Multiarea Distributed State Estimation Using a Dual-Stream Neural Network With Collaborative Learning in Smart Grids

As smart grids expand in scale, connectivity, and intelligence, centralized state estimation can no longer meet the demands of complex coordination and robust operation. To address these challenges, this article proposes a distributed state estimation (DSE) solver based on a deep neural network, featuring an edge intelligence and collaborative computing architecture. First, a DSE model is developed based on the proposed dual-stream neural network, embedding node state ranges as constraints to improve cross-region learning of nonlinear measurement–state relationships. Second, during offline training, DSE training begins with intraregion local learning to capture regional characteristics and enhance estimation accuracy. It then proceeds to interregion collaborative estimation through supervised learning with round control, where a global zero vector-based state mapping method unifies partitioned measurement models and a joint loss enforces cross-region synchronization, ensuring boundary consistency and improving global robustness. During online operation, a timestamp-based delay-aware credible interpolation strategy is proposed to dynamically adjust the reliability of cross-region boundary information, while a joint loss function supports real-time fine-tuning to improve estimation accuracy and robustness. Finally, a DSE experimental platform is implemented on multiple ARM-based motherboards, and the effectiveness of the proposed solver is validated through comparative experiments on a partitioned IEEE 118-bus system.