Machine learning based-predictive modeling with control mechanisms for enhanced biogas yield through internet of things-enabled sensor networks

Abstract

In the recent decade, global energy demand has increased in importance driven by industrialization, urbanization, and population growth. The government of Rwanda has been placing significant emphasis on adopting renewable energy sources as essential elements for sustainable energy strategies, to reduce reliance on fossil fuels. Bioenergy is a renewable energy source globally available with agricultural residues being a primary supplier to bioenergy production. This makes it a promising alternative energy source, as its characteristics ensure accessibility and affordability for the local community. Recently, Rwanda like other Eastern African Countries (EAC) has promoted biogas as an alternative source of cooking energy through various initiatives, specifically for rural communities. Despite various governments’ support policies, the adoption and diffusion of biogas technology have been considerably low. This research reported a lack of computing technology controlling the operating parameters and predicting biogas yield within biogas production supply chains as challenges for efficient biogas production. This research aims to achieve three objectives, first to develop an Internet of Things (IoT)based system for controlling biogas production and analyse the correlation between environmental data and biogas output. Second, to identify the most suitable power harvesting method to sustain the designed sensor nodes, and third, to design a machine learning (ML) model for predicting the biogas yield, and compare its performance against traditional models using the collected dataset. The research was executed in three sequential phases. For the first stage, an IoT prototype for data acquisition was developed, and tested to collect real-time data. During this phase, the thresholds for environmental parameters were determined experimentally and the actuation mechanisms were set to regulate the optimum condition. Additionally, an assessment of environmental data correlation was made with statistics and regression model, the variables are correlated in such a way that gives insights, and variability in biogas production is explained by R Squared (R2) 73.4% for the environment parameters explored, indicating a relatively good fit. In the second stage, an analysis of the power harvesting approach was made to ensure the sustainability of the deployed sensor node, and the power harvesting system specification was XIV derived from the sensor node energy consumption calculation. Further, a mathematical model predicting solar panel size was derived from two simultaneous functions, and global horizontal irradiation (GHI) as experimental input data. As a result, the designed sensor node can be powered by a solar panel size varied from 17.8 cm2 to 21.7 cm2. The model was tested on values data and it is generic to be adopted anywhere. In the third phase, the research focuses on improving biogas yield prediction using a hybrid ML approach. This approach integrates two data-driven models such as a light gradient-boosting machine (LightGBM) and categorical boosting (CatBoost) as based models. The hyperparameter turning using random search optimization with 5-fold cross-validation is adopted to avoid overfitting in each model. Further, the evolutionary strategy optimization technique is adopted to optimize the metal model. The hybrid model is applied to environmental data from biogas facilities, the model achieved superior performance with a mean squared error (RMSE) of 0.004 and mean absolute error (MAE) of 0.0024, surpassing k-nearest neighbor (KNN), random forest (RF), and decision tree (DT) models. The findings underscore the potential of accurate biogas yield prediction for optimizing energy production. This research demonstrates the contribution of emerging technology solutions as a way to meet growing global energy needs and advance sustainable biogas operations. In summary, developing and integrating IoT and predictive modeling for biogas production monitoring and prediction aligns with the International Energy Agency's (IEA) Net Zero by 2050 resolution. First, optimization of biogas production can lead to a reduction in greenhouse gas emissions, as the captured methane can be used for energy generation instead of being released into the atmosphere. Moreover, integrating these technologies can have a broader impact on the waste management sector, as optimizing biogas production from organic waste can incentivize the diversion of waste from landfills, decreasing the overall emissions and environmental impact.