Effective Adam-Optimized LSTM Neural Network for Electricity Price Forecasting

摘要

Electrical energy, considered to be a clean energy source, has made a significant contribution to humanity. To make better use of electric energy, great efforts have been paid by electricity market researchers and practitioners on electricity price forecasting. Long short-term memory (LSTM), a type of recurrent neural network, performs well in many areas, such as language modeling and speech recognition. However, the performance of applying the LSTM model to process time series and nonlinear regression problems is not so satisfactory. Stochastic gradient-based optimization has core practical importance in many scientific and engineering fields. Adam, a method for efficient stochastic optimization, has combined the advantages of two popular optimization methods: AdaGrad and RMSProp, it makes LSTM model perform even better. In this study, two examples were listed to verify the performance of the Adam-optimized LSTM neural network, and the dataset from New South Wales of Australia were adopted to illustrate the excellence of model. The results show that the proposed model can significantly improve the prediction accuracy.