运用卷积神经网络预测中国太阳能路灯市场需求  

Prediction of Market Demand for Solar Street Lights in China Based on CNN

作　　者：舒服华[1] SHU Fuhua(School of Continuing Education of Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学继续教育学院,湖北武汉430070

出　　处：《中国照明电器》2024年第7期99-103,共5页China Light & Lighting

基　　金：湖北省自然科学基金项目“大数据与机器学习技术应用研究”(2020CFB177)。

摘　　要：由于我国太阳能路灯市场需求呈非线性、不光滑的复杂分布,传统人工神经网络预测效果难如人意。卷积神经网络(CNN)通过使用过滤器,让相邻层神经元采用部分连接,同一特征平面的神经元参数相同,使所有神经元实现参数共享,性能优良,预测精度高。运用卷积神经网络对我国太阳能路灯市场需求进行了预测。结果显示,平均预测误差仅为0.958 8%,比支持向量机的1.170 2%减小了18.065 3%,比随机森林的1.082 2%减小了11.402 7%。利用模型预测了2023—2027年我国太阳能路灯市场需求,通过分析,表明这一预测结果有一定的可信度。Since the data of demand in the solar street lamp market in China,the prediction performance of traditional artificial neural networks is unsatisfactory.The convolutional neural network(CNN)uses filters to make the neurons of adjacent layers partially connected,and the neuron parameters of the same feature plane are the same,so that all neurons can share parameters,with excellent performance and high prediction accuracy.The convolution neural network was used to predict the market demand for solar street lights in China.The results show that the average prediction error is only 0.9588%,18.0653%less than 1.1702%of SVM,and 11.4027%less than 1.0822%of random forest.The model was used to predict the market demand for solar street lights in China from 2023 to 2027,and analysis shows that this prediction result has high reliability.

关 键 词：太阳能 路灯 需求量 预测 卷积神经网络 

分 类 号：TM-9[电气工程]