Research On Oilfield Well Group Injection-recovery Law And Dynamic Development Index Calculation Model Based On ANN Identification Theory

Experience formula and fitting formula based on a large number of original data are used in traditional development index calculation.They can't get the satisfied identification result and global optimal solution for nonlinear dynamic system.The attracting of neural networks for nonlinear dynamic system identification are multi-forward neural networks can approximate to nonlinear reflection by any accuracy gives a new method for complicated system modeling;The special study ability makes it fit to the change of system and environment;The parallel calculation feature makes it have potentiality to accomplish a lot of complex calculation fast; The distributed information storage and process construction make it have fault-tolerance;multi-input and multi-output construction is convenient to multi- variable system identification.so neural networks has very important research value and wide application foreground in nonlinear system identification.Oilfield well group injection-recovery system belongs to nonlinear dynamic system.It has uncertainty and .so it is fit for using identification theory of neural networks to build identification model for system.Research on high efficient identification algorithm and then predict the development index.The research of this issue has important meaning in production and management of oil deposits and production prediction and planning.The basic theory,characteristic and procedure of neural networks identification is researched in the thesis. Identification model of stratum pressure prediction is built.Improved BP algorithm is presented.The model and algorithm are used in oilfield well group injection-recovery system. The prediction result proves the effectiveness.Aiming at the disadvantages of traditional neural networks making the model and algorithm complicated when it solves the problems of system process inputs and the dependence of time order,a process neural networks model with time-varying inputs and outputs used to system identification model is presented,and the learning algorithm based on function base expanded integrated with grades descending and genetic simulated annealing are given for the identification algorithm. Predicting the oil yield and the water yield,it is satisfied with the result.Aiming at traditional Elman networks is hard to identify the dynamic system which the input and output are all time-varying function,the thesis presents a Elman feedback process neural networks model.it is similar to traditional Elman networks in construction.The difference is the input and link weights of Elman feedback process neural networks can be time-varying function.It is the expand of traditional Elman networks in timing area.The thesis presents the topological construction and learning algorithm of Elman feedback process neural networks.Model and algorithm are proved effectively through the recovery percent prediction.