多样性PSO_SVR油气操作成本时间序列预测模型

Diversity PSO_SVR Oil-gas Operating Cost Time Series Forecasting Model

为解决现行油气操作成本预测方法存在的油气操作成本影响因素难确定，传统模型的构建需采集大量多维数据且对具有非线性特征的油气操作成本时间序列预测精度较低的问题，首先采用CAO算法对油气操作成本时间序列进行空间重构并构建模型所需样本。其次，为解决支持向量机回归模型（SVR）所需参数难确定及原始粒子群优化算法（PSO）易陷入局部最优的问题，利用多样性PSO算法（DPSO）并结合K—fold交叉验证最小误差准则进行参数寻优，建立基于空间重构的DP-SO—SVR油气操作成本时间序列预测模型。最终，以某油田1983～2004年油气操作成本为数据，将所提出的DPSO—SVR模型与网格搜索SVR模型、PSO—SVR模型、Levenberg—Marquardt BP神经网络模型及背景值优化GM（1，1）模型进行对比实验，并对各模型的预测能力进行对比及原因分析。结果表明所提出的DPSO—SVR模型能够解决上述问题，具有更高的预测精度且能够帮助油田管理者做出更好的决策。

To address the deficiencies existing in the current oil-gas operating cost prediction methods, Cao algo- rithm was used to reconstruct the oil-gas operating cost time series and create the sample data for the model. Second- ly, in order to overcome the challenges to determine the parameters needed in the support vector regression model （SVR） and that the original particles swarm optimization algorithm （PSO） has the tendency to local optimum solu- tion, the diversity PSO algorithm （DPSO） and the minimum error criterion of K-fold cross validation were used to optimize the parameters, and the DPSO_SVR oil-gas operating cost time series forecasting model based on space re- construction was proposed. Finally, with the oil-gas operating cost data of an oilfield from 1983 to 2004, the pro- posed DPSO_SVR model was used and compared with grid search SVR model, PSO-SVR model, Levenberg-Mar- quardt BP neural network model and optimization background value GM （ 1,1 ） model. These models＂ forecasting per- formances were compared and analyzed. The results show that the proposed DPSO_SVR model can solve the example problems, has a higher prediction precision, and can help the oilfield managers make better decisions.