Reducing energy consumption through the implementation of AI to control aeration intensity in the composting process

The paper presents the results of a study on the use of artificial neural network (ANN) to optimize aeration intensity during the composting of municipal sewage sludge. Field experiments were carried out at a composting facility in Tczew, where a two-stage, six-week composting process was applied to a mixture of sewage sludge and wood chips in concrete reactors with forced aeration. Process monitoring encompassed 41 full-scale composting cycles under industrial conditions. For modeling purposes, a dataset was compiled that included physicochemical input variables (e.g., organic matter content, carbon, nitrogen, and pH) and online sensor data (temperature, moisture content, oxygen concentration, and airflow rate), together with output variables describing fan operation. A deep learning long short-term memory (LSTM) neural network for sequential data analysis was used to predict aeration intervals. The model was trained on 554 cases, approximately 12% of which were reserved for independent testing. The results demonstrated good agreement between ANN predictions and actual operating data, as well as the model's ability to capture the gradual decrease in oxygen demand as the composting process progressed. Under operational conditions, the ANN-based control algorithm reduced the specific electricity consumption for aeration from 4.1 to 3.1 kWh/Mg, corresponding to an approx. 24% reduction compared with conventional interval-based control. The developed model allows for optimizing the operating interval of the compost aeration fan without significantly affecting the rate of organic matter decomposition.