基于测井参数优选的煤层含气量预测模型

Prediction model of coalbed methane content based on well logging parameter optimization

随着煤层气勘探的不断深入,对煤层含气量预测精度提出了更高的要求。基于煤层含气量测井响应特征,分析测井参数与含气量的相关性,提出MIV(MeanImpactValue)技术与LSSVM(Least Squares Support Vector Machine)结合的测井参数优选策略,优选最优测井参数作为网络建模的输入自变量组合,通过粒子群算法优化LSSVM网络核心参数,最后构建一套适用于煤层含气量预测的MIV-PSO-LSSVM模型。在此基础上,分别对比分析LSSVM、PSO-LSSVM、MIV-LSSVM和MIV-PSO-LSSVM模型对煤层含气量的预测性能,并与传统多元回归方法进行了对比,利用拟合优度和均方根误差对此5类模型进行评价。结果表明:PSO优化下的LSSVM模型预测精度得到有效提升,结合MIV方法优选测井参数可大幅度改善神经网络建模性能,MIV-PSO-LSSVM模型可实现煤层含气量高精度预测,为煤层气勘探及其储层评价提供新的技术支撑,且本研究的建模策略及思想可广泛应用于其他机器学习建模研究领域。

With the development of coalbed methane(CBM)exploration,higher accuracy of CBM content prediction is required.Based on the response characteristics of CBM logging,the correlation between logging parameters and gas content is analyzed,and the optimization strategy of logging parameters by combining MIV technology with LSSVM is proposed.The optimal logging parameters are selected as the input independent variables of network modeling,and the core parameters of LSSVM(Least Squares Support Vector Machine)network are optimized by particle swarm optimization.Finally,a set of MIV-PSO-LSSVM model suitable for CBM content prediction is constructed.The prediction performances of LSSVM,PSO-LSSVM,MIV-LSSVM,MIV-PSO-LSSVM and traditional multiple regression method are compared and analyzed respectively.The results show that the prediction accuracy of LSSVM model optimized by PSO is increased,and the modeling performance of neural network is improved significantly with MIV method to optimize logging parameters.MIV-PSO-LSSVM model could realize high-precision prediction of CBM content,providing new technical support for CBM exploration and reservoir evaluation.And the modeling strategy of this research can be widely used in other ML(machine learning)modeling research fields.