Abstract: Machine Learning for Electricity Market Clearing
Solving the âOptimal Power Flowâ (OPF) problem in real-time allows for system operators to make minimum cost, feasible decisions regarding generator dispatch, and provides additional information such as locational marginal prices (LMPs). In this work, we expedite the solution of OPF using machine learning. In particular, given a power grid, learning a priori which lines are saturated at their thermal limits yields complementary slackness KKT conditions. When combined with the KKT condition of stationarity at the optimum, and the global requirement of balance between load and generator dispatch, these constitute a relatively low-dimensional, transparent system of linear equations which may be solved for generator dispatch, as well as the dual variables associated with the balance constraint and thermal limits. Subsequently, LMPs may be computed. In this talk, we outline this âIdentification-then-Solvingâ (ITS) procedure, and demonstrate how, if a neural network is effectively used to identify saturated lines, the procedure allows for efficient, accurate estimates of dispatch and LMPs. Furthermore, we seek solutions which adhere to financial coherency conditions of revenue adequacy, cost recovery, and efficiency via strong duality. Provided these conditions are met, ITS is a promising approach for real-time, cost-effective decision making by system operators, and in providing LMPs, it is additionally informative to electricity market participants.
