Dec 2025· arXiv.org· Vol abs/2512.22287· 1 citation· 41 references
Computer Science
TL;DR
The proposed Cluster Aggregated GAN framework is established, a hybrid generative approach that routes each appliance to a specialized branch based on its behavioral characteristics that consistently outperforms baseline methods across metrics measuring realism, diversity, and training stability.
Abstract
Synthetic appliance data are essential for developing non-intrusive load monitoring algorithms and enabling privacy preserving energy research, yet the scarcity of labeled datasets remains a significant barrier. Recent GAN-based methods have demonstrated the feasibility of synthesizing load patterns, but most existing approaches treat all devices uniformly within a single model, neglecting the behavioral differences between intermittent and continuous appliances and resulting in unstable training and limited output fidelity. To address these limitations, we propose the Cluster Aggregated GAN framework, a hybrid generative approach that routes each appliance to a specialized branch based on its behavioral characteristics. For intermittent appliances, a clustering module groups similar activation patterns and allocates dedicated generators for each cluster, ensuring that both common and rare operational modes receive adequate modeling capacity. Continuous appliances follow a separate branch that employs an LSTM-based generator to capture gradual temporal evolution while maintaining training stability through sequence compression. Extensive experiments on the UVIC smart plug dataset demonstrate that the proposed framework consistently outperforms baseline methods across metrics measuring realism, diversity, and training stability, and that integrating clustering as an active generative component substantially improves both interpretability and scalability. These findings establish the proposed framework as an effective approach for synthetic load generation in non-intrusive load monitoring research.
Accurate identification of individual appliance activation and power consumption is essential for effective energy consumption analysis in residential buildings. Non-Intrusive Load Monitoring (NILM) allows the disaggregation of appliance consumption using the aggregate household power without the need for sensing. Despite the usefulness of NILM, the existing models struggle with the tasks of generalization, overlapping appliance signatures, and the presence of varying operating conditions. In this paper, a novel MoE-BiGRU-Transformer architecture is proposed. This architecture includes the combination of a mixture-of-expert routing method, bi-directional temporal information modeling, and self-attention under the paradigm of multi-task NILM framework, representing a unified integration of existing components for NILM. For this purpose, the MoE-BiGRU-Transformer model captures bidirectional temporal dependencies, global contextual features, and appliance-specific patterns through a self-attention process using the Transformer encoder component and learns appliance specialization through a gating function learning approach using the MoE component. A multi-task loss function is defined to train our model. Extensive experiments were conducted using two benchmark datasets: REDD and UK-DALE, while varying the number of epochs trained and the random seed. These results showed that the proposed model had average accuracy over 99%, high f1-score for most appliances, and low mean absolute error (MAE). Cross-dataset experiments showed high generalization, with accuracy and f1-score largely preserved when models were trained on one dataset and evaluated on the other. Ablation tests also showed that BiGRU and normalization components contributed to improved temporal resolution and stability of convergence.
M. Abbas, Malik Intisar Ali Sajjad, Gianfranco Chicco et al.· Proceedings of the Instituti...· 0 citations
A three-layer “Aggregator-Edge-End user” architecture leveraging a data-driven workflow and a hybrid deep learning model combining a Temporal Convolutional Network (TCN) and a Bidirectional Long Short-Term Memory (BiLSTM) network is introduced.
The power injections variability due to volatile renewable energy sources and large dynamic loads (e.g., Plug-in Electric Vehicles and Heat Pumps) may cause excessive voltage fluctuations and power system instability. To mitigate these problems, accurate load profiles are needed to support both grid operation and planning. However, real load profiles are not always readily and fully available due to technical and privacy constraints, limiting their applicability and the possibility of extrapolating consumption patterns for prospective studies. Synthetic load profile generation offers a practical alternative to address these limitations, while preserving data privacy and accessibility. This paper presents a deep learning framework that combines Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to produce artificial, but data-driven seasonal weekly load profiles (SWLP). The training process for each cluster uses separate WGANs, which implement the Wasserstein loss function together with gradient penalty to ensure stable training, while preventing mode collapse. The quality of the synthetic profiles is evaluated using statistical and distribution-based metrics. The proposed WGAN-GP achieved an average Wasserstein distance of 0.042 with a pattern correlation coefficient of about 0.98 with respect to the real load profiles derived from an Irish residential dataset. In comparison, the WGAN without gradient penalty returned am averageWasserstein distance of 0.352, while a Variational Autoencoder (VAE) used as a benchmark achieved an average Wasserstein distance of 0.108 with a pattern correlation coefficient of 0.92. Mean profile comparisons and load distribution analyses showed a good agreement between real and synthetic data across all identified consumption-pattern clusters. These results demonstrate the capability of the proposed framework to generate SWLPs preserving the statistical and temporal characteristics of real electricity consumption data.
S. Narayanan, M. Nair, D. Macii et al.· Energies· 0 citations
The proliferation of smart grids has led to an abundance of unlabeled residential electricity consumption time series data, which contains rich user behavior information. However, traditional profiling methods struggle to capture its inherent complexity. This paper proposes a novel deep self-supervised learning framework, hereafter referred to as Electricity Mixture of Experts (EleMoE), based on a Mixture of Experts (MoE) architecture for unsupervised profiling of residential electricity consumption behavior. Our framework innovatively integrates various complementary time series deep learning models (including MLP, LSTM, TCN, Transformer, and Mamba) as “experts.” These experts are trained via a self-supervised task of time series reconstruction to learn deep embeddings of consumption patterns. The MoE architecture enables the model to dynamically select or combine experts based on the characteristics of input segments, thereby capturing the complexity and diversity of electricity consumption behavior more comprehensively and robustly. After obtaining high-quality embeddings, we employ the K-Means algorithm for unsupervised clustering. The optimal number of clusters is systematically determined using analytical techniques like the Elbow and Silhouette methods, enhancing the objectivity of the profiling. Experimental results demonstrate that our proposed framework exhibits superior performance. Through in-depth analysis of average load curves and multi-dimensional radar charts, we successfully identify and explain various business-valuable residential electricity consumption profiles. This research offers an efficient, automated, interpretable, and robust new methodology for user behavior analysis in smart grids.
Xi-Jun Lin, Kang-Qian Huang, Zhen-Zhen Liu et al.· 電腦學刊· 0 citations
Electricity load forecasting is essential for balancing energy supply and demand, reducing energy waste, and maintaining power grid stability. Accurate forecasts enable power utilities to optimize energy dispatch and mitigate the risk of supply shortages or outages. However, conventional forecasting methods often struggle to capture highly nonlinear local fluctuations in electricity consumption and long-term temporal dependencies. To address these challenges, this study proposes MSCNN-ResLSTM, a hybrid model for multi-step electricity load forecasting. The proposed model integrates Multi-Scale Convolutional Neural Networks (MSCNNs) to extract local time-series features at multiple temporal scales, residual blocks (ResBlocks) to enhance feature representation through residual connections, and Long Short-Term Memory (LSTM) networks to model long-range temporal dependencies. To comprehensively evaluate its effectiveness, a cross-paradigm experimental framework is established in which MSCNN-ResLSTM is compared with seven representative benchmark models from three methodological categories: traditional machine learning (Extreme Gradient Boosting-XGBoost), classical recurrent and convolutional neural networks (LSTM, Temporal Convolutional Network-TCN, CNN-LSTM, MSCNN-LSTM, and Direct LSTM (Seq2Seq)), and self-attention-based architectures (Transformer). Experimental results show that MSCNN-ResLSTM achieves higher forecasting accuracy and greater stability across the full 24-step prediction horizon, consistently outperforming all competing baselines while effectively suppressing recursive error propagation.
Accurate power load forecasting is critical for the efficient operation of industrial microgrids. However, raw meteorological and consumption data typically exhibit non-stationary characteristics, complicating the hyperparameter tuning of deep learning models, and subsequently degrading the prediction accuracy of these frameworks. To address the aforementioned challenges, a new hierarchical forecasting structure denoted as INRBO-SSA-LSTM is proposed in this paper. First, Pearson correlation analysis is employed for feature reduction, identifying the four main factors to mitigate the dimensionality curse. Building upon this foundation, a refined Newton-Raphson-Based Optimizer (INRBO) is introduced, integrating a cosine adaptive t-distribution perturbation, a boundary-aware non-uniform steering scheme, and a fitness-aware hybrid perturbation mechanism. Evaluated against the CEC2022 benchmark suite, comprehensive evaluations reveal that the INRBO demonstrates superior global exploration and local refinement capabilities compared to baseline algorithms when assessed on the CEC2022 benchmark suite for foundational optimization performance. Furthermore, rigorous testing on the CEC2017 suite across 10, 30, and 50 dimensions successfully validates its exceptional robustness and search capabilities in high-dimensional spaces. INRBO functions as a dual-stage optimizer within the proposed framework; in the initial phase, it dynamically calibrates the parameters of Singular Spectrum Analysis (SSA) to extract deterministic load patterns, achieving a maximum signal-to-noise ratio of 15.87 dB; in the second phase, it optimizes the global hyperparameters of the Long Short-Term Memory (LSTM) network. Validated using actual industrial microgrid data in Jiangsu Province, China, the proposed method significantly outperforms traditional baseline models across all indicators; specifically, the prediction error (RMSE = 10.9764, MAPE = 3.7866%) is substantially minimized, and the coefficient of determination (R2 = 0.9741) is highly optimal. This adaptable framework effectively accommodates temporal demand variations, offering a robust foundation for the advancement of intelligent power management technology.
Jinming Luo, Fu Chen, Lingshan Kong et al.· Electronics· 0 citations
