Guangzhou Zhao

A new game model for distributed optimization problems with directed communication topologies

Abstract In this paper, the distributed optimization problems of multi-agent systems with directed communication topologies are investigated by using the game theory. In particular, a new general non-cooperative game model, termed state based weakly acyclic game, is provided to solve the problem. Based on this approach, the desired global objective is achieved by designing local objective function for individual agent to make coordination decisions. It is worth noting that all the obtained equilibria are thus solutions to the proposed distributed optimization problems with directed and time-varying communication topologies. Simulations on consensus problem in multi-agent systems are provided to verify the validness of the proposed methodology.

A game theoretical formulation for distributed optimization problems

The focus of this paper is to develop a theoretical framework for analysis and design of distributed optimization problem in multi-agent systems by using the language of game theory and cooperative control methodology. In the framework, a piecewise-constant and binary-valued matrix in the cooperative control theory is introduced to describe the sensing/communication among agents and to cope with the practical situations where the information sharing may be in a distributed, dynamically changing and local manner. Based on information acquisition/communication model, state based ordinal potential game is designed to capture the optimal solution to distributed optimization problems in multi-agent systems by appropriately specifying local objective function for each individual decision maker. It is worth noting that the proposed analysis and design methodology has the advantages that the resulted equilibriums are capable of solving the distributed optimization problems even if the corresponding communication topologies is local, time-varying and intermittent. Meanwhile, the minimal requirement for the communication among the agents is provided to ensure the global objective is desirable under the new framework.

Recursive Model Predictive Control for Fast Varying Dynamic Systems

A well known drawback of model predictive control (MPC) is that it can only be adopted in slow dynamics, where the sample time is measured in seconds or minutes. The main reason leads to the problem is that the optimization problem included in MPC has to be computed online, and its iterative computational procedure requires long computational time. To shorten computational time, a recursive approach based on Iterative Learning Control (ILC) and Recursive Levenberg Marquardt Algorithm (RLMA) is proposed to solve the optimization problem in MPC. Then, recursive model predictive control (RMPC) is proposed to realize MPC for fast varying dynamic systems. Simulation results show the effectiveness of RMPC compared with conventional MPC.

Distributed Optimization and State Based Ordinal Potential Games

The focus of this paper is to develop a theoretical framework to analyze and address distributed optimization problem in multi-agent systems based on the cooperative control methodology and game theory. First the sensing/communication matrix is introduced and the minimal communication requirement among the agents is provided. Based on the matrix communication model, the state based ordinal potential game is designed to capture the optimal solution. It is worth noting that the proposed methodology can guarantee the distributed optimization problem converge to desired system level objective, even though the corresponding communication topologies may be local, time-varying and intermittent. Simulations on a multi-agent consensus problem are provided to verify the validness of the proposed methodology.

A Parameter Identification Scheme for Second-Order Highway Traffic Model Based on Differential Algebraic Methodology

In this paper, we develop a fast on-line and parameter identification scheme for the second-order macroscopic traffic flow model. The proposed parameter identification scheme is devised in the framework of the algebraic identification, with differential algebra and operational calculus as major mathematical tools. Compared to conventional methods, the new identification scheme allows the parameters of second-order macroscopic traffic model, namely free speed and critical density, be estimated in an on-line and computationally efficient fashion are identified by means of differential algebra and operational calculus. The simulation example of a hypothetical scenario demonstrates these advantages numerically.

Speed control for a permanent magnet synchronous motor with an adaptive self-tuning uncertainties observer

This paper presents a robust speed control method for a permanent magnet synchronous motor (PMSM) drive. The controller designed from conventional field-oriented vector control method with constant parameters will result in an unsatisfactory performance of a PMSM due to dynamic uncertainties such as changes in load and inertia. In this paper an adaptive self-tuning (ST) observer that estimates dynamic uncertainties on-line has been developed, where output is fed forward as compensation to the PMSM torque. The stability of the observer is studied using the discrete-time Lyapunov theory. A performance comparison of the proposed controller with the constant parameter based conventional field-oriented vector controller is presented though a simulation study, which illustrates the robustness of the proposed controller for PMSMs.

A Game Strategy for Power Flow Control of Distributed Generators in Smart Grids

We consider the distributed power control problem of distributed generators(DGs) in smart grid. In order to ensure the aggregated power output level to be desirable, a group of DGs with local and directed communications are expected to operate at the specified same ratio of their maximal available power output. To that end, the non-cooperative game is introduced and the DGs are modeled as self-interested game players. A new game model, termed state based weakly acyclic game, is developed to specify decision making architecture for each DGs, and at the point of the equilibrium of the game, the global objective of the power control problem can be achieved through autonomous DGs that are capable of making rational decisions to optimize their own payoff functions based on the local and directed information from other DGs. The validness of the proposed methodology is verified in simulation.

MMSVC: An efficient unsupervised learning approach for large-scale datasets

We propose a multi-scale, hierarchical framework to extend the scalability of support vector clustering (SVC). Based on multi-sphere support vector clustering, the clustering algorithm called multi-scale multi-sphere support vector clustering (MMSVC) works in a coarse-to-fine and top-to-down manner. Given one parent cluster, the next learning scale is generated by a secant-like numerical algorithm. A local quantity called spherical support vector density (sSVD) is proposed as a cluster validity measure to describe the compactness of the cluster. It is used as a terminate term in our framework. When dealing with large-scale dataset, our method benefits from the easy parameters tuning (robustness of parameters with respect to the clustering result) and the learning efficiency. We took 1.5 million tiny images to evaluate the method. Experimental result demonstrated that our method greatly improved the scalability and learning efficiency of support vector clustering.

DP-PMK: an improved pyramid matching kernel for approximating correspondences in high dimensions

As the feature dimension increases, the original pyramid matching kernel suffers from distortion factors that increase linearly with the feature dimension. This paper proposes a new method by consistently dividing the feature space into two subspaces while generating several levels. In each subspace of the level, the original pyramid matching is used. Then, a weighted sum of every subspace at each level is made as the final measurement of similarity. Experiments on data set Caltech-101 and ETH-80 show that compared with other related algorithms which need hundreds of times of original computation time, dimension partition pyramid matching kernel only needs about 4–6 times less of original computation time to obtain the similar accuracy.

An improved adaptive sliding mode observer for sensorless control of PMSM

In this paper, an improved adaptive sliding mode observer is presented for achieving sensorless vector control of permanent magnet synchronous machine. Compared to conventional sliding mode observer, the proposed observer based on sliding mode observer and model reference adaptive system can effectively improve the precision of the rotor position estimation. The simulation and experiment results show that the improved adaptive sliding mode observer is feasible and effective.