Analysis method and analysis device based on expected function safety analysis system

Abstract

The invention provides an analysis method and device based on an expected function safety analysis system, and the method comprises the steps: inputting the to-be-analyzed information of a vehicle into the expected function safety analysis system, so as to enable the expected function safety analysis system to analyze the to-be-analyzed information based on an analysis process, obtaining an expected function safety analysis result of the vehicle; the expected function security analysis result comprises analysis results aiming at a plurality of analysis dimensions, the expected function security analysis system comprises a cue word system and an expected function security language model, and the cue word system determines corresponding cue word information based on each analysis step; and the expected function security language model determines an analysis result corresponding to the cue word information based on the cue word information. Through the analysis method and the analysis device, the analysis efficiency and accuracy are greatly improved, personal errors are reduced, and the consistency and reliability of analysis results are ensured.

Description

Technical Field

[0001] The present application relates to the field of vehicle testing technology, and in particular to an analysis method and an analysis device based on an expected functional safety analysis system. Background Art

[0002] With the rapid development of autonomous driving technology, safety issues have gradually become the focus of the industry. In autonomous driving systems, in addition to traditional functional safety (Functional Safety), intended functional safety (Safety of the Intended Functionality, SOTIF) has also become particularly important. SOTIF mainly focuses on safety issues caused by insufficient system performance or changes in the external environment when the system is fault-free. These problems are particularly prominent in autonomous driving systems, because autonomous driving systems need to make fast and accurate decisions in complex and changing environments, and any slight functional deficiencies may lead to serious safety accidents. However, SOTIF analysis is more complex and cumbersome than traditional failure mode analysis, and requires in-depth analysis and evaluation of multiple aspects such as system functions and design operating conditions (ODC).

[0003] At present, most SOTIF analyses still rely on manual work. This method is not only time-consuming and labor-intensive, but also easily affected by personal experience and subjective bias, resulting in inconsistent and unreliable analysis results. Another problem with manual analysis is that it is difficult to scale and standardize. Different engineers will have different analysis methods and standards, which is particularly prominent in large projects. Therefore, how to improve the efficiency and accuracy of SOTIF analysis has become an urgent problem to be solved. Summary of the invention

[0004] In view of this, the purpose of this application is to provide an analysis method and analysis device based on an expected functional safety analysis system. The expected functional safety analysis system includes an expected functional safety language model and a prompt word system, which can automatically and comprehensively identify and evaluate the expected functional safety issues of the autonomous driving system under fault-free conditions, and generate a detailed SOTIF analysis report, which greatly improves the efficiency and accuracy of the analysis, reduces human errors, and ensures the consistency and reliability of the analysis results.

[0005] In a first aspect, an embodiment of the present application provides an analysis method based on an expected functional safety analysis system, the analysis method comprising:

[0006] Determine the analysis process of the current expected functional safety analysis task; wherein the analysis process includes multiple analysis steps, each analysis step represents a different analysis dimension;

[0007] Acquire information to be analyzed of the vehicle, and input the information to be analyzed into an expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain an expected functional safety analysis result of the vehicle; wherein the expected functional safety analysis result includes analysis results for multiple analysis dimensions, and the expected functional safety analysis system includes a prompt word system and an expected functional safety language model, the prompt word system determines the corresponding prompt word information based on each analysis step, and the expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

[0008] Further, the inputting the information to be analyzed into the expected functional safety analysis system so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle includes:

[0009] For each analysis step in the analysis process, determine the input data corresponding to the analysis step; wherein, when the analysis step has a previous analysis step connected in series with it, the input data of the analysis step is the output of the expected functional safety language model based on the prompt word information corresponding to the previous analysis step;

[0010] Inputting the input data into the prompt word system to determine the target prompt word information corresponding to the analysis step;

[0011] Inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step;

[0012] The target analysis results of each analysis step are spliced ​​in the order of each analysis step to obtain the expected functional safety analysis results of the vehicle.

[0013] Furthermore, the step of inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step includes:

[0014] Determine a filling template corresponding to the analysis step; wherein the filling template includes a content template and an associated template;

[0015] The target prompt word information is input into the expected functional safety language model, an original analysis result output by the expected functional safety language model is obtained, and the original analysis result is filled into the filling template to obtain the target analysis result.

[0016] Furthermore, after determining the target analysis result corresponding to the analysis step, the analysis method further includes:

[0017] When there are other analysis steps before this analysis step, the target analysis result corresponding to this analysis step is compared with the target analysis results corresponding to the other analysis steps. If there is a difference, the target analysis result corresponding to this analysis step is changed.

[0018] Furthermore, the expected functional safety language model is trained by the following steps:

[0019] Acquire expected functional safety analysis case data, and perform data cleaning and data structuring on the expected functional safety analysis case data to determine a training data set and a test data set;

[0020] Building an expected functional safety primitive language model based on the training data set;

[0021] A performance evaluation is performed on the expected functional safety original language model according to the test data set, and an expected functional safety original language model whose evaluation index reaches a preset standard is determined as the expected functional safety language model.

[0022] Further, after determining the expected functional safety language model, the analysis method further includes:

[0023] Acquire new analysis data, perform data cleaning and data structuring on the new analysis data, and determine a new data set;

[0024] Incremental learning and fine-tuning are performed on the expected functional safety language model based on the newly added data set, and the updated expected functional safety language model is used as the expected functional safety language model.

[0025] In a second aspect, an embodiment of the present application further provides an analysis device based on an expected functional safety analysis system, the analysis device comprising:

[0026] An analysis process determination module is used to determine the analysis process of the current expected functional safety analysis task; wherein the analysis process includes multiple analysis steps, each analysis step represents a different analysis dimension;

[0027] An analysis result determination module is used to obtain information to be analyzed of the vehicle, and input the information to be analyzed into an expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle; wherein the expected functional safety analysis result includes analysis results for multiple analysis dimensions, and the expected functional safety analysis system includes a prompt word system and an expected functional safety language model, the prompt word system determines the corresponding prompt word information based on each analysis step, and the expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

[0028] Further, when the analysis result determination module is used to input the information to be analyzed into the expected functional safety analysis system so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle, the analysis result determination module is also used to:

[0029] For each analysis step in the analysis process, the input data corresponding to the analysis step is determined; wherein, when the analysis step has a previous analysis step connected in series with it, the input data of the analysis step is the analysis result output by the expected functional safety language model based on the prompt word information corresponding to the previous analysis step;

[0030] Inputting the input data into the prompt word system to determine the target prompt word information corresponding to the analysis step;

[0031] Inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step;

[0032] The target analysis results of each analysis step are spliced ​​in the order of each analysis step to obtain the expected functional safety analysis results of the vehicle.

[0033] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the analysis method based on the expected functional safety analysis system as described above are performed.

[0034] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the analysis method based on the expected functional safety analysis system as described above are executed.

[0035] An analysis method and an analysis device based on an expected functional safety analysis system are provided in an embodiment of the present application. First, an analysis process of a current expected functional safety analysis task is determined; wherein the analysis process includes multiple analysis steps, and each analysis step represents a different analysis dimension; then, information to be analyzed of the vehicle is obtained, and the information to be analyzed is input into the expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain an expected functional safety analysis result of the vehicle; wherein the expected functional safety analysis result includes analysis results for multiple analysis dimensions, and the expected functional safety analysis system includes a prompt word system and an expected functional safety language model, wherein the prompt word system determines corresponding prompt word information based on each analysis step, and the expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

[0036] This application decomposes the complex SOTIF analysis task into multiple steps to form a coherent analysis link. The prompt word system guides the language model to focus on the key aspects of system design through a series of preset prompt words. The prompt word system and the analysis process mechanism ensure the consistency and logic of the analysis process, can comprehensively identify and evaluate the potential safety hazards of the autonomous driving system, and can promptly discover and correct the deviations in the analysis process to ensure that the final analysis report has a high degree of consistency and reliability. The SOTIF analysis system based on LLM can automatically process and analyze a large amount of SOTIF case data and generate a detailed SOTIF analysis report. The system includes an expected functional safety language model (SOTIF AI) and a prompt word system, which can automatically and comprehensively identify and evaluate the expected functional safety issues of the autonomous driving system in a fault-free situation, generate a detailed SOTIF analysis report, greatly improve the efficiency and accuracy of the analysis, reduce human errors, and ensure the consistency and reliability of the analysis results.

[0037] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1A flowchart of an analysis method based on an expected functional safety analysis system provided in an embodiment of the present application;

[0040] Figure 2 An example diagram of an analysis process of an expected functional safety analysis task provided in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of the structure of an analysis device based on an expected functional safety analysis system provided in an embodiment of the present application;

[0042] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0044] First, the application scenarios to which the present application is applicable are introduced. The present application can be applied in the field of vehicle testing technology.

[0045] With the rapid development of autonomous driving technology, safety issues have gradually become the focus of the industry. In autonomous driving systems, in addition to traditional functional safety (Functional Safety), intended functional safety (Safety of the Intended Functionality, SOTIF) has also become particularly important. SOTIF mainly focuses on safety issues caused by insufficient system performance or changes in the external environment when the system is fault-free. These problems are particularly prominent in autonomous driving systems, because autonomous driving systems need to make fast and accurate decisions in complex and changing environments, and any slight functional deficiencies may lead to serious safety accidents. However, SOTIF analysis is more complex and cumbersome than traditional failure mode analysis, and requires in-depth analysis and evaluation of multiple aspects such as system functions and design operating conditions (ODC).

[0046] Research has found that, at present, most SOTIF analyses still rely on manual work. This method is not only time-consuming and labor-intensive, but also easily affected by personal experience and subjective bias, resulting in inconsistent and unreliable analysis results. Another problem with manual analysis is that it is difficult to scale and standardize. Different engineers have different analysis methods and standards, which is particularly prominent in large projects. Therefore, how to improve the efficiency and accuracy of SOTIF analysis has become an urgent problem to be solved.

[0047] Based on this, the embodiment of the present application provides an analysis method based on the expected functional safety analysis system, which can automatically and comprehensively identify and evaluate the expected functional safety issues of the autonomous driving system under fault-free conditions, and generate a detailed SOTIF analysis report, which greatly improves the efficiency and accuracy of the analysis, reduces human errors, and ensures the consistency and reliability of the analysis results.

[0048] See also Figure 1 , Figure 1 This is a flow chart of an analysis method based on an expected functional safety analysis system provided in an embodiment of the present application. Figure 1 As shown in, the analysis method provided in the embodiment of the present application includes:

[0049] S101, determine the analysis process of the current expected functional safety analysis task.

[0050] For the above step S101, in the specific implementation, the current expected functional safety analysis task is first initialized to determine the analysis process of the current expected functional safety analysis task. Specifically, the analysis process includes multiple analysis steps, and each analysis step represents a different analysis dimension. The direction and content filling of the SOTIF analysis task are implemented in the form of an analysis process. Each analysis step in the analysis process guides the expected functional safety analysis system to complete the analysis of a specific analysis dimension. The output of the previous analysis step of the analysis process serves as the input of the subsequent analysis step, so the analysis of each step needs to consider the context to ensure the coherence and logic of the entire analysis process.

[0051] According to the embodiments provided by the present application, the analysis dimensions may include potential harmful behaviors HB and hazards HZ, potential functional deficiencies FI, potential trigger conditions TC, functional modification FM strategies, verification and confirmation V&V strategies, hazard scenarios HS and SOTIF targets SG, so as to comprehensively identify and evaluate the potential safety hazards of the system. Here, the above-mentioned multiple analysis dimensions all refer to the key links and key contents in the SOTIF activities. Potential harmful behaviors HB refer to system behavior patterns that may cause safety risks, and hazards HZ refer to specific unsafe states that may be caused by these behaviors. Potential functional deficiencies FI involve situations where the system fails to fully realize the expected functions, which are caused by insufficient design specifications, performance limitations or other reasons. Potential trigger conditions TC refer to factors or events that may trigger potential harmful behaviors. Verification and confirmation V&V strategies are a series of methods and techniques used to prove the effectiveness of system design and changes to ensure that the modified system can achieve the expected safety level. Hazard scenarios HS refer to the process of describing in detail how potential harmful behaviors develop into actual hazards in combination with specific trigger conditions and background conditions. SOTIF target SG means that through a series of analysis and improvement measures, the safety of the autonomous driving system under the designed operating conditions is significantly improved, and potential hazards can be effectively prevented or mitigated even in the absence of faults. Functional modification FM strategy refers to the system improvement measures proposed to mitigate or eliminate identified risks and achieve the expected functional safety goals.

[0052] See also Figure 2 , Figure 2 This is an example diagram of the analysis process of an expected functional safety analysis task provided in an embodiment of the present application. Figure 2, Step 1: Hazardous behavior HB identification. This step uses the input of ODC and the autonomous driving functional architecture as prompt words, and generates hazardous behaviors based on the template of hazardous behavior HB. Hazardous behavior HB refers to unexpected behaviors with potential hazards performed by the ego vehicle, such as unexpected acceleration, wrong lane change direction, etc. It is related to the ODC and functional architecture of the ego vehicle. Step 2: Hazard HZ identification. This step uses the hazardous behavior HB and the above input as prompt words, and generates hazards based on the template of hazard HZ. Hazard HZ refers to the hazards caused by the ego vehicle in the scene due to the hazardous behavior HB performed by itself. Hazard HZ can be between the ego vehicle and other vehicles (too small distance), between the ego vehicle and the road boundary (running out of the road boundary), and between the road and other traffic participants (too small distance with pedestrians and animals). Step 3: Functional insufficiency FI and trigger condition TC identification. This step uses the input of the autonomous driving functional architecture as a prompt word to generate the functional insufficiency FI and trigger condition TC of the autonomous driving system. Functional insufficiency FI is directly associated with the input system functional architecture, and at the same time associates hazard HZ with hazardous behavior HB; trigger condition TC associates functional insufficiency FI. Functional insufficiency FI includes performance limitations (such as the camera's perception distance limitation, accuracy limitation, etc.) and insufficient specifications (such as insufficient definition of important targets in the perceived target recognition algorithm). FI directly points to a certain component or algorithm, which leads to deficiencies in the system including the component and algorithm, affecting the performance of the entire vehicle. Trigger condition TC refers to a triggering factor in the scenario that causes the functional insufficiency of the entire vehicle. TC acts on FI to cause the vehicle to behave in a hazardous manner, resulting in harm. Step 4: Verify and confirm the generation of V&V strategy. This step uses the insufficient autonomous driving function and trigger conditions, as well as the previous text as prompts to generate a V&V strategy based on the template of the verification and confirmation V&V strategy. The verification and confirmation V&V strategy is a verification strategy based on the test scenario for functional insufficiency. It is directly associated with the functional insufficiency FI, and through the association with FI, it can be associated with TC, HZ, HB, etc.; in the verification and confirmation V&V strategy, "Validation" refers to the confirmation of qualitative or quantitative goals for a certain strategy. In terms of process, when the hazard caused by a certain functional deficiency in the test scenario is verified, the hazard can be accepted (qualitative: expert experience, jury, etc. consider it acceptable; quantitative: the result of the hazard assessment is less than the acceptable threshold), it means that a verification and confirmation V&V strategy is completed. Step 5: Generate hazard scenario HS. This step uses the verification and confirmation V&V strategy and the previous text as prompts to generate hazard scenario test cases based on the scenario description template. The hazard scenario is a test carrier that verifies whether the hazard HZ caused by the hazardous behavior HB of the vehicle under test when a certain functional deficiency FI is triggered by the trigger condition TC of the scenario can be accepted by the confirmation target given in the verification and confirmation V&V strategy. Step 6: Generate the expected functional safety goal SG. This step uses the description of the hazard scenario and the previous text as prompts to determine the expected functional safety goal SG under each hazard scenario.The SOTIF target SG refers to the expected behavior that the vehicle should perform in the hazard scenario HS after the expected functional safety is achieved. Step 7: Functional modification FM strategy generation. This step uses the expected functional safety goal of the hazard scenario and the previous text as prompts to determine the functional modification FM strategy required to achieve the expected functional safety goal SG of each hazard scenario. The functional modification FM is associated with FI, HZ, SG, etc. Functional modifications can be modifications to rules, functions, definitions, etc., including one or more modification methods such as system modifications, functional restrictions, and authority transfer. After the functional modification FM is executed, the SG target and the V&V confirmation target should be achieved when testing again in the hazard scenario HS.

[0053] For the above step S101, the analysis process can also use a fully connected prompt network, and the above analysis steps are a relatively simplified series + local parallel structure. In theory, the SOTIF analysis link can be fully connected, so a fully connected prompt network can be designed, in which each prompt node can be directly connected to other nodes to form a large network structure. The fully connected prompt network can better capture the complex relationship between different analysis steps, improve the reasoning ability of the model and the comprehensiveness of the analysis. For example, the identification results of functional deficiencies FI can directly affect the identification of the expected functional safety goals SG, rather than just indirectly through intermediate steps. The fully connected prompt network has higher requirements for computing power and large model processing capabilities, but it can also achieve a more comprehensive and coupled SOTIF analysis effect, with the same or even better functions and capabilities.

[0054] S102, obtaining information to be analyzed of the vehicle, and inputting the information to be analyzed into an expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain an expected functional safety analysis result of the vehicle.

[0055] Here, the information to be analyzed of the vehicle may include the vehicle's autonomous driving functional architecture and design operating conditions (ODC). The functional specifications and architecture of the autonomous driving system refer to a series of tasks that the system is expected to perform and its internal structure, including but not limited to sensor configuration, decision-making algorithms, actuators, and the interaction between components. The specifications and architecture are intended to ensure that the system can operate safely and effectively according to the predetermined goals, while providing a basic framework for subsequent safety analysis. For example, the functional specifications may describe in detail how the vehicle makes decisions in specific situations, such as avoidance strategies when encountering pedestrians or obstacles; the architecture covers sensing technology, perception, decision-making system design, etc. Design operating conditions refer to the environmental parameters and constraints under which the autonomous driving system is expected to work, which are determined by the manufacturer. The design operating conditions include factors such as geographical area, road type, weather conditions, traffic density, and the internal state of the vehicle, which define the conditions under which the system can operate normally. The expected functional safety analysis results include analysis results for multiple analysis dimensions.

[0056] Regarding the above step S102, during the specific implementation, the information to be analyzed of the vehicle is obtained, and the information to be analyzed is input into the expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle.

[0057] Specifically, the expected functional safety analysis system consists of two parts, the prompt word system and the expected functional safety language model, namely SOTIF AI. The expected functional safety language model refers to a SOTIF large language model. After a large amount of expected functional safety analysis data training, the model has the ability to understand the functional specifications and architecture of the autonomous driving system, and can fill in the SOTIF analysis content of the entire process for the autonomous driving system. Here, the expected functional safety language model adopts the large language model (LLM, Large Language Model) technology, has powerful natural language processing and generation capabilities, and can realize intelligent applications in multiple fields. The advantage of the large language model is that it can understand and process complex text information, perform logical reasoning and knowledge expansion, which makes it have great potential in SOTIF analysis. The prompt word system refers to the calling tool of the expected functional safety language model, which is used to provide specific analysis task instructions to the expected functional safety language model to ensure that the model can perform accurate SOTIF analysis for a specific autonomous driving system. Specifically, the prompt word system determines the corresponding prompt word information based on each analysis step, and inputs the prompt word information into the expected functional safety language model. The expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

[0058] In this application, the expected functional safety analysis results can be presented in a variety of forms, such as table format, document format or other forms. Based on the efficient processing capabilities of the expected functional safety analysis system, the expected functional safety analysis results can be output in batches, with each batch focusing on a specific analysis dimension to ensure that each analysis step can receive in-depth and detailed review.

[0059] Here, this application realizes the automation of SOTIF analysis by introducing large language model (LLM) technology. Traditional manual analysis methods are not only time-consuming and labor-intensive, but also easily affected by personal experience and subjective bias, resulting in inconsistency and unreliability of analysis results. The intelligent analysis system based on the SOTIF language model can quickly process and analyze a large amount of SOTIF case data and generate a detailed SOTIF analysis report. This greatly improves the efficiency and accuracy of the analysis, reduces human errors, and ensures the consistency and reliability of the analysis results.

[0060] In this way, according to the above steps S101-S102, the present application decomposes the complex SOTIF analysis task into multiple steps to form a coherent analysis link. The prompt word system guides the language model to focus on the key aspects of system design through a series of preset prompt words. The prompt word system and the analysis process mechanism ensure the coherence and logic of the analysis process, can comprehensively identify and evaluate the potential safety hazards of the autonomous driving system, and can promptly discover and correct the deviations in the analysis process, ensuring that the final analysis report has a high degree of consistency and reliability. The SOTIF analysis system based on LLM can automatically process and analyze a large amount of SOTIF case data and generate a detailed SOTIF analysis report. It can automatically and comprehensively identify and evaluate the expected functional safety issues of the autonomous driving system in the absence of faults. The automation of SOTIF analysis is realized, and a structured and complete analysis report can be generated to provide support for the safety assessment of the autonomous driving system.

[0061] Specifically, with respect to the above step S102, the inputting of the information to be analyzed into the expected functional safety analysis system so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle includes:

[0062] Step 1021, for each analysis step in the analysis process, determine the input data corresponding to the analysis step.

[0063] For the above step 1021, in the specific implementation, for each analysis step in the analysis process, the input data corresponding to the analysis step is determined. Here, when the analysis step has a previous analysis step connected in series with it, the input data of the analysis step is the analysis result output by the expected functional safety language model based on the prompt word information corresponding to the previous analysis step. As an example, when the analysis step is to analyze the potential harmful behavior HB, such as Figure 2 As shown in , since there is no previous analysis step in series with this step, the input data corresponding to this analysis step is the design operating condition ODC in the data to be analyzed. When the analysis step is to analyze the potential hazard HZ, as shown in Figure 2 As shown in , this analysis step has a previous analysis step connected in series, namely, the analysis step of the potential hazardous behavior HB. At this time, the input data of this analysis step is the analysis result output by the expected functional safety language model based on the prompt word keyword corresponding to the analysis step of the potential hazardous behavior HB.

[0064] Step 1022: input the input data into the prompt word system to determine the target prompt word information corresponding to the analysis step.

[0065] Step 1023: input the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step.

[0066] For the above steps 1022 to 1023, in the specific implementation, after the input data corresponding to the analysis step is determined, the input data is input into the prompt word system to determine the target prompt word information corresponding to the analysis step. Then, the target prompt word information is input into the expected functional safety language model to determine the target analysis result corresponding to the analysis step.

[0067] As an optional embodiment, for the above step 1023, inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step includes:

[0068] Step 10231, determine the filling template corresponding to the analysis step.

[0069] For the above step 10231, during the specific implementation, determine the filling template corresponding to the analysis step. During the specific implementation, the expected functional safety language model needs to output text that conforms to the template rather than free text. The filling template refers to the filling template for the analysis content of each analysis step in the expected functional safety analysis, which includes two parts, one is the content template, and the other is the association template. Here, continuing the example in the above steps, the content template is specifically: Hazardous behavior HB and hazard HZ: list structure or tree structure, HB and HZ are mutually associated. Functional deficiency FI: list and or tree structure, FI is associated with the system functional architecture. Trigger condition TC: list and or tree structure, TC is associated with functional deficiency FI. Function modification FM: list and or tree structure, FM is associated with functional deficiency FI. Verification and confirmation V&V: list or tree structure, V&V is associated with functional deficiency FI. Hazard scenario HS: list or tree structure, the description template of the scenario can be a four-layer, six-layer or seven-layer scenario structure, HS is associated with FI, TC, HB, and HZ. The association template refers to the association between the above contents (i.e., table cells, nodes of tree structures). The association of analysis contents can be carried out and executed through system analysis models such as HARA (Hazard Analysis and Risk Assessment), STPA (Systems-Theoretic Process Analysis), FTA (Fault Tree Analysis) in the field of functional safety. Different degrees of template adaptation are required for different analysis models. For example, in SPTA, the control model is the system model of the object under test, and the key node UCA (Unsafe Control Action) corresponds to the hazardous behavior HB.

[0070] Step 10232: input the target prompt word information into the expected functional safety language model, obtain the original analysis result output by the expected functional safety language model, and fill the original analysis result into the filling template to obtain the target analysis result.

[0071] For the above step 10232, in the specific implementation, the target prompt word information corresponding to the analysis step is input into the expected functional safety language model, the original analysis result output by the expected functional safety language model is obtained, and the original analysis result is filled into the filling template to obtain the target analysis result. In this way, a structured filling template is designed as the framework of the analysis result, so that the output content conforms to the predetermined format, the output format of each analysis step is standardized, and the standardization and consistency of the analysis result are ensured.

[0072] Step 1024, concatenate the target analysis results of each analysis step in the order of each analysis step to obtain the expected functional safety analysis results of the vehicle.

[0073] For the above step 1024, in the specific implementation, after the target analysis result of each analysis step is determined, the target analysis result of each analysis step is spliced ​​in the order of each analysis step to obtain the expected functional safety analysis result of the vehicle. In this way, the expected functional safety language model is called multiple times to generate analysis result text fragments and integrate them to eventually form a complete text of the target length, and the context is aligned to ensure continuity. Since the basic model of the expected functional safety language model has a maximum output text length limit, but the context continuity is required in the SOTIF analysis process, a long text splicing method is needed to integrate the texts output multiple times. As an example, the pseudo code of the long text generation function is as follows:

[0074]

[0075]

[0076] The `FUNCTION GenerateLongText` is a functional module that generates long texts segment by segment. It generates and integrates text segments by repeatedly calling the SOTIF AI model to eventually form a complete text of the target length. The function is suitable for generating longer text content and uses overlapping contexts during the generation process to ensure the continuity of the generated text. `model` represents the text generation model called by the generation module, namely SOTIF AI, which is used to generate text segments of a specific length based on the input context. `initial_prompt` represents the initial input prompt of the generation module, which serves as the starting content for the first generation of text segments. `segment_length` represents the maximum length of each generated text segment, limiting the number of text characters generated by each generation call. `overlap_length` represents the overlapping length between generated segments, allowing part of the content of the current generated segment to be retained in the next generation call, thereby establishing text continuity between segments. In the logic of the `GenerateLongText` function module, first, `initial_prompt` is assigned to the `context` variable, which is used as the context information of the current generated segment. Then initialize the empty list `generated_text` to store the text segments of each generation call. The logical process of the `WHILE` loop is that, under the premise that the total length of `generated_text` has not reached the target length, the generation module calls `model` to generate a new text segment based on the `context` and `segment_length` parameters, and stores the generated result as `new_segment`. Subsequently, the `new_segment` is added to the `generated_text` list. When there are multiple text segments in the `generated_text` list, the `last_segment` is spliced ​​with the `new_segment` by taking out the `overlap_length` characters at the end of the previous segment and assigning them to `last_segment` to update the value of `context`, thereby ensuring the continuity between the generated text segments; if `generated_text` contains only a single segment, `new_segment` is directly assigned to `context`. After each loop ends, if the total length of all text segments in `generated_text` reaches or exceeds the target length, the loop is jumped out. Finally, the concatenation result of all text fragments is returned, that is, the complete generated text of the target length, and the expected functional safety analysis result is obtained.

[0077] In this way, according to the above steps 1021-1024, the complex SOTIF analysis task is decomposed into multiple steps, and the analysis results of each step are used as input for the subsequent steps, forming a coherent analysis chain. In this way, the multi-step analysis method makes the analysis of each step more focused and accurate, and can timely discover and correct deviations in the analysis process. Therefore, the expected functional safety analysis system can comprehensively identify and evaluate the potential safety hazards of the autonomous driving system, ensuring the comprehensiveness and depth of the analysis.

[0078] As an optional embodiment, after determining the target analysis result corresponding to the analysis step, the analysis method further includes:

[0079] When there are other analysis steps before this analysis step, the target analysis result corresponding to this analysis step is compared with the target analysis results corresponding to the other analysis steps. If there is a difference, the target analysis result corresponding to this analysis step is changed.

[0080] In the operation process of the expected functional safety analysis system, interactive adjustment is an important verification and correction mechanism to ensure the consistency and accuracy of the entire analysis process. Interactive adjustment refers to the interactive comparison of related contents. If there is a difference, the earlier output content is taken as the correct item and the later output is corrected. For the above steps, in the specific implementation, after obtaining the target analysis result corresponding to the analysis step, when there are other analysis steps before the analysis step, the target analysis result corresponding to the analysis step is compared with the target analysis results corresponding to the other analysis steps. If there is a difference, the target analysis result corresponding to the analysis step is changed. In this way, when the system is performing a multi-step SOTIF analysis, the output of each step will be interactively compared with the output of other related steps. For example, the output generated in the hazard behavior HB identification stage will be compared with the output of the subsequent hazard HZ identification, functional deficiency FI and trigger condition TC identification steps to check whether there are logical contradictions or information inconsistencies. If a difference is found during the comparison process, the earlier output content will be used as a benchmark to make necessary corrections to the later output.

[0081] In this way, the SOTIF analysis method of this application supports interactive checking of the analysis process, with each batch focusing on specific analysis content or system modules to ensure that each part can receive in-depth and detailed review. Interactive checking can effectively avoid the uncertainty of the output content of large models and ensure that the final analysis report has a high degree of accuracy, reliability and contextual consistency.

[0082] As an optional embodiment, according to the analysis method provided by the present application, the expected functional safety language model is trained through the following steps:

[0083] A: Obtain the expected functional safety analysis case data, and perform data cleaning and data structuring on the expected functional safety analysis case data to determine the training data set and the test data set.

[0084] Here, the expected functional safety analysis case data refers to a case set that contains the analysis results of the SOTIF activities for a certain autonomous driving system. A SOTIF analysis case usually consists of multiple documents. The document usually directly points to a certain analysis topic of SOTIF (such as functional deficiency analysis, trigger condition analysis). There is a certain correlation between the various documents through SOTIF activities. Thematic analysis and content relevance are the key to SOTIF analysis. The expected functional safety analysis case data should be sufficient to ensure coverage of the SOTIF full process analysis. At the same time, it should be diverse. It is necessary to process SOTIF analysis cases for multiple autonomous driving systems or functions (such as HWP, AVP, AEB and other functions) to construct basic training data.

[0085] Data cleaning refers to checking and processing the collected data, and it is necessary to verify the quality, completeness, consistency, and diversity of the data. Data structuring refers to the preprocessing of SOTIF analysis case data, and the collected data needs to be organized into a format that can be received by the large model tuning interface, such as the Prompt-Completion format. To this end, different contents in the case data need to be processed in different ways. Specifically, long and difficult sentences and large paragraphs are chunked, and long texts are reasonably segmented to ensure that the Prompt-Completion pairs can focus on a single clear SOTIF activity theme. In addition, terminology and proper nouns need to be uniformly processed to ensure that the model can correctly understand and apply these concepts. In the process of data collection and processing, NLP technology can also be introduced to automatically identify and extract key information, such as potential harmful behaviors HB, hazards HZ, and functional deficiencies FI.

[0086] The training dataset and the test dataset refer to two data subsets used to train and evaluate the performance of the expected functional safety language model, respectively. The training dataset refers to a specific domain dataset prepared for fine-tuning the LLM basic model. After selection and processing, these data can help the model better understand the expertise and analysis methods in the SOTIF field, and guide the language model to learn the relevant knowledge and skills of SOTIF analysis. The test dataset is used to objectively measure the performance of the language model on unseen data to ensure that the model has good generalization ability.

[0087] For the above step A, during the specific implementation, the training and test data are processed. First, the expected functional safety analysis case data is obtained, and the expected functional safety analysis case data is cleaned and structured to form the SOTIF training data set and the test data set.

[0088] B: Constructing an expected functional safety original language model based on the training data set.

[0089] For the above step B, in the specific implementation, the original language model of expected functional safety is constructed based on the training data set determined in the above step A. Here, the basic model is tuned by injecting the SOTIF training data set based on the Fine-tuning training method to obtain the original language model of expected functional safety focusing on SOTIF. Specifically, Fine-tuning tuning refers to further training on the pre-trained basic large model using the SOTIF training data set in a specific field, so that the model can better adapt to and understand the specific needs of the SOTIF field, enhance the performance of the model in SOTIF analysis tasks, improve the recognition accuracy of key elements such as potential functional deficiencies, hazardous behaviors, and trigger conditions, and improve the quality and professionalism of the analysis report generated by the model. In the process of Fine-tuning, the training effect is optimized by adjusting hyperparameters such as the learning rate and batch size to ensure that the model can obtain good generalization capabilities even with limited training data.

[0090] Thus, according to the above step B, this application uses a fine-tuning training method for SOTIF analysis, and fine-tunes the basic large model to enable it to better adapt to the specific needs of the SOTIF field. The method includes selecting a suitable LLM basic model, processing and structuring SOTIF analysis case data, forming a training data set and an evaluation set, and optimizing the training effect by adjusting hyperparameters such as learning rate and batch size. The fine-tuning training method is a key step to improve model performance. Through fine-tuning, the model can better understand and process the expertise in the SOTIF field, improve the recognition accuracy of key elements such as potential functional deficiencies, harmful behaviors, and trigger conditions, and ensure that the generated analysis report is highly professional and reliable.

[0091] In the above step B, in addition to the traditional fine-tuning method, transfer learning and multi-task learning techniques can be introduced. Transfer learning can transfer knowledge from pre-trained models in other related fields to improve the performance of the model on SOTIF analysis tasks. Multi-task learning allows the model to learn multiple related tasks at the same time, such as hazardous behavior identification, functional deficiency identification, etc., share the underlying representation, and improve the generalization ability and robustness of the model. This type of learning method can also enable large models to have the ability to learn and understand in specific knowledge fields.

[0092] C: Performing a performance evaluation on the expected functional safety original language model according to the test data set, and determining the expected functional safety original language model whose evaluation index reaches a preset standard as the expected functional safety language model.

[0093] For the above step C, in the specific implementation, the performance of the expected functional safety original language model is evaluated according to the test data set determined in the above step A, and the expected functional safety original language model whose evaluation indicators meet the preset standards is determined as the expected functional safety language model. Here, when conducting performance evaluation, a variety of indicators can be selected to comprehensively evaluate the performance of the model, including but not limited to traditional machine learning evaluation indicators such as accuracy, recall rate, F1 score, and customized indicators for SOTIF analysis characteristics, such as the completeness, logic and professionalism of the analysis report. In addition, an expert review panel needs to be introduced to review the SOTIF output report to verify whether the SOTIF large model has complete output capabilities. After tuning, the expected functional safety language model is obtained.

[0094] As an optional embodiment, after determining the expected functional safety language model, the analysis method further includes:

[0095] I: Acquire new analysis data, perform data cleaning and data structuring on the new analysis data, and determine a new data set.

[0096] For the above step I, in the specific implementation, when there is a new SOTIF analysis demand, or when new data is accumulated in actual application, first obtain the newly added analysis data, and pre-process the newly added analysis data, including data cleaning and data structuring, to ensure that it meets the requirements of model input, and obtain a new data set. Here, the method of data cleaning and data structuring is the same as that in the above step A, and can achieve the same technical effect, which will not be repeated here.

[0097] II: Incrementally learning and fine-tuning the expected functional safety language model based on the newly added data set, and using the updated expected functional safety language model as the expected functional safety language model.

[0098] Model iteration is a continuous optimization process. By constantly introducing new data and new requirements, the model can be kept up to date to adapt to the ever-changing application scenarios. For the above step B, in the specific implementation, after the new data set is determined, the existing model is incrementally learned using these new data. This process can strengthen the model's understanding and processing capabilities of new data on the basis of maintaining the original model knowledge. After incremental learning, the model also needs to be fine-tuned. The performance of the model can be further optimized by adjusting the learning rate, increasing the number of training rounds, etc. After fine-tuning, the model is comprehensively evaluated again to ensure that its performance on both new and old data meets the expected standards. Finally, the updated expected functional safety language model is used as the expected functional safety language model and redeployed to the expected functional safety analysis system, and its performance in actual applications is monitored, and further adjustments and optimizations are made when necessary.

[0099] Here, as an optional embodiment, during the deployment phase of the expected functional safety language model, it is necessary to comprehensively record the key information of the entire training process, including the training data set used, the initial parameter settings of the model, the hyperparameter adjustment during the fine-tuning process, and the performance evaluation results after each iteration. After completing the information recording, the final trained expected functional safety language model is exported and adapted according to the interface specifications of the expected functional safety analysis system to ensure that the model can be seamlessly integrated into the system. In addition, the operating environment of the model needs to be configured during the deployment process to ensure that it can run stably and efficiently in the production environment.

[0100] The analysis method based on the expected function safety analysis system provided in the embodiment of the present application first determines the analysis process of the current expected function safety analysis task; wherein the analysis process includes multiple analysis steps, and each analysis step represents a different analysis dimension; then, the vehicle's information to be analyzed is obtained, and the information to be analyzed is input into the expected function safety analysis system, so that the expected function safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected function safety analysis result of the vehicle; wherein the expected function safety analysis result includes analysis results for multiple analysis dimensions, and the expected function safety analysis system includes a prompt word system and an expected function safety language model, the prompt word system determines the corresponding prompt word information based on each analysis step, and the expected function safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

[0101] This application decomposes the complex SOTIF analysis task into multiple steps to form a coherent analysis link. The prompt word system guides the language model to focus on the key aspects of system design through a series of preset prompt words. The prompt word system and the analysis process mechanism ensure the consistency and logic of the analysis process, can comprehensively identify and evaluate the potential safety hazards of the autonomous driving system, and can promptly discover and correct the deviations in the analysis process to ensure that the final analysis report has a high degree of consistency and reliability. The SOTIF analysis system based on LLM can automatically process and analyze a large amount of SOTIF case data and generate a detailed SOTIF analysis report. The system includes an expected functional safety language model (SOTIF AI) and a prompt word system, which can automatically and comprehensively identify and evaluate the expected functional safety issues of the autonomous driving system in a fault-free situation, generate a detailed SOTIF analysis report, greatly improve the efficiency and accuracy of the analysis, reduce human errors, and ensure the consistency and reliability of the analysis results.

[0102] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of an analysis device based on an expected functional safety analysis system provided in an embodiment of the present application. Figure 3 As shown in , the analysis device 300 includes:

[0103] The analysis process determination module 301 is used to determine the analysis process of the current expected functional safety analysis task; wherein the analysis process includes multiple analysis steps, each analysis step represents a different analysis dimension;

[0104] The analysis result determination module 302 is used to obtain the vehicle's information to be analyzed and input the information to be analyzed into the expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle; wherein the expected functional safety analysis result includes analysis results for multiple analysis dimensions, and the expected functional safety analysis system includes a prompt word system and an expected functional safety language model, the prompt word system determines the corresponding prompt word information based on each analysis step, and the expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

[0105] Further, when the analysis result determination module 302 is used to input the information to be analyzed into the expected functional safety analysis system so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle, the analysis result determination module 302 is also used to:

[0106] For each analysis step in the analysis process, the input data corresponding to the analysis step is determined; wherein, when the analysis step has a previous analysis step connected in series with it, the input data of the analysis step is the analysis result output by the expected functional safety language model based on the prompt word information corresponding to the previous analysis step;

[0107] Inputting the input data into the prompt word system to determine the target prompt word information corresponding to the analysis step;

[0108] Inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step;

[0109] The target analysis results of each analysis step are spliced ​​in the order of each analysis step to obtain the expected functional safety analysis results of the vehicle.

[0110] Furthermore, when the analysis result determination module 302 is used to input the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step, the analysis result determination module 302 is also used to:

[0111] Determine a filling template corresponding to the analysis step; wherein the filling template includes a content template and an associated template;

[0112] The target prompt word information is input into the expected functional safety language model, an original analysis result output by the expected functional safety language model is obtained, and the original analysis result is filled into the filling template to obtain the target analysis result.

[0113] Furthermore, the analysis device 300 further includes a comparison and modification module. After determining the target analysis result corresponding to the analysis step, the comparison and modification module is used to:

[0114] When there are other analysis steps before this analysis step, the target analysis result corresponding to this analysis step is compared with the target analysis results corresponding to the other analysis steps. If there is a difference, the target analysis result corresponding to this analysis step is changed.

[0115] Furthermore, the analysis device 300 further includes a model training module, which is used to train the expected functional safety language model through the following steps:

[0116] Acquire expected functional safety analysis case data, and perform data cleaning and data structuring on the expected functional safety analysis case data to determine a training data set and a test data set;

[0117] Building an expected functional safety primitive language model based on the training data set;

[0118] A performance evaluation is performed on the expected functional safety original language model according to the test data set, and an expected functional safety original language model whose evaluation index reaches a preset standard is determined as the expected functional safety language model.

[0119] Furthermore, the analysis device 300 further includes a model incremental learning module. After determining the expected functional safety language model, the model incremental learning module is used to:

[0120] Acquire new analysis data, perform data cleaning and data structuring on the new analysis data, and determine a new data set;

[0121] Incremental learning and fine-tuning are performed on the expected functional safety language model based on the newly added data set, and the updated expected functional safety language model is used as the expected functional safety language model.

[0122] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown in , the electronic device 400 includes a processor 410 , a memory 420 and a bus 430 .

[0123] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, the above-mentioned Figure 1 The steps of the analysis method based on the expected functional safety analysis system in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0124] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The steps of the analysis method based on the expected functional safety analysis system in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0125] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0126] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0127] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0129] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application can essentially be embodied in the form of a software product, or in other words, the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0130] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An analysis method based on an expected functional safety analysis system, characterized in that: The analysis method comprises: Determine the analysis process of the current expected functional safety analysis task; wherein the analysis process includes multiple analysis steps, each analysis step represents a different analysis dimension; Acquire information to be analyzed of the vehicle, and input the information to be analyzed into an expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain an expected functional safety analysis result of the vehicle; wherein the expected functional safety analysis result includes analysis results for multiple analysis dimensions, and the expected functional safety analysis system includes a prompt word system and an expected functional safety language model, the prompt word system determines the corresponding prompt word information based on each analysis step, and the expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

2. The analysis method according to claim 1, characterized in that The step of inputting the information to be analyzed into the expected functional safety analysis system so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle includes: For each analysis step in the analysis process, the input data corresponding to the analysis step is determined; wherein, when the analysis step has a previous analysis step connected in series with it, the input data of the analysis step is the analysis result output by the expected functional safety language model based on the prompt word information corresponding to the previous analysis step; Inputting the input data into the prompt word system to determine the target prompt word information corresponding to the analysis step; Inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step; The target analysis results of each analysis step are spliced ​​in the order of each analysis step to obtain the expected functional safety analysis results of the vehicle.

3. The analysis method according to claim 2, characterized in that The step of inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step includes: Determine a filling template corresponding to the analysis step; wherein the filling template includes a content template and an associated template; The target prompt word information is input into the expected functional safety language model, an original analysis result output by the expected functional safety language model is obtained, and the original analysis result is filled into the filling template to obtain the target analysis result.

4. The analysis method according to claim 2, characterized in that After determining the target analysis result corresponding to the analysis step, the analysis method further includes: When there are other analysis steps before this analysis step, the target analysis result corresponding to this analysis step is compared with the target analysis results corresponding to the other analysis steps. If there is a difference, the target analysis result corresponding to this analysis step is changed.

5. The analysis method according to claim 1, characterized in that The expected functional safety language model is trained by the following steps: Acquire expected functional safety analysis case data, and perform data cleaning and data structuring on the expected functional safety analysis case data to determine a training data set and a test data set; Building an expected functional safety primitive language model based on the training data set; A performance evaluation is performed on the expected functional safety original language model according to the test data set, and an expected functional safety original language model whose evaluation index reaches a preset standard is determined as the expected functional safety language model.

6. The analysis method according to claim 5, characterized in that After determining the expected functional safety language model, the analysis method further includes: Acquire new analysis data, perform data cleaning and data structuring on the new analysis data, and determine a new data set; Incremental learning and fine-tuning are performed on the expected functional safety language model based on the newly added data set, and the updated expected functional safety language model is used as the expected functional safety language model.

7. An analysis device based on an expected functional safety analysis system, characterized in that: The analysis device comprises: An analysis process determination module is used to determine the analysis process of the current expected functional safety analysis task; wherein the analysis process includes multiple analysis steps, each analysis step represents a different analysis dimension; An analysis result determination module is used to obtain information to be analyzed of the vehicle, and input the information to be analyzed into an expected functional safety analysis system, so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle; wherein the expected functional safety analysis result includes analysis results for multiple analysis dimensions, and the expected functional safety analysis system includes a prompt word system and an expected functional safety language model, the prompt word system determines the corresponding prompt word information based on each analysis step, and the expected functional safety language model determines the analysis result corresponding to the prompt word information based on the prompt word information.

8. The analysis device according to claim 7, characterized in that When the analysis result determination module is used to input the information to be analyzed into the expected functional safety analysis system so that the expected functional safety analysis system analyzes the information to be analyzed based on the analysis process to obtain the expected functional safety analysis result of the vehicle, the analysis result determination module is further used to: For each analysis step in the analysis process, the input data corresponding to the analysis step is determined; wherein, when the analysis step has a previous analysis step connected in series with it, the input data of the analysis step is the analysis result output by the expected functional safety language model based on the prompt word information corresponding to the previous analysis step; Inputting the input data into the prompt word system to determine the target prompt word information corresponding to the analysis step; Inputting the target prompt word information into the expected functional safety language model to determine the target analysis result corresponding to the analysis step; The target analysis results of each analysis step are spliced ​​in the order of each analysis step to obtain the expected functional safety analysis results of the vehicle.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the analysis method based on the expected functional safety analysis system as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the analysis method based on the expected functional safety analysis system according to any one of claims 1 to 6 are executed.

Images