A Dynamic Soft-sensing Method Based on Impulses Response Template and Parameter Estimation with Modified DE Optimization

As major experimental modeling methods, static soft-sensing methods have been widely used in modern chemical production process now. In fact, for the sampling frequency to output variable by laboratory analyzer off-line is rather low and uniform, the computed results, which gained from those methods or the existed major dynamic methods such as neural networks, are difficult to satisfy the requirements of dynamic control on-line. A dynamic soft-sensing method (DSSM) based on impulse response template (IRT) and parameter estimation using differential evolution (DE) optimization is presented in this paper. However, for a multi-variables system, learning of template parameters still takes large computational cost, and is not only slow in the convergence speed but also easy to be trapped into local optima so as to enlarge the modeling errors. To account for these problems, the original DE (ODE) is modified in the aspects of scaling factor and crossover rate, which could dynamically change with iterative loops. Subsequently, a complete implementation of the modified DE (MDE) is presented. Experiment based on hysys simulation of a primary tower system to build a three-inputs-single-output model is carried out, under various impulse response length and noise standard, and the final comparison results demonstrate the effectiveness and robustness of this method.