accelerometers) combined with a road material recognition algorithm, with an accuracy of ±0.05.

[0097] Environmental data is used to describe the external conditions of vehicle driving, directly affecting the selection of driving modes and the evaluation of system performance. It mainly includes the following parameters:

[0098] Weather data: Categorized into sunny, rainy, snowy, and foggy categories. Weather information is obtained through on-board meteorological sensors (such as rain sensors and temperature sensors) or V2X (vehicle-to-everything) communication. Weather classification accuracy must exceed 95%.

[0099] Visibility: Measured in meters (m), this value indicates the visible distance in the current environment. This value is estimated using an onboard LiDAR or camera combined with image processing algorithms. The measurement error must be within ±10m.

[0100] Traffic density: Measured in vehicles per kilometer, it indicates the density of vehicles on a road. This data is based on vehicle-to-everything (V2X) communication, onboard radar, or cameras, and is combined with road length estimation to achieve an accuracy of ±5 vehicles per kilometer.

[0101] Vehicle status data is used to describe the dynamic behavior of the vehicle and is an important basis for mode selection and evaluation. It mainly includes the following parameters:

[0102] Vehicle speed: Indicates the vehicle's real-time speed in km / h. This value is obtained through an on-board speed sensor or GPS module, with a measurement error within ±0.5 km / h.

[0103] Acceleration: Indicates the vehicle's longitudinal and lateral acceleration in m / s². Measured by the vehicle's IMU, the accuracy must reach ±0.01 m / s².

[0104] Collected data is preprocessed, perhaps using Kalman filtering or low-pass filtering to remove sensor noise and ensure data smoothness and consistency. Multi-source data can also be aligned using timestamps to ensure temporal consistency across sensors. Synchronization errors must be controlled within ±10ms. Anomalous data should be identified and removed using statistical methods (such as the Z score) or machine learning algorithms (such as the Isolation Forest algorithm). The proportion of anomalous data should be kept below 1%.

[0105] The cognitive computing-based operational paradigm tree is a decision-making framework based on cognitive computing. By constructing a multi-level decision tree, it decomposes complex driving scenarios into multiple subtasks and dynamically selects the optimal driving mode based on input parameters. The specific design is as follows:

[0106] The cognitive computing operation paradigm tree is set to consist of a root node and intermediate nodes, and the input is road posture, environmental data and vehicle state parameters. Among them, the root node represents the driving mode selection, and the intermediate nodes are used for subtask decomposition, that is, the intermediate nodes include subtasks such as road posture analysis, environmental assessment, and vehicle state judgment.

[0107] At each intermediate node, a cognitive computing model (such as Bayesian networks, fuzzy logic, or deep neural networks) is used to reason about the input data and output the priority and confidence of the subtask;

[0108] The driving modes include high-speed defense, baseline cruise, and environmental penetration speed. High-speed defense prioritizes safety over efficiency and comfort, prioritizing safety in high-risk scenarios (such as steep slopes and slippery roads) to minimize accident risks. Trigger conditions include: road attitude: slope > q1 (e.g., q1 = 10%), curvature > s1 (e.g., s1 = 0.1 / m), and road friction coefficient < d1 (e.g., d1 = 0.5); environmental data: severe weather (e.g., rain, snow, fog), visibility < n1 (e.g., n1 = 100m), and traffic density > g1 (e.g., g1 = 50 vehicles / km); and vehicle status: speed > c1 (e.g., c1 = 80km / h), and acceleration rate > w1 (w1 = 3.0m / s³).

[0109] The baseline cruise mode prioritizes efficiency over safety and comfort, optimizing travel time and energy consumption while maintaining basic safety and comfort. Trigger conditions are: In road posture, slope ≤ q1 (e.g., q1 = 10%), curvature ≤ s1 (e.g., s1 = 0.1 / m), and road friction coefficient ≥ d1 (e.g., d1 = 0.5); in environmental data, weather data is moderate (e.g., clear or light rain), visibility ≥ n1 (e.g., n1 = 100m), traffic density ∈ (g2, g1) (e.g., g1 = 50 vehicles / km, g2 = 20 vehicles / km); in vehicle status, speed ∈ (c2, c1) (e.g., c1 = 80km / h, c2 = 30km / h), and acceleration change rate ≤ w1 (w1 = 3.0m / s³).

[0110] The environmental penetration speed mode prioritizes comfort over safety and efficiency to improve user experience (such as stability). Trigger conditions are: road attitude: slope ≤ q2 (e.g., q2 = 5%), curvature > s2 (e.g., s2 = 0.2 / m, parking lot or narrow road); environmental data: visibility < n2 (e.g., n2 = 50m), traffic density > g3 (e.g., congestion); vehicle status: speed < c2, acceleration change rate ≤ w2 (w2 = 2m / s³).

[0111] Among them, q1, q2, s1, s2, d1, n1, n2, g1, g2, g3, c1, c2, w1 and w2 are all pre-set values; q1>q2, s1<s2; d1; n1>n2; g1>g2>g3; c1>c2; w1>w2, which can be calculated by taking the average value or weighted average based on real-time data analysis, or can be predicted by a machine model (such as deep learning).

[0112] The vehicle's road posture, environmental data, and vehicle status parameters are collected in real time at a fixed frequency hb, and input into the cognitive computing operation paradigm tree to output the selected driving mode. If the currently output driving mode is the same as the previous output, there is no need to switch the driving mode. If the currently output driving mode is different from the previous output, a gradual state machine is used to smoothly switch the driving mode.

[0113] To avoid sudden changes when switching driving modes (such as sudden acceleration, deceleration, or steering wheel shaking), the system uses a gradual state machine to achieve smooth switching of driving modes. The method of using a gradual state machine for smooth switching of driving modes includes:

[0114] Smooth switching includes switching the weight of driving modes and switching the control parameters of the vehicle;

[0115] For driving mode weight switching, preset the weights of safety, efficiency, and comfort in each driving mode, add a switching time window T (for example, T = 5 seconds), and gradually adjust each weight within the window using linear interpolation until the preset weight value in the driving mode is reached;

[0116] For vehicle control parameter switching, an intermediate state is introduced during the switch. This intermediate state represents the intermediate values of the control parameters between the two modes. These control parameters include vehicle speed, acceleration, and following distance. For example, when switching from high-speed defense to baseline cruise control, the vehicle speed switches from 10 km / h to 30 km / h. The intermediate value of the intermediate state can be set to 20 km / h. The RNN algorithm is used to evaluate the scene hop (i.e., the complexity of the scene, which can be determined by the evaluation value output by the RNN algorithm. The larger the evaluation value, the more complex the scene. When using the RNN algorithm, road posture and environmental data can be used as input data, and historical data can be used for pre-training the RNN algorithm). The switching time window is lengthened by a multiplier based on the scene hop. The adjusted switching time window is obtained and used as the final switching time window. This is then applied to the control parameter switching process. For example, assuming a scene hop of 5, the pre-set lengthening multiplier is 3 for scene hops between 4 and 6, meaning that the final switching time window is three times the length of the initial switching time window.

[0117] The above solution is designed to address the potential for large control parameter jumps caused by direct switching in certain complex scenarios. To this end, an intermediate state is introduced as a transition, and the switching time window is doubled to ensure smoothness.

[0118] Each mode switching event (including trigger conditions, switching time, weight changes, control parameter adjustments, etc.) is recorded in the log. The log format includes timestamp, input parameters, mode change, switching smoothness score, etc.

[0119] Using a cognitive computing-based operational paradigm tree, the system dynamically selects three driving modes (High Speed Defense, Baseline Cruise, and Environmental Penetration Speed) based on real-time collected road posture, environmental data, and vehicle status parameters. Each mode is assigned specific weights and hard constraints for different scenarios (e.g., inclement weather, high-speed cruise, and comfortable driving), ensuring robustness in diverse real-world driving scenarios and meeting the needs of different users (e.g., those prioritizing safety or efficiency).

[0120] A gradual state machine enables smooth driving mode switching, including weight switching (linear interpolation) and control parameter switching (introduction of intermediate states). The length of the switching time window is dynamically adjusted based on the scenario hops assessed by the RNN algorithm, ensuring a smooth and seamless switching process. This reduces the impact of driving mode switching on vehicle control, improves driving safety and comfort, and enhances the user experience.

[0121] The design method of the multi-layer perception evaluation system includes:

[0122] Step L1: Construct a three-dimensional association network;

[0123] Taking safety, efficiency, and comfort as key dimensions, pre-set sub-dimensions under the key dimensions. The sub-dimensions are decomposed into sub-sub-dimensions. The key dimensions, sub-dimensions, and sub-sub-dimensions are distributed on the same level. This continues in this way until all levels are filled. Each level has dimensions. All dimensions are simulated into nodes. Any two nodes are connected by directed edges, and attribute values are attached to each directed edge, forming a three-dimensional network structure.

[0124] Directed edges represent the mutual influence between dimensions, pointing from the influencer to the influenced. For example, if perception directly influences decision-making ability, an edge exists. The direction of the edge indicates that perception points to decision-making ability, and the attribute value represents the strength of the influence, which can be positive or negative (i.e., positive influence strength is represented by a positive number, and negative influence strength is represented by a negative number).

[0125] For example, the multi-layer perception evaluation system consists of three layers. The first layer is the key dimension node, including safety, efficiency, and comfort. The second layer is the sub-dimension node. For example, the sub-dimensions under safety include perception capability, decision-making capability, control capability, and fault tolerance capability. The sub-dimensions under perception capability are on the third layer, including object detection capability and dynamic target tracking capability.

[0126] Among them, it is necessary to design the calculation method for each dimension. For example, perception ability refers to the accuracy and robustness of the system's perception of the environment. Object detection ability is reflected by object recognition rate (the proportion of correctly identified targets), false detection rate (the proportion of incorrectly identified targets), and missed detection rate (the proportion of unrecognized targets). Specifically, in the scene, labeled data sets (such as pedestrians, vehicles, and obstacles) are used for testing and to calculate the above indicators.

[0127] Step L2: The last level of the dimensional association network defines quantifiable metrics. However, these metrics are no longer measured independently, but instead serve as inputs to the network to calculate the performance of the entire system. For example, the object detection capability of object recognition rate (the proportion of correctly identified objects), false positive rate (the proportion of incorrectly identified objects), and missed detection rate (the proportion of unrecognized objects) are all quantifiable metrics defined and used as inputs to the dimensional association network.

[0128] Initialize the attribute value of each directed edge and the energy value of each node, and use the influence evolution propagation mechanism to update the attribute value of the directed edge in the three-dimensional association network, thereby evolving the energy value of the node;

[0129] The energy values of the nodes corresponding to the key dimensions in the three-dimensional association network are obtained as the final output values, that is, the scores of safety, efficiency and comfort are obtained.

[0130] By building a multi-layered perception evaluation system and a three-dimensional correlation network, the technical solution can break down the three key dimensions of safety, efficiency, and comfort into sub-dimensions and sub-sub-dimensions, ultimately yielding quantifiable metrics (such as emergency braking response time and path length optimization rate). This fine-grained decomposition results in more comprehensive and specific evaluation results, avoiding the ambiguity or overgeneralization inherent in traditional evaluation methods. Furthermore, these fine-grained metrics can be used to pinpoint performance bottlenecks (such as insufficient object detection recall in the perception module) and enable targeted system optimization.

[0131] The method for initializing the attribute value of each directed edge and the energy value of each node includes:

[0132] Pre-train machine learning models (such as neural networks) in different modes, and use historical data to input the model in the corresponding mode to obtain the attribute value of each directed edge;

[0133] The expert scoring method is used to obtain the initial score of each node and perform normalization.

[0134] The influence evolution propagation mechanism is used to update the attribute values of directed edges in the three-dimensional association network, thereby evolving the energy values of the nodes. The methods include:

[0135] Since the actual driving environment is not independent, changes in each dimension will cause ripples and affect the surrounding dimensions. Therefore, when updating the attribute values of directed edges and the energy values of nodes, a comprehensive propagation model is needed. This design is based on the powerful expressive power of graph neural networks (GNNs) and combines the functions of gated propagation models, discrete-time dynamic propagation models, energy propagation models, and diffusion propagation models. Specifically, based on the graph attention mechanism (GAT), it will incorporate some functions of the gating mechanism (used to dynamically adjust the influence intensity of propagation to avoid excessive propagation or accumulation of negative influences), discrete-time dynamics (used to simulate the dynamic evolution of propagation and ensure the convergence of cyclic dependencies), energy propagation (used to introduce damping and saturation effects to simulate energy loss during propagation), and diffusion propagation (used to introduce score difference-driven smooth propagation to simulate the balance process between dimensions) to form a comprehensive influence evolution propagation mechanism. The details are as follows:

[0136] The indicators of the last level in the stereo association network are collected as input, the evaluation period is preset, and the indicators are discretized into time steps for dynamic iteration;

[0137] At the current time step, the update direction is passed from the lowest level to the highest level in sequence. For any two nodes connected by a directed edge, the influence evolution propagation mechanism is used to update until the entire three-dimensional association network is updated. The iteration is repeated at the next time step until the preset evaluation cycle time is exhausted. The iteration stops and the energy value of each node and the attribute value of the directed edge are updated.

[0138] Then the energy values of the nodes corresponding to safety, efficiency and comfort on the first level are output as the scores of the key dimensions;

[0139] Construct a comprehensive scoring function to evaluate the overall system performance. This involves normalizing the energy values of the nodes corresponding to safety, efficiency, and comfort, and performing a weighted summation based on the corresponding weights under the selected driving mode to obtain a comprehensive score.

[0140] The scores of key dimensions and the comprehensive scores are output as the evaluation scores of the intelligent driving system.

[0141] We set up a variety of test scenarios, including high-risk scenarios (such as pedestrians crossing the road, inclement weather), standard scenarios (such as urban roads and highways), and low-speed scenarios (such as parking lots and congested roads). We calculated scores for each test scenario and analyzed the differences in system performance in different scenarios. For example, a system may perform well on highways but poorly on congested roads.

[0142] The influence evolution propagation mechanism is designed based on the graph attention network, including:

[0143] For any two nodes in the three-dimensional association network, at each time step, the attention weights of the nodes corresponding to the influencer and the influenced are calculated using the graph attention network (GAT) according to the direction of the directed edge. The specific calculation process is to use a learnable linear transformation matrix to map the energy value of the node to the high-order feature space, and then use the softmax function to calculate the attention weight of the node. This represents the importance of influence propagation and serves as the new attribute value of the directed edge. In addition, the saturation effect of energy propagation is added. The new energy value received by the corresponding node of the influenced node is the weighted sum of the attention weights of all predecessor nodes (such as all the influencer nodes corresponding to the influenced node), the initial attribute value of the directed edge, and the energy value.

[0144] In the process of influence transmission, a gating mechanism is introduced to dynamically adjust the influence intensity of the propagation. The gating mechanism is similar to the update gate in GRU, which is used to control the fusion ratio of the new energy value and the current energy value. The specific method is as follows: the update gate is the Sigmoid function value calculated by the weighted sum of the current energy value and the new energy value, which is used to indicate the degree of acceptance of the new energy value; the candidate state is the hyperbolic tangent function value calculated by the weighted sum of the current energy value and the new energy value, which is used to indicate the state after the new energy value is fully received. The weighted sum of the current energy value and the candidate state is used to update the node energy value. The weight is determined by the update gate to achieve the update of the node energy value.

[0145] Based on the node energy value update, a score difference-driven diffusion propagation is introduced to simulate smooth propagation between dimensions. This involves obtaining the energy value difference between the nodes corresponding to the influencer and the influenced, taking the weighted sum of the initial energy values of all predecessor nodes corresponding to the influenced node and the energy value difference, and then multiplying the product with the diffusion coefficient (used to control the diffusion speed) to obtain the diffusion term. The updated energy value is then accumulated with the diffusion term to obtain the corrected energy value of the node.

[0146] Based on the node energy value correction, the damping effect of energy propagation is introduced to simulate the energy loss during the propagation process, including the introduction of a damping coefficient to control the energy attenuation rate. The absolute value is obtained by subtracting the damping coefficient from 1. For the current time step, the final state of the energy value at the node in the next time step is the product of the corrected energy value and the absolute value.

[0147] The saturation effect of energy propagation includes calculating the Sigmoid function value of the energy value of the node corresponding to the influencer and mapping it to the (0, 1) interval to simulate the saturation effect and avoid infinite growth of the energy value.

[0148] To update the stereoscopic association network, we introduced an influence evolution propagation mechanism based on a graph attention network, combining saturation effects (sigmoid mapping), gating mechanisms (updating gates and candidate states), score difference drive (diffusion propagation), and damping effects. This mechanism can dynamically simulate the propagation and evolution of influences between dimensions.

[0149] The graph attention network captures the important influence relationships between dimensions through attention weights, avoiding the limitations of traditional equal dissemination.

[0150] The saturation effect and damping effect prevent the energy value from being infinitely amplified or rapidly decayed, thus improving the stability of the model.

[0151] The gating mechanism and score difference drive enhance the nonlinear expression ability of the propagation process, making the evaluation results closer to the real scene.

[0152] Therefore, the complex interactions between modules in the intelligent driving system can be better simulated, significantly improving the authenticity and robustness of the evaluation.

[0153] Use graph visualization tools (such as Gephi and NetworkX) to draw a three-dimensional network structure, showing the relationships between nodes and edges and the performance of each dimension, to help developers and users intuitively understand system performance. Nodes at each level are the same size, and starting from the first level, the node shapes gradually decrease. Directed edges are drawn with different colors based on the attribute values, and positive and negative impacts are set based on the colors.

[0154] Use dynamic visualization tools to demonstrate the updating process of the stereo correlation network structure;

[0155] The entire evaluation process is arranged on a visual interface for interaction with users.

[0156] Use visualization tools to intuitively identify key trade-offs between dimensions. For example, if an edge's attribute value is high and negative, it indicates that improving security will result in a significant decrease in efficiency, and system design optimization is needed to mitigate this conflict. If a node has a high impact on multiple nodes but a low score, it indicates that this node is a system bottleneck and requires priority optimization. If an edge's attribute value is negative and large, it indicates a significant conflict between the two, and algorithmic or policy adjustments are needed to mitigate the conflict. Example 2

[0157] See also Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A multi-dimensional fine-grained intelligent driving system evaluation system is provided, including:

[0158] Data acquisition module: used to collect road posture, environmental data and vehicle status parameters during intelligent driving;

[0159] Driving mode dynamic switching module: Dynamically selects three driving modes: high-speed defense, baseline cruise, and environmental penetration speed based on a cognitive computing-based operation paradigm tree, and uses a gradual state machine to smoothly switch driving modes;

[0160] Multi-dimensional fine-grained evaluation module: Based on the driving mode of each switch selection, a multi-layer perception evaluation system is used to evaluate the intelligent driving system's safety, efficiency, and comfort scores, as well as the overall score, and conduct visual interaction. Example 3

[0161] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the multi-dimensional fine-grained intelligent driving system evaluation method provided above is implemented.

[0162] Since the electronic device introduced in this embodiment is an electronic device used to implement a multi-dimensional fine-grained intelligent driving system evaluation method in the embodiment of this application, based on the multi-dimensional fine-grained intelligent driving system evaluation method introduced in the embodiment of this application, those skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the embodiment of this application, it falls within the scope of protection of this application.

[0163] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0164] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A multi-dimensional fine-grained intelligent driving system evaluation method, characterized by: include: The system collects road posture, environmental data, and vehicle status parameters during intelligent driving, dynamically selects three driving modes: high-speed defense, baseline cruise, and environmental penetration speed based on a cognitive computing-based operation paradigm tree, and uses a gradual state machine to smoothly switch between driving modes. A pre-set multi-layer perception evaluation system is used to switch driving modes in various test scenarios. Based on each selected driving mode, the system is used to evaluate the safety, efficiency, and comfort scores of the intelligent driving system, as well as a comprehensive score, and to provide visual interaction. The multi-layer perception evaluation system includes a stereo association network, which forms a multi-level stereo representation by refining the granularity of multiple dimensions and forms a complete stereo association network through node simulation and directed edge connection. For any node in the three-dimensional association network, an influence evolution propagation mechanism is used to update it. The influence evolution propagation mechanism is designed based on the graph attention network, and the saturation effect of energy propagation, the gating mechanism, the score difference drive of diffusion propagation, and the damping effect of energy propagation are introduced to realize the true diffusion of the influence interaction between evolutionary dimensions. The influence evolution propagation mechanism is designed based on the graph attention network and includes: For any two nodes in the three-dimensional association network, at each time step, the graph attention network is used to calculate the attention weights of the influencer and the influenced node according to the direction of the directed edge, which are used as the new attribute value of the directed edge. The energy propagation and saturation effect are added. The new energy value received by the influenced node is the weighted sum of the attention weights of all predecessor nodes, the initial attribute value of the directed edge, and the energy value. In the process of influence transfer, a gating mechanism is introduced to control the fusion ratio of the new energy value and the current energy value. The specific method is as follows: the update gate is the Sigmoid function value calculated by the weighted sum of the current energy value and the new energy value; the candidate state is the hyperbolic tangent function value calculated by the weighted sum of the current energy value and the new energy value. The node energy value is updated using the weighted sum of the current energy value and the candidate state, and the weight is determined by the update gate; Based on the node energy value update, a score difference drive for diffusion propagation is introduced. This involves obtaining the energy value difference between the influencer and the affected node, taking the weighted sum of the initial energy values of all predecessor nodes corresponding to the affected node and the energy value difference, then multiplying the result by the diffusion coefficient to obtain the diffusion term. The updated energy value is then added to the diffusion term to obtain the corrected energy value of the node. On the basis of the node energy value correction, the damping effect of energy propagation is introduced, including the introduction of the damping coefficient. The absolute value is obtained by subtracting the damping coefficient from 1. For the current time step, the final state of the energy value at the node in the next time step is the product of the corrected energy value and the absolute value. The saturation effect of energy propagation includes calculating the Sigmoid function value of the energy value of the node corresponding to the influencer and mapping it to the interval (0, 1).

2. A multi-dimensional fine-grained intelligent driving system evaluation method according to claim 1, characterized in that: The three driving modes of high-speed defense, baseline cruise and environmental penetration speed include: In each driving mode, a weight value is used to quantify the priority of each key dimension, and the primary dimension has the largest weight value. Among them, the key dimensions are safety, efficiency and comfort, and the primary dimension is the most prioritized key dimension in each driving mode; Hard constraints are set for key indicators for each driving mode to meet minimum performance requirements in all situations. For the high-speed defense mode, the safety constraints are set to emergency braking response time less than A1, target detection recall rate greater than B1, and braking distance less than C1; the efficiency constraint is set to path length optimization rate greater than D1, and the comfort constraint is set to acceleration change rate less than E1. For the baseline cruise mode, the safety constraints are set as emergency braking response time less than A2, target detection recall rate greater than B2, and braking distance less than C2; the efficiency constraint is set as path length optimization rate greater than D2; and the comfort constraint is set as acceleration change rate less than E2. For the environmental penetration speed mode, the safety constraints are set as emergency braking response time less than A3, target detection recall rate greater than B3, and braking distance less than C3; the efficiency constraint is set as path length optimization rate less than D3; and the comfort constraint is set as acceleration change rate less than E3. Among them, A1, A2, A3, B1, B2, B3, C1, C2, C3, D1, D2, D3, E1, E2 and E3 are pre-set values, and the values meet the requirements of A1<A2<A3; B1>B2>B3; C1>C2>C3; D1>D2>D3; E1>E2>E3.

3. A multi-dimensional fine-grained intelligent driving system evaluation method according to claim 2, characterized in that: The method for collecting road posture, environmental data, and vehicle status parameters during intelligent driving and dynamically selecting the three driving modes of high-speed defense, baseline cruising, and environmental penetration speed based on a cognitive computing operation paradigm tree includes: Road posture data includes slope, curvature and road friction coefficient; environmental data includes weather data, visibility and traffic density; vehicle state parameters include speed and acceleration; The cognitive computing operation paradigm tree is composed of a root node and intermediate nodes, and the input is road posture, environmental data and vehicle state parameters. The root node represents the driving mode selection, and the intermediate nodes are used for subtask decomposition. At each intermediate node, a cognitive computing model is used to reason about the input data and output the priority and confidence of the subtask; The driving modes include high-speed defense, baseline cruise, and environmental penetration speed. The high-speed defense mode prioritizes safety and is triggered by the following conditions: slope > q1, curvature > s1, and road friction coefficient < d1 in the road posture; weather data in the environmental data is severe weather, visibility < n1, and traffic density > g1; and vehicle speed > c1 and acceleration change rate > w1 in the vehicle status. The baseline cruise mode prioritizes efficiency, and the triggering conditions are: slope ≤ q1, curvature ≤ s1, and road friction coefficient ≥ d1 in the road posture; weather data in the environmental data is moderate weather, visibility ≥ n1, and traffic density ∈ (g2, g1); vehicle speed ∈ (c2, c1), and acceleration change rate ≤ w1 in the vehicle state; The environmental penetration speed mode prioritizes comfort, and the triggering conditions are: slope ≤ q2, curvature > s2 in road posture; visibility < n2, traffic density > g3 in environmental data; vehicle speed < c2, acceleration change rate ≤ w2 in vehicle status; Wherein, q1>q2, s1<s2; d1; n1>n2; g1>g2>g3; c1>c2; w1>w2, and q1, q2, s1, s2, d1, n1, n2, g1, g2, g3, c1, c2, w1 and w2 are all preset values; The vehicle's road posture, environmental data, and vehicle status parameters are collected in real time at a fixed frequency hb, and input into the cognitive computing operation paradigm tree to output the selected driving mode. If the currently output driving mode is the same as the previous output, there is no need to switch the driving mode. If the currently output driving mode is different from the previous output, a gradual state machine is used to smoothly switch the driving mode.

4. A multi-dimensional fine-grained intelligent driving system evaluation method according to claim 3, characterized in that: The method for smoothly switching the driving mode using the gradual state machine includes: Smooth switching includes switching the weight of driving modes and switching the control parameters of the vehicle; For driving mode weight switching, the weights of safety, efficiency, and comfort in each driving mode are preset, and a switching time window is added. Within the window, each weight is gradually adjusted using linear interpolation until the preset weight value in the driving mode is reached; For the vehicle control parameter switching, an intermediate state is introduced in the middle of the switching. The intermediate state is the intermediate value of the control parameters in the two modes. The control parameters include vehicle speed, acceleration and following distance. The RNN algorithm is used to evaluate the scene jump. The lengthening multiple of the switching time window is set according to the scene jump. The adjusted switching time window is obtained and used as the final switching time window and applied to the control parameter switching process.

5. A multi-dimensional fine-grained intelligent driving system evaluation method according to claim 4, characterized in that: The design method of the multi-layer perception evaluation system includes: Step L1: Construct a three-dimensional association network; Taking safety, efficiency, and comfort as key dimensions, pre-set sub-dimensions under the key dimensions. The sub-dimensions are decomposed into sub-sub-dimensions. The key dimensions, sub-dimensions, and sub-sub-dimensions are distributed on the same level. This continues until all levels are filled. Each level has dimensions. All dimensions are simulated into nodes. Any two nodes are connected by directed edges, and attribute values are attached to each directed edge, forming a three-dimensional network structure. Among them, directed edges represent the mutual influence relationship between dimensions, from the influencer to the influenced, and the attribute value is used to represent the intensity of the influence, including positive and negative influences; Step L2: The dimensions of the last level in the stereo association network define quantifiable indicators, and these indicators are used as inputs of the stereo association network; Initialize the attribute value of each directed edge and the energy value of each node, and use the influence evolution propagation mechanism to update the attribute value of the directed edge in the three-dimensional association network, thereby evolving the energy value of the node; Obtain the energy value of the node corresponding to the key dimension in the three-dimensional association network as the final output value.

6. A multi-dimensional fine-grained intelligent driving system evaluation method according to claim 5, characterized in that: The method for initializing the attribute value of each directed edge and the energy value of each node includes: Pre-train machine learning models in different modes, and use historical data to input the model in the corresponding mode to obtain the attribute value of each directed edge; The expert scoring method is used to obtain the initial score of each node and perform normalization.

7. The multi-dimensional fine-grained intelligent driving system evaluation method according to claim 6, characterized in that: The method of using the influence evolution propagation mechanism to update the attribute value of the directed edge in the three-dimensional association network, thereby evolving the energy value of the node, includes: The indicators of the last level in the stereo association network are collected as input, the evaluation period is preset, and the indicators are discretized into time steps for dynamic iteration; At the current time step, the update direction is passed from the lowest level to the highest level in sequence. For any two nodes connected by a directed edge, the influence evolution propagation mechanism is used to update until the entire three-dimensional association network is updated. The iteration is repeated at the next time step until the preset evaluation cycle time is exhausted. The iteration stops and the energy value of each node and the attribute value of the directed edge are updated. Then the energy values of the nodes corresponding to safety, efficiency and comfort on the first level are output as the scores of the key dimensions; Construct a comprehensive scoring function to evaluate the overall system performance. This involves normalizing the energy values of the nodes corresponding to safety, efficiency, and comfort, and performing a weighted summation based on the corresponding weights under the selected driving mode to obtain a comprehensive score. The scores of key dimensions and the comprehensive scores are output as the evaluation scores of the intelligent driving system.

8. The multi-dimensional fine-grained intelligent driving system evaluation method according to claim 7 is characterized in that: The visual interaction method includes: Use graph visualization tools to draw a three-dimensional association network structure. The nodes in each layer are the same size. Starting from the first layer, the node shapes gradually become smaller. The thickness of the directed edges is drawn according to the size of the attribute value. Directed edges are drawn using different colors, and positive and negative influences are set based on the color. Use dynamic visualization tools to demonstrate the updating process of the stereo correlation network structure; The entire evaluation process is arranged on a visual interface for interaction with users.

Images