基于深度学习的香菇菌棒生产成本预测模型研究  

Research on Production Cost Prediction Model of Lentinula edodes Logs Based on Deep Learning

作　　者：卢翠红 张峰 吴秋兰[1] Lu Cuihong;Zhang Feng;Wu Qiulan(College of Information Science and Engineering,Shandong Agricultural University,Taian 271018,China)

机构地区：[1]山东农业大学信息科学与工程学院,山东泰安271018

出　　处：《山东农业科学》2025年第1期174-180,共7页Shandong Agricultural Sciences

基　　金：山东省重大科技创新工程项目(2022CXGC010609)。

摘　　要：针对香菇菌棒生产成本管控难、成本预测精度低等问题,本研究在深入剖析香菇菌棒生产成本关键影响因素的基础上,提出了基于时间卷积神经网络(TCN)、双向长短期记忆网络(BiLSTM)和注意力(Attention)机制的香菇菌棒生产成本预测模型。首先利用轻量级梯度提升机(LightGBM)筛选出与香菇菌棒生产成本相关的重要特征,降低预测模型的输入维度;然后构建TCN网络与BiLSTM网络对输入数据进行特征提取,并将提取的特征进行融合;最后在上述基础上添加Attention机制,使用全连接层得到最终的香菇菌棒生产成本预测结果。实验结果表明,该模型的预测均方根误差、平均绝对百分比误差、平均绝对误差分别为0.0841、2.2526、0.0738,香菇菌棒生产成本的预测曲线接近真实的曲线,具有良好的预测效果,可以有效满足香菇菌棒生产企业对成本预测的要求。Aiming at the problems of difficulty in cost control and low cost forecasting accuracy for the production of Lentinula edodes logs,on the basis of in-depth analysis of the key factors affecting the production cost of L.edodes logs,a production cost prediction model based on temporal convolutional neural network(TCN),biodirectional long short-term memory(BiLSTM)and Attention mechanism were proposed in this study.First,the important features related to the production cost of L.edodes logs were screened out by LightGBM to reduce the input dimension of the prediction model;then the TCN network and BiLSTM network were constructed to extract features from the input data,and the extracted features were fused;finally,based on the above,the Attention mechanism was added,and the full connection layer was used to obtain the final prediction result of the production cost of L.edodes logs.The prediction root mean squared error,mean absolute percentage error and mean absolute error of the model were 0.0841,2.2526,and 0.0738,respectively.The production cost prediction curve of L.edodes logs was close to the real curve.The experimental results showed that the model had a good prediction effect and could effectively meet the cost prediction requirements of L.edodes logs production enterprises.

关 键 词：香菇菌棒 双向长短期记忆网络 时间卷积神经网络 注意力机制 成本预测 深度学习 

分 类 号：S126[农业科学—农业基础科学] S646.1