A Hybrid Federated Learning Framework With Dynamic Task Allocation for Multi-Party Distributed Load Prediction
Haizhou Liu, Xuan Zhang, Xinwei Shen, Hongbin Sun, Mohammad Shahidehpour
Abstract
In distributed load prediction problems, training datasets on load consumption and load-related features are scattered within various districts and parties. To fully exploit the underlying patterns of these decentralized data, federated learning is needed as a privacy-preserving distributed load prediction scheme. However, current federated learning frameworks address either the horizontal or vertical separation of data, and tend to overlook the case where both are present. To this end, we propose a hybrid federated learning framework for multi-party distributed load prediction. We seamlessly integrate horizontal and vertical federated learning to address the hybrid scenario where features are scattered within local heterogeneous parties and samples are scattered within homogeneous districts. Moreover, we develop an optional dynamic task allocation scheme to improve fairness among parties and boost training efficiency. A follow-up case study is presented to justify the necessity of adopting the proposed framework. The advantages of the framework in convergence of training, losslessness, data security, accuracy, generalization performance, information fairness and computational efficiency are also theoretically and/or experimentally confirmed.