Electrical demand forecast with arima and XGBoost models

Electric energy is a type of energy found in secondary energy groups that society needs in every field. Electrical energy is becoming more widely available in different fields in line with technological developments. Generating electrical energy in a way to meet the increasing needs and transferring it to the users in the right way at the right time is a very important problem. In case of excessive generation of electrical energy, storage is very inefficient in terms of cost and feasibility. Likewise, if electrical energy cannot meet the need, many problems may arise in terms of industry and social life. Due to the nature of electrical energy, it must be supplied as needed and delivered to industry and society. In this respect, countries or other organizations need to plan their energy policies, investments and operations sufficiently before in order to provide safe and efficient electricity. It may take years to implement a technical or strategic decision made in this area. To realize these processes, a country, organization, companies, etc. It needs to forecast future electricity demand and then plan every detail about it. In this study, a short term electricity demand forecast has been made. ARIMA and XGBoost models were used in the estimation. Australia's state of Victoria demanded nuclear water without question from 2000 to the 2019 market. It is intended to predict only the first week of January. This route is only in the first week of summer, these all models are analyzed for short term forecasting. While applying the ARIMA model, ACF and PACF graphics were used and the ARIMA model was created. The XGBoost model was chosen by comparison. For the training of the XGBoost model, a parameter set was created as time information and then these end parameters were added and all models were compared. As a result, it is seen that ARIMA and XGBoost models give similar results when first compared. It is seen that the XGBoost model increases its value and the parameters it learns give better results. The MAPE value was reduced from 0.10 to 0.05, the error was reduced. It is understood that with this massage, artificial intelligence models can be increased with the parameter to be added, taking into account the data. For future research, it can be made to compare real intelligence and similar models with the same factors in external factors such as parameters, temperature, population, obtained from time series.