The invention discloses a gravitation-search RNA-GA-applied catalytic cracking main fractionate tower neural network modeling method, and belongs to the intelligent modeling field. The catalytic cracking main fractionate tower is a multiple-input-multiple-output system. The gravitation-search RNA-GA-applied catalytic cracking main fractionate tower neural network modeling method includes the steps: establishing an RBF neural network model through the input-output data of samples; taking one part of the sample data as a training sample, taking the sum of the absolute values of error between the output data of the RBF neural network model and the corresponding practical data as a fitness function; utilizing the gravitation-search RNA genetic algorithm to perform optimization searching on the radial primary function center of the catalytic cracking main fractionate tower neural network model to obtain the optimal solution of the radial primary function center of the RBF neural network model; and then obtaining a neural network model for the catalytic cracking main fractionate tower. The gravitation-search RNA-GA-applied catalytic cracking main fractionate tower neural network modeling method can obtain an ideal effect in the experiment, and is also suitable for modeling of other complicated systems.