电缆地层测试泵抽关键参数预测方法研究OA

In the operation of sample extraction by cable formation testing pump,key parameters such as the breakthrough time of oil and gas,the stable pumping time and the purity of the sample directly affect the sampling decision.At present,the acquisition of these parameters mainly relies on the subjective experience of the operators,which easily leads to poor sampling timing and makes it difficult to balance sample quality and operation efficiency.Based on the pump sampling data of 158 wells in the western South China Sea,this paper uses the Pearson correlation coefficient method to determine the sensitive factors of oil and gas breakthrough time and pump stable time.On this basis,combined with the weight coefficients of each sensitive factor determined by the random forest algorithm,a multi-parameter fusion pump factor expression is constructed,and then a prediction model for oil and gas breakthrough time,pump stable time,and sample purity suitable for different instruments is established.The research results show that:①Porosity,mobility,pump speed,and drilling fluid specific gravity/formation pressure coefficient are the sensitive factors affecting pump sampling.Porosity,mobility,and pump speed are negatively correlated with oil and gas breakthrough time and pump stable time,while drilling fluid specific gravity/formation pressure coefficient is positively correlated with oil and gas breakthrough time and pump stable time.However,the determination coefficients of each single factor are relatively low,and the fitting effect is limited.②The determination coefficients of the multi-parameter fusion pump factor with oil and gas breakthrough time and pump stable time are all greater than 0.75,and the fitting degree is much better than that of a single sensitive factor.③Through on-site application verification at 16 pump sampling points of 12 wells in the western South China Sea,the errors of the oil and gas breakthrough time and pump stable time predicted by the prediction model based on the pump factor and the actual values are all less than or equal to 10 minutes,and the error of the predicted sample purity and the experimental analysis results is less than or equal to 10%.The conclusion is that the prediction model proposed in this paper can meet the rapid decision-making needs of on-site operations and solve the problem of the lack of quantitative guidance in pump sampling operations to a certain extent.