Hybrid Neuro-Fuzzy Modeling for Electricity Consumption Prediction in a Middle-Income Household in Gauteng, South Africa: Utilizing Fuzzy C-means Method

Machine learning (ML) models, renowned for their precision, are increasingly utilized in forecasting electricity consumption, a crucial aspect for empowering utilities with insights to optimize system performance in terms of productivity and efficiency. The limitations of conventional methods have fueled the adoption of ML-driven approaches for modeling energy consumption. This study introduces a fusion of fuzzy c-means (FCM)-clustered adaptive neuro-fuzzy inference system (ANFIS) with evolutionary algorithms (EAs) to predict energy consumption in middle-income households, focusing on Gauteng province, South Africa. FCM clustering was selected for its ability to enhance computational efficiency and speed. The model's simulation involved fine-tuning ANFIS structure through particle swarm optimization (PSO) and genetic algorithm (GA), with the optimal model selected through performance evaluation utilizing relevant statistical metrics such as root mean square error (RMSE), mean absolute error (MAE), coefficient of root mean square (CVRMSE), mean absolute deviation (MAD), and coefficient of variation (RCoV). The best-performing model obtained in the initial scenario underwent further enhancement by integrating a scaling factor (SF) into the GA. Subsequently, the modified GA-ANFIS-FCM (with 2 clusters) yielded the best performance, with values of RMSE, MAD, MAE, RCoV, and CVRMSE at 0.1715, 0.1272, 0.1251, 0.3619, and 73.5171, respectively. This study highlights the potential of the optimal model to serve as a dependable tool for accurately predicting energy consumption and forecasting.