Machine Learning for Power Consumption Forecasting

The article proposes a model for medium-term forecasting of daily active power schedules in Moscow, it is formed on the basis of machine learning using a regression decision tree algorithm, taking into account average daily data on air temperature, cloudiness and natural light. This model predicts the values of daily electricity consumption time series taking into account meteorological factors based on simple decision rules derived from the characteristics of the data. The initial data were archives of average daily values of electricity consumption time series, air temperature, cloudiness and natural light for the territory of Moscow for the period from January 1, 2018 to December 31, 2019. Forecasts of average daily active power values were made using a forecast model based on the decision tree method for selected periods in April, June and October. The results of forecasting the active powers average daily values show that in most test examples the values obtained were no more than 5%. In addition, the developed algorithm allows the use of seasonal patterns between daily active power schedules and meteorological factors. Software implemented as a modeling tool was Python.