Abstract: Some large and medium-sized irrigation water pumps in China have run for a long time, and their existing dynamic characteristics are significantly different from the design conditions. To obtain more accurate dynamic characteristics of these pumps, first we obtained the raw data of dynamic characteristics from the raw material of pump source, design data, and efficiency curve of the motor, etc. Second, the dynamic characteristics were corrected using a static correction method of exponential decay and a dynamic training method of the extension neural network in combination with the measurements of pump sets in practical operation. Finally, we developed a large and medium-sized irrigation pumps dynamic characteristics analysis code V3.0 that integrates the two methods into a single module. Practical application shows that the modified equation of pump power characteristics that is used in this code, can describe the real engineering conditions. And typical characteristic curves produced by the code have a reasonable spacing and follow the same trend without intersecting. The data fitting errors of the code are low and the actual results of the code applications are satisfactory. This study would lay a basis for further study on the optimization of joint operation of the pumps and the gates.