With the development of the economy and the improvement of people’s living standards, electric power load demonstrates a trend of rapid growth. Effective forecasting of power load plays a crucial role in energy management and the normal operation of the smart grid. In traditional machine learning schemes, data is typically collected centrally for training power load forecasting models. However, power data often contains a significant amount of user privacy, and this approach faces the risk of privacy leakage. How to fully utilize the potential value of power load data without compromising user privacy has become a significant challenge. This paper proposes a privacy-preserving federated learning scheme to achieve power load forecasting while protecting user privacy. Our scheme introduces smart contracts and blockchain technology to replace the central server in traditional federated learning architecture. Additionally, it incorporates secret sharing technology to protect the privacy of the intermediate model parameters. We also conduct experiments to verify the feasibility and effectiveness of the scheme.