The invention provides a reactive power compensation method of a power network. The method comprises the steps of setting different types of line faults on an objective power network system model; setting corresponding reactive power compensation schemes for various line faults on the basis of the line faults; obtaining transient data before and after reactive power compensation separately by using a transient stability simulation program on the basis of the line faults and the reactive power compensation schemes; taking the transient data before and after reactive power compensation as input and output items separately, carrying out training by using a neural network and obtaining a trained reactive power compensation classifier; and combining new line fault information with various reactive power compensation schemes, carrying out grading by using the trained reactive power compensation classifier and obtaining the scheme with the highest score as an optimal scheme. The beneficial effects of accurately and quantitatively determining reactive power compensation places and total amount, reducing human intervention and greatly improving the stability and the security of the power network are achieved.