Auto-Regressive and Neural Network Models for Weather-Informed Load Forecasts

In this paper, we propose auto-regressive and neural network models to forecast load profiles based on weather measurements without a need for manual calibration. The accuracy of the auto-regressive model is compared to the accuracy of the predictions of the neural network model with optimized parameters. The methods are compared using a test set of thirteen loads, each with hourly load time-series data and an additional data-set containing twenty weather features also sampled hourly. The results of the study show an average of a ten percent increase in accuracy when using the proposed auto-regressive model compared to the optimized neural network strategy.