Forecasting Electricity Consumption using Long Short Term Memory and Prophet Algorithm

Electricity has become one of the main human needs today because all environments, whether at home, at work, or in factories, use electrical energy. Every year the use of electricity always increases, this cause an increase in electricity prices which in turn makes electricity expensive. With the increase in tariffs, this should be an impetus for the public to be aware of saving electricity use. This study aims to compare the two models using two algorithms, namely LSTM and Prophet, then measure the level of accuracy and draw conclusions using the statistical metrics Mean Absolute Error (MAE) method to forecast electricity consumption in a period of thirty days or about one month. The datasets used are the consumption of electricity use in Germany during the period 2006 – 2017. This data includes the total daily consumption of electricity in GWh, daily column in day – month – year format, wind power production in GWh, production solar power in GWh, as well as the total sum of wind and solar power production in GWh. In this case, the researcher only uses daily column data in the format of days – months – years and data on total daily consumption of electricity as parameters to estimate electricity use for the next month. This data is provided by Open Power System Data (OPSD) and is available on the “kaggle.com” website. The data used in this study is very useful for time series analysis. Based on the results of testing with the LSTM algorithm with the SGD optimizer, the MAE value is 0.198987. The test results with the Prophet algorithm produce an MAE with a value of 40.