神经网络算法在致密油压裂后产能预测中的应用

Application of neural network algorithm in tight oil productivity prediction after fracturing

开展致密油压裂后产能预测方法研究,指导采油工程方案中压裂施工参数优选。传统方法主要是应用解析公式或数值模拟方法进行产能分析与评价,存在求解公式繁琐、参数获取困难、计算过程复杂等问题。为此,引入灰色关联分析、遗传算法改进神经网络,建立致密油压裂后产能预测模型。收集99口直井缝网压裂井的测井参数、压裂施工参数及压裂后第一年累计产油量,建立样本数据集。利用灰色关联分析算法,明确影响产能的地质因素和工程因素。应用遗传算法优化神经网络的层间权值和层内阈值,构建3层神经网络模型,测试样本产能预测相对误差为7.25%,满足工程应用的精度要求。模型应用结果表明,基于神经网络算法预测致密油压裂后产能是可行的,能够定量优化压裂施工参数,在采油工程方案编制中具有一定的实用性和推广性。

Research on tight oil productivity prediction method after fracturing has been conducted to guide the optimization of fracturing parameters in oil production engineering scheme.The traditional method mainly uses analytical formula or numerical simulation method to analyze and evaluate the productivity,which has many problems,such as complicated formula,difficult parameter acquisition and complex calculation process.Therefore,grey correlation analysis and genetic algorithm were introduced to improve the neural network to establish the tight oil productivity prediction model after fracturing.The logging parameters,fracturing operation parameters and cumulative oil production in the first year after fracturing were collected from 99 fractured wells with vertical fracture-network to establish the sample data sets.The geological factors and engineering factors affecting the productivity were clarified by using the grey correlation analysis algorithm.The genetic algorithm was used to optimize the weight between layers and the threshold within layers of neural network,and the three-layer neural network model was constructed.The relative error of test samples for productivity prediction was 7.25%,which met the accuracy requirements of engineering applications.The application results of the model showed that it was feasible to predict the productivity of tight oil after fracturing based on neural network algorithm,which could quantitatively optimize the fracturing operation parameters,with certain practicability and popularization in the preparation of oil production engineering scheme.