A novel mid- and long-term time-series forecasting framework for electricity price based on hierarchical recurrent neural networks

This paper proposes an innovative end-to-end time-series decomposition-forecasting framework based on recurrent neural networks called DF-RNN for electricity price. The framework combines an RNN-based decomposition model and an RNN-based forecasting model to perform mid- and long-term time-series forecasting tasks effectively. A hierarchical RNN-based time-series decomposition model is introduced to decompose time series data into trend, seasonal, and residual components. The RNN-based forecasting models generate forecasts for each component series, aggregated to produce the time-series forecast. The DF-RNN model simultaneously optimizes the objective functions of both the time-series decomposition model and the forecasting model, resulting in an optimal time-series forecasting model. The DF-RNN model has a flexible network structure and clear interpretability, making it easy to analyze and understand the results. The effectiveness of the proposed framework is evaluated in a real-world electricity-price forecasting in the European Power Exchange France. The experimental results demonstrate that DFRNN is a promising and effective mid- and long-term time-series forecasting model.