Operation rule derivation of hydropower reservoir by k-means clustering method and extreme learning machine based on particle swarm optimization

Abstract

In practice, the rational operation rule derived from historical information and real-time working condition can help the operators make the quasi-optimal scheduling plan of hydropower reservoirs, leading to significant improvements in the generation benefit. As an emerging artificial intelligence method, the extreme learning machine (ELM) provides a new effective tool to derivate the reservoir operation rule. However, it is difficult for the standard ELM method to avoid falling into local optima due to the random determination of both input-hidden weights and hidden bias. To enhance the ELM performance, this research develops a novel class-based evolutionary extreme learning machine (CEELM) to determine the appropriate operation rule of hydropower reservoir. In CEELM, the k-means clustering method is firstly adopted to divide all the influential factors into several disjointed sub-regions with simpler patterns; and then ELM optimized by particle swarm intelligence is applied to identify the complex input-output relationship in each cluster. The results from two reservoirs of China show that our method can obtain satisfying performance in deriving operation rules of hydropower reservoir. Thus, it can be concluded that the model’s generalization capability can be improved by isolating each subclass composed of similar dataset.