Abnormal Data Detection Based on Dual-Factor Weighted SVDD for Multimode Batch Processes

Current methods for detecting abnormal data in batch processes using Support Vector Data Description (SVDD) overlook the varying importance of different training samples to the hypersphere. This always results in model overfitting, thereby affecting the accuracy of abnormal data detection. To address this issue, this paper proposes a dual-factor weighted SVDD (DFWSVDD) method for detecting abnormal data in multimode batch processes. First, density peak clustering is used to partition the process into different modes. On this basis, a distance calculation function for high-dimensional data samples is constructed to obtain the local density value of each sample point and calculate the relative distance of each sample to the center of its mode. By combining local density and relative distance as dual-factor weight for samples, a dual-factor weighted SVDD model is constructed to detect abnormal data.The effectiveness of the proposed method is validated through experimental results from the penicillin fermentation process.