Multi-target Evolutionary Hyperparameter Optimization for Multi-task Predictive Modeling in Biomass Pyrolysis

Introduction. Conventional pyrolysis modeling of biomass relies on single-output configurations that predict one process parameter at a time, such as bio-char or bio-oil yield. This approach fails to capture the complex interactions between process conditions and the full spectrum of reaction subproducts, limiting both predictive accuracy and the physical consistency of model outputs. This work presents a multi-target modeling framework for the simultaneous prediction of four pyrolysis outcomes: bio-char yield, gas yield, calorific value, and carbon retention potential.

Methods. Evolutionary algorithms were applied to the hyperparameter optimization of Random Forest, Support Vector Regression, and Artificial Neural Network models, all configured for multiple-output regression. Input variables included organic matter composition, pyrolysis temperature, heating rate, and residence time. The evolutionary search operated without manual architecture tuning, relying on population-based optimization to identify optimal model configurations across the combined output space.

Results. Predictive performance of the multi-target models was compared against conventional single-output methods using mean squared error and the coefficient of determination. The multi-target approach produced improved accuracy and greater physical consistency across all four output variables relative to single-output baselines. The evolutionary optimization captured underlying thermochemical interactions that single-output configurations systematically missed.

Conclusions. Evolutionary hyperparameter optimization provides a viable and technically sound strategy for shifting environmental modeling of pyrolysis processes from traditional single-output simplifications toward more complete, physically coherent analyses. The proposed framework supports more informed decision-making in bioenergy production by providing a simultaneous evaluation of the energetic and environmental value of pyrolysis products. These findings have direct relevance to geo-resources and the broader energy transition context.