Research On Model And Algorithm For Prediction Of Oilfield Development Indexes Based On Artificial Neural Network

At present，many development blocks of oilfields has entered the adjusted developmentphase of high moisture content, significant increase in water content, natural decline rateincreases,so it is important to study the dynamic changes of oilfield, and to apply these lawsto adjust and improve the oilfield development plan. Analysis and prediction of oilfielddevelopment index is an important part of oilfield dynamic analysis, the reliability ofprediction accuracy directly affects the adaptation and implementation to oilfielddevelopment planning programs. The choices of index prediction’s methods and modelshave a great impact to the prediction accuracy of the results. Find and explore the optimalindex prediction method has also been an important research topic of oil workers.This topic uses intelligent modeling methods and techniques, based on artificial neuralnetworks, intelligent algorithms to research the analysis and forecasting of oilfielddevelopment index, early production warning and forecast, forecasting of inefficient welleffectiveness of measures, to solve practical problems in oil oilfields. For various practicalproblems, I have studied three different prediction models and realized organization andmanagement of the development index data. I have been deeply studied Elman neuralnetwork model, the process neural network model, radial basis function neural networkmodel. According to the characteristics and advantages of each model, applied to differentpractical problems, and analyzed the optimization algorithm-genetic algorithm and particleswarm optimization algorithm. According to the actual situation, improved Genetic Algorithmand used the improved genetic algorithm to optimize the Elman neural network is applied tothe prediction of moisture problems, built predictive models. For oil field early productionwarning and forecast, selected process neural network as a predictive model, used swarmalgorithm for process neural network training, built forecast and early warning indicatorsystem, and built early warning models. I have studied dynamic K-means clusteringalgorithm, used RBF neural network to predict the effectiveness of measures, and have builta predictive indicator system for the effectiveness of measures, built predictive models.This paper presents three prediction models all took real historical data as a model fieldtraining samples, used the actual data from the oil production plant to validate the accuracyof the model. Experiments show that the three models have achieved good prediction. Thetopic research can not only enrich the index forecasting methods, and can effectively guidethe oilfield development, assisted field planning program design.