Rule-based expert system for energy optimization : detection and identification of relationships between rules in knowledge base

This master thesis presents the design of a rule-based expert system that uses specific energy observations to provide energy recommendations to its users. The aim of this thesis was to improve a similar recommendation AI developed by WeSmart, a company that provides support to energy communities. Since the quality of the recommendations issued are highly dependent on the rules in the knowledge base, the main focus of the work was to develop a method to help the human experts in the design of such rule sets. To reach this objective, I have produced an algorithm that identifies different types of relationships between rules, basing my work on an existing technique created for firewall policies. I have also implemented a prototype of tool using that algorithm. This program, Relationship Identification Tool (RIT), allows to manage a rule set by identifying the different relationships between the rules as well as providing support for the creation and modification of the rule set. This manuscript starts with an explanation of the context of energy communities and a presentation of WeSmart. Then comes an assessment of the existing AI developed by WeSmart and suggestions for improvements. After that, a review of literature from different fields is presented alongside the conclusions made regarding the development of the chosen solution. Those fields are associative classification, expert systems and firewalls. After this description of the process that lead to it comes the presentation of the results.