Abstract: Support vector machine method is based on the principle of structural risk minimization, and suitable for small samples, nonlinearity and local minimun by Kernel function processing technology, it overcomes the limitation of conventional statistical method, avoids dimentionality disaster, compromises the commonality and popurization of the model on the basis of limited samples and solves the problems of learning performance and popularization effectively with higher prediction accuracy. In practical production, there are a lots factors influncing reservoir productivity, and the factors impact each other. In view of comprehensively considering formation factors, well logging productivity prediction parameters are sorted out, the productivity is predicted by support vector machine, the prediction result is correlated and compareed with the result from multi- regression and BP neural network processing. Practice suggests that the support vector machine is better than the latter two methods and is worthyto be popularized.

Key words: support vector machine, structural risk minimization, learning performance, generalization, predication of productivity