Automobile V process software development workflow arrangement method, system, medium and equipment

Abstract

The invention discloses an automobile V process software development workflow arrangement method and system, a medium and equipment. The method comprises the following steps: acquiring a root node of automobile V workflow arrangement; performing workflow node search and generation according to the root node cycle based on a Monte Carlo tree search algorithm; evaluating the target workflow generated in each cycle based on a multi-stage evaluation method; if the evaluation is passed, taking the target workflow generated by the current cycle as a final workflow; and if the assessment is not passed, reversely propagating the assessment result to all nodes of the workflow and carrying out the next cycle until convergence is completed, and generating a final workflow. Based on the data processing flow, the thought of Monte Carlo tree search is introduced into the software V development flow, so that the problems encountered by traditional SOP workflow and existing automatic workflow arrangement are solved.

Description

Technical Field

[0001] This invention relates to the field of automotive electronic software development technology, and in particular to a method, system, medium, and device for orchestrating automotive V-process software development workflow. Background Technology

[0002] Currently, workflow implementation is mainly divided into two types: customized workflows (SOPs) and automatically generated and orchestrated workflows. In the development of intelligent automotive electronic software, the end-to-end toolchain collaboration of the V-process (requirements analysis → system design → software implementation → unit testing → integration testing → system verification) faces the following technical bottlenecks: Traditional industrial workflow methods, such as those used in Jenkins and Jira engines, rely on experts defining specific workflows for each scenario. The advantage of SOPs is that they ensure stable operation and high reliability for repetitive processes. However, SOPs cannot be automatically updated to meet new business requirements or changes, requiring experts to rewrite and review them. Adding a new functional domain to a customized workflow necessitates the development of multiple customized workflows by experts, resulting in significant time costs and a high degree of specialization.

[0003] The current methods for building automated workflows mainly consist of the following: 1. It is constructed through prompt words, relying entirely on the capabilities of the large language model itself.

[0004] This method has high versatility, but it is not suitable for task orchestration in V-software workflow production. Firstly, large models naturally have lower prior knowledge of V-software workflows compared to generalized tasks. Each step in a V-software workflow has clearly defined upstream and downstream tasks, resulting in high dependencies and specialized outputs that cannot be pre-generated by a large model. Secondly, V-software workflows are typically complex and cumbersome, making it difficult for large language models to directly orchestrate a complete and reliable workflow.

[0005] 2. In academia, a common approach is to modularize some parts of the workflow and then select the corresponding modules to splice together using a large model – hereinafter referred to as AFLOW. See: Jiayi Zhang, Jinyu Xiang, Zhaoyang Yu, et al. "AFlow: Automating Agentic Workflow Generation." arXiv preprint arXiv:2410.10762v4. 2025.

[0006] Specifically, AFLOW combines the Monte Carlo tree search method with the concept of operators. AFLOW treats each generated work step in the workflow as a node in a tree, while the user's input is the root node of the workflow. In addition, AFLOW integrates seven common work sets called operators. These operators include the following: (1) Generate, (2) Format, (3) Review, (4) Integrate, (5) Test, (6) Program, and (7) Custom. For the current workflow and user query, AFLOW first selects nodes, then selects operators from the existing seven operators by calling the large model to splice the workflow into the next node, evaluates the newly generated workflow, and then performs backpropagation to influence the next node selection.

[0007] It is important to note that when selecting nodes in AFLOW, you can select any node (such as an empty node) to start generating a new workflow. The advantage of doing this is that it avoids the negative impact of generating incorrect workflows in the previous steps on the final result.

[0008] This method solves the problem of task orchestration accuracy to some extent, but there are still at least four points that are not applicable to task orchestration in automotive V software processes: A. Although AFLOW defines custom workflows for extension, the proportion of custom workflows in the workflow orchestration generated by AFLOW is very small; most work steps are composed of the first six operators. For automotive V software processes, up to 70 custom workflows are required. At this point, the operator search space and orchestration complexity increase significantly, making it difficult to achieve effective orchestration directly. B. For a workflow, AFLOW needs to actually execute the current work step to simulate it and get the current workflow feedback to update it: Each single step of the workflow in the V software process may take several minutes or more to execute, and the overall time cost is too high. C. This method abstracts the workflow step by step into some general methods, such as voting, automatic correction, self-reflection and other functions. These functions are difficult to distinguish literally in the actual V software process development. For example, Checksheet IV review and Checklist self-check are difficult to distinguish and arrange directly. D. AFLOW's workflow automation approach is primarily suitable for divergent scenarios, including but not limited to code generation, resource searching, web page operations, and question answering. These divergent scenarios are characterized by clear task objectives and relatively short overall workflow steps; therefore, six general operators and one additional custom operator can solve most problems. For V-type software workflows, the overall workflow is longer, and the repetition and similarity of each task step are relatively low. Generalization methods for general tasks have high repetition between steps in their automatically orchestrated workflows, making them unsuitable for V-type software workflow scenarios.

[0009] In summary, the above-mentioned popular automated workflow generation solutions have some drawbacks, such as the simplicity of workflow generation, low dependency on workflow sequence, and high dependence of workflow evaluation on real-time execution results.

[0010] Therefore, this invention provides a Monte Carlo tree search-based workflow orchestration scheme for automotive V-process software development, introducing the concept of Monte Carlo tree search into the software V-process to solve the aforementioned problems encountered in traditional SOP workflows and current automated workflow orchestration. Summary of the Invention

[0011] This invention provides a method, system, medium, and device for orchestrating automotive V-process software development workflows. It introduces the concept of Monte Carlo tree search into the software V-development process to solve the problems encountered in traditional SOP workflows and current automated workflow orchestration. Firstly, a workflow orchestration method for automotive V-process software development is provided, including: Obtain the root node of the automotive V workflow orchestration; Based on the Monte Carlo tree search algorithm, workflow nodes are searched and generated cyclically according to the root node; The target workflow generated in each iteration is evaluated based on a multi-level evaluation method; If the evaluation passes, the target workflow generated in the current cycle will be used as the final workflow. If the evaluation fails, the evaluation result is backpropagated to all nodes in the workflow and the next loop is performed until convergence is achieved, at which point the final workflow is generated.

[0012] In some embodiments, the Monte Carlo tree search algorithm performs workflow node search and generation cyclically based on the root node, including: Starting from the root node of the automotive V workflow orchestration, a leaf node is located along the search tree based on the upper confidence boundary algorithm. An unvisited leaf node is selected from the located child nodes, and the located leaf node is added to the search tree to obtain a workflow generated in one loop.

[0013] In some embodiments, the upper confidence boundary algorithm is shown in the following formula:

[0014] In the formula, Q represents the cumulative reward; N represents the number of visits; is a hyperparameter; S is the current workflow that has been expanded and generated; A is the current leaf node that has been located; Relevant(S,A) is the matching degree between the current workflow that has been expanded and generated and the most similar workflow in the preset standard workflow knowledge base; UCB is the upper confidence boundary value.

[0015] In some embodiments, the evaluation of the target workflow generated in each iteration based on the multi-level evaluation method includes: For each target workflow generated in a cycle, evaluate whether the target workflow meets the preset target standard indicators; If the preset target standard indicators are not met, the workflow will be generated in the next cycle. If the target standard indicators are met, the target workflow is scored based on the LLM debate method, and it is determined whether the score result is greater than or equal to the score threshold. If the score is less than the scoring threshold, the workflow will be generated in the next cycle. If the score is greater than or equal to the scoring threshold, then it is determined whether the target workflow meets the preset scenario requirements. If the preset scenario requirements are not met, the workflow will be generated in the next cycle. If the preset scenario requirements are met, the target workflow generated in the current cycle is deemed to have passed the evaluation.

[0016] In some embodiments, the scoring of the target workflow based on the LLM debate method includes: Based on the first major model, the correctness of the target workflow and its consistency with the expected goal are positively demonstrated, and positive demonstration results are generated. The potential inaccuracies of the target workflow are backtested based on the second major model, generating backtesting results. The positive and negative arguments are evaluated based on the third model, generating a quantitative score between 0 and 1.

[0017] In some embodiments, the preset target standard indicators include ASPICE standard and ISO 26262 standard indicators.

[0018] In some embodiments, generating the final workflow upon proceeding to the next iteration until convergence is complete includes: Get the current loop count; If the current number of iterations exceeds the threshold, then the next iteration will be stopped, and the workflow generated in the current iteration will be used as the final workflow. If the current number of iterations has not exceeded the threshold, then proceed to the next iteration.

[0019] Secondly, a workflow orchestration system for automotive V-process software development is provided, including: The root node acquisition module is used to acquire the root node of the automotive V workflow orchestration. The node search module is communicatively connected to the root node acquisition module and is used to search for and generate workflow nodes cyclically based on the root node using the Monte Carlo tree search algorithm. The evaluation module, which is communicatively connected to the node search module, is used to evaluate the target workflow generated in each cycle based on a multi-level evaluation method. An evaluation module, communicatively connected to the evaluation module, is used to, if the evaluation passes, take the target workflow generated in the current cycle as the final workflow; and, The evaluation failure module is connected in communication with the evaluation module. If the evaluation fails, it backpropagates the evaluation result to all nodes in the workflow and performs the next loop until convergence is achieved, thus generating the final workflow.

[0020] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the automotive V-process software development workflow orchestration method as described above.

[0021] Fourthly, embodiments of the present invention provide an electronic device, including a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, wherein the processor executes the computer program to implement the automotive V-process software development workflow orchestration method as described above.

[0022] Compared with existing technologies, the advantages of this invention are as follows: It obtains the root node of the automotive V workflow orchestration; it uses a Monte Carlo tree search algorithm to iteratively search and generate workflow nodes based on the root node; it evaluates the target workflow generated in each iteration using a multi-level evaluation method; if the evaluation passes, the target workflow generated in the current iteration is taken as the final workflow; if the evaluation fails, the evaluation result is backpropagated to all nodes of the workflow and the next iteration is performed until convergence is achieved, generating the final workflow. Based on the above data processing flow, this invention introduces the idea of ​​Monte Carlo tree search into the software V development process to solve the problems encountered in traditional SOP workflows and current automated workflow orchestration. Attached Figure Description

[0023] Figure 1 This is a schematic flowchart of an embodiment of the automotive V-process software development workflow orchestration method of the present invention; Figure 2 This is a schematic diagram of another embodiment of the automotive V-process software development workflow orchestration method of the present invention; Figure 3 This is a schematic diagram of the structure of an automotive V-process software development workflow orchestration system according to the present invention. Detailed Implementation

[0024] Referring now to specific embodiments of the invention, examples of which are illustrated in the accompanying drawings. Although the invention will be described in conjunction with specific embodiments, it will be understood that it is not intended to limit the invention to the described embodiments. Rather, it is intended to cover variations, modifications, and equivalents included within the spirit and scope of the invention as defined by the appended claims. It should be noted that the method steps described herein can be implemented by any functional block or functional arrangement, and any functional block or functional arrangement can be implemented as a physical entity or a logical entity, or a combination of both.

[0025] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Note: The examples described below are merely specific examples and are not intended to limit the embodiments of the present invention to the specific steps, values, conditions, data, order, etc. Those skilled in the art can utilize the concepts of the present invention to construct more embodiments not mentioned herein by reading this specification.

[0027] Please see Figure 1 and 2 As shown in the diagram, this invention provides a flowchart of a workflow orchestration method for automotive V-process software development. The method includes: Step S100: Obtain the root node of the automotive V workflow orchestration; Step S200: Based on the Monte Carlo tree search algorithm, the workflow node is searched and generated cyclically according to the root node; Specifically, in this embodiment of the invention, workflow refers to the process of doing work, which naturally conforms to the tree structure in the Monte Carlo Tree Search (MCTS) algorithm. The corresponding relationship is that the user's input is the root node, and the edges connecting trees represent the parent-child dependencies, corresponding to the order of events in the workflow. Furthermore, the Monte Carlo tree evaluates the current state in real time and feeds it back to previous nodes when generating the workflow, thus perfectly suited for the task of automatically generating workflows. For the traditional Monte Carlo Tree Search algorithm, the algorithm steps are as follows: 1. Node Selection Phase: Based on the Upper Confidence Bound (UCB) algorithm, select the node to be expanded from the end nodes of the current workflow branch. For parallel development tasks, multi-objective optimization is performed on the end nodes of each branch workflow; for linear development tasks, a depth-first strategy is used to select the latest generated node.

[0028] 2. Node Extension Phase: Leveraging the V-process domain knowledge base, new work step nodes conforming to ASPICE standards are generated based on the current workflow status. A real-time compliance check mechanism is introduced during the extension process to ensure that the new nodes meet ISO 26262 functional safety requirements.

[0029] 3. Simulation and Evaluation Phase: Establish a multi-dimensional evaluation indicator system to assess the status of some generated workflows. A weighted scoring method is used to comprehensively consider development efficiency, resource utilization, and compliance indicators. Real-time evaluation is performed immediately after each new work step is generated.

[0030] 4. Backpropagation Phase: Design a value backtracking mechanism to propagate the execution effect of the current work step back along the path to previous nodes. Update the node value function using the Bellman equation to achieve distributed evaluation of the workflow execution effect.

[0031] As can be seen from the traditional process described above, the Monte Carlo tree search process closely matches the automatic workflow generation process. First, the current work step is determined; then, new work steps are constructed and generated; the current workflow result is evaluated; and finally, the evaluation result is passed to each previous work step to complete a loop. When the workflow evaluation result meets expectations, or the loop reaches the expected number of iterations, the process terminates and a new workflow is generated.

[0032] The application of the Monte Carlo tree search algorithm in automated workflow construction is mentioned in the AFLOW paper, but the constraints mentioned earlier prevent its direct transfer to V software workflows. Therefore, this invention improves the Monte Carlo tree search algorithm so that it can be transferred to V software workflows.

[0033] The Monte Carlo tree search algorithm performs workflow node search and generation iteratively based on the root node, including: Starting from the root node of the automotive V workflow orchestration, a leaf node is located along the search tree based on the upper confidence boundary algorithm. An unvisited leaf node is selected from the located child nodes, and the located leaf node is added to the search tree to obtain a workflow generated in one loop.

[0034] The upper confidence boundary algorithm is shown in the following formula:

[0035] In the formula, for each generated leaf node, this algorithm is used to traverse and calculate and select the leaf node; Q is the cumulative reward; N is the number of visits; is a hyperparameter; S is the expanded and generated current workflow; A is the located current leaf node; Relevant(S,A) is the matching degree between the expanded and generated current workflow and the most similar workflow in the preset standard workflow knowledge base. If a similar expert workflow can be matched, the score here will be improved, specifically the cosine similarity of the most similar workflow; UCB is the upper confidence boundary value.

[0036] For a given leaf node, the most similar workflow content retrieved from the preset standard workflow knowledge base is used as a prompt word to be input into the following large model to generate text, allowing the large model to evaluate the workflow status of each cycle.

[0037] It should be noted that the preset standard workflow knowledge base is a workflow knowledge base compiled by business experts based on user requirements or scenarios.

[0038] Step S300: Evaluate the target workflow generated in each iteration based on a multi-level evaluation method, including: S310, For each target workflow generated in each cycle, evaluate whether the target workflow meets the preset target standard indicators; the preset target standard indicators include ASPICE standard and ISO 26262 standard indicators.

[0039] The large model should evaluate whether the currently generated target workflow conforms to the ASPICE standard (Automotive Software Process Improvement and Capability Determination, a model and assessment method used in the automotive industry to evaluate and improve R&D process capabilities) and the ISO 26262 standard indicators. Since both standards must be met in the automotive V software process, if the large model determines that the current workflow does not meet the standards, it should directly return a score of 0.

[0040] S320, if the preset target standard indicators are not met, the workflow will be generated in the next cycle; S330, if the preset target standard indicators are met, the target workflow is scored based on the LLM debate (Large Language Model Debate) method. The scoring of the target workflow based on the LLM debate method includes: Based on the first major model, the correctness of the target workflow and its consistency with the expected goal are positively demonstrated, and positive demonstration results are generated. The potential inaccuracies of the target workflow are backtested based on the second major model, generating backtesting results. The positive and negative arguments are evaluated based on the third model, generating a quantitative score between 0 and 1. A score of 0 indicates that the argument is completely infeasible, while a score of 1 indicates that it is completely feasible.

[0041] And determine whether the scoring result is greater than or equal to the scoring threshold; If the score is less than the scoring threshold, the workflow will be generated in the next cycle. Therefore, step S330 is only performed after step S310 has passed, and the scoring result of S330 is directly used as the score of the currently generated target workflow.

[0042] S340, if the score threshold is greater than or equal to the target workflow, then determine whether the target workflow meets the preset scenario requirements; If the preset scenario requirements are not met, the workflow will be generated in the next cycle. If the preset scenario requirements are met, the target workflow generated in the current cycle is deemed to have passed the evaluation.

[0043] Step S400: If the evaluation passes, the target workflow generated in the current cycle is taken as the final workflow.

[0044] Specifically, the large model is retrieved to determine whether the current target workflow meets the user's scenario requirements. If the current workflow score and workflow already meet the user's expected scenario, the process is stopped in advance, and the currently generated target workflow is taken as the final workflow.

[0045] In step S500, if the evaluation fails, the evaluation result is backpropagated to all nodes of the workflow and the next loop is performed until convergence is achieved, at which point the final workflow is generated.

[0046] Specifically, this step involves passing the impact of previous steps on the current workflow to each previous node, updating the current score to the previous nodes, and thus affecting the next evaluation.

[0047] Simultaneously, after backpropagation, the current loop count is checked. If the loop count exceeds the expected number, the task is stopped and the final workflow is output to prevent excessive resource waste. The specific steps are as follows: When proceeding to the next iteration until convergence is achieved, the final workflow is generated, including: Get the current loop count; If the current number of iterations exceeds the threshold, then the next iteration will be stopped, and the workflow generated in the current iteration will be used as the final workflow. If the current number of iterations has not exceeded the threshold, then proceed to the next iteration.

[0048] In summary, the core innovation of this invention lies in the deep integration of the improved MCTS algorithm with the software V-process expert knowledge base, specifically reflected in: Domain knowledge-guided search strategy: Construct a pre-defined standard workflow database confirmed by experts to guide the decision-making process and node expansion stages of the algorithm.

[0049] Domain constraint handling mechanism: Real-time compliance verification is introduced during the node expansion phase to ensure that the generated workflow complies with automotive electronics development specifications and meets domain standards such as ASPICE process domain constraints and ISO 26262 safety requirements.

[0050] For real-time workflows, use LLM debate for evaluation instead of actual workflow execution.

[0051] See also Figure 3 As shown, this embodiment of the invention also provides an automotive V-process software development workflow orchestration system, including: The root node acquisition module is used to acquire the root node of the automotive V workflow orchestration. The node search module is communicatively connected to the root node acquisition module and is used to search for and generate workflow nodes cyclically based on the root node using the Monte Carlo tree search algorithm. The evaluation module, which is communicatively connected to the node search module, is used to evaluate the target workflow generated in each cycle based on a multi-level evaluation method. An evaluation module, communicatively connected to the evaluation module, is used to, if the evaluation passes, take the target workflow generated in the current cycle as the final workflow; and, The evaluation failure module is connected in communication with the evaluation module. If the evaluation fails, it backpropagates the evaluation result to all nodes in the workflow and performs the next loop until convergence is achieved, thus generating the final workflow.

[0052] In summary, the main advantages of this invention are as follows: First, compared to the traditional SOP method, it solves the problem of automated generation and orchestration of software V-process workflows.

[0053] Secondly, common tree-structure search methods for generating workflows are not applicable to specific scenarios in software V-process operations. This patent combines a pre-set standard workflow knowledge base to improve the accuracy and reliability of workflow generation.

[0054] Third, the large-scale model debate method replaces the original method of evaluating workflows, which required actual workflow execution and was too time-consuming.

[0055] Fourth, by combining the ISO26262 standard and ASPICE metrics, the safety and professionalism of the overall workflow are ensured.

[0056] Specifically, this embodiment corresponds one-to-one with the above method embodiments. The functions of each module have been described in detail in the corresponding method embodiments, so they will not be repeated here.

[0057] Based on the same inventive concept, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements all or part of the method steps of the above method.

[0058] The present invention can implement all or part of the processes in the above methods, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0059] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a memory and a processor. The memory stores a computer program that runs on the processor. When the processor executes the computer program, it implements all or part of the method steps described above.

[0060] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the computer device, connecting all parts of the computer device through various interfaces and lines.

[0061] Memory can be used to store computer programs and / or modules. The processor performs various functions of the computer device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system and at least one application program required for a function (e.g., sound playback, image playback, etc.); the data storage area can store data created based on the use of the mobile phone (e.g., audio data, video data, etc.). Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMedia Cards (SMC), Secure Digital (SD) cards, Flash Cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0062] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0063] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for workflow orchestration of automotive V-process software development, characterized in that, The method comprises the following steps: acquiring a root node of a vehicle V workflow arrangement; performing workflow node search and generation based on a Monte Carlo tree search algorithm according to the root node in a loop; evaluating a target workflow generated in each loop based on a multi-level evaluation method; if the evaluation is passed, taking the target workflow generated in the current loop as a final workflow; if the evaluation is not passed, performing reverse propagation of the evaluation result to all nodes of the workflow and performing the next loop until convergence is completed to generate the final workflow.

2. The automotive V-process software development workflow orchestration method of claim 1, wherein, The performing of the workflow node search and generation based on the Monte Carlo tree search algorithm according to the root node in the loop comprises the following steps: starting from the root node of the vehicle V workflow arrangement, locating a leaf node along a search tree based on an upper confidence bound algorithm, selecting an unvisited leaf node from the located sub-nodes, and adding the located leaf node to the search tree to obtain a workflow generated in one loop.

3. The automotive V-flow software development workflow orchestration method of claim 2, wherein, The upper confidence bound algorithm is as shown in the following formula: In the formula, Q is a cumulative reward; N is a number of visits; is a hyperparameter; S is an expanded generated current workflow; A is a located current leaf node; Relevant(S,A) is a matching degree between the expanded generated current workflow and a most similar workflow in a preset standard workflow knowledge base; and UCB is an upper confidence bound value.

4. The automotive V-flow software development workflow orchestration method of claim 1, wherein, The evaluation of the target workflow generated in each loop based on the multi-level evaluation method comprises the following steps: for the target workflow generated in each loop, evaluating whether the target workflow meets preset target standard indicators; if the preset target standard indicators are not met, performing the next loop generation of the workflow; if the preset target standard indicators are met, scoring the target workflow based on an LLM debate method, and determining whether the score result is greater than or equal to a score threshold; if the score result is less than the score threshold, performing the next loop generation of the workflow; if the score result is greater than or equal to the score threshold, determining whether the target workflow meets preset scene requirements; if the preset scene requirements are not met, performing the next loop generation of the workflow; if the preset scene requirements are met, determining that the evaluation of the target workflow generated in the current loop is passed.

5. The automotive V-flow software development workflow orchestration method of claim 4, wherein, The scoring of the target workflow based on the LLM debate method comprises the following steps: positively arguing the correctness and consistency of the target workflow with an expected target based on a first large model to generate a positive argument result; negatively arguing the possible incorrectness of the target workflow based on a second large model to generate a negative argument result; judging the positive argument result and the negative argument result based on a third large model to generate a quantitative score between 0 and 1.

6. The automotive V-flow software development workflow orchestration method of claim 4, wherein, The preset target standard indicators comprise ASPICE standard and ISO 26262 standard indicators.

7. The automotive V-flow software development workflow orchestration method of claim 1, wherein, The performing of the next loop until convergence is completed to generate the final workflow comprises the following steps: acquiring a current loop number; if the current loop number exceeds a number threshold, stopping the next loop and taking the workflow generated in the current loop as the final workflow; if the current loop number does not exceed the number threshold, performing the next loop.

8. An automotive V-process software development workflow orchestration system, characterized by, The method comprises the following steps: a root node acquisition module is configured to acquire a root node of a vehicle V workflow arrangement; a node search module is in communication connection with the root node acquisition module and is configured to perform workflow node search and generation based on a Monte Carlo tree search algorithm according to the root node in a loop; An evaluation module, in communication connection with the node search module, is configured to evaluate the target workflow generated in each cycle based on a multi-level evaluation method; An evaluation passing module, in communication connection with the evaluation module, is configured to, if the evaluation passes, take the target workflow generated in the current cycle as the final workflow; and, An evaluation failing module, in communication connection with the evaluation module, is configured to, if the evaluation fails, propagate the evaluation result to all nodes of the workflow and perform the next cycle until convergence is completed to generate the final workflow.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the automobile V-process software development workflow orchestration method according to any one of claims 1 to 7.

10. An electronic device comprising a storage medium, a processor, and a computer program stored in the storage medium and operable on the processor, characterized in that, The processor, when running the computer program, implements the automobile V-process software development workflow orchestration method according to any one of claims 1 to 7.

Images