Continuous Modeling of Power Plant Performance with Regularized Extreme Learning Machine
Rui Xu, Weizhong Yan · 2019
Power plant modeling is critically important for power plant operation optimization and cost reduction. The inherently nonstationary characteristics of power plants raise a big challenge to the learning mechanisms and require the learning algorithms to adapt effectively and promptly to the continuously drifting environments. In our previous study, we proposed an online ensemble regression approach, with extreme learning machine (ELM) as the base model, to model power plant performance in a dynamic environment, which can autonomously update models to respond to environmental changes, either gradual or abrupt. However, one drawback we observed for the proposed approach is that the algorithm performance is not stable due to the randomness nature of ELMs. In this paper, we address this issue by applying regularized ELM as the base model within the online ensemble framework. The empirical results on three real power plant data sets demonstrate that the proposed modification can lead to more stable generalization performance of the algorithm. At the same time, the algorithm consistently achieves performance with mean average percentage error less than the required 1% threshold in real field operations.