Study on light scattering characteristics of micro-nano particles based on SSA-BP model

Background. Accurate prediction of micro-nano particle concentration and size distribution in industrial environments is of great significance for mitigating dust explosion risks, especially in confined spaces. Traditional approaches, including light scattering methods and mechanistic models, are constrained by their inability to effectively handle the complex non-linear correlations between particle properties and environmental variables. Thus, there is an imperative demand for a non-contact, real-time technique to characterize micro-nano particles. Methods. A multi-layer backpropagation neural network (BP-NN) is proposed in this study to characterize the complex interdependencies between micro-nano particle properties (i.e., size, concentration) and optical parameters (i.e., extinction coefficient, scattered light intensity, refractive index). To improve the network’s prediction accuracy and avoid local optima, the Sparrow Search Algorithm (SSA) is incorporated to optimize the initial weights and thresholds of the BP-NN. Furthermore, the random forest (RF) algorithm is utilized to quantify the relative importance of input parameters, providing insights into their regulatory effects on prediction performance. Results. Although the BP-NN outperforms conventional cubic function methods, it suffers from prediction errors exceeding 5% for specific particle sizes of 1.16 µm, 1.30 µm, and 2.47 µm, which exerts a negative impact on the accuracy of concentration estimation. The SSA-BP hybrid model, by contrast, delivers substantially enhanced performance: it reduces micro-nano particle size prediction errors to within 3% and achieves highly robust concentration estimation with the training set showing RMSE = 0.6825 and R² = 0.9624, and the test set attaining RMSE = 0.2434 and R² = 0.9865. Additionally, RF-based feature importance analysis reveals that the refractive index (importance score of 0.74) and scattered light intensity of 0.6 are the dominant factors influencing concentration prediction, alongside particle size of 0.6, whereas the extinction coefficient of 0.2 has a negligible effect. Conclusion. These findings underscore the feasibility of weak-light scattering techniques for safe optical characterization in flammable and explosive industrial settings. Through the integration of neural networks with metaheuristic optimization algorithms, this study enhances the precision of micro-nano particle monitoring, presenting a scalable and reliable solution for industrial safety applications—especially in mitigating dust explosion hazards via rapid, non-contact optical detection. Furthermore, the identification of dominant optical parameters delivers actionable insights to streamline detection system configurations and prioritize key variables in industrial monitoring practices.