基于遗传BP神经网络的Al/Fe_(2)O_(3)铝热剂热力学性能预测方法研究  

Research on the Prediction Method of Thermodynamic Properties of Al/Fe_(2)O_(3) Aluminum Thermite Based on Genetic BP Neural Network

作　　者：薛美晨 薛园园[2] 胡耀立 欧阳的华[3] 彭鹏 罗彬杰 XUE Mei-chen;XUE Yuan-yuan;HU Yao-li;OUYANG Di-hua;PENG Peng;LUO Bin-jie(School of Resources Engineering,Xi’an University of Architecture and Technology,Xi’an,710055;Shaanxi Applied Physics and Chemistry Research Institute,Xi’an,710061;School of Chemistry and Chemical Engineering,Nanjing University of Science and Technology,Nanjing,210094;Hubei Handan Electromechanical Co.,Ltd,Xiangyang,441000)

机构地区：[1]西安建筑科技大学资源工程学院,陕西西安710055 [2]陕西应用物理化学研究所,陕西西安710061 [3]南京理工大学化学与化工学院,江苏南京210094 [4]湖北汉丹机电有限公司,湖北襄阳441000

出　　处：《火工品》2025年第2期45-52,共8页Initiators & Pyrotechnics

基　　金：陕西省自然科学基金面上项目(2024JC-YBMS 414)。

摘　　要：为了探索Al/Fe_(2)O_(3)铝热剂热力学性能的预测方法,采用遗传算法对BP神经网络的初始权值和阈值进行优化,利用HSC Chemistry软件计算了不同温度及配比下的Al/Fe_(2)O_(3)吉布斯自由能及反应焓变,以此作为基础数据,建立了基于遗传BP神经网络的Al/Fe_(2)O_(3)铝热剂燃烧热力学性能预测模型,分别得到了124组Al/Fe_(2)O_(3)吉布斯自由能和化学反应焓变的训练集样本数据及31组预测集样本数据,并计算了模型误差。结果表明:该模型预测的Al/Fe_(2)O_(3)吉布斯自由能及反应焓变的测试集均方根误差(RMSE)分别为0.499 1和0.702 7,平均绝对误差(MAE)分别为0.533 2和0.4411,决定性系数R^(2)分别为0.982 7和0.988 5;遗传BP神经网络能够用于Al/Fe_(2)O_(3)铝热剂热力学性能的预测。In order to explore the prediction method for the thermodynamic properties of Al/Fe_(2)O_(3) aluminum thermite,genetic algorithm was used to optimize the initial weights and thresholds of BP neural network.HSC Chemistry software was used to calculate the Gibbs free energy and reaction enthalpy change of Al/Fe_(2)O_(3) at different temperatures and ratios.Based on this,a prediction model for the combustion thermodynamic properties of Al/Fe_(2)O_(3) aluminum thermite was established using genetic BP neural network.124 sets of training set sample data and 31 sets of prediction set sample data for Gibbs free energy and chemical reaction enthalpy change of Al/Fe_(2)O_(3) were obtained,and the model error was calculated.The results show that the root mean square errors(RMSE)of the Gibbs free energy and reaction enthalpy change predicted by the model for Al/Fe_(2)O_(3) are 0.4991 and 0.7027,respectively,with mean absolute errors(MAE)of 0.5332 and 0.4411,and determination coefficients R^(2) of 0.9827 and 0.9885,respectively;Genetic BP neural network can be used for predicting the thermodynamic properties of Al/Fe_(2)O_(3) aluminum thermite.

关 键 词：铝热剂 吉布斯自由能 反应焓变 遗传算法 BP神经网络 

分 类 号：TJ55[兵器科学与技术—军事化学与烟火技术] TQ560.7[化学工程—炸药化工]