Prediction of gas‐liquid‐solid product distribution after solid waste pyrolysis process based on artificial neural network model

Abstract

Four solid wastes including sludge, watermelon rind, corncob, and eucalyptus and their demineralized samples were selected to conduct pyrolysis experiments under different experimental conditions, including temperature, residence time, carrier gas flow rate, and heating rate, respectively. The co-pyrolysis experiment was carried out after mixing different types and different ratios of solid wastes to investigate the influence of different factors on yield of char, tar, and gas. A three-layer artificial neural network (ANN) based on back propagation (BP) algorithm was developed and trained to simulate and predict the yield of products. The experimental conditions and characteristic parameters of samples, including the content of C, H, K, Ca, Mg, Fe, volatile, ash, and fixed carbon, were selected as input parameters while the yields of char, tar, and gas were selected as output parameters. The effect of input parameters including residence time, carrier gas flow rate, heating rate, and the content of metal elements on the network performance was investigated in detail to optimize the ANN model. It was found that the metal element has the greatest influence, and its importance to the yield of each product exceeds 25%. The pyrolysis experimental data of single-component solid waste were selected as the model training set for model learning and training and then use the trained model to predict the co-pyrolysis experimental data of multi-component solid waste. Good agreement was achieved between experimental and predicted results; the correlation coefficient was 0.9836, and the root mean square error (RMSE) was only 2.4273, which revealed that the ANN method can be effectively used in predicting of the gas-liquid-solid product distribution after solid waste pyrolysis process. Novelty Statement In this manuscript, based on the principle of artificial neural network, experimental conditions and sample characteristic parameters are used as input variables to model the distribution of different sample pyrolysis products. The model is used to predict the co-pyrolysis data, and the influences of input parameters on the accuracy of the model are analyzed. The model values are in good agreement with the experimental data, indicating that the model is a high accurate general model and can be applied to different samples.