油田开发动态指标多步预测模型研究

为了实时了解单井、井组及区块开发状况,需要对油田各类动态指标作准确而快速的短期预测。油田生产数据一般都是非线性的小样本时间序列,而且各个指标之间的关系都是非常复杂的。根据单变量时间序列的相空间重构思想,在选取模型参数的时候综合考虑了油田的实际适用性,建立了基于支持向量机的时间序列多步预测模型。应用该模型,针对江苏陈堡油田某单井月度含水率的时间序列,分别在含水率低、中、高等不同时期进行了多步预测,一步预测有很高的精度,5步范围内绝对误差不超过10%。对江苏陈堡油田老井月度产油量的时间序列进行了多步预测,在峰谷处相对误差较大,而峰谷之间预测较好。实例计算表明,该模型具有较强的自适应能力和良好的多步预测效果,能够准确而快速地短期预测油田开发动态指标值。

Multi-step forecasting model of dynamic index for oilfield development

In order to know about the development situations of single well,well group and block in real time,it is necessary to forecast the various dynamic indexes quickly and accurately.Field production data are usually non-linear and less data time series,among which they have complicated relationship.Based on the idea of phase space reconfiguration of single variable time series,this paper establishes the multi-step forecasting model of time series from support vector machine,in which the practicability of model data is taken into account when selecting the model parameters.Aimed at the time series of monthly water cut rate for a well in Chenbao Oilfield,the multi-step forecasting is done at the different periods of low,middle and high water cut.The one-step forecasting has a high accuracy and the absolute error of five-step forecasting will be less than 10%.Aimed at the time series of monthly oil rate for the mature wells in Chenbao Oilfield,the multi-step forecasting is done.The error of forecasting is large at peak and valley,and the result is good between peak and valley.Example shows that this model has good performance on self-adaption and multi-step forecasting results and can forecast the value of dynamic index for oilfield development accurately and immediately.