Investigating the impact of co-processing municipal solid waste in cement kilns on coal savings and emissions using machine learning

The use of refuse-derived fuel (RDF) from municipal solid waste (MSW) promotes sustainable development and carbon neutrality in the cement industry. The development of a model for energy consumption and emissions from MSW co-processing in cement kilns (MCKC) is challenging due to the varying characteristics of MSW and the complexity of calcination system, which affects coal saving efficiency and technology optimisation. This paper presents a comparative analysis of several algorithms and demonstrates that the Long Short-Term Memory (LSTM) algorithm provides the most accurate results when developing an MCKC energy consumption and pollutant emission simulation model. The Shapley Additive Explanation is used to further identify key influencing factors and patterns. A case study in China reveals a coal-saving trend with RDF-to-coal blending ratios over 17% and an RDF water content under 25%. Better control of PM, SO₂, and NO_X emissions is achieved with the blending ratio of sieved waste to cement clinker below 7% and water content of sieved waste under 30%. This emphasises the need to optimise pretreatment processes and to source-separate MSW, which will require efforts such as the construction of source separation facilities. While LSTM shows promise, data requirements and training complexity should be considered for model reliability. Further research is needed to generalise the findings beyond China. This study lays the foundation for data-driven MCKC models to assist regions in simulating and optimising MCKC systems with local data. Future studies can improve model accuracy by considering additional variables and sample sizes in the complex MCKC system.