You can design PI controller for either for load current or for load voltage connected to the system or for both loops.

In my experience there is no way to calculate analytically the parameters of a PID in a PWM inverter that feeds and AC-type network. Perhaps many approaches based on the mean model of the network connected to the inverter and the storage element in the DC side of the inverter have been addressed by many authors. The mean model is far away from the real model of the connected inverter due the non-linearity behavior of the system. Many papers have been written about this topic, but I'll give you and advice. Close your loop with an unitary gain, then observe the behavior of the error signal, add a proportional constant to the error signal and start doing some variations, if the passive network elements are stable in a wide range of frequencies you can make variation in 10% of the initial value of Kp and watch to the response of the system. Then add an integrator with a very little gain, 1% of the Kp constant, this integrator only can help with load changes, remember to reset it each cycle of the fundamental frequency. Finally if you want to add a derivative action, please take into account that this action could increase the frequency (or add noise) in the control signal "u" of the PID, and if that signal have bigger frequency than the carrier of your PWM, you won't have the expected modulation. 

Let me know if you find a more formal approach to set these parameters, without approximations like the mean model, or tuning the parameters like I do.

But, if you want to start with an analytical approach you can take a glance to this work: --Xiaoming Sun; Dichen Liu; Yuqiong Wang; Jie Zhao, "A quasi-PID controller for complex waveform generating inverters,"-- at http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4758007&isnumber=4757911