Multiple Delay Identification Method Based On Ga And Its Applications In The Carbonization Process

The inherent property like multi long time-delay causes the outputs of the long production process incapable of reflecting the variation of input values and control signals timely. Though the actions of regulator and adjusting institution haven't time delay, the controlled variables are affected and controlled after a pure lag time, which is the characteristic of production process. Since the regulation doesn't work on time, it may enlarge overshoot of the output, increase regulation time, degrade performance of the transition process and make system unstable. Consequently, the control performance and the quality of products are influenced directly, and energy and resource are wasted. Therefore, the identification of multi-time delay parameters is a significant research area of the long production control system.Genetic algorithm (GA) is an optimization method based on biological evolution theory. Its optimization function isn't required strictly in continuity and differentiability. In this work, GA is applied to identify the multi-time delay parameters of the long production process. For the characteristics of individual coding and objective function design, the identification problem is transformed into optimization problem. Using the input and output data of the production process, the identification of delay parameter is achieved by efficient parallel searching in the feasible solution space. The influences of operation parameters on the identification precision of the time delay parameters and the time complexity are researched by Matlab simulation.According to alumina continuous carbonation decomposition process (ACCDP) reaction mechanism analysis and technological characteristic, the dynamic model of decomposition rate gradient determined by the carbon dioxideis addition amount in each tank is established. Then the identification of multi-time delay parameters based on GA is applied to identify the multi-time delay of the dynamic model. Finally, experiments on the industrial input and output data have demonstrated the accuracy of the proposed identification algorithm.