Performance Assessment Theory And Application Of Non Gaussian Systems Based On Dynamic Data Reconciliation

With the improvement of the automatic level of industrial production,the real-time and accuracy requirements of the control algorithm are higher and higher.In order to keep the control system running in the best state for a long time,it is necessary to evaluate its control performance accurately.Therefore,the performance evaluation of control system is of great significance to industrial process design.The system will inevitably be affected by non-Gaussian noise in the process of operation.The existing performance assessment standards based on variance are not suitable for the actual industrial process because they cannot fully describe the non-Gaussian random characteristics.In recent years,a minimum entropy performance evaluation benchmark has been proposed for non-Gaussian stochastic control systems.However,the influence of measurement noise is not considered in the established benchmark.In fact,due to the bad effects of voltage instability and sensor aging in industrial environment,the measurement noise is inevitable and may not obey Gaussian distribution.Data quality is the basis of control performance assessment,and the existence of measurement noise would lead to the establishment of control performance assessment benchmark cannot accurately measure the quality of control performance.Dynamic data reconciliation(DDR)can improve the quality of transient measurement data and has great potential in the performance monitoring of control system.In this paper,the DDR method and the minimum entropy control technology are combined to establish a new benchmark of control performance assessment for non-Gaussian stochastic system.The main contributions can be summarized as follows:(1)For non-Gaussian systems with single input and single output(SISO),a DDR method is proposed to deal with non-Gaussian measurement noise.Based on DDR,a minimum rational entropy controller is designed.Minimum rational entropy(MRE)is used to formulate a new benchmark for control performance assessment(CPA)of SISO non-Gaussian systems.The proposed CPA method is applied to a single input single output wind energy conversion system.The simulation results show that the proposed DDR method is superior to theextended Kalman filter.Compared with the minimum variance CPA benchmark,the proposed CPA benchmark can more accurately reflect the control performance of the system.(2)At present,most of the actual industrial processes are multi-input multi-output(MIMO)systems,and the variables are coupled with each other,so it is difficult to design the controller.In this paper,for multivariable non-Gaussian systems,considering that the inverse of multi-dimensional noise covariance matrix may not exist,leading to the problem that the reconciled data cannot be obtained,a multivariable dynamic data reconciliation(MDDR)method is proposed,aiming at the coupling between system variables,a minimum joint rational entropy(MJRE)controller is designed,based on which a new CPA benchmark of multivariable non-Gaussian systems is established.The proposed CPA method is applied to two input two output wind power generation system,and compared with the multivariable minimum variance CPA benchmark,the effectiveness and superiority is verified of the proposed CPA benchmark.(3)The performance assessment framework based on DDR-minimum rational entropy is applied to the organic Rankine cycle(ORC)system.Considering that the complexity of the system will slow down the convergence rate of parameters in the DDR method,and then affect the real-time control effect,the DDR method is improved,and the accelerated dynamic data reconciliation(ADDR)method is proposed.The simulation results show that the convergence rate of ADDR method is faster than the convergence rate before the improvement,and for the organic Rankine cycle system,the CPA benchmark of the proposed method is better than the minimum variance CPA benchmark.