Theoretical Foundation and Directions of Electric Power Artificial Intelligence (I): Hypothesis Analysis and Application Paradigm

To achieve carbon emissions peak and carbon neutrality, artificial intelligence(AI) is becoming the essential technique in the process of building a power network based on new energy. In recent years, AI applications in different electric power scenarios have varied widely in applicability and performance, due to the theoretical hypothesis and inherent limitation of AI algorithms, as well as the specific needs of different electric power applications. In order to address these issues, this paper summarizes four paradigms of electric power artificial intelligence, or electric power AI in short, including deep connectionist, symbolist, reinforced- behaviorist and ensemblist, by analyzing the fundamental hypothesis and due limitations of their core algorithms, matching the characteristics and needs of electric power scenarios, and sorting out the specific algorithms with better performance and accordingly technical parameters. Furthermore, this paper identifies the common technical bottlenecks faced by electric power AI, which includes trustworthy ethnics bottleneck, data distribution bottleneck and evolutionary migration bottleneck. Facing with these obstacles, three electric power AI mechanisms including data-knowledge fusion-driven mechanism, parallel interaction mechanism and model evolving mechanism are analyzed. In the following article, these three mechanisms would be discussed in details, with a more systematic electric power AI development mechanism being proposed, in order to promote the self-organization, self-coordination, self-learning and self-evolving capabilities of electric power AI. © 2023 Chinese Society for Electrical Engineering. All rights reserved.