Research on Risk Evaluation and Prediction of Salt Deposition in Gas Storage Based on Machine Learning Models and Methods

With the rapid growth of China’s economy and the gradual transformation of the energy consumption structure, the demand for natural gas is rising rapidly. Experience in natural gas peak-shaving shows that underground gas storage is the most economical and effective method to peak nature gas shaving and ensure the gas smoothly supply. W underground gas storage (UGS) is the largest UGS in central and eastern China. Due to the high salinity of formation water, the gas wells of W UGS may face more serious risk of salt deposition, which seriously affects the operation safety of UGS. In order to diagnose the salt deposition risk of UGS in real time, a more applicable and accurate salt deposition risk evaluation model is established based on machine learning method. Firstly, the modeling data set is obtained using the wellbore salt deposition prediction model. In order to increase the accuracy of the model, the salt deposition risk evaluation model of gas well in UGS is divided into two parts: the wellbore salt deposition judgment model and the wellbore salt deposition risk degree prediction model. Further, the decision tree algorithm is used to establish the wellbore salt deposition judgment model to judge whether the gas well occurs salt deposition. If the salt deposition happened, the wellbore salt deposition risk degree prediction model developed by K-Means algorithm and KNN algorithm is used to predict the gas well salt deposition risk level. Finally, the production parameters of gas well are obtained by numerical simulation method and input into the salt deposition risk evaluation model to evaluate the salt deposition risk of UGS wells. The accuracy of the model is more than 95%, which can accurately predict the salt deposition risk of UGS wells. The model is applied in W UGS, and the result shows that the salt deposition risk of UGS wells are low after two cycles of W UGS operation.