An effective multi-objective community development algorithm and its application to identify control model of supercritical units

The community development algorithm (CDA) performs well in solving numerical optimization problems and practical engineering applications. To better utilize CDA, this paper combines it with non-dominated sorting and clustering-based special crowding distance to form a multi-objective community development algorithm (MOCDA). In addition, the helper selection mechanism is devised to select the more suitable learning objects for particles. A series of comprehensive examinations prove that MOCDA is better than the other 9 state-of-the-art competitors on the multimodal multi-objective Congress on Evolutionary Computation 2020 and imbalanced distance minimization benchmark problems. Quantitatively, MOCDA leads MMODE_CSCD by 22.44%, demonstrating a strong ability to solve multimodal multi-objective optimization problems. For engineering practice, MOCDA is employed to identify the three-input, three-output control model of supercritical units by regarding the data of multiple time periods as multiple objectives, and the experimental results show that this approach is more effective than the direct summation of the single objective algorithm. During the encoding process, an additional position is added for the solution’s chromosome to control whether or not a delay link works. Experimental results show that this method has a lower root mean square error and significantly reduces the maximum error at the initial moment compared to the encoding scheme with a fixed delay link. Most importantly, the identification accuracy of MOCDA is much higher than that of other algorithms, indicating its superiority in solving challenging multi-objective problems in the real world. The source code of MOCDA is publicly available at: https://github.com/Mingliang-Wu/MOCDA.git.