Intelligent system for automated monitoring of biomaterial quality for digestion in a biogas plant [in Ukrainian]

To address the challenge of inefficient and non-real-time control of biogas plants, an intelligent automated monitoring system was developed. The objective is to enhance the stability and productivity of anaerobic digestion through the implementation of proactive data analysis and adaptive control. The study was conducted on a pilot biogas plant using fermented biomaterial. An ESP32-based IoT device equipped with temperature sensors and a camera was used for data collection. Data analysis was performed on the AWS cloud infrastructure, utilizing S3, DynamoDB, and Fargate services, along with an ensemble of machine learning (XGBoost, LSTM) and computer vision (ResNet-50) models. During a 6-month testing period, the system demonstrated high reliability, collecting 98.6 % of planned measurements, and efficiency, with an average data processing time of 3.2 minutes. For substrate image analysis, the Otsu segmentation method with Contrast Limited Adaptive Histogram Equalization (CLAHE) was validated and selected, showing the lowest coefficient of variation at 3.63 %, indicating its stability. The initial predictive machine learning model achieved a coefficient of determination (R²) of 0.85 and a Root Mean Square Error (RMSE) of 0.61 m³/day. A key result is the effectiveness of the adaptive retraining mechanism: after the first update cycle with 100 new, operator-validated records, the model's accuracy significantly increased. The R² score rose to 0.91, and the RMSE decreased by 42.6 % to 0.35 m³/day. This proves that the system can effectively learn from real operational data, continuously improving its predictive accuracy and, consequently, the quality of its management recommendations.