Researches On Modeling And Algorithms Of Data Reconciliation For Process Industry

Data reconciliation is one of the key elements of CIMS (Computer Integrated Manufacture System) in process industry. By using the redundancy in measurements and process models, it can eliminate mesaurements with gross error (MGEs). reduce random measurment errors and estimate unmeasured variables by using statistical methods. So, data reconciliation can provide reliable information for CIMS.Model is the basis of data reconciliation, and many difficulties encountered in data reconciliation industry applications have strong relationship with models. Consequently, not only from application point of view, but also from theoretical point of view, it is important to do researches on the model-oriented data reconciiation approaches. The main ideas in this dissertation are motivated from industry applications, which are listed as follows:1. Quasi-steady-state linear data reconciliation is defined for mass balance of a whole factory by taking the storage capacity of devices into consideration. Then in order to dealing with the low redundancy of process industry sensor network, the procedure of building an accurate and efficient quasi-steady-state linear mass balance model is discussed in three steps: rough model establishment, model impovement and model validation.2. A method of selecting MGEs that can be deleted is proposed. Because too much MGEs' deletion will result in an ineffective model, MGEs can not be deleted totally at one time. They must be treated differently in order to be deleted as many as possible to ensure the reconciliation precision. MGEs that can be deleted are selected one by one according to their redundacy degrees. In this way, the information in the data can be used sufficiently and the reconciliation precision is ensured. Simulation result demonstrates the efficiency and consistency of the proposed method.3. The application of the proposed method to a refinery is discussed. Application results show the proposed method can deal with MGEs effectively and provide useful suggestion on instruments checking. Thus the number of instruments that must be checked is minimized, the maintenance effort andexpense are reduced.4. A novel objective function of sensor network design is proposed. Both of the prices of raw materials and products and reconciliation precision are integrated into one objective-economic loss. Then, sensor network design is defined as a problem of multi-objective optimization. Both of the cost and economic loss are used as objectives. Objectives evaluation function is made by using virtual objective method and min-max method. Genetic algorithms are used to solve the optimization problem quickly and sufficiently. In simulation part, the effect of three genetic parameters on convergence speed is compared among 56 sets, and the set that achieves the fastest convergence speed is selected for further simulation. And the relative reconciliation precision is defined according to sensor precision, which makes it easy to evaluate the design results. Simulation result demonstrates its feasibility and shows this method accompanied with data reconciliation can significantly reduce sensor network investment.5. A new formulation of .data reconciliation is proposed for hybrid systems as a quasi-steady-state bilinear data reconciliation problem. For those nodes where the scheduling events may happen, scheduling-equations are established and added to the model. In this way, the model's redundancy is improved, which enhances the solvability of the data reconciliation problem. Simulation results demonstrate the efficiency and consistency of the proposed approach.The dissertation is concluded with a summary and prospect of future data reconciliation researches.