An artificial intelligence‐based non‐intrusive load monitoring of energy consumption in an electrical energy system using a modified K‐Nearest Neighbour algorithm

Abstract

Abstract Energy profligacy and appliance degradation are the apex reasons accounting for the continuous rise in power wastage and high energy bills. The decline in energy conservation and management in residences has been largely attributed to the financial implications of using intrusive methods. This work aimed to resolve the challenges of intrusive load monitoring by introducing artificial intelligence and machine learning to optimise load monitoring. To solve this challenge, a non‐intrusive approach was proposed where modalities for load prediction and classification were achieved with a Bagging regressor and a modified multiclass K‐Nearest Neighbour algorithms. This developed supervised learning models produced a 0.9624 R2 score and 78.24% accuracy for prediction and classification, respectively, when trained and tested on a Dutch Residential Energy Dataset. This work seeks to provide a cost‐effective approach to the optimisation of energy using steady state active power features. Essentially, the adoption of this non‐intrusive technique for load monitoring would effectively aid customers on the distribution network save cost on energy bills, facilitate the detection of faulty appliances, provide recommendations for smart homes and buildings with the required information for efficient decision making and planning of energy needs. In the long term, easing the pressure on power generation to meet demand would translate to reduction in carbon emissions based on a wide‐scale implementation of this proposed system. Hence, these are important parameters in realising the development of smart sustainable cities and sustainable energy systems in this current industrial revolution.

Keywords

• artificial intelligence, data analytics, data structures and machine learning
• smart cities
• smart cities applications
• smart power grids