High-precision multi-target prediction and interpretability analysis of biomass gasification via ensemble machine...

Area of Science:

• Thermochemical processes
• Biomass energy conversion
• Computational modeling

Background:

• Machine learning (ML) offers powerful nonlinear fitting for complex thermochemical processes like gasification.
• Data preprocessing, including imputation, encoding, and standardization, is crucial for enhancing data quality in ML models.

Purpose of the Study:

• To evaluate the performance of six mainstream ML models for predicting biomass gasification product composition, yield, and efficiency.
• To compare hyperparameter optimization strategies and ensemble methods for improving predictive accuracy.
• To provide reliable decision support for gasification technology selection and design.

Main Methods:

• Data imputation, categorical variable encoding, and standardization were applied to improve data quality.
• Six ML models were trained and evaluated using four hyperparameter optimization strategies.
• Ensemble methods, including Weighted Averaging, were employed to integrate individual model strengths.
• Global and local interpretability techniques were utilized to analyze model predictions.

Main Results:

• Most ML models achieved a coefficient of determination (R ² ) exceeding 0.90 after optimization.
• Ensemble models demonstrated superior performance compared to single-model approaches.
• The Weighted Averaging ensemble achieved an average test R ² of 0.9251, outperforming individual models.
• Interpretability techniques provided insights into the prediction drivers.

Conclusions:

• Machine learning, particularly ensemble methods, offers high-precision prediction capabilities for biomass gasification.
• Optimized ML models and interpretability tools enhance decision-making for gasification process design and investment.
• This study validates the effectiveness of advanced ML techniques in optimizing thermochemical process simulations.