Parallel Simulator for Large Power Systems
Network description, where the state equations of the circuit are constructed, and every dynamic element of the network is then replaced by its companion model. In this phase also the partitioning of the system and the load balancing among the DSPs are performed. Components solution, where the initial values for every component of the circuit are set and the companion models are solved on each DSP. Network solution collects the values found for each element of the circuit, and performs the triangular substitution on the factorized network matrix, in order to obtain the system solution (see Figure 2).
As the factorization process of the network matrix is performed off-line, and the on-line triangular substitution phase only depends on the execution time of the components solution process, the most time critical task is highly affected by the selection of the integration method for the time discretization and linearization phases, inside the network description process. We propose to use partitioning techniques, so that the forward and backward substitution processes will be distributed among the DSPs. The approach presents four main advantages:
• The solutions of the single components are distributed between different DSPs.
• The elaboration time is reduced and exactly predicted, given that no multiple iterations are required.
• The size of the matrix that describes the system is reduced, since components solved using state equation do not present internal node.
• The process of building the solution is easy to automatize, since the resistive companion frame work is used.
Some preliminary results, performed on a CPU architecture, have shown that our approach may be effectively used for distributing real-time simulation on different DSPs. We are now preceding with an analytic assessment of the stability analysis at system level, and with tests of the proposed approach applied to different components that, present particular challenges (machines and power converters).