Potential of Biochar as Soil Amendment: Prediction of Elemental Ratios from Pyrolysis of Agriculture Biomass Using Artificial Neural Network

The rapid growth of the agriculture sector has been facing environmental issues with agriculture waste generation. Agriculture biomass is a good source for biochar production through the pyrolysis process. Biochar is a highly carbonaceous material and has been widely studied on its potential to improve soil quality. It is essential to understand and have a good prediction of biochar quality for biomass pre-screening. The elemental ratios and surface area both play an important role in determining the suitability of biochar as a soil amendment. In this study, a feedforward neural network (FFNN) with a backpropagation algorithm was developed to model the pyrolysis process in predicting the elemental ratios and surface area of various types of biochar using literature data. The O/C and H/C ratio are important parameters in soil quality to determine the stability of biochar in soil. Surface area is equally important to determine the porosity of biochar on its capability to retain water and nutrients. The optimization of the model was done by comparing the algorithm, transfer function, and hidden neurons. The prediction of the elemental ratios and surface area were based on the effect of pyrolysis temperature, heating rate, residence time, ultimate and proximate analysis. It was found that Levenberg–Marquardt backpropagation with ultimate analysis as input variable had the best results in terms of MSE (0.0087 and 0.0278), MAE (0.0594 and 0.0999), MAPE (17.835 and 11.891%), and R² (0.8601). A validation test was done on the developed model to test its capability to predict the outputs on a wide range of biomass feedstock. The test has shown good alignment with experimental data as a low MSE of 0.0161 is obtained. The model has the capability to achieve high accuracy in prediction with a high overall R² value and low MSE, MAE, and MAPE.