针对非线性最小二乘法的传统算法在皮肤组织太赫兹波段德拜模型参数提取时, 对参数初始值的要求很高, 且所得模型与原始数据偏差较大的问题。将遗传算法应用于非线性最小二乘法对原始数据进行拟合, 并提取参数。遗传算法随机产生初始解, 不需要设置参数初始值, 能简化参数提取过程; 其搜索过程是从多个可能解开始, 而不是单一初始值, 所以不易陷入局部最优, 基本能实现全局优化。以皮肤组织德拜参数提取为例, 将遗传算法应用于非线性最小二乘法进行参数提取时不仅减小了文献中的模型在0.15~0.6THz频段与原始数据的偏差, 而且在整个频率范围与原始数据都能较好吻合, 验证了遗传算法提取德拜参数的有效性和皮肤组织太赫兹波段德拜模型的正确性。

The existing extraction procedures of nonlinear least squares can generate the double Debye model of skin tissue, but the extraction procedures need an exact initial value of the parameter, the model tracks the measurements poorly at low frequencies. The genetic algorithm is applied to data fitting to extract the parameters of double Debye model of skin tissue. There is no need for genetic algorithm to set initial value of the parameter because it can generated possible initial parameters randomly, which is simplified parameters extraction process. The search process starts from a number of possible parameters, rather than a single initial value. So it can be uneasy to fall into local optimum, and the global optimization is achieved. The model optimized by genetic algorithm not only reduces the deviation at the low frequencies from 0.15THz to 0.6THz, but also is consistent with the original data at the whole frequencies. The results confirm the effectiveness of the method and viability of the double Debye model.