Data driven tensor product-based modeling and control of a damped Lotka-Volterra predator-prey process

In the world of engineering, controlling a machine or a biological system often feels like trying to steer a ship through a storm while the map keeps changing. To guide a system—whether it is a robot arm, a chemical reactor, or a population of animals—engineers need a mathematical model, a set of rules that predicts how the system will react to different inputs. For simple, predictable machines, these rules are fixed and easy to write down. But for complex, living systems where things change based on their current state, the rules are fluid. The challenge has long been how to create a reliable map for these shifting landscapes using only the data we can measure, without needing to know every hidden detail of how the system works inside. If we can build a good enough map from observation alone, we can then design a controller that gently nudges the system onto a desired path, keeping it stable even when it tries to wander off.

This is the core puzzle tackled by Árpád Varga, a researcher at Obuda University in Hungary, who set out to test a new way of modeling and controlling a classic biological scenario: the predator-prey relationship. The specific system he studied is a damped version of the Lotka-Volterra model, a famous mathematical description of how populations of predators and prey rise and fall together. In nature, if there are too many predators, they eat too many prey, causing the prey population to crash, which eventually leads to the predators starving and their numbers dropping, allowing the prey to recover. This cycle can be chaotic, but in this study, the system was "damped," meaning it naturally settles down over time rather than oscillating forever. The researcher's goal was to see if he could take measurements of these populations, build a simplified model of their behavior using a specific mathematical technique called tensor product modeling, and then use that model to force the populations to stay at a specific, desired level, effectively taming the wild swings of nature.

To do this, Varga first had to figure out how to describe the system without knowing the exact biological constants that govern it. Instead of guessing the rules, he treated the system as a black box that he could observe. By measuring the current number of prey and predators, and also measuring how fast those numbers were changing at every moment, he could calculate a snapshot of the system's behavior. He used a mathematical trick, known as a pseudo-inverse operation, to reverse-engineer the relationship between the current state and the change in state. This gave him a rough, instantaneous picture of the system's internal logic. However, this raw data was messy and hard to use for control. To make it useful, he fitted smooth polynomial curves to the data points, essentially drawing a continuous surface through the scattered measurements to represent how the system behaves across different population levels.

The researcher then tested a more advanced method to represent this same data: tensor product modeling. Imagine the data as a complex, multi-dimensional shape. Instead of describing this shape with a single, massive equation that requires storing thousands of numbers, tensor product modeling breaks the shape down into a grid of smaller, simpler building blocks. It calculates the system's behavior by looking at how these blocks combine based on the current population numbers. The idea is that this approach might be more efficient, especially for very complex systems with many variables, because it can capture the essential behavior with fewer stored numbers than a traditional polynomial equation. In this study, the researcher compared the traditional polynomial fit against this new tensor product approach to see which one created a better map for the controller to use.

The results of the computer simulations showed that both methods worked well within the range of the data they had seen. The models could accurately predict how the system would behave, allowing a control algorithm, based on a principle called computed torque control, to guide the populations to a target number. This control method works by constantly calculating the difference between where the system is and where it should be, then applying a corrective force to close that gap. In the simulations, both the polynomial model and the tensor product model successfully kept the predator and prey populations steady at the desired levels, with very small errors. The tensor product model did show a slight disadvantage in this specific, simple case: because the system only had two variables (predators and prey), the tensor method actually required storing more numbers than the polynomial method. The researchers noted that the advantage of the tensor approach is likely to appear only in much more complex systems with many more variables, where the traditional method would become too heavy and slow to use.

To prove that this wasn't just a computer fantasy, Varga built a physical version of the experiment using an analog computer, a device that uses electrical circuits to simulate mathematical relationships in real time. He connected this hardware to a modern industrial controller, creating a hybrid system where the physical circuit represented the predator-prey dynamics and the digital controller applied the corrections. The setup included voltage signals representing the populations, with positive voltages adding to the population and negative voltages representing hunting or removal. The controller used the tensor product model, loaded into its memory, to calculate the necessary adjustments. When the system was left alone, the populations oscillated as expected. But when the controller was switched on, it successfully guided the populations from their chaotic starting points to the specific, steady values the researcher had chosen.

The physical experiment confirmed that the data-driven approach works in the real world, not just in theory. The controller managed to stabilize the populations, though the tracking was slightly less perfect than in the computer simulations, which is typical when moving from a clean digital environment to a physical one with electrical noise and hardware limitations. The researchers found that the system could handle different target values, moving the populations to new steady states smoothly. They also demonstrated that the controller could adapt to changing targets in real time, a crucial capability for practical applications. The study concluded that while the tensor product method did not offer a memory advantage for this simple two-variable system, the overall methodology of building a model directly from measurements and using it for control is robust and effective.

The work highlights a shift in how engineers might approach complex systems. Instead of spending years trying to derive exact equations for every biological or chemical interaction, it is possible to observe the system, learn its behavior from the data, and construct a functional model on the fly. This approach is particularly powerful for systems where the underlying physics are too complicated to write down perfectly but where the behavior can be measured. The study suggests that for simpler systems, traditional methods might still be sufficient, but as systems grow in complexity, the ability to break down the behavior into manageable, weighted components could become essential. The research stands as a proof of concept that data-driven modeling can successfully tame the unpredictable rhythms of nature, offering a new tool for managing everything from wildlife populations to industrial processes.