ABSTRACT

This paper describes the development of metaheuristic based artificial neural network (i.e., ANN optimized with PSO, ANN-PSO) and multilinear regression models to predict the higher heating values (HHV) of solid fuels based on the parameters of proximate and ultimate analyzes of the solid fuels. Three hundred data points of HHVs, proximate and ultimate analyzes obtained from published papers on solid fuels are used in this study. The results of the proximate and ultimate analyzes are used in the training and development of ANN-PSO models as well as the development of multilinear regression models. The models were tested for performance validation. The performances of the proposed models were evaluated using mean absolute error (MAE), average absolute error (AAE) and average biased error (ABE). Based on good agreement between results and other statistical performance parameters, ANN-PSO models perform better than multilinear regression models. The ANN-PSO models can predict the higher heating values of solid fuels for practical applications. The ANN-PSO models demonstrated excellent predictive ability showing predicted experimental HHV ratio that is close to 1.00.

摘要

本文介绍了基于元启发式的人工神经网络（即，用 PSO 优化的 ANN，ANN-PSO）和多线性回归模型的发展，以基于对固体燃料的近似和最终分析的参数来预测固体燃料的高热值 (HHV)。固体燃料。本研究使用了从已发表的固体燃料论文中获得的三百个 HHV 数据点、近似分析和最终分析。近似分析和最终分析的结果用于训练和开发 ANN-PSO 模型以及开发多线性回归模型。对模型进行了性能验证测试。使用平均绝对误差 (MAE)、平均绝对误差 (AAE) 和平均偏差误差 (ABE) 来评估所提出模型的性能。基于结果与其他统计性能参数之间的良好一致性，ANN-PSO 模型比多线性回归模型表现更好。ANN-PSO 模型可以预测实际应用中固体燃料的更高热值。ANN-PSO 模型表现出优异的预测能力，显示预测的实验 HHV 比率接近 1.00。