| 摘 要： |
Producing the accurate forecasting of solar radiation time series is a difficult assignment. To accomplish it, this paper employs the wavelet artificial neural network model for long-term forecasting of daily global solar radiation. To achieve this goal, observed time series of solar radiation data from four stations of Nebraska State (central United States) with different altitudes were used for multi-time-step ahead forecasting on time scales from 1 to 28 days. To determine the appropriateness of the models, four statistical indices were used. The performance of the proposed hybrid model was also compared to the artificial neural network. The results show that using wavelet transform as a data preprocessing method improved the efficiency of the artificial neural network model. The accuracy of the wavelet artificial neural network technique for different mother wavelets and various combinations of input sets were evaluated in this study. Results indicated that the performance of the model is dominated by input sets and mother wavelets. However, the Coiflet family of wavelets granted the best overall performance for 1-day-ahead forecasting. Investigating the results of 1-day-ahead forecasting also revealed that 1-week time lag and 6-day time lag presented the best performance among all input sets for artificial neural networks (ANN) and Wavelet-ANN models respectively. Moreover, the use of 1-year time lag was found to be more effective for forecasting daily global solar radiation in the regions with higher altitudes. This result underlines the importance of investigating the abilities of Wavelet-ANN in different climatic and geographical conditions. |
在线留言
联系我们
所长信箱