Application of Hybrid Neuro-Wavelet Models for Effective Prediction of Wind Speed

Abstract

Severe energy crisis and depletion of fossil fuels necessitates more number of installations of wind farms. Accurate wind forecast is crucial in the efficient utilization and power management of wind farms connected to a grid or in conjunction with other sources such as solar, DG, battery, etc. This paper proposes a hybrid neuro-wavelet predictive tool to predict wind speed which combines the advantages of both wavelet decomposition and neural network. Wavelet decomposition is used to filter out the high frequency outliers in the wind speed, thus making a smooth data to make the prediction accurate. The filtered data is used to train the neural network. Four different models are proposed. NAR-TS model and NAR-Wavelet models are univariate models with past values of wind speed as input. In NAR-TS model time series values are directly applied as input to neural network, whereas in NAR-Wavelet model input to the neural network is the wavelet decomposed data. In a similar way NARX-TS and NARX-Wavelet models are developed with multivariate neural network, where the inputs are air temperature, relative humidity and wind speed which is the feed back. Each of these models are used to predict 4.5 hours ahead and 18 hours ahead predictions. The Mean Average Percentage Error (MAPE) values are calculated for each model and the results are compared.