Auto tuning for new energy dispatch problem: A case study

• We propose a real-size rolling model to predict the Unit Commitment scheduling.
• We formulate the model to maximize new energy generation.
• We implement an auto-tuning solver CMIP for our model.
• We use auto-tuning techniques to optimize presolving, primal heuristics, etc.
• We compare the performance of other solvers and our auto-tuning solver.

The new energy dispatch problem has aroused more and more attention. In this paper, we investigate the problem of determining the optimal usage of generating power during a scheduling period. A set of MIP formulations are adopted for precise modeling of the variety of power systems (different power generation units) and the actual situation in china. Based on these formulations, we construct a new energy dispatch model which includes many MIP sub-problems. An auto-tuning MIP solver CMIP is given to effectively improve the performance of solving the proposed model. The CMIP focuses on optimizations for presolver, the LP solver for corresponding relaxation problem, and the primal heuristics. Actual predict data is used in performance experiments. Computational results conform to the viability of optimization. Our optimizations further reduce 27.6% of the average execution time compared to CPLEX.