Online Tuning of Cloud-based Wide-Area Controllers with Variations in Network Traffic

We present an online tuning algorithm for wide-area controllers that are implemented over cloud-based shared communication networks. Synchrophasor measurements from individual generators are first communicated to virtual computing machines or VMs in the cloud. Each VM transmits its received state information to other VMs using software-defined communication (SDN) in the cloud, and computes the control law for its assigned generator. The network delays in the VM-to-VM communication, however, may cause the closed-loop performance of the grid to degrade. Perfect delay bounds can almost never be guaranteed no matter how robust the SDN controllers are, in which case the best option might be to drop that link from the control loop while still ensuring that the closed-loop performance remains close to optimal. To address this problem, we present a sparsity-constrained LQR algorithm that starts from an ideal all-to-all connected communication network, and progressively drops links if the traffic in those links are predicted to be high. Each drop is accompanied with a simultaneous optimal tuning of the control gains associated with the existing links using a method called Geromel's algorithm. We validate the effectiveness of our sparse optimal controller using simulations of the IEEE 39-bus power system model.