Developing a residential consumption forecast method

Microgrids are going to take a major part of the future electricity distribution system, where they will form a local, controllable entity, consisting of various consumers, local energy producers and network energy storages. To be able plan, forecast and calculate the consumption and regulation energy values in advance by an algorithm - it is very important to have an accurate forecast, which takes the PV production too, into consideration. However, because of the low number of consumers and prosumers - the ordinary forecasting methods - which are applied in the transmission system-level consumption prediction - are not easily applicable in this situation, therefore a new approach is recommended. This paper focuses on developing such a method, that is capable of predicting a microgrid's electricity consumption in a D-1 and D-2 basis either from only the historical consumption data, or the history data and externalities. To do this, several real-life consumption data packages are analyzed and used to train a neural network. This network's parameters are subject to change throughout the process, according to the characteristics of the actual data, which is being predicted by it. Furthermore, in order to validate and evaluate the neural network's predictions - a classical ARIMA prediction model is also implemented and evaluated.