No abstract

Fossil fuels cause environmental and ecosystem problems. Hence, fossil fuels are replaced by nonpolluting, renewable, and clean energy sources such as wind energy. The stochastic and intermittent nature of wind speed makes it challenging to obtain accurate predictions. Long short term memory (LSTM) networks are proved to be reliable models for time series forecasting. Hence, an improved deep learning-based hybrid framework to forecast wind speed is proposed in this paper. The new framework employs a stacked autoencoder (SAE) and a stacked LSTM network. The stacked autoencoder extracts more profound and abstract features from the original wind speed dataset. Empirical tests are conducted to identify an optimal stacked LSTM network. The extracted features from the SAE are then transferred to the optimal stacked LSTM network for predicting wind speed. The efficiency of the proposed hybrid model is compared with machine learning models such as support vector regression, artificial neural networks, and deep learning based models such as recurrent neural networks and long short term memory networks. Statistical error indicators, namely, mean absolute error, root mean squared error, and R2, are adopted to assess the performance of the models. The simulation results demonstrate that the suggested hybrid model produces more accurate forecasts.

No abstract

ABSTRACT

Wind energy, one of the greatest progressing renewable energy sources, becomes more significant for sustainable development and environmental protection. Its intermittent nature makes accurate and reliable predictions very challenging. Currently, hybrid models are extensively employed for wind speed forecasting and have been established to perform superior to traditional single forecast models. Hence, in this paper, a hybrid multi-step wind speed forecasting framework that combines the features of Wavelet Transform (WT), Long Short Term Memory (LSTM), and Support Vector Regression (SVR) is proposed. The prediction accuracy of the model is enhanced by denoising the dataset using wavelet transforms, which decomposes the data into low and high-frequency sub-series. The low-frequency sub-series is forecasted using LSTM network, and the high-frequency sub-series using SVR. Each forecasting outcomes are summed up to get the final forecasting results. The simulation results reveal that the forecast accuracy has significantly improved for the proposed wavelet-based hybrid model.