A prior-knowledge-based time series model for heat demand prediction of district heating systems

Time series prediction methods are effective for predicting energy demands of district heating systems. Due to the thermal inertia of systems, historical heat load is usually applied as the most crucial input for prediction. However, it results in lateral errors (phase lags) between actual and predicted heat demand. To address this issue, a hybrid time series model is proposed based on autoregressive integrated moving average model and resistance-capacitance model. To reduce the lateral errors, estimated current and future heat demands are introduced as exogenous inputs of the autoregressive integrated moving average model, and they perform as prior knowledge. To generate estimated heat demand, a piecewise resistance-capacitance model is proposed, which considers various climatic and operating conditions. Lateral error of the hybrid model is decreased by 7.1% compared with conventional autoregressive integrated moving average model. Moreover, the hybrid model retains good post-hoc interpretability, which proves that exogenous inputs can effectively reduce lateral errors.