Consensus of multiagent systems with kalman filter applications

In recent years, interest in systems or tools that can make their own decisions and implement these decisions has begun to grow. Especially with the realization of the Industrial 4.0 revolution, many applications that raise productivity in production, improve in-plant organization and raise workers' health are left to robotic or unmanned solutions. These robotic or unmanned systems consisting of a large number of elements or vehicles are called multi-agent systems in the technical literature. It has been shown in this study that these multi-agent systems, which are formed by bringing more than one vehicle or element according to the neighborhood relation, are more advantageous than the systems consisting of the elements which are independent of each other and which decide independently. Increased interest in autonomous systems has also increased the need to know and instantly control the parameters (speed, location, etc.) of the elements that make up these systems. Taking individual information of the elements in the system and controlling them according to this information is especially troublesome for systems consisting of a large number of elements. In order to solve this problem, multi-agent systems with network topology will be discussed. In the thesis study, detailed information about specific properties of multi-agent systems with network topology is provided. According to this, the elements forming these systems have network topology and share the information they have with the elements forming the system with the help of various relation rules. In the thesis study, this neighborhood relation is emphasized and detailed information about information transfer topology is provided The members of the multi-agent systems have to work together to accomplish a certain task or achieve a goal. Detailed information about the meaning and conditions of consensus and cooperation is provided in this study. In order to be able to cooperate in such systems, a specific consensus rule or protocol has to be established. In the thesis study, it has been shown how consensus and cooperation are provided according to the parameters and equations in systems with different dynamics. These rules, which allow the elements in the system to agree on a common value and act accordingly, are all called the consensus algorithm. The definition of consensus and conditions under which consensus is achieved is given in detail. Detailed information about graph theory and consensus conditions under the graph theory is provided. Consensus of multi-agent systems is analyzed with respect to graph theoretic concepts. Sample problems and solutions have been included in order to better show how and under what conditions a specific system reaches consensus. In the study conducted, the advantages and disadvantages of the multi-agent system components according to the individual studies of the collective coordination and working together are also revealed. It has been observed that the performance of multi-agent systems based on the consensus algorithm is higher. In the thesis study, it is explained how the elements of multi-agent system structure share the information they have with other members. It is analyzed how invidual agents behave when noise is added to the communication links between them. Despite the noise in the system, Kalman filter applications have been implemented in order to ensure that the elements in the system are nevertheless neatly connected to the system dynamics. The main work in this thesis is to investigate the effect of noise (process and measurement noise) in consensus and how to suppress this noise using Kalman filter. Three different systems are analyzed to ensure the consensus of these multi-agent systems are obtained under noisy conditions. Detailed information and equations is given to know the dynamics and consensus conditions for these multi-agent systems. In order to better explain the theory of the applied Kalman filter, information has been given about probability and distributions. In particular, detailed information has been provided about the normal distribution which will be evaluated as a disturbing input. Therefore, it is shown how the estimates depend on the parameters. At the same time, parameters of how the linear state estimation methods are derived and connected are explained. Detailed informations about the parameters used in Kalman filter equations are given. The operation diagram of the Kalman filter algorithm is described. In case the system is nonlinear, standard Kalman filter cannot be used. In this case extended Kalman filter can be employed to estimate system states. In the thesis, all these information data are explained in detail with multi-agent systems with 3 different dynamics and different parameters in various directions. The equations and parameters of the mentioned systems are explained. The information transfer topology and applied consensus protocols for these systems are described in detail. Noise is added to the information exchanged between these elements and the transfer of this information, and Kalman filter applications are performed to suppress this noise. The obtained results are illustrated by graphs. The values of the number of agents are changed and the results are given comparatively. The elements in the multi-agent systems will share the information they have by communicating within a specific neighborhood relationship. The probability of systems being able to communicate, that is, their neighbors, will be determined randomly, depending on their proximity. Each system considered has a specific control protocol. The parameters required for this control protocol are set iteratively by means of adaptive PD controller which is tuned by the gradient descent algorithm. It is assumed that the information exchange between agent is noisy and therefore, the systems are simulated by adding noise to these informations. Consequently, signals to observe the effects of this noise, Kalman filter is applied to suppress this noise, so that smoother information is transmitted to the elements forming the multi-agent.