Integrated approach to predicting rock fracture intensity based on radon tracing and artificial neural network

In geological and engineering practices, determining fracture intensity of rock masses is critical for the exploitation of resources such as oil, natural gas, uranium, and geothermal energy. Due to the lack of technological means to directly measure the distribution of rock fractures, it is very difficult to obtain the rock fracture intensity. This paper proposes an integrated approach to predicting rock fracture intensity based on artificial neural network (ANN) and radon tracing. Firstly, a radon migration model was established to numerically simulate radon exhalation rate of fractured rock masses under different fracture parameters. In the model, rock fractures were generated using the discrete fracture network (DFN). 900 sets of data were numerically calculated as learning data for the ANN using the model. The proposed method has good prediction accuracy with a coefficient of determination of 0.907. The number of hidden layers and neurons are key factors determining the accuracy of model prediction. Finally, the model was used to predict the fracture intensity of a fractured rock mass with outcrop. The predicted fracture intensity is close to the measured value, with a difference of 7.5 %.