A graphical solution of multimodal optimization to improve food properties

Abstract

When the search spaces were set excessively broad, the random‐centroid optimization (RCO) was difficult to home‐in on the global optimum in the case of multimodal functions. A preliminary narrowing strategy was introduced based on the general trend of steeper slopes toward the global optimum than those toward local optima. The search spaces were narrowed to half for factors at either end by comparing response values at both extreme terminals (0.1 and 0.9 in the full space of 0‐1.0) of the search space (half‐space design). The differences (or ratios) of these two values represented the slopes. This new method was more reliable than repeating probabilistic random search to narrow search spaces prior to RCO as reported in the previous paper (Nakai et al., 1998a). However, for most of the ordinary optimization projects with reasonable sizes of the search spaces, there was no need of using this half‐space design search. Application of the RCO, by changing the amount of FeCl2 to add to fish actomyosin at a low concentration and heating temperature and time, efficiently maximized the breaking strength of the gels formed.