GA-BP神经网络对含油饱和度的动态预测研究

Research on Dynamic Prediction of Oil Saturation with GA-BP Neural Network

为克服传统BP神经网络收敛速度慢、易陷入局部极小值等不足,采用遗传算法对其进行了优化。之后利用遗传算法优化的BP神经网络—GA-BP实现了对江苏油田庄2断块某小层含油饱和度的动态预测:首先确定GA-BP神经网络的输入、输出层神经元,接着建立经验公式,实现了输入神经元时间T的定量化,然后选取训练样本对GA-BP神经网络进行训练,最终建立起含油饱和度的动态预测模型并利用该模型对5年以后的含油饱和度进行了预测。该预测结果对油田现场下一步的生产实践具有重要的指导意义。

In order to overcome the slower convergence rate and falling into local minimal value easily of traditional BP neural network, the genetic algorithm is used for optimization. Then the oil saturation of a certain layer of Z2 fault-block in Jiangsu oilfield is predicted. Firstly ,the input and output layer neurons are established. Secondly ,an empirical formula is given that realizes the quantification of the in- put neuron-time. Lastly,the dynamic prediction model on oil saturation of the GA-BP neural network is established after the training with the selected samples. And the oil saturation of 5 years later is predicted with the dynamic prediction model. The prediction result of oil saturation has an important guiding significance to the production practice.