Electricity Demand Forecasting using Dual Stream TBATS-CNN-LSTM Architecture

Abstract

The problem is the model's reliability, accuracy, and meaningfulness to convince decision-makers of the actions to be taken when seasonality is one of the features in the findings of the existing deep learning forecasts. The purpose of this chapter is to come up with a novel dual-stream hybrid architecture that is capable of predicting electricity demand and accessing its accuracy levels by benchmarking it with individual architecture model's forecasting accuracy levels using out-of-sample time series. The approach in this chapter uses time series and convolutional neural network (CNN)-based long short-term memory with various configurations to construct a forecasting model for short- to medium-term aggregate load forecasting. The obtained results show that the TBATS-CNN-LSTM-based model has shown high accuracy as compared to the base learner, and the model is optimised with hyperparameter tuning. Only optimally selected time-lag features captured all the characteristics of complex time series in South Africa.