Abstract: In order to construct an efficient, low-cost and highly adaptable coal blending solution that tackles the problems of traditional methods such as reliance on artificial experience, computational complexity, and poor result stability, a structured coal blending model based on adaptive particle swarm optimization algorithm is proposed. This study takes coal samples of different grades from seven mining areas in the Cixi Coalfield. First, the coal quality indicators of blended coal—including ash content, moisture content, volatile matter, sulfur content, and caking index—are experimentally measured to analyze prediction errors in the traditional linear weighting model. For the indices of ash content, volatile matter and sulfur content, a linear prediction model is used for prediction; aiming at the nonlinear characteristics of the caking index, the principle of support vector machine is introduced to construct a nonlinear prediction model containing Gaussian function terms. Subsequently, a multi-constraint coal blending structural model is established with the objectives of minimizing costs, minimizing high-quality coal ratio, and maximizing low-quality coal ratio, Genetic Algorithm (GA), Particle Swarm Optimization (PSO1), and Adaptive Particle Swarm Optimization (PSO2) are used for solution optimization, with a focus on improving the dynamic adjustment strategies of learning factors and inertia weights. The results show that: ash content, volatile matter, and sulfur content can be effectively predicted by the linear prediction model (R² values close to 0.9). The R² value of the nonlinear prediction model for the caking index reaches 0.927, significantly better than traditional weighted formula. In model solutions, the PSO2 algorithm demonstrates superior performance compared to GA and PSO1, exhibiting reduced prediction errors and achieving the lowest coal blending cost of 1502.80 yuan. Furthermore, the iterative process of PSO2 is more stable, with performance essentially stabilizing after 20 generations. The structured coal blending model based on adaptive particle swarm optimization algorithm can effectively achieve high-precision prediction of blended coal quality indicators and rapid optimization of blending schemes. It can provide technical support for the clean and efficient utilization of coal and has good engineering application value.