Predict-IT - Forecasting District Heating Loads

Abstract

In the realm of many thermal energy systems, and particularly within district heating networks, heat load forecasts play a pivotal role in optimizing system operation and efficient infrastructure usage. While district heating operators routinely log measurement data, its potential remains underutilized. One essential application of such data is forecasting a network’s heat load based on historical data records. Such forecasts can improve the efficient usage of plant infrastructure and facilitate predictive operational strategies. This paper introduces ”Predict-IT”, a web-based platform designed to standardize the entire forecasting pipeline, making the generation of predictions largely independent of expert knowledge. The Predict-IT platform is powered by a state-of-the-art long short-term memory (LSTM) based neural network algorithm which only requires very little inputs (measured heat load and ambient temperature) to deliver satisfying forecasting accuracy, even a couple of days ahead. The prediction algorithm is validated on two data sets from local Austrian district heating networks, showing the general applicability of the LSTM-based neural network, given an appropriate set of hyperparameters. The Predict-IT platform simplifies the process of forecasting heat loads into a few discrete steps: data upload, algorithm training, heat load forecast generation, and visualization of forecasts. The source code will be open-source, and deployment and installation will be facilitated by an easily installable Docker solution.