Grey Theory And Neural Network Model In Oilfield Production Prediction

Accurate prediction of oil production is the important condition for oil companies to develop a reasonable plan,avoiding blind investment and guaranteeing the success of the sustainable development.It also provides a powerful scientific protection for petroleum management to make a reasonable development plan.However,oil production forecast has many uncertain factors,these factors including less data,incomplete data,etc.It also makes forecasting the oil production difficult.Grey prediction has the merit of processing the data with small sample or poor information and forecasting with good results,so it is widely used in management,optimization,forecasting and other fields.According to the historical data of No.10 Oil Production Plant at Daqing Oilfield,a traditional model based on the gray anticipation technology and two improved models were put up:Unbiased GM models and GM model based on initial correction.They can respectively predict the annual oil production.The results have shown that:the prediction accuracy of the improved model is not the best,in addition,using different data and different methods,the results are not the same.We can draw a conclusion that the resulted from single method are not very stable,there are certain risks and limitations.At the same time,the system collected and compiled all the data which can affect petroleum's production in the recent years.It uses grey correction analysis method of the grey theory to analysis main factors of the production of petroleum every year.These main factors include utilization rate of the well,years of injection-production ratio,synthesis of moisture content,production speed,annual output of liquid,recovery degree,natural decline rate.It uses these factors as neural network's input and uses annual petroleum production indicatorsas neural network's output,training the network over and over again until get the best structure,then predicting production,the results shows that prediction accuracy is higher,however these factors need a high requirement on the data,less available information of general production,it doesn't meet the statistical rule.So it is hard to use it to do scientific and reasonable forecast with the petroleum's production.The grey theory and neural network have different advantages and disadvantages in the prediction of petroleum's production.The grey theory is based on amount of data comparison,it uses in the short petroleum's prediction whose volatility is not strong and trend is obvious,it has better predicted results.Even though the neural network is not as good as grey theory in the short prediction,it has obvious advantages in long prediction whose original data is unordered and volatility is stronger.In this paper,gray theory and neural network models the characteristics of each,create a combination model,complementary advantages.Daqing Oilfield ten plants utilize a class of block 1994-2004 annual oil indicator data to establish a one-dimensional sequence,respectively,through this one-dimensional sequence of traditional gray GM(1,1)model,and its improved unbiased GM model,the improved model GM initial correction value obtained three groups improved gray neural network model as an input,the original sequence data as the actual production output of the network,the network until repeated training to get the best network structure.This model is applied to the Daqing Oilfield ten blocks of a class of plant oil production forecast 2005-2010,the results showed that:(1)improve the combined model using only an annual oil production index factors in making predictions of their data need to be less,the average relative error less than 5%,the prediction accuracy is relatively high,predicted good results.(2)using the same data set of traditional GM(1,1)model,gray Unbiased GM model,gray forecasting model GM initial correction,the average relative error of 7.09%,7.46%,27.18%,4.13%,3.85%,and a single gray prediction model and neural network model,the improved model mix adaptability,ability to promote good prediction,the prediction accuracy is relatively high,for the accurate prediction of oil production provides an effective and scientific methods,with some practical value.