Optimizing the Reservoir Connection Structure Using Binary Symbiotic Organisms Search Algorithm: A Case Study on Electric Load Forecasting

Abstract

ESN (Echo state network) is a novel recurrent neural network and it is also acknowledged as a powerful temporal processing method, especially in real-valued, time-series forecasting fields. The current research results believe that the connection structure of the reservoir has significant effect for ESN’s forecasting performance. However, the randomly generated reservoir is hard to establish a optimal reservoir structure for a given task. Optimizing the connection structure of reservoir can be considered as a feature selection issue and this issue can be solved by binary optimization algorithm. SOS (Symbiotic organisms search) is a recently proposed heuristic algorithm and its superior performance is confirmed via many mathematical benchmark functions and engineering design problems. It’s worth noting that the original SOS is only suitable for continuous numerical optimization problems. In this paper, a binary SOS, called BSOS, is employed to optimize the connection structure of reservoir of the standard ESN. To verify the effectiveness of the proposed model, a real electric load series derived from New South Wales in Australia is adopted as benchmark dataset. The experimental results demonstrate that the proposed model can significantly improve the forecasting accuracy and it is a hopeful forecasting model.