Research On Predictive Mechanisms Among Dynamic Behaviors Of Biological Flocks/Swarms

A large number of biological studies show that natural biological flocks/swarms usually have some predictive intelligence. Each agent in biological groups may not only make the next step behavioral decision based on the currently acquired information, but also predict leaders and neighbors'future behaviors according to locally historical information. Thus predictive mechanism has the potential to greatly improve the collective behavior of flocks/swarms.There are two different types in natural groups, social groups and industrial multi-agents systems: One is with leaders, and the other is without leaders. Two predictive mechanisms are designed accordingly in this thesis, analyzing in detail and verifying improvements in flocks/swarms properties such as the synchronization, the cohesion and the effective communication cost reduction.Therefore, for groups with leaders, the thesis makes use of small-world-type network structural properties. It realizes a novel predictive control strategy in flocks with leaders, which is carried out by the A / R (Attractive/Repulsive) model. The strategy reduces the negative influence of the time delay among the flocks by introducing some long range connections, which therefore improve the consensus properties (including velocity consensus and position consensus). Statistical physical experiments may prove the strategy's existence. Another interesting rule is the complementary relations between the number of long links and predictive steps. Increasing predictive steps could compensate for insufficient long links'number, by contrast, adding long links could compensate for the lack of adequate network predictive ability.On the other hand, an effective Model Predictive Consensus Protocol - MPC Consensus Protocol for flocks/swarms without leaders is designed in the thesis. Specifically, for a traditional consensus strategy proposed by Olfati-Saber and Murray, the future states differences among flocks/swarms could be minimized by predicting several future steps states on each node, using the method of the receding horizon optimization. The predictive mechanism is capable of greatly improving the consensus convergence'speed and sharply decreasing sampling frequency, which remarkably reduces flocks/swarms communication costs required to achieve a prescribed consensus velocity. It has been proved by means of matrix theory.Consequently, the investigation is of strategic significance to breaking through bottleneck of flocking performance. The research from the aspect of natural science provides strong evidences for verifying the existence of natural flocks/swarms predictive intelligence, and reveals that the natural flocks/swarms collective communication means is the interval rather than continuous. Moreover, the predictive mechanism may make a crucial effect in consensus performance, thus play an important role in the formation and evolution of flocks/swarms. From an industrial application point of view, the thesis's theoretical achievement may have the suitable promoted value in the domain of some relevant practical engineering like multi-agent robots systems, unmanned air vehicles (UAVs), sensor networks, traffic flow system, social network and congestion control in communication networks.