High-resolution peak demand estimation using generalized additive models and deep neural networks

This paper covers predicting high-resolution electricity peak demand features given lower-resolution data. is a relevant setup as it answers whether limited higher-resolution monitoring helps to estimate future loads when the data no longer available. That question particularly interesting for network operators considering replacing predictive models due economic considerations. We propose predict half-hourly minima and maxima of (every minute) load while model inputs are lower resolution (30 min). combine ptedictions generalized additive (GAM) deep artificial neural networks (DNN), which popular in forecasting. extensively analyze prediction models, including input parameters’ importance, focusing on load, weather, seasonal effects. The proposed method won competition organized by Western Power Distribution, British distribution operator. In addition, we provide rigorous evaluation study that goes beyond frame models’ robustness. results show methods superior benchmark concerning out-of-sample root mean squared error (RMSE). holds regarding month supplementary study, an additional eleven months. Overall, our combination reduces RMSE 57.4% compared benchmark.