Data-driven Predictive Control Methods For Complex Process Systems

With the development of industrial technology,the industrial process systems become more complex than before and it's more difficult in building the accurate mechanism models of these process systems.Hence,the data-driven approach has been obtained widespread attention since it emerged.Data-driven control also turns into focus of study.Simply,the data-driven control is a method from data to design controller directly.Model predictive control(MPC)has been attractive for decades in control theory field.It has become more established as the one of the choices for the control architecture in the industry,especially with the improvement of computational capabilities of processors.The most important features of the MPC are: it can handle multivariable control problems,it considers input and output constraints,and it adapts structural changes.Subspace identification is one of the available system identification algorithms for the state-space modeling,which does not require the use of the tedious modeling mechanism because the certain state-space model is obtained just once and there is enough input-output data for the processing.Moreover,the subspace predictors obtained by the subspace identification algorithm can be used to derive the prediction output for predictive controllers,eliminating the intermediate step of process model identification and providing a method of data-driven predictive control which has been applied in some industrial processes and has achieved good results.The main works and contributions of this thesis are as follows:(1)The subspace identification method is introduced and the PO-MOESP(past output-multivariable output error state space)algorithm for the industrial processes is presented.The PO-MOESP is derived from the basic subspace identification algorithm and is of interest when one wants to jointly model the deterministic and stochastic part of the system.The efficiency of this algorithm is illustrated by the data of negative pressure in the power plant.(2)A new data-driven adaptive predictive control method ensuring adaptation of closed-loop systems is presented.The method can offer an attractive alternative for industrial nonlinear,time-varying systems of long period in closed-loop condition and there is no need for obtaining the system explicit model which can reduce the complexity.As the prediction model output is the key element of the predictive controller,we propose to derive such model based on the subspace predictors which are obtained through the closed-loop subspace identification algorithm driven by input-output data.Taking advantage of transformational system model,the closed-loop data is effectively processed in this subspace algorithm.By combining the merits of receding window and recursive identification methods,an adaptive mechanism for online updating subspace predictors is given.The subspace predictors are derived by recursive method using a fixed modest size of data set with receding window method.The proposed mechanism can sufficiently fade the influence of the old data better than only recursive method and bring less computation load than only receding window method.Further,the data inspection strategy is introduced to eliminate the negative impact of the harmful(or useless)data on the system performance.The problems of online excitation data inaccuracy and closed-loop identification in adaptive control are well solved in the proposed method.Simulation results show the efficiency of this method.(3)For continuous-time system,linear parameter varying(LPV)system and continuous-time LPV system,we are able to extend the subspace identification to design the predictive controller and obtain key subspace predictors to design the data-driven predictive controller in continuous-time,LPV and continuous-time LPV systems respectively.It makes the controller more attractive in practical industrial processes.1)A model predictive control method for continuous-time systems based on subspace identification is presented.Firstly,it's developed by reformulating the continuous-time systems using Laguerre filters to obtain the subspace prediction output.Next,the subspace predictors are derived by QR decomposition from input-output and Laguerre matrix.They are then used to construct the incremental future prediction output and the continuous subspace predictive controller is designed.Finally,the process control simulations of an industrial evaporator system show the effectiveness of the proposed method.2)A model predictive control method based on subspace identification for LPV systems is presented.The LPV systems are first introduced to derive the subspace prediction output through subspace identification algorithm.Then,the subspace predictors can be obtained by QR decomposition from input-output matrix.They are used to design the model predictive controller.It's shown that the integrated action is incorporated in the controller to eliminate the steady error.Finally,the simulation of flapping dynamics of a wind turbine system is provided to verify the effectiveness of the proposed controller.3)A data-driven predictive control method based on subspace identification for continuous-time LPV systems is presented.It is developed by reformulating the continuous-time LPV system which utilizes Laguerre filters to obtain the subspace prediction of output.The subspace predictors are derived by QR decomposition of input-output and Laguerre matrices obtained by input-output data.The predictors are then applied to design the model predictive controller.We control the continuous-time LPV systems to obtain the attractive performance with the proposed data-driven predictive control method.The proposed method is applied to a wind turbine to verify its effectiveness and feasibility.(4)A novel data-driven predictive control method based on the subspace identification of the Hammerstein-Wiener systems operating in the open-and closed-loop conditions is presented.The main features of the proposed method can be summarized as:1)Only input and output data(present and past recent data stored in the system)are used to generate the required predictive control action,and there is no need for precise mathematical model e.g.(A,B,C,D)for control design;2)It is applicable for the both open-and closed-loop systems that they are transformed into the uniform forms in order to obtain the subspace predictions of the outputs via recursive substitution;3)The subspace predictors obtained through the QR factorization are employed to make the method simple and robust;4)The incremental nonlinear functions is directly integrated into the MPC cost function,significantly facilitating the design procedures and simplifying the online computations.The method is developed using the reformulation of the open-and closed-loop Hammerstein-Wiener systems.The subspace predictions of the outputs are derived for both conditions using recursive substitution of the Hankel matrices.Since the output nonlinearity is presented by polynomial representation,the subspace predictors are obtained using the QR decomposition and additional algebra methods.The predictors are applied to the model predictive controller,wherein the integrated action is successfully incorporated.A forth-order model simulation test is introduced in order to show that the identified model,obtained from the subspace predictors,performs well with the proposed method.The effectiveness and feasibility of the proposed controller is verified by control application to the fermentation bioreactor system.The key subspace predictors are all obtained through the above three data-driven predictive control methods,but the system models are different,the Hankel matrices are different,and the subspace predictors are also different.In(2),the subspace predictors are just derived through QR decomposition from input-output data matrices.In(3),The subspace predictors are derived from input-output and Laguerre matrices obtained by input-output data.In(4),The subspace predictor can be obtained from the nonlinear function matrices.