Estimation of bio-oil yield for biomass using hybrid models

The bio-oil yield from biomass during pyrolysis, influenced by multiple chemical and process parameters, is particular for gaining sustainable production of energy, necessitating accurate predictive models. This study employs a Gradient Boosting Machine (GBM) framework, augmented by four sophisticated optimization algorithms: Batch Bayesian Optimization (BBO), Evolution Strategies (ES), Bayesian Probability Improvement (BPI), and Gaussian Processes Optimization (GPO). The model leverages a dataset comprising 400 experimental samples, with 90 % allocated for training and 10 % for testing, using input variables such as content of carbon, content of nitrogen, content of hydrogen, content of ash, content of oxygen, crystallinity index, BET surface area, catalyst-to-biomass ratio, residence time, temperature and to predict bio-oil yield. To mitigate overfitting, k-fold cross-validation is useful during training. The performance of every optimization algorithm is assessed through computational runtime and metrics such as R-squared (R²), mean squared error (MSE), and average absolute relative error (AARE%). Correlation analysis reveals varied relationships, with BET surface area (0.18) and oxygen content (0.14) showing positive associations with bio-oil yield, while temperature (-0.26) and ash content (-0.22) exhibit notable negative correlations. Among the optimization approaches, GBM-BBO achieves the highest accuracy, with an R² of 0.99 for training set and 0.94 for test set, surpassing other methods. Regarding computational efficiency, GPO is the fastest, requiring 171.9 s, whereas BBO is the slowest at 298.2 s. SHAP analysis identifies BET surface area, ash content, and temperature as powerful factors affecting bio-oil yield, underscoring the efficacy of data-driven methodologies in addressing intricate systems. These models offer reliable tools for estimating bio-oil yield, reducing reliance on expensive, time-consuming, and resource-intensive experimental processes.