Short-Term Load Forecasting in DSO Substation Networks with Dimensionality Reduction Techniques

Electric load forecasting is a crucial task for Distribution System Operators (DSOs) in power system for many applications to face future RES increasing penetration with respect to different time horizons related to optimal operations, congestion prevention or future planning. In this field, neural computing and proper reduction techniques can be used to implement useful management tools to cope with uncertainty and future smart grid services. In this manuscript the combined use of Recurrent Neural Networks and Principal Component Analysis techniques is validated to guarantee the forecasting capability of electrical loads related to a grid substation. A historical series of measured Medium Voltage data is considered with the aims to demonstrate the efficiency of the designed procedure. Following all reported numerical results, the authors discuss a novel approach to be extended also to real end user loads in LV network including also thermal consumption, with a reduced dimension of the input dataset. A comparative analysis with the SoA is reported to support the proposed methodology.