With the increased rate of industrialization in the world, the consumption of electricity has been rising steadily in all sectors. Technological progress has brought a multitude of electrical appliances to everyday life, which has led to increasing dependence on energy-intense appliances for comfortable and convenient life. Consequently, the consumer level energy demand has increased significantly in this fast paced modern society. In response to this growing demand, extensive deliberations have been made in terms of energy conservation and management, demand side management and energy efficient practices. One of the things that can be done to balance the energy demand is changing the consumption behaviour, optimising the schedule of use and adaptation to the energy generation resources. Energy conservation not only assists economic efficiency and demand regulation, but also means that it plays an important role in avoiding carbon emissions.

The Building and Climate Change report underscores the fact that residential and commercial buildings in India are responsible for 39 per cent of the overall carbon emissions, which is more than the emissions from the transportation and industrial sectors (Khan et al., 2025). Further research states that, in countries like India, almost 93 percent of the carbon emissions come from residential buildings (Chen et al., 2019). These findings highlight the need to focus on residential energy efficiency to tackle environmental problems caused by the rapid urbanisation.

The development of smart grid technologies has allowed greater capabilities in power system monitoring and control. Smart metering systems enable consumers to monitor how much electricity they use in a day, and real-time feedback of energy consumption encourages consumers to use their appliances more efficiently. Next-generation smart metering infrastructures enable fine-grained monitoring of load by using new and advanced cloud platforms and intelligent learning algorithms (Lemos et al., 2025). Recent research has shown that real time, appliance level feedback can lead to energy savings of more than 12% (Lin, 2022).

Moreover, real-time identification of loads allows energy providers to provide adequate grid services depending on the usage pattern of consumers. Therefore, online real-time load monitoring systems offer great potential to achieve energy efficiency in terms of better utility support services (Ray, 2025).

The main objective of this research is to achieve efficient Non-Intrusive Load Monitoring (NILM) framework through the use of appropriate energy disaggregation algorithm for smart home applications. The proposed model of NILM is further integrated with a microcontroller based hardware prototype for real-time monitoring and control of appliances towards enhanced energy-efficient operations of the grid (Chen et al., 2021).

Within the modern power networks, the development of smart grid technologies has reinforced the possibilities of data acquisition in real time and load management. Appliance specific usage patterns are an important parameter in the analysis of energy consumption in demand side management applications (Alsalemi et al., 2021).

Load disaggregation or non-intrusive appliance load monitoring is an important source of data for direct feedback control strategies in smart home energy management systems. Existing NILM methodologies can be broadly grouped into intrusive and non-intrusive monitoring methodologies. Intrusive monitoring is the method used to place sensors at the appliance node level, while non-intrusive monitoring uses advanced signal processing and machine learning algorithms to reconstruct individual appliance consumption from the aggregated meter data (Lin et al., 2022).

The framework of NILM improves user awareness of the use of electricity and effective control strategies of devices. The performance of NILM relies on the choice of suitable energy disaggregation algorithms depending on data availability, characteristics of appliances and load profile fluctuations (Joha et al., 2024).

In this study a NILM developments based on a decision tree is developed using benchmark datasets, namely the Reference Energy Disaggregation Dataset (REDD) and the Retrofit Decision Support Tools for UK Homes using Smart Home Technology (REFIT) dataset. A model is proposed which is then tested in terms of the normal performance indicators to prove its effectiveness. Furthermore, in order to validate the experimental results, a microcontroller-based hardware setup which is integrated with the ThingSpeak cloud platform is developed and connected with the proposed NILM model, so as to facilitate real-time monitoring and control of appliances for better energy efficiency (Sayed et al., 2021).

Non-intrusive load monitoring (NILM) has received much attention in academic research and industrial practice during the past two decades, because of its great potential to improve building level energy efficiency (Jahid, 2025). Consequently, significant efforts have been made in developing more robust machine learning models, which constitute a very central role in the energy disaggregation process and are crucial for making energy savings possible. This research trend is basically gone from classical supervised learning methods to deep learning based methods and unsupervised methods such as Hidden Markov Models (HMMs) are still an active research topic because of their performance in the problem of dis-agglomeration (Stogia 2025). With the constant development of technology, modern electrical appliances become increasingly sophisticated and their operation states become more complex and less distinguishable. In this context, the advent of deep learning is a significant breakthrough in research for energy disaggregation. Three deep neural networks architectures - namely: Long Short-Term Memory (LSTM) networks, denoising autoencoders, and a predictive model for estimating the appliance activation time, deactivation time, and average power consumption (Hu et al., 2020). Their results showed real improvements over the conventional methods in terms of accuracy and adaptability to unseen configurations of houses. Extending this work, proposed a sequence to point learning (Seq2point) framework to deal with the Single-Channel Mix Source Separation (MSS) problem (Yuan et al., 2020). By reformulating the learning task in order to simplify the mapping for the neural nets, their technique achieved better prediction accuracy and better performance on the real-world datasets by the automatic extraction of relevant signal features, which were previously handcrafted. In accordance with these developments, adaptive CoBiLSTM (Co.-attentive Bidirectional Long Short.-Term Memory) model in order to address the limitations of static disaggregation approaches (Lin 2020). By taking advantage of bidirectional LSTMs, the model captures the context of the variations in consumption patterns and hence gives extra flexibility and accuracy in estimating the appliance. Despite these methodological improvements, NILM still has its own substantial challenges, especially caused by the diversity of appliances and differences in user behavior. Variability prevents development of universally applicable NILM models (Franco et al., 2021). Additionally, the limited temporal resolution of smart meter measurements to limit the performance of disaggregation for some categories of appliances (Gheorghe, 2025). In order to overcome these limitations, probabilistic methods such as factorial hidden Markov models (FHMMs) have shown great potential. Bonfigli et al. suggested an improved FHMM framework coupled with modified AFAMAP algorithm by exploiting measurements of both active and reactive power for improved accuracy of disaggregation (Lin 2025 et al., 2025). Similarly, Modified FHMM (MFHMM), which decreases the computational complexity while enhancing the segmentation and identification of appliance operating states and demonstrated a good performance on publicly available datasets (Gopinath and Kumar 2023). In parallel, Wu et al. tried to reduce the dependency of HMM-based models on past information about appliances by proposing an adaptive clustering-based approach coupled with FHMMs, which further improved the accuracy in terms of disaggregation (Mari et al., 2023). Besides probabilistic models, hybrid learning strategies and ways of transfer learning have come out as promising directions. Introducing two transfer-learning strategies which can improve the generalisation of the model and decrease the amount of training data needed (Stefani et al., 2025). A hybrid CNN-LSTM model that takes mutual advantage of spatial and temporal features in load signatures, leading to better accuracy of load disaggregation (Papaioannou et al., 2025). The question of the scalability and adaptability of NILM solutions is a critical issue in the research. Pereira and Nunes emphasized the need for the design of methods that can be deployed on a large scale in various operating environments (Wang et al., 2022). At the same time there is increasing attention to privacy and ethical issues, for example, the work (Franco et al., 2023) which explored the trade-off between the analytical advantages and the user's privacy in smart-metering systems. Although a lot of progress has been made in the field of NILM research, there are important challenges that remain, in particular with respect to model generalisation and the reliance on labelled training data. These challenges to the optimisation of the whole power distribution network using smart meter data underlining the fact that energy disaggregation is only a part in the bigger picture of energy system efficiency improvement (Serna et al., 2025). PROPOSED METHOD The proposed framework works as an intelligent power management system, and thus allows for remote monitoring and control by consumers of electrical appliances in their homes by Internet of Things (IoT) enabled applications such as mobile applications. In conjunction with real time visualization of energy consumption and budgetary management functionalities, the system provides remote as well as manual control (on/off switching mechanisms) of devices. It provides data communication integrity between the user interface and the control unit, includes an emergency alert capability, and integrates a Google Map interface to help utility providers identify the locales with higher than normal energy consumption rates. These features are carefully designed to give consumers as well as the utility operators the ability to move towards sustainable energy practices by allowing ongoing remote monitoring and scheduling of appliance operate, which can optimise energy utilisation. The hardware structure of the proposed system consists of a client unit as shown in Figure 1. In the present implementation the microcontroller interface i.e. Arduino board is used as the client unit and is responsible for obtaining the measurements from the connected sensors. These measurements are then sent via a central database for storage and further processing. Apart from data acquisition, microcontroller also connects the electrical appliances by controlling them. The central unit is implemented with ESP32 module, with database stored within ESP32 module and web-based services such as monthly energy budgeting and notification alert are delivered.