DEVELOPMENT OF MULTIFACTORIAL SHORT-TERM POWER LOAD FORECASTING MODEL FOR ENUGU LOAD CENTRE USING ARTIFICIAL NEURAL NETWORK CONCEPT

Abstract

This study presents a multifactorial short-term power load forecasting model for Enugu load center using artificial neural network (ANN) concept. The aim is to improve daily forecasting accuracy by introducing more features such as temperature, income per capita, and load category to the model feature set. Historical load data, temperature data, income per capita, and load category for the months of August 2012 – October 2012 were used. The stages include data collation, data pre-processing, the neural network modeling, the input variables selection, model training, and model testing cum forecasting. The previous day data and previous week data were used as inputs to the ANN model. The modeled network has two layers, i.e., a hidden layer, and an output layer. The number of neurons in the hidden layer can be varied for the different performance of the network, while the output layer has a single neuron. The optimal layer architecture obtained for the data is 50-1. The model was trained with historical load data, together with other factors (temperature, income per capita, and load category). The trained model was able to identify the complex relation in the historical data, and when tested with new set of historical data, it was able to forecast electric load for 24 hours in advance. The performance of the model was analysed in terms of the mean squared error (MSE) in the forecast which gave an average of 1.30% over one week’s forecast. On average, this represents a high degree of accuracy in the load forecast. Further, the ANN model was also trained with only historical load data, and when tested, gave an average of MSE of 2.66% over one week forecast. The results when compared with that of the former shows that the ANN short-term load forecasting model, which employed load data together with other factors (temperature, etc.) performed better than the model which employed load data alone. In conclusion, an ANN model architecture was provided, which is reliable for predicting hourly load on a daily basis for the studied load center in particular and Nigerian networks in general.