Aiming at the problems such as difficult determination of net structure

over fitting

low convergence rate

and local minimization in pattern recognition of traditional neural networks

as well as the problem of ignoring the importance of each sample in the standard support vector machine for regression（SVR）

the improved support vector machine for regression was applied in gas sensor array signal pattern recognition by combing with multi-component gas monitoring sensor array for transformer oil.The experimental results show that this pattern recognition method has better forecasting accuracy and generalization than the traditional neural network and the standard SVR

and it effectively solve the cross-sensitivity problem of the multi-component gas sensor in monitoring.