Search

Rapid pyrolysis is a promising technique to convert biomass into fuel oil, where NO X emission remains a substantial environmental risk. NH 3 and HCN are top precursors for NO X emission. In order to clarify their migration path and provide appropriate strategies for their controlling, six up-to-date machine learning (ML) models were established to predict the NH 3 and HCN yield during rapid pyrolysis of 26 biomass feedstocks. Cross-validation and grid search methods were used to determine the optimal hyperparameters for these ML models. The support vector regression (SVR) model achieved optimal accuracy among them. The optimal root means square error (%), mean absolute error (%), and R 2 of test set for NH 3 /HCN yield were 1.2901/1.1531, 1.0501/0.84712, and 0.98253/0.96152, respectively. In addition, based on the results of Pearson correlation analysis, the input variables with a weak linear correlation with the target product were eliminated, which was found capable of improving the prediction accuracy of almost all ML models except SVR. While after input variables elimination, the SVR model still showed the optimal NH 3 and HCN yield prediction accuracy. It reflects SVR's great significance and potential for predicting the yield of NO X precursors during rapid biomass pyrolysis., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2023 Elsevier B.V. All rights reserved.)

• Hyperspectral data contain enough but obscure information for MSW characterization. • Machine learning helps utilizing the full range of hyperspectral pattern. • Accuracy for organic/inorganic components classification reached nearly 100%. • Predicting accuracies of C/H/O contents and LHV reached 92.6%, 86.9%, 80.4%, 90.5%. • This work provides promising fundamentals for downstream treatment of MSW. Determining thermochemical properties and eliminating inorganic components of municipal solid waste (MSW) are crucial to its thermochemical treatment. Traditional characterization and classification technologies have shortcomings including long duration, complex operation, and inevitable sample consumption. This study proposed a hyperspectral imaging and machine learning models based method to solve these problems. Under the optimal parameter conditions, the identification accuracy of inorganic components by F1 scoring reached nearly 100% in MSW, and the prediction accuracy of carbon, hydrogen, oxygen, nitrogen contents and low heating value (LHV)of organic components by mean relative error value reached 92.6%, 86.9%, 80.4%, 54.7% and 90.5%, respectively. The results validated the hypothesis that combination of hyperspectral imaging and machine learning models are promising to accomplish fast characterization and classification of components in MSW, where principal component analysis was capable to abstract crucial information from the spectral pattern, and artificial neural network presented satisfactory classification and regression performance. [Display omitted] [ABSTRACT FROM AUTHOR]

• A LIBS and machine learning based system was proposed to characterize residual wastes. • The optimal classification and characterization accuracy could reach 95%. • The working mechanism, robustness and features of the system were demonstrated. • The potential application scene and limitations of the system were discussed. • The system is promising to enhance downstream utilization of residual wastes. Elemental composition and heating value are essential properties of residual wastes (RW) for its energy utilization. This paper proposed a highly efficient approach to distinguish inorganic components and characterize organic compounds in RW via laser-induced breakdown spectroscopy (LIBS) and machine learning (ML) models. LIBS data of various RW samples were collected to train and test the hybrid model, which includes a data pretreatment module, a classification module and a regression module. Impacts of different ML model categories and parameters were investigated and discussed. Under optimal conditions, the accuracy for predicting C content, H content, O content and lower heating value reached 96.70%, 92.21%, 87.11% and 94.28%, respectively. The robustness of this system was validated. The future application of the model and their limitation were also discussed. This method provides innovative technical ideas for the identification and characterization of RW, and has important potential value for the energy treatment and utilization of RW. [Display omitted] [ABSTRACT FROM AUTHOR]

[Display omitted] • Combining machine learning and physical insights to study biochar active sites. • Internal and external characters constructed multidimensional fingerprint descriptor. • C-C/C=C, C=O and defect sites were all predicted with over 90 % accuracy. • The descriptors enable high-throughput screening of preparation parameters. Highly active biochar has great application potential in heterogeneous catalysis and adsorptive processes. The complexity of carbonization process makes it difficult to construct target active sites. This work put forward a reactive descriptor based on pyrolysis parameters and intrinsic composition of biomass. Results show that the model showed better predictive performance for C-C/C=C (R2 = 0.85), C=O (R2 = 0.85) and defect (R2 = 0.91) sites. The SHapley Additive exPlanation analysis shows that the pyrolysis parameters and the higher heating values are equally important for the active sites. The predictive performance and guiding role of the descriptor were validated by experiments. The descriptors proposed in this study integrated significant advantages of simplicity and easy accessibility, which would break the bottleneck of accurate construction of active sites and provide a theoretical basis for high-value resource utilization of biomass. [ABSTRACT FROM AUTHOR]

Model pyrolysis oil (bio-oil) is a promising bio-fuel with complex and unstable contents, making its fast characterization an essential demand in industrial production. This study proposed an integrated machine learning framework to predict elemental composition, low heating value, and unsaturated concentration from infrared spectra, achieving fast and interpretable characterization of bio-oil. In the integrated framework, a peak loading-based strategy was used to dimensionally reduce the spectral data. Bayesian optimized random forest (RF) and extreme gradient boosting (XGBoost) models were used to predict bio-oil properties from dimensionally reduced spectral data. Ensemble learning was used to combine RF and XGBoost models together for better predicting performance. Results showed that the proposed characterization method achieved an average accuracy of 99.53 %, a low RMSE value of 0.726, and an R2 of 0.98. The Shapley value analysis revealed that the vibration of NH 2 stretch (1594 cm−1), C-H stretch (2868 cm−1), and C-N stretch in the aromatic ring (1229 cm−1) have a significant contribution to the characterization results. The working mechanism of the proposed characterization method was interpreted by the internal relationship among spectral peak location/height, functional group species/amount, and the predicted characteristics. The results are hoped to serve quality control in production of bio-oil. [Display omitted] [ABSTRACT FROM AUTHOR]

[Display omitted] • The descriptor γ was constructed by integrating the C%, O%, I D /I G , and SSA of BBC. • γ can describe the activity of BBC accurately for triggering NR in PMS system. • O% played the most vital role in facilitating NR contribution. • C% (50–60 %), O% (30–50 %), I D /I G (0.4–1.2), and SSA (579–1259 m2/g) promoted NR. The peroxymonosulfate (PMS) activation was influenced strongly by active sites on a catalyst surface. To some extent, biochar characters, such as structure, elemental composition, and specific surface area (SSA), can reflect the composition of active sites. However, the understanding of active centers lacked accuracy in the structure–activity relationships of biochar-based catalysts (BBC) due to the consideration of one single feature. In this study, XGBoost (XGB), random forest regression (RF), and supporting vector regression (SVR) models were employed to construct descriptors to predict non-radical (NR) contribution in BBC/PMS systems. Consequently, the XGB model provided the best prediction with an R2 value of 0.81. Thereby, activation preference (γ), as a characteristic descriptor of BBC, was developed using the XGB model. Interestingly, the descriptor fully considered carbon content (C%), oxygen content (O%), defects (I D /I G), and SSA. Besides, O% played the most critical role in predicting NR contribution. Noticeably, the NR oxidation pathway could be promoted within the range of 50–60 % for C%, 30–50 % for O%, 0.4–1.2 for I D /I G , and 579–1259 m2/g for SSA in the BBC/PMS system. Herein, machine learning (ML) was applied to develop a comprehensive descriptor, further providing an innovative strategy for high-throughput screening of BBCs to activate PMS efficiently. The comprehensive descriptor offered a new perspective for elucidating the structure–activity relationship of BBC, facilitating the application of BBC/PMS systems in water treatment. [ABSTRACT FROM AUTHOR]

This study proposed a fast characterization method of bio-oil via the combination of attenuated total reflection flourier transformed infrared spectroscopy (ATR-FTIR) and machine learning models. The input to the model is high-dimensional infrared spectral data. Unsaturated concentration, effective hydrocarbon ratio, low calorific value, C content, H content, and O content are all relevant bio-oil indicators. The model parameters were optimized based on prediction accuracy and correlation coefficient. By comparing the sole support vector regression (SVR) model versus principal component analysis (PCA) preprocessed SVR model, the results showed that PCA preprocessing can significantly improve the overall performance of SVR model towards prediction of bio-oil characteristics. Under optimal parameters, the predicted accuracies for unsaturated concentration, effective hydrocarbon ratio, low calorific value, C content, H content, and O content reached 91.98%, 97.44%, 99.50%, 98.65%, 98.56%, and 97.88%, respectively. The correlation coefficient of sole SVR model was 0.3, and the correlation coefficient of PCA preprocessed SVR model was 0.9. Furthermore, the characteristic peaks of the infrared spectra at the optimal PC were analyzed, and PC6 and PC7 were found to have the most influence on the predicting performance. [Display omitted] [ABSTRACT FROM AUTHOR]

Anaerobic digestion is a promising approach to dispose of biodegradable waste and wastewater, generating biogas as an alternative energy resource. This work proposed a so-called M-CADM1 for continuous anaerobic digestion simulation, which combined the machine learning and anaerobic digestion model No.1 (ADM1). The detailed reaction path and intermediate products in different stages of anaerobic digestion are specified in ADM1. The kinetic parameters were modified by machine learning. The characteristics (elemental composition) of feedstocks were used to predict kinetic parameters. A total of 75 biomass samples were used to establish for machine learning models. Five element contents (C, H, O, N, S), feedstock feed rate, and anaerobic digestion temperature were used as the input. The kinetic parameters were set as output. The sensitivities of 17 kinetic parameters were evaluated. 7 kinetic parameters with the highest sensitivities were selected as ADM1 model inputs by sensitivity analysis. The R2 and RMSE were used as the index to evaluated the accuracy of machine learning model. The best R2 and RMSE reached 0.84 and 0.196. The TIC was used as the index to evaluated the accuracy of M-CADM1. By comparing the simulated value with the experimental value, the accuracy of the overall M-CADM1 expressed by TIC of kitchen waste was 0.036. The organic acid content and pH in the reactor were considered as indicators to study the accuracy and stability of the M-CADM1. Trends in organic acids, free ammonia or hydrogen inhibition, and pH were consistent with experimental continuous anaerobic digestion results. [Display omitted] • M-CADM1 model was proposed to simulate biomass continuous anaerobic digestion. • Four machine learning models were tested for crucial kinetic parameters prediction. • Predicted kinetic parameters were embed in ADM1 for products prediction. • The R2 and RMSE for kinetic parameters predicting model reached 0.84 and 0.196. • The optimal overall TIC of established M-CADM1 model was 0.036. [ABSTRACT FROM AUTHOR]

• A new method is proposed to characterize chemical features of biomass and waste. • The method is based on infrared spectroscopy and machine learning models. • The method is fast and the optimal accuracy could reach as high as 95.54%. • The robustness of this method is validated, and its properties are discussed. • The method can enhance sorting of biomass and waste for downstream utilization. Heterogeneity is a most serious obstacle for treatment and utilization of biomass and waste (BW). This paper proposed a fast characterization method based on infrared spectroscopy and machine learning models, thus to roughly predict the elemental composition and heating value of BW. The fast characterization results could be used to sort different BW components by their suitable downstream utilization techniques. The infrared spectra based hybrid model contained a feature compression section to extract core information from raw infrared spectra, a classification section to distinguish inorganic dilution, and a regression section to generate the elemental composition and heating value results. By parameters optimization, the accuracy of this hybrid model reached 95.54%, 85.53%, 92.40%, and 92.49% for C content, H content, O content, and low heating value prediction, respectively. The robustness analysis was conducted by completely rearranging the training and test sets, and it further validated the hypothesis that the infrared spectra contains enough qualifying and quantifying information to characterize these properties of BW. Compared with previous literature, the C–H, C–O, and O–H correlations in BW were also well kept in the predicted results. This work is hoped to enhance upstream sorting system design for treatment and utilization of BW. [ABSTRACT FROM AUTHOR]