Application of Deep Learning Technique based Load Forecast for Frequency Regulation

A primary goal of a power system operation and control strategy is keeping the system’s cumulative generation and load in sync. Load frequency control (LFC) has become an important component of contemporary power systems to attain this aim. The researchers have proposed several controllers to enhance the LFC scheme. In this paper, an active disturbance rejection control (ADRC) based control technique bridges the gap between traditional and modern control practices. The ADRC method provides better disturbance rejection. Furthermore, the load forecasting approach is incorporated to make the system resistant to future load disturbances. An improved long short-term memory (LSTM) based day ahead forecasting scheme is suggested in this work. The suggested method solves the error accumulation problem of the iterate-based forecasting technique that may occur while predicting multiple future steps. This forecasting-based control strategy is tested on the IEEE- 39 bus system with the active participation of renewable energy sources. The test system is divided into three areas. A real-time load data of Delhi NCR recorded at every 5-minute interval from three different distribution companies (DISCOs) is used as a load profile for the three-area test system. A day ahead prediction for all three areas generates a disturbance signal used as LFC input. The simulation result shows that the suggested forecasting scheme gives a satisfactorily predicted load compared to the actual load. The controller is shown to improve the frequency regulation and tie-line power deviation over time in a multi-area test power system.