Design of a Data Driven Reactive Power Forecasting for an Active Cross-Voltage Level Reactive Power Management

Abstract:
In the context of the discussed approaches for cross-voltage level system control, the short-term forecasting of grid-related quantities e.g. reactive power is of elementary importance. This paper presents a complete design for a data-driven process for short-term forecasting of the reactive power demand of individual substations and the transmission- and distribution grid interface. The environment designed for this purpose and its features are presented in the scope of reactive power forecasts with real grid data. For this a machine-learning forecast algorithm is used, which is implemented on the basis of an autoregressive recurrent network. The advantages and disadvantages of the environment design as well as selected forecasting algorithm are discussed with respect to further development approaches.