PREDICTION OF MICROBIAL ACTIVITY DURING BIOSOLIDS COMPOSTING USING ARTIFICIAL NEURAL NETWORKS

In order to determine the relationships among temperature, moisture content, and microbial activity during biosolids composting, well-controlled composting experiments were conducted using 2-factor factorial design with six temperatures (22.C, 29.C, 36.C, 43.C, 50.C, and 57.C) and five moisture contents (30%, 40%, 50%, 60%, and 70%). The microbial activity was indicated by O2 uptake rate (mg gdry matter -1 h-1) and cumulative O2 uptake (mg gdry matter -1) over time. An artificial neural network (ANN) model to predict microbial activity was developed using 8760 data patterns. After experimentation with different ANN architectures (i.e., standard backpropagation, Jordan, and Ward) and model parameter settings (i.e., data partitioning strategy, number of hidden nodes, learning rate, and momentum), the best prediction model was a combined Ward network for both O2 uptake rate and cumulative O2 uptake. An additional composting experiment at 34.C with five moisture contents was conducted, and the corresponding 1460 data patterns were used for model evaluation. The model evaluation with 34.C data showed good prediction results for both O2 uptake rate (r2 = 0.928, MAE = 0.08 mg g-1 h -1) and cumulative O2 uptake (r2 = 0.986, MAE = 5.28 mg g -1).