Research on prediction algorithm of shale gas reservoir production

China is rich in shale gas resources, and it is still in the initial stage of development with huge development potential. However, shale gas is different from conventional gas reservoirs in that the physical properties of the reservoir are very poor, and the storage method is special. Multi-stage fractured horizontal wells are often used for development. The development cost and technical requirements are high, and the development is difficult. Our model uses the most realistic and latest shale gas production data in the Weiyuan area. According to the reservoir information and fracturing information parameter values of each well, the commonly used machine learning methods are used to predict the test production of different wellheads respectively. The mean absolute error (MAE) is used to compare different machine learning prediction algorithms. Experiments and results show that the gradient-enhanced regression fitting effect is better, the accuracy of the predicted test production is relatively high, the generalization ability, robustness, and various types of data can be flexibly processed. At the same time, it takes less time to adjust parameters while ensuring the accuracy of the forecast. In addition, the use of big data analysis can help shale reduce mining costs, increase shale gas production, and optimize drilling parameters and fracturing parameters.