摘要
由于PM2.5浓度预测中的影响因素过于复杂,影响因素的高维性与非线性对预测结果有着很大的干扰,容易产生PM2.5浓度预测误差高和模型泛化能力差等问题。针对上述缺陷,可通过一种基于随机森林-粒子群优化-极限学习机(RF-PSO-ELM)的PM2.5浓度预测模型解决。该模型首先使用随机森林算法对影响因素进行特征选择,选择出对于PM2.5浓度重要性高的因素构成特征;再利用提取得到的特征作为PSO-ELM算法的输入;最后对上海市的PM2.5浓度做出预测,从最终的实验数据中可以看出:该模型比支持向量机(SVM)、未优化的极限学习机(ELM)和反向神经网络(BPNN)等预测模型在预测精度和泛化能力方面有着显著的提高。
Because the factors affecting the PM2.5 concentration value are too complicated,the high dimensionality and nonlinearity of the influencing factors have a great interference with the final result of the prediction.The performance of the prediction model is prone to problems such as high prediction error of PM2.5 concentration value and poor generalization ability.In view of the above mentioned defects,which can be solved by proposing a PM2.5 concentration prediction model based on random forest-particle swarm optimization-extreme learning machine(RF-PSO-ELM).The model first uses the random forest algorithm to select the influencing factors,selects the characteristics that are important to the PM2.5 concentration,and uses the extracted features as the input of the PSO-ELM algorithm to predict the PM2.5 concentration.The prediction of PM2.5 concentration in Shanghai shows significant improvements that the model is more accurate than the traditional support vector machine(SVM)and unoptimized extreme learning machine(ELM)and back propagation neural network(BPNN)and so on in terms of prediction accuracy and generalization capabilities.
作者
王鑫圆
曹春萍
WANG Xin-yuan;CAO Chun-ping(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《软件》
2020年第6期12-18,62,共8页
Software
基金
国家自然科学基金资助项目(61703278)
国家自然科学基金青年项目(61803264)。
