Research On Semi-Svm Regression And Harmony Search Algorithm With Its Application

Large amounts of data are generated in modern complex industry, and these data contain a lot of useful information. In order to extract useful information and use them to optimize industrial processes, two kind of intelligent algorithms are considered, which can improve the efficiency of industrial production. Semi-supervised Support Vector Machines recently rise in research,also the Harmony Search algorithm, an optimization algorithm, is recently proposed. Therefore, this paper focuses on the improvement of semi-supervised support vector machine algorithm and harmony search algorithm, and industrial application is conducted in two typical batch processes. The main achievements are divided into the following areas.1. In order to solve the modeling problem without online measurement, a semi-supervised incremental support vector machine regression algorithm based on kernel estimation (SE-INC-SVM) was proposed. Kernel technique was used to label and updated in training process, which can solve the difficulty that unlabeled data are hard to be labeled accurately in the traditional semi-supervised support vector regression. The incremental algorithm was employed to solve the semi-supervised support vector machines, which can largely improve the efficiency. It is shown from the simulation test on manual data sets that, the improved algorithm not only can effectively use the unlabeled data, but also has high generalization ability.2. Two different improved harmony search algorithms are proposed to solve the optimal problem. A parallel dynamic harmony memory harmony search algorithm (PDHS) was proposed to overcome the drawbacks which are common in traditional harmony algorithm. Harmony memory grouping strategy is given for contributing to the slowness and premature here. And dynamical search space harmony search (DSHS) algorithm was proposed in order to solve the problem that search space is fixed in the standard harmony search algorithm. The DSHS employs two strategies which are dynamically changing the search space and ancillary harmony-memory to increase the diversity of harmonies and the global search ability. The simulation results of typical test function show that PDHS and DSHS can greatly improve the search accuracy and speed.3. Two typical batch processes are considered as application object, penicillin fermentation and exothermic batch reaction. First semi-supervised incremental support vector machine regression are used to establish black-box model between process variables and quality. Then DSHS and PDHS are used to calculate the optimal control law in these two different processes, respectively. Better modeling accuracy and improved control effectiveness were shown in this simulation application.