Prediction of Electric Power Consumption Using RNN Networks

Abstract

The electric power service in the Syrian Arab Republic suffers from many difficulties resulting from the lack of resources (fuel), in addition to the sabotage of many generation centers by terrorist groups, which led to the implementation of rationing programs in the governorates according to the consumption of those governorates and the production centers located in them. (factories, pumping centers, hospitals and the population).

Forecasting electric energy consumption also requires knowledge of daily consumption quantities, consumption times and other influencing factors that constitute large amounts of data. Predicting the exact electrical load is still a challenging task due to many problems such as the non-linear nature of the time series or the seasonal patterns it displays, which are very time consuming and affect the accuracy of the prediction performance. The process can be improved by using RNNs.[2]

Initially, the optimal and appropriate consumption for the region was determined, compared with production and the possibility of passing the surplus to other backup operations or providing production centers with the surplus that could be obtained through the previous forecasting process.

Also, Recurrent Neural Networks (RNN) were used, which are time series based on data sequences according to time indices and their ability to predict future values ​​based on past data. Then the performance of those networks was compared with DNN networks (Dense Neural Network) to obtain an optimal future prediction that can be served by the Ministry of Electricity in the Syrian Arab Republic and to solve the problem of predicting the electrical load compared to previous studies.

The time-based successive division method has also been adopted, which has the ability to work more accurately for randomly sampled data. For cases of low regulation of the hourly data for wattage consumption, we can sample a set of data over time and take 20 percent of the data for example as training and test samples.

Based on the prediction values ​​resulting from this study, work is being done to distribute electrical energy in the most appropriate manner and in accordance with the importance of higher usage.