Figure 5 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Energy used for making predictions with the proposed model in comparison to VGG11, TanoniCRNN and VAE-NILM.
Source publication
Non-intrusive load monitoring (NILM) is the process of obtaining appliance-level data from a single metering point, measuring total electricity consumption of a household or a business. Appliance-level data can be directly used for demand response applications and energy management systems as well as for awareness raising and motivation for improve...
Context in source publication
Context 1
... 0.75 · 10 6 1.11 · 10 9 1.97 MJ VAE-NILM [19] 3.8 · 10 6 0.42 · 10 9 13.2 -263 MJ * * For VAE-NILM, it is only possible to compute a range of values based on the reported epochs [19] between 5 and 100. Group using batch size 128 are displayed in In addition to the energy used during the training, energy consumed for making predictions can also be significant when the number of requests for predictions is high as depicted in Figure 5. On the x-axis the figure plots the number of predictions from 0 to 10 million while on the y-axis it plots the consumed energy in mega Joules. ...
Citations
... This method requires less data and no need for data augmentation (DA) approaches. Reference [17] proposes a novel Convolutional transpose Reccurrent Neural Network (CtRNN) architecture focusing on reduced computational complexity and improving energy efficiency. Reference [18] proposes the few-shot Transfer learning (TL) based on metalearning and relational network to improve the load recognition generalization performance, which does not require complex inference and recurrent structures. ...
The loads that have several working states cannot be accurately distinguished by the conventional Non-Intrusive Load Monitoring (NILM) methods. This paper proposed an improved NILM method based on the Resnet18 Convolutional Neural Network (CNN) and Support Vector Machine (SVM) algorithm to address the misidentification of multi-state appliances. The V-I trajectories of loads are at first classified with Resnet18. Then, load features with low redundancy is obtained through the Max-Relevance and Min-Redundancy (mRMR) feature selection algorithm from various operating states of loads that were not successfully classified. The SVM algorithm is developed for two-stage identification to achieve high accuracy of classification for identifying the multi-state appliances quickly. This proposed NILM method can significantly improve the accuracy of identification for multi-state loads. Finally, the Plaid dataset is acquired to validate the effectiveness and accuracy of the proposed method.
Due to growing population and technological advances, global electricity consumption is increasing. Although CO2 emissions are projected to plateau or slightly decrease by 2025 due to the adoption of clean energy sources, they are still not decreasing enough to mitigate climate change. The residential sector makes up 25% of global electricity consumption and has potential to improve efficiency and reduce CO2 footprint without sacrificing comfort. However, a lack of uniform consumption data at the household level spanning multiple regions hinders large-scale studies and robust multi-region model development. This paper introduces a multi-region dataset compiled from publicly available sources and presented in a uniform format. This data enables machine learning tasks such as disaggregation, demand forecasting, appliance ON/OFF classification, etc. Furthermore, we develop an RDF knowledge graph that characterizes the electricity consumption of the households and contextualizes it with household-related properties enabling semantic queries and interoperability with other open knowledge bases like Wikidata and DBpedia. This structured data can be utilized to inform various stakeholders towards data-driven policy and business development.
Demand Response (DR) has become a key strategy for enhancing energy system sustainability and reducing costs. Deep Learning (DL) has emerged as crucial for managing DR's complexity and large data volumes, enabling near real-time decision-making. DL techniques can effectively tackle challenges such as selecting responsive users, understanding consumption behaviours, optimizing pricing, monitoring and controlling devices, engaging more consumers in DR schemes, and determining fair remuneration for participants. This research work presents an integrated architecture for smart grid energy management, combining a Deep Attention-Enhanced Sequence-to-Sequence Model (AES2S) with Energy-Aware Optimized Reinforcement Learning (EAORL). The objective is to design a system that performs non-intrusive load monitoring and optimizes demand response to enhance energy efficiency while maintaining user comfort. The AES2S module accurately performs appliance state identification and load disaggregation using convolutional layers, Enhanced Sequence-to-Sequence Model networks, and an attention mechanism. The EAORL module employs a multi-agent system, where each agent uses a Deep Q-Learning Network to learn optimal policies for adjusting energy consumption in response to grid conditions and user demand. The system uses an Iterative Policy Update mechanism, where agents update their policies sequentially, ensuring stable and effective learning. The integration ensures seamless data flow, with AES2S outputs enhancing EAORL state representations. Validated in a simulated smart grid environment, the architecture dynamically adjusts energy consumption, demonstrating significant improvements in energy efficiency, cost reduction, and user comfort. Evaluation metrics confirm the system's effectiveness, making AES2S-EAORL a robust solution for smart grid energy management and demand response optimization.
