Comparative Analysis of Feedforward Neural Network and LSTM Model for Load Forecasting

Load forecasting is crucial and very vital in the management and in the control of power systems for appropriate distribution. There is need to have an efficient and accurate system to avoid wastage of energy and at the same time to forecast the demand in order to ensure that there is a reliable supply. Efficient model can generate better result and also enhance reliability. This paper aims at comparing between two neural network-based models, namely LSTM network and feedforward neural network to determine the most appropriate model to use in load forecasting. These models are developed with data of a power station in Beijing that contains temperature, humidity, wind speed and load data. The outcome of each model is assessed using ‘mean squared error’ (MSE), ‘coefficient of determination’ (R2) and the ‘mean absolute error’ (MAE). The results obtained infer that the LSTM model is better than the feedforward neural network in capturing temporal dependencies found in load data to give better predictions.