Federated Deep Learning Multiple Households Energy Consumption Forecasting Considering Data Privacy-Preserving

Energy consumption forecasting is vital for the development of smart grids. Currently, much of the research in this field concentrates on forecasting energy consumption at the individual household level. Few methods focus on forecasting energy consumption for multiple households, and those that do exist suffer from poor accuracy, ignore user privacy, and require high communication resources. This paper proposes a method for forecasting energy consumption across multiple households, which combines federated learning(FL), bidirectional long short-term memory(BiLSTM) network, and attention mechanism(AM). The bidirectional context of the input sequence is captured using forward and backward long short-term memory, while the attention mechanism is employed to weigh inputs and emphasize key information, thereby improving prediction accuracy. The federated learning framework safeguards against privacy breaches of household energy consumption data during the training process. Experimental results show that the proposed method can implement simultaneous prediction of multiple households energy consumption with a consistent model via a transmission control protocol, with reduced mean absolute error(MAE), mean absolute percentage error(MAPE), and root mean square error(RMSE) compared with traditional household energy consumption forecasting methods. Additionally, it has improved prediction accuracy, reduced computing time, and enhanced generalization capability.