Assuming you already decided that load shedding is the solution to the problem (instead or in addition to reactive compensation, new transmission lines, etc.), then I suggest to re-study the network such that the non-convergence state is reached but incrementally. A voltage stability study would identify contingencies that collapse the network at some incremental power transfer.
In PSSe there is no API function to provide eigenvalues of the Jacobian matrix but by monitoring the voltage decrease at buses as the network reaches its singular point, a severely impacted region can be identified, then it can be said that the load in that region is good candidate for load shedding. Tools like the "fast voltage stability index" can be easily coded in python to get a list of weak buses in the base case. The ACCC, n-1-1 and 'AC Corrective Actions' activities can calculate load shedding under contingency to resolve the non-convergence state.
Tools like VSAT can analyze the network to find 'voltage collapse' conditions and provide a list of participating buses after a modal analysis run. Loads in those buses are candidates for load shedding.
Next, a dynamic simulation using under voltage load shedding relays would verify the effectiveness of the 'UVLS' schema, with appropriate load fraction and time delays, such the voltage collapse does not occur under contingency.
