准确及时的检测原油含水率对注水策略调整、 原油开采能力评估、 油井开发寿命预测等均具有重要意义。 然而, 当前我国大多数油田均已进入高含水的开发中晚期, 含水率测量难度大且准确率不高。 在此背景下, 开展了高含水情况下利用近红外光谱进行原油含水率测量的研究。 首先介绍了目前原油含水率检测的常用方法, 分析了它们的优劣。 理论上, 由于水的近红外光吸收带与原油中C—H键的吸收带有明显区别, 根据Lambert-Beer吸收定律和吸光度线性叠加定律可知, 不同含水率高含水原油近红外光谱会存在较强响应差异。 为此, 对高含水原油进行近红外光谱检测, 建立原油含水率与近红外光谱响应间的非线性映射模型, 可实现高含水原油含水率的精确测量。 为了验证该方法的有效性, 搭建了近红外光谱数据采集实验装置: 采用白炽灯作为光源, 经过光路调节成平行光后垂直射入样品池, 用近红外光谱仪(海洋光学NIR512)采集光谱用于分析。 其中, 接收光谱仪带宽为900～1 700 nm, 平均分成512个波段。 光谱数据利用光谱仪配套软件储存在电脑中。 样本采用相同厚度不同比例的油水混合物, 样本含水率范围为70%～99%, 共采集数据60组, 每组重复3次取平均值。 得到原始数据后, 先进行原始数据预处理, 以减少数据采集时来自高频随机噪音及温度不稳定、 样本不均匀、 基线漂移、 光散射等不利因素的影响。 分别选用了S-G滤波、 一阶导数和S-G滤波+一阶导数作为数据预处理的方法, 利用连续投影算法(SPA)对光谱数据进行降维, 并利用偏最小二乘法(PLS)和多元线性回归(MLR)进行建模, 模型精度通过计算均方根误差值(RMSE)和相关系数(r)来验证。 对比发现, 使用S-G滤波+一阶导数建立的模型RMSE值最小(RMSE=0.007 0, r=0.998 3)。 使用SPA降维后的模型要优于全波段PLS模型(RMSE=0.083 3, r=0.920 6)与MLR模型(RMSE=0.099 9, r=0.967 1)。 利用SPA提取出的31个特征波长建立的模型仅占全波段的6.05%, 并获得了较好的精度。 证明了利用光谱检测高含水原油含水率可行性, 并且得到了满意的精度, 为高含水原油的含水率检测提供了新的方法, 为进一步利用近红外光进行高含水原油的快速检测与在线监测提供参考。

Accurately and timely measuring water content of the crude oil is of great significance for water injection strategy adjustment, crude oil exploitation capacity assessment, and oil well development lift prediction. However, at present, most of China’s oil fields have entered the mid- or late- development stage with high water content. And the corresponding water content is difficult to measure accurately. Under this circumstance, this paper carried out research on the measurement of water content of the crude oil using near-infrared spectroscopy. Specifically, commonly employed methods for measuring water content of the crude oil were introduced, and advantages and disadvantages of these methods were analyzed. Theoretically, since the near-infrared absorption band of water is significantly different from the absorption of C—H bond in crude oil, according to Lambert-Beer’s law of absorption and linear law of absorbance, there is a strong response difference in the near-infrared spectrum of high water cut crude oil with different water content. Therefore, we proposed to use near-infrared spectroscopy to accurately measure the crude oil with high water content. And then, by analyzing the measured near-infrared spectrum, non-linear mapping between the water content of the testing crude oil and the near-infrared spectrum can be established. With the obtained non-linear mapping model, water content of the crude oil can be accurately calculated. In order to evaluate the performance of this method, we constructed a hardware platform for collecting near-infrared data. In this platform, Incandescent lamp was employed as a light source, and near-infrared spectrometer (Ocean Optics NIR512) was used to collect near-infrared in range 900～1 700 nm with 512 uniformly divided sub bands. The collected data were stored in the computer using the spectrometer supporting software. With the obtained near-infrared data, the raw data preprocessing was performed to reduce the influence of temperature and high frequency random noise, sample unevenness, baseline drift, light scattering, and et al. In this paper, S-G filtering, or first order derivative, or S-G filtering+first order derivative techniques were employed as the preprocessing method; Successive Projection Algorithm (SPA) was used to reduce the dimension of the raw data; Partial Least Square (PLS) and Multiple Linear regression (MLR) were employed to construct the corresponding non-linear mapping model; Root Mean Square Error (RMSE) and Correlation coefficient (R) were used to evaluate the quantitative measuring performance. Experimental results illustrated that: model constructed using S-G filtering+first order derivative as preprocessing method can achieve the best RMSE (RMSE=0.007 0, r=0.998 3); Model constructed with reduced dimensional data using SPA method is better than the one (RMSE=0.083 3, r=0.920 6) constructed by PLS with full band data and the one (RMSE=0.099 9, r=0.967 1) constructed by MLR with full band. Obviously, although the 31 dimensionality-reduced feature bands obtained by SPA method are only 6.05% of the full band data, the corresponding water content measuring accuracy of the crude oil is very promising. In general, we validate the feasibility of using spectroscopy technique to measure water content of the high water content crude oil, and satisfactory accuracy can be achieved. Therefore, it can be said that this paper provides a new method for water content measurement of high water content crude oil, and provides reference for accurately and timely measuring high water content crude oil using near-infrared spectroscopy.