Automatic test method and system for recurrence problem of vehicle-mounted navigation system

Abstract

The invention relates to the technical field of system testing, and discloses an automatic test method and system for a recurrence problem of a vehicle-mounted navigation system, and the method comprises the following steps: building a static road network topology through analyzing an urban high-precision map, generating a dynamic traffic flow model through combining historical traffic data, and injecting a random traffic event; the method comprises the following steps: establishing a three-layer scene model fused with a spatio-temporal evolution law, pre-calculating a basic complexity difficulty value of each scene, forming a virtual scene library with a label, establishing a multi-modal noise model based on physical characteristics of a sensor, and synthesizing a cross-dimensional interference mode by using a generative adversarial network. According to the method, the exploration limitation of a traditional method is broken through by constructing a high-fidelity virtual scene and multi-modal interference collaborative injection mechanism and adopting an intelligent evolutionary strategy, and the continuously optimized test capability is formed by combining interpretable diagnosis and closed-loop feedback, so that the double improvement of the reliability verification efficiency and depth of the vehicle-mounted navigation system is realized.

Description

Technical Field

[0001] The present invention relates to the field of system testing, and more specifically, to an automatic testing method and system for reproducing problems in a vehicle-mounted navigation system. Background Art

[0002] With the rapid development of intelligent connected vehicles, the testing complexity faced by vehicle-mounted navigation systems has increased exponentially. Traditional methods based on real road tests are limited by high costs, low scenario coverage, etc., and cannot truly obtain the state of multi-dimensional urban road conditions pressure. When conducting relevant vehicle-mounted navigation system tests, it is difficult to effectively cope with the navigation failure problem in high-density urban road conditions. Summary of the Invention

[0003] The present invention provides an automatic testing method and system for reproducing problems in a vehicle-mounted navigation system, and solves the technical problem in the related art that path planning cannot be realized due to multiple changes in multi-dimensional urban road conditions.

[0004] The present invention provides an automatic testing method for reproducing problems in a vehicle-mounted navigation system, including the following steps: S100, constructing a static road network topology by parsing a high-precision urban map, generating a dynamic traffic flow model in combination with historical traffic data, injecting random traffic events, establishing a three-layer scenario model integrating spatio-temporal evolution laws, and pre-computing the basic complexity difficulty values of each scenario to form a tagged virtual scenario library; S200, establishing a multi-modal noise model based on the physical characteristics of sensors, synthesizing cross-dimensional interference patterns using a generative adversarial network, dynamically adjusting the interference intensity and time frequency according to the scenario difficulty, controlling the spatio-temporal synchronization of multi-sensor data streams, and ensuring the rationality of interference through fidelity verification; S300, dynamically scheduling the test scenario difficulty using a curriculum learning mechanism, improving the exploration reward function of the reinforcement learning strategy to prevent premature convergence, generating adversarial edge scenarios to supplement test cases, integrating multiple exploration strategies for co-evolution, and reversely correcting the scenario difficulty evaluation system with the test results; S400, aggregating multi-source data of test logs, interference parameters, and scenario features, constructing a component-level fault propagation map to quantify the influence weights of system modules, generating a visualization diagnosis report integrating spatio-temporal associations, and optimizing the scenario difficulty parameters and test strategies through a closed-loop feedback mechanism.

[0005] Further, S100 specifically includes the following steps: S110, constructing the static layer of the road topology: parsing the high-precision map data, extracting the road network topology structure, and constructing static scenario elements; S120, generating dynamic traffic flow: generating time-varying traffic flow based on historical traffic data and simulating vehicle behavior interactions; S130, Injection into the emergency event layer: Randomly generate traffic abnormal events and their spatio-temporal influence scopes; S140, Spatio-temporal graph network modeling: Construct a probabilistic spatio-temporal graph network to quantify the coupling relationship of each layer element; S150, Initial difficulty calculation: Pre-calculate the basic difficulty of the scenario for subsequent learning and scheduling; The calculation formula for the basic difficulty of the scenario is as follows: ; Where represents the basic difficulty of the scenario, represents the number of intersections, represents the length of road section ; represents the traffic flow increase ratio during peak hours, represents the total number of emergency events, represents the influence intensity of event m, represents the event number, represents the intersection density, represents the peak coefficient, represents the event intensity; The output product is a structured scenario description file: ; Where represents the total duration of the scenario, represents the static road network layer, represents the dynamic traffic flow layer at time t, represents the emergency event layer, represents the scenario description file.

[0006] Furthermore, in S200, it specifically includes the following steps: S210, Sensor noise baseline modeling: Establish the background noise model of each sensor; S220, Multi-modal interference generation: Generate cross-dimensional interference patterns based on IWGAN; Generator objective function: ; Discriminator constraint: ; Where represents the generator network, represents the discriminator network, represents the random noise input, represents the noise distribution, represents the real data sample, represents the generated fake sample, denotes the gradient penalty coefficient, denotes the Lipschitz constraint boundary, denotes the gradient operator with respect to z, denotes the discriminator's output for real samples, denotes the discriminator's output for generated samples; S230, Difficulty Adaptive Jamming Injection: Dynamically adjust the jamming intensity according to the scene's basic difficulty; S240, Cross-Sensor Synchronization Control: Ensure the spatio-temporal consistency of multi-sensor jamming; S250, Jamming Effect Verification: Pre-verify the rationality of jamming in a virtual environment.

[0007] Furthermore, in S300, it specifically includes the following steps: S310, Course Difficulty Dynamic Scheduling: Adjust the scene difficulty based on the scene's basic difficulty and real-time test feedback; S320, Maximum Entropy Exploration Strategy Optimization: Improve the DRL reward mechanism to prevent premature strategies; S330, Adversarial Edge Scenario Generation: Generate high-value test samples using the actual noise intensity; S340, Multi-Strategy Co-Evolution: Integrate three exploration strategies for synchronous evolution; S350, Scene Difficulty Closed-Loop Feedback: Update the scene difficulty library with the test results in reverse.

[0008] Furthermore, in S320, the calculation formula for maximum entropy exploration strategy optimization is as follows: Entropy-Enhanced Reward Function: ; Policy Update Rule: ; where denotes the original reward, denotes the enhanced reward, denotes the policy entropy, denotes the state visit frequency, denotes the entropy reward coefficient, denotes the state penalty coefficient, denotes the policy network parameters, denotes the parameterized policy, denotes the state-action value function, denotes the baseline function, denotes the policy gradient, denotes the expectation operator.

[0009] Furthermore, in S330, the calculation formula for adversarial edge scenario generation is as follows: Perturbation gradient calculation: ; Projection constraint: ; where represents the adversarial perturbation, represents the perturbation step size, represents the adversarial sample, represents the original sample, represents the Q-value of the adversarial sample, respectively represent the minimum and maximum value ranges of the sample, represents the clipping function, represents the sign function, represents the gradient operator for x.

[0010] Furthermore, in S340, the calculation formula for multi-strategy co-evolution is as follows: Policy weight update: ; Scoring function: ; where represents the weight of policy i, represents the score of policy i, represents the policy selection temperature, represents the policy the number of unique faults discovered, represents the execution times of policy i, represents the entropy value of policy i.

[0011] Furthermore, in S350, the calculation formula for scenario difficulty closed-loop feedback is as follows: Difficulty correction: ; Complexity update: ; where represents the updated difficulty value, represents the base difficulty value, represents the failure rate impact factor, represents the number of failures, represents the total number of tests, represents the complexity of scenario s, represents the DRL exploration steps of scenario s, represents the average exploration steps, represents the updated scenario complexity.

[0012] Further, in S400, it specifically includes the following steps: S410, Multi-source fault data fusion: Aggregate test logs, interference parameters, and static scenario features; S420, Fault propagation graph construction: Construct a fault impact network at the system component level based on GNN; S430, Component contribution quantification: Calculate the contribution degree of each system component to the fault; S440, Explainable report generation: Generate a visual diagnostic report including spatio-temporal associations; S450, Closed-loop feedback optimization: Update the scenario library in S100 and the strategies in S300 in reverse with the diagnostic results.

[0013] The present invention also provides an automatic test system for reproducing problems in a vehicle navigation system, including: Virtual scenario construction module: Parse the urban high-precision map to construct a static road network topology, combine historical traffic data to generate a dynamic traffic flow model, inject random traffic events, and finally establish a three-layer scenario model integrating spatio-temporal evolution laws. At the same time, pre-compute the basic complexity difficulty values of each scenario to form a labeled virtual scenario library; Multi-modal interference simulation module: Establish a multi-modal noise model based on the physical characteristics of sensors, use a generative adversarial network to synthesize cross-dimensional interference patterns, dynamically adjust the interference intensity and time frequency according to the scenario difficulty, strictly control the spatio-temporal synchronization of multi-sensor data streams, and ensure the rationality of interference through fidelity verification; Intelligent test evolution engine: Adopt a curriculum learning mechanism to dynamically schedule the test scenario difficulty, improve the exploration reward function of the reinforcement learning strategy to prevent premature convergence, generate adversarial edge scenarios to supplement test cases, integrate multiple exploration strategies for co-evolution, and reverse-correct the scenario difficulty evaluation system with the test results; Fault diagnosis feedback module: Aggregate multi-source data such as test logs, interference parameters, and scenario features, construct a component-level fault propagation map to quantify the influence weights of system modules, generate a visual diagnostic report integrating spatio-temporal associations, and optimize the scenario difficulty parameters and test strategies through a closed-loop feedback mechanism; Virtual-real fusion verification module: Compare the virtual test and real vehicle road test results in a digital twin environment, verify the test effectiveness through multi-source data alignment, generate a visual test report, and provide decision support information such as a fault reproduction path and a system vulnerability heat map.

[0014] The beneficial effects of the present invention are as follows: By constructing a high-fidelity virtual scenario and a multi-modal interference co-injection mechanism, the present invention significantly improves the test coverage rate and fault reproduction accuracy under complex urban road conditions; Explore the limitations of traditional methods by adopting an intelligent evolution strategy, and effectively discover potential defects of the navigation system in edge scenarios; Combine interpretable diagnosis with closed-loop feedback to form a continuously optimized testing ability, ultimately achieving a double improvement in the efficiency and depth of the reliability verification of in-vehicle navigation systems, and providing a strong guarantee for intelligent driving safety. Description of the Drawings

[0015] Figure 1 is a flowchart of an automatic testing method for reproducing problems of an in-vehicle navigation system proposed by the present invention; Figure 2 is a structural block diagram of an automatic testing system for reproducing problems of an in-vehicle navigation system proposed by the present invention. Detailed Embodiments

[0016] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0017] As Figure 1 shown, an automatic testing method for reproducing problems of an in-vehicle navigation system includes the following steps: S100, Spatiotemporal decoupling scenario modeling: Construct a static road network topology by parsing the urban high-precision map, generate a dynamic traffic flow model in combination with historical traffic data, and inject random traffic events. Finally, establish a three-layer scenario model that integrates spatiotemporal evolution laws, and pre-compute the basic complexity difficulty values of each scenario to form a virtual scenario library with labels; In an embodiment of the present invention, the specific steps are as follows: S110, Construction of the static layer of the road topology: Parse the high-precision map data, extract the road network topology structure, and construct static scenario elements; ; ; ; ; Wherein represents a road node (coordinate + type), represents the three-dimensional coordinates of the node, represents the node type, represents the road segment connection attribute, represents the road segment width, Indicates the number of lanes, Indicates the road section slope, where Indicates the total number of road segments, Indicates the set of node types, Indicates static scene features; S120, Dynamic traffic flow generation: Generate time-varying traffic flow based on historical traffic data and simulate vehicle behavior interactions; ; Among them, the vehicle kinematic model: ; Among them Indicates the dynamic traffic flow characteristics at time t, Indicates the vehicle number, Indicates the number of dynamic vehicles, Indicates the simulation time step (default 0.1s), Indicates the position vector of vehicle k at time t, Indicates the position vector of vehicle k at time t+1, Indicates the velocity vector of vehicle k at time t, Indicates the velocity vector of vehicle k at time t+1, Indicates the acceleration vector of vehicle k at time t; S130, Emergency event layer injection: Randomly generate traffic abnormal events (accidents, construction areas, etc.) and their spatio-temporal influence ranges; ; Event impact function: ; Among them Indicates the emergency event layer characteristics, Indicates the total number of emergency events, Indicates the start time of event m, Indicates the end time of event m, Indicates the influence area of event m, Indicates the impact function of event m, Indicates the impact intensity at position (x,y) and time t, Indicates the central position coordinates of event m, Indicates the impact intensity coefficient (0.3~1.0), Indicates the spatial influence radius (50~200m); S140, Spatio-temporal graph network modeling: Construct a probabilistic spatio-temporal graph network (PSTGN) to quantify the coupling relationship of elements in each layer; ; Edge weight update rule: ; where represents the total number of nodes in the graph network, represents the scene state transition probability, represents the scene state at time t + 1, represents the scene state at time t, represents the state of node i at time t + 1, represents the set of parent nodes of node i at time t, represents the set of neighbor nodes of node i, represents the node similarity (based on spatial distance and functional attributes), represents the graph network temperature coefficient, represents the edge weight from node i to j at time t; S150, Initial Difficulty Calculation: Pre - calculate the basic scene difficulty for subsequent course learning scheduling; ; where represents the basic scene difficulty, represents the number of intersections, represents the road segment length (in meters), represents the traffic flow increase ratio during peak hours (0.0 - 2.0), represents the total number of emergencies, represents the impact intensity of event m, represents the event number, represents the intersection density, represents the peak coefficient, represents the event intensity; The output product is a structured scene description file: ; where represents the total scene duration, represents the static road network layer, represents the dynamic traffic flow layer at time t, represents the emergency layer, represents the basic scene difficulty, represents the scene description file; S200, Sensor Interference Co - injection: Establish a multi - modal noise model based on the physical characteristics of sensors, use a generative adversarial network to synthesize cross - dimensional interference patterns, dynamically adjust the interference intensity and time frequency according to the scene difficulty, strictly control the spatio - temporal synchronization of multi - sensor data streams, and ensure the rationality of interference through fidelity verification; In one embodiment of the present invention, the specific steps are as follows: S210, Sensor Noise Baseline Modeling: Establish the background noise models for each sensor (GPS / IMU / camera); ; Among them, the weights of the Gaussian mixture components satisfy: ; Among them represents the sensor type index, represents the Gaussian component index, represents the sensor the th mean value of the Gaussian component, represents the corresponding standard deviation, from associated road section noise, represents the weight of the k-th Gaussian component, represents the Gaussian distribution function, represents the baseline noise model of sensor s, represents being proportional to; S220, Multi-modal Interference Generation: Generate cross-dimensional interference patterns (space + time + logic) based on IWGAN; Generator objective function: ; Discriminator constraint: ; Among them represents the generator network, represents the discriminator network, represents the random noise input, represents the noise distribution, represents the real data sample, represents the generated fake sample, represents the gradient penalty coefficient, represents the Lipschitz constraint boundary, represents the gradient operator for z, represents the output of the discriminator for the real sample, represents the output of the discriminator for the generated sample; S230, Difficulty Adaptive Interference Injection: Dynamically adjust the interference intensity according to the basic difficulty of the scenario ; ; Interference frequency in the time dimension: ; Among them represents the actual noise intensity of sensor s, represents the sensor Reference noise intensity, Indicates the basic difficulty of the scenario, respectively represent the minimum and maximum values of the scenario difficulty, Indicates the interference injection frequency, Indicates the reference interference frequency; S240, Cross-sensor synchronization control: Ensure the spatio-temporal consistency of multi-sensor interference; Synchronization error constraint: ; Spatial correlation constraint: ; Wherein Indicates the set of sensors, Indicates the sensor 's time offset, Indicates the sensor 's time offset, Indicates the maximum time deviation, Indicates the sensor 's spatial error, Indicates the sensor 's spatial error, Indicates the sensor 's standard deviation, Indicates the sensor 's standard deviation, Indicates the minimum spatial correlation coefficient, Indicates the covariance operator; S250, Interference effect verification: Pre-verify the interference rationality in the virtual environment; Fidelity index: ; Through threshold control: ; Wherein Indicates the mean vector of the real data, Indicates the mean vector of the simulation data, Indicates the covariance matrix of the real data, Indicates the covariance matrix of the simulation data, Indicates the maximum allowable degradation coefficient, Indicates the baseline FID value, Indicates the Fréchet Inception Distance, Indicates the square of the L2 norm, Indicates the trace of the matrix; S300, Adaptive Test Execution and Evolution: Adopt a curriculum learning mechanism to dynamically schedule the difficulty of test scenarios, improve the exploration reward function of the reinforcement learning strategy to prevent premature convergence, generate adversarial edge scenarios to supplement test cases, integrate multiple exploration strategies for co-evolution, and reverse-correct the scenario difficulty evaluation system with test results; In one embodiment of the present invention, the specific steps are as follows: S310, Dynamic Scheduling of Curriculum Difficulty: Based on and real-time test feedback to adjust the scenario difficulty; ; ; Dynamic Difficulty Update:

[0018] where represents the initial complexity weight, represents the complexity weight at time t, represents the rarity weight at time t, represents the curriculum progress coefficient, represents the difficulty increment amplitude, represents the number of failed scenarios, represents the total number of test scenarios, represents the updated scenario difficulty; S320, Optimization of the Maximum Entropy Exploration Strategy: Improve the DRL reward mechanism to prevent premature strategies; Entropy-Enhanced Reward Function: ; Policy Update Rule: ; where represents the original reward, represents the enhanced reward, represents the policy entropy, represents the state visit frequency, represents the entropy reward coefficient, represents the state penalty coefficient, represents the policy network parameters, represents the parameterized policy, represents the state-action value function, represents the baseline function, represents the policy gradient, represents the expectation operator; S330, Generation of Adversarial Edge Scenarios: Generate high-value test samples using the actual noise intensity; Perturbation Gradient Calculation: ; Projection constraint: ; where represents the adversarial perturbation, represents the perturbation step size, represents the adversarial sample, represents the original sample, represents the Q-value of the adversarial sample, respectively represent the minimum and maximum value ranges of the sample, represents the clipping function, represents the sign function, represents the gradient operator for x; S340, Multi-strategy co-evolution: Integrate three exploration strategies to evolve synchronously. The exploration strategies include UCB, Boltzmann, and NoisyNet strategies; Policy weight update: ; Scoring function: ; where represents the weight of policy i, represents the score of policy i, represents the policy selection temperature, represents the policy the number of unique faults discovered, represents the execution times of policy i, represents the entropy value of policy i; S350, Scene difficulty closed-loop feedback: Update the scene difficulty library with the test results in reverse; Difficulty correction: ; Complexity update: ; where represents the updated difficulty value, represents the base difficulty value, represents the failure rate impact factor, represents the number of failures, represents the total number of tests, represents the complexity of scene s, represents the DRL exploration steps of scene s, represents the average exploration steps, represents the updated scene complexity; S400, Interpretability Diagnosis and Verification: Aggregate multi-source data such as test logs, interference parameters, and scenario features, construct a component-level fault propagation graph to quantify the influence weights of system modules, generate a visualization diagnosis report integrating spatio-temporal associations, and optimize scenario difficulty parameters and test strategies through a closed-loop feedback mechanism; In one embodiment of the present invention, the specific steps are as follows: S410, Multi-source Fault Data Fusion: Aggregate test logs, interference parameters, and static scenario features; ; Time Alignment Constraint: ; Wherein represents the total number of data samples, represents the i-th fault data, represents the actual noise intensity of the i-th sample, represents the basic difficulty of the i-th sample, represents the static scenario feature of the i-th sample, [[ID=२६]]represents the fault timestamp of sensor s, represents the scenario timestamp of sensor s, represents the multi-source data alignment threshold, , represents the fused dataset; S420, Fault Propagation Graph Construction: Construct a fault influence network at the system component level based on GNN; Node State Update: ; [[ID=४३]]Edge Weight Calculation: ; Wherein represents the hidden state of node v at the l-th layer, represents the learnable parameter matrix, represents the neighbor set of node v, represents the GELU activation function, represents the multi-layer perceptron, represents the edge weight from node u to v, represents the hidden state of node v at the (l + 1)-th layer, represents the hidden state of node u at the l-th layer, represents the feature vector of node u, represents the feature vector of node v, represents the feature vector of node k; S430, Component Contribution Degree Quantification: Calculate the contribution degree of each system component to the fault; It should be noted that in the above translation, the Chinese character "२६" in the original text seems to be an incorrect character. It is translated as "represents the fault timestamp of sensor s" for the sake of translation integrity. If it is a specific symbol or has a correct form, the translation should be adjusted accordingly.; Gradient significance: ; where represents the i-th system component, represents all propagation paths through component the edge weights on the path, represents the fault loss function, represents the input features of component i, represents the loss of the n-th fault sample, represents the number of fault samples, represents the contribution score of component i, represents the gradient operator for the input features of component i, represents the gradient of the fault loss with respect to the input features of component i, represents a fault propagation path, represents an edge on the path; S440, interpretable report generation: Generate a visual diagnostic report including spatio-temporal correlations; S440, interpretable report generation: Generate a visual diagnostic report including spatio-temporal correlations; Key metric calculation: ; Spatio-temporal correlation: ; where represents the relative impact index of component the contribution score of component represents the number of faults related to component i, the contribution score of component represents the total number of tests, represents the basic difficulty of the k-th sample, represents the average scenario difficulty, represents the relative impact index of the k-th sample, represents the average relative impact index, represents the spatio-temporal correlation coefficient, represents the spatio-temporal correlation coefficient, represents the component index; S450, closed-loop feedback optimization: Reverse update the Step1 scenario library and Step3 strategy with the diagnostic results; Scenario difficulty correction: ; Policy entropy value constraint: ; where represents the updated difficulty value, represents the basic difficulty value, represents the difficulty correction coefficient, represents the relative impact index, represents the number of failures, represents the total number of tests, represents the original policy entropy value, represents the updated policy entropy value, represents the policy entropy adjustment factor, represents the number of new types of faults.

[0019] Such as Figure 2 shown, the present invention also proposes an automatic test system for reproducing problems of a vehicle-mounted navigation system, including: Virtual scene construction module: By parsing the high-precision city map, construct a static road network topology, combine historical traffic data to generate a dynamic traffic flow model, and inject random traffic events, and finally establish a three-layer scene model integrating spatio-temporal evolution laws, and pre-compute the basic complexity difficulty values of each scene to form a tagged virtual scene library; Multi-modal interference simulation module: Based on the physical characteristics of sensors, establish a multi-modal noise model, use a generative adversarial network to synthesize cross-dimensional interference patterns, dynamically adjust the interference intensity and time frequency according to the scene difficulty, strictly control the spatio-temporal synchronization of multi-sensor data streams, and ensure the rationality of interference through fidelity verification; Intelligent test evolution engine: Adopt a curriculum learning mechanism to dynamically schedule the test scene difficulty, improve the exploration reward function of the reinforcement learning strategy to prevent premature convergence, generate adversarial edge scenes to supplement test cases, integrate multiple exploration strategies for co-evolution, and reverse-correct the scene difficulty evaluation system with the test results; Fault diagnosis feedback module: Aggregate multi-source data such as test logs, interference parameters, and scene features, construct a component-level fault propagation map to quantify the influence weights of system modules, generate a visual diagnostic report integrating spatio-temporal associations, and optimize the scene difficulty parameters and test strategies through a closed-loop feedback mechanism; Virtual-real fusion verification module: Compare the virtual test and real vehicle road test results in a digital twin environment, verify the test effectiveness through multi-source data alignment, generate a visual test report, and provide decision-making support information such as fault reproduction paths and system vulnerability heat maps.

[0020] Based on the above automatic test method for reproducing problems of a vehicle-mounted navigation system, the following example is given, the continuous lane-changing scene of an urban overpass; Step 1: Scene modeling Input: High-precision map of a certain elevated overpass; Generated elements: Static layer: 3-layer ring overpass, with 8 ramps and a speed limit of 60 km / h; Dynamic layer: Traffic flow density during peak hours is 120 vehicles per minute; Sudden layer: Temporary construction on Ramp 2 (influence radius 150m); Initial difficulty: = 0.68 (complex topology + high traffic flow); Step 2: Interference injection GPS interference: Simulate the positioning drift caused by overpass occlusion (lateral error ±5m); Visual interference: Add rain and fog weather rendering, visibility < 50m; Logical interference: Inject outdated map data and hide the closed information of Ramp 2; Step 3: Test evolution Course learning: Initially select the task of three consecutive lane changes; DRL exploration: Path planning deadlock is found at the 100th iteration; Generate adversarial scenario: Suddenly appear construction roadblocks 200m before the exit; Strategy coordination: NoisyNet strategy discovers the optimal detour path; Step 4: Diagnostic feedback Root cause analysis: The update frequencies of the positioning module and path planning do not match (200ms vs 500ms); Visual perception misidentifies temporary markings under low visibility; Dynamic feedback: Increase the difficulty of this scenario to = 0.75; Add the semantic label of "high-frequency lane change area" to the static layer.

[0021] Through multi-step collaborative optimization, the test coverage rate and diagnostic accuracy for complex scenarios are significantly improved.

[0022] The embodiments of the present invention are described above, but the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of the present invention.

Claims

1. An automatic test method for reproducing problems of a vehicle-mounted navigation system, characterized in that, It includes the following steps: S100. Construct a static road network topology by parsing the high-precision city map, generate a dynamic traffic flow model in combination with historical traffic data, inject random traffic events, establish a three-layer scenario model integrating spatio-temporal evolution laws, and pre-compute the basic complexity difficulty values of each scenario to form a labeled virtual scenario library; S200. Establish a multi-modal noise model based on the physical characteristics of sensors, synthesize cross-dimensional interference patterns using a generative adversarial network, dynamically adjust the interference intensity and time frequency according to the scenario difficulty, control the spatio-temporal synchronization of multi-sensor data streams, and ensure the rationality of interference through fidelity verification; S300. Adopt a curriculum learning mechanism to dynamically schedule the test scenario difficulty, improve the exploration reward function of the reinforcement learning strategy to prevent premature convergence, generate adversarial edge scenarios to supplement test cases, integrate multiple exploration strategies for co-evolution, and reverse-correct the scenario difficulty evaluation system with the test results; S400. Aggregate multi-source data of test logs, interference parameters, and scenario features, construct a component-level fault propagation map to quantify the influence weights of system modules, generate a visualization diagnosis report integrating spatio-temporal associations, and optimize the scenario difficulty parameters and test strategies through a closed-loop feedback mechanism.

2. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 1, wherein In S100, it specifically includes the following steps: S110. Construction of the static layer of road topology: Parse the high-precision map data, extract the road network topology structure, and construct static scenario elements; S120. Generation of dynamic traffic flow: Generate time-varying traffic flow based on historical traffic data and simulate vehicle behavior interactions; S130. Injection of the emergency event layer: Randomly generate traffic abnormal events and their spatio-temporal influence ranges; S140. Modeling of the spatio-temporal graph network: Construct a probabilistic spatio-temporal graph network to quantify the coupling relationship of elements in each layer; S150. Initial difficulty calculation: Pre-compute the basic scenario difficulty for subsequent learning and scheduling; The calculation formula for the basic scenario difficulty is as follows: ； wherein represents the basic difficulty of the scenario, represents the number of intersections, represents the road section length, represents the increase rate of vehicle flow during peak hours, represents the total number of emergencies, represents the impact intensity of event m, represents the event number, represents the intersection density, represents the peak factor, represents the event intensity; The output product is a structured scenario description file: ； Among them represents the total duration of the scenario, represents the static road network layer, represents the dynamic traffic flow layer at time t, represents the emergency event layer, represents the scenario description file.

3. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 2, characterized in that, In S200, it specifically includes the following steps: S210. Baseline modeling of sensor noise: Establish the background noise model of each sensor; S220. Generation of multi-modal interference: Generate cross-dimensional interference patterns based on IWGAN; Generator objective function: ； Discriminator constraint: ； Among them represents the generator network, represents the discriminator network, represents the random noise input, represents the noise distribution, represents the real data sample, represents the generated fake sample, represents the gradient penalty coefficient, represents the Lipschitz constraint boundary, represents the gradient operator for z, represents the output of the discriminator for real samples, represents the output of the discriminator for generated samples; S230. Difficulty-adaptive interference injection: Dynamically adjust the interference intensity according to the basic scenario difficulty; S240. Cross-sensor synchronization control: Ensure the spatio-temporal consistency of multi-sensor interference; S250. Verification of interference effect: Pre-verify the rationality of interference in a virtual environment.

4. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 3, wherein In S300, it specifically includes the following steps: S310. Dynamic scheduling of curriculum difficulty: Adjust the scenario difficulty based on the basic scenario difficulty and real-time test feedback; S320. Optimization of the maximum entropy exploration strategy: Improve the DRL reward mechanism to prevent premature strategies; S330. Generation of adversarial edge scenarios: Generate high-value test samples using the actual noise intensity; S340. Co-evolution of multiple strategies: Integrate three exploration strategies for synchronous evolution; S350. Closed-loop feedback of scenario difficulty: Update the scenario difficulty library with the test results in reverse.

5. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 4, characterized in that, In S320, the calculation formula for optimizing the maximum entropy exploration strategy is as follows: Entropy-enhanced reward function: ； Policy update rule: ； Among them represents the original reward represents the enhanced reward represents the policy entropy represents the state visit frequency represents the entropy reward coefficient represents the state penalty coefficient represents the policy network parameters represents the parameterized policy represents the state - action value function represents the baseline function represents the policy gradient represents the expectation operator 6. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 5, characterized in that In S330, the calculation formula for generating adversarial edge scenarios is as follows: Perturbation gradient calculation: ； Projection constraint: ； Among them represents the adversarial perturbation represents the perturbation step size represents the adversarial sample represents the original sample represents the Q-value of the adversarial sample respectively represent the minimum and maximum value ranges of the sample represents the clipping function represents the sign function represents the gradient operator for x 7. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 6, characterized in that, In S340, the calculation formula for multi-strategy co-evolution is as follows: Policy weight update: ； Scoring function: ； where represents the weight of strategy i, represents the score of strategy i, represents the strategy selection temperature, represents the strategy the number of unique faults discovered, represents the execution times of strategy i, represents the entropy value of strategy i.

8. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 7, wherein In S350, the calculation formula for closed-loop feedback of scenario difficulty is as follows: Difficulty correction: ； Complexity update: ； wherein represents the updated difficulty value, represents the basic difficulty value, represents the failure rate impact factor, represents the number of failures, represents the total number of tests, represents the complexity of scenario s, represents the DRL exploration steps of scenario s, represents the average exploration steps, represents the updated scenario complexity.

9. The automatic test method for reproducing problems of an in-vehicle navigation system according to claim 8, wherein In S400, it specifically includes the following steps: S410, Multi-source fault data fusion: Aggregate test logs, interference parameters, and static scenario features; S420, Fault propagation graph construction: Build a fault impact network at the system component level based on GNN; S430, Component contribution quantification: Calculate the contribution degree of each system component to the fault; S440, Interpretable report generation: Generate a visual diagnostic report containing spatio-temporal associations; S450, Closed-loop feedback optimization: Update the scenario library in S100 and the strategies in S300 in reverse with the diagnostic results.

10. An automatic test system for reproducing problems of a vehicle-mounted navigation system, characterized in that, Execute the steps in an automatic test method for reproducing problems of a vehicle navigation system as described in any one of claims 1-9, including: Virtual scenario construction module: Build a static road network topology by parsing a high-precision urban map, generate a dynamic traffic flow model in combination with historical traffic data, inject random traffic events, and finally establish a three-layer scenario model integrating spatio-temporal evolution laws, and pre-calculate the basic complexity and difficulty values of each scenario to form a labeled virtual scenario library; Multi-modal interference simulation module: Establish a multi-modal noise model based on the physical characteristics of sensors, synthesize cross-dimensional interference patterns using a generative adversarial network, dynamically adjust the interference intensity and time frequency according to the scenario difficulty, strictly control the spatio-temporal synchronization of multi-sensor data streams, and ensure the rationality of interference through fidelity verification; Intelligent test evolution engine: Adopt a curriculum learning mechanism to dynamically schedule the test scenario difficulty, improve the exploration reward function of the reinforcement learning strategy to prevent premature convergence, generate adversarial edge scenarios to supplement test cases, integrate multiple exploration strategies for co-evolution, and reverse-correct the scenario difficulty evaluation system with the test results; Fault diagnosis feedback module: Aggregate multi-source data such as test logs, interference parameters, and scenario features, construct a component-level fault propagation graph to quantify the influence weight of system modules, generate a visual diagnostic report integrating spatio-temporal associations, and optimize the scenario difficulty parameters and test strategies through a closed-loop feedback mechanism; Virtual-real fusion verification module: Compare the virtual test and real vehicle road test results in a digital twin environment, verify the test effectiveness through multi-source data alignment, generate a visual test report, and provide decision support information such as the fault reproduction path and the system vulnerability heat map.

Images