27 February 2020 in IEEE Access
Electric load data are essential for data-driven approaches (including deep learning) in smart grid, and advanced smart meter technologies provide fine-grained data with reliable communications. Despite the recent development of smart metering devices, however, missing data still arise due to unexpected device power off, communication failure, measuring error, or other unknown reasons. In this paper, we investigate a deep learning framework for missing imputation of smart meter data by leveraging a denoising autoencoder (DAE). Then, we compare the performance of the proposed DAE with traditional methods as well as other recently developed generative models, e.g., variational autoencoder and Wasserstein autoencoder. The proposed DAE based imputation shows significantly better results compared to other methods in terms of root mean square error (RMSE) by up to 28.9% for point-wise error, and by up to 56% for daily-accumulated error.
21 August 2024 in IEEE Transactions on Sustainable Energy
In recent years, power systems integrated with distributed energy resources (DERs) have been considered to mitigate climate change. However, this makes power systems even more uncertain and complex, so uncertainty-aware accurate forecasting needs to be considered for the massive penetration of renewable energy. To this end, we propose a scalable and missing-insensitive framework for probabilistic multi-site photovoltaic (PV) power forecasting, specifically focused on large-scale PV sites and space-time missing data. By leveraging the graph neural network (GNN), the proposed scalable graph learning mechanism with random coarse graph attention and probabilistic spatio-temporal learning performs efficiently for large-scale PV sites in terms of forecasting accuracy and model training complexity. At the same time, our framework adaptively imputes the missing PV data in the space and time domain, respectively. Ablation study results demonstrate that our framework is effective for extracting complex spatial-temporal features across large-scale PV sites. Under extensive experiments, our framework shows 7 $-$ 10% and 6 $-$ 25% improvement on average for over 1600 PV sites and three types of space-time missing data, which ensures accurate and stable forecasting.
26 April 2024 in IEEE Transactions on Industrial Informatics
In this article, we propose FedAND, a unified federated learning optimization algorithm, to tackle client drift and server drift issues under partial client participation. Federated learning is gaining popularity due to privacy concerns and mobile computing, but it still faces challenges due to heterogeneous and distributed data. FedAND leverages consensus alternating direction method of multipliers (ADMM) and resolves the server drift caused by the server state in the global update. Under partial participation, we prove that FedAND preserves the strong convergence properties of ADMM while suppressing the server drift, which in turn reduces the client drift and thus achieves better convergence. Our empirical results demonstrate superior performance compared to other methods such as FedProx, FedADMM, FedPD, and FedDyn in diverse scenarios of statistical and system heterogeneity under partial client participation.
28 May 2024 in IEEE Transactions on Power Systems
Conventional solvers are often computationally expensive for constrained optimization, particularly in large-scale and time-critical problems including AC optimal power flow (OPF) problems. While this leads to a growing interest in using neural networks (NNs) as fast optimal solution approximators, incorporating the constraints with NNs is challenging. In this regard, we propose deep Lagrange dual with equation embedding (DeepLDE), a framework that learns to find an optimal solution without using labels. To ensure feasible solutions, we embed equality constraints into the NNs and train the NNs using the primal-dual method to impose inequality constraints. The equality constraints correspond to power flow equations, and the inequality constraints include the operational limits of generators and transmission lines. We prove the convergence of DeepLDE and show that the previous primal-dual learning method cannot solely ensure equality constraints without the help of equation embedding. Simulation results on non-convex and AC-OPF problems show that the proposed DeepLDE achieves the smallest optimality gap among all the NN-based approaches while always ensuring feasible solutions. Furthermore, the computation time of the proposed method is up to 35 times faster than the baselines in solving constrained non-convex optimization, and/or AC-OPF.