Combining Integrated Simulation Software with Advanced Optimization Methods
We love to hear from our users on how they are using the decision-making power of Simul8 in their projects, to improve processes, tackle big challenges and make a real impact.
It was great to hear from Jonas and Angel, part of the ICSO-Meta research group on how they are combining simulation software with metaheuristics and reinforced learning via python, to demonstrate the potential of integrated simulation software with advanced optimization methods to aid decision-making.
Can you tell us a little bit about you and your university?
Of course. We work as a part of the ICSO-Meta research group. The research group is an interdisciplinary team of researchers, both PhD and PhD candidates, from top universities in Spain. We have experience in Data Analysis, System Optimization and Reinforcement Learning in various contexts such as Logistics and Transportation, Manufacturing, Finance, Intelligent Cities or Healthcare.
Can you tell us about your project/the aim?
The idea of the project is to bring together reputed simulation companies with our research group, in order to generate new solution methods for real life complex problems. These new methods are specially well suited for concepts such as Intelligent Systems, Industry 4.0 and Digital Twins.
Can you describe the metaheuristic technique in simple terms/your own words?
When problems become big and complex, with many interacting components, like in a warehouse, the performance of classical solution methods is limited. Furthermore, in real life systems, the uncertainty about some of the parameters that controls such systems is always present. The type of metaheuristic that we are developing is called “sim-heuristic” and it combines simulation with heuristics, to optimize the performance of such complex and realistic systems.
What’s the aspect of the project that excites you the most?
Probably the fact that we could provide high quality solutions to very complex problems that are faced by people working in industrial contexts, which are also robust to real life variability.
So, how does Simul8 play a part in your project?
Simul8 can have a significant role in this project. First of all, it can help saving a significant amount of time during the creation of the simulation component of the sim-heuristic. Rich discrete-event scenarios can be modelled quickly and effectively with Simul8. Secondly, it can help with the validation of the scenarios and the solutions, since experts can interrogate the model created and easily interpret the results obtained.
How do you measure the value of the project?
The main value of the project is the research innovation produced. This means that both the academic impact of our work, and the implementation of these methodologies in different real-life customers applications can be used to measure the impact of this research.
Were there any challenges?
One challenge is the fact that there needs to be an efficient communication protocol between the simulation component and the heuristic component. Finally, the alignment between the heuristic and the simulation is important, since an excessively simplified heuristic or a very complicated simulation model can impede finding good quality solutions. Finding a balance between both components and still represent accurately the modelled environment is a real challenge.
How could this research impact the simulation industry?
We believe that it can stimulate the collaboration between simulation companies and universities and research groups. This could, for instance, enhance the simulation products by integrating these optimization methodologies in their software. Also, the fact that what simulation companies are willing to make their product being able to interact with other technologies, can encourage decision makers to consider simulation as the high value-added activity that it is.
The main difference between the Simheuristic method that we are currently developing and other applications like Optquest is the level of interaction between the simulation and the optimization. On the one hand, in applications like Optquest, only the final result after the simulation has run to completion is used for finding the best combination of parameters. In that sense, they are two separated and sequential components. On the other hand, in our Simheuristic proposal, partial information is extracted as the simulation progresses, in order to better guide the optimization algorithm. This means a closer interaction between both the simulation and optimization components.
How will this research change the way decisions can be made?
One possible area that could change is the degree of automation for some strategic decisions, where human decision-making is important nowadays. We humans are good at dealing with uncertainty, but our capacity for calculating combinations of different options is limited compared to computers. Providing methods that can be easily interpreted and trusted by decision-makers can change the current paradigm to a more automated one.