Eat the cake and have it too
High detail city-scale electro-mobility simulations require significant computational resources and time to produce results. In spite of the availability of dedicated HPC and cloud-based resources, optimization is essential to obtain results in a timely and cost-effective manner. However, at the same time care has to be taken that the fidelity of the simulation does not suffer. Significant time and resources can be saved by diverting computational resources from areas of less activity towards those that require more resources for accurate simulation. How exactly can this be achieved? Multi-resolution modeling.
So, in the context of simulation, what exactly is resolution? Resolution refers to the level of detail of the model. Traffic simulations are traditionally categorized into three, or sometimes, four types based on resolution,. At one end of the spectrum, in macroscopic models, the concept of individual vehicles (agents) is not present; instead, the traffic of a city or region is quantified in terms of traffic density and flow. This is ideal when the goal is to get an overview of the traffic network as a whole. At the other end of the spectrum lies microscopic models, in which vehicles are modelled individually. Here driver-vehicle units use decision algorithms for planning paths, changing lanes and overtaking. In between these two extremes are mesoscopic models, which can have, depending on the use-case a more microscopic approach for some features and a macroscopic approach for others. An example is, when the internal components of a vehicle are of interest, whereas the behavior (acceleration, lane change,...) is not the main purpose of the simulation. The traffic conditions are simulated macroscopically and only the current traffic velocity on a certain road is taken into consideration for the internal model simulation of a vehicle. All vehicles on a road have the same velocity input.
To rephrase what was said earlier: the trick to optimal use of computational resources lies in using a lower resolution models for areas that require lesser level of detail and fidelity (accuracy) and a higher resolution model in other cases. There are several interesting research problems to be solved in order to achieve this kind of fast high-fidelity multi-resolution traffic simulation. One of the biggest is what happens when a microscopic agent travels from a zone of higher resolution into a zone of lower resolution and vice? The latter (aggregation) is trivial; the former (disaggregation) is a difficult problem since new information has to be generated . Moreover, different strategies of when and where lower resolution can be used without compromising fidelity have to be explored. Within the upcoming articles in this topic we will present approaches to the above problems as well as some results of how much speedup can be achieved with little loss of fidelity.