Simulation model for routes in public transport systems using parallel computing in IoT environments

Abstract

Background: There are a large number of tools for the simulation of traffic and routes, which take into account the different simulation models (macroscopic, microscopic, and mesoscopic). Unfortunately, these simulation tools are limited when simulating a complete public transport system, which includes all its buses and routes (up to 270 for the London Underground). The processing times for this type of simulation increase in an unmanageable way, since all the relevant variables that are required to simulate consistently and reliably the system behavior must be evaluated. Different studies nationwide have encountered this problem. In these, tools have been generated for the simulation and optimal allocation of routes in mass transport systems such as Transmilenio in Bogot´a, Colombia, concluding that current simulation models cannot simulate systems of public transport completely. This problem is because they represent a large-scale NP-hard problem, in which the time required for simulation increases exponentially when a new element is added. At present, the processing capacity of modern computers is no longer measured based on the clock frequency of their CPUs, since the limit allowed by silicon has been reached (around 4GHz). Today the processing capacity depends more on the number of cores and the amount and speed of RAM. Unfortunately, to the date, there are no enough models for the simulation of the behavior of routes in public transport systems that take advantage of the benefits of parallel processing provided by modern computer systems, such as GPUs, or other parallel processing architectures. Therefore, we found that there is a lack of simulation models that can handle the behavior of all routes of a public transport system efficiently and consistently, taking advantage of parallel computing. Aims: The main aim of this project is the purpose of a new public transport systems’ routes simulation model for parallel computing architecture in IoT environments. This aim comprises the following objectives: • Identify the input and output variables for a routes simulation model in public transport systems, from IoT environments. • Define a parallel computing architecture that is efficient for the execution of route simula-tion models in public transport systems. • Construct a new simulation model for the defined parallel computing architecture, which, through the identified input and output variables, allows the simulation of routes in public transport systems to be carried out efficiently and consistently. • Validate the consistency of the new simulation model using other simulation models already validated. • Evaluate the performance of the new simulation model, in contrast to other models executed in different architectures. Methods: A scientometric tool called ScientoPy was built to support the state of the art analysis. In state of the art, we reviewed the applications for the Internet of Things focusing on Smart Cities and then in Intelligent Transport Systems. Then, we review the applications for the three main parallel processing architectures: GPU, FPGA, and multi-core to find parallel implementations for public transport simulation. Next, a validation public transport simulator was build entirely in Python with dynamic lists to emulate the passengers’ queues in the stops and buses. We named this as Pure Python simulator. The behavior of this simulator was validated with the simulation output data consistency and with the graphical real-time 3D output that shows the passengers and buses’ movement in the system. Then, we designed and built a parallel simulation model called Masivo PSC (Parallel Simulation Core) then performs concurrently the arrival, boarding, and alighting operations per each stop in a separate work item that runs in OpenCL. Furthermore, we validated the Masivo PSC simulation outputs with the Pure Python simulator outputs for 4 scenarios. Finally, we extracted the simulator performance indicators for the most complicated scenario with 300 stops and near to 500,000 passengers. Results: In the validation results, we found that the relative error for the total alighted passengers and the total average commute time is not greater than 0.7 % in all of the 4 tested scenarios. The performance results of Masivo show a speed-up factor of 10.2 compared with the simulator model running with one compute unit, and speed-up factor of 278 times faster than the Pure Python validation simulator. The real-time factor achieved was 3050 times faster than the 10 hours simulated duration. Conclusions: A new simulation model for routes in public transport systems using parallel computing in IoT environments, called Masivo PSC was generated. Masivo works with a predefined public transport system conditions, which include the stops total number, stops’ capacity, and the OD matrix. This OD matrix and routes information can be updated to this model via CSV files. Masivo gets the simulation results for total alighted passengers and average commute time. Similarly, it shows the performance indicators.