An hour-ahead scheduling problem for a system with wind resource

Hour-ahead scheduling (HAS), currently as a third piece of the two-settlement market structure, allows parties to submit further bids, issues advisory schedules and prices, and plays a key role at updating the day-ahead (DA) scheduling results with renewed load forecast data. With the integration of wind resource, HAS will be more important for the reliability and economics of power system operation. Since the variability and uncertainty of wind power output make system operators difficult to access accurate wind data at DA time, and thus unsuitable to have the power output of conventional units with minimal dispatch costs at DA market, system operators need to give an optimal dispatch at hour-ahead time with the scheduled unit commitment and updated hour-ahead wind forecast. Furthermore, since HAS is the last “firewall” for system operator before 5-min real-time (RA) market to correct previous dispatch result, in order for enhancing the system reliability and reducing the total cost payment, especially frequency regulation cost at RA market, a good HAS is required to consider the load following (LF) capacity of conventional units with quantifying the physical characteristics of wind hour-ahead uncertainty and sub-hourly variability. Thus, in this paper, we states a HAS problem which gives a schedule result including optimal LF capacity considering the above new issues caused by wind. The results not only integrate intra-hourly ramping limits and marginal opportunity cost of conventional units, but also reflect the quality of LF service weighed by a probabilistic performance index when the conventional units are following 5-min net load. This will reduce inefficient or unduly discriminatory payment to LF service. A real small system in East China Power Grid is used as a numerical case backing the conclusion effectively with given assumptions on wind sub-hourly variability and speed forecast error. However, the proposed statement can be also implemented to any other distribution assumptions and thus extend the application.