In order to explore the energy efficiency of geothermal energy application under karst geological conditions in hot-summer and cold-winter zone, the system energy efficiency ratio and unit energy efficiency ratio of ground source heat pump system are predicted based on the optimized back propagation neural network model by genetic algorithm (GA-BP) under the condition that system load rate is below 30% in summer. It is verified that the GA-BP model has higher prediction accuracy by the prediction error evaluation indicators. The relationship between short-term test and medium and long-term evaluation of energy efficiency ratio is studied by the prediction model. The results show that the relative error of the predicted value and the calculated value is ±5%. The predicted values of system energy efficiency ratio and unit energy efficiency ratio are predicted by the GA-BP neural network model. All prediction error evaluation indices are smaller than BPNN prediction model. This shows that GA-BP model can be used to predict energy efficiency of ground source heat pump system under karst geological conditions. The optimal time for conducting short-term monitoring is 14:00~16:00 of the day or 13 days in total in July and August and meet the unit load rate of 60% ~ 70% based on the prediction model. The short-term predicted values of system energy efficiency ratio and unit energy efficiency ratio for are applied to the medium and long-term efficiency ratio evaluation. The relative error of the evaluated energy efficiency ratio is within the allowable range.