Comparison of Optimization Techniques for Application to Food Product and Process Development

摘要

ABSTRACTThe super‐simplex optimization was modified by incorporating a quadratic regression subroutine. The new algorithm was found to be more efficient in converging at the optimum than the following algorithms: fractional factorial designs, sequential single‐factor search, pattern‐search method, Morgan‐Deming simplex and super‐simplex. The efficiency of the modified super‐simplex optimization was dependent on the‘fitness of the regression equation to the model equation. Response surface analysis was accurate only when the model equation was quadratic factorial and the boundary covered the optimum. When a constraint was imposed to restrict the search within the boundary, unlike other simplex algorithms including the weighted centroid method, the modified super‐simplex algorithm circumvented a problem of the search stalling at the boundary. Since it is not unusual for optimization to be restricted within boundaries, the modified super‐simplex algorithm could be useful in optimizing food processing and analysis. When the modified super‐simplex technique is applied to ‐food research, even higher optimization efficiency could be expected by incorporating backward stepwise multiple regression analysis instead of the quadratic regression subroutine with regression equations of set formulae. To obtain a regression equation fit better to the true response surface, all information, including graphic