Classification of electrical loads to increase energy efficiency in micro-grids

Depending on the developing technology and the increasing global population, energy has become a basic need in all areas of life. The choices made by human beings when using energy are so important that they affect both the environment and the life of living things. The phone charger left in the socket, the preference of using lamps and electronic devices that do not have a saving feature, and even the refrigerator door left open cause unnecessary electricity consumption and cause the natural balance of the world to deteriorate. Unnecessary energy consumption brings with it many problems such as climate change, global warming and infertility of the soil. In addition to sustainable and renewable energy sources, energy saving also plays an important role in reducing environmentally harmful energy production. Energy efficiency is the use of methods and techniques that require less energy to perform the same function. Based on this definition, it can be said that there is a close connection between the concepts of energy efficiency and energy saving. In order for the building occupants to use energy efficiently and to save energy, they should be familiar with the devices that are frequently used in homes and offices. Energy management is facilitated in consumption devices whose power consumption characteristics are determined. In this study, 5 electrical devices that are frequently used in homes and offices are selected and classified according to their consumption characteristics in order to use electrical energy more efficiently. In the experimental studies, active and reactive power consumptions and current and voltage harmonics of the fan, electric teapot, monitor, laptop computer and printer on the consumption side were taken as reference. Hioki PW3198 power analyzer was used as the measuring instrument and 15 signal samples were taken from each device. Thirty-one classes were determined by considering the different operating variations of the five selected devices. A total of 465 signal samples were obtained. In the feature matrix extraction, arithmetic mean, median value and standard deviation functions and one-dimensional local binary pattern technique were used. Recently widely used machine learning algorithms for classification were compared and the performance of the five algorithms that gave the best results were analyzed. According to the results obtained, the linear support vector machine method gave the highest accuracy rate with 83.4% accuracy. The support vector machine technique was the second algorithm to complete the classification in the shortest time with a time of 52.88 seconds.