RISP: Tunable Fault-Tolerance for Distributed Iterative Numerical Solvers for the Smart Grid

This paper presents a technique we call redundant iterative semantic paths (RISP) that can be applied to a class of existing distributed iterative numerical solvers to achieve agent or communication link fault tolerance. We transformed an existing distributed iterative power flow solver for radial power distribution systems using RISP and tested the result for fault tolerance. RISP achieves fault-tolerance by expanding the normal scope of local knowledge for each agent to include that of its neighboring peers (i.e. input clustering) and adapting the usual messaging scheme to allow the redundant computation of voltages and current for all buses. The solver requires the decoupling of the input sensing function from the compute function. This decoupling enables making knowledge otherwise local to an agent available to its neighbors, even if the agent itself, or its communication link, is not available. The level of clustering can be controlled incrementally, according to the desired level of fault-tolerance. Results for our solution, as applied to the 13-node IEEE test feeder at various redundancy levels, show significant improvement in agent and communication fault-tolerance, as compared to the non-RISP solution. Further work is needed to understand how the increased complexity of RISP can be managed in real systems to minimize any resulting adverse effects on the availability of infrastructure resources.