A modified deep residual network for short-term load forecasting

The electrical load has a prominent position and a very important role in the day-to-day operations of the entire power system. Due to this, many researchers proposed various models for forecasting load. However, these models are having issues with over-fitting and the capability of generalization. In this paper, by adopting state-of-the-art of deep learning, a modified deep residual network (deep-ResNet) is proposed to improve the precision of short-term load forecasting and overcome the above issues. In addition, the concept of statistical correlational analysis is used to identify the appropriate input features extraction ability and generalization capability in order to progress the accuracy of the model. Two utility (ISO-NE and IESO-Canada) datasets are considered for evaluating the proposed model performance. Finally, the prediction results obtained from the proposed model are promising as well as accurate when compared with the other existing models in the literature.

Automated summary

In this paper, a modified deep residual network (deep-ResNet) is proposed to improve the precision of short-term load forecasting by adopting state-of-the-art deep learning techniques and overcome the issues of over-fitting and generalization. The concept of statistical correlational analysis is used to identify appropriate input features extraction ability and generalization capability, leading to improved accuracy of the model. Two utility datasets are used to evaluate the proposed model performance, and the results show promising and accurate prediction compared to existing models in the literature.