Research On Control Strategy Of Premixed Natural Gas Engine Based On Model Predictive Control

Distributed generators driven by natural gas engines have the advantages of fast startstop process and quick load response,and play an important role in power supply and smart grid construction.The stable operation of natural gas engines is of great significance to reduce emission pollution,improve uninterrupted power quality,and meet power generation requirements.Aiming at the inaccurate modeling of the traditional control method of natural gas engine and the difficulty of system coupling,the model predictive control method is adopted to improve the engine’s reference tracking and anti-load disturbance performance.The main research contents are as follows:Aiming at the problem that the traditional single-input premixed air-fuel ratio(AFR)model does not consider the influence of intake mass flow,resulting in insufficient model accuracy,a dual-input AFR model is proposed.On the basis of considering the fuel throttle opening,the model derives the AFR expression under different intake mass flow,and proposes an AFR correction parameter related to the intake gas mass to integrate the above expressions to form a double-input formation.The complete engine model including the new AFR dual-input model is verified by bench test,and the results show that the accuracy of the new model is significantly improved compared with the traditional model.Aiming at the problem of many iterations and heavy computing load in online solution of optimal problem in the nonlinear model predictive control method,a switching mechanism strategy based on warm start is proposed.This strategy uses a unconstrained single-cycle algorithm to solve the problem when the optimal solution does not trigger the constraints,and uses the sequential quadratic programming(SQP)algorithm only when the constraints are triggered,and uses the warm start strategy to obtain an initial value which is closer to the optimal solution to speed up the iteration process.The simulation results show that the switching mechanism strategy based on warm start can effectively speeds up the iteration speed of the solution algorithm and reduces the number of iterations.Aiming at the problem of insufficient control performance in the loading and unloading process due to the unidentifiable load torque in the linear model predictive control method,a gain scheduling strategy based on an adaptive Kalman filter is proposed.The adaptive Kalman filter automatically adjusts the covariance coefficient to quickly obtain smooth estimated load torque and equilibrium point information.The gain scheduling strategy uses the equilibrium point information to calculate the gain scheduling coefficient matrix and corrects the controller output to improve transient response performance.The simulation results show that the performance of this method is better than that of nonlinear model predictive control at a fixed speed point.A hybrid model predictive control strategy was proposed to solve the problem of insufficient reference tracking performance of linear model predictive control in the case of speed operating points switching.In this strategy,a speed nonlinear model predictive controller is designed separately for the case of speed operating points switching.For the case of a fixed operating point of the AFR in the whole process,it is separated from the original model and simplified into a linear model,and the linear model predictive controller is designed.After splitting and simplifying the model and reallocating computing resources,the prediction horizons of these two controllers are all extended.Simulation results show that hybrid model predictive control has better comprehensive control performance than single linear or nonlinear model predictive control.Finally,based on the actual engine bench test,the above control strategies are all verified.The results show that the engine has good reference tracking performance and load disturbance resistance,and the comprehensive control performance of the hybrid model predictive control strategy is the best.