South African Power Distribution Network Load Forecasting Using Hybrid AI Techniques: ANFIS and OP-ELM

South Africa has been a late participant in the previous industrial revolutions. With the fourth industrial revolution upon us, South Africa cannot afford to be left behind. Artificial Intelligence (AI) and big data, which are at the center of this revolution, are improving humans lives. Load forecasting has been shown to have benefits in power systems maintenance and operation. The study of load foreasting in South African (SA) power distribution networks using AI is limited. This paper presents a comparative study of hybrid AI techniques, adaptive neuro-fuzzy inference systems (ANFIS) and OP-ELM, in South African distribution load forecasting. This is achieved with a case study on a real South African large power consuming substation using three performance measures, root mean square error (RMSE), mean absolute error (MAE) and symmetric mean absolute percentage error (sMAPE). The paper also investigates the impact of cleaning up loading data and the inclusion of temperature on the two techniques' models' performance. OP-ELM achieved the lowest error in comparison to ANFIS, achieving an sMAPE of 3.83%, MAE of 5.32% and RMSE of 6.52%. Hybrid AI techniques can thus be used to forecast load in South African distribution networks. This application of AI can lead to costs savings for S.A. power utilities.