Session: 09-17-01: AI for Energy

Paper Number: 164714

Meta-Heuristics Optimized Neuro-Fuzzy Modelling and Feature Importance Analysis of Bio-Oil Yield via Biomass Pyrolysis 

The consumption of fossil fuel-based energy is increasing owing to the growth of emerging countries and a rising global population. In the face of the climate crisis, we must strive in many ways to make a successful transition from today's fossil-based economy to a future circular bioeconomy. Consequently, biomass energy, which is abundant and sustainable has emerged as a viable sustainable alternative. Among the various conversion techniques for biomass, the production of bio-oil through the pyrolysis technique is gaining traction. If appropriately upgraded, bio-oil could serve as an alternative to fossil fuels. However, the characteristics and yield of bio-oil is contingent on the composition of the biomass feedstock and the pyrolysis reaction conditions. The bio-oil production process involves a complex non-linear relationship amongst multiple variables including the properties of the biomass and the operating parameters of the pyrolysis process

This study develops an Adaptive Neuro-Fuzzy Inference System (ANFIS) model with prominent metaheuristic algorithms namely Ant Colony Optimization (ACO), Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) to optimize and fine-tune fuzzy parameters of ANFIS such as membership functions, rule base parameters, and training process model towards enhance the predictive accuracy of bio-oil yield modeling. The selection of PSO, GA, and ACO for optimizing ANFIS is driven by their proven effectiveness in global optimization, adaptive learning, and nonlinear system modeling. A comprehensive assessment of the relative impact and contribution of these variables is critical to enhance the predictive capability of the model as well as optimizing the pyrolysis process towards optimal bio-oil yield. To this end, the Gini importance metrics of the decision tree algorithms were utilized for the feature importance analysis of the biomass properties and the pyrolysis variables

The performance of the hybrid ANFIS model was assessed using relevant statistical metrics such as root mean square error (RMSE), mean absolute deviation (MAD), mean absolute error (MAE), mean absolute percentage error (MAPE), the variance accounted for (VAF), room mean bias error (rMBE) and correlation co-efficient (R²). Based on the metrics, ANFIS-PSO with a triangular membership function of grid partitioning (GP) clustering outperformed ANFIS-GA and ANFIS-ACO with RMSE, MAD, MAE, MAPE, VAF, rMBE and R²-values of 3.433, 0.435, 1.345, 7.9533, 56.634, 0.341 and 0.89354. Feature importance assessment revealed ash content as the most important predictor of bio-oil yield based on a GI-value of 0.319668 (about 35% cumulative importance). Additionally, carbon and nitrogen, along with ash content, dominated the bio-oil prediction and exhibited the highest predictive power on HHV with about 69% cumulative importance.  The feature importance analysis aids in feature selection and as well enhances the interpretability of the neuro-fuzzy model, providing insights into the key drivers of bio-oil production. The study’s outcome indicates that metaheuristic optimization significantly enhances ANFIS performance by refining its fuzzy rule base and reducing prediction errors.

This research contributes to advancing machine learning applications in bioenergy, providing an interpretable and optimized soft computing framework for bio-oil yield prediction. The findings are expected to benefit researchers, bioenergy engineers, and policymakers by offering a data-driven approach to enhance biomass pyrolysis processes, optimize feedstock selection, and improve bio-oil production efficiency.

Presenting Author: Tien-Chien Jen Univ Of Johannesburg

Presenting Author Biography: Prof Tien-Chien Jen is a Professor at the Department of Mechanical Engineering Science of the University of Johannesburg.

Authors:

Oluwatobi Adeleke University of Johannesburg
Tien-Chien Jen Univ Of Johannesburg