Application of Long Short-Term Memory for Energy Load Prediction in the Microgrid Using Python Software

Abstract

The development of distributed power supply systems, microgrids is recognized as relevant and requires intensive study. Research on microgrid management systems is inextricably linked with data science. The paper presents a study of the use of software for predicting the load consumed by a typical microgrid over a monthly interval. The formulation of the problem of forecasting time series, applied to classical stationary series, is described. The process of data processing using the open-source machine software libraries NumPy, Keras is presented. A class is developed in the Python environment based on the use of recurrent neural networks-long short-term memory, the applicability for the task is shown. The model was trained using iterative optimization of the series value, and the data sampling window. The satisfactory accuracy of forecasting based on the developed model is shown. The conclusions for further study of the applicability of this algorithm in the practice of managing distributed power supply systems are presented.