Prediction of the Contents of Four Polycyclic Aromatic Hydrocarbons in Smoked Sausage Using Back Propagation Neural Network Optimized by Particle Swarm Optimization Algorithm

A predictive model based on a back propagation artificial neural network (BP-ANN) optimized by particle swarm optimization (PSO) algorithm was developed to predict the contents of four polycyclic aromatic hydrocarbons (PAHs) (benzo(a)pyrene, benzo(a)anthracene, benzo(b)fluoranthene, and chrysene) in smoked sausage. Smoking temperature, smoking time, fat/lean meat ratio and smoked sausage color (a^* and b^* values) were used as input layer parameters, and the measured contents of four PAHs as output layer parameters. The PSO-BP-ANN model was used to optimize the initial weight and threshold to obtain the best parameters. The results showed that the mean square error (MSE) of the proposed predictive model was 0.018, and the correlation coefficients (R²) for training, validation, test and global data sets were 0.951, 0.929, 0.933 and 0.940 respectively. All these parameters were better that those of the BP-ANN model, indicating that the PSO-BP-ANN model had better accuracy and robustness.