Driving safety evaluation system and method for unmanned vehicle and vehicle
By combining fuzzy logic and deep learning technology, the problem that existing systems cannot accurately evaluate uncertainty and ambiguity in complex traffic environments is solved, and higher accuracy and robustness of safety assessment are achieved, and the flexibility and reliability of the system are enhanced.
Patent Information
- Application Number
- CN202411850269.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-16
- Publication Date
- 2025-04-11
AI Technical Summary
Existing autonomous vehicle safety assessment systems cannot accurately assess uncertainty and ambiguity in complex traffic environments, and a single evaluation method is difficult to fully capture the complexity and variability in driving.
A security assessment system combining fuzzy logic and deep learning technology is adopted to collect and analyze vehicle status and environmental information in real time through modules such as data acquisition, preprocessing, fuzzy logic assessment and deep learning assessment, dynamically match the weights of risk factors, and generate safety assessment data through the decision fusion module.
It significantly improves the accuracy and robustness of safety assessment of driverless vehicles in complex traffic environments, enhances the flexibility and reliability of the system, can better adapt to the changing traffic environment, and reduce the risk of potential traffic accidents.
Smart Images
Figure CN119809324A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of unmanned driving control technology, and in particular to a system, method and vehicle for evaluating the driving safety of an unmanned vehicle. Background Art
[0002] The existing driverless vehicle safety assessment system, as an important support for the development of autonomous driving technology, has achieved a preliminary assessment of vehicle driving safety to a certain extent, but its core mechanism still has certain limitations. Most of these systems rely on preset hard rules and simple logical processing, such as fixed distance thresholds, speed limits, and hard enforcement of traffic rules. When dealing with relatively simple traffic scenarios, such systems can still cope, but when faced with complex and changeable real traffic environments, their limitations are clearly revealed.
[0003] Existing systems cannot accurately assess uncertainty and ambiguity. In real traffic environments, vehicles face many uncertain factors, such as pedestrian behavior prediction and other vehicles’ driving intention judgment, which are often highly ambiguous and uncertain. Existing systems can only make simple judgments based on fixed rules and logic, and cannot effectively deal with these complex situations, resulting in inaccurate assessment results in a changing traffic environment.
[0004] In addition, a single evaluation method is difficult to fully capture the complexity and variability of the driving process. When driving, unmanned vehicles need to consider multiple factors such as road conditions, weather, traffic signals, other traffic participants, etc. The interaction and influence between these factors make the driving process extremely complicated. Existing systems often use a single evaluation method, such as rule-based evaluation, model-based evaluation, etc. These methods can only focus on one aspect of safety factors, but cannot fully capture the complexity and variability of the driving process. Summary of the invention
[0005] The purpose of the present invention is to provide a driving safety assessment system, method and vehicle for unmanned vehicles, which can accurately assess the driving safety of the vehicle.
[0006] In a first aspect, a driving safety assessment system for an unmanned vehicle comprises: Data acquisition module, used to collect vehicle status information and surrounding environment information in real time; Data preprocessing module, used to denoise and normalize the collected data, and extract risk factors and key features related to vehicle driving; A fuzzy logic assessment module is used to define fuzzy sets of risk factors, define membership functions for each set, dynamically match the weights of risk factors according to traffic scenarios, and calculate fuzzy logic risk scores based on the membership functions and weights of each risk factor; The deep learning assessment module uses the trained deep learning model to conduct risk assessment on pre-processed key features and historical accident data, and outputs a deep learning risk score; The decision fusion module is used to fuse the fuzzy logic risk score with the deep learning risk score to obtain safety assessment data.
[0007] Optionally, the fuzzy logic evaluation module includes: A fuzzy set definition unit is used to define the fuzzy set of risk factors; A membership function definition unit, which defines a membership function for each fuzzy set; The fuzzy logic risk score calculation unit calculates the fuzzy logic risk score according to the membership function and the dynamically adjusted weight.
[0008] Optionally, the deep learning evaluation module includes: A data preprocessing unit for collecting, cleaning, standardizing or normalizing historical traffic data; Feature selection unit, which determines the key features that affect traffic risks in historical traffic data; The weight assignment unit assigns weights to each feature based on historical data and experience; The deep learning model training unit adopts a convolutional neural network model and uses the pre-processed historical traffic data as a training set to train the convolutional neural network model to identify risk levels under different feature combinations; The deep learning model training and evaluation unit uses a trained convolutional neural network model to perform evaluation based on the key features related to vehicle driving output by the data preprocessing module to obtain a deep learning risk score.
[0009] Optionally, the decision fusion module adopts a weighted average formula to fuse the fuzzy logic risk score and the deep learning risk score, specifically: Finalscore = (FLscore * 10 * w1 + DLscore * w2) / (10 * w1 + w2), Among them, Finalscore is the security assessment data, FLscore is the fuzzy logic risk score, DLscore is the deep learning risk score, w1 and w2 are the weights of the fuzzy logic score and deep learning score respectively, and w1 + w2 = 1.
[0010] Optionally, it also includes a safety decision module, which is used to generate corresponding safe driving decisions based on the safety assessment data.
[0011] In a second aspect, a method for evaluating the driving safety of an unmanned vehicle according to the present invention comprises the following steps: Step 1. Collect vehicle status information and surrounding environment information in real time; Step 2. De-noise and normalize the collected data, and extract risk factors and key features related to vehicle driving; Step 3. Define the fuzzy set of risk factors, define the membership function for each set, dynamically match the weights of risk factors according to traffic scenarios, and calculate the fuzzy logic risk score based on the membership function and weight of each risk factor; Step 4. Use the trained deep learning model to conduct risk assessment on the preprocessed key features and historical accident data, and output the deep learning risk score; Step 5. Fusion the fuzzy logic risk score with the deep learning risk score to obtain security assessment data.
[0012] Optionally, the step 3 is specifically: Step 31. Define fuzzy sets: Define the set of risk factors with close distance as set A; Define the set with high speed as the risk factor as set B; Step 32. Define the membership function: The membership function of set A is defined as μA(x), where x represents the actual distance between vehicles; The membership function μA(x) is defined as: When x ≤ dmin, μA(x) = 1, indicating that x completely belongs to set A; When dmin < x ≤ dmax, μA(x) = (dmax - x) / (dmax - dmin), which means that the degree to which x belongs to set A gradually decreases; When x > dmax, μA(x) = 0, indicating that x does not belong to set A; The membership function of set B is defined as μB(v), where v represents the speed of the vehicle; The membership function μB(v) is defined as: When v ≤vmin, μB(v) = 0, indicating that v does not belong to set B; When vmin < v ≤vmax, μB(v) = (v - vmin) / (vmax - vmin), which means that the degree to which v belongs to set B gradually increases; When v > vmax, μB(v) = 1, indicating that v completely belongs to set B; Among them, dmin and dmax are the minimum and maximum values of distance respectively, and vmin and vmax are the minimum and maximum values of speed respectively; Step 33: Calculate the fuzzy logic risk score: Dynamically match W according to traffic scenarios A and W B , where W A and W B are the weight values of set A and set B, respectively, indicating the importance of different risk factors in risk assessment; Fuzzy Logic Risk Score = W A * μA(x) + W B * μB(v).
[0013] Optionally, step 4 specifically includes: Step 41. Collect and pre-process historical traffic data; Step 42: Determine key features that affect traffic risk based on historical traffic data, including vehicle speed, weather conditions, road type, traffic density, and historical accidents; Step 43. Assign a weight to each key feature affecting traffic risk based on historical traffic data and experience; Step 44. Use the convolutional neural network model and the preprocessed historical traffic data as a training set to train the convolutional neural network model to identify risk levels under different feature combinations; Step 45. Based on the key features related to vehicle driving output in step 2, use the trained convolutional neural network model to perform evaluation to obtain a deep learning risk score.
[0014] Optionally, the step 5 is specifically: The weighted average formula is used to integrate the fuzzy logic risk score and the deep learning risk score. The formula is: Finalscore = (FLscore * 10 * w1 + DLscore * w2) / (10 * w1 + w2), Among them, Finalscore is the security assessment data, FLscore is the fuzzy logic risk score, DLscore is the deep learning risk score, w1 and w2 are the weights of the fuzzy logic score and deep learning score respectively, and w1 + w2 = 1.
[0015] In a third aspect, the vehicle described in the present invention adopts the unmanned vehicle driving safety assessment system described in the present invention.
[0016] Beneficial effects of the present invention: (1) The present invention significantly improves the accuracy of safety assessment of unmanned vehicles in complex traffic environments by combining fuzzy logic and deep learning technology.
[0017] (2) The present invention effectively improves the accuracy and robustness of the evaluation by combining fuzzy logic and deep learning. Fuzzy logic has unique advantages in dealing with uncertainty and ambiguity, and can better simulate the human decision-making process, especially when faced with complex and incomplete information. Deep learning can recognize and learn complex patterns and features through training with large amounts of data. Combining the two can not only utilize the powerful data processing capabilities of deep learning, but also the ability of fuzzy logic to process uncertain information, thereby achieving higher accuracy and robustness in the evaluation process.
[0018] (3) The present invention introduces a dynamic weight adjustment mechanism to improve the flexibility of the assessment. The importance of various risk factors may change in different traffic scenarios. For example, in urban traffic, the dynamic behavior of pedestrians and non-motor vehicles may be more important than in highway scenarios. By dynamically adjusting the weights of different risk factors, the system can adapt to various traffic environments more flexibly, thereby making more accurate risk assessments.
[0019] (4) The present invention also uses decision fusion technology to further improve the reliability of the evaluation. Decision fusion refers to the integration of evaluation results from different sources or different algorithms to obtain a more comprehensive and reliable evaluation conclusion. In the present invention, by integrating the evaluation results of fuzzy logic and deep learning, the limitations that may exist in a single method can be effectively avoided, thereby improving the reliability of the overall evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 It is one of the principle block diagrams of the driving safety assessment system for unmanned vehicles described in the embodiments of the present application; Figure 2 is a principle block diagram of the fuzzy logic evaluation module described in the embodiment of the present application; Figure 3 It is a principle block diagram of the deep learning evaluation module described in the embodiment of the present application; Figure 4 This is the second principle block diagram of the driving safety assessment system for unmanned vehicles described in the embodiments of the present application; Figure 5 is a flow chart of a method for evaluating driving safety of an unmanned vehicle described in an embodiment of the present application; Figure 6 is a flowchart of the fuzzy logic evaluation workflow in an embodiment of the present application; Figure 7 is a flowchart of the deep learning evaluation workflow in an embodiment of the present application; Figure 8 It is a flowchart of the decision fusion workflow in the embodiment of the present application. DETAILED DESCRIPTION
[0021] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.
[0022] like Figure 1 As shown, in an embodiment of the present application, a driving safety assessment system for unmanned vehicles includes a data acquisition module, a data preprocessing module, a fuzzy logic assessment module, a deep learning assessment module and a decision fusion module. The data acquisition module is used to collect vehicle status information (such as vehicle speed, acceleration, steering angle and other status information) and surrounding environment information (such as the position, speed and position of other vehicles and pedestrians) in real time through on-board sensors (such as radar, lidar, camera). The data preprocessing module is used to denoise and normalize the collected data, and extract risk factors and key features related to vehicle driving. The fuzzy logic assessment module is used to define a fuzzy set of risk factors, define a membership function for each set, dynamically match the weights of risk factors according to traffic scenarios, and calculate the fuzzy logic risk score according to the membership function and weight of each risk factor. The deep learning assessment module uses the trained deep learning model to perform risk assessment on the preprocessed key features and historical accident data, and outputs the deep learning risk score. The decision fusion module is used to fuse the fuzzy logic risk score with the deep learning risk score to obtain safety assessment data.
[0023] like Figure 2 As shown, in a possible embodiment, the fuzzy logic evaluation module includes a fuzzy set definition unit, a membership function definition unit and a fuzzy logic risk score calculation unit. The fuzzy set definition unit is used to define the fuzzy set of risk factors. The membership function definition unit defines a membership function for each fuzzy set. The fuzzy logic risk score calculation unit calculates the fuzzy logic risk score according to the membership function and the dynamically adjusted weight.
[0024] like Figure 3As shown, in a possible embodiment, the deep learning evaluation module includes a data preprocessing unit, a feature selection unit, a weight allocation unit, a deep learning model training unit and a deep learning model training evaluation unit. The data preprocessing unit is used to collect, clean, standardize or normalize historical traffic data. The feature selection unit determines the key features that affect traffic risks in the historical traffic data. The weight allocation unit assigns weights to each feature based on historical data and experience. The deep learning model training unit adopts a convolutional neural network model, and uses the preprocessed historical traffic data as a training set to train the convolutional neural network model to identify the risk level under different feature combinations. The deep learning model training evaluation unit uses the trained convolutional neural network model to evaluate the key features related to vehicle driving output by the data preprocessing module to obtain a deep learning risk score.
[0025] like Figure 4 As shown, in a possible embodiment, a driving safety assessment system for an unmanned vehicle further includes a safety decision module for generating a corresponding safe driving decision based on the safety assessment data.
[0026] like Figure 5 As shown, in an embodiment of the present application, a method for evaluating the driving safety of an unmanned vehicle includes the following steps: Step 1. Collect vehicle status information and surrounding environment information in real time.
[0027] Step 2. De-noise and normalize the collected data, and extract risk factors and key features related to vehicle driving.
[0028] Step 3. Define fuzzy sets of risk factors, define membership functions for each set, dynamically match the weights of risk factors according to traffic scenarios, and calculate the fuzzy logic risk score based on the membership function and weight of each risk factor.
[0029] Step 4. Use the trained deep learning model to conduct risk assessment on the preprocessed key features and historical accident data, and output the deep learning risk score.
[0030] Step 5. Fusion the fuzzy logic risk score with the deep learning risk score to obtain security assessment data.
[0031] The following is a detailed description of each step: Step 1: Data Collection The vehicle's speed, acceleration, steering angle and other status information are collected in real time through on-board sensors (such as radar, lidar, and cameras), as well as surrounding environment information such as the location and speed of other vehicles and the location of pedestrians. The sensor data must be calibrated with high precision to ensure the accuracy and reliability of the data.
[0032] Step 2: Data Preprocessing The collected data is denoised and normalized, and risk factors and key features related to vehicle driving are extracted.
[0033] Step 3: Fuzzy Logic Evaluation Define fuzzy sets of risk factors, such as "close distance", "high speed", etc., and define membership functions for each set. Use a formula to calculate the risk assessment value, which needs to consider the weights of different risk factors. The weight values are dynamically adjusted according to the traffic scenario.
[0034] like Figure 6 As shown in the figure, the specific implementation steps of fuzzy logic evaluation are as follows: Step 31. Define fuzzy sets: Define the set of risk factors that are close to each other as set A.
[0035] The set whose risk factor is high speed is defined as set B.
[0036] Step 32. Define the membership function: The membership function of set A is defined as μA(x), where x represents the actual distance between vehicles.
[0037] The membership function μA(x) is defined as: When x ≤ dmin, μA(x) = 1, indicating that x completely belongs to set A.
[0038] When dmin < x ≤dmax, μA(x) = (dmax - x) / (dmax - dmin), which means that the degree to which x belongs to set A gradually decreases.
[0039] When x > dmax, μA(x) = 0, indicating that x does not belong to set A.
[0040] The membership function of set B is defined as μB(v), where v represents the speed of the vehicle. dmin and dmax are the minimum and maximum values of the distance, respectively.
[0041] The membership function μB(v) is defined as: When v ≤vmin, μB(v) = 0, indicating that v does not belong to set B.
[0042] When vmin < v ≤vmax, μB(v) = (v - vmin) / (vmax - vmin), which means that the degree to which v belongs to set B gradually increases.
[0043] When v > vmax, μB(v) = 1, indicating that v completely belongs to set B.
[0044] Among them, vmin and vmax are the minimum and maximum values of the speed respectively.
[0045] Step 33: Calculate the fuzzy logic risk score: Dynamically match W according to the traffic scenario A and W B , where W A and W B are the weight values of set A and set B respectively, representing the importance of different risk factors in risk assessment.
[0046] Fuzzy logic risk score = W A * μA(x) + W B * μB(v).
[0047] For dynamically matching W A and W B according to the traffic scenario, specifically: For example, on a highway, the speed factor may be more important, so the value of W B can be increased; while in a congested urban section, the distance factor may be more important, so the value of W A can be increased.
[0048] The following is an example: Define fuzzy sets: Set A: "Short distance", membership function μA(x), where x represents the actual distance between vehicles.
[0049] Set B: "High speed", membership function μB(v), where v represents the speed of the vehicle.
[0050] Define membership functions: For set A: "Short distance", the membership function μA(x) is defined as: When x ≤ 10, μA(x) = 1.
[0051] When 10 < x ≤ 30, μA(x) = (30 - x) / (30 - 10).
[0052] When x > 30, μA(x) = 0.
[0053] For set B: "High speed", the membership function μB(v) is defined as: When v ≤ 60, μB(v) = 0.
[0054] When 60 < v ≤ 120, μB(v) = (v - 60) / (120 - 60).
[0055] When v>120, μB(v) = 1.
[0056] Matching weight value: On the highway, W A = 0.4, W B = 0.6.
[0057] In congested urban areas, W A = 0.6, W B = 0.4.
[0058] Calculate the risk assessment value: Risk Assessment Value = W A * μA(x) + W B * μB(v).
[0059] Assume that the distance between the vehicles is x = 15 and the speed of the vehicles is v = 100.
[0060] μA(x) = (30 - 15) / (30 - 10) = 0.5.
[0061] μB(v) = (100 - 60) / (120 - 60) = 0.6667.
[0062] Highway risk assessment value = 0.4 * 0.5 + 0.6 * 0.6667 = 0.5333.
[0063] The risk assessment value of urban congested road sections = 0.6 * 0.5 + 0.4 * 0.6667 = 0.5333.
[0064] Step 4: Deep Learning Evaluation Use a trained deep learning model (such as convolutional neural network (CNN)) to evaluate the preprocessed data and output a deep learning risk score. The deep learning model needs to be trained with a large amount of traffic scene data to ensure the accuracy and robustness of the evaluation results.
[0065] like Figure 7 As shown in Figure 1, the specific steps of deep learning evaluation include: (1) Collect and preprocess historical traffic data: Collect historical traffic data including vehicle speeds, weather conditions, road types, traffic density, accident history, etc.
[0066] Clean the data to remove outliers and missing values.
[0067] Standardize or normalize your data to ensure that the data input to your deep learning model is in a consistent format.
[0068] (2) Feature selection: Determine the key features that affect traffic risks based on historical traffic data, specifically: Identify key characteristics that influence traffic risk, such as: Feature I: Vehicle speed (the faster the speed, the higher the risk); Feature II: Weather conditions (bad weather increases risk); Feature III: road type (highways are more risky than urban roads); Feature IV: Traffic density (the greater the density, the higher the risk); Feature V: Historical accidents (high risk in accident-prone areas).
[0069] (3) Weight distribution: Based on historical traffic data and experience, a weight is assigned to each key feature that affects traffic risk, such as: The weight of feature I: 0.3; The weight of feature II: 0.2; The weight of feature C: 0.1; The weight of feature D: 0.25; The weight of feature E: 0.15.
[0070] (4) Training using a convolutional neural network (CNN) model: Construct a CNN model, which includes multiple convolutional layers, pooling layers, and fully connected layers.
[0071] Using historical traffic data as a training set, the CNN model is trained to identify risk levels under different feature combinations.
[0072] (5) Risk score calculation Based on the key features related to vehicle driving output in step 2, the trained convolutional neural network model is used for evaluation to obtain a deep learning risk score.
[0073] Assume that the input real-time data feature values are as follows: The value of feature I: Assuming the vehicle speed is 80 km / h, the value of feature I is 80; The value of feature II: Assuming it is a sunny day, the value of feature II is 1; The value of feature III: Assuming the road type is an urban road, the value of feature III is 0; The value of feature IV: Assuming the traffic density is medium, the value of feature IV is 2; The value of feature V: Assuming that there is no historical accident, the value of feature V is 0.
[0074] The trained model is as follows: Calculate the deep learning risk score based on the values of the weights and features: Assume that the weight of feature I is 0.3, the weight of feature II is 0.2, the weight of feature III is 0.1, the weight of feature IV is 0.25, and the weight of feature V is 0.15.
[0075] Deep learning risk score = 0.3 * (0.3 * 80) + 0.2 * (0.2 * 1) + 0.1 * (0.1 *0) + 0.25 * (0.25 * 2) + 0.15 * (0.15 * 0).
[0076] Deep learning risk score = 0.3*24 + 0.2*0.2 + 0.1*0 + 0.25*0.5 + 0.15*0.
[0077] Deep learning risk score = 7.2 + 0.04 + 0 + 0.125 + 0.
[0078] Risk score = 7.365.
[0079] Step 5: Decision Fusion In this step, the fuzzy logic risk score is integrated with the deep learning risk score to obtain more comprehensive and accurate security assessment data.
[0080] In one possible embodiment, the fuzzy logic risk score is first multiplied by 10 to reduce the gap between it and the deep learning risk score so that it can play a greater role in the fusion process.
[0081] Let the fuzzy logic risk score be FLscore, the deep learning risk score be DLcore, and the fused security assessment data be Finalscore. In order to more flexibly fuse the two scores, a new weighted average formula is adopted, which allows different weights to be assigned to the fuzzy logic score and the deep learning score. The specific formula is as follows: Finalscore = (FLscore * 10 * w1 + DLscore * w2) / (10 * w1 + w2); Among them, w1 and w2 are the weights of fuzzy logic score and deep learning score respectively, and w1 + w2 = 1. By adjusting the values of w1 and w2, the influence of the two scores in the fusion process can be flexibly controlled.
[0082] In the fusion process, the complementarity of the two evaluation methods is taken into account. The fuzzy logic evaluation method is usually based on experience and rules and can handle uncertainty and ambiguity, while the deep learning evaluation method trains the model through a large amount of data and can capture complex patterns and features in the data. By fusing the results of these two evaluation methods, more comprehensive and accurate security assessment data can be obtained.
[0083] Finally, the weighted average fusion score is selected as the final output for subsequent security decision-making and risk assessment. In practical applications, the values of w1 and w2 can be adjusted according to specific needs and scenarios to obtain the optimal fusion effect.
[0084] In the embodiments of the present application, the purpose of the security assessment data is: Generate corresponding safe driving decisions based on safety assessment data, such as adjusting vehicle speed, changing driving path, or performing emergency braking, etc. The decision-making process needs to consider the vehicle's dynamic characteristics and the constraints of traffic rules to ensure the rationality and feasibility of the decision.
[0085] This system can significantly improve the safety of driverless vehicles. By accurately assessing uncertainty and ambiguity, driverless vehicles can make more accurate safety assessments in complex traffic environments, thereby effectively avoiding potential traffic accidents and improving driving safety.
[0086] In addition, the system enhances the adaptability of unmanned vehicles to changing traffic environments. The system can dynamically adjust the weight of risk factors according to different traffic scenarios, so that unmanned vehicles can better adapt to various complex traffic environments and improve their reliability and efficiency in practical applications.
[0087] The system also improves the decision-making quality of autonomous vehicles. Based on comprehensive risk assessment results, autonomous vehicles can make higher-quality safe driving decisions, thereby improving overall safety. This not only helps protect the safety of the vehicle itself, but also helps protect the safety of other road users, ultimately achieving a safer, more efficient and intelligent autonomous transportation system.
[0088] In an embodiment of the present application, a vehicle adopts the unmanned vehicle driving safety assessment system as described in the embodiment of the present application.
[0089] The above-mentioned embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above-mentioned embodiments. Any other changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement modes and shall be included in the protection scope of the present invention.
Claims
1. A driving safety assessment system for unmanned vehicles, characterized in that: The system comprises: Data acquisition module, used to collect vehicle status information and surrounding environment information in real time; Data preprocessing module, used to denoise and normalize the collected data, and extract risk factors and key features related to vehicle driving; A fuzzy logic assessment module is used to define fuzzy sets of risk factors, define membership functions for each set, dynamically match the weights of risk factors according to traffic scenarios, and calculate fuzzy logic risk scores based on the membership functions and weights of each risk factor; The deep learning assessment module uses the trained deep learning model to conduct risk assessment on pre-processed key features and historical accident data, and outputs a deep learning risk score; The decision fusion module is used to fuse the fuzzy logic risk score with the deep learning risk score to obtain safety assessment data.
2. The driving safety assessment system for unmanned vehicles according to claim 1, characterized in that: The fuzzy logic evaluation module includes: A fuzzy set definition unit is used to define the fuzzy set of risk factors; A membership function definition unit, which defines a membership function for each fuzzy set; The fuzzy logic risk score calculation unit calculates the fuzzy logic risk score according to the membership function and the dynamically adjusted weight.
3. The driving safety assessment system for unmanned vehicles according to claim 1, characterized in that: The deep learning evaluation module includes: A data preprocessing unit for collecting, cleaning, standardizing or normalizing historical traffic data; Feature selection unit, which determines the key features that affect traffic risks in historical traffic data; The weight assignment unit assigns weights to each feature based on historical data and experience; The deep learning model training unit adopts a convolutional neural network model and uses the pre-processed historical traffic data as a training set to train the convolutional neural network model to identify risk levels under different feature combinations; The deep learning model training and evaluation unit uses a trained convolutional neural network model to perform evaluation based on the key features related to vehicle driving output by the data preprocessing module to obtain a deep learning risk score.
4. The driving safety assessment system for unmanned vehicles according to claim 1, characterized in that: The decision fusion module uses a weighted average formula to fuse the fuzzy logic risk score and the deep learning risk score, specifically: Finalscore = (FLscore * 10 * w1 + DLscore * w2) / (10 * w1 + w2), Among them, Finalscore is the security assessment data, FLscore is the fuzzy logic risk score, DLscore is the deep learning risk score, w1 and w2 are the weights of the fuzzy logic score and deep learning score respectively, and w1 + w2 = 1.
5. The driving safety assessment system for unmanned vehicles according to claim 1, characterized in that: It also includes a safety decision module, which is used to generate corresponding safe driving decisions according to the safety assessment data.
6. A method for evaluating the driving safety of an unmanned vehicle, characterized in that: The following steps are involved: Step 1. Collect vehicle status information and surrounding environment information in real time; Step 2. De-noise and normalize the collected data, and extract risk factors and key features related to vehicle driving; Step 3. Define the fuzzy set of risk factors, define the membership function for each set, dynamically match the weights of risk factors according to traffic scenarios, and calculate the fuzzy logic risk score based on the membership function and weight of each risk factor; Step 4. Use the trained deep learning model to conduct risk assessment on the preprocessed key features and historical accident data, and output the deep learning risk score; Step 5. Fusion the fuzzy logic risk score with the deep learning risk score to obtain security assessment data.
7. The method for evaluating driving safety of an unmanned vehicle according to claim 6, characterized in that: The step 3 is specifically as follows: Step 31. Define fuzzy sets: Define the set of risk factors with close distance as set A; Define the set with high speed as the risk factor as set B; Step 32. Define the membership function: The membership function of set A is defined as μA(x), where x represents the actual distance between vehicles; The membership function μA(x) is defined as: When x≤dmin, μA(x) = 1, indicating that x completely belongs to set A; When dmin<x≤dmax, μA(x) =(dmax - x) / (dmax - dmin), which means that the degree to which x belongs to set A gradually decreases; When x>dmax, μA(x) = 0, indicating that x does not belong to set A; The membership function of set B is defined as μB(v), where v represents the speed of the vehicle; The membership function μB(v) is defined as: When v≤vmin, μB(v)= 0, which means v does not belong to set B; When vmin<v≤vmax, μB(v) = (v- vmin) / (vmax - vmin), which means that the degree to which v belongs to set B gradually increases; When v>vmax, μB(v)= 1, indicating that v completely belongs to set B; Among them, dmin and dmax are the minimum and maximum values of distance respectively, and vmin and vmax are the minimum and maximum values of speed respectively; Step 33: Calculate the fuzzy logic risk score: Dynamically match W according to traffic scenarios A and W B , where W A and W B are the weight values of set A and set B, respectively, indicating the importance of different risk factors in risk assessment; Fuzzy Logic Risk Score = W A * μA(x) + W B * μB(v).
8. The method for evaluating driving safety of an unmanned vehicle according to claim 6, characterized in that: The step 4 specifically includes: Step 41. Collect and pre-process historical traffic data; Step 42: Determine key features that affect traffic risk based on historical traffic data, including vehicle speed, weather conditions, road type, traffic density, and historical accidents; Step 43. Assign a weight to each key feature affecting traffic risk based on historical traffic data and experience; Step 44. Use the convolutional neural network model and the preprocessed historical traffic data as a training set to train the convolutional neural network model to identify risk levels under different feature combinations; Step 45. Based on the key features related to vehicle driving output in step 2, use the trained convolutional neural network model to perform evaluation to obtain a deep learning risk score.
9. The method for evaluating driving safety of an unmanned vehicle according to claim 6, characterized in that: The step 5 is specifically as follows: The weighted average formula is used to integrate the fuzzy logic risk score and the deep learning risk score. The formula is: Finalscore = (FLscore * 10 * w1+DLscore * w2) / (10 * w1+w2), Among them, Finalscore is the security assessment data, FLscore is the fuzzy logic risk score, DLscore is the deep learning risk score, w1 and w2 are the weights of the fuzzy logic score and deep learning score respectively, and w1+w2 =1.
10. A vehicle, characterized in that: A driving safety assessment system for an unmanned vehicle as described in any one of claims 1 to 5 is used.