摘要
为了预测一种固废声屏障的材料性能,并优化材料配比,研究提出了一种基于卷积神经网络和双向长短期记忆网络的混合神经网络模型,并引入注意力机制,采用试验数据对模型进行了训练。结合Pareto最优解的思想和北方苍鹰算法得到了多目标北方苍鹰优化算法,并基于该算法对模型中的关键参数进行了优化,使模型对主要性能指标的预测误差都降到0.8%以下,而未经优化的模型对部分指标的预测误差高达3.97%左右。以材料性能为目标,基于提出的高精度预测模型和多目标北方苍鹰算法对框架材料和吸声屏材料的配比进行了优化。经过配比优化后,框架材料和吸声屏材料的不同性能之间得到了较好的协调和总体提升,同时保证了较高的渣土利用率。相比于正交试验所得数据,优化后的框架材料在相近的渣土质量分数下,28D抗压强度提高了11.62%,28D抗折强度相差不多。在具有相近吸声系数的前提下,吸声屏材料的计权隔声量提高了65.82%,渣土质量分数提高了9.30百分点。研究提出的材料配比优化方法取得了较好的效果,能够有效提高新型固废声屏障材料的研发效率,降低研发成本。
To predict the material properties of a solid waste noise barrier based on its material proportions and optimize these proportions,we propose a hybrid neural network model that combines a convolutional neural network with a bidirectional long short-term memory network,incorporating an attention mechanism.The model is trained using orthogonal experimental data,which provides comprehensive intrinsic correlation information with minimal sample data,ensuring that it does not overfit within the upper and lower limits of the experimental dataset.A multi-objective Northern Goshawk Optimization(NGO)algorithm was developed by integrating the concept of Pareto optimal solutions with the NGO algorithm.Utilizing this approach,key parameters within the model were optimized,resulting in a reduction of prediction errors for the main performance indicators to below 0.8%.In contrast,the unoptimized model exhibited prediction errors for certain indicators as high as approximately 3.97%.Utilizing the proposed high-precision prediction model and the multi-objective NGO algorithm,the proportions of frame materials and sound-absorbing screen materials were optimized with material performance as the objective.Following the optimization,the distinct properties of both the frame material and the sound-absorbing screen material were effectively harmonized and enhanced,while also ensuring a high utilization rate of waste soil.After optimization,the 28-day compressive strength of the frame material increased by 11.62%compared to the data obtained from orthogonal experiments,while the 28-day flexural strength remained similar at comparable slag content levels.With sound absorption coefficients held constant,the weighted sound insulation improved by 65.82%,along with a 9.30 percentage points increase in slag content.The proposed material performance prediction model and proportion optimization method have yielded positive results,significantly enhancing the research and development efficiency of new solid waste noise barrier materials while reducing associated costs.
作者
曹芳
胡超
郭家舜
尚庆芳
吴桐
CAO Fang;HU Chao;GUO Jiashun;SHANG Qingfang;WU Tong(Hunan Changzhu Highway Development Co.,Ltd.,Changsha 410011,China;Key Laboratory of Road Construction Technology Equipment,State Ministry of Education,Chang'an University,Xi'an 710064,China;Detong Intelligent Technology Co.,Ltd.,Xuchang 461000,Henan,China;CCCC Hemei Environmental and Ecological Construction Co.,Ltd.,Wuhan 430000,China)
出处
《安全与环境学报》
北大核心
2025年第10期4032-4042,共11页
Journal of Safety and Environment
关键词
环境工程学
声屏障
固废材料
卷积神经网络
配比优化
environmental engineering
noise barrier
solid waste
convolutional neural network
proportions optimization
