Day-ahead aggregated load forecasting based on household smart meter data

Abstract

Smart meters provide much energy consumption information at the residential level, making it possible to improve short-time load forecasting accuracy by identifying more specific load patterns for each consumer type. This paper proposes a day-ahead load forecasting approach that uses the smart meter data aggregated by residential customers’ power consumption characteristics. First, the long-term trend information and daily fluctuation information are extracted from the residential load time series. According to the load characteristics reflected by the daily load fluctuation information, the residential consumers are clustered into several groups using the K-means algorithm. The non-linear autoregressive neural network is used to forecast each cluster of consumers to capture their specific load patterns. Finally, the aggregated load at the system level is obtained by combining each cluster’s forecasting results. The proposed method’s forecasting performance is evaluated on the data from Irish household customers’ actual smart meters. The outcomes show that the suggested approach can improve forecasting performance under a specific clustering scale.

Keywords

• Aggregated load forecasting
• Smart meter
• K-means
• Neural network