Toward explainable heat load patterns prediction for district heating

Abstract

Heat networks play a vital role in the energy sector by offering thermal energy to residents in certain countries. Effective management and optimization of heat networks require a deep understanding of users' heat usage patterns. Irregular patterns, such as peak usage periods, can exceed the design capacities of the system. However, previous work has mostly neglected the analysis of heat usage profiles or performed on a small scale. To close the gap, this study proposes a data-driven approach to analyze and predict heat load in a district heating network. The study uses data from over eight heating seasons of a cogeneration DH plant in Cheongju, Korea, to build analysis and forecast models using supervised machine learning (ML) algorithms, including support vector regression (SVR), boosting algorithms, and multilayer perceptron (MLP). The models take weather data, holiday information, and historical hourly heat load as input variables. The performance of these algorithms is compared using different training sample sizes of the dataset. The results show that boosting algorithms, particularly XGBoost, are more suitable ML algorithms with lower prediction errors than SVR and MLP. Finally, different explainable artificial intelligence approaches are applied to provide an in-depth interpretation of the trained model and the importance of input variables.

Predict-IT - Forecasting District Heating Loads

“…For time series forecasting, various data-driven algorithm approaches are available, broadly categorized into two groups: Statistical models of the ARIMA/SARIMA class, which model time correlations among input variables using statistical methods with a well-established theoretical foundation, and machine learning algorithms such as boosting and support vector regression (see e.g. [2], [7]). In Predict-IT, a state-of-theart LSTM-based neural network is adopted due to its capability to represent time series data with complex system dependencies [1], [8].…”

“…Forecasting the heat load of district heating plants remains an active research area, as evidenced by recent works such as [1], [2]. While existing literature primarily focuses on creating and refining prediction algorithms, this paper takes a different approach.…”

Enabling Alarm-Based Fault Prediction for Smart Meters in District Heating Systems: A Danish Case Study

“…In contrast, DH research has extensively delved into predictive models for other aspects, such as heat load prediction, as evidenced by studies like [11][12][13][14]. These load predictions can be used to understand the nature of future heat loads for better day-ahead planning in the system from both a production and a grid standpoint to avoid faults due to misconfiguration of set points, which could lead to not enough supply to consumers.…”