Anaerobic fermentation gas production prediction based on SCSSA-CNN-BiLSTM neural network

Anaerobic fermentation, as a highly efficient organic waste treatment technology, can convert agricultural waste into biogas for resource recycling and sustainable energy supply. The anaerobic fermentation process is affected by factors such as the carbon-to-nitrogen ratio of the reaction substrate, pH, volatile fatty acids, ammonia nitrogen concentration, and chemical oxygen demand. To explore the patterns of anaerobic fermentation, gas production experiments of anaerobic fermentation with mixed raw materials were carried out, using mixed raw materials with cow dung-to-corn stover ratios of 1∶1, 2∶1, and 3∶1. Three parallel experiments for each group were conducted to ensure the reliability and reproducibility of the experimental results. Subsequently, a Sine Cosine and Cauchy mutation strategy-enhanced Sparrow Search Algorithm (SCSSA) was proposed to optimise the hyperparameters of a Convolutional Neural Network integrated with Bidirectional Long Short-Term Memory (CNN-BiLSTM). The reaction time, cow dung to corn stover ratios, pH, volatile fatty acids, ammonia nitrogen concentration, and COD were selected as input parameters to the model, and daily gas production and daily methane yield were selected as output parameters. The results showed that the highest methane production was achieved with a 3∶1 ratio of cow dung to corn stover, followed by a 1∶1 ratio, and the smallest yield was achieved with a 2∶1 ratio in the experimental group. The SCSSA-CNN-BiLSTM hybrid feedstock anaerobic fermentation gas production prediction model achieved high prediction accuracy, with 95.29% for daily gas production and 95.87% for daily methane yield, and a goodness-of-fit (R²) of 0.972. The approach solves the problem of premature convergence into the local optimum in the traditional sparrow searching algorithm and enhances the global searching capability, which provides the basis for the practical experiments.