Discrimination method of biomass slagging tendency based on particle swarm optimization deep neural network (DNN)

Abstract

The slagging problem that occurs during the combustion of biomass fuels is difficult to deal with, affecting the safe operation of the boiler and reducing the combustion efficiency. Predicting the slagging tendency of biomass combustion can guide the selection of fuel, reduce the experimental cost and improve the production efficiency of the boiler. In this paper, 114 kinds of biomass are selected as the data set. Through the visualization of the overall distribution of the data set, it is found that the data distribution has certain rules, but the whole overlap and cross distribution. The quantitative relationship between slagging degree classification and each element was obtained by Spearman correlation analysis. The higher the content of Ca, Mg and P elements, the smaller the slagging degree, while the more Si, Cl and Ash content, the more serious slagging. Taking 13 kinds of chemical elements and biomass ash content (%) as input and slagging type as output, and the network models of Deep Neural Network, Recurrent Neural Network and Long Short Term Memory Neural Network were built by python 3.6 in TensorFlow virtual environment. The four indexes of Accuracy, Precision, Recall and F1_score were used to evaluate the model. The accuracy of Deep Neural Network model was highest, which was 0.8260869565217391, 0.827380523809524, 0.8425925925926, 0.8322021116138764, respectively. The particle swarm optimization algorithm is used to optimize the highest precision DNN model, and the optimization accuracy is improved to 0.9130434. The optimized model was used to predict the slagging types of other 571 biomass, and the results provide guidance for the actual operation of biomass fuel boilers, thus saving experimental time and reducing experimental costs. At the same time, this study provides a new method for predicting the slagging trend of biofuels.