Development of a deep learning-based framework for operational optimisation of municipal solid waste incinerators

Combustion efficiency of Municipal Solid Waste (MSW) incinerators depends on numerous operational parameters like air flowrates, boiler feedwater temperature, conveyer speed etc. Optimising these operational parameters can lead to higher efficiency, reduce emissions and maximise waste-to-energy conversion however, the complex interdependence of these parameters makes it difficult to identify the optimal conditions on which to run the power plant. In this study, we develop a Deep Learning (DL) based framework to optimise the operation of MSW incinerators. Historical operational data from a 600 tonne/day MSW incinerator has been collected and ranked based on feature importance using Gradient Boosting Decision Trees (GBDT). The dimensionally reduced dataset is used to train a Backpropagation Neural Network (BPNN) model, characterizing highly non-linear relationship between operational parameters and steam production from the MSW incinerator, achieving a mean relative error of 7.79% and prediction accuracy of 92.21%. Finally, Particle Swarm Optimization (PSO) is then employed to optimise the operational parameters. The optimisation process converged within 650 iterations (~3 minutes), yielding increase in steam production from 2.7t/t to 3.11t/t waste, which is equivalent to 15.2% increase in the thermal efficiency of the MSW incinerator. The proposed DL-PSO framework enables automated optimisation of the operational parameters, minimising dependency on operator experience, providing a novel, practical and computationally efficient tool for enhancing the combustion performance of MSW incinerators and reducing emissions.