Evolutionary-based neuro-fuzzy modelling of combustion enthalpy of municipal solid waste

Abstract

The viability of thermal waste-to-energy (WTE) plants and its optimal performance have informed intelligent predictive modelling of its significant variables critical to optimal energy recovery and plant operational planning using machine learning approach. However, the optimality of hyper-parameters is significant to accurate modelling of combustion enthalpy of waste in neuro-fuzzy models. In this study, the significant effect of hyper-parameters tuning of different clustering techniques, vis-à-vis fuzzy c-means (FCM), subtractive clustering (SC) and grid partitioning (GP), on the performance of the ANFIS model in its standalone and hybridized form was investigated. The ANFIS model was optimized with two evolutionary algorithms, namely particle swarm optimization (PSO) and genetic algorithm (GA), for predicting the lower heating value (LHV) of waste using the city of Johannesburg as a case study. The optimal model for LHV prediction was selected based on minimum error criteria after testing the models’ performance using relevant statistical metrics like root mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute deviation (MAD), relative mean bias error (rMBE) and coefficient of variation (RCoV). The result revealed a better performance of the hybridized ANFIS model than the standalone ANFIS model. Also, a significant variation in all models’ performance at different clustering technique was noted. However, all GP-clustered models gave the most accurate prediction than others. The most accurate model was obtained using a GP-clustered PSO-ANFIS model with triangular input membership function (tri-MF) giving RMSE, MAD, MAPE, rMBE and RCoV values of 0.139, 0.064, 2.536, 0.071 and 0.181, respectively. This study established the significance of municipality-based LHV prediction model to enhance the efficiency of thermal WTE plants and the robustness of evolutionary-based neuro-fuzzy model for heating value prediction.