CN-121981491-A - Generating type AI operation decision system based on organization strategy and vehicle context

CN121981491ACN 121981491 ACN121981491 ACN 121981491ACN-121981491-A

Abstract

The application discloses a generating type AI operation decision system based on an organization strategy and a vehicle context, which integrates a multi-source situation data access module, a strategy knowledge graph construction module, a multi-mode situation fusion module, a generating type AI decision module, a digital twin simulation verification module, a strategy compliance arbitration module, a decision optimization and interpretation generation module and a decision execution and monitoring module into a whole, dynamically generates candidate decisions by structuring the organization strategy and deeply fusing the organization strategy with real-time data, and the feasibility and risk are predicted through high-fidelity simulation, the decision is ensured to be highly consistent with the rule through automatic compliance verification, and finally the reliable decision which is optimized through multiple targets and interpreted by the attached natural language is output, so that the defects of unreasonable problem, lack of constraint and unpredictable output of the generated AI applied to the vehicle operation are overcome, and the intelligent operation decision which is safe, reliable, controllable in compliance and stable and interpretable is realized.

Inventors

LIN TAO
TANG BO

Assignees

北京中科慧居科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260213

Claims (10)

1. A generated AI operation decision system based on an organization policy and a vehicle context, comprising: The multi-source situation data access module is used for acquiring organization strategy data, vehicle context data and external environment data in real time and adding a time stamp and a space identifier for the acquired data; the strategy knowledge graph construction module is used for converting the organization strategy data into a structured strategy knowledge graph, wherein the strategy knowledge graph at least comprises strategy nodes, rule nodes, target nodes and trigger relationships, constraint relationships, priority relationships and conflict relationships among the strategy nodes, the rule nodes and the target nodes; The multi-mode context fusion module is used for carrying out space-time alignment and semantic matching on the strategy knowledge graph, the vehicle context data and the external environment data and generating fusion context characterization according to causal analysis and priority ordering, wherein the fusion context characterization at least comprises a physical layer, a rule layer and a target layer, and the fusion context characterization is output in a mixed form of a structured field and a natural language abstract; The generation type AI decision-making module is used for dynamically constructing contextualization prompts according to the fusion situation characterization, and inputting the contextualization prompts into a generation type AI model which is finely adjusted in advance to generate a plurality of candidate operation decision-making schemes, wherein each candidate operation decision-making scheme at least comprises an executable instruction set, an application premise, a risk prompt and a cited rule clause mark; The digital twin simulation verification module is used for converting each candidate operation decision scheme into a simulation instruction sequence and carrying out deduction in a digital twin simulation environment of the high-fidelity vehicle, wherein repeated deduction of a disturbance term is carried out at least once for each candidate operation decision scheme so as to obtain risk probability or robustness indexes, and a simulation evaluation report is generated; The strategy compliance arbitration module is used for carrying out semantic matching and logic reasoning on each candidate operation decision scheme and the strategy knowledge graph and generating a compliance arbitration report, wherein the compliance arbitration report at least comprises hard constraint verification, soft constraint verification and target compliance evaluation; The decision optimization and interpretation generation module is used for carrying out multi-objective grading and sorting on the candidate operation decision schemes which are not removed based on the simulation evaluation report and the compliance arbitration report, and selecting an optimal scheme as a final decision; The decision execution and monitoring module is used for converting the final decision into a vehicle networking issuing instruction or a dispatching work order and issuing the instruction or the dispatching work order to vehicles or related personnel, monitoring the execution deviation in real time and recording whole-course audit data comprising situation snapshots, candidate schemes, simulation evaluation reports, compliance arbitration reports, manual approval records and execution logs.
2. The organizational policy and vehicle context based generated AI operation decision system of claim 1, further comprising: The system comprises a feedback learning and system optimization module, a strategy knowledge graph, a training sample generation module and a vehicle digital twin simulation model generation module, wherein the feedback learning and system optimization module is used for acquiring real effect data after decision execution and performing deviation calculation with the simulation evaluation report, updating weights of applicable conditions or priority relationships of at least one rule node in the strategy knowledge graph based on the deviation calculation result, constructing a training sample and performing incremental fine adjustment for generating an AI model, and calibrating energy consumption parameters or traffic flow parameters of the vehicle digital twin simulation model to reduce the follow-up simulation prediction error.
3. The system for generating AI operation decision-making based on organization policy and vehicle context according to claim 1, wherein when the policy knowledge graph construction module converts the organization policy data into a structured policy knowledge graph, the system specifically comprises preprocessing the organization policy data, identifying entity and semantic relations in the preprocessed organization policy data, constructing a knowledge graph according to the entity and semantic relations, and obtaining the structured policy knowledge graph.
4. The organizational policy and vehicle context based generation AI operation decision system of claim 1, wherein the multimodal context fusion module utilizes a causal inference model to analyze causal links among different factors when performing causal analysis.
5. The system for generating AI operation decisions based on organizational policies and vehicle contexts of claim 1, wherein the multimodal context fusion module, when prioritizing, prioritizes a plurality of activation policies and identifies core constraints based on a priority relationship in the policy knowledge graph and a current organizational goal.
6. The organizational policy and vehicle context based generated AI operation decision system according to claim 1, wherein the fine tuning of the generated AI model is performed by supervised fine tuning and reinforcement learning fine tuning of a large language model on traffic operation domain data.
7. The organizational policy and vehicle context based generated AI operational decision system of claim 1, wherein said simulated assessment report comprises a scene description, index data, risk assessment, visualization data, and composite score.
8. The organizational policy and vehicle context based generated AI operational decision system of claim 1, wherein said compliance arbitration report comprises a checklist, a violation list, a deviation specification, a target compliance, a composite score, and an approval record.
9. The organizational policy and vehicle context based generation type AI operation decision making system of claim 1, wherein the decision making and interpretation generation module evaluates dimensions in performing multi-objective optimization including security, compliance, efficiency, economy, customer satisfaction, and environmental friendliness.
10. A method for generating AI operation decisions based on an organization policy and a vehicle context, wherein the method is applied to the system for generating AI operation decisions based on an organization policy and a vehicle context according to any one of claims 1 to 9, and comprises: step 1, acquiring organization strategy data, vehicle context data and external environment data in real time, and adding a time stamp and a space identifier to the acquired data; Step 2, converting the organization strategy data into a structured strategy knowledge graph, wherein the strategy knowledge graph at least comprises strategy nodes, rule nodes, target nodes, triggering relations, constraint relations, priority relations and conflict relations among the strategy nodes, the rule nodes and the target nodes; Step 3, carrying out space-time alignment and semantic matching on the strategy knowledge graph, the vehicle context data and the external environment data, and generating fusion situation characterization according to causal analysis and priority ordering, wherein the fusion situation characterization at least comprises a physical layer, a rule layer and a target layer, and the fusion situation characterization is output in a mixed form of a structured field and a natural language abstract; Step 4, dynamically constructing contextualized prompts according to the fusion situation characterization, and inputting the contextualized prompts into a generated AI model to generate a plurality of candidate operation decision schemes, wherein each candidate operation decision scheme at least comprises an executable instruction set, an application premise, a risk prompt and a cited rule clause mark; Step 5, converting each candidate operation decision scheme into a simulation instruction sequence and carrying out deduction in a high-fidelity vehicle digital twin simulation environment, wherein repeated deduction of introducing disturbance items is carried out at least once for each candidate operation decision scheme so as to obtain risk probability or robustness indexes, and a simulation evaluation report is generated; step 6, carrying out semantic matching and logic reasoning on each candidate operation decision scheme and the strategy knowledge graph and generating a compliance arbitration report, wherein the compliance arbitration report at least comprises hard constraint verification and soft constraint verification and target compliance evaluation; Step 7, performing multi-objective grading and sorting on the candidate operation decision schemes which are not removed based on the simulation evaluation report and the compliance arbitration report, and selecting an optimal scheme as a final decision; And step 8, converting the final decision into an instruction issued by the Internet of vehicles or a dispatching work order, issuing the instruction to vehicles or related personnel, monitoring the execution deviation in real time, and recording whole-course audit data comprising a situation snapshot, a candidate scheme, a simulation evaluation report, a compliance arbitration report, a manual approval record and an execution log.

Description

Generating type AI operation decision system based on organization strategy and vehicle context Technical Field The application relates to the technical field of intelligent traffic, in particular to a generation type AI operation decision system based on an organization strategy and a vehicle context. Background With the rapid development of intelligent traffic systems and internet of vehicles, vehicle operation management is gradually evolving from a traditional manual scheduling mode to a data-driven and intelligent decision-making direction. In application scenes such as sharing travel, logistics transportation, public transportation and the like, how to efficiently allocate vehicle resources, optimize a driving path, reduce operation cost and improve user satisfaction is a core problem of industry attention. Under the background, the vehicle operation decision system is generated by relying on big data analysis, an artificial intelligent algorithm and a real-time communication technology to comprehensively perceive and intelligently process multidimensional information such as vehicle states, traffic environments, user demands and the like, so that dynamic, accurate and efficient operation decision support is realized. At present, a vehicle operation decision system mainly depends on a preset rule engine and a traditional optimization algorithm, and has the following technical defects: The strategy understanding capability is insufficient, namely the existing system cannot understand unstructured organization strategy texts, such as 'priority guarantee brand reputation in this quarter', and other high-level targets, so that the strategy level and the execution level are disjointed. The code is manually modified when the strategy is changed, and the response period is up to weeks or even months. The traditional rule engine adopts static IF-THEN logic, and is difficult to cope with complex and changeable real scenes. For example, the preset "detour on congestion" rule does not distinguish between nuances such as whether an emergency order is made, whether the customer is willing to afford additional costs, whether the current monthly KPI guide is met, etc. The decision creativity is lost, in the face of unexpected scenes (such as extreme weather superimposed sudden epidemic situation), the traditional algorithm can only search for a suboptimal solution in a predefined solution space, and the capability of generating innovative coping schemes is lacking. Compliance verification is lagged, the existing system depends on whether the post-hoc manual inspection decision accords with hundreds of policy regulations, is low in efficiency and easy to miss, and causes illegal operations to be discovered after execution. The decision black box problem is that the decision process of the traditional optimization algorithm is opaque, a manager cannot understand why the system makes the decision, the trust is low, and manual intervention is still needed at the key moment. The lack of a continuous learning mechanism, the rule and the parameter need to be manually and regularly adjusted, and the system cannot automatically learn and optimize from the historical decision effect. In recent years, generative artificial intelligence (large language models) has demonstrated powerful semantic understanding, contextual reasoning and content generation capabilities, but there is a significant risk to directly apply to vehicle operation: Phantom problems the model may generate decisions that look reasonable but violate physical laws or safety regulations, such as suggesting vehicles traveling long distances on road segments without charging facilities. Lacking the constraint that pure generation AI is difficult to strictly adhere to the specific policy terms of an organization, schemes may be generated that are contrary to corporate strategic goals. Unpredictability, namely, the model output has randomness, different decisions can be generated under the same situation, and the reliability requirement of an operation system is difficult to meet. Disclosure of Invention Therefore, the application provides a generating type AI operation decision system based on an organization strategy and a vehicle context, so as to solve the problems that the prior art easily generates illusion, lacks constraint and has unpredictability when the generating type artificial intelligence is applied to the vehicle operation. In order to achieve the above object, the present application provides the following technical solutions: In a first aspect, a generated AI operation decision system based on an organization policy and a vehicle context, comprising: The multi-source situation data access module is used for acquiring organization strategy data, vehicle context data and external environment data in real time and adding a time stamp and a space identifier for the acquired data; the strategy knowledge graph construction module is used for converting