Statistical Approximations of Seasonal Peak Loads for Commercial Areas in South Africa

Load forecasting is important process, which ensures that power consumption at the consumer end, is effectively delivered. The selection of an appropriate statistical distribution for a dataset is one of the approaches required to analyse load profiles. Typically, probability distributions associated with the daily peak load measurements is often of interest to Forecasters. In this paper, the daily peak load at a commercial location in Johannesburg, South Africa is examined on a seasonal cyclical basis. The measurement, undertaken over a period of about two years, between 2019 and 2020, is classified into four seasons: summer, autumn, winter, and spring. Two probabilistic models, lognormal and normal distributions, are applied to approximate the behaviour of daily peak loads. Results from this investigation suggest that normal and lognormal distributions are suitable models for understanding daily peak load profiles during summer and spring seasons. However, both autumn and winter seasons are found to exhibit a different trend, for which these two distributions are found as unstable fits. Further observation suggests that the annual distributions of the peak loads at this location is multimodal in behaviour.