In a paper published in the journal Sustainability, researchers showcase the significance of solid biofuels and IoT technologies in smart city development. Solid biofuels, derived from organic sources like wood and agricultural waste, offer renewable energy solutions. Integrating IoT with solid biofuel classification enhances production control. The paper introduces Solid Biofuel Classification using Sailfish Optimizer Hybrid Deep Learning (SBFC-SFOHDL), an Internet of Things (IoT)-based method that employs deep learning for biofuel classification. Simulation results demonstrate its superiority over existing models.
Study: Advancing Solid Biofuels Classification in IoT-driven Smart Cities. Image credit: jamesteohart/Shutterstock
Background
Agricultural and forestry waste results in substantial biomass waste, raising environmental concerns globally. Despite being discarded, waste like fruit peels and crop residues has the potential to be used as feedstock. Fuel type is pivotal in energy applications, affecting processing and results. Incineration is a common power generation choice, emphasizing emission reduction. The choice of technique depends on the type of fuel and its impact on the environment. Machine learning (ML) enables prediction without requiring explicit coding and finds extensive application in various fields, including biomass gasification.
Related work
Numerous studies have focused on efficient energy extraction from agricultural waste. An intelligent ensemble technique is developed for solid fuel classification, integrating deep learning (DL) methods like Convolutional Neural Networks and Long Short-Term Memory (CNN-LSTM), Gated Recurrent Unit (GRU), and LSTM. Other research explores topics such as digestate-based hydrochar production, energy mining from agricultural waste, biomass briquette viability, traceability systems for pruning biomass, and optimized co-pelletization processes.
Proposed method
The present study focuses on enhancing the SBFC-SFOHDL model within the IoT framework. The primary objective of this paper is to proficiently detect and classify solid biofuels derived from agricultural residues in the IoT environment. The SBFC-SFOHDL system encompasses IoT-based data collection, data preprocessing, The Multihead Self Attention-based Convolutional Bidirectional Long Short-Term Memory model (MSA-CBLSTM) based solid fuel classification, and Sailfish Optimizer (SFO) based hyperparameter tuning.
Initially, input data undergo preprocessing to make them compatible for analysis. Subsequently, the MSA-CBLSTM model is employed for solid fuel classification, categorizing the biofuels into distinct classes. Finally, the SFO algorithm optimizes the hyperparameters of the MSA-CBLSTM model to enhance its classification performance.
In the preprocessing phase, three approaches are used: data normalization, data transformation, and class labeling. These steps prepare the data for analysis, converting categorical values into mathematical ones and assigning suitable class labels. The MSA-CBLSTM model is applied for classification, and its bidirectional nature allows dynamic data extraction from preceding and subsequent segments.
To optimize the hyperparameters of the MSA-CBLSTM model, this study utilizes the SFO algorithm, which draws inspiration from the hunting behavior of sailfish. This algorithm's distinct features include its flexibility in exploration and diversification, enabling it to efficiently search the solution space for better solutions. The SFO technique involves several steps, such as population initialization, elitism evaluation, and sailfish position updating, ultimately improving the model's convergence and performance.
Experimental results
The study assesses the performance of the SBFC-SFOHDL model using a dataset that encompasses four distinct classes: manufactured biomass, coal, wood, and agricultural residues. Notably, the accuracy of the SBFC-SFOHDL model increases as the number of epochs rises—achieving 94.27% accuracy for 500 epochs, 97.01% for 1500 epochs, and a peak of 98.63% for 3000 epochs. Furthermore, metrics such as precision-recall values, Fscore, and Matthews Correlation Coefficient (MCC) scores all exhibit consistent improvements.
Comparatively, the SBFC-SFOHDL technique outperforms several other models, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), flat classifier, Hierarchical Clustering (HC), and Intelligent Ensemble of Voting-Based Solid Fuel Classification (IEVB-SFC) models, across precision, recall, accuracy, and Fscore metrics. This superiority is attributed to the incorporation of SFO-based hyperparameter tuning, which significantly contributes to the model's enhanced accuracy and overall performance. In summary, the SBFC-SFOHDL approach presents a robust solution with the potential to outperform existing methods in solid fuel classification tasks.
Contributions of the paper
The key contributions of this study can be summarized as follows:
- Innovative Classification Method: The paper introduces SBFC-SFOHDL, a novel method for classifying solid biofuels from agricultural residues within the context of an IoT environment.
- IoT and Deep Learning Fusion: By integrating IoT technology and deep learning, the study advances the accuracy of solid biofuel classification, enhancing the potential for sustainable energy extraction.
- MSA-CBLSTM Model: The proposed technique employs the MSA-CBLSTM model, combining convolutional and bidirectional LSTM components for effective feature extraction and robust classification of solid biofuels.
- SFO Hyperparameter Tuning: The inclusion of the SFO algorithm optimizes the MSA-CBLSTM model's performance by fine-tuning hyperparameters, improving accuracy and classification outcomes.
- Comprehensive Performance Assessment: Through thorough evaluation using various metrics, including accuracy, precision-recall, Fscore, and MCC, the paper demonstrates the superior performance of SBFC-SFOHDL in comparison to existing models, contributing to the field of sustainable energy generation.
Conclusion
To sum up, this study focuses on enhancing the SBFC-SFOHDL model within the IoT platform for proficiently detecting and classifying solid biofuels from agricultural residues. The proposed algorithm includes IoT-based data collection, preprocessing, MSA-CBLSTM-based fuel classification, and SFO-based hyperparameter tuning. Leveraging the SFO technique aids in optimal hyperparameter selection, boosting MSA-CBLSTM's classification accuracy. Simulation results validate SBFC-SFOHDL's superiority over recent DL methods, with higher accuracy, precision, recall, and Fscore scores. Future endeavors may explore feature fusion-based DL techniques for further enhancing solid biofuel classification.
IoT and Machine Learning for Pure Gas Detection