| Aiming the interbedded disturb problem of multilayer heterosphere oil field, most oil fields adopt the waterflood development to improve recovery ratio, but it can create greatly divergence and cause grave interfere problem in every stratum. A new sucker-rod pumping system of double-acting separate recovery is developed in order to solve the problem, now it is used in most oil fields in our country. The present fault diagnosis technique can't well suit to diagnose the pumping system. So far, the public reports in the field are few at home and overseas. The artificial neural networks grow rapidly and become to be one of the most advanced problems lately. In comparison with traditional expert system, the artificial neural nets expert system is possessed of obvious superiority in the aspects of knowledge acquisition, parallel reasoning, adaptability studying, error containing capability and so on. It satisfies the needs to recognize the down-hole dynagraph of sucker-rod pumping system that is complicated, nonsteady and disturbed to some extent.In many kinds of artificial neural networks, BP neural nets is one of the most pioneer and common models, it is successfully applied to equipment fault diagnosis. In the paper, BP neural nets is briefly introduced, and based on the shortcoming of the BP neural nets are summed up, expert system of neural networks are established by using improved BP momentum algorithm.Double-acting pump is a new type pump for separate layer recovery, in this paper, the structure and work principle of double-acting pump are briefly introduced, and the mechanics characteristics of separate layer recovery are stressly analysised, the prediction model has been established, we can gain the forced condition of each section of sucker rod by using the display difference method to divide, and can paint the dynagraph of each section of sucker rod string. This will establish base for analyzing static and dynamic behavior, selecting parameters, optimizing design and diagnosing fault. In the paper, the diagnosis atlas of double-acting pumping system is established by calculating, simulating and uniting the diagnosis atlas of rule pumping system. The atlas is used to established modules of neural networks as specimen input set. A pretreatment method of MATLAB-Toolbox-based is proposed so that a computer can recognize the real dynagraph and the total scheme of double-acting fault diagnosis expert system is presented. Function of various modules and run mechanism of the expert system is discussed. The course to initialize, train and test for neural network modules has been illustrated in detail. The fault diagnosis expert system software has been designed with the MATLAB-Toolbox, then the correctness of diagnostic conclusions for the expert system is verified with the real dynagraph of experimental well. Programs of diagnosis system has been designed, it may provide a powerful tool for fault diagnosis of separate layer recovery system, and has an important sense for the technique development of separate layer recovery. |
