Intelligent Energy Disaggregation and Optimization in Smart Homes Using NILM and AI-Based Predictive Analytics

With rising usage of IoT-devices in human society raised the demand for better energy management and monitoring systems in residential buildings. Many techniques for designing energy monitoring systems are proposed earlier in literature. Non-Intrusive Load Monitoring (NILM) technique is one of these known techniques. Generally, this technique is used in industrial energy monitoring systems. It utilizes appliance-wise energy disaggregation using single smart meter. This eliminates the need for intrusive sensor installation. This work presents the application of NILM on Indian residential infrastructures. It involves modelling and implementation of an adaptive machine learning (ML)-based NILM framework. The proposed method is tailored for daily domestic energy profiles. This utilizes feature-level signal decomposition which is followed by a hybrid classification approach. The method is compared to using Hidden Markov Models (HMM) and Random Forest (RF) algorithms. Incorporation of adaptive learning component in it allows incremental improvement. It eliminates the need to retrain the model from scratch whenever new appliance signatures are added. For analyzing the effectiveness of the proposed method, real-Time deployment is performed on live smart-meter data. The data is collected from an Indian residential home. Work also compares the performance of HMM and RF models on proposed technique. The results show promising performance of NILM technique in residential buildings.