Soft Sensor of Biomass in Fermentation Process Based on Robust Neural Network

In the course of soft sensor modeling of biomass in fermentation process using neural network, it will usually make the modeling accuracy and estimation performance of soft sensor model worsened when there are outliers in modeling data. To solve this problem, a soft sensor modeling method based on robust neural network is proposed in this paper. Firstly, the anomaly degree of each modeling data pairs is calculated using k-nearest neighbor algorithm, and the weight of each modeling data pairs is determined according to the calculated anomaly degrees. Then, the soft sensor model of biomass based on robust neural network is developed. Simulation is performed using the production data from Nosiheptide fermentation process, and the simulation results show the effectiveness of the proposed method.