着急呀，大家帮帮忙啊

It is ok for residuals to be either positive or negative. Actually that must be the case since by construction sum of residual = 0 (this is the first order condition for minimizing residual sum squre if intercept term is included). sum of residual = 0 cannot hold if all residuals >0.

模型太简单了吧，试一下非线性的

一般解决过拟合的方法有两种，一是增加样本个数，而是用正则化方法例如岭回归的方式，都可以解决你的问题。可以试试

这么几个数据去做回归模型太粗糙了吧？