It is to stabilize the variance via BC transformation. For different parameter in this transformation, you have different sum of residual squares, then choose the parameter that gives the smallest SSR. In many case, the parameter means a log transformation.

谢谢zhaosweden，我主要是不明白为何除以几何平均数之后，导致线性模型和双对数线性模型的残差平方和具有可比性。