Ocean surface wind prediction Based on GA-LIESN On-line Learning Model

Abstract

The winds that often occur on the ocean surface have a great impact on economic activities such as the development of ports and harbor engineering. The accurate forecast of wind speeds on the ocean surface is of great significance. A Genetic Algorithm-Leaked Integral Echo State Network(GA-LIESN) On-line Learning Model for ocean surface wind prediction is proposed in this paper. First of all, the model uses correlation analysis to select the features which are most relevant to the wind speed label as the input of the forecast model. Then introduce leaky integral neurons into echo state network (ESN) to improve the model's short-term memory for time series. Next, an On-line Learning algorithm is used to correct the model in real time by introducing new samples to make the model adaptive. Finally, in order to overcome the blind selection of network parameters, genetic algorithm (GA) is used to optimize the network parameters. The meteorological observation data which are measured from a station in the Bohai Bay are selected as experimental data. Compared with Auto-Regressive and Moving Average Model (ARMA) and ESN, the GA-LIESN On-line Learning Model shows a higher prediction accuracy.