基于非线性自回归神经网络模型对生活垃圾产生量的预测  

Prediction of Domestic Waste Output Based on Nonlinear Autoregressive Neural Network Model

作　　者：朱远超 王晓燕 田光 ZHU Yuan-chao;WANG Xiao-yan;TIAN Guang(Beijing Municipal Institute of City Management,Beijing 100028,China;Beijing Key Laboratory of Municipal Solid Waste Detection Analysis&Evaluation,Beijing 100028,China)

机构地区：[1]北京市城市管理研究院,北京100028 [2]生活垃圾检测分析与评价北京市重点实验室,北京100028

出　　处：《四川环境》2024年第3期149-153,共5页Sichuan Environment

摘　　要：旨在建立生活垃圾产生量预测模型,更好的预测生活垃圾产生量,以便有序筹划生活垃圾处置设施和构建灵活的收运调配体系。方法采用非线性自回归神经网络(NAR),通过调整延迟阶数和隐含层神经元个数等模型参数,建立基于生活垃圾产生量的历史时间序列预测模型。实验结果显示,NAR神经网络时间序列模型对于北京市生活垃圾产生量有较好的预测能力,当延迟阶数为5,隐含神经元个数为10时,预测模型测试集的r值为0.9717,平均绝对百分比误差为3.385%,均方根误差为5051.831 t/w,预测模型通过了残差序列非自相关检验,预测效果较好。结论表明针对生活垃圾产生量数据可以开展NAR神经网络模型非线性自回归预测,且可不用考虑其它相关影响因素数据的可获得性,具有一定的便利和实际应用意义。In this paper,a prediction model for the domestic waste output was established to better predict the domestic waste output,which can help building a plan for domestic waste disposal facilities and a flexible collection and transportation system.By using the non-linear auto-aggressive neutral network(NAR)and testing the model parameters such as delay order and the number of hidden layer neurons,a prediction model based on historical time series of domestic waste output was built.The results showed:The NAR neural network time series model had good predictability for the domestic waste output in Beijing.When the delay order was 5 and the number of hidden neurons was 10,the model achieved an r-value of 0.9717,a mean absolute percentage error of 3.385%,and a root mean square error of 5051.831 t/w.The residual sequence also passed the non-autocorrelation test,indicating a good prediction performance.The conclusion also indicated that NAR neural network model can be used for nonlinear autoregressive prediction of domestic waste output,without considering the availability of other related influencing factors,and had specific convenience and practical application significance.

关 键 词：生活垃圾 预测模型 非线性自回归 神经网络 

分 类 号：X33[环境科学与工程—环境工程] X799.3