DSCE-GEP算法在PM2.5浓度预测中的应用

Study on the Prediction of PM2.5 Concentration by Double System Co-evolutionary Gene Expression Programming

摘要:

雾霾防治是目前空气质量保护问题研究的热点，PM2.5浓度预测是雾霾防治的关键之一。文章采用一种双系统协同进化的基因表达式编程算法（DSCE-GEP）进行PM2.5浓度预测，该算法在GEP算法中引入人工干预操作来提高算法进化速度以及解的质量。DSCE-GEP算法是对人类进化的模拟，不仅具有强大的模型学习能力，而且能得到模型的显式函数表达式。文中以西安地区逐日PM2.5浓度预测为例，将DSCE-GEP算法与传统基因表达式编程算法（GEP）、文献中分类回归树和极限学习机组合模型（CART-EELM）以及卷积神经网络和长短期记忆神经网络组合模型（CNN-LSTM）进行了对比实验。实验结果表明，DSCE-GEP算法拟合度更高，是一种具有竞争力的智能预测算法。

Abstract:

Haze prevention and control is a hot topic of air quality protection research at present, and PM2.5 concentration prediction is one of the keys to haze prevention and control. In this paper, a dual system co-evolution gene expression programming algorithm (DSCE-GEP) was used to predict PM2.5 concentration. In this algorithm, manual intervention was introduced into the GEP algorithm to improve the algorithm evolution speed and the quality of the solution. DSCE-GEP algorithm is a simulation of human evolution. It not only has strong ability of model learning, but also can get explicit function expression of the model. In this paper, the daily PM2.5 concentration prediction in Xi 'an area is taken as an example, and the DSCE-GEP algorithm is compared with the traditional gene expression programming algorithm (GEP), the classification regression tree and extreme learning machine combination model (CART-EELM) in literature, and the convolutional neural network and long and short-term memory neural network combination model (CNN-LSTM). The experimental results show that the DSCE-GEP algorithm has higher fitting degree and is a competitive intelligent prediction algorithm.