目的 :研究色谱重叠峰解析新方法。　方法 :构造了以修正高斯模型 (EMG)为基函数的径向基函数神经网络 (EMG-RBFNN) ,在网络学习算法中提出采用基于共享小生境技术的约束最优保留两阶段遗传算法 :结构学习和参数最速梯度下降进化 ,从而使EMG-RBFNN具有结构学习能力 ,使该神经网络能够适应组分数未知的色谱重叠峰解析。尤其是在将气相和高效液相色谱EMG模型参数τ、σ和tR 之间的近似线性关系作为可行域约束条件引入算法后 ,极大地限制了算法可行的解空间 ,减少了局部最优解出现的概率 ,提高了算法运行效率。　结果 :将此新方法分别应用于计算机仿真色谱和 45例中药材HPLC ,算法可以以较高的精度解析出仿真色谱 ,实际色谱峰解析也有很高的解析精度。　结论 :该方法非常适用于组分数未知的各种色谱 (重叠 )峰的解析 Objective: To study a new method of chromatographic overlap analysis. Methods: A radial basis function neural network (EMG-RBFNN) based on the modified Gaussian model (EMG) was constructed. In the network learning algorithm, a constrained optimal two-phase constrained genetic algorithm based on shared niche technology was proposed. Structure Learning and the most steep gradients of parameters evolve, so that EMG-RBFNN has the ability of structure learning, which makes the neural network adapt to the analysis of chromatographic overlapping peaks with unknown composition. Especially, introducing the approximate linear relationship between the parameters τ, σ and tR in gas and high performance liquid chromatography (EMG) model as the feasible region constraint condition greatly limits the feasible solution space of the algorithm and reduces the appearance of local optimal solution The probability of improving the efficiency of the algorithm. Results: The new method was applied to the computer simulation of chromatography and 45 cases of Chinese herbal medicine HPLC, the algorithm can be resolved with higher precision simulation chromatography, the actual chromatographic peak resolution also has high resolution accuracy. Conclusion: This method is very suitable for the analysis of various chromatographic (overlapping) peaks with unknown composition