Abstract
Electricity load forecasting has seen increasing importance recently, especially with the effectiveness of deep learning methods growing. Improving the accuracy of electricity load forecasting is vital for public resources management departments. Traditional neural network methods such as long short-term memory (LSTM) and bidirectional LSTM (BiLSTM) have been widely used in electricity load forecasting. However, LSTM and its variants are not sensitive to the dynamic change of inputs and miss the internal nonperiodic rules of series, due to their discrete observation interval. In this paper, a novel neural ordinary differential equation (NODE) method, which can be seen as a continuous version of residual network (ResNet), is applied to electricity load forecasting to learn dynamics of time series. We design three groups of models based on LSTM and BiLSTM and compare the accuracy between models using NODE and without NODE. The experimental results show that NODE can improve the prediction accuracy of LSTM and BiLSTM. It indicates that NODE is an effective approach to improving the accuracy of electricity load forecasting.
Subject
Computer Networks and Communications
Reference27 articles.
1. Methods and Models for Electric Load Forecasting: A Comprehensive Review;Hammad;Logist. Sustain. Transp.,2020
2. gil Koo, B., Kim, M.S., Kim, K.H., Lee, H.T., Park, J., and Kim, C.H. (2013, January 4–5). Short-term electric load forecasting using data mining technique. Proceedings of the 2013 7th International Conference on Intelligent Systems and Control (ISCO), Coimbatore, India.
3. Short-Term Load Forecasting With Deep Residual Networks;Chen;IEEE Trans. Smart Grid,2019
4. Bunn, D.W., and Farmer, E.D. (1986). Comparative Models for Electrical Load Forecasting, Wiley.
5. Fallah, S.N., Ganjkhani, M., Shamshirband, S., and Chau, K.w. (2019). Computational Intelligence on Short-Term Load Forecasting: A Methodological Overview. Energies, 12.
Cited by
3 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献