Search

Renewable energy penetration increases the power grid's operational uncertainty, threatening the economic effectiveness and reliability of the grid. In this article, we examine how uncertainty affects unit commitment (UC), a classical electricity market procedure. Stochastic programming has helped handle uncertainty for UC and performed well with distribution knowledge, but the lack of such information in practice deteriorates the effectiveness. Such a dilemma becomes more pronounced when dealing with joint chance constraints solely based on samples. To address this issue, we introduce statistical feasibility into UC and develop robust sample-based algorithms employing appropriate uncertainty sets to hedge uncertainty without distribution dependence. We also propose a learn-to-optimize acceleration method to convexify UC. Furthermore, we construct an optimization kernel to boost computational efficiency.

The high penetration of renewable energy brings significant uncertainty to the power grids. Taking economic dispatch (ED) as an example, the inaccurate prediction of renewable energy generations dramatically increases the dispatch cost and risks the power grid's reliable operation. The accurate distribution knowledge of the renewable generations enables modeling the ED as stochastic programming with joint chance constraints, which various classical methods can tackle. However, in practice, such distribution knowledge is inaccessible, and we can only observe samples from some unknown distribution. This makes conducting effective ED solely based on the observed samples challenging. It is particularly true when we need to handle the joint chance constraints. To tackle these challenges, we introduce the notions of statistical feasibility and statistically feasible ED to guarantee the satisfaction of the joint chance constraints. Specifically, we first propose a sample-adaptive robust optimization (RO) to decouple the joint constraints. We then identify that the inaccurate uncertainty set leads to RO's conservativeness, and then reconstruct the constraint-specific uncertainty sets. We design the corresponding sample-adaptive reconstruction-based RO (ReconRO) based on the reconstructed uncertainty sets to further enhance the ED's effectiveness.

Transitioning to a low-/zero-carbon energy ecosystem requires a thorough and accurate understanding of how energy is consumed on the demand side. To achieve this goal, user profiling has become a crucial data analytical tool to understand consumers' energy activities and characterize the patterns/trends of different consumers. Though useful, user profiling remains a challenge due to the high irregularity and volatility inherited in consumer energy activities. Moreover, traditional user profiling requires time-series load consumption data to be collected and processed in a centralized server, which increases processing budgets and privacy leakage risks. To address these challenges, we propose a federated shift-invariant dictionary learning clustering approach to enable distributed and computationally efficient user profiling. The proposed approach elucidates the characteristics of each user by decomposing its time-series load consumption data into a linear combination of electricity consumption patterns. Furthermore, with the aid of a federated learning framework, the proposed approach allows most user profiling steps to be completed locally. Simulation studies based on real-world dataset show that compared with conventional federated clustering methods, the proposed approach is capable of better revealing the patterns behind the load series data, which improves the accuracy and effectiveness of user profiling.

Sharing economy is believed to be able to improve the efficiency of the electricity sector by sharing the underutilized electricity storage. However, this is a delicate task because the optimal decision making often requires the distributional information of demand, which is often hard to obtain and characterize. Hence, the decision making has to rely on the historical samples. While the literature has investigated the sample-based storage control, little is known about historical samples’ theoretical impact on the storage sharing performance. To this end, we theoretically conduct the performance analysis for the sample-based storage sharing decisions. To highlight the impact of the samples, we consider two kinds of historical samples: one is pure demand information, and the other is additionally associated with external feature information. We investigate how these two kinds of samples enable the decision making with a performance guarantee. We further compare the theoretical performance bounds with the empirical results, indicating the effectiveness of the proposed methods and the tightness of the derived bounds. Our theoretical analysis provides a new perspective for efficient demand sample collection and cost-effective sharing mechanism design.