Research On Analysis And Prediction Model Of Oilfield Development Data Based On Dynamic Neural Network

Oilfield development, as a comprehensive technology, including reservoir geology,mining, oil production and so on, is a kind of complex nonlinear dynamic system. Through the analysis of typical relations and change rules between variables, it can provide decision basis for scientifically making oil production and planning scheme. The analysis and forecast of the oilfield data acts as an important content of oilfield geological development research work, the credibility of predicted results affect directly the oilfield development plans. And the correct choice of model and methods is of critical importance. In this way, how to build and choose the prediction model, and analyze the calculation complexity, the accuracy, the credibility of the model has become an important research topic in oilfield.Basing on dynamic neural network and intelligent algorithm, this topic studies analysis and prediction model of oilfield development data. Firstly, it analyzes oilfield development data from two aspects: qualitative and quantitative, and also expounds the data preprocessing technique. And it constructs the Elman neural network model based on genetic algorithm through systematic analyzing of the Elman neural network model, and then proposes an improved genetic algorithm, and carries out the prediction model of single well water index. It establishes self-organizing feature map neural network model and blends with dynamic K-mean improving optimization model for improving stability and nature of leaning of the network from mechanism, and applies the model to the prediction of classifying abnormal well, application results are good. At last, for the problem of yield prediction, it builds a kind of the neural network with two hidden layers model, using particle swarm optimization in the network model and improving nature of leaning of the network model, and it also carries out the experimental analysis.For the three kinds of prediction models which are Elman neural network model,self-organizing feature map neural network model and process neural network model,prediction processing is carried out according to actual data in this topic. The experimental results show that the three models are operable, practicable and feasible, and their predictive effects are good. Therefore, the research of this paper provides a new method for oilfield development prediction, and helps to analyze and decide of oilfield production and management, which have great theoretical and application value.