Consumer-oriented heat consumption prediction
Maciej Grzenda · 2012
Abstract: The advent of modern low-cost monitoring and wire-less transmission systems results in unprecedented availability of measurement data potentially available in near real-time mode. In particular, some of the remote meter reading systems can be used to collect data on an hourly or even sub-hourly basis. This allows the utility companies to model and predict consumer behaviour more precisely than before. In this study, the way the monitoring data can be used to model heat consumption at individual premises supplied with heat by a district heating system, is proposed. The proposed algorithm is based on customer partitioning used to devise a number of group models serving the needs of consumers sharing similar consumption profiles. Self-organising maps are used to group averaged long-term time series, while the short-term time series provide a basis for group prediction models. Particular atten-tion has been paid to a wider hydraulic modelling perspective, as the application of the proposed method to provide assumed demand for hydraulic model of a district heating system is envisaged. The approach has been validated using a real data set. Results show that in spite of a limited number of monitored consumers, group predic-tion models, constructed using the algorithm proposed in this study, can significantly reduce demand prediction error.