Internet of things (IoT) based adaptive energy management system for smart homes

Abstract

Internet of things enhances the flexibility of measurements under different environments, the development of advanced wireless sensors and communication networks on the smart grid infrastructure would be essential for energy efficiency systems. It makes deployment of a smart home concept easy and realistic. The smart home concept allows residents to control, monitor and manage their energy consumption with minimal wastage. The scheduling of energy usage enables forecasting techniques to be essential for smart homes. This thesis presents a self-learning home management system based on machine learning techniques and energy management system for smart homes. Home energy management system, demand side management system, supply side management system, and power notification system are the major components of the proposed self-learning home management system. The proposed system has various functions including price forecasting, price clustering, power forecasting alert, power consumption alert, and smart energy theft system to enhance the capabilities of the self-learning home management system. These functions were developed and implemented through the use of computational and machine learning technologies. In order to validate the proposed system, real-time power consumption data were collected from a Singapore smart home and a realistic experimental case study was carried out. The case study had proven that the developed system performing well and increased energy awareness to the residents. This proposed system also showcases its customizable ability according to different types of environments as compared to traditional smart home models. Forecasting systems for the electricity market generation have become one of the foremost research topics in the power industry. It is essential to have a forecasting system that can accurately predict electricity generation for planning and operation in the electricity market. This thesis also proposed a novel system called multi prediction system and it is developed based on long short term memory and gated recurrent unit models. This proposed system is able to predict the electricity market generation with high accuracy. Multi Prediction System is based on four stages which include a data collecting and pre-processing module, a multi-input feature model, multi forecast model and mean absolute percentage error. The data collecting and pre-processing module preprocess the real-time data using a window method. Multi-input feature model uses single input feeding method, double input feeding method and multiple feeding method for features input to the multi forecast model. Multi forecast model integrates long short term memory and gated recurrent unit variations such as regression model, regression with time steps model, memory between batches model and stacked model to predict the future generation of electricity. The mean absolute percentage error calculation was utilized to evaluate the accuracy of the prediction. The proposed system achieved high accuracy results to demonstrate its performance.