A 3D numerical model of the municipal solid waste incineration (MSWI) process was constructed based on a grate furnace with a daily processing capacity of 800 tons. Fluent was used for analyzing key factors affecting the concentration and diffusion level of particulate matter (PM). According to the actual MSWI plant working condition, a 3D model of the incinerator and the waste heat boiler has been constructed under benchmarks. Key factors affecting PM generation were determined by combining mechanistic knowledge and experts’ experience. They were the combustion temperature of solid phase municipal solid waste (MSW), the wall’s PM collision mode, and the second baffle length. Subsequently, the process of resolving the 3D numerical model was delineated. Then, a univariate analysis of the aforementioned 3D model was conducted for the three pivotal factors mentioned above. Conclusively, the effect of the important factors on the number of particles at the outflow of the incinerator was analyzed via orthogonal experiments to obtain the optimal combination. PM concentration initially diminished and then rose with the increased combustion temperature of the solid-phase MSW. Furthermore, a noteworthy reduction in PM concentration was observed when the second baffle length was 12.45–12.95 m. The greatest influence on the PM concentration of the outlet was posed by the wall’s PM collision mode, followed by the second baffle length. The appropriate adjustment of the combustion temperature of the solid-phase MSW, selection of wall materials, and design of the second baffle length were beneficial for diminishing PM concentration and ensuring long-term stable operation of the MSWI process. The combinative optimality of the three key factors was acquired via orthogonal experiments, which proved the subsequent optimal control of PM concentration at the outlet.

Artificial intelligence (AI) has found widespread application across diverse domains, including residential life and product manufacturing. Municipal solid waste incineration (MSWI) represents a significant avenue for realizing waste-to-energy (WTE) objectives, emphasizing resource reuse and sustainability. Theoretically, AI holds the potential to facilitate optimal control of the MSWI process in terms of achieving minimal pollution emissions and maximal energy efficiency. However, a noticeable shortage exists in the current research of the review literature concerning AI in the field of WTE, particularly MSWI, hindering a focused understanding of future development directions. Consequently, this study conducts an exhaustive survey of AI applications for optimal control, categorizing them into four fundamental aspects: modeling, control, optimization, and maintenance. Timeline diagrams depicting the evolution of AI technologies in the MSWI process are presented to offer an intuitive visual representation. Each category undergoes meticulous classification and description, elucidating the shortcomings and challenges inherent in current research. Furthermore, the study articulates the future development trajectory of AI applications within the four fundamental categories, underscoring the contribution it makes to the field of MSWI and WTE.

Broad learning system based on neural network (BLS-NN) has poor efficiency for small data modeling with various dimensions. Tree-based BLS (TBLS) is designed for small data modeling by introducing nondifferentiable modules and an ensemble strategy to the traditional broad learning system (BLS). TBLS replaces the neurons of BLS with the tree modules to map the input data. Moreover, we present three new TBLS variant methods and their incremental learning implementations, which are motivated by deep, broad, and ensemble learning. Their major distinction is reflected in the incremental learning strategies based on: 1) mean square error (mse); 2) pseudo-inverse; and 3) pseudo-inverse theory and stack representation. Therefore, this study further explores the domain of BLS based on the nondifferentiable modules. The simulations are compared with some state-of-the-art (SOTA) BLS-NN and tree methods under high-, medium-, and low-dimensional benchmark datasets. Results show that the proposed method outperforms the BLS-NN, and the modeling accuracy is remarkably improved with the small training data of the proposed TBLS.