Evolutionary extreme learning machine for energy price forecasting

摘要

Accurate electricity price forecasting is a key area in the electricity market. This paper proposes a hybrid model, Evolutionary-Improved Cuckoo Search Extreme Learning Machine (E-ICSELM) for day and week ahead prediction of a highly volatile financial time series data i.e. electricity price for six different energy markets such as Hourly Ontario Electricity Price (HOEP), Pennsylvania Jersey Maryland (PJM), New England, Nord Pool, California and Spain. In this model, Improved Cuckoo Search (ICS), a meta-heuristic, population based optimization techniques used to select input weights and hidden biases and Moore-Penrose (MP) generalized inverse to analytically determine the output weights. Experimental results show the superiority of the proposed E-ICSELM model when it is compared with simple ELM and Evolutionary-Cuckoo Search based ELM (E-CSELM).