Click the text at the bottom left corner “Read the Original” to jump to IEEE Xplore to view the paper on your mobile!
✦
✦
This paper addresses the issue of how to improve the effectiveness of power grid control by enhancing the intelligence level of agents. It proposes a theory and method for the objective quantification assessment of agent intelligence levels, and based on this, it presents a method for the autonomous optimal evolution of agents guided by intelligent assessment results. The implementation of these methods has led to significant improvements in power grid control effectiveness.
Parallel System Based Quantitative Assessment and Self-Evolution for Artificial Intelligence of Active Power Corrective Control
Active Power Corrective Control Agent Based on Parallel Systems: Intelligent Assessment and Autonomous Evolution
Authors: Zhang Tianyun, Zhang Jun, Wang Feiyue, Xu Peidong, Gao Tianlu, Zhang Haoran, Si Ruiqi
DOI: 10.17775/CSEEJPES.2023.00190
Article Link: https://ieeexplore.ieee.org/document/10375965
01
Challenges of Artificial Intelligence Technology in Power System Dispatch Control:
Agent Evaluation and Evolution Optimization
The lack of objective quantification assessment methods for agent intelligence levels has made it difficult for researchers to evaluate and compare the performance of power grid control agents with clear and intuitive standards, leading to challenges in optimization. The currently used reinforcement learning model algorithms also face issues such as low flexibility and high time and labor costs in finding agents with better power grid control effectiveness. These problems greatly restrict the application of artificial intelligence technology in power system dispatch control. Therefore, this paper proposes a theory for the objective quantification assessment of agent intelligence levels and a method for the autonomous evolution of agents based on intelligent assessment results.
02
How to objectively evaluate the intelligence level of agents
and utilize intelligent assessment results
to promote the evolution of power grid control agents?
The essence of artificial intelligence is the imitation of human intelligence. The evaluation of human intelligence can be demonstrated through assessments of specific abilities in specific scenarios. Similarly, the quantitative assessment of machine intelligence levels can be achieved through testing and evaluating the power grid control capabilities and effects in typical (power grid) scenarios. Following the theoretical framework of imitating human intelligence, this paper innovatively constructs a system of intelligent quantitative assessment indicators and proposes a theoretical method for the objective quantification assessment of agent intelligence levels. Based on this, using automatic reinforcement learning technology as the framework for autonomous evolution and Bayesian optimization algorithms as the evolutionary algorithms, it proposes a method for autonomous evolution based on intelligent assessment results. Using the above methods, a parallel system-based intelligent assessment and autonomous evolution system (PLASE system) was constructed as the experimental platform. Using the PLASE system as the working platform, this paper achieves the objective quantification assessment of the intelligence level of power grid control agents, a clear and intuitive description of power grid control effects, and autonomous optimal evolution based on intelligent assessment.
Figure 1 Intelligent Quantitative Assessment Method
Figure 2 Autonomous Evolution Based on Intelligent Assessment
03
Evolution Effect of Power Grid Control Agents Guided by Intelligent Assessment Results
This paper takes the comparative transfer index as the intelligent index to be tested, and the operational cost score as the scoring formula for the indicator. It utilizes the PLASE system for the objective quantification assessment of the intelligence level of corrective control agents and for the autonomous optimal evolution of agents guided by intelligent assessment results. The operational results of the PLASE system indicate that during the training process, the changes in the intelligence level of the corrective control agents are consistent with the improvements in the agents’ power grid control capabilities. Experiments demonstrate that using the intelligent autonomous optimal evolution method based on intelligent quantitative assessment can yield agents with higher intelligence levels and better power grid control effects. From the perspective of improving intelligence levels, the evolution effect improvement rate of agents reaches as high as 88%. Using the methods proposed in this paper, accurate directional evolution of power grid control agents can be achieved.
Figure 3 3D Diagram of Autonomous Evolution Effects of Power Grid Control Agents Based on Intelligent Assessment
Citation Information
T.Y. Zhang, J. Zhang, F.Y. Wang, P.D. Xu, T.L. Gao, H.R. Zhang, R.Q. Si, “Parallel System Based Quantitative Assessment and Self-evolution for Artificial Intelligence of Active Power Corrective Control”, CSEE Journal of Power and Energy Systems, vol. 10, no. 1, pp. 13–28, Dec. 2023.
Author Introductions
Zhang Tianyun: PhD student. Mainly engaged in research on the application of hybrid intelligence technology in power system dispatch control.
Zhang Jun: Professor, corresponding author. Currently employed at the School of Electrical Engineering and Automation, Wuhan University, mainly engaged in research on signal processing, big data, artificial intelligence, blockchain, and their applications in complex intelligent systems.
Wang Feiyue: Professor. Currently employed at the Institute of Automation, Chinese Academy of Sciences, mainly engaged in research on parallel systems, social computing, and knowledge automation.
Xu Peidong: Postdoctoral researcher. Mainly engaged in research on power system dispatch control, artificial intelligence, and intelligent vehicles.
Gao Tianlu: PhD student. Mainly engaged in research on simulation technology, artificial intelligence, processing, and blockchain applications in power system control.
Zhang Haoran: PhD student. Mainly engaged in research on power grid control and the application of machine learning in power grids.
Si Ruiqi: PhD student. Mainly engaged in research on power grid optimization, reinforcement learning, and multi-agent systems.
Serious Statement:
According to the relevant regulations of the National Copyright Administration, the reproduction and compilation of this WeChat work in print media, websites, Weibo, and WeChat official accounts must include the name of this WeChat account, QR code, and other key information, and the original name “CSEE JPES” must be indicated at the beginning. Individuals are requested to forward and share according to the original text of this WeChat account.
Production: Zhu Zhirun
Editor: Zhu Tengyi
Online Submission Website
http://mc03.manuscriptcentral.com/cjpes
