基于压缩感知和深度学习的农产品价格预测  

Agricultural product price prediction based on compressed sensing and deep learning

作　　者：周志轩 陈仲民 邓君丽 ZHOU Zhixuan;CHEN Zhongmin;DENG Junli(College of Informatics,Huazhong Agricultural University,Wuhan 430070,China)

机构地区：[1]华中农业大学信息学院,湖北武汉430070

出　　处：《武汉大学学报（工学版）》2024年第9期1327-1334,共8页Engineering Journal of Wuhan University

基　　金：中央高校基本科研业务费专项资金资助项目(编号:2662019YJ003)。

摘　　要：采用一种结合压缩感知(compressive sensing,CS)和深度学习的组合模型,利用CS算法的正交匹配追踪(orthogonal matching pursuit,OMP)算法和深度学习模型的长短期记忆(long short-term memory,LSTM)模型,基于近年来可能存在噪声的农产品价格原始数据,通过稀疏表示、设计观测矩阵、信号重构等对原始数据降噪,进而结合降噪后的重构数据与LSTM模型对农产品价格趋势进行预测。该组合模型具有对数据存储要求低、对微小噪声不敏感等优势,与支持向量回归(support vector regression,SVR)传统模型相比,预测数据的精确度高约13%,与其他模型相比也具有较高精确度,在1 a的时间跨度内能够取得较传统模型更准确的预测效果。In this paper,a combined model that combines compressed sensing(CS)and deep learning is adopted.The orthogonal matching pursuit(OMP)algorithm is selected from the compressed sensing algorithm,and the long short-term memory(LSTM)model is selected from the deep learning model.Based on the original agricultural product price data that may have noise in recent years,the original data are denoised by sparse representation,design of observation matrix,signal reconstruction and other steps,and then the reconstructed data after noise reduction and LSTM model are combined to predict the price trend of agricultural products.The combined model has the advantages of low requirements for data storage and low sensitivity to small noise.Compared with the traditional support vector regression(SVR)model,the accuracy of the prediction data is about 13%higher.Compared with other models,the combined model also has higher accuracy,and can achieve more accurate prediction results than the traditional models within a one-year time span.

关 键 词：压缩感知 正交匹配追踪 深度学习 长短期记忆 组合模型 价格预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]