Real-time actuation of scheduled reserve capacity in power system operations with high penetration of wind power is prone to failure on account of unexpected shortcomings in network transfer capability. In this paper, a graph partitioning-based reserve zoning method is incorporated into the security-constrained unit commitment to improve the deliverability of operating reserves in a reserve zone and mitigate possible congestions caused by uncertain wind power. A graph representation of power system is proposed in which the edge weights are quantified by the likelihood of secure transmission utilization for each line. The probability distribution of line flows are characterized by the uncertainty of multiple correlated wind farm output forecasts as well as credible line outage contingencies reflected on the line flows using distribution factors. The minimum k-cut problem using the Gomory-Hu equivalent tree is addressed as a simple and efficient method to solve the NP-complete partitioning problem. The resultant zones can assure reduced risk of congested operating conditions and thus, provide a new approach to efficient management of intra-zonal congestions.