Research On Distributed Environment Detection Algorithm Based On Multi-agent Flocking

Multi-agent system has much more advantages than a single agent, as a result, it attracts much attention of researchers to study collaborative multi-agent systems in recent years. And environment field detection based on multi-agent flocking has very good realistic meanings. Some collaborative agents can find maximum value of the environmental concentrations by tracking the gradient of the scalar field (such as nuclear pollution, algal blooms in a lake, toxic gases leakage). Moreover, they can also track some characteristic values to obtain distribution information of the environment field which is very useful to help us to handle environmental pollution and emergent event, and to understand nature.In general, it is impossible to get the prior distribution model of the scalar field. In fact, each agent can only measure the scalar value of environment field, but cannot measure gradient information. Therefore, the agents need to estimate relative algorithm is needed to gradient of the unknown field of interest.Firstly, a distributed environment field maximum location detection algorithm based on multi-agent flocking has been proposed in the thesis. In this algorithm, agents get measurements of the unknown field of interest from their neighbors and themselves. Then, every agent estimates gradient of environment field based on the measurements. Depending on the estimated gradient of environment field, we utilized Newton-Raphson consensus for distributed optimization to obtain input for driving multi-agent to the location of the maximum. At the same time, in combination with multi-agent flocking algorithm, the group of agents located the maximum point of the unknown environment field finally. Compared with existed algorithms, the first advantage of the proposed algorithm is that it does not need prior model of environment field’s distribution, and the second is that agents which perform environment detect task needn’t to keep a fix formation. The validity and robustness of the proposed algorithm has been verified by simulations. Secondly, in order to get more information of the unknown field, we proposed a contour tracking algorithm based on multi-agent flocking on the basis of the maximum location detection algorithm. In the contour tracking algorithm, every agent obtains orthogonal vectors of estimated gradients, then perform consensus average on orthogonal vectors to obtain input for tracking contours. At the same time, in combination with multi-agent flocking algorithm, the group of agents finished specific contour tracking finally. This algorithm discusses two different cases based on whether the measurement larger than expectation value or not at initial position. The validity and applicability of our proposed contour tracking algorithm has been verified by simulations.