Research On Gas Sensor Fault Diagnose Method Based On PCA And LS_SVM

Gas sensor is one of the most important parts in coal mine monitoring system, and the ability to detect the sensor output signal safe or not is the guarantee of the whole monitoring system. Once the sensor fails and Tfannot timely and accurately reflect the situation of gas under the mine, it can cause great economic loss, even casualties would happen. Visible, the reliability of gas sensor directly affect the safety of underground properties and personnel's life. Therefore, studying of gas sensor fault diagnosis technology and timely find the possible faults in the operation has great practical significance on the safety production of modern coal mine monitoring system.The fault types and the existing fault diagnosis methods are analyzed in this paper, and on the basis of the diagnostic accuracy and speed, this paper put forward a new method based on the principal component analysis (PCA) and least squares support vector machine (LSSVM) algorithm for gas sensors fault classification diagnosis.First of all, use the principal component analysis (PCA) to demesne the coal mine gas sensor collected data points, and improve the traditional principal component analysis method. According to the mean values method calculate different samples to carry the original data information ability, and then do the reduction operation. This makes sample data to carry on the secondary reduction, simplifies the sample structure, reducing the redundancy, and ensures the data after reduction carry enough original sample data information.Then, taking the data after dimensionality reduction by PCA as sample, do classification modeling researches of gas sensor under the least squares support vector machine (LS_SVM) method. In order to improve the accuracy of classification, this paper regards the polynomial kernel function, the radial basis function (RBF) kernel function, Sigmoid kernel function as the basis of three kinds of kernel function, and combined with support vector machine (SVM) classification algorithm to optimal model parameters. Through repeated experiments, modeling based on radial basis function (RBF) kernel function and multiple classification algorithm based on binary tree classification model has a higher rate of correct judgment.Finally, build a gas sensor fault diagnose model based on PCA and LS_SVM algorithm, and compared with BP neural network algorithm. The results show that fault diagnosis model based on PCA and the LS_SVM classification not only have faster speed but also have higher classification accuracy, it meets the original design purpose of this paper.