Scholarlink

<p>Anaerobic digestion (AD) offers a sustainable valorization route for agricultural residues, yet lipid-rich oilseed wastes introduce process variability (e.g., acidification, foaming) that complicates control. This study developed a data-driven artificial neural network (ANN) to forecast daily biogas production rates using six routine bench-scale measurements: pH, temperature (T), total solids (TS), volatile solids (VS), biochemical oxygen demand (BOD), and chemical oxygen demand (COD). Data were collected over a 41-day mesophilic batch run using an oilseed-and-straw substrate. To ensure robust generalization despite the limited sample size, the dataset was partitioned randomly, and a compact “nano-scale” multilayer perceptron architecture (6-2-1 topology) was employed. Weights and biases were initialized using a Genetic Algorithm (GA) to prevent local minima entrapment, followed by benchmarking of ten backpropagation optimizers. The Levenberg–Marquardt algorithm achieved the best performance (MSE = 7.72 × 10⁻<sup>4</sup>; RMSE = 0.0278 L.d<sup>−1</sup>) with stable generalization on the test set. Furthermore, a sensitivity analysis using Garson’s algorithm revealed that biodegradable organics (BOD, 31.6%; COD, 26.5%) and solids loading (TS, 15.1%) were the dominant predictors of gas production, outweighing pH and temperature in the stable mesophilic range. These findings establish a clear control hierarchy for lipid-rich digesters, prioritizing substrate loading management over reactive pH correction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Estimation of Biogas Production Rate from Oilseed and Straw in Anaerobic Digestion Laboratory Using Artificial Neural Network and Determination of the Most Effective Parameters on the Bio Gas Production Rate Using the Garson Algorithm

  • 期刊论文

  • 发表日期：

    2026年4月10日

查看全文
  • Abolghasem Pazoki,
  • Maryam Pazoki,
  • Marjan Khaez,
  • Fatemeh Mohammadzadehchali,
  • Reza Ghasemzadeh,
  • Babak Sheydaei,
  • Sanaz Tajziehchi

摘要

Anaerobic digestion (AD) offers a sustainable valorization route for agricultural residues, yet lipid-rich oilseed wastes introduce process variability (e.g., acidification, foaming) that complicates control. This study developed a data-driven artificial neural network (ANN) to forecast daily biogas production rates using six routine bench-scale measurements: pH, temperature (T), total solids (TS), volatile solids (VS), biochemical oxygen demand (BOD), and chemical oxygen demand (COD). Data were collected over a 41-day mesophilic batch run using an oilseed-and-straw substrate. To ensure robust generalization despite the limited sample size, the dataset was partitioned randomly, and a compact “nano-scale” multilayer perceptron architecture (6-2-1 topology) was employed. Weights and biases were initialized using a Genetic Algorithm (GA) to prevent local minima entrapment, followed by benchmarking of ten backpropagation optimizers. The Levenberg–Marquardt algorithm achieved the best performance (MSE = 7.72 × 10⁻4; RMSE = 0.0278 L.d−1) with stable generalization on the test set. Furthermore, a sensitivity analysis using Garson’s algorithm revealed that biodegradable organics (BOD, 31.6%; COD, 26.5%) and solids loading (TS, 15.1%) were the dominant predictors of gas production, outweighing pH and temperature in the stable mesophilic range. These findings establish a clear control hierarchy for lipid-rich digesters, prioritizing substrate loading management over reactive pH correction.