Electrochemiluminescence Sensor Based on Electrospun Three-Dimensional Carbon Nanofibers for the Detection of Difenidol Hydrochloride

An optimized back propagation neural network on small samples spectral data to predict nitrite in water

Accurate detection of pollutant levels in water bodies using fusion algorithms combined with spectral data has become a critical issue for water conservation. However, the number of samples is too small and the model is unstable, which often leads to poor prediction and fails to achieve the measurement goal well. To address these challenges, this paper proposes a practical and effective method to precisely predict the concentrations of nitrite pollution in aquatic environments. The proposed method consists of three steps. Firstly, the dimension of the spectral data is reduced using Kernel Principal Component Analysis (KPCA), followed by sample augmentation using Generative Adversarial Network (GAN) to reduce calculation cost and increase the diversity and scale of the data. Secondly, several improvement strategies, including multi-cluster competitive and adaptive parameter updating, are introduced to enhance the capability of the Particle Swarm Optimization (PSO) algorithm. The improved PSO algorithm is then applied to optimize the initialization weights and biases of the Back Propagation neural network, thereby improving the model fitting and training performance. Finally, the developed prediction model is employed to predict the test set samples. The result suggests that the R2, RMSE, and MAE values are 0.976290, 0.008626, and 0.006617, which outperform the state-of-the-art and provided a promising model for the prediction of nitrite concentration in water.