基于时序动态分析的油井产量预测研究

针对目前常用的油井产量预测方法效果并不理想的问题，开展时间序列分析来进行油井产量动态预测研究。采用时间序列分析结合残差修正方法，建立具有时序动态分析能力的产量差分自动回归移动平均模型（Autoregressive Integrated Moving Average，ARIMA），得出预测初始值与真实油井产量的残差；通过构建支持向量机（Support Vector Machine，SVM）时序预测模型进行残差修正，获得油井产量组合预测值；并将长短期记忆网络（Long Short-Term Memory，LSTM）模型与上述方法进行对比。实验表明，组合预测模型、LSTM模型的预测结果平均相对误差率分别为9.81%和32.44%。说明组合模型预测更精准，为油井产量的动态预测提供了一种有效方法，可作为油井在生产计划时的快速实时辅助依据，具有实用价值。

A Study on Oil Well Production Prediction Based on Time Series Dynamic Analysis

The effect of oil well production prediction method currently used is not ideal. This study aims at dynamic prediction of oil production, using time series analysis combined with residual error correction. We build an ARIMA (Autoregressive integrated moving average) model with the ability of time series dynamic analysis, to predict the initial value and the real residual oil well production; the residual error was corrected by constructing the (SVM) Support Vector Machine time series prediction model to obtain the predicted value of oil well production combination. The LSTM (Long Short-Term Memory) model is compared with the above methods. The experimental results show that the average relative error rates of the combined prediction model and the LSTM model are 9.81% and 32.44% respectively. The conclusion is that the combined model prediction is more accurate, and provides an effective method for the dynamic prediction of oil well production, which can be used as a fast and real-time auxiliary basis for oil well production planning and has practical value.