Research on PM_(2.5) Concentration Prediction Algorithm Based on Temporal and Spatial Features

PM2.5 has a non-negligible impact on visibility and air quality as an important component of haze and can affect cloud formation and rainfall and thus change the climate,and it is an evaluation indicator of air pollution level.Achieving PM2.5 concentration prediction based on relevant historical data mining can effectively improve air pollution forecasting ability and guide air pollution prevention and control.The past methods neglected the impact caused by PM2.5 flow between cities when analyzing the impact of inter-city PM2.5 concentrations,making it difficult to further improve the prediction accuracy.However,factors including geographical information such as altitude and distance and meteorological information such as wind speed and wind direction affect the flow of PM2.5 between cities,leading to the change of PM2.5 concentration in cities.So a PM2.5 directed flow graph is constructed in this paper.Geographic and meteorological data is introduced into the graph structure to simulate the spatial PM2.5 flow transmission relationship between cities.The introduction of meteorological factors like wind direction depicts the unequal flow relationship of PM2.5 between cities.Based on this,a PM2.5 concentration prediction method integrating spatial-temporal factors is proposed in this paper.A spatial feature extraction method based on weight aggregation graph attention network(WGAT)is proposed to extract the spatial correlation features of PM2.5 in the flow graph,and a multi-step PM2.5 prediction method based on attention gate control loop unit(AGRU)is proposed.The PM2.5 concentration prediction model WGAT-AGRU with fused spatiotemporal features is constructed by combining the two methods to achieve multi-step PM2.5 concentration prediction.Finally,accuracy and validity experiments are conducted on the KnowAir dataset,and the results show that the WGAT-AGRU model proposed in the paper has good performance in terms of prediction accuracy and validates the effectiveness of the model.

A Hybrid Deep Learning Approach for PM2.5 Concentration Prediction in Smart Environmental Monitoring

Nowadays,air pollution is a big environmental problem in develop-ing countries.In this problem,particulate matter 2.5(PM2.5)in the air is an air pollutant.When its concentration in the air is high in developing countries like Vietnam,it will harm everyone’s health.Accurate prediction of PM2.5 concentrations can help to make the correct decision in protecting the health of the citizen.This study develops a hybrid deep learning approach named PM25-CBL model for PM2.5 concentration prediction in Ho Chi Minh City,Vietnam.Firstly,this study analyzes the effects of variables on PM2.5 concentrations in Air Quality HCMC dataset.Only variables that affect the results will be selected for PM2.5 concentration prediction.Secondly,an efficient PM25-CBL model that integrates a convolutional neural network(CNN)andBidirectionalLongShort-TermMemory(Bi-LSTM)isdeveloped.This model consists of three following modules:CNN,Bi-LSTM,and Fully connected modules.Finally,this study conducts the experiment to compare the performance of our approach and several state-of-the-art deep learning models for time series prediction such as LSTM,Bi-LSTM,the combination of CNN and LSTM(CNN-LSTM),and ARIMA.The empirical results confirm that PM25-CBL model outperforms other methods for Air Quality HCMC dataset in terms of several metrics including Mean Squared Error(MSE),Root Mean Squared Error(RMSE),Mean Absolute Error(MAE),and Mean Absolute Percentage Error(MAPE).

改进灰狼算法优化GBDT在PM_(2.5)预测中的应用

针对灰狼算法易陷入局部最优解和全局搜索能力不足的问题,通过霍尔顿序列(Halton Sequence)搜索算法初始化狼群位置,避免灰狼算法陷入局部最优解和重复运算;引入莱维飞行和随机游动策略对灰狼算法的寻优过程进行优化,以增加算法的全局搜索能力;利用粒子群算法模拟灰狼种群得出的最佳适应度以用于惩罚项改进灰狼算法中的头狼更新策略。使用改进算法优化的梯度提升树(Gradient Boosting Decision Trees,GBDT)模型对北京市大气污染物监测数据中PM_(2.5)质量浓度进行预测,采用3种评估函数对各模型以及混合模型预测效果得分进行评估。结果显示,本文改进的灰狼算法对梯度提升树的优化效果优于其他算法,均方根误差E RMS为6.65μg/m^(3),平均绝对值误差E MA为3.20μg/m^(3),拟合优度(R^(2))为99%,比传统灰狼算法优化结果的均方根误差减少了19.19μg/m^(3),平均绝对值误差降低了10.03μg/m^(3),拟合优度增加了9百分点;与霍尔顿序列和莱维飞行改进的(Levy Flight-Halton Sequence,LHGWO)相比,改进的灰狼算法预测得分的均方根误差降低了10.39μg/m^(3),平均绝对值误差减小了6.71μg/m^(3),拟合优度提高了5百分点。研究表明了预测模型优化的有效性,为未来城市改善空气质量提供了科学依据和技术支持。

基于改进机器学习的PM_(2.5)浓度预测模型研究

针对现有机器学习模型预测PM_(2.5)浓度存在模型过于复杂、没有考虑时空信息和缺失值填补不准确而导致模型性能下降的问题,利用随机森林取代统计学方法填补缺失值,并纳入时空因素提升模型精度.建立了综合遥感数据、气象及协同污染物数据,适用于沿海城市的PM_(2.5)浓度预测模型(K-means-RF-XGBoost模型),模型预测耗时仅为BP神经网络的4%.利用2019年大连市实时监测数据对模型PM_(2.5)浓度预测进行训练和测试,结果表明,建立的K-means-RF-XGBoost模型预测PM_(2.5)浓度有很高的准确性,与没有考虑时空信息的同种模型相比均方根误差(erms)降低了约48%,决定系数(R^(2))提升了约10%;能有效地预测高PM_(2.5)浓度并适用于波动范围大的情况,如春季模型在测试集中R^(2)可达0.935;同时在日级预测上表现优异,R^(2)可达0.819.该研究为沿海城市PM_(2.5)浓度预测提供了新思路.

基于注意力机制的CNN-ILSTM地铁站PM_(2.5)预测建模

为提高PM_(2.5)的预测精度,提出一种基于卷积神经网络(CNN)、改进长短期记忆网络(ILSTM)和注意力机制(attention)组合的预测模型。ILSTM删除LSTM中的输出门,改进其输入门和遗忘门,并引入转换信息模块(CIM),以防止学习过程中的过饱和。该模型将一维卷积神经网络的特征提取和改进长短期记忆网络学习序列依赖性的能力相结合,得到过去不同时间特征状态对未来PM_(2.5)浓度的影响,可以有效模拟PM_(2.5)在时间和空间上的依赖性,并通过注意力机制自动权衡过去的特征状态,进一步提升PM_(2.5)预测的准确度。实验结果表明:CNN-ILSTM-attention模型的拟合度达到98.5%,与LSTM模型、CNN-LSTM模型和CNN-ILSTM模型相比,分别提高26%、9.2%和6.2%,具有较高的预测精度和应用价值。

Multi-Scale Variation Prediction of PM2.5 Concentration Based on a Monte Carlo Method

Haze concentration prediction,especially PM2.5,has always been a significant focus of air quality research,which is necessary to start a deep study.Aimed at predicting the monthly average concentration of PM2.5 in Beijing,a novel method based on Monte Carlo model is conducted.In order to fully exploit the value of PM2.5 data,we take logarithmic processing of the original PM2.5 data and propose two different scales of the daily concentration and the daily chain development speed of PM2.5 respectively.The results show that these data are both approximately normal distribution.On the basis of the results,a Monte Carlo method can be applied to establish a probability model of normal distribution based on two different variables and random sampling numbers can also be generated by computer.Through a large number of simulation experiments,the average monthly concentration of PM2.5 in Beijing and the general trend of PM2.5 can be obtained.By comparing the errors between the real data and the predicted data,the Monte Carlo method is reliable in predicting the PM2.5 monthly mean concentration in the area.This study also provides a feasible method that may be applied in other studies to predict other pollutants with large scale time series data.

基于生成对抗网络模型的小样本PM_(2.5)预测

针对目前数据驱动的方法在小样本下PM_(2.5)预测准确率较低的问题,提出一种基于生成对抗性网络(GAN)模型PME-GAN,用于在线预测PM_(2.5)浓度值。在生成器中加入长短期记忆网络(LSTM)并用于提取输入数据的时序特征,在判别器中加入多层感知机网络(MLP),通过生成器对PM_(2.5)浓度值进行预测。与LSTM、GRU、CNN-LSTM和CNN-GRU 4种模型进行对比实验,结果表明,该方法在小样本数据集上具有更高的预测准确率,对保定测试集的后25%数据开始预测,预测效果很好。

基于BP神经网络的PM_(2.5)浓度值预测模型

PM_(2.5)对空气质量的恶劣影响和对生命健康的严重威胁日益引起了各界的关注。揭示PM_(2.5)时间分布规律,对其浓度进行有效预测,有助于大众及时采取防控措施和降低污染暴露强度。故以西安市为研究区,基于反向传播神经网络(BP神经网络),应用2014年1月1日至2017年11月4日的1400组大气污染物监测数据进行训练学习,并用2017年11月4日至2018年8月31日的300组数据进行测试和检验,最终建立了精度较高的PM_(2.5)浓度预测模型,用以预测次日PM_(2.5)浓度值,并针对偏差较大的预测结果,进行了成因分析和讨论。

联合Transformer注意力机制的PM_(2.5)浓度预测网络研究

应用深度学习技术进行PM_(2.5)浓度预测方法的研究,提出联合Transformer注意力机制的循环预测网络。模型的核心是多头注意力-长短期记忆网络(MH-LSTM),通过构建统一记忆单元捕捉时空特征关联。MH-LSTM单元使用LSTM联合Transformer注意力机制对时间变化和全局空间特征统一建模形成记忆信息。记忆信息“之”字流向跨越堆叠的MH-LSTM模块,高层记忆信息辅助下一时刻低层记忆信息的获取。应用该模型结合河北省生态环境监测中心提供的PM_(2.5)浓度数据开展预报试验,结果表明,相对于卷积LSTM网络(ConvLSTM)、预测循环神经网络(PredRNN)、双重记忆网络(MIM),该模型预测的平均绝对误差分别减小了18.13%、10.23%、9.62%,实现了同时捕捉PM_(2.5)浓度的时空相关性,具有更优预测性能。

Analysis of PM2.5 concentrations in Heilongjiang Province associated with forest cover and other factors

Atmospheric particulate matter(PM2.5) seriously influences air quality. It is considered one of the main environmental triggers for lung and heart diseases. Air pollutants can be adsorbed by forest. In this study we investigated the effect of forest cover on urban PM2.5 concentrations in 12 cities in Heilongjiang Province,China. The forest cover in each city was constant throughout the study period. The average daily concentration of PM2.5 in 12 cities was below 75 lg/m^3 during the non-heating period but exceeded this level during heating period. Furthermore, there were more moderate pollution days in six cities. This indicated that forests had the ability to reduce the concentration of PM2.5 but the main cause of air pollution was excessive human interference and artificial heating in winter. We classified the 12 cities according to the average PM2.5 concentrations. The relationship between PM2.5 concentrations and forest cover was obtained by integrating forest cover, land area,heated areas and number of vehicles in cities. Finally,considering the complexity of PM2.5 formation and based on the theory of random forestry, we selected six cities and analyzed their meteorological and air pollutant data. The main factors affecting PM2.5 concentrations were PM10,NO_2, CO and SO_2 in air pollutants while meteorological factors were secondary.

A New Model Using Multiple Feature Clustering and Neural Networks for Forecasting Hourly PM2.5 Concentrations,and Its Applications in China

Particulate matter with an aerodynamic diameter no greater than 2.5 lm(PM2.5)concentration forecasting is desirable for air pollution early warning.This study proposes an improved hybrid model,named multi-feature clustering decomposition(MCD)–echo state network(ESN)–particle swarm optimization(PSO),for multi-step PM2.5 concentration forecasting.The proposed model includes decomposition and optimized forecasting components.In the decomposition component,an MCD method consisting of rough sets attribute reduction(RSAR),k-means clustering(KC),and the empirical wavelet transform(EWT)is proposed for feature selection and data classification.Within the MCD,the RSAR algorithm is adopted to select significant air pollutant variables,which are then clustered by the KC algorithm.The clustered results of the PM2.5 concentration series are decomposed into several sublayers by the EWT algorithm.In the optimized forecasting component,an ESN-based predictor is built for each decomposed sublayer to complete the multi-step forecasting computation.The PSO algorithm is utilized to optimize the initial parameters of the ESN-based predictor.Real PM2.5 concentration data from four cities located in different zones in China are utilized to verify the effectiveness of the proposed model.The experimental results indicate that the proposed forecasting model is suitable for the multi-step high-precision forecasting of PM2.5 concentrations and has better performance than the benchmark models.

VARIATION CHARACTERISTICS OF THE PLANETARY BOUNDARY LAYER HEIGHT AND ITS RELATIONSHIP WITH PM2.5 CONCENTRATION OVER CHINA

The planetary boundary layer height(PBLH) was calculated using the radiosonde sounding data, including120 L-band operational sites and 8 GPS sites in China. The diurnal and seasonal variations of PBLH were analyzed using radiosonde sounding(OBS-PBLH) and ERA data(ERA-PBLH). Based on comparison and error analyses, we discussed the main error sources in these data. The frequency distributions of PBLH variations under different regimes(the convective boundary layer, the neutral residual layer, and the stable boundary layer) can be well fitted by a Gamma distribution and the shape parameter k and scale parameter s values were obtained for different regions of China. The variation characteristics of PBLH were found in summer under these three regimes for different regions. The relationships between PBLH and PM_(2.5) concentration generally follow a power law under very low or no precipitation conditions in the region of Beijing, Tianjin and Hebei in summer. The results usually deviated from this power distribution only under strong precipitation or high relative humidity conditions because of the effects of hygroscopic growth of aerosols or wet deposition. The OBS-PBLH provided a reasonable spatial distribution relative to ERA-PBLH.This indicates that OBS-PBLH has the potential for identifying the variation of PM_(2.5) concentration.

基于深度学习的沈阳市春节期间PM_(2.5)浓度预测研究

本文利用2016年—2022年沈阳市春节期间逐时空气质量监测数据和气象因子观测资料,基于贝叶斯参数优化,建立了GRU、LSTM深度学习模型以及LGBM、XGBOOST、RF、GBDT树集成学习模型,并且通过4种错误度量标准,将模型进行对比进而得出预测PM_(2.5)浓度的最优模型。结果表明:GRU模型的PM_(2.5)浓度预测准确度最高、训练速度最快、模型最简单,其MSE为32.160,R2为0.973,其次为LSTM模型,GBDT模型的预测效果最差。同时整体来看,深度学习模型要优于常见的树集成学习模型。

Analysis on Thermal Environment of Underlying Surface and PM2.5 Concentration in Community Park of Beijing in Winter

Community park is one of the most important landscape spaces for urban people to live outdoors,and people’s perception of environmental microclimate is a direct factor affecting the use frequency and experience of community parks.In this paper,Shijingshan Sculpture Park of Beijing was taken as experimental object.Using the method of fi eld measurement,9-d winter test for 3 months was conducted in three kinds of landscape architecture spaces,including waterfront plaza,open green space and square under the forest.Via regression analysis method,the measured air temperature(Ta),relative humidity of air(RH),particulate matter(PM2.5)were analyzed.It is found that winter sunshine is main infl uence factor of garden microclimate,and there is a negative correlation between local temperature and humidity;local temperature and humidity can regulate the local PM2.5 concentration,and temperature shows negative correlation with PM2.5 concentration,while humidity shows positive correlation with PM2.5 concentration.Meanwhile,via comparative analysis of temperature,humidity and PM2.5 concentration in different types of garden spaces,the infl uence of different space forms,planting forms and materials on thermal environment of underlying surface and PM2.5 concentration was summarized,and design strategy was optimized,to be as benefi cial reference of reconstruction design of community parks.

基于RFE-RF模型的太原市PM_(2.5)浓度预测研究

为了更好的利用高空气象要素采用机器学习的方法对太原市PM_(2.5)的浓度进行预测。采用太原市2015~2018年环境空气质量数据和NCEP再分析数据,将空气污染物PM_(2.5)浓度作为标签、根据RFE特征选择的结果,将最利于提升模型表现的预报因子作为输入,选用随机森林(RF)回归模型进行预测,同时构建3个对比模型以进一步验证RF模型的预测准确率。结果表明:RF模型的MAE、MAPE、RMSE分别为17.19、38.17%和26.0,与Lasso模型相比,分别降低了7.7%、5.1%和2.7%;相比于SVM预测模型的MAE、MAPE、RMSE分别降低了23.1%、15.3%和29.9%;相比于KNN预测模型,RF模型的MAE、MAPE、RMSE分别降低了17.2%、19.8%和15.2%。RF模型具有良好的预测效果,R^(2)达0.71,4种模型预测值与实测值的相关系数依次为0.76、0.78、0.82和0.84,RF模型的预报效果均好于Lasso模型、KNN模型和SVM模型。通过选取最优的RF预测模型应用到日常的环境空气质量预报业务中,将进一步提高太原市PM_(2.5)浓度预报的准确率,同时也为加强太原市的空气污染防治,实现环境综合管理和决策科学化提供了的重要科技手段。

Manufactured nanoparticle:A prediction model for understanding PM2.5 toxicity to human

Epidemiological studies have demonstrated that chronic exposure to polluted concentration of fine ambient particulate matter(PM2.5)can induce markedly harmful effects on human health,however,an enormous research effort is still need to the comprehensive understanding of PM2.5 induction of new negative health outcomes.Recently,Maher and colleges[1]from Environmental Magnetism and Paleomagnetism at Lancaster University

北京区域环境气象数值预报系统及PM_(2.5)预报检验

基于北京地区快速更新循环同化预报系统(BJ-RUC)、WRF-Chem模式和优选的能见度参数化方案,建立了北京区域环境气象数值预报系统。对2014年全年PM_(2.5)浓度、能见度和APEC(Asia-Pacific Economic Cooperation)期间预报效果检验结果表明:该系统对京津冀及周边地区PM_(2.5)浓度的预报效果较好,大部分站点的相关系数在0.6以上,特别足北京的部分站点可达0.8以上,预报结果相比观测总体偏低,随着预报时效的延长,24 h之后预报效果略有下降。相比人工观测,能见度预报结果与自动观测能见度更加接近,对持续性低能见度过程预报与实况吻合较好,对于小时能见度低于10 km的分级检验显示,预报准确率从77%左右逐级下降,2 km以下在40%左右。2014年APEC期间,系统很好地预报出北京地区空气质量指数、PM_(2.5)浓度和能见度的时空演变特征,为APEC期间环境气象预报服务提供了有力的技术支撑。

气象参数对基于BP神经网络的PM_(2.5)日均值预报模型的影响

建立了基于BP神经网络的PM_(2.5)质量浓度预报模型,对广州市5个监测点2012年6月—2013年5月的PM_(2.5)质量浓度日均值进行预报,分析了总体预报误差、不同风速和降雨量下的预报误差,以及天气预报误差对PM_(2.5)质量浓度预报误差的影响。结果表明,BP神经网络模型对5个站点的PM_(2.5)预报结果稳定,平均相对误差为29.71%。在有利于PM_(2.5)扩散的气象条件下预报误差较大,风速较大时与风速较小时预报误差的差异高达15%,而不同降雨量情况下的预报误差较相近。修正天气预报后,各站点的预报误差平均降低了4.67%。这表明可从空气质量数据质量等方面入手改进模型。

新型现代有轨电车内PM 2.5浓度实时监测系统设计

设计了一种新型的基于GPRS的现代有轨电车内PM 2.5浓度实时监测系统。系统包括车载硬件终端和中心平台,车载硬件终端实现对电车内PM 2.5浓度数据采集;中心平台以TCP协议的Socket通信为基础,采用Visual Basic软件设计,能实时显示和记录车内PM 2.5浓度的动态曲线以及历史数据。车载硬件终端与中心平台间采用GPRS网络模块SIM900A通信。该系统通过与车内空气净化器组合使用,可提高车内空气质量。

改进的快速跟踪回声状态网络及PM2.5预测

针对递归最小二乘回声状态网络在噪声环境中预测精度不高的问题,提出了一种改进的快速跟踪回声状态网络。首先在递归最小二乘回声状态网络结构的基础上,将自适应调节的可变遗忘因子加入其代价函数中,用改进的递归最小二乘法对网络输出权值进行训练,得到快速跟踪回声状态网络;然后利用经典Lorenz混沌系统验证快速跟踪回声状态网络的有效性;最后利用灰关联法分析各相关变量与PM 2.5的关联度,建立PM 2.5浓度值辅助变量集合,将辅助变量集合输入到快速跟踪回声状态网络进行PM 2.5浓度值预测。实验表明,与传统回声状态网络、递归最小二乘回声状态网络预测效果相比,快速跟踪回声状态网络的预测方法精度佳,抗噪声能力强。