Reducing the need for communication in remote, natural waveform co-simulations of electrical power systems: An adaptive approach

Electrical power systems are becoming more interconnected and technologically diverse to accommodate ever increasing shares of non-dispatchable generation. These changes are imposing new requirements on the simulation of electrical power systems. One of these requirements is that simulations integrate models of different subsystems, developed by different experts, from different organizations, which may not wish to disclose the information embedded in their models. This, to study the interactions between neighboring, interconnected grids, or between existing grids and new devices. Another requirement is that they reproduce phenomena in a wider range of timescales, to study the interactions between subsystems with slow and fast dynamic behavior. One way for electrical power system simulations to comply with these requirements is with remote, natural waveform co-simulation. Co-simulation is a model integration approach in which each subsystem is simulated in a different simulator. These simulators exchange interface variables, at runtime, to represent interactions between the subsystems. Since the simulators can interact remotely, over a communication network, co-simulation has the advantage that the organization that owns the model needs not disclose it. It also has the advantage that each model can be simulated with the simulator for which it was intended, so organizations that use different simulators can collaborate without having to translate their models. If such a co-simulation is performed using natural waveform models, at a high time resolution, then it is also possible to reproduce a wide range of timescales, from slow to very fast phenomena. But the fact that such a co-simulation is performed remotely, with the communication delays this entails, and that its high time resolution translates into a high communication rate, make it rather slow. Thus, it is desirable to reduce the need for inter-simulator communication. In this thesis I explore a solution to this communication challenge, based on the hypothesis that slower phenomena are easier to predict. If the co-simulated phenomena can be classified as predictable according to some criterion, it should be possible to find expressions that predict interface variables, and that each simulator can use to compute its own inputs instead of expecting inputs to be communicated. I propose a criterion for classifying phenomena as predictable or unpredictable, as well as methods for finding these expressions based on an interpolated Fourier transform and Taylor-Kalman filters. Additionally, I propose a co-simulation algorithm where the simulators compute their own inputs while the co-simulated phenomena are predictable. After applying these ideas to the co-simulation of two different test systems, I was able to reduce the need for communication up to 60%. A co-simulation framework with these characteristics is a step towards more descriptive models and better performing simulations, and a tool that increases our ability to take better advantage of existing energy infrastructure, as well as to develop it further.