Prediction of thermal system parameters based on PSO-ELM hybrid algorithm
Liangyu Ma, Lijuan Zhao, Xiaoxia Wang · 2017
Parameter prediction with high precision is of great importance for real-time condition monitoring and fault diagnosis of the thermal system during variable load process. This paper presents a performance enhancement scheme for the extreme learning machine (ELM) to predict the operating parameters of the thermal system using particle swarm optimization (PSO). ELM is a feed-forward neural network with single hidden layer, which has well generalization ability and fast learning ability. However, the number of the hidden layer nodes of ELM could not be automatically obtained. The optimal selection of ELM parameters can improve its performance. In this paper, the discrete-valued PSO is applied for optimizing the number of the hidden nodes to enhance ELM performance. The simulation results show that the proposed hybrid algorithm is more accurate and effective for predicting thermal system parameters.