A hybrid electric load forecasting model based on decomposition considering fisher information

Accurate and efficient short-term load forecasting plays an important role in the stable operation of power grids and the economic operation of society. Among those factors that effect the electric load, meteorology is one of the most important factors that changes electric load fluctuation. This research innovatively adds the influence factor of multi day continuous meteorological conditions to the daily peak load forecasting method of the power grid, the forecasting method consists of four modules: feature selection, time-frequency signal decomposition, deep learning deterministic model forecasting, and probabilistic model forecasting. Firstly, screening meteorology factors that related to electric load, using fisher information method to quantify the continuous meteorological effect on electric load. Secondly, this study used VMD to decompose the electric load into discrete sub-modes and then expand them into a sequence. Thirdly, deep learning neural network LSTM was used to predict subsequences and reconstruct them. Time-frequency signal decomposition methods can reduce the impact of noise and distinguish the features of the electric load time series. Finally, the deterministic load forecasting results were extended to probabilistic forecasting results by using Gaussian process regression. To validate the performance of the proposed method, ten different load forecasting methods were demonstrated in this study. These simulation results with respect to load forecasting of three different data sets in a certain Province of China and Australia. Simulation experiments shown that the deterministic model proposed has a maximum improvement of 74.32% and a minimum improvement of 0.55% compared to the comparative model, on the indicator RMSE. The probability prediction results shown that the prediction interval of the proposed model covered most of the observed values of the load, indicated that the proposed model can not only provides the accurate deterministic electric load forecasting data, but also provides the reliable probabilistic electric load forecasting results.