Controlling complex systems with artificial intelligence

Controlling complex systems with artificial intelligence

Researchers at ETH Zurich and the Frankfurt Faculty have developed an artificial neural network that can fix challenging handle problems. The self-​learning technique can be made use of for the optimization of source chains and generation procedures as well as for intelligent grids or website traffic manage programs.

Image credit score: Pixabay (Cost-free Pixabay license)

Energy cuts, money community failures and supply chain disruptions are just some of the numerous of difficulties ordinarily encountered in advanced methods that are incredibly tricky or even unattainable to manage employing present approaches. Management units primarily based on artificial intelligence (AI) can help to optimise sophisticated processes – and can also be utilised to build new company designs.

Alongside one another with Professor Lucas Böttcher from the Frankfurt University of Finance and Management, ETH scientists Nino Antulov-​Fantulin and Thomas Asikis – each from the Chair of Computational Social Science – have created a versatile AI-​based command procedure referred to as AI Pontryagin which is designed to steer advanced devices and networks in the direction of sought after goal states. Employing a mixture of numerical and analytical solutions, the scientists demonstrate how AI Pontryagin immediately learns to handle techniques in near-​optimal strategies even when the AI has not previously been knowledgeable of the ideal remedy.

Self-​learning management method

Fluctuations in intricate programs are capable of triggering cascades and blackouts. To keep away from this kind of incidents and enhance resilience, technique professionals have devised a broad assortment of regulate mechanisms and regulations usual apps involve voltage regulate in ability grids, for case in point, or pressure screening in economic institutions. And nonetheless it is not generally doable to control advanced dynamic programs by guide intervention.

In their paper, the researchers exhibit how AI Pontryagin quickly learns quasi-​optimal manage indicators for complex dynamic methods. The researchers’ assessment lays a lot of the important groundwork further exploration is still expected to ascertain the system’s applicability to unique, genuine-​world cases. At current, command techniques are commonly utilised to, for instance, shield power grids from fluctuations and outages, handle epidemics, and optimise source chains.

Source-​chain regulate as achievable application

To use AI Pontryagin as meant, the AI should 1st be offered with info on the goal system’s dynamics. In offer chains, this could possibly incorporate information of the number of probable suppliers, as properly as obtaining fees and turnaround times. This details is used to identify which places involve dynamic optimisation.

People ought to also provide information on the system’s original standing, this kind of as recent stock concentrations, and its preferred (goal) position, these as the requirement to replenish inventory to particular stages whilst minimising the use of methods.

The textual content is dependent on a push release of the Frankfurt College of Finance and Administration


Böttcher L, Antulov-​Fantulin N, Asikis T, AI Pontryagin or how synthetic neural networks master to control dynamical techniques, DOI: 10.1038/s41467-​021-27590-

Source: Eidgenössische Technische Hochschule Zürich