Aiming at characteristics of complexity and nonlinear in enterprise financial data, this paper puts forward a financial early warning model based on General Regression Neural Network which is optimized by Logistic chaos mapping Variable step size Fruit Fly Optimization Algorithm. Firstly, the Logistic chaos mapping is used in Fruit Fly Optimization Algorithm to modify the initial value. Secondly, based on optimal initial value, we modify the step size of FOA in order to find the best Spread. Finally, we analyze the forecast data and select representative indicators. LVFOA shows better ability of global optimization and fast convergence, and it improves the prediction accuracy of GRNN. The simulation results show that the warning accuracy of new model is higher than GRNN model and FOA-GRNN model, better fitting the complex financial data.