Incorporating prior knowledge in a fuzzy least squares support vector machines model

The least squares support vector machines (LS-SVM) is sensitive to noises or outliers. To address the drawback, a least squares support vector machines model incorporated with a prior knowledge on data is presented. Information of noise distribution for samples is introduced in the training process. A strategy based on the sample affinity is presented to discriminate data and noises. A fuzzy membership is automatically generated and assigned to each corresponding data point in the sample set by using the strategy and the noise model. The performance of FLS-SVM is improved to resist against noises. The flexibility increase to treat data points with noises or outliers. The proposed method is applied to fault diagnosis for the lubricating oil refining process. The experiment result shows better robust of the proposed method.