Vehicle expected function safety monitoring method and system

Abstract

The invention provides a vehicle expected function safety monitoring method and system, and the method comprises the steps: constructing a Bayesian network according to vehicle multi-source heterogeneous data and vehicle safety factors; determining a deviation correction factor according to the deviation between the measured value and the estimated value of the dynamic parameter of the monitored vehicle; according to the multi-source heterogeneous data set of the monitored vehicle, determining the posterior probability of the Bayesian network for each risk level under the multi-source heterogeneous data set; and correcting the posterior probability by using the deviation correction factor, and determining the risk level of the monitored vehicle according to the corrected posterior probability. According to the invention, vehicle expected function safety real-time high-precision monitoring fused with vehicle dynamics is realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent connected vehicle safety technology, and in particular to a method and system for monitoring the safety of vehicle expected functions. Background Art

[0002] With the widespread adoption of end-to-end advanced intelligent driving technologies and the widespread use of high-power motors, intelligent driving systems are improving travel efficiency and driving experience, but they are also exposing significant safety risks. Current intelligent driving systems rely on artificial intelligence models with black-box characteristics, making their decision-making logic difficult to explain. Functional safety testing also suffers from a long-tail problem, making it difficult for the system to operate stably under all expected operating environments and conditions. This makes it difficult for drivers to accurately control the system's behavior in complex scenarios. In this context, monitoring various vehicle information and the safety of its intended functions is crucial.

[0003] Currently, the industry primarily relies on traditional testing and verification methods, such as real-vehicle road testing, closed-field testing, and software simulation testing. However, these methods are unable to effectively cover long-tail scenarios and struggle to meet the stringent functional safety and information security requirements of intelligent driving systems. Therefore, to ensure system functional completeness, minimize the risk of functional deficiencies, and guarantee expected functional safety and information security, an effective method is urgently needed that can monitor vehicle dynamics and communication information in real time to promptly identify potential safety hazards. Summary of the Invention

[0004] The present invention provides a vehicle expected functional safety monitoring method and system, which are used to solve the defects in the existing technology, such as the difficulty in tracing the control logic caused by the unexplainability of the artificial intelligence model of the intelligent driving system, and the insufficient safety guarantee under extreme working conditions caused by the long-tail problem of the expected functional safety test, to achieve real-time and high-precision monitoring of the vehicle's expected functional safety that integrates vehicle dynamics.

[0005] The present invention provides a method for monitoring the safety of vehicle expected functions, comprising:

[0006] Construct a Bayesian network based on multi-source heterogeneous vehicle data and vehicle safety factors;

[0007] determining a deviation correction factor based on a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle;

[0008] Determining, based on the multi-source heterogeneous dataset of the monitored vehicles, a posterior probability of the Bayesian network for each risk level under the multi-source heterogeneous dataset;

[0009] The posterior probability is corrected using the deviation correction factor, and the risk level of the monitored vehicle is determined based on the corrected posterior probability.

[0010] According to a vehicle expected functional safety monitoring method provided by the present invention, a Bayesian network is constructed based on vehicle multi-source heterogeneous data and vehicle safety factors, including:

[0011] Construct a hierarchical node variable system based on vehicle multi-source heterogeneous data and vehicle safety factors;

[0012] Based on the expertise of experts in vehicle networks, perception algorithms, and control strategies, we systematically sort out the causal logic between the communication, perception, decision-making, correction, and risk assessment layers to construct an initial directed acyclic graph.

[0013] Based on the initial directed acyclic graph, combined with a complete data set of information covering all levels including the actual risk conditions and collision risk levels of vehicles, edges of the initial directed acyclic graph are added, deleted, or adjusted using a K2 algorithm to search for an optimal structure of the initial directed acyclic graph;

[0014] Based on the climbing algorithm, the edges of the initial directed acyclic graph are randomly adjusted to obtain a better directed acyclic graph as a Bayesian network.

[0015] According to a vehicle expected function safety monitoring method provided by the present invention, the maximization structure learning scoring function of the K2 algorithm is:

[0016]

[0017] Wherein, S(G:D) represents the score of the Bayesian network G relative to the multi-source heterogeneous dataset, P(D|G) is the probability that the Bayesian network G appears in the multi-source heterogeneous dataset, m represents the number of samples in the multi-source heterogeneous dataset, and |G| represents the number of edges in the Bayesian network G.

[0018] According to a vehicle expected function safety monitoring method provided by the present invention, a deviation correction factor is determined based on the deviation between the measured value and the estimated value of the dynamic parameter of the monitored vehicle, including:

[0019] determining the absolute and relative deviations between the measured and estimated values ​​of the kinetic parameters;

[0020] Determining the severity of the deviation of the kinetic parameter based on the absolute deviation and the relative deviation;

[0021] The deviation correction factor is determined based on the severity of the deviation of the kinetic parameters, safety, and response timeliness.

[0022] According to a vehicle expected function safety monitoring method provided by the present invention, the deviation correction factor is determined according to the severity of the deviation of the dynamic parameters, safety, and response timeliness using the following formula:

[0023] λ i =C i ·T i ·S i γ

[0024] Among them, the deviation correction factor λ i is the deviation correction factor of the i-th kinetic parameter, C i represents the safety of the ith kinetic parameter, T i represents the response time of the i-th kinetic parameter, S i represents the severity of the deviation of the ith kinetic parameter, γ represents the normalization factor,

[0025] According to a vehicle expected functional safety monitoring method provided by the present invention, based on a multi-source heterogeneous dataset of the monitored vehicle, determining the posterior probability of each risk level of a Bayesian network under the multi-source heterogeneous dataset, the method includes:

[0026] Calculating the conditional probability of each node in the Bayesian network using maximum likelihood estimation on the complete data in the multi-source heterogeneous data set;

[0027] The conditional probability of each node in the Bayesian network is calculated using Bayesian parameter estimation and Dirichlet prior distribution for the missing data in the multi-source heterogeneous dataset, and the BIC criterion is used for verification;

[0028] According to the conditional probability of each node under each risk level, the posterior probability of each risk level is obtained.

[0029] According to a vehicle expected function safety monitoring method provided by the present invention, the maximum likelihood estimation formula is:

[0030]

[0031] Among them, θ ijk In the Bayesian network, M represents the conditional probability when node i takes the jth value and its parent node takes the kth combination value. ijk In the multi-source heterogeneous dataset, the number of samples where node i takes the jth value and its parent node takes the kth combination value, r i Represents the number of combinations of parent node values ​​of node i, It represents the total number of samples when node i takes the jth value under all combinations of values ​​of node i's parent node.

[0032] According to a vehicle expected function safety monitoring method provided by the present invention, the posterior probability is corrected using the deviation correction factor, and the risk level of the monitored vehicle is determined based on the corrected posterior probability, including:

[0033] The maximum value of the corrected posterior probability for each risk level is determined by the following formula:

[0034]

[0035] Among them, L is the maximum value of the corrected posterior probability of each risk level, r is the value of the risk level, N 16 is the node variable of the risk assessment layer in the node variable system, E is the multi-source heterogeneous data set, P(N 16 =r|E) represents the posterior probability of risk level r under a given multi-source heterogeneous dataset E, and n is the number of the kinetic parameters;

[0036] The maximum value is compared with a preset threshold to determine the risk level.

[0037] The present invention also provides a vehicle expected function safety monitoring system, comprising:

[0038] A network construction module is used to construct a Bayesian network based on multi-source heterogeneous vehicle data and vehicle safety factors;

[0039] a first calculation module, configured to determine a deviation correction factor based on a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle;

[0040] A second calculation module is used to determine the posterior probability of each risk level of the Bayesian network under the multi-source heterogeneous data set according to the multi-source heterogeneous data set of the monitored vehicle;

[0041] A correction evaluation module is used to correct the posterior probability using the deviation correction factor, and determine the risk level of the monitored vehicle according to the corrected posterior probability.

[0042] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for monitoring the expected functional safety of a vehicle as described above is implemented.

[0043] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the vehicle expected functional safety monitoring method described in any one of the above is implemented.

[0044] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described methods for monitoring the safety of expected vehicle functions.

[0045] The vehicle intended functional safety monitoring method and system provided by the present invention utilize a multi-dimensional monitoring approach. Through SOME / IP packet monitoring, autonomous driving sensor numerical monitoring, and dynamic estimation data, the system monitors the vehicle in real time from multiple dimensions, including communication, sensing, and vehicle dynamics. This comprehensive coverage of every aspect of the intelligent driving system where safety issues may arise enables timely detection of potential safety hazards, significantly improving the comprehensiveness and accuracy of monitoring. Furthermore, the independent safety controller, independent of the autonomous driving system, utilizes its own independent monitoring and decision-making mechanisms to make safety assessments and interventions when the system experiences a malfunction or anomaly, avoiding situations where safety assurance cannot be guaranteed due to issues inherent in the autonomous driving system. This enhances the reliability and stability of the entire safety monitoring system and provides dual protection for vehicle driving safety. Furthermore, the real-time display system presents complex monitoring data in an intuitive interface, making it easier for drivers to understand the vehicle's safety status in real time and for technicians to monitor, debug, and maintain the system in real time, improving the system's usability and maintainability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 1 is a flow chart of a vehicle expected function safety monitoring method provided by the present invention;

[0048] Figure 2 Schematic diagram of the overall framework of the vehicle expected function safety monitoring method provided by the present invention;

[0049] Figure 3 is a schematic diagram of an initial directed acyclic graph in the vehicle expected function safety monitoring method provided by the present invention;

[0050] Figure 4 It is a structural diagram of the vehicle expected function safety monitoring system provided by the present invention. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0052] The following combination Figure 1 A vehicle intended functional safety monitoring method according to the present invention comprises:

[0053] Step 101: constructing a Bayesian network based on multi-source heterogeneous vehicle data and vehicle safety factors;

[0054] Step 102 , determining a deviation correction factor based on a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle;

[0055] Step 103 , determining the posterior probability of each risk level of the Bayesian network under the multi-source heterogeneous dataset of the monitored vehicle;

[0056] Step 104 : Correct the posterior probability using the deviation correction factor, and determine the risk level of the monitored vehicle based on the corrected posterior probability.

[0057] like Figure 2 As shown in the figure, based on the requirements of intelligent vehicle safety monitoring, the multi-source heterogeneous data collection of vehicles is realized by deploying a multi-source data collection module, specifically including:

[0058] SOME / IP packet monitoring: A libpcap-based packet capture and SOME / IP protocol parsing module is deployed within the vehicle's communication network architecture. This module, in accordance with the SOME / IP protocol specification, captures data packets transmitted between vehicle modules in real time, decodes them, extracts key data such as vehicle status information and control commands, and transmits this data to an independent safety controller.

[0059] Autonomous Driving Sensor Data Monitoring: A data acquisition and preprocessing unit is installed at the vehicle's sensor installation locations. This unit collects raw data from various sensors in real time and performs preprocessing operations such as filtering and calibration to remove noise and errors. The processed data is transmitted in real time to an independent safety controller. Simultaneously, a status monitoring device is installed near the sensors to monitor their operating status, such as power supply and signal strength. The detected sensor operating status is transmitted to the independent safety controller and displayed on the Android real-time display system.

[0060] Dynamics Estimation Data: A Kalman filter-based vehicle dynamics model is embedded in the vehicle's electronic control unit (ECU). Real-time driving data, such as speed, acceleration, steering angle, and tire pressure, is acquired through the vehicle's CAN bus and other communication methods. This data is then fed into the dynamics model for calculation, resulting in the vehicle's longitudinal, lateral, and vertical dynamic parameters. The estimated dynamics data, along with actual measured data, is transmitted in real time to an independent safety controller.

[0061] The innovative introduction of Bayesian networks within independent safety controllers allows for the deep integration of heterogeneous data from multiple sources, including SOME / IP packets, autonomous driving sensors, and dynamics estimation. By building a network structure and determining probabilistic relationships, vehicle risks can be accurately assessed.

[0062] Independent safety controller hardware was designed and installed. This controller features a high-performance processor, large-capacity storage, and a stable communication interface. A Bayesian network and a bias correction algorithm were implemented within the safety controller, enabling it to receive data from SOME / IP packet monitoring, autonomous driving sensor data monitoring, and dynamics estimation, and perform comprehensive analysis and judgment.

[0063] In terms of information security monitoring, we regularly update our intrusion detection rule base to enhance our ability to identify new cyberattacks. Regarding expected functional safety monitoring, we continuously optimize security rules and algorithm parameters based on actual testing and deviation correction. When a security risk is detected, the security controller issues an early warning signal, which can be displayed in real time on the Android system. We regularly test and update the security controller's software and hardware to ensure stable and reliable performance.

[0064] The real-time display system develops a real-time display application for vehicle expected functional safety based on embedded systems, such as Android or QNX. Designed with an intuitive interface, it displays SOME / IP packet monitoring data, sensor values, dynamic estimation data, and safety controller operating status in real time using charts, text, and indicator lights. Connecting to an independent safety controller via a wireless network enables real-time data transmission and display, and supports customizable alarm thresholds and display preferences.

[0065] This method achieves multi-dimensional data monitoring and analysis by building a complete monitoring system: deploying libpcap-based packet capture and SOME / IP protocol parsing modules to obtain vehicle communication data, using data acquisition and preprocessing units to collect and process autonomous driving sensor data, and embedding algorithms in the ECU to complete vehicle dynamics data estimation; transmitting multi-source data to an independent safety controller, monitoring information security and expected functional safety through preset security rules and algorithms, and issuing timely warnings when risks are discovered; at the same time, using a real-time display system to intuitively present key monitoring data, allowing users to understand the vehicle's safety status in real time, and supporting user-defined settings and historical data queries, thereby ensuring the expected functional safety of the vehicle under various driving conditions.

[0066] The independent safety controller uses timestamp matching and sensor calibration technology to eliminate the deviations in sampling frequency and installation position between different data sources and build a unified space-time coordinate system. The coordinate transformation formula is as follows:

[0067]

[0068] Among them, x′, y′, z′ are the coordinates of the point in the target coordinate system, x, y, z are the coordinates of the point in the source coordinate system, R is the rotation matrix, and T is the translation vector.

[0069] Z-score normalization is used for numerical parameters (such as vehicle speed and pressure), and one-hot encoding is performed on categorical data (such as message type) to ensure data scale consistency. Principal component analysis (PCA) is used to reduce data dimensionality, and continuous parameters are divided into discrete states (such as normal or abnormal) using equal-width binning, forming a symbolic representation suitable for Bayesian network processing. The formula for equal-width binning is as follows:

[0070] in,

[0071] Where x is the continuous parameter value to be discretized, i is the bin index, k is the number of bins, min(x) is the minimum value of the continuous parameter x in the dataset, and max(x) is the maximum value of the continuous parameter x in the dataset.

[0072] Based on multi-source heterogeneous data and vehicle safety factors, a node variable system covering communication, perception, decision-making, correction and risk assessment levels is defined; then, expert knowledge is combined with data-driven methods such as the K2 algorithm and hill climbing algorithm to construct a directed acyclic graph structure reflecting causal relationships; finally, maximum likelihood estimation is applied to the complete data set, and the missing data is combined with Bayesian parameter estimation and Dirichlet prior distribution to generate a conditional probability table, which is then corrected for bias and accurately calculated to determine the risk level.

[0073] This embodiment constructs a Bayesian network based on multi-source heterogeneous data and vehicle safety factors, determines a deviation correction factor according to the deviation between the measured and estimated values ​​of the dynamic parameters of the monitored vehicle, uses the deviation correction factor to correct the deviation of the Bayesian network's posterior probability for each risk level, accurately calculates the risk level, and realizes real-time, high-precision monitoring of the vehicle's expected functional safety that integrates vehicle dynamics.

[0074] Based on the above embodiment, this embodiment constructs a Bayesian network based on vehicle multi-source heterogeneous data and vehicle safety factors, including:

[0075] Construct a hierarchical node variable system based on vehicle multi-source heterogeneous data and vehicle safety factors;

[0076] Based on the expertise of experts in vehicle networks, perception algorithms, and control strategies, an initial directed acyclic graph is constructed by systematically sorting out the causal logical connections between the communication, perception, decision-making, correction, and risk assessment layers.

[0077] Based on the initial directed acyclic graph, combined with a complete data set of information covering all levels including the actual risk conditions and collision risk levels of vehicles, edges of the initial directed acyclic graph are added, deleted, or adjusted using a K2 algorithm to search for an optimal structure of the initial directed acyclic graph;

[0078] Based on the climbing algorithm, the edges of the initial directed acyclic graph are randomly adjusted to obtain a better directed acyclic graph as a Bayesian network.

[0079] Based on the characteristics of multi-source heterogeneous data and vehicle safety factors, a hierarchical node variable system is constructed as shown in Table 1.

[0080] Table 1

[0081]

[0082] During the expert knowledge initialization phase of Bayesian network construction, experts in vehicle networks, perception algorithms, and control strategies draw on their expertise to build an initial directed acyclic graph (DAG) based on their understanding of the causal logic between communication, perception, decision-making, and correction layers. For example, experts based their experience on establishing causal relationships such as "CAN frame anomaly (N1) → camera blur (N5)" and "radar failure (N6) → control command conflict (N10)," forming the foundational framework for algorithm optimization. Figure 3 It is an initialized DAG graph constructed based on the upper nodes according to expert knowledge.

[0083] First, based on the expert's initial DAG, the preprocessed dataset D (including node annotation data at each level) is input. The dataset is obtained by preprocessing the collected multi-source data, combining it with the actual risk situation of the vehicle, and annotating labels such as collision risk level to form complete data covering information at all levels.

[0084] The K2 algorithm searches for the optimal structure, attempting to add, delete, or adjust edges and iteratively updating the DAG. For example, the algorithm might find that the edge weight from "Packet Loss Rate (N2) → Radar Failure (N6)" is higher, increasing the S(G:D) score and strengthening this causal relationship. Or, it might determine that the edge from "Camera Blur (N5) → Direction Deviation (N14)" added by the expert does not help improve the data fit, decreasing the S(G:D) score and deleting the edge.

[0085] Because K2 has a local optimal problem, it is necessary to introduce a hill climbing algorithm. Starting from the initial DAG, the edges are randomly adjusted (to ensure there are no cycles), the scores are calculated, the improved changes are retained, and it is iterated until convergence to obtain a better DAG.

[0086] Based on the above embodiment, the maximum structure learning scoring function of the K2 algorithm in this embodiment is:

[0087]

[0088] Wherein, S(G:D) represents the score of the Bayesian network G relative to the multi-source heterogeneous dataset, P(D|G) is the probability of the Bayesian network G appearing in the multi-source heterogeneous dataset given the network structure, and the probability refers to the conditional probability table. m represents the number of samples in the multi-source heterogeneous dataset, and |G| represents the number of edges in the Bayesian network G.

[0089] Based on the above embodiment, in this embodiment, a deviation correction factor is determined based on the deviation between the measured value and the estimated value of the dynamic parameter of the monitored vehicle, including:

[0090] determining the absolute and relative deviations between the measured and estimated values ​​of the kinetic parameters;

[0091] Determining the severity of the deviation of the kinetic parameter based on the absolute deviation and the relative deviation;

[0092] The deviation correction factor is determined based on the severity of the deviation of the kinetic parameters, safety, and response timeliness.

[0093] The absolute deviation and relative deviation of the kinetic parameters were calculated as evidence for the deviation correction. The calculation formulas for the absolute deviation and relative deviation are as follows:

[0094] AbsDev=|estimated value - measured value|

[0095]

[0096] Where AbsDev refers to the absolute deviation and RelDev refers to the relative deviation.

[0097] Set the parameter deviation threshold. For example, if RelDev is less than 10% and AbsDev is less than 5, the parameter deviation level is N (normal) and the parameter deviation severity is S (normal). i =0; RelDev is less than 20% and AbsDev is less than 10, the parameter deviation level is M (mild abnormality), the parameter deviation severity is S i =0.3; RelDev is greater than or equal to 20% and AbsDev is greater than or equal to 10, the parameter deviation level is S (severe abnormality), the parameter deviation severity is S i =0.7.

[0098] Based on the above embodiment, in this embodiment, the deviation correction factor is determined according to the severity of the deviation of the kinetic parameters, safety, and response timeliness by the following formula:

[0099] λ i =C i ·T i·S i γ

[0100] Among them, the deviation correction factor λ i is the deviation correction factor of the i-th kinetic parameter, C i represents the safety of the i-th dynamic parameter, referring to the Automotive Safety Integrity Level (ASIL) of the parameter in ISO 26262; T i represents the response timeliness of the i-th kinetic parameter, based on the safety time requirement of ISO 26262; S i represents the severity of the deviation of the ith kinetic parameter, γ represents the normalization factor, max(C i ·T i ·S i ) represents C i ·T i ·S i where i represents different dynamic parameters such as velocity, acceleration, etc.

[0101] Based on the above embodiment, in this embodiment, the posterior probability of each risk level of the Bayesian network under the multi-source heterogeneous data set of the monitored vehicle is determined according to the multi-source heterogeneous data set, including:

[0102] Calculating the conditional probability of each node in the Bayesian network using maximum likelihood estimation on the complete data in the multi-source heterogeneous data set;

[0103] The conditional probability of each node in the Bayesian network is calculated using Bayesian parameter estimation and Dirichlet prior distribution for the missing data in the multi-source heterogeneous data set, and finally verified by the BIC criterion to generate a conditional probability table;

[0104] According to the conditional probability of each node under each risk level, the posterior probability of each risk level is obtained.

[0105] Based on the above embodiment, the formula for maximum likelihood estimation in this embodiment is:

[0106]

[0107] Among them, θ ijk In the Bayesian network, M represents the conditional probability when node i takes the jth value and its parent node takes the kth combination value. ijk In the multi-source heterogeneous dataset, the number of samples where node i takes the jth value and its parent node takes the kth combination value, r i Represents the number of possible value combinations of the parent node of node i, It represents the total number of samples when node i takes the jth value under all possible combinations of values ​​of node i's parent node.

[0108] Based on the above embodiment, this embodiment uses the deviation correction factor to correct the posterior probability, and determines the risk level of the monitored vehicle according to the corrected posterior probability, including:

[0109] The maximum value of the corrected posterior probability for each risk level is determined by the following formula:

[0110]

[0111] Among them, L is the maximum value of the corrected posterior probability of each risk level, r is the value of the risk level, N 16 is the node variable of the risk assessment layer in the node variable system, E is the multi-source heterogeneous data set, P(N 16 =r|E) represents the posterior probability of risk level r under a given multi-source heterogeneous dataset E, n is the number of the kinetic parameters, represents the sum of the deviation correction factors of n kinetic parameters;

[0112] The maximum value is compared with a preset threshold to determine the risk level.

[0113] During the inference phase, a real-time, multi-source, heterogeneous dataset is input into the Bayesian network as evidence E. Evidence E collects multi-source data in real time from the communication layer (such as CAN frame anomaly rate and packet loss rate), the perception layer (camera ambiguity and radar fault status), and the decision layer (control command conflicts and trajectory planning errors). After preprocessing such as timestamp alignment and normalization, an evidence set is formed. The posterior probability is calculated using variable elimination. The posterior probability calculation formula is as follows:

[0114]

[0115] Among them, P(N 16 =r|E) represents the posterior probability of risk level r given evidence E, which is obtained by multiplying the conditional probabilities; P(E) is the sum of the joint probabilities corresponding to evidence E at all risk levels.

[0116] For example, when the final output optimal risk level decision result L is less than 0.4, it is low risk; when L is between 0.4 and 0.7, it is medium risk; and when L is greater than 0.7, it is high risk.

[0117] The real-time display system integrates SOME / IP communications, sensor, dynamics, and safety controller data, presenting it in a variety of graphical formats. Wireless networks establish data channels with independent safety controllers, enabling instant information synchronization and simultaneous display of risk assessment results, meeting the need for real-time monitoring of vehicle safety status.

[0118] In summary, through the three modules of multi-source data acquisition module, independent safety controller module and real-time display system, real-time functional safety monitoring of autonomous driving vehicles in complex traffic environments is achieved, and risk level estimation is made.

[0119] The vehicle expected function safety monitoring system provided by the present invention is described below. The vehicle expected function safety monitoring system described below and the vehicle expected function safety monitoring method described above can refer to each other.

[0120] like Figure 4 As shown, the system includes a network construction module 401, a first calculation module 402, a second calculation module 403 and a correction evaluation module 404, wherein:

[0121] The network construction module 401 is used to construct a Bayesian network based on vehicle multi-source heterogeneous data and vehicle safety factors;

[0122] The first calculation module 402 is used to determine a deviation correction factor according to a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle;

[0123] The second calculation module 403 is used to determine the posterior probability of each risk level of the Bayesian network under the multi-source heterogeneous data set of the monitored vehicle;

[0124] The correction evaluation module 404 is configured to correct the posterior probability using the deviation correction factor, and determine the risk level of the monitored vehicle according to the corrected posterior probability.

[0125] This embodiment constructs a Bayesian network based on multi-source heterogeneous data and vehicle safety factors, determines a deviation correction factor according to the deviation between the measured and estimated values ​​of the dynamic parameters of the monitored vehicle, uses the deviation correction factor to correct the deviation of the Bayesian network's posterior probability for each risk level, accurately calculates the risk level, and realizes real-time, high-precision monitoring of the vehicle's expected functional safety that integrates vehicle dynamics.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A vehicle expected function safety monitoring method, characterized in that: include: Construct a Bayesian network based on multi-source heterogeneous vehicle data and vehicle safety factors; determining a deviation correction factor based on a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle; Determining, based on the multi-source heterogeneous dataset of the monitored vehicles, a posterior probability of the Bayesian network for each risk level under the multi-source heterogeneous dataset; The posterior probability is corrected using the deviation correction factor, and the risk level of the monitored vehicle is determined based on the corrected posterior probability.

2. The vehicle expected function safety monitoring method according to claim 1, characterized in that: A Bayesian network is constructed based on multi-source heterogeneous vehicle data and vehicle safety factors, including: Construct a hierarchical node variable system based on vehicle multi-source heterogeneous data and vehicle safety factors; Based on the expertise of experts in vehicle networks, perception algorithms, and control strategies, an initial directed acyclic graph is constructed by systematically sorting out the causal logical connections between the communication, perception, decision-making, correction, and risk assessment layers. Based on the initial directed acyclic graph, combined with a complete data set of information covering all levels including the actual risk conditions and collision risk levels of vehicles, edges of the initial directed acyclic graph are added, deleted, or adjusted using a K2 algorithm to search for an optimal structure of the initial directed acyclic graph; Based on the climbing algorithm, the edges of the initial directed acyclic graph are randomly adjusted to obtain a better directed acyclic graph as a Bayesian network.

3. The vehicle expected function safety monitoring method according to claim 2, characterized in that: The maximum structure learning scoring function of the K2 algorithm is: Wherein, S(G:D) represents the score of the Bayesian network G relative to the multi-source heterogeneous dataset, P(D|G) is the probability that the Bayesian network G appears in the multi-source heterogeneous dataset, m represents the number of samples in the multi-source heterogeneous dataset, and |G| represents the number of edges in the Bayesian network G.

4. The vehicle expected function safety monitoring method according to claim 1, characterized in that: Determining a deviation correction factor based on a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle includes: determining the absolute and relative deviations between the measured and estimated values ​​of the kinetic parameters; Determining the severity of the deviation of the kinetic parameter based on the absolute deviation and the relative deviation; The deviation correction factor is determined based on the severity of the deviation of the kinetic parameters, safety, and response timeliness.

5. The vehicle expected function safety monitoring method according to claim 4, characterized in that: The deviation correction factor is determined by the following formula based on the severity of the deviation of the kinetic parameters, safety, and response timeliness: l i =C i ·T i ·S i ·c Among them, the deviation correction factor λ i is the deviation correction factor of the i-th kinetic parameter, C i represents the safety of the ith kinetic parameter, T i represents the response time of the i-th kinetic parameter, S i represents the severity of the deviation of the ith kinetic parameter, γ represents the normalization factor, 6. The vehicle expected function safety monitoring method according to claim 1, characterized in that: Determining, based on the multi-source heterogeneous dataset of the monitored vehicle, a posterior probability of the Bayesian network for each risk level under the multi-source heterogeneous dataset, including: Calculating the conditional probability of each node in the Bayesian network using maximum likelihood estimation on the complete data in the multi-source heterogeneous data set; The conditional probability of each node in the Bayesian network is calculated using Bayesian parameter estimation and Dirichlet prior distribution for the missing data in the multi-source heterogeneous dataset, and the BIC criterion is used for verification; According to the conditional probability of each node under each risk level, the posterior probability of each risk level is obtained.

7. The vehicle expected function safety monitoring method according to claim 6, characterized in that: The formula for the maximum likelihood estimation is: Among them, θ ijk In the Bayesian network, M represents the conditional probability when node i takes the jth value and its parent node takes the kth combination value. ijk In the multi-source heterogeneous dataset, the number of samples where node i takes the jth value and its parent node takes the kth combination value, r i Represents the number of combinations of parent node values ​​of node i, It represents the total number of samples when node i takes the jth value under all combinations of values ​​of node i's parent node.

8. The vehicle expected function safety monitoring method according to claim 2, characterized in that: Correcting the posterior probability using the deviation correction factor, and determining the risk level of the monitored vehicle according to the corrected posterior probability, including: The maximum value of the corrected posterior probability for each risk level is determined by the following formula: Among them, L is the maximum value of the corrected posterior probability of each risk level, r is the value of the risk level, N 16 is the node variable of the risk assessment layer in the node variable system, E is the multi-source heterogeneous data set, P(N 16 =r|E) represents the posterior probability of risk level r under a given multi-source heterogeneous dataset E, and n is the number of the kinetic parameters; The maximum value is compared with a preset threshold to determine the risk level.

9. A vehicle expected function safety monitoring system, characterized in that: include: A network construction module is used to construct a Bayesian network based on multi-source heterogeneous vehicle data and vehicle safety factors; a first calculation module, configured to determine a deviation correction factor based on a deviation between a measured value and an estimated value of a dynamic parameter of the monitored vehicle; A second calculation module is used to determine the posterior probability of each risk level of the Bayesian network under the multi-source heterogeneous data set according to the multi-source heterogeneous data set of the monitored vehicle; A correction evaluation module is used to correct the posterior probability using the deviation correction factor, and determine the risk level of the monitored vehicle according to the corrected posterior probability.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the vehicle expected function safety monitoring method as described in any one of claims 1 to 8 is implemented.

Images