基于联邦学习的时间感知非侵入式负载监测研究

在现代社会中，能源消耗成为一个重要议题，而在家庭中，电器的能耗监测尤为关键。通过充分监测每个电器的能耗，能够帮助用户更好地参与节能实践。非侵入式负载监测（NILM）是一种能源管理技术，它能够将整个家庭用电量分解成每个单独电器的能耗。近年来，深度学习模型在NILM任务中的应用取得了显著进展，然而，面对未知家庭时，现有模型的性能仍然存在严重下降的问题。因此，为了提高模型的监测性能和泛化能力，本文对NILM展开了研究。

在研究过程中，本文观察到时域信息有助于增强模型的泛化能力。因为不同家庭的电器使用模式往往遵循相似的时间规律，因此利用时域信息可以提高模型性能。基于以上观察，本文提出了一种基于时间感知的双卷积神经网络架构用于NILM（TimeNILM），其中双卷积神经网络分别用于回归和分类，然后通过基于注意力机制的特征融合方法使回归子网络从分类子网络中获取关于关键时刻的知识，从而增强TimeNILM模型对这些时刻的准确识别，进而提高监测性能，同时本文还探讨了在无监督迁移场景下的负载监测研究。实验结果表明，TimeNILM模型在两个真实世界数据集（REDD和UK-DALE）上表现优异，在未知家庭中实现了8%-27%的平均绝对误差（MAE）增益和8%-35%的信号聚合误差（SAE_𝛿）增益。

为了保护用户数据的隐私安全，本文将TimeNILM模型扩展到联邦学习的框架，由于相同的家电在不同家庭可能存在相似的使用模式和状态转换，本文对分类子网络的知识进行共享，并提出一种共同学习和聚合的联邦策略，以平衡性能提升和数据隐私保护之间的关系。此外，本文成功在资源受限的边缘设备上部署TimeNILM模型，并探讨不同的量化方案对总体性能的影响，通过实验验证表明该模型能够在树莓派上进行实时的负载监测，为实际应用提供了可行的解决方案。

综上所述，本文针对家庭中应用NILM所面临的难点和问题，提出了基于时间感知的TimeNILM模型，有效地提高了监测性能和泛化能力，并将该模型扩展到联邦学习的框架以保护用户数据的隐私安全。通过在真实世界数据集上进行实验验证，并将TimeNILM模型部署到资源受限的边缘设备上，本文证明了该模型在实际应用中的可行性和有效性。这项研究对于推动智能家居节能技术的发展具有重要意义，为用户提供了更好的能源管理和节能实践的支持。

Energy consumption has become a significant concern in modern society, with monitoring household appliance energy consumption being particularly crucial. Thorough monitoring of the energy consumption of each appliance can assist users in engaging more effectively in energy-saving practices. Non-intrusive load monitoring (NILM) is an energy management technology that disaggregates the total household electricity consumption into the energy consumption of individual appliances. In recent years, significant progress has been made in applying deep learning models to NILM tasks. However, existing models still suffer from severe performance degradation when faced with unknown households. Therefore, to enhance the monitoring performance and generalization ability of the model, this paper conducts research on NILM.

During the research process, the paper observes that temporal information can help enhance the model's generalization ability. Since the usage patterns of different households' appliances often follow similar temporal patterns, utilizing temporal information can improve model performance. Based on this observation, the paper proposes a time-aware dual convolutional neural network architecture for NILM (TimeNILM), where dual convolutional neural networks are used for regression and classification, respectively. Then, the feature fusion method based on the attention mechanism is employed to enable the regression subnetwork to acquire knowledge about crucial moments from the classification subnetwork, thereby enhancing the TimeNILM model's accurate identification of these moments and improving NILM performance. Additionally, the paper also explores load monitoring research in unsupervised transfer scenarios. Experimental results demonstrate that the TimeNILM model achieves excellent performance gains of 8% to 27% in mean absolute error (MAE) and 8% to 35% in signal aggregated error (SAE_𝛿) on two real-world datasets (REDD and UK-DALE) in unknown households.

To safeguard user data privacy, the paper extends the TimeNILM model to the framework of federated learning. As similar household appliances may exhibit similar usage patterns and state transitions in different households, the paper shares knowledge of the classification subnetwork and proposes a federated strategy for joint learning and aggregation to balance the relationship between performance improvement and data privacy protection. Furthermore, the paper successfully deploys the TimeNILM model on resource-constrained edge devices and investigates the impact of different quantization schemes on overall performance. Experimental validation demonstrates that the model can perform real-time load monitoring on Raspberry Pi, providing a feasible solution for practical applications.

In summary, this paper addresses the challenges and issues faced in applying NILM in households by proposing the time-aware TimeNILM model, which effectively enhances monitoring performance and generalization ability. Moreover, the model is extended to the framework of federated learning to protect user data privacy. Through experimental validation on real-world datasets and deployment on resource-constrained edge devices, the paper demonstrates the feasibility and effectiveness of the model in practical applications. This research is of significant importance in advancing smart home energy-saving technologies and providing better support for users in energy management and energy-saving practices.

[1] BOUCKAERT S, PALES A, MCGLADE C, et al. Net zero by 2050: A roadmap for the global energy sector[Z]. 2021.
[2] EHRHARDT-MARTINEZ K, DONNELLY K, LAITNER J. Advanced metering initiatives and residential feedback programs: a meta-review for household electricity-saving opportunities.[J]. American Council for an Energy-Efficient Economy, 2010.
[3] DARBY S. The effectiveness of feedback on energy consumption[M]. Environmental Change Institute, University of Oxford, 2006.
[4] RIDI A, GISLER C, HENNEBERT J. A survey on intrusive load monitoring for appliance recognition[C]//2014 22nd international conference on pattern recognition. IEEE, 2014: 3702-3707.
[5] ZHONG M, ZHANG R, LUAN W. Lightweight Non-Intrusive Load Monitoring: A Short Review[J/OL]. SSRN Electronic Journal, 2022. http://dx.doi.org/10.2139/ssrn.3974727.
[6] HIMEUR Y, ALSALEMI A, BENSAALI F, et al. Recent trends of smart nonintrusive load monitoring in buildings: A review, open challenges, and future directions[J]. International Journal of Intelligent Systems, 2022, 37(10): 7124-7179.
[7] SCHIRMER P A, MPORAS I. Non-intrusive load monitoring: A review[J]. IEEE Transactions on Smart Grid, 2022, 14(1): 769-784.
[8] KASELIMI M, PROTOPAPADAKIS E, VOULODIMOS A, et al. Towards trustworthy energy disaggregation: A review of challenges, methods, and perspectives for non-intrusive load monitoring[J]. Sensors, 2022, 22(15): 5872.
[9] KELLY J, KNOTTENBELT W. The UK-DALE dataset, domestic appliance-level electricity demand and whole-house demand from five UK homes[J]. Scientific data, 2015, 2(1): 1-14.
[10] HART G W. Nonintrusive appliance load monitoring[J]. Proceedings of the IEEE, 1992, 80(12): 1870-1891.
[11] JIMENEZ Y, DUARTE C, PETIT J, et al. Steady state signatures in the time domain for nonintrusive appliance identification[J]. Ingeniería e Investigación, 2015, 35: 58-64.
[12] KLEMENJAK C, EGARTER D, ELMENREICH W. Yomo: the arduino-based smart metering board[J]. Computer Science-Research and Development, 2016, 31: 97-103.
[13] LAUGHMAN C, LEE K, COX R, et al. Power signature analysis[J]. IEEE power and energy magazine, 2003, 1(2): 56-63.
[14] KOLTER J Z, JOHNSON M J. REDD: A public data set for energy disaggregation research[C]//Workshop on data mining applications in sustainability (SIGKDD), San Diego, CA: Vol. 25. 2011: 59-62.
[15] FORNEY G D. The viterbi algorithm[J]. Proceedings of the IEEE, 1973, 61(3): 268-278.
[16] YAN L, SHEIKHOLESLAMI M, GONG W, et al. Challenges for real-world applications of nonintrusive load monitoring and opportunities for machine learning approaches[J]. The Electricity Journal, 2022, 35(5): 107136.
[17] SARASWAT P. Supervised machine learning algorithm: a review of classification techniques[J]. Integrated Emerging Methods of Artificial Intelligence & Cloud Computing, 2021: 477-482.
[18] NAEEM S, ALI A, ANAM S, et al. An unsupervised machine learning algorithms: Comprehensive review[J]. International Journal of Computing and Digital Systems, 2023.
[19] ESTHER VARMA C, PRASAD P S. Supervised and Unsupervised Machine Learning Approaches—A Survey[C]//ICDSMLA 2021: Proceedings of the 3rd International Conference on Data Science, Machine Learning and Applications. Springer, 2023: 73-81.
[20] SHI X, MING H, SHAKKOTTAI S, et al. Nonintrusive load monitoring in residential households with low-resolution data[J]. Applied Energy, 2019, 252: 113283.
[21] MACHLEV R, LEVRON Y, BECK Y. Modified cross-entropy method for classification of events in NILM systems[J]. IEEE Transactions on Smart Grid, 2018, 10(5): 4962-4973.
[22] ALSHAREEF S, MORSI W G. Application of wavelet-based ensemble tree classifier for nonintrusive load monitoring[C]//2015 IEEE Electrical Power and Energy Conference (EPEC). IEEE, 2015: 397-401.
[23] LIN Y H, TSAI M S, CHEN C S. Applications of fuzzy classification with fuzzy c-means clustering and optimization strategies for load identification in NILM systems[C]//2011 IEEE international conference on fuzzy systems (FUZZ-IEEE 2011). IEEE, 2011: 859-866.
[24] DEBNATH R, BARDHAN R, MISRA A, et al. Lockdown impacts on residential electricity demand in India: A data-driven and non-intrusive load monitoring study using Gaussian mixture models[J]. Energy Policy, 2022, 164: 112886.
[25] LIU Q, NAKOTY F M, WU X, et al. A secure edge monitoring approach to unsupervised energy disaggregation using mean shift algorithm in residential buildings[J]. Computer Communications, 2020, 162: 187-195.
[26] LYU Z, ZHU Q, ZHANG H, et al. Non-invasive Load Identification and Decomposition Based on Affinity Propagation and Selected Bayes Classification[C]//2021 13th International Conference on Measuring Technology and Mechatronics Automation (ICMTMA). IEEE, 2021: 832-835.
[27] PEREIRA L, NUNES N. An empirical exploration of performance metrics for event detection algorithms in non-intrusive load monitoring[J]. Sustainable Cities and Society, 2020, 62: 102399.
[28] LAZZARETTI A E, RENAUX D P B, LIMA C R E, et al. A multi-agent NILM architecture for event detection and load classification[J]. Energies, 2020, 13(17): 4396.
[29] ZHANG L, JING Z, AUGER F, et al. Abrupt change detection of home electrical appliance signals using mathematical morphology gradient[C]//2019 IEEE Innovative Smart Grid Technologies-Asia (ISGT Asia). IEEE, 2019: 726-730.
[30] ZHU Z, ZHANG S, WEI Z, et al. A novel CUSUM-based approach for event detection in smart metering[C]//IOP Conference Series: Materials Science and Engineering: Vol. 322. IOP Publishing, 2018: 072014.
[31] PARSON O, GHOSH S, WEAL M, et al. Non-intrusive load monitoring using prior models of general appliance types[C]//Proceedings of the AAAI Conference on Artificial Intelligence: Vol. 26. 2012: 356-362.
[32] KONG W, DONG Z Y, HILL D J, et al. A hierarchical hidden Markov model framework for home appliance modeling[J]. IEEE Transactions on Smart Grid, 2016, 9(4): 3079-3090.
[33] HUANG X, YIN B, WEI Z, et al. An online non-intrusive load monitoring method based on Hidden Markov model[C]//Journal of Physics: Conference Series: Vol. 1176. IOP Publishing, 2019: 042036.
[34] LUAN W, YANG F, ZHAO B, et al. Industrial load disaggregation based on hidden Markov models[J]. Electric Power Systems Research, 2022, 210: 108086.
[35] KOLTER J Z, JAAKKOLA T. Approximate inference in additive factorial hmms with application to energy disaggregation[C]//Artificial intelligence and statistics. PMLR, 2012: 1472-1482.
[36] MAKONIN S, POPOWICH F, BAJIC I V, et al. Exploiting HMM Sparsity to Perform Online Real-Time Nonintrusive Load Monitoring[J/OL]. IEEE Transactions on Smart Grid, 2016, 7(6): 2575–2585. http://dx.doi.org/10.1109/tsg.2015.2494592.
[37] BONFIGLI R, PRINCIPI E, FAGIANI M, et al. Non-intrusive load monitoring by using active and reactive power in additive Factorial Hidden Markov Models[J]. Applied Energy, 2017, 208: 1590-1607.
[38] BONFIGLI R, SEVERINI M, SQUARTINI S, et al. Improving the performance of the AFAMAP algorithm for Non-Intrusive Load Monitoring[C]//2016 IEEE Congress on Evolutionary Computation (CEC). IEEE, 2016: 303-310.
[39] EGARTER D, BHUVANA V P, ELMENREICH W. PALDi: Online load disaggregation via particle filtering[J]. IEEE Transactions on Instrumentation and Measurement, 2014, 64(2): 467-477.
[40] SIROJAN T, PHUNG B T, AMBIKAIRAJAH E. Deep neural network based energy disaggregation[C]//2018 IEEE International conference on smart energy grid engineering (SEGE). IEEE, 2018: 73-77.
[41] HUBER P, CALATRONI A, RUMSCH A, et al. Review on deep neural networks applied to low-frequency nilm[J]. Energies, 2021, 14(9): 2390.
[42] ANGELIS G F, TIMPLALEXIS C, KRINIDIS S, et al. NILM applications: Literature review of learning approaches, recent developments and challenges[J]. Energy and Buildings, 2022, 261: 111951.
[43] AZAD M I, RAJABI R, ESTEBSARI A. Non-intrusive load monitoring (nilm) using deep neural networks: A review[C]//2023 IEEE International Conference on Environment and Electrical Engineering and 2023 IEEE Industrial and Commercial Power Systems Europe (EEEIC/I&CPS Europe). IEEE, 2023: 1-6.
[44] KELLY J, KNOTTENBELT W. Neural nilm: Deep neural networks applied to energy disaggregation[C]//Proceedings of the 2nd ACM international conference on embedded systems for energy-efficient built environments. 2015: 55-64.
[45] BONFIGLI R, FELICETTI A, PRINCIPI E, et al. Denoising autoencoders for non-intrusive load monitoring: improvements and comparative evaluation[J]. Energy and Buildings, 2018, 158: 1461-1474.
[46] TSANTEKIDIS A, PASSALIS N, TEFAS A. Recurrent neural networks[M]//Deep learning for robot perception and cognition. Elsevier, 2022: 101-115.
[47] NOH S H. Analysis of gradient vanishing of RNNs and performance comparison[J]. Information, 2021, 12(11): 442.
[48] CHUNG J Y, GULCEHRE C, CHO K, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[C]//Neural Information Processing Systems 2014 Workshop on Deep Learning. 2014.
[49] LINDEMANN B, MÜLLER T, VIETZ H, et al. A survey on long short-term memory networks for time series prediction[J]. Procedia Cirp, 2021, 99: 650-655.
[50] KIM J, KIM H, et al. Classification performance using gated recurrent unit recurrent neural network on energy disaggregation[C]//2016 international conference on machine learning and cybernetics (ICMLC): Vol. 1. IEEE, 2016: 105-110.
[51] KRYSTALAKOS O, NALMPANTIS C, VRAKAS D. Sliding window approach for online energy disaggregation using artificial neural networks[C]//Proceedings of the 10th Hellenic Conference on Artificial Intelligence. 2018: 1-6.
[52] WANG T, JI T, LI M. A new approach for supervised power disaggregation by using a denoising autoencoder and recurrent LSTM network[C]//2019 IEEE 12th International symposium on diagnostics for electrical machines, power electronics and drives (SDEMPED). IEEE, 2019: 507-512.
[53] KASELIMI M, DOULAMIS N, VOULODIMOS A, et al. Context aware energy disaggregation using adaptive bidirectional LSTM models[J]. IEEE Transactions on Smart Grid, 2020, 11(4): 3054-3067.
[54] XIA M, WANG K, SONG W, et al. Non-intrusive load disaggregation based on composite deep long short-term memory network[J]. Expert Systems with Applications, 2020, 160: 113669.
[55] CONG S, ZHOU Y. A review of convolutional neural network architectures and their optimizations[J]. Artificial Intelligence Review, 2023, 56(3): 1905-1969.
[56] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
[57] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 779-788.
[58] VAN DEN OORD A, DIELEMAN S, ZEN H, et al. WaveNet: A Generative Model for Raw Audio[J/OL]. Computer Research Repository, 2016. http://arxiv.org/abs/1609.03499.
[59] BAI X, SHI B, ZHANG C, et al. Text/non-text image classification in the wild with convolutional neural networks[J]. Pattern Recognition, 2017, 66: 437-446.
[60] ZHANG C, ZHONG M, WANG Z, et al. Sequence-to-point learning with neural networks for non-intrusive load monitoring[C]//Proceedings of the AAAI conference on artificial intelligence: Vol. 32. 2018.
[61] BARBER J, CUAYÁHUITL H, ZHONG M, et al. Lightweight non-intrusive load monitoring employing pruned sequence-to-point learning[C]//Proceedings of the 5th international workshop on non-intrusive load monitoring. 2020: 11-15.
[62] SHIN C, JOO S, YIM J, et al. Subtask gated networks for non-intrusive load monitoring[C]//Proceedings of the AAAI Conference on Artificial Intelligence: Vol. 33. 2019: 1150-1157.
[63] HE J, LIU J, ZHANG Z, et al. MSDC: exploiting multi-state power consumption in nonintrusive load monitoring based on a dual-CNN model[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2023: 5078-5086.
[64] LAFFERTY J D, MCCALLUM A, PEREIRA F C N. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data[C]//Proceedings of the Eighteenth International Conference on Machine Learning. San Francisco, CA, USA: Morgan Kaufmann, 2001: 282–289.
[65] LIM W Y B, LUONG N C, HOANG D T, et al. Federated learning in mobile edge networks: A comprehensive survey[J]. IEEE Communications Surveys & Tutorials, 2020, 22(3): 2031-2063.
[66] LI T, SAHU A K, TALWALKAR A, et al. Federated learning: Challenges, methods, and future directions[J]. IEEE signal processing magazine, 2020, 37(3): 50-60.
[67] LI Q, WEN Z, WU Z, et al. A survey on federated learning systems: Vision, hype and reality for data privacy and protection[J]. IEEE Transactions on Knowledge and Data Engineering, 2021, 35(4): 3347-3366.
[68] YANG Q, LIU Y, CHEN T, et al. Federated machine learning: Concept and applications[J]. ACM Transactions on Intelligent Systems and Technology (TIST), 2019, 10(2): 1-19.
[69] PÖTTER H, LEE S, MOSSÉ D. Towards privacy-preserving framework for non-intrusive load monitoring[C]//Proceedings of the Twelfth ACM international conference on future energy sys tems. 2021: 259-263.
[70] WANG X, LI W. Mtfed-nilm: Multi-task federated learning for non-intrusive load monitoring[C]//2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech). IEEE, 2022: 1-8.
[71] ZHANG Y, TANG G, HUANG Q, et al. FedNILM: Applying Federated Learning to NILM Applications at the Edge[J]. IEEE Transactions on Green Communications and Networking, 2023, 7(2): 857-868.
[72] MARI S, BUCCI G, CIANCETTA F, et al. Advanced architecture for training and testing NILM systems[C]//2022 IEEE International Instrumentation and Measurement Technology Conference (I2MTC). IEEE, 2022: 1-6.
[73] BRAUWERS G, FRASINCAR F. A general survey on attention mechanisms in deep learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2021, 35(4): 3279-3298.
[74] NIU Z, ZHONG G, YU H. A review on the attention mechanism of deep learning[J]. Neurocomputing, 2021, 452: 48-62.
[75] SOYDANER D. Attention mechanism in neural networks: where it comes and where it goes[J]. Neural Computing and Applications, 2022, 34(16): 13371-13385.
[76] BAHDANAU D, CHO K, BENGIO Y. Neural Machine Translation by Jointly Learning to Align and Translate[C]//3rd International Conference on Learning Representations. 2015.
[77] SUDHA D, RAMAKRISHNA M. Comparative study of features fusion techniques[C]//2017 International Conference on Recent Advances in Electronics and Communication Technology (ICRAECT). IEEE, 2017: 235-239.
[78] KIM D, CHOI J. Unsupervised representation learning for binary networks by joint classifier learning[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022: 9747-9756.
[79] MCMAHAN B, MOORE E, RAMAGE D, et al. Communication-efficient learning of deep networks from decentralized data[C]//Artificial intelligence and statistics. 2017: 1273-1282.
[80] NGUYEN H V, BAI L. Cosine similarity metric learning for face verification[C]//Asian conference on computer vision. Springer, 2010: 709-720.
[81] SINGH R H, MAURYA S, TRIPATHI T, et al. Movie recommendation system using cosine similarity and KNN[J]. International Journal of Engineering and Advanced Technology, 2020, 9(5): 556-559.
[82] PARK K, HONG J S, KIM W. A methodology combining cosine similarity with classifier for text classification[J]. Applied Artificial Intelligence, 2020, 34(5): 396-411.
[83] AHMED M, SERAJ R, ISLAM S M S. The k-means algorithm: A comprehensive survey and performance evaluation[J]. Electronics, 2020, 9(8): 1295.
[84] PAPAKYRIAKOU D, BARBOUNAKIS I S. Benchmarking and Review of Raspberry Pi (RPi) 2B vs RPi 3B vs RPi 3B+ vs RPi 4B (8GB)[J/OL]. International Journal of Computer Applications, 2023, 185(3): 37–52. http://dx.doi.org/10.5120/ijca2023922693.
[85] LIU X, XING Z, LIU H, et al. Combination of UAV and Raspberry Pi 4B: Airspace detection of red imported fire ant nests using an improved YOLOv4 model[J]. Mathematical biosciences and engineering: MBE, 2022, 19(12): 13582-13606.
[86] WEI Y, HONG S, MA M, et al. Raspberry Pi 4B-based cloud-based robot design and demonstration platform construction[C]//2023 IEEE 3rd International Conference on Power, Electronics and Computer Applications (ICPECA). IEEE, 2023: 1736-1739.
[87] WU Q. Research Control a Smart Car Based on Chinese Voice Recognition Technology through the Raspberry Pi 4B Platform[C]//2023 IEEE 3rd International Conference on Power, Electronics and Computer Applications (ICPECA). IEEE, 2023: 1465-1473.
[88] DAI J. Real-time and accurate object detection on edge device with TensorFlow Lite[C]//Journal of physics: conference series: Vol. 1651. IOP Publishing, 2020: 012114.
[89] CRUZ T S, GOMES J P, MARTINS L F, et al. Extensible hardware inference accelerator for fpga using models from tensorflow lite[C]//2023 IEEE International Conference on Consumer Electronics (ICCE). IEEE, 2023: 1-5.
[90] PRANEETH R S, AKASH K C S, SREE B K, et al. Scaling object detection to the edge with yolov4, tensorflow lite[C]//2023 7th International Conference on Computing Methodologies and Communication (ICCMC). IEEE, 2023: 1547-1552.