Temperature Prediction in District Heating Systems with cFIR models

Current methodologies for the optimal operation of district heating systems are based on model predictive control. In complement to load forecasts, accurate predictions (up to 12-hour ahead) of the water temperature at critical points of the networks are crucial for meeting constraints related to consumers while minimizing the production costs for the heat supplier. The paper introduces a new forecasting methodology based on a conditional Finite Impulse Response (cFIR) model, for which the model coefficients are replaced by nonparametric or semi-parametric coefficient functions of the water flux at the supply point and of the time of day. This allows for nonlinear variations of the time delays in the FIR model. The coefficients functions can be adaptively estimated with a method that combines local polynomial regression, exponential forgetting, recursive weighted least squares and Tikhonov regularization. Results are given for the test case of the Roskilde district heating system, over a period of more than 6 years. The advantages of the proposed forecasting methodology in terms of a higher forecast accuracy, in terms of its use for simulation purposes, or alternatively for better understanding transfer functions of district heating systems, are clearly shown.