Study On Neural Network Model And Optimal Algorithm For The Oil Pumping Control System In Oil Field

Oil extraction control system in the oil field is mainly used to measure thelevel of the oil underground and extract oil intelligently. According to the geologyof oil field in our country, most of the oil wells appear empty after exploitation forsome time. But when this happens, the electromotor still works on all day and allnight that results in the phenomenon of "a big horse hauling a small carriage". Soit makes the difficulties in retrenching the energy sources because of the largeamount of the power energy used. So this system in the dissertation is essentialthat it can not only meet with the need of intelligent oil extraction, but also offerthe effective plan of design for the research, development and production of thenew-type oil extraction control system.In order to resolve the "relatively light load in a long term" and "pumpemptiness" problems that many wells faced. The intermittent operation for thecontrol system of oil pumping is presented in this paper. It isn't linear relationbetween armature current and electromotor load according to the workcharacteristic of electromotor, but the change of fluid surface effects the change ofarmature current directly. So designer sampling armature current signal, then getthe information of fluid surface after dealing with the signal. And then designereduced the connection between fluid surface and time on the basis of theconnection between current and time.Then, in this dissertation the main research is greatly focused on the softwarealgorithms.General BP neural network are studied in depth on the system modeling.Sigmoid activation function is used in BP algorithm;it has high gain in the middleof interval and low gain at two terminal intervals. When data are trained far awayfrom zero, study rate of convergence is slow. So the standardization method oforiginal training data is applied. In this paper standard normalization and nonlinearnormalization are applied to standard original data. Simulation results indicate thatnonlinear normalization is superior to standard normalization. To improve theconvergence speed of conventional BP Neural Network and overcome itsdrawbacks of getting stuck at local minima, nonlinear homotopy BP NeuralNetwork is presented to identify oil-pumping model. According the theoreticalbasis of homotopy BP algorithm, the main merit of homotopy BP algorithm is thatthe homotopy function is inducted, so the capability of avoiding getting into thelocal minimum is enhanced. But the signal change is too small in every section;BP algorithm has low rate of convergence and the running times of homotopy BPalgorithm amounts to the summation of the running times of every general BPalgorithm, so the convergence speed of homotopy BP neural network still shouldbe enhanced. Therefore a new advanced homotopy BP algorithm which combinedhomotopy algorithm and quick BP algorithm is put forward. In the end, advancedhomotopy BP algorithm is used to model the relation between fluid height andcurrent, and the curve of the oil cumulation. The results of simulation show thatthe advanced homotopy BP algorithm not only has the advantages of fastconvergence and but also avoids falling into local minimum.The key of the intermittent operation of oil pumping is the determination ofthe best outwork time of electromotor. In order to optimizing the outwork time, thegeneral GA is researched thoroughly and the corresponding simulation is done.The general GA has the problems of greatness calculation, slow rate of globaloptimization, falling into local optimization and low solution precision, so theimproved GA algorithm is studied, that is accelerating GA, and the research insimulation is done and the improved GA is proved to be better than standard GA.At last the best outwork time is ascertained, and then the validity of the algorithmsis proved by the simulation result. All referred above are significant and valuableto the theoretical guidance.