CN-122019548-A - Energy source hosting method based on search enhancement generation and multi-agent cooperation

CN122019548ACN 122019548 ACN122019548 ACN 122019548ACN-122019548-A

Abstract

The invention relates to an energy source hosting method based on search enhancement generation and multi-agent cooperation, which comprises the following steps of S1, synchronizing a plurality of time sequence operation data of hosting equipment in real time, fitting to obtain a plurality of straight line segments and circular arc segments to form continuous and smooth alternative curves, S2, converting geometric description of the alternative curves into natural language description by combining offline document data, vectorizing to construct a mixed index knowledge base comprising a vector database and a knowledge graph, S3, receiving a user instruction, disassembling the user instruction into a plurality of subtasks, S4, utilizing concepts related to query entities in the knowledge graph to conduct query expansion through the mixed index knowledge base when in vector search, reserving high-scoring search results as high-quality knowledge segments, and S5, generating an energy source equipment control strategy by the high-quality knowledge segments, real-time data and safety constraint of the hosting equipment through a large language model.

Inventors

Wu Mourong
LI WEINA
XU ZEFENG

Assignees

桦熙新能源科技(福建)有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (8)

1. The energy source hosting method based on the search enhancement generation and multi-agent cooperation is characterized by comprising the following steps of: The method comprises the steps of S1, screening a plurality of time sequence operation data of the real-time synchronous hosting equipment, wherein abnormal parameters with monotonous and smooth change process in the time sequence operation data are used as alternative parameters, performing piecewise linear fitting on the alternative parameters to obtain a plurality of straight line segments, performing circular arc fitting on data points between adjacent straight line segments to obtain circular arc segments, adjusting the circular arc segments to form continuous and smooth alternative curves with the corresponding straight line segments; S2, acquiring offline document data of the hosting equipment, converting the geometric description of the alternative curve into natural language description by combining the offline document data, carrying out semantic segmentation and vectorization processing on the offline document data and the natural language description, and constructing a mixed index knowledge base comprising a vector database and a knowledge graph; s3, receiving a user instruction or a system triggering event, carrying out intention recognition based on a large language model, and disassembling a target task into a plurality of subtasks; S4, searching in the mixed index knowledge base according to the structural information contained in each subtask, carrying out query expansion by utilizing concepts related to query entities in a knowledge graph through the mixed index knowledge base during vector searching, carrying out query expansion by utilizing new entities extracted from fragments recalled during vector searching during knowledge graph searching through the mixed index knowledge base, filtering and grading two types of search results, and reserving a high-grade search result as a high-quality knowledge fragment; S5, generating an energy device control strategy through a large language model by using the real-time data and the safety constraint of the high-quality knowledge segments and the hosting device.
2. The method for generating the energy source hosting collaborative with multiple agents based on retrieval enhancement according to claim 1, wherein the alternative parameters comprise a part or all of parameters including a single body temperature, an SOC, a single body pressure difference, a charge-discharge multiplying power of an energy storage battery, an exhaust temperature, a condensation pressure, a chilled water outlet temperature, a current of a water chilling unit, an output power of a photovoltaic inverter, a return air temperature of an end air conditioner and a power factor of a power distribution system.
3. The method for generating the energy source hosting method cooperated with the multi-agent based on the retrieval enhancement according to claim 1, wherein the steps of performing piecewise linear fitting on the candidate parameters to obtain a plurality of straight line segments, performing circular arc fitting on data points between adjacent straight line segments to obtain circular arc segments, and adjusting the circular arc segments to form continuous and smooth candidate curves with the corresponding straight line segments include: Simplifying the alternative parameters into folding lines connected by a plurality of key points, wherein the length of each folding line is longer than a preset duration time, and obtaining the straight line segment; and carrying out least square arc fitting by using transition data points between adjacent straight-line segments, taking the directions of the front and rear adjacent straight-line segments as tangent constraints at the starting point and the end point of the arc segments, and obtaining the arc segments which are continuously and smoothly connected with the adjacent straight-line segments by adjusting the number of the data points and the range of a data window involved in fitting.
4. The method for generating the energy source hosting collaborative with multiple agents based on the retrieval enhancement according to claim 1, wherein the step S2 comprises the following steps: S2.1, introducing the geometric description of the alternative curve into equipment operation rules, parameter constraints and failure mechanisms in offline document data, and carrying out semantic enhancement processing on the geometric description so as to generate specialized description fused with physical meaning and treatment suggestion; S2.2, carrying out semantic segmentation on the offline document data slice and the specialized description, then carrying out vectorization processing, and writing into a vector database; S2.3, extracting entity relations containing equipment, parameters, anomalies, reasons, measures and constraints from the offline document data slice and the specialized description to construct a knowledge graph; s2.4, simultaneously carrying out vectorization storage and entity relation extraction on the vector database and knowledge units belonging to the same text in the knowledge graph, and establishing bidirectional metadata association between the vector database record and the knowledge graph node, thereby forming a mixed index knowledge base combining semantic indexes and structure indexes.
5. The method for generating energy source hosting in coordination with multiple agents based on retrieval enhancement as claimed in claim 4, wherein step S2.1 comprises the steps of: S2.1.1, describing the alternative curve as a geometric description text comprising the geometric description by a time-series data textualization technique; S2.1.2 retrieving from said offline document material relevant knowledge corresponding to said geometric description text; S2.1.3, fusing the geometric description text and the related knowledge, further retrieving and adding the description of possible reasons, processing measures and risk levels of the abnormality from the offline document data, and generating the specialized description.
6. The method for generating energy source hosting in coordination with multiple agents based on retrieval enhancement as claimed in claim 4, wherein step S2.4 comprises the steps of: Simultaneously carrying out vectorization storage and entity relation extraction on text contents of the same knowledge unit, so that the knowledge unit forms a semantic index in a vector database and forms a structural index in a knowledge graph; And establishing a mapping relation between the semantic index and the structural index through the knowledge unit identification, so as to construct a mixed index knowledge base combining the semantic index and the structural index.
7. The method for generating the energy source hosting collaborative with multiple agents based on retrieval enhancement according to claim 1, wherein step S4 comprises the following steps: Searching the nearest text fragment in the vector database according to the structural information contained in each subtask by using an RAG engine, and expanding the text fragment according to the triples in the knowledge graph corresponding to the text fragment; searching the closest triplet in the knowledge graph according to the structural information contained in each subtask, and expanding the triplet according to the text fragment in the vector database corresponding to the triplet; And (3) performing relevance scoring on the recall fragments by adopting a Cross-Encoder Cross encoder, and filtering according to keywords irrelevant to subtasks to obtain a plurality of search results with highest scores as high-quality knowledge fragments.
8. The method for generating the energy source hosting collaborative with multiple agents based on the retrieval enhancement according to claim 1, wherein the step S5 comprises the following steps: S5.1, dynamically screening the high-quality knowledge segments according to the parameter values and the change trend in the real-time data, and reserving the knowledge segments matched with the current working condition; S5.2, limiting boundaries of the screened knowledge segments according to preset safety constraints, and eliminating or correcting suggestions which violate physical limits, operation specifications or logic constraints; s5.3, the knowledge segments after screening, removing or correcting, the real-time data abstract and the safety constraint are assembled into a structured prompt word, a large language model is input, and an energy equipment control strategy containing executable control instructions is generated.

Description

Energy source hosting method based on search enhancement generation and multi-agent cooperation Technical Field The invention relates to the field of energy source hosting, in particular to an energy source hosting method based on search enhancement generation and multi-agent cooperation. Background The energy source hosting is carried out by specialized teams on the energy source system of the energy unit, links such as equipment operation, maintenance and energy consumption optimization are covered, full chain hosting of the photovoltaic energy storage unit, the water chilling unit and the tail end air conditioning equipment can be realized, and meanwhile, technology and equipment updating are carried out on each equipment, so that the purposes of saving energy and energy cost are achieved. The existing energy source hosting method mainly depends on manual experience, takes group control of a water chilling unit as an example, and the load and unload strategy and fault diagnosis are highly dependent on personal experience of operation and maintenance specialists, so that the response speed is low and the operation standardization degree is low. Meanwhile, the data island of each device is serious, and real-time operation data (such as evaporation pressure/condensation pressure/current percentage of a cold water host), device documents (operation manual/maintenance guide), historical worksheets and industry standards are stored in different systems in a scattered manner, so that the association analysis and comprehensive utilization cannot be performed. The invention aims at solving the problems existing in the prior art and designing an energy source hosting method based on the cooperation of search enhancement generation and multiple agents. Disclosure of Invention The invention aims to solve at least one problem existing in the prior art by providing an energy hosting method based on the cooperation of search enhancement generation and multi-agent. The technical scheme of the invention is as follows: an energy hosting method for generating cooperation with multiple agents based on retrieval enhancement comprises the following steps: The method comprises the steps of S1, screening a plurality of time sequence operation data of the real-time synchronous hosting equipment, wherein abnormal parameters with monotonous and smooth change process in the time sequence operation data are used as alternative parameters, performing piecewise linear fitting on the alternative parameters to obtain a plurality of straight line segments, performing circular arc fitting on data points between adjacent straight line segments to obtain circular arc segments, adjusting the circular arc segments to form continuous and smooth alternative curves with the corresponding straight line segments; S2, acquiring offline document data of the hosting equipment, converting the geometric description of the alternative curve into natural language description by combining the offline document data, carrying out semantic segmentation and vectorization processing on the offline document data and the natural language description, and constructing a mixed index knowledge base comprising a vector database and a knowledge graph; s3, receiving a user instruction or a system triggering event, carrying out intention recognition based on a large language model, and disassembling a target task into a plurality of subtasks; S4, searching in the mixed index knowledge base according to the structural information contained in each subtask, carrying out query expansion by utilizing concepts related to query entities in a knowledge graph through the mixed index knowledge base during vector searching, carrying out query expansion by utilizing new entities extracted from fragments recalled during vector searching during knowledge graph searching through the mixed index knowledge base, filtering and grading two types of search results, and reserving a high-grade search result as a high-quality knowledge fragment; S5, generating an energy device control strategy through a large language model by using the real-time data and the safety constraint of the high-quality knowledge segments and the hosting device. Further, the alternative parameters include some or all of the parameters including the cell temperature, SOC, cell pressure difference, charge-discharge rate of the energy storage cell, and the exhaust temperature, condensing pressure, chilled water outlet temperature, current of the chiller, and output power of the photovoltaic inverter, and the return air temperature of the terminal air conditioner, and the power factor of the power distribution system. Further, performing piecewise linear fitting on the alternative parameters to obtain a plurality of straight line segments, performing circular arc fitting on data points between adjacent straight line segments to obtain circular arc segments, and adjusting the circular arc segments to form continuous an