| Due to the influence of many factors, the chemical process data obtained through chemical manufacture devices or measuring meters may corrupted by kinds of noises, which makes the physical equations of chemical units cannot be met. Error of data can be divided into random error and gross error. And accurate data is a solid foundation for effective monitoring, modeling, scheduling and decision analysising. In order to obtain more accurate and reliable data, it is necessary to do data reconciliation. The object of data reconciliation is to eliminate random errors and gross errors in measurement data using statistic methods, recognition technologies and optimization technologies. The research jobs of this paper focus on gross error detection, optimization algorithm for data reconciliation and the application of data reconciliation technology in chemical process.Firstly, the development and the application of data reconciliation are reviewed. Secondly, according to the fact that the measured variables with gross errors sometimes exist in the steady state process, a gross error detection method based on LM-BP neural network is developed. Two examples verify the feasibility of this algorithm. Thirdly, a modified algorithm (CMDE) is proposed in this paper to improve the searching performance of differential evolution (DE) algorithm, and CMDE is tested on large number of benchmark functions and compared with the basic DE algorithm and other two modified DE algorithms. The simulation results demonstrate that CMDE is superior to other three algorithms. And then it is applied in data coordinate of a chemical process without gross error. Finally, a simplified model is established based on the research of the reaction principles and flow of the methanol synthesis process. The data which needs to be reconciliation is reconciled with the modified algorithm in this paper. The reconciled results indicate the validity of the method in this paper. |
PDF Full Text Request