31 December 2019 in IEEE Transactions on Power Systems
Power demand forecasting with high accuracy is a guarantee to keep the balance between power supply and demand. Due to strong volatility of industrial power load, ultra-short-term power demand is difficult to forecast accurately and robustly. To solve this problem, this paper proposes a Long Short-Term Memory (LSTM) network based hybrid ensemble learning forecasting model. A hybrid ensemble strategy---which consists of Bagging, Random Subspace, and Boosting with ensemble pruning---is designed to extract the deep features from multivariate data, and a new loss function that integrates peak demand forecasting error is proposed according to bias-variance tradeoff. Experimental results on open dataset and practical datasets show that the proposed model outperforms several state-of-the-art time series forecasting models, and obtains higher accuracy and robustness to forecast peak demand.
1 February 2020 in Applied Energy
23 January 2025 in IEEE Transactions on Cybernetics
The coordination of distributed energy resources (DERs) within virtual power plants (VPPs) is expected to generate significant economic benefits and enhance the operational stability of modern power systems. However, achieving massive coordination of heterogeneous and uncertain DERs remains a challenge in current research. To address this issue, this article proposes a novel bi-level optimization approach based on mean-field reinforcement learning (MFRL) to enable the coordination of massive DERs in VPPs. The problem is decomposed into multiple subproblems: the upper-level subproblem models power dispatch among integrated energy systems (IESs) in response to coordinated demand, while a series of lower-level subproblems determine the operational schemes of DERs within individual IESs. Considering the large decision space, an MFRL algorithm with fast Shapley credit allocation is developed to efficiently solve the upper-level optimization. Meanwhile, the lower-level subproblems are formulated as small-scale mixed-integer linear programming (MILP) problems, addressing the difficulties caused by IES heterogeneity in applying mean-field approximation. Simulation results show that the proposed approach significantly improves convergence speed and reduces the global cost of VPP operation, especially in massive-scale scenarios. In test scenarios ranging from 10 to 500 agents, the proposed bi-level optimization approach improves the objective by 4.8%–26.6%, compared to the advanced baseline method.
27 May 2024 in IEEE Transactions on Transportation Electrification
Electric vehicles (EVs), as flexible distributed energy storage units, play an important role in maintaining the balance between supply and demand in modern power systems. However, the stochasticity of their arrival times and the uncertainty of the demand for charging will affect the accuracy of any scheduling. Therefore, an Adaptive Robust Optimization model considering the uncertainty of EV charging demands is proposed. Firstly, an adaptive data-driven uncertainty set is constructed to address the issue of unknown probability distribution associated with the demand for charging, effectively capturing the multimodal characteristics of the uncertain data. Secondly, to eliminate extreme scenarios during real-time scheduling, a scenario reduction strategy based on the Jensen-Shannon divergence is proposed. The strategy effectively reduces the model’s conservatism. Thirdly, to prevent grid overload resulting from the large-scale integration of EVs, a hybrid demand response strategy is proposed, which guides EV users to charge in an orderly manner. A new column-and-constraint generation algorithm with a parallel alternating optimization procedure (C&CG-PAOP) is employed to obtain the optimal solution. Experimental results demonstrate the effectiveness of our method in reducing the operating costs of charging stations.
5 May 2024 in Journal of Building Engineering
21 December 2023 in IEEE Transactions on Instrumentation and Measurement
Non-intrusive load monitoring (NILM) enables the real-time monitoring and data analysis of energy consumption, providing more accurate, efficient and intelligent data support for energy management. However, traditional NILM models can only identify known load (KL) presented in the training dataset; they cannot perform unknown load (UKL) detection, and they have problems with feature selection, including information loss and limitations in capturing load signal information, which make them less suitable for the user-side. In this paper, a new method based on bi-modal fusion and one-class classification is proposed for NILM. The bi-modal features include two-dimensional image features and multi-dimensional sequence features. To enrich the feature representation, the bi-modal features are fused by long short-term memory (LSTM) network. To address UKL detection, a simple and lightweight one-class (OC) model based on deep support vector data description (DSVDD) is innovatively proposed. The proposed model is also validated on multiple publicly available datasets, demonstrating its significant advantages and applicability both on KL identification and UKL detection.