Integrated Multi-Horizon Power and Energy Forecast for Aggregated Electric Water Heaters

Abstract

Domestic electric water heaters (DEWH) are common residential loads and good candidates for direct load control owing to their energy storage capacity. Power prediction for aggregated DEWHs is of great importance for controlling them to follow desired load profile without negative impact to the normal end use. Accurate energy prediction provides extra information for optimization during specified time periods. However hysteresis band of thermostats and randomness of the end-users' behavior make prediction difficult, especially when the aggregation size is small. This paper constructs an integrated forecasting framework suitable for aggregated loads that comprises both multi-horizon power prediction and energy forecast. K-mean and wavelet decomposition-based neural networks are employed for power prediction. To mitigate forecasting error accumulation when calculating energy according to predicted power consumption, an error reduction algorithm, based on Markov models with an additional compensation coefficient is reported to refine energy prediction. The aggregation effects on performance are analyzed based on data from 95 DEWHs. The experimental results demonstrate the effectiveness of proposed algorithms. Mean absolute errors are about 4.6kW to 6.1kW for 5 minutes to 6 hours-ahead power forecasts of the 95 DEWHs, and nearly 60% improvement is achieved for 6-hour energy forecasts with the proposed error reduction algorithm.