CONSTRUCTION OF AN ENERGY CONSUMPTION ANALYSIS SOFTWARE SYSTEM USING AN ONTOLOGICAL APPROACH AND LARGE LANGUAGE MODELS

Abstract

This study aims to investigate the potential of using large language models for interaction between human and a digital twin that models a building’s energy consumption management system. A review of existing works has shown that the literature on the application of large language models to the field of buildings and energy is currently quite limited, but research in this area is growing rapidly.

The work develops an approach for building a buildings energy usage digital twin based on the use of the proposed ontological model, a graph database, and large language models, which allows obtaining results on energy efficiency through a direct user communication interface with the knowledge database.

As part of the study, a knowledge database with domain knowledge about the relationships between buildings, meters, meter readings and consumption, weather stations, and climate data was built using Neo4j graph database and the ontological model we proposed earlier. A prototype of a chatbot built using the Lang Chain framework, an agent approach, and Retrieval Augmented Generation technology was developed, and the ability of the Gpt-4o-mini model to respond to queries related to energy consumption using semantic relationships in the schema of the built knowledge database, to automatically generate Cypher queries to the Neo4j database, execute them, and respond considering additional context, using defined prompts for the agent, was analysed.

The paper presents the results of the research, the analysis showed that the proposed ontological approach for building an energy consumption analysis system, which is based on knowledge about the features of the domain area, offers a formalized structure and description of the system for representing knowledge, suitable for machine reading, can help people understand the system through natural language interaction, providing better planning and decision-making. It was established that digital twins can model, simulate, help manage the energy consumption system, identify anomalies, predict and identify problems in the system.