Pathmind Reinforcement Learning for Simulation webinar

Mining simulation

Pathmind is a platform that enables AnyLogic users to integrate reinforcement learning into their AnyLogic simulations. This webinar demonstrates how to set up Pathmind reinforcement learning in an AnyLogic model and train an AI policy in the Pathmind web application. We also share two example models that showcase how reinforcement learning can outperform baseline heuristics and other optimization tools.

The webinar is hosted by AnyLogic Simulation/AI Program Lead Arash Mahdavi. Pathmind’s Head of Customer Success Edward Junprung and Lead Simulation Engineer Sahar Esmaeilzadeh highlight the benefits of Pathmind reinforcement learning, go over the workflow, and break down the example models.

Reinforcement Learning for Simulation

Simulations that contain variability, large state spaces, and multiple contradictory objectives will see the most benefits from reinforcement learning. Reinforcement learning is also not limited by human preconceptions on how a problem should be solved and does not require substantial rewrites when information changes, making it a powerful asset for simulation modelers.

The Pathmind/AnyLogic Workflow

You can think of Pathmind as a bridge between AnyLogic and the reinforcement learning ecosystem. Pathmind requires no prior experience with AI or Data Science and takes care of the reinforcement learning so you can focus on your simulation.

During the webinar, we show you the simple steps of adding reinforcement learning capabilities to an AnyLogic model using the Pathmind Helper palette item. AnyLogic’s RL Experiment feature is then used to export the model directly into the Pathmind Training web application. Once training is complete, the AI policy is loaded in AnyLogic for validation.

Pathmind Reinforcement Learning for Simulation webinar. AnyLogic Simulation/AI Program Lead Arash Mahdavi hosts Pathmind’s Head of Customer Success Edward Junprung and Lead Simulation Engineer Sahar Esmaeilzadeh. Watch on YouTube.

Pathmind Example Models

At the conclusion of the webinar, we go over two example models that showcase how Pathmind reinforcement learning can outperform other optimization methods.

Multi-Echelon Supply Chain features a network of manufacturing centers, distributors, and retailers with an inventory optimization problem. The network needs to figure out which location to order goods from and how much product should be ordered to balance inventory holding and transportation costs. The baseline for the model is an (r, Q) inventory optimization hybridized with a nearest-neighbor heuristic. Pathmind reinforcement learning beat that baseline to increase order serviceability from 65% to 85% and profit by 34%.

Automated Guided Vehicle (AGV) Powered By AI uses reinforcement learning to optimize dispatching routes for a fleet of AGVs in a manufacturing center. Parts must be dropped off at the correct machines in a specific processing sequence to maximize total output. Variables such as processing times, equipment failures, and supply arrivals make a static heuristic ineffective for optimizing the AGVs. Pathmind AI outperformed a shortest-queue heuristic to improve results by nearly 78%.

To take a closer look at how we integrated reinforcement learning into these models, visit the Multi-Echelon Product Delivery and Automated Guided Vehicle (AGV) tutorials. You can also sign up for a free Pathmind account and start adding AI to your own models.

Related posts