Mechanical Property Prediction of <em>Cunninghamia lanceolata</em> Using a Quantitative-Causality-Based BP Neural Network Optimized by Adaptive Fractional-Order PSO

To enable non-destructive evaluation of wood mechanical properties, this study proposes a backpropagation neural network (BP) optimized by an adaptive fractional-order particle swarm optimization algorithm (AFPSO), termed LK-BP-AFPSO. This model accurately predicts four key mechanical properties—longitudinal tensile strength (SPG), modulus of elasticity (MOE), bending strength (MOR), and longitudinal compressive strength (CSP)—by extracting limited physical features without damaging the wood structure. The proposed method demonstrates strong generalization and robustness across various Cunninghamia lanceolata types, including fast-growing (YKS), red-heart (CSH, XXH), and iron-heart (XXT) variants. A key challenge lies in effectively identifying informative features from indirect measurements. To address this, the Liang–Kleeman (L-K) information flow theory—a first-principle-based, efficient causality analysis method—is introduced. By quantifying causal influence through the coefficient of variation, L-K information flow enables effective feature ranking, which further enhances the model’s prediction accuracy and efficiency. This integrated framework offers a reliable solution for non-destructive mechanical property prediction of wood, contributing to intelligent assessment in forestry engineering.