This paper presents a methodology for modeling the load demand of plug-in hybrid electric vehicles (PHEVs). Due to the stochastic nature of vehicle arrival time, departure time and daily mileage, probabilistic methods are chosen to model the driving pattern. However, these three elements of driving pattern are correlated with each other, which makes the probability density functions (pdfs)-based probabilistic methods inaccurate. Here a fuzzy logic based stochastic model is built to study the relationship between the three elements of driving pattern. Moreover, a load profile modeling framework (LPMF) for PHEVs is proposed to synthesize both the characteristics of driving pattern and vehicle parameters into a load profile prediction system. Based on this stochastic model of PHEV, a two-layer evolution strategy particle swarm optimization (ESPSO) algorithm is proposed to integrate PHEVs into a residential distribution grid. A novel business model is developed for PHEVs to provide ancillary service and participate in peak load shaving. A virtual time-of-use rate is used to reflect the load deviation of the system. Then, an objective function is developed to aggregate the peak load shaving, power quality improvement, charging cost, battery degradation cost and frequency regulation earnings into one cost function. The ESPSO approach can benefit the system in four major aspects by: 1) improving the power quality; 2) reducing the peak load; 3) providing frequency regulation service; and 4) minimizing the total virtual cost. Finally, simulations are carried out based on different control strategies and the results have demonstrated the effectiveness of the proposed algorithm.