Hydropower Generation Forecasting via Deep Neural Network

With the advances of deep learning, its applications in our daily life have attracted considerable attention from both academics and industry. However, most of the existing works focus on computer vision and natural language processing, while few studies in the real industrial manufactures. The main reasons are that the data resources are difficult to obtain and the relationships between industrial data are too complex to be modeled. In this paper, we propose a deep neural network based approach for hydroelectric power generation prediction, which, to our knowledge, is the first attempt modeling power generation data with the combination of residual neural networks and recurrent neural networks. Furthermore, we consider different grains of the hydropower generation by dividing the data into four-levels, i.e. recent, daily, weekly, and time-series sequences, which can greatly improve the prediction performance. To this end, we employ a multi-information fusion method to fuse the four components (i.e. closeness, period, trend, long-period) predicted results, among which different component is learned with different weights to determine their influence on final hydropower prediction. Experiments conducting on the real hydropower generation prove the effectiveness of the proposed model, which significantly outperform the baselines. We hope this research will open a new perspective of improving data usage in the industry, especially in power generation areas.