Planning Electric Power Generation: A Nonlinear Mixed Integer Model Employing Benders Decomposition

Abstract

This paper describes the development and application of an optimization program that is used to help electric utilities plan investments for power generation. For each year over a planning horizon the program determines what types and sizes of generating plants should be constructed, so as to minimize total discounted cost while meeting reliably the system's forecasted demands for electricity. The problem is formulated as a large-scale, chance constrained, mixed integer program. The solution algorithm employs Benders' Partitioning Principle, a mixed integer linear programming code, and a successive linearization procedure. Computation costs are low and, in the important area of sensitivity analysis, the program offers special economies which make it attractive to power system planners. Computational results are presented for a full sized generation planning problem for the six New England states where the algorithm is currently being used for planning generating facilities.