A Hybrid Empirical Mode Decomposition based ARIMA and MLP Models for Wind Power Time Series Forecasting

Hybrid models is one of the most well-known solution introduced for obtaining more accurate and more reliable results in real world modelling and forecasting problems. One of the main purpose of these methods is to eliminate the drawbacks of single models in modelling various latent patterns in data. Thus in this study, in order to improve forecasting accuracy of Auto-regression Integrated Moving Average (ARIMA) and multilayer perceptron (MLP) models and reduce the complexity of time series forecasting calculations, a novel hybrid model based on the principle of divide and conquer is represented. In the proposed method, in the first modelling phase, the time series, which is potentially has complexity, high volatility, high frequency and consists of several patterns, is decomposed to its Intrinsic Mode Functions (IMFs) by employing Ensemble Empirical Mode Decomposition (EEMD). In the next step, each of these simplified decomposed components is forecasted using the ARIMA and MLP models. Finally, the forecasted value of each of the IMFs are combined to obtain final hybrid result. The numerical forecasting results for a benchmark dataset, that is, the wind power indicated that the proposed methods especially EEMD-MLP method can improve the performance of the ARIMA and MLP model in forecasting two volatile bench mark data sets. Thus, the proposed hybrid model can introduce as an effective predicting tool for high variation time series.