Privacy-preserving short-term load forecasting with convolutional neural networks using secure computation

This paper aims to explore the use of privacy-preserving computation and convolutional neural networks for short-term electricity load forecasting, with the goal of addressing the challenges of privacy leakage in data sharing. To achieve this, techniques such as homomorphic encryption are adopted to encrypt historical load data, protecting the privacy of the electricity load data, while the convolutional neural network is utilized to extract the data's characteristics and predict future electricity loads through a CNN model. This approach ensures the privacy and security of data while simultaneously improving forecasting accuracy. In this study, electricity load data from a city in Southwest China is used for experimentation. The results show that our proposed method can achieve high prediction accuracy (up to 95%) while preserving data privacy. Therefore, the results of this research are of significant importance to the reliability of the power system and privacy protection research.