Coordination of networked microgrids (MGs) offers a promising solution to utilize distributed resources flexibilities and accommodate renewable energy. This paper studies real-time coordination of distribution system operation (DSO) and MGs considering multivariate uncertainty. Current researches suffer from inadaptability to dynamic system uncertainty, extensive iterations, and dependence on prediction. To fill these gaps, A novel multi-agent learning based stochastic dynamic programming (MASDP) is proposed to obtain the optimal policy for entities. Specifically, transactive energy control (TEC) mechanism, which requires only market-based information interactions, is employed to facilitate coordination between MG and DSO. A data-driven offline self-learning is proposed for entities to learn how to manage resources in response to system uncertainty. After sufficient offline learning, online operation of MASDP can be run in both non-iterative and iterative manners, by which near-optimal/optimal real-time solutions of DSO and MGs can be given sequentially and distributedly. Numerical comparisons with state-of-art policies and TEC algorithms verify the optimality, efficiency, adaptability, and scalability of MASDP.

With the rapid development of economy and technology, large-scale integrated energy buildings account for an increasing proportion of urban load. However, the randomness of EV owner behaviors, electricity price and outdoor temperature have brought challenges to the energy management of integrated energy buildings. This paper proposes a stochastic dynamic programming-based online algorithm to address the energy management of integrated energy buildings with electric vehicles and flexible thermal loads under multivariate uncertainties. First, an online energy management framework is established, which is further formulated as a multi-stage stochastic sequential decision-making problem. To address the complexities of the problem, a novel stochastic dynamic programming is employed to develop a distribution-free, computationally efficient, and scalable solution. By using extensive training samples, the algorithm is trained offline to learn how to deal with multivariate uncertainties and get the approximate optimal solution, which no longer depends on intraday forecast information. Numerical tests demonstrate the effectiveness of the proposed algorithm compared with other online algorithms in terms of optimality and computation efficiency.

Efficient fault diagnosis of power transformer can effectively ensure the safe and stable operation of power system. Considering that the effect of traditional Random Forest (RF) is seriously affected by the initial parameters, this paper proposes a RF algorithm based on Grey Wolf Optimization (GWO-RF) to improve the accuracy of transformer fault identification. The algorithm uses GWO to optimize the total number of decision trees and the depth of the maximum decision tree, effectively balancing the accuracy of the RF model with the ability to generalize. In this paper, the Dissolved Gas in oil is taken as the fault characteristic quantity, and 335 sets of data are used to form the fault set. Three different data sets are verified by GWO-RF, traditional RF and IEC three ratio method. The experimental results show the effectiveness of GWO-RF method.

Community energy sharing is an important way to encourage prosumers energy exchanges and exploit distributed resource flexibility. Previous researches suffer from low generalization, privacy leaking, heavy communication and computational burden. In this paper, a behavior-based and fast convergence energy sharing mechanism based on prosumers’ response information is proposed to fill these gaps. Specifically, a continuous description of prosumers’ behavior (CDPB) is applied to characterize their optimal response in a small price range. And the distributed algorithm based on response constraints (DBRC) is applied to achieve energy sharing with convergence and optimality, where only price and CDPB information are exchanged. Then, DBRC with accelerated convergence (DBRC-A) is proposed to reduce the time complexity and realize fast convergence. The efficiency, scalability, practicality of DBRC-based energy sharing mechanism are verified by numerical simulations on communities with different prosumers.

Energy sharing can activate prosumer flexibility to support power system operations. However, in practice, communication delays are inevitable and seriously threaten the real-time energy sharing equilibrium. While many asynchronous methods aim to address this, they ignore the learnability of prosumer’s response and often struggle to balance convergence speed and stability. To tackle this issue, this paper proposes a novel asynchronous algorithm based on best response (BR) learning. First, we demonstrate that the prosumer’s best response in ADMM-based energy sharing can be represented as the piecewise linear function. Besides, we propose a batch-fitting method to learn the BR function with high accuracy and low computational cost. Building on this, we develop an asynchronous ADMM algorithm with a practical error criterion to ensure strict convergence of energy sharing under severe delays. Numerical studies indicate that the proposed algorithm can achieve faster convergence with fewer iterations and greater reliability across various energy sharing scales and asynchronous rates than existing asynchronous algorithms.