Some Researches On Data Rectification For Process Industries

As an intelligent data processing technology to ensure the precision, reliability, and integrity of data, data rectification is one of the key elements of computer integrated manufacture system in process industry. Utilizing the redundancy and process model, it filters the gross errors in the measurements by statistical theory, optimally adjusts the measurements and potential model uncertainty (such as parameters, etc.) to improve the quality of the measurements and estimate unmeasured variables by optimization theory. With the complexity of systems and the growth of demand for information technology, the process model which data rectification is based on is extending to nonlinear and dynamic. However, measurement network, process model and corresponding data rectification algorithm will have significant impacts on the result of data rectification. Therefore, it is urgent to study efficient data rectification algorithms and measurement network design methods for different process models to meet the needs of industrial applications.In this thesis, for complex process model such as bilinear process model and dynamic process model, some researches on data rectification algorithms and measurement network design methods are done. The main works are as follows:1. After presenting the significance of data rectification, various data rectification techniques are briefly mentioned and some major concepts in data rectification are introduced. Finally, the advances in the application of data rectification are reviewed.2. An efficient bilinear data reconciliation algorithm based on multi-component mass balances is proposed to improve the precision of the flows and components. Firstly, utilizing the bilinear structure of the multi-component mass balances, a new unobservable variable elimination approach is developed to decompose and regularize the reconciliation problem. Then, the problem is solved by PSO algorithm after elimination of the constraint equations. The simulation results demonstrate that this algorithm may give more accurate data of the flows and components, avoid theproduction of meaningless data and have good computing efficiency.3. For generalized linear dynamic system, a robust linear dynamic data rectification algorithm is proposed. Firstly, the dynamic system is transformed and discretized to discrete state space model. The state space equation represents the temporal redundancy information, while the output equation represents the spatial redundancy information. Then, the robust estimation theory based on Huber function is used to solve the transformed problem. This algorithm is robust to gross errors or outliers, and it is very useful for online processing. Finally, a case study for a linear dynamic system proves the benefits of the proposed data rectification algorithm.4. Dynamic data rectification has a great relationship with the measurement network. To improve the accuracy of dynamic data reconciliation, a new measurement network design method for generalized linear dynamic system is studied. With the given investment, this method can generate a measurement network which guarantees the highest accuracy of data rectification and observability of all unmeasured process variables.Finally, the thesis is concluded with a summary and discussions of future and prospective data rectification researches.