A Novel Sequence to Sequence Data Modelling Based CNN-LSTM Algorithm for Three Years Ahead Monthly Peak Load Forecasting
Osaka Rubasinghe, Xinan Zhang, Tat Kei Chau, Yau Hing Chow, Tyrone Fernando, Herbert Ho‐Ching Iu
Abstract
Long-term load forecasting (LTLF) models play an important role in the strategic planning of power systems around the globe. Obtaining correct decisions on power network expansions or restrictions based on predictions help substantially reduce the power grid infrastructure costs. The classical approach of LTLF is limited to the usage of artificial neural networks (ANN) or regression-based approaches along with a large set of historical electricity load, weather, economy and population data. Considering the drawbacks of classical methods, this article introduces a novel sequence to sequence hybrid convolutional neural network and long short-term memory (CNN-LSTM) model to forecast the monthly peak load for a time horizon of three years. These drawbacks include, lack of sensitivity to changing trends over long time horizons, difficulty of fitting large number of variables and complex relationships, etc. (Velicer and Plummer, 1998). Forecasting time interval plays a key role in LTLF. Therefore, using monthly peak load avoids unnecessary complications while providing all essential information for a good long-term strategical planning. The accuracy of the proposed method is verified by the load data of “New South Wales (NSW)”, Australia. The numerical results show that, proposed method has achieved higher prediction accuracy compared to the existing work on long-term load forecasting.