Electrical Load Forecasting to Plan the Increase in Renewable Energy Sources and Electricity Demand: a CNN-QR-RTCF and Deep Learning Approach

Abstract

This research develops a new electric charge prediction method by using Convolutional Neural Networks with Quantile Regression (CNN-QR) combined with the Rainbow Technique for Categorical Features (RTCF) and using Deep Learning to create layers for the architecture of the neural network. This combination captures both local and global interdependencies within the load data. In particular, RTCF employs advanced natural language processing (NLP) techniques to convert several important categorical features into a single feature called “category,” which is tailored to the various attributes of the Baja California Sur system, in Mexico, taking into consideration climatic conditions, local circumstances and a significant increase in energy consumption. Furthermore, this research compares CNN-QR with classical quantile regression and shows that CNN-QR works better at capturing sophisticated load patterns and producing probabilistic estimates. The above methodology uses hourly data from January 2019 to October 2020. The results obtained provide valuable information for policy formulation in the energy sector, specifically in the areas of load forecasting and expansion of renewable energy and electricity consumption. Finally, it is worth mentioning that the utilization of CNN-QR with RTCF not only improves the accuracy of load forecasting, but also provides a strategic framework for energy management and resource planning in dynamic energy systems, which demonstrates its substantial importance for market participants and authorities, as well as regulators.

Keywords

• Electric Load Forecasting, Convolutional Neural Networks, Quantile Regression, Rainbow Technique for Categorical Features, Deep Learning