Deep Learning Based Hybrid Assisted Stochastic Unit Commitment With Transportable Energy Storage

The computation load of the Transportable Energy Storage System with Stochastic Unit Commitment (TESS-SUC), when modelled as Mixed-Integer Linear Programming (MILP) problem, is very demanding, especially with complex transportation topology and large power system. This paper proposes a Deep Learning (DL) based model to reduce the optimization time of TESS-SUC while retaining close-to-optimal solutions. Hybrid Assistance (HA), a novel approach that leverages DL to support the optimization of the TESS-SUC MILP problem, is proposed. HA ensures a reduction in computation time, keeping the solution close-to-optimal and avoids the violation of constraints made by DL's prediction. Simulations on the IEEE 118-bus system with a Transportable Energy Storage (TES) showed that 94.52% of computation time reduction can be achieved while retaining the optimality of the solution with about 0.06% gap from the optimal solution, without any violation of constraints. Results further revealed that among various types of binary variables, assisting the solver with unit commitment decision variables is the most effective strategy for reducing computation time. Furthermore, correlation analysis indicated that optimizing precision and recall metrics for UC and TES binary variables, respectively, can further enhance the accuracy of DL predictions and ensure greater reduction of computation time.