Heating, ventilation, and air conditioning (HVAC) systems constitute a large proportion of building energy consumption and provide considerable potential for power grid regulation. While the HVAC power consumption forecasting task is generally straightforward with sufficient historical data, it becomes challenging when dealing with scarce data. Such situation is common in cases of intermittent data collection or early system implementations, where precise forecasting is required despite limited data available. Considering accessible datasets from nearby or similar HVAC systems through energy management systems, this paper proposes an adaptive transfer learning framework to tackle this issue. Specifically, the framework leverages diverse source domains, employing model-level regularizers to quantify domain discrepancies and an adaptive parameter regulation mechanism to dynamically align source domains with the target domain. Embedded within the framework, a unique deep learning architecture with attention mechanisms is proposed, capable of identifying complex temporal patterns and hierarchical features in HVAC systems. Experiments on public HVAC datasets demonstrate the generalization, accuracy and robustness of our methodology under diverse data-scarce scenarios.
24 May 2021 in IEEE Transactions on Instrumentation and Measurement
In this article, we introduce the Intever public database for arcing event detection (IAED) along with novel feature extraction methods and an end-to-end deep neural network. First, we describe the details of arc simulation process and database specifications. Second, we handcraft a set of 29 efficient arcing features, including novel nonlinear dynamic features extracted from reconstructed phase space. Third, we propose a customized multiscale convolutional neural network (Multi-Scale CNN) model. Finally, we present the benchmark testing results on six IAED cohorts and 25 load settings and analyze the time complexity of each algorithm. Experimental results show that the highest F1 scores on high-resolution cohort and low-resolution cohort are achieved by Multi-Scale CNN at 94.16% and 87.59%, respectively. Also, Multi-Scale CNN gives superior results on various challenging load settings, such as induction cooker connected with air purifier in high resolution and microwave connected with refrigerator in low resolution, and is proved run faster than traditional arcing detection methods.
The household energy management system (HEMS) has become an important system for energy conservation and emission reduction. In this study, home energy management considering carbon quota has been established. Firstly, the household photovoltaic output model, load model of various electrical appliances, battery load model, and charging and discharging of electric vehicles (EVs) model are established. Secondly, the carbon emission and carbon quota of household appliances and EVs are considered in these models. Thirdly, the energy optimization model of minimum the household user’s total comprehensive operation cost with the minimum total electricity consumption, carbon trading cost, battery degradation cost, and carbon quota income are proposed, taking into account constraints such as the comfort of users’ energy use time. Subsequently, the improved particle swarm optimization (IPSO) algorithm is used to tackle the problem. Compared to the standard particle swarm optimization (PSO), the IPSO has significantly improved the optimization effect. By comparing the optimization results in different scenarios, the effectiveness of the strategy is verified, and the influence of different carbon trading prices on optimal energy scheduling has been analyzed. The result shows that the comprehensive consideration of carbon trading cost and total electricity cost can reduce the household carbon emissions and the total electricity cost of the household user. By increasing the carbon trading price, the user’s carbon trading income and the EV carbon quota income increase, and the overall operating cost decreases; the guidance and regulation of carbon trading price can make a valuable contribution to HEMS optimization. Compared to the original situation, the household carbon emissions are reduced by 14.58 kg, a decrease of over 21.47%, while the total comprehensive operation cost are reduced by 14.12%. Carbon quota trading can guide household users to use electricity reasonably, reducing household carbon emissions and the total cost of household electricity.
To improve the operational efficiency and reliability of photovoltaic power stations, this paper introduces a novel approach to detect outliers in photovoltaic arrays using a Vine-Copula method. The procedure is divided into two distinct phases. Initially, it identifies deviations in the direct current (DC) component of the photovoltaic (PV) system. The following phase extends this by pinpointing irregularities in the DC voltage of the array. To model the interconnection between the PV current, irradiance, and temperature, the Vine-Copula is employed in this process. The optimisation of this function is based on the Akaike information criterion. Subsequently, a conditional probability model for the PV current is developed along with a formula to determine the quantile of this probability. This interval is then employed as the primary metric for detecting and eliminating current deviations. After refining the current data, a similar approach is taken to address voltage irregularities. The results of the simulation tests indicate that this proposed method is more effective, showing lower error rates and higher accuracy in detecting outliers, compared to other methods.
Modern distribution networks undergo frequent topology reconfiguration, volatile bi-directional flows, and noisy measurements, making five-class line-loss anomaly classification both valuable and challenging. In this study, PC-LossGNN is proposed—a physics-consistent spatiotemporal graph neural network for edge-level classification into Normal, Infrastructure, Documentation, Metering, and Theft. A static topology prior is fused with a measurement-adaptive graph and confidence-aware multi-source features; power-flow physics is injected via residual-guided attention using active/reactive balance, voltage-drop, and ohmic-loss residuals. A dual-path decoder is employed to yield calibrated probabilities and interpretable class evidence, trained under an uncertainty-weighted curriculum objective. On six months of real utility data, macro-F1 of 0.8503 and accuracy of 0.9915 are achieved, surpassing XGBoost, LSTM, GCN, STGCN, and two recent physics-aware spatiotemporal GNN baselines including ST-RGNN and PA-STGCN. Ablation indicates that physics-consistent regularization is pivotal, while adaptive topology and interactive temporal encoding further improve performance. Robustness tests with injected Gaussian noise show more graceful degradation than baselines. These results suggest that PC-LossGNN provides accurate, physically plausible, and interpretable five-way line-loss diagnostics suitable for real-world operations.