1. There is no substantial difference between linear regression and nonlinear regression, both are looking for suitable parameters to meet the laws of the existing data. The fitted equation (model) is generally used for the calculation of the internal difference or a small range of external differences.2. There is generally an internal connection between Y and X, such as E=m*c^2. So relevant information can be collected before the regression, or can be directly applied.3. Scatterplots are made between Y and each X, and their correspondences are observed. If it is linear, change the parameters can be applied to linear regression; otherwise, non-linear regression can be considered.4. Linear regression can be directly calculated by the method of least squares of the corresponding coefficients, the coefficients of the hypothesis test (H0: b = 0, Ha: b0), to exclude the influence of the variables of the small, and then regression can be considered; non-linear can be considered to the transformation of X or Y, such as logarithmic, squared, square, exponential and so on, as far as possible, transformed into a linear regression can be.5. 5. Refer to the goodness-of-fit R^2 and variance S, to have a certain understanding of the accuracy of the model.