Predicting Stability of a Decentralized Power Grid Linking Electricity Price Formulation to Grid Frequency Applying an Optimized Data-Matching Learning Network to Simulated Data

Abstract

The stability of decentralized electricity grids is influenced by real-time electricity prices and the cost sensitivity and reaction times of power producers and consumers. The decentral smart grid control (DSGC) system is designed to provide demand-side control of decentralized electricity grids by linking real-time electricity prices to changes in grid frequency over the time scale of a few seconds. This stimulates electricity demand-side consumption / production on similar time scales. Grid stability of DSGC systems can be simulated by considering a wide range of assumptions for the electricity volumes consumed / produced (P) by each grid participant, their cost-sensitivity (G) and reaction times (Tau) to changing grid conditions. Such a simulation (10,000 cases) published for a simple four-node star decentralized grid configuration with randomized values for P, G and Tau quantifies dynamic grid stability (Stbin) in terms of grid mechanical and pricing influences. This study applies an optimized data-matching machine-learning algorithm, the, transparent open box (TOB) learning network to predict Stbin (ranging from −0.0808 to +0.1094 s−2) for this published simulation from its independent variables. TOB manages to predict Stbin to a high degree of accuracy (RMSE ~0.016 s−2; R2 ~0.85) for this grid configuration in which independent variables P, G and Tau are poorly correlated with Stbin. By involving average G and Tau values for the three consumers as input variables TOB prediction accuracy is further improved (RMSE ~0.0075 s−2; R2 ~0.90). The study highlights the importance of compound feature selection when predicting grid stability of decentralized electricity grids.