CN-121979987-A - Intelligent tower crane-oriented natural language interaction AI proxy system and method thereof

CN121979987ACN 121979987 ACN121979987 ACN 121979987ACN-121979987-A

Abstract

The invention provides a natural language interaction AI proxy system and a natural language interaction AI proxy method for an intelligent tower crane. The agent system comprises a natural language understanding module, a task planning module, a tool scheduling engine, a large language model interface, a data adaptation layer, a result generation module and a learning optimization module, and the agent method comprises the steps of S1, natural language input processing, S2, intention understanding and task planning, S3, data acquisition and preprocessing, S4, intelligent analysis and reasoning, S5, result verification and optimization, S6, natural language answer generation, and S7, feedback learning and optimization. According to the intelligent tower crane system and the intelligent tower crane system, the special AI proxy architecture is constructed, natural language interaction between a user and the intelligent tower crane system is realized, the large language model is automatically called to conduct intelligent analysis on the running state of the tower crane, and an analysis result and decision advice which are easy to understand are provided for the user.

Inventors

GUAN ZHENCHANG
CHEN FENGJIN
FENG QI
LUO JIANBIN
HUANG YONG
SONG YUNBO
DONG PENG
XIE LEI
Xu Linzheng
HUANG CHAO
MA XIAOBIN

Assignees

中建一局集团东南建设有限公司
福州大学
福建省榕圣建设发展有限公司
福建省二建建设集团有限公司

Dates

Publication Date: 20260505
Application Date: 20260116

Claims (10)

1. An intelligent tower crane oriented natural language interaction AI proxy system is characterized by comprising: the natural language understanding module is used for receiving natural language input of a user, and carrying out semantic analysis and user intention recognition; The task planning module is used for making a corresponding task plan according to the identified user intention; The tool scheduling engine is used for automatically selecting and calling a corresponding analysis tool according to the task plan; The large language model interface is used for calling the large language model to carry out deep reasoning and analysis; The data adapting layer is used for converting the original data of the tower crane system into a format which can be understood by the large language model; The result generation module is used for converting the analysis result into a natural language answer and returning the natural language answer to the user; and the learning optimization module is used for continuously optimizing the system performance according to the user feedback.
2. The intelligent tower crane oriented natural language interactive AI proxy system as set forth in claim 1, wherein the natural language understanding module employs a deep learning model architecture based pre-training language model with the following features: integrating a professional vocabulary library in the field of building engineering; A few sample learning technology is adopted to quickly adapt to a specific expression mode in the field of tower cranes; multiple rounds of dialogue context understanding are supported that are capable of processing successive related queries.
3. The intelligent tower crane oriented natural language interactive AI proxy system as set forth in claim 1, wherein the task planning module employs a hierarchical task decomposition strategy comprising: the intention classification layer classifies the user inquiry into four categories of state inquiry, risk assessment, efficiency analysis and knowledge consultation; the task decomposition layer is used for decomposing the complex query into an executable atomic task sequence; The dependency analysis layer is used for analyzing the dependency relationship among tasks and determining the execution sequence; And the priority ranking layer ranks the priorities according to the importance and the emergency degree of the tasks.
4. The intelligent tower oriented natural language interactive AI proxy system as set forth in claim 1, wherein the tool scheduling engine comprises the following tool categories: The data query tool set comprises a sensor data query tool, a historical data retrieval tool and a real-time state acquisition tool; The analysis and calculation tool set comprises a safety assessment tool, an efficiency analysis tool and a trend prediction tool; the knowledge retrieval tool set comprises a standard query tool, an operation manual retrieval tool and a case library query tool; The visual generation tool set comprises a chart generation tool, a report generation tool and a 3D visual tool.
5. The intelligent tower crane oriented natural language interactive AI proxy system as set forth in claim 1, wherein the large language model interface adopts a model-as-a-service architecture, the model-as-a-service architecture having the following characteristics: supporting unified call of multiple large language models; Model load balancing and fault switching are realized, and high availability of service is ensured; integrating a model performance monitoring and cost control mechanism; hot plug and version management of the model is supported.
6. The intelligent tower crane oriented natural language interactive AI proxy system as set forth in claim 1, wherein the data adaptation layer adopts a multi-modal data fusion technique, the multi-modal data fusion technique having the following characteristics: The structured data processing is to convert the structured data such as sensor data, control parameters and the like into natural language description; Time sequence data processing, namely converting time sequence data into trend description and anomaly annotation; Image data processing, namely converting video monitoring data into scene description and object recognition results; context enhancement-adding domain knowledge background and interpretation information to the data.
7. The intelligent tower crane oriented natural language interactive AI proxy system as set forth in claim 1, wherein the result generation module combines templating and generation, and has the following characteristics: The safety key information adopts a predefined template to ensure accuracy and consistency; Analyzing and interpreting information is generated by adopting a large language model, and a personalized expression mode is provided; Confidence labels provide confidence scores for each analysis result; The diversified output supports various output forms such as text, charts, voice and the like.
8. The intelligent tower crane-oriented natural language interaction AI proxy method is characterized by adopting the intelligent tower crane-oriented natural language interaction AI proxy system as claimed in any one of claims 1-7, and comprises the following steps: S1 natural language input processing Receiving natural language query input of a user, performing voice recognition and text preprocessing, and extracting key information and context of the query; s2, intention understanding and task planning Performing intention recognition and semantic analysis based on a natural language understanding module, mapping user inquiry to specific analysis task types, and making a detailed task execution plan and a tool call sequence; S3, data acquisition and preprocessing Invoking a corresponding data query tool according to task requirements, acquiring related data of the tower crane system, and cleaning, formatting and enhancing the context of the original data; s4, intelligent analysis and reasoning Inputting the preprocessed data into a large language model, and carrying out deep analysis and reasoning by combining domain knowledge to generate a preliminary analysis result and a preliminary conclusion; S5, result verification and optimization Performing rationality test on the analysis result, comparing and verifying with preset safety rules and constraint conditions, and adjusting and optimizing the analysis conclusion according to the verification result; S6, natural language answer generation Converting the analysis result into natural language expression which is easy to understand by the user, adding necessary explanation and suggestion, generating a final answer and returning the final answer to the user; s7, feedback learning and optimization And collecting feedback of the user on answer quality, analyzing system performance and user satisfaction, and updating model parameters and optimization strategies.
9. The natural language interactive AI proxy method for intelligent tower crane as set forth in claim 1, wherein in the step S2, the intention recognition adopts a multi-level classification strategy, specifically as follows: Primary classification, namely identifying the basic type of the query; secondary classification, namely identifying the specific field of inquiry; three-level classification, namely identifying the accurate intention of the query; In the step S4, the large language model call adopts a thinking chain reasoning method, and the thinking chain reasoning method comprises the following characteristics: problem decomposition, namely decomposing a complex problem into a plurality of sub-problems; step-by-step reasoning, namely gradually analyzing each sub-problem according to a logic sequence; the result synthesis, namely synthesizing and summarizing the analysis results of all the sub-problems; confidence assessment provides a confidence score for the final conclusion.
10. The natural language interactive AI proxy method for an intelligent tower crane as set forth in claim 1, wherein the method further comprises a security mechanism, the security mechanism comprising the following features: Inputting security check, namely performing security check on user input and filtering malicious query; verifying the inquiry authority of the user, and ensuring the safety of data access; Output content auditing, namely conducting security auditing on the answer generated by the AI, and avoiding misleading information; Operation boundary limit-ensure that the AI agent does not perform any device control operations, only provide analysis suggestions.

Description

Intelligent tower crane-oriented natural language interaction AI proxy system and method thereof Technical Field The invention relates to the technical field of artificial intelligence and natural language processing, in particular to an AI proxy system based on a large language model, which is used for carrying out intelligent analysis and decision support on an intelligent tower crane in a natural language interaction mode. Background With the rapid development of artificial intelligence technology, AI agents (AI agents) are widely used in various fields as intelligent systems capable of autonomously sensing environments, planning and performing actions. However, in the field of intelligent industrial equipment, in particular intelligent interaction of construction engineering machinery, the following problems exist in the prior art: 1. The interaction mode is single and behind, the man-machine interaction of the traditional industrial equipment mainly depends on a professional operation interface, an instrument panel and an alarm system, a user needs to have professional knowledge to understand complex technical parameters, the learning cost is high, and the operation threshold is large. 2. The intelligent understanding capability is lacking, namely the existing construction engineering mechanical system cannot understand natural language query of a user, cannot convert daily language of the user into a specific equipment analysis task, and the usability and popularity of the system are limited. 3. Analysis capability fragmentation, namely, although industrial equipment generates a large amount of data, a unified intelligent analysis framework is lacking, and various analysis functions are independent from each other and cannot provide comprehensive decision support for users. 4. The self-adaptive learning mechanism is lacking, the existing construction engineering mechanical system is mostly driven by static rules, cannot be self-optimized according to user feedback and use habits, and is difficult to adapt to individual demands of different users. Therefore, a natural language interaction AI proxy system specially oriented to industrial equipment such as intelligent tower cranes and the like is needed, natural language query of users can be understood, a large language model is automatically called for intelligent analysis, and valuable decision information is returned. Disclosure of Invention Aiming at the problems, the invention aims to provide a natural language interaction AI proxy system and a method thereof for an intelligent tower crane, which are used for realizing natural language interaction between a user and the intelligent tower crane system by constructing a special AI proxy architecture, automatically calling a large language model to carry out intelligent analysis on the running state of the tower crane, and providing an easy-to-understand analysis result and decision suggestion for the user. The invention firstly provides a natural language interaction AI proxy system oriented to an intelligent tower crane, which comprises: the natural language understanding module is used for receiving natural language input of a user, and carrying out semantic analysis and user intention recognition; The task planning module is used for making a corresponding task plan according to the identified user intention; The tool scheduling engine is used for automatically selecting and calling a corresponding analysis tool according to the task plan; The large language model interface is used for calling the large language model to carry out deep reasoning and analysis; The data adapting layer is used for converting the original data of the tower crane system into a format which can be understood by the large language model; The result generation module is used for converting the analysis result into a natural language answer and returning the natural language answer to the user; and the learning optimization module is used for continuously optimizing the system performance according to the user feedback. Further, the natural language understanding module adopts a pre-training language model based on a deep learning model (transducer) architecture, and the pre-training language model has the following characteristics: A professional vocabulary library in the field of integrated building engineering, which comprises tower crane terms, component names, operation specifications and the like; a few sample Learning (Few-shot Learning) technology is adopted to quickly adapt to a specific expression mode in the field of tower cranes; multiple rounds of dialogue context understanding are supported that are capable of processing successive related queries. Further, the task planning module adopts a hierarchical task decomposition strategy, and the hierarchical task decomposition strategy comprises the following characteristics: the intention classification layer classifies the user inquiry into four categories of sta