Predictor-based Neural Dynamic Surface Control For A Class Of Strict-feedback Nonlinear System

Strict-feedback nonlinear systems are common in civilian and national defense construction,such as manipulators,aircraft,ships,etc.,which have important research significance for the intelligent control of this kind of systems.Neural network(NN)-based adaptive control method is widely used in the control of strict-feedback nonlinear systems.However,there are still some problems such as poor transient control performance and a large number of learning parameters.This treatise focuses on these problems,combined with predictor,minimal number of learning parameters(MNLPs)and other techniques,and predictor-based NN adaptive dynamic surface control(DSC)strategies are proposed.According to the context from single system to multi-agent system(MAS),the output tracking,bipartite consensus and formation of strict feedback nonlinear systems are considered respectively.The main work of this paper includes the following aspects:1)For a class of strict feedback nonlinear single-input-single-output systems with unknown dynamics and unknown control gain,a predictor-based NN adaptive DSC strategy is designed.The predictor,Nussbaum function and NN DSC method are combined to deal with the output tracking control problem.Due to the small initial error and additional adjustable parameters,prediction errors have faster convergence speed compared with surface errors.The prediction error is used to replace the error surface to update the weights of NNs,which can avoid the high frequency signal with large adaptive gain due to the large initial error and the slow convergence speed of the surface error.This strategy not only extends the limitation of the existing predictor based control methods on the unit control gain,but also combines with the MNLPs technique to reduce the number of learning parameters.By a Lyapunov function,it is proved that the closed-loop signal is ultimately bounded when the initial state of the system is bounded.2)For a class of strict-feedback nonlinear MASs,a predictor-based NN dynamic surface bipartite consensus control strategy is designed.According to the knowledge of structurally balanced topological graph theory,the MNLPs technique based on neighbor information is considered,and the predictor-based NN DSC is introduced into the bipartite consensus of MASs.Furthermore,for the strict-feedback MASs under sensor attack,the output feedback secure bipartite consensus control strategy is designed according to the security control condition of sensor sparse attack,and the predictor-based NN control strategy is applied to the output feedback problem.To deal with the system subject to nonsymmetric input saturation,an auxiliary system with the same order as the system is designed,which acts on the control law of each step of the backstepping to compensate for the input constraint.With the help of an input-to-state practically stability(ISp S)-Lyapunov function,it is proved that the bipartite consensus error converges to the small neighborhood of the origin and the closed-loop system signal is bounded.3)For a class of nonlinear multi-agent systems with external disturbances,a bipartite formation control strategy based on disturbance observer is designed.A nonlinear disturbance observer with predictor-based NN is proposed to approximate the generalized disturbance including external disturbance.By using the NN weight norm learning method,the MNLPs technique combined with predictor is improved,and the lemma based on continuous function separation theorem in the control strategy design is cancelled,which reduces the difficulty of control strategy design and stability analysis.Through ISp S analysis,it is proved that the bipartite formation error converges to the small neighborhood of the origin and the closed-loop system signal is bounded.4)For a class of second-order multi-dimensional nonlinear MASs,a formation control strategy of collision avoidance and connectivity preservation is designed by combining formation error transformation and artificial potential field method.The formation control strategy framework is constructed by NN DSC based on predictor and disturbance observer.The formation error transformation equation is used to achieve the desired formation,collision avoidance and connectivity preservation among agents.A logarithmic potential function based on obstacle avoidance distance polynomial is proposed to provide repulsive force for agent obstacle avoidance.Compared with the existing potential function,the proposed potential function is continuously differentiable at the breakpoint,and the function changes smoothly when the obstacle enters and leaves the obstacle detection range.It avoids the sharp increase/decrease of the existing potential function in the control law when the obstacle enters/leaves the detection range.By combining the control strategy of collision avoidance and connectivity preservation based on error transformation and artificial potential field method,the use of potential functions is reduced.By constructing Lyapunov equations,the formation error convergence without obstacles and the obstacle avoidance effect with obstacles are analyzed respectively.