基于Wavelet-LSTM模型的北京空气污染物浓度预测

文章为了达到精准预测北京市空气污染物浓度目的,应用小波分解变换(wavelet transform)和长短期神经记忆网络(long short-term memory,LSTM)相结合的方法,建立Wavelet-LSTM空气污染物浓度预测模型,对北京市6项空气污染物浓度预测。研究首先通过小波分解变换将日空气污染物浓度的历史时间序列分解为不同频率并重新组合为高维训练数据集合;其次使用高维数据集训练LSTM预测模型,重复试验调整参数,获得最优预测模型。研究结果表明,组合模型对于污染物浓度预测比传统LSTM模型的预测精度和稳定性更高。     

小波支持向量机在大气污染物浓度预测中的应用

用小波分解重构和支持向量机相结合的方法,建立大气污染物浓度预测模型。通过小波分解,将大气污染物浓度序列分解为不同频段的小波系数序列,再对各层的小波系数序列重构到原尺度上。利用相关分析的方法构建出低频小波系数a3和中频小波系数d3的支持向量机模型输入因子为前一天小波系数a3和7个气象因子;高频小波系数d2和d1以前三日的小波系数为输入因子,然后对各小波系数序列采用相应的支持向量机模型进行预测,各小波系数均使用ν-支持向量回归机(ν-SVR)算法和径向基函数,最后通过小波重构合成大气污染物浓度序列的最终预测结果。通过对大气SO2浓度预测实例证明,该大气污染物浓度预测模型具有推广能力较强、预测精度较高、训练速度快、便于建模等优点,具有良好的应用前景。     

基于小波分解与自适应多级残差修正的最小二乘支持向量回归预测模型的PM2.5浓度预测

采用小波分解(WD)将济南市科干所监测站PM_(2.5)浓度的一维时间序列(2013年1月1日—2017年8月15日)分解为高维信息,获得了该监测站附近PM_(2.5)浓度的时频变化特征,重点分析了PM_(2.5)的随机性和趋势性问题.然后构建了基于小波分解的多级残差修正的最小二乘支持向量回归预测模型(AMLRC-WLSSVR),结果发现,该模型能够很好地对济南市PM_(2.5)浓度做出预测,特别是针对重污染天气的预测有很好的精度.为了避免预测结果的不确定性问题,提出了一种基于方差估计给出预测值置信区间上界的方法,同时,有效弥补了单点预测的不稳定性及预测精度不足的缺点,该方法能够为实际空气污染预警提供技术支持.     

小波网络在大气污染物浓度预测中的应用

提出一种“分解-重构-预测”小波网络的大气污染物浓度的预测方法。通过小波分解,把浓度序列分解为不同频段的小波系数序列,再对各层的小波系数子序列重构到原尺度上,然后对小波系数序列采用相匹配的BP神经网络模型进行预测,最后合成浓度序列的最终预测结果。经对二氧化硫浓度预测证明,该方法预测模型推广能力强,预测精度高。     

XGBoost-LSTM变权组合模型支持下短期PM2.5浓度预测——以上海为例

为进一步提高PM_2.5浓度预测的精度,基于XGBoost和LSTM进行改进得到变权组合模型XGBoost-LSTM（Variable）.过对预测因子进行相关性分析,得到其它大气污染物和气象因素对PM_2.5浓度的影响,确定最优PM_2.5浓度预测因子,再将预处理后数据集输入LSTM模型和XGBoost模型分别进行预测,采用基于残差改进的自适应变权组合方法得到最终预测结果.结果表明,污染物变量的相对重要性高于气象因子变量,其中当前PM_2.5和CO浓度的相对重要性较高,而平均风速和相对湿度重要性较低.XGBoost-LSTM（Variable）模型的RMSE、MAE和MAPE值为1.75、1.12和6.06,优于LSTM、XGBoost、SVR、XGBoost-LSTM（Equal）和XGBoost-LSTM（Residual）模型.分季节预测结果表明,XGBoost-LSTM（Variable）模型在春季预测精度最好,而夏季预测精度较差.模型预测精度高的原因在于其不仅考虑了数据的时间序列特征,又兼顾了数据的非线性特征.     

基于聚类与多元回归的空气质量预报模型开发

为了方便准确地预测空气污染物浓度,基于气象因子聚类与多元回归的方法,以广东省顺德区为例开发了空气质量统计预报模型。预报模型能够较好地模拟出顺德区NO2、SO2、CO、PM10、PM2.5日均浓度和O3日最大8 h浓度水平和变化趋势,模型的模拟结果与实测值具有较高的相关性(相关系数R约为0.76),标准化平均偏差为1.2%~13.4%,标准化平均误差为14.2%~30.3%,模型普遍略为高估各项污染物浓度水平。预报模型具有简单易行、节约人力物力、准确可靠等优点,适用于地级市及区县空气污染物的预报。     

基于ARIMA模型的PM_(2.5)预测

PM2.5的精确预测是大气污染评价和治理的关键性工作。本文针对PM2.5浓度变化的时间序列分布特征,结合环境监测站提供的相关数据,应用自回归移动平均模型(ARIMA(p,d,q))预测短期PM2.5的日平均浓度。结果表明:由于PM2.5浓度变化受气象场、排放源、复杂下垫面、理化生过程的耦合等多种因素的影响,不同时段内的变化模式存在巨大差异,因此采用分时段序列预测模型可以提高PM2.5的预测精度;通过将分时段序列模型与灰色GM(1,1)模型和全年时间序列模型的预测结果进行对比,发现该模型预测效果更好。     

小波分析在PM10浓度时间序列分析中的应用

以西安市PM10日平均浓度时间序列为例,根据小波分析的基本原理,应用小波分解和重构对PM10浓度时间序列的变化进行了分析,得到PM10的年变化趋势和突变特征。研究结果表明,用小波分析应用于大气污染物浓度时间序列的分析是可行的。     

一种基于小波变换和LSTM的大气污染物浓度预测模型

针对现有大气污染物浓度预测模型存在预测精度不高、污染物种类单一等不足的问题,通过小波分解将高维大气污染物数据转换为低维数据,再对分解序列建立长短期记忆网络（LSTM）预测模型,最后通过小波重构将分解序列重构为污染物时间序列,建立了1种基于小波变换（WT）的LSTM大气污染物预测模型（WT-LSTM）,用以预测目标区域内的次日平均ρ（PM_2.5）、ρ（PM₁₀）、ρ（SO₂）、ρ（NO₂）和ρ（O₃）。采用长沙市2015—2018年10处国控站点的数据进行验证,结果表明:相对于LSTM、多元线性回归（MLR）和基于WT的WT-MLR模型,WT-LSTM的均方根误差和绝对平均误差均下降了50%,其对PM_2.5、PM₁₀、SO₂、NO₂和O₃的污染等级预测准确率均在80%以上。     

基于小波分解和SVM的大气污染物浓度预测模型研究

针对大气污染物浓度的精准预测问题,运用小波分解将污染物浓度一维序列分解为高维信息,结合气象及污染物浓度数据,构建了基于小波分解的支持向量机预测模型.最后将模型应用于长沙市2018年PM_2.5和O_{3-8 h}的浓度预测.结果表明：①在其他参数不变的条件下,该模型在平均绝对误差（MAE）、平均绝对百分比误差（MAPE）、均方根误差（RMSE）、一致性水平（IA）和相关系数（R）指标上均优于未经小波分解的预测模型;②在考虑其他污染物对PM_2.5浓度的影响后,预测模型评价指标MAE、MAPE和RMSE分别减少了5.57%、9.91%和3.44%,有着更小的误差;③在考虑气象因素对O_{3-8 h}浓度的影响后,预测模型评价指标MAE、MAPE和RMSE分别减少了1.59%、3.54%和0.82%,同样也有更小的误差.由此可以看出,本文所提模型能够有效预测大气污染物浓度,为相关研究提供了方法参考.     

Distribution and ecological effect of mercury in Laogang landfill, Shanghai, China

Laogang landfill near Shanghai is the largest landfill in China, and receives about 10000 t of daily garbage per day, Samples of topsoil and plants were analyzed to evaluate mercury pollution from the landfill. For topsoil samples, there were significant correlations among total mercury （HgT）, combinative mercury （Hgc） and gaseous mercury （HgG）, and content of total organic carbon （TOC）, but, no significantly relationship was found between Hg content and filling time. Hg content changes in vertical profiles with time showed that the average Hgv of profiles 1992, 1996, and 2000 was similar, but their average HgG was quite different. HgT was significantly correlated with Hgc in profile 1992 and 2000, and Hgv was significantly correlated with Hg6 in profile 1996. HgG/Hgv ratio in profile samples decreased in the order of （HgG,/HgT）1992〉（HgG/HgT）1996〉〉（HgG/HgT）2000. A simple outline of Hg release in landfill could be drawn： with increasing of filling time, degradation undergoes different biodegradation, accordingly, gaseous mercury goes through small, more, and small proportion to total mercury. Distribution of Hg in plants was inhomogeneous, following the order of leaf〉root〉stem. The highest value of leaf may be associated with higher atmospheric Hg from landfill. Ligneous plants （e.g. Phyllostachys glanca, Prunus salicina and Ligustrum lucidum） are capable of enriching more Hg than herbaceous plants.     

EDTA-enhanced phytoremediation of lead contaminated soil by Bidens maximowicziana

Phytoremediation is a potential cleanup technology for the removal of heavy metals from contaminated soils.Bidens maximowicziana is a new Pb hyperaccumulator,which not only has remarkable tolerance to Pb but also extraordinary accumulation capacity for Pb.The maximum Pb concentration was 1509.3 mg/kg in roots and 2164.7 mg/kg in overground tissues.The Pb distribution order in the B. maximowicziana was:leaf>stem>root.The effect of amendments on phytoremediation was also studied.The mobility of soil Pb and the Pb concentrations in plants were both increased by EDTA application.Compared with CK(control check),EDTA application promoted translocation of Pb to overground parts of the plant.The Pb concentrations in overground parts of plants was increased from 24.23-680.56 mg/kg to 29.07-1905.57 mg/kg.This research demonstrated that B.maximowicziana appeared to be suitable for phytoremediation of Pb contaminated soil,especially,combination with EDTA.     

Decomposition of alachlor by ozonation and its mechanism

Decomposition and corresponding mechanism of alachlor, an endocrine disruptor in water by ozonation were investigated. Results showed that alachlor could not be completely mineralized by ozone alone. Many intermediates and final products were formed during the process, including aromatic compounds, aliphatic carboxylic acids, and inorganic ions. In evoluting these products, some of them with weak polarity were qualitatively identified by GC-MS. The information of inorganic ions suggested that the dechlorination was the first and the fastest step in the ozonation of alachlor.     

Enzymatic oxidation of aqueous pentachlorophenol

The influence of the nonionic surfactant Tween 80 on pentachlorophenol （PCP） oxidation catalyzed by horseradish peroxidase was studied. The surfactant was tested at concentrations below and above its critical micelle concentration （CMC）. Enhancement of PCP removal was observed at sub-CMCs. The presence of Tween 80 in the reaction mixture reduced enzyme inactivation which occurred through a combination of free radical attack and sorption by precipitated products. A simple first-order model was able to simulate time profiles for enzyme inactivation in the presence or absence of Tween 80. At supra-CMCs, the surfactant caused noticeable reductions in PCP removal, presumably through micelle partitioning of PCP which precluded the hydrophobic PCP molecule from interacting with the enzyme.     

基于综合水质标识指数和对应分析法的校园河道水质评价

以三峡大学的校园河道求索溪为研究对象,利用综合水质标识指数法确定求索溪水质类别,分析其水质时空变化规律,并利用对应分析法得出求索溪中不同监测点的主要污染因子.研究结果表明:求索溪整体的综合水质标识指数为7.423,整体水质为劣V类(地表水环境质量标准GB 3838-2002)且黑臭.从时间变化来看,求索溪4月份的水质最差,5月份次之,4、5月份所有监测点的水质都劣于V类且黑臭;8月份水质最好,水质为Ⅳ类;从空间分布来看,8个监测点综合水质标识指数均超过6.0,水质为劣V类,其中6号监测点的水质相对最好,监测点3号的水质相对最差;对应分析法得出求索溪的整体水体污染程度受总氮因子的影响最大,其次为总磷.该研究拟为求索溪及类似校园河道的水环境治理研究提供基础依据和参考.     

Effects of chitosan on growth of an aquatic plant (Hydrilla verticillata) in polluted waters with different chemical oxygen demands

Effects of chitosan on a submersed plant, Hydrilla verticillata, were investigated. Results indicated that H. venicillata could prevent ultrastructure phytotoxicities and oxidativereaction from polluted water with high chemical oxygen demand (COD). Superoxide dismutase (SOD) activity and malondialdehyde (MDA) contents in H. verticillata treated with 0.1% chitosan in wastewater increased with high COD (980 mg/L) and decreased with low COD (63 mg/L), respectively. Ultrastructural analysis showed that the stroma and grana of chloroplast basically remained normal. However, plant cells from the control experiment (untreated with chitosan) were vacuolated and the cell interval increased. The relict of protoplast moved to the center, with cells tending to disjoint. Our findings indicate that wastewater with high COD concentration can cause a substantial damage to submersed plant, nevertheless, chitosan probably could alleviate the membrane lipid peroxidization and ultrastructure phytotoxicities, and protect plant cells from stress of high COD concentration polluted water.     

Removal of cadmium using MnO2 loaded D301 resin

MnO2 loaded weak basic anion exchange resin D301 （Anion exchange resin, macroreticular weak basic styrene） as adsorbent has been prepared and applied to the removal of cadmium. The adsorption characteristics have been investigated with respect to effect of pH, equilibrium isotherms, removal kinetic data, and interference of the coexisting ions. The results indicated that the Cd^2＋ could be efficiently removed using MnO2 loaded D301 resin in the pH range of 3-8 from aqueous solutions with the co-existence of high concentration of alkali and alkaline-earth metals ions. The saturate adsorption capacity of the Cd^2＋ was 77.88 mg/g. The adsorption process followed the pseudo first-order kinetics. The equilibrium data obtained in this study accorded excellently with the Langmuir adsorption isotherm.     

Effect of coagulation pretreatment on the fouling of ultrafiltration membrane

The purpose of this study is to understand the effect and mechanism of preventing membrane fouling, by coagulation pretreatment, in terms of fractional component and molecular weight of natural organic matter （NOM）. A relatively higher molecular weight （MW） of hydrophobic compounds was responsible for a rapid decline in the ultrafiltration flux. Coagulation could effectively remove the hydrophobic organics, resulting in the increase of flux. It was found that a lower MW of neutral hydrophilic compounds, which could remove inadequately by coagulation, was responsible for the slow declining flux. The fluxes in the filtration of coagulated water and supernatant water were compared and the results showed that a lower MW of neutral hydrophilic compounds remained in the supernatant water after coagulation could be rejected by a membrane, resulting in fouling. It was also found that the coagulated flocs could absorb neutral hydrophilic compounds effectively. Therefore, with the coagulated flocs formed on the membrane surface, the flux decline could be improved.     

Penetration analysis of elements and bioleaching treatment of spent refractory for recycling

Acidithiobacillus ferrooxidans ATCC23270 and Acidithiobacillus thiooxidans TM-32 were used for bioleaching of spent refractories of aluminium and copper melting furnaces for their recycling.Firstly,penetration of elements into aluminium melting furnace refractory was investigated and it was found that up to 7 cm from surface was contaminated.Comparison on leaching efficiency by the strains ATCC23270 and TM-32 found that the strain ATCC23270 could treat larger amount of the refractories than the strain TM-32 could do.In the experiment of bioleaching of spent refractory aluminium melting furnace by the strain ATCC23270,high leaching efficiency were obtained on Al,Si,and Ca,and extremely low leaching performance was,however,shown on the rest of elements i.e.,Na,Mn,and Zn.Under the strain TM-32 use,relatively high leaching performance was recognized on Al,Si,Ca,Na,Mn,and Zn.In the experiment of bioleaching for spent refractory copper melting furnace,almost the same leaching trends were shown on Cu,Zn,Al,and Si under the strains ACTT23270 and TM-32 uses.     

Simultaneous removal of ammonium and phosphate by zeolite synthesized from coal fly ash as influenced by acid treatment

Zeolite synthesized from fly ash （ZFA） without modification is not efficient for the purification of NH4＋ and phosphate at low concentrations that occur in real effluents, despite the high potential removal capacity. To develop an effective technique to enhance the removal efficiency of ammonium and phosphate at low concentrations, ZFA was modified with acid treatment and the simultaneous removal of ammonium and phosphate in a wide range of concentration was investigated. It was seen that when compared with untreated ZFA, only the treatment by 0.01 mol/L of H2SO4 significantly improved the removal efficiency of ammonium at low initial concentrations. The behavior was well explained by the pH effect. Treatment by more concentrated H2SO4 led to the deterioration of the ZFA structure and a decrease in the cation exchange capacity. Treatment by 0.01 mol/L H2SO4 improved the removal efficiency of phosphate by ZFA at all initial P concentrations, while the treatment by concentrated H2SO4 （≥0.9 mol/L） resulted in a limited maximum phosphate immobilization capacity （PIC）. It was concluded that through a previous mild acid treatment （e.g. 0.01 mol/L of H2SO4）, ZFA can be used in the simultaneous removal of NH4＋ and P at low concentrations simulating real effluent.