改进LSSVM在水平井产能预测中的应用

Application of LSSVM Improved Horizontal Well Productivity Prediction

在水平井开发可行性论证及水平井优化设计中,产能预测是重要依据,常规产能预测方法,由于样本较少和影响因素多,预测精度难以保证。为此,采用最小二乘支持向量机方法回归出预测模型进行水平井产能预测。最小二乘支持向量机对标准支持向量机进行了改进,把不等式约束改为等式约束,把误差平方和损失函数作为训练集的经验损失,把解二次规划问题转化为求解线性方程组问题,较好地解决了水平井产能预测样本少、影响因素多的问题。引入粒子群优化算法来优选最小二乘支持向量机中的参数组合,既克服了交叉验证法耗时长的缺点,又发挥了最小二乘支持向量机的小样本学习能力强和计算简单的特点。以大庆油田某一区块10口水平井的生产资料作为样本,采用最小二乘支持向量机方法回归出预测模型,对该区块两口水平井的产能进行了预测,结果表明,预测产能与实际产能的最大相对误差小于15%,能够满足工程需要。

The horizontal well productivity prediction is an important basis for decision making in the feasibility study of horizontal well development and optimum design.However,it is difficult to ensure the prediction accuracy due to limited sample data and complicated influencing factors.The least squares support vector machine regresssion was used to predict horizontal well productivity.Least squares support vector machine improved standard SVMs in which the inequality constraints were changed to equality constraints.The square error and loss function were regarded as a training set of empirical loss.The solutions of quadratic programming problem is transformed into the problem of solving linear equations.All these solved the problems of limited sample data and the complicated impacting factors.The particle swarm optimization algorithm to optimize least square support vector machine parameters overcome the shortcomings of time-consuming in cross-validation method,but also shown the characteristics of learning ability of support vector machine＇s small samples and simple calculation.Using 10 horizontal wells from Daqing Oilfield as samples,the predcition model was obtained using least squares support vector machine regression,and the horizontal wells＇ preductivity were predicted using the regressed model.The prediction results shown that the maximum relative error of prediction is less than 15% which meets the engineering requirements.