Forecasting Long-Term Electricity Demand Time Series Using Artificial Neural Networks

Abstract

Accurate forecast of electricity load are increasingly important. We present a method to forecast long-term weather-dependent hourly electricity load using artificial neural networks. The fully connected dense artificial neural networks with 5 hidden layers and 1,024 hidden nodes per layer are trained using historic data from 2006 to 2015. Input parameters comprise calendrical information, annual peak loads and weather data. The results are benchmarked against the method to forecast electric loads used in the current mid-term adequacy forecasts published by the European Network of Transmission System Operators (entso-e). For validation year 2016, our approach shows a mean absolute percentage error of 2.8%, whereas the common approach as used by entso-e shows an average error of 4.8% using peak load scaling. Further, we conduct forecasts for Germany, Sweden, Spain, and France for scenario year 2025 and assess parameter variations. Our approach can serve to increase prediction accuracy of future electricity loads.