«Systems of Systems»

To create complex systems, for example, for managing a large plant, “Russian Railways” trains or international supply chains, our company is developing a network-centric approach and building a multi-agent new-class «system of systems», designed to show, in fact for the first time ever, the possibilities of co-evolution in self-organized systems.

In this approach, any complex multi-agent resource management system, designed to solve problems of distribution, planning, optimization and control of resources in real time, recursively unfolds as a self-organizing network of self-similar multi-agent schedulers with peer-to-peer interaction. These systems use service architecture and a common data bus for coordinating decisions (information highway through which systems exchange data required for making decisions, counter offers, requests and acknowledgments, etc.).

The final solution in a network-centric system is obtained by matching individual solutions of subsystems, each of which works for its purpose and performs its tasks.


For example, a complex multi-agent planning system (hereinafter referred to as the «scheduler») of a large plant with 20-30 workshops, may consist of several schedulers of individual shops interacting with each other in the same way as individual agents interact within one such scheduler. However, in this case, the whole «swarm» interacts with other «swarms».

Network-Centric Approach-fig 1

Network-centric architecture of a distributed smart enterprise resource management system in real time

In practice, this means that an unforeseen event in one of the enterprise schedulers will be processed as soon as possible and taken into account in the plans of this scheduler. However, if this does not work and the solution affects the plans of other schedulers, then the interaction process is initiated, possibly with a wave of negotiations for resolving the conflict. In case of resolution and settlement of this conflict, it allows such a network of schedulers to continuously maintain relevance of interconnected plans even with any turbulence changes in the environment.

For example, let us consider possible schemes of interaction between schedulers of production and transport shops of a large industrial enterprise.

Network-Centric Approach-fig 2-en

Example 1: Production shop delays execution of a previously agreed product, since a new urgent order has appeared. In this case, the truck that was previously planned for transportation of the finished product to the client is re-planned so as not to stand «at the gate» waiting for the production shop, and not lose money during idle time. In this situation, the production shop may be offered another, less profitable, truck, as a substitute (it can be of a larger volume and therefore more expensive), and the shop will incur the costs, saving overall profitability at the expense of the new order.

Example 2: Transport, which was planned for delivering the finished product to the client, is a little late due to breakdown of the truck. Then the production shop reschedules its work, and has enough time to additionally execute the new order that has just appeared: it can deliver the new order together with the first order, reducing transportation costs.

Example 3: For an important order, the completion date is scheduled and now its delivery to the client is being planned. During the planning, it turns out that transportation can take so long that the delivery time will be violated and the plant will have to pay a penalty for delay. In this regard, the production scheduler must find a way to finish the product earlier and assess how this will affect other orders. After that, the scheduler will decide which option is more profitable.

These examples clearly represent the key principle of network-centric approach: «Solve problems as local as possible and as global as required».

History and Prospects of Development

The network-centric approach began to actively develop in military applications, since it is not possible to solve all problems through the center during deployment of a large-scale military campaign. Moreover, no one can predict in advance where the problem might arise and which of the interacting systems would have to respond first, offering a solution and coordinating it with other systems.

It is important to note that the systems under consideration must be initially developed as multi-tier systems, involving not only horizontal but also vertical interactions and negotiations, for example: an industry as a network of plants; a plant as a network of shops; a shop as a network of sections; a section as a network of workers’ schedulers.

This approach makes it possible to build fundamentally new complex «systems of systems» with fractal-like structures from autonomous but coordinated schedulers that fully utilize principles of self-organization and evolution at all levels, and that is why they form a single complex «organism» of the enterprise, operating on the basis of new intelligent real-time systems.

Implementation of the discussed principles of system interaction is State-of-the-Art — the highest level of complexity and skill in development and programming of modern systems, requiring a combination of knowledge and skills in the theory of complex adaptive systems, mathematical models, methods and algorithms for Distributed Problem Solving, design of next-generation autonomous intelligent systems, parallel and asynchronous programming, and real-time decision support.

At the same time, we consider this approach as a fundamentally new basis for building smart Internet of Things that creates prerequisites for Industry 5.0.

Advantages of This Approach

Advantages of smart resource management systems based on the network-centric approach are the following:

  • Openness to adding new subsystems;
  • High quality of solutions: interrelated plans are coordinated;
  • Flexibility of planning — changes can be initiated from either side;
  • Data reliability: support of relevance of plans between subsystems;
  • System productivity (if necessary, your server — for each scheduler);
  • System scalability (lower levels are not visible to the higher ones);
  • Reliability: failure of one subsystem does not stop the work of the entire system;
  • Survivability: reaching the result (with less accuracy) even with loss of lower-level systems;
  • Efficiency: less development and support costs.

 Additional Information

For more details on the principles of the proposed network-centric systems, please contact our training and consulting center at info@kg.ru.

We also highly recommend our book «Managing Complexity», published by WIT Press in 2014.