Based On Time-varying System Support Vector Machine Prediction Model And Its Application

With the gradual increase of the technical difficulties of development for oil&gas, the dynamic change of development is more complicated. Nowadays, how to effectively achieve reasonable,efficient and sustainable development for oil&gas is one of the key questions which our country faces.The prediction research of the oil&gas development system is the important assurence to realize the reasonable and effective development; the optimization research of the oil&gas development system is the important way to realize suatainable development. To conduct optimal control for oil&gas development system can make oil&gas development dynamic in line with people's expectation as far as possible in the following of internal mechanism, and then achieve more economic benefits. In recent years, many oil experts have research and set up various of dynamic prediction methods for oil,such as neural network, numerical simulation method, adaptive prediction method and muti-gas curve method, and so on. But these methods all exist some questions such as large amount of data, complexity computation, low forecasting precision and difficult parameter selection. Support vector regression machine is for prediction of small sample data, and effectively overcoming the "dimension disaster" and over learning in traditional methos. Therefore, it is widely applied in prediction of oil&gas production, oil recovery and reservoir sensitivity,and so on. However, the applied articles which is used in oil exploratin which combined time series and support vector in systematic and depth are not seen.Support Vector Machine is a new machine learning method which based on statistical principles and recurred to optimization method to solve problems.In recent years, along with the deep theory research in support vector machine, the algorithm have made breakthrough progress, but there are some factors affecting the accuracy of the prediction in real data research process. For the complex and non-stationary data, how to improve the prediction effect have become people's research target. Build different forecasting model through analysing the features of input and output data sample can better fit data and improve the accuracy of prediction data.This paper firstly introduces the basic principle of suppor vector machine in systenmatical, summarizes the selection method for kernel function and kernel parameter, and gives the advantages and disadvantages of each selection methods. Secondly,in order to realize the dynamic prediction research of oil&gas production dynamic prediction better and to improve the accuracy of prediction data, this paper puts forwards a support vector machine prediction model based on time-varying system.This model is proposed by the prediction problem in oil&gas production dynamic and using support vector regression machine and time series analysis theory. In this model, the dynamic data of oil&gas production is decomposed into multiple subsequence in different scale level by using discrete wavelet transform, analyses and reveals the information contained within the prediction variables, and do time series analysis to each subsequence. And then take the analytical subsequence as input vector, construct support vector regression model by taking support vector regression machine as a tool. Finally reconstruct the prediction results of each subsequence, thereby the prediction results can be obtained, and then the prediction results and predictive error are discussed, so the support vector machine prediction model based on time-varying is effective and feasible.