Abstract: Pump detection period is an important index to reflect the working reliability of pumping wells. Accurate prediction of pump detection period is of great practical significance to improve oil well production efficiency and economic benefits. Aiming at the low accuracy of pump detection period prediction in oil field, a pump detection period prediction method based on feature fusion is proposed. This method introduces SVR to extract the static characteristics of oilfield data, reconstructs the characteristics of oilfield dynamic data, uses convolution neural network to learn the dynamic characteristics of oilfield data, introduces multi-modal compression bilinear pooling to fuse the static and dynamic characteristics, and uses discriminant model to train the fusion characteristics to realize the accurate prediction of pump detection cycle. The experimental results verify the effectiveness and feasibility of the model.

Key words: feature extraction, feature fusion, pump detection period predicting, support vector regression, convolutional neural network