摘要
采用神经网络方法对垃圾热值进行了预测。通过对垃圾组分与热值的相关性分析得知城市生活垃圾的热值与塑料和纸的关系最密切 ,并采用多元线性回归方法得出热值与物理组成的关系。针对垃圾成份的复杂多变性 ,采用神经网络方法对城市生活垃圾的热值进行了预测。神经网络以垃圾的物理组成 (塑料、纸、食品、草木和织物 )作为输入 ,采用 10 8组数据和BP算法对网络进行训练 ,发现采用隐层单元数为 7,学习速率为 0 1时 ,网络收敛速度较快 ,同时给出均方差随计算次数的变化关系 ,并将计算结果与实验测量值进行了比较。结果显示 10 8组数据中仅有 4组数据与测量值的相对误差超过 5 % ,其余数据均在 5 %误差范围内 ,比多元线性回归方法有较大改善。
A neural network method is used to forecast the heating value of municipal solid waste (MSW). Through a correlation analysis of the MSW components and their heating value it is discovered that the heating value of the MSW is immensely dependent on such materials as plastics and paper. With the help of a multiple linear regression method the relationship of the heating value of MSW and its physical composition was obtained. In light of the complexity and high variation of MSW composition a neural network method is employed to forecast the heating value of MSW. With the physical composition of MSW (plastics, paper, foodstuff, grass wood and textile fabric) serving as the input of the neural network a training course was conducted of the neural network with the use of 108 groups of data and the BP algorithm. It has been found that the convergence speed of the network is relatively high when hidden units totaling 7 and the learning rate of 0. 1 were used. Meanwhile, given is the variation of mean square deviation with the number of calculation times. The calculated results were compared with the test and measured ones. It is found that among the 108 groups of data only four groups have a relative error higher than 5% as compared with measured values. The error of the remaining groups of data does not exceed 5%, which represents a relatively great improvement as compared with the multiple linear regression method.
出处
《热能动力工程》
EI
CAS
CSCD
北大核心
2002年第3期275-278,共4页
Journal of Engineering for Thermal Energy and Power
基金
国家教育部科学技术研究重点基金资助项目 (99174)
高等学校重点实验室访问学者基金资助项目 (80 0 13 16)
国家教育部<跨世纪优秀人才计划>基金资助项目 (K980 0 2 6
江苏省教育厅"青蓝工程"基金资助项目 (JS980 8)
关键词
神经网络法
预测
城市
生活垃圾
热值
Algorithms
Composition
Heating
Neural networks
Regression analysis
