Frequency constrained unit commitment

Abstract

The unit commitment (UC) problem deals with the short-term schedule of the electrical generation to meet the power demand. The main objective is to minimize the production cost, while respecting technical and security constraints. In addition to the system load, a specific amount of spare capacity is committed to cope with uncertainties, such as forecasting errors and unit outages; this is called reserve and it has been traditionally specified following a static reliability criterion. In a system with a conventional generation mix, this security constraint allows one to obtain UC solutions that naturally provide an acceptable transient response. However, the increasing penetration of variable generation sources, such as wind and solar, can lead to UC solutions that no longer ensure system security. Thus, enhanced security constraints have been proposed to consider the power system dynamics when optimising the day-ahead generation schedule. Some published works are focused on the formulation of these constraints in a mixed-integer linear programming structure to apply classic optimisation techniques. Nevertheless, power system dynamics is a non-linear problem, and, to the authors’ knowledge, the limits of these linear approximations have not been discussed in literature. This work examines the ability of different UC models to produce secure schedules when facing unit outages, through the implementation of a set of primary reserve and energy co-optimisation models. These models are built based on linear approximations of dynamic constraints that are available in recent literature. Then, dynamic simulations are performed for every conceivable outage to observe the transient response of a test system and to quantify the risk of under frequency load shedding. It is shown that the system’s need for frequency regulation depends on the operating point and that enforcing a fixed reserve, inertia or maximal frequency slope constraint is not cost/effective to limit the dynamic risk. If used, these values should be set on an hourly basis by iteration between the optimisation and the dynamic simulator.