Air pollution forecasting based on attention-based LSTM neural network and ensemble learning

With air pollution having become a global concern, scientists are committed to working on its amelioration. In the field of air pollution prediction, there have been good results in experimental research so far, but few studies have integrated weather forecast information and the properties of air pollution drift. In this work, we propose a novel wind-sensitive attention mechanism with a long short-term memory (LSTM) neural network model to predict the air pollution - PM2.5 concentrations by considering the influence of wind direction and speed on the changes of spatial–temporal PM2.5 concentrations in neighbouring areas. Preliminary predictions for PM2.5 are then made by an LSTM neural network regarding neighbouring pollution; these predictions are “paid attention to” and we finally apply an ensemble learning method based on e X treme G radient B oosting (XGBoost) to combine the preliminary predictions with weather forecasting to make second phase predictions of PM2.5. The experiment is conducted using PM2.5 data and weather forecast data. Our results illustrate that the proposed method is superior to other methods in predicting PM2.5 concentrations, including multi-layer perceptron, support vector regression, LSTM neural network, and extreme gradient boosting algorithm.     

基于多元线性回归的雾霾预测方法研究

提出了一种在线样本更新的多元线性回归分析的雾霾预测方法。首先搜集了北京市天气状况,包括平均气温、湿度、风级等气象数据以及PM2.5、CO、NO₂、SO₂等大气成分浓度数据,然后通过散点图对这些因素进行主要影响因素分析,筛选出对雾霾影响比较明显的因素作为雾霾预测的依据。通过在线样本更新的多元线性回归建立了PM2.5含量预测模型,并将气象要素作为雾霾的判断标准。最后给出实际例子,利用多元线性回归对北京未来一天、三天及一周的PM2.5含量进行较为精确的预测。     

基于特征向量的最小二乘支持向量机PM2.5浓度预测模型

针对大气中细颗粒物(PM2.5)浓度预测的问题,提出一种预测模型。首先,通过引入综合气象指数综合考虑风力、湿度、温度等因素;然后,结合实际二氧化硫（SO2）浓度、二氧化氮（NO2）浓度、一氧化碳（CO）浓度和PM10浓度等,构成特征向量;最后,利用特征向量和PM2.5浓度数据来建立最小二乘支持向量机（LS-SVM）预测模型。经2013年城市A和城市B环境监测中心的数据预测分析表明,引入综合气象指数后预测的准确性提高,误差降低近30%。说明该模型能够较为准确地预测PM2.5浓度,并具有较高的泛化能力。此外还分析了PM2.5浓度与住院率、医院门诊量的关系,发现了它们的高度相关性。     

Forecasting of the daily meteorological pollution using wavelets and support vector machine

The paper presents the method of daily air pollution forecasting by using support vector machine (SVM) and wavelet decomposition. Based on the observed data of NO₂, CO, SO₂ and dust, for the past years and actual meteorological parameters, like wind, temperature, humidity and pressure, we propose the forecasting approach, applying the neural network of SVM type, working in the regression mode. To obtain the acceptable accuracy of forecast we decompose the measured time series data into wavelet representation and predict the wavelet coefficients. On the basis of these predicted values the final forecasting is prepared. The paper presents the results of numerical experiments on the basis of the measurements made by the meteorological stations, situated in the northern region of Poland.     

基于多元回归分析的PM2.5预测研究

PM2.5对人体健康和大气环境质量的影响众所周知,分析、预测PM2.5浓度对污染天气防治与干预有着非常重要的作用。利用灰色关联度、多元回归分析等方法对全国各大城市空气质量进行了研究,分析了影响PM2.5浓度的主要因素并进行了影响程度排序,构建了PM2.5预测模型并进行了预测实践,为我国环境空气质量预报和污染天气防治干预提供了有效的决策信息。     

A combination of hidden Markov model and fuzzy model for stock market forecasting

This paper presents a novel combination of the hidden Markov model (HMM) and the fuzzy models for forecasting stock market data. In a previous study we used an HMM to identify similar data patterns from the historical data and then used a weighted average to generate a ‘one-day-ahead’ forecast. This paper uses a similar approach to identify data patterns by using the HMM and then uses fuzzy logic to obtain a forecast value. The HMM's log-likelihood for each of the input data vectors is used to partition the dataspace. Each of the divided dataspaces is then used to generate a fuzzy rule. The fuzzy model developed from this approach is tested on stock market data drawn from different sectors. Experimental results clearly show an improved forecasting accuracy compared to other forecasting models such as, ARIMA, artificial neural network (ANN) and another HMM-based forecasting model.     

改进Pi-sigma神经网络及其算法在短期负荷预测中的应用

针对电力负荷的特点,综合考虑历史负荷、天气、日类型等因素的影响,将模糊逻辑和神经网络的长处融合在一起,构建了基于改进Pi-sigma神经网络及其算法的短期负荷预测模型,用于预测预报日的各小时负荷,其中在学习速率的选择、隶属度函数参数的更新等多处进行了改进,进一步减小了预测误差.地区电网的实际应用证明了该算法的有效性.     

基于 HY-FMV 模型的雷暴天气预测研究

本文以极端天气中的雷暴天气为研究对象,基于历史气象数据预测未来三小时是否发生雷暴。为预测雷暴是否发生,本文分别对极端天气气象数据的采样、数据预处理、特征选择,以及建模分析进行了研究,最终提出一种基于机器学习方法的HY-FMV模型框架对雷暴天气进行预测。该模型采用混合模型进行数据预处理,基于概率分布与模型评价进行特征的选择和构建,并使用梯度提升树算法对极端天气进行预测分类。最后,本文以2010年到2015年福建和广东两省数据为例,分别使用本文所提出的HY-FMV模型,和随机森林算法等进行雷暴天气预测,结果表明,本文所提出的HY-FMV模型在F1指标上精度达到78%,相比其他算法,在雷暴天气预测精度上提高了0.5%-0.6%。     

模糊化的Modular模糊神经网络降水预报模型研究

以广西西南部前汛期5、6月25个气象站平均逐日降水量作为预报对象,采用自然正交分解方法和模糊化方法对输入因子预处理后,结合Modular模糊神经网络建立了一种新的降水预报模型,并进行了逐日业务预报应用试验.结果表明,该降水预报模型比常规Modular模糊神经网络方法及逐步回归方法有更高的预报精度,具有较好的业务应用前景.     

基于层次聚类和极限学习机的母线短期负荷预测

利用传统方法预测母线负荷时,通常选取离待测日相近的一段时间作为历史相似日进行模型训练,没有考虑其天气情况、星期类型、节假日等因素的影响,相似日与待测日特征相差较大。为解决以上问题,提出一种基于层次聚类（HC）和极限学习机（ELM）的母线负荷预测算法。首先使用层次聚类法将母线历史日负荷进行聚类,然后对层次聚类得出的聚类结果建立决策树,其次根据待测日的温度、湿度、星期和节假日类型等日属性在决策树中匹配出训练极限学习机预测模型的历史日负荷,最后建立极限学习机预测模型,对待测日母线日负荷进行预测。对两条不同母线的负荷进行了预测,与传统单一的极限学习机相比,所提算法的平均绝对百分比误差（MAPE）分别降低了1.4和0.8个百分点。实验结果表明,所提算法预测母线负荷具有更高的预测精度和稳定性。     

基于改进粒子群算法的模糊神经网络及其在短时天气预报中的应用

提出一种基于改进粒子群模糊神经网络进行短时天气预测的方法,将粒子群算法与模糊人工神经网络进行融合,充分发挥粒子群算法全局寻优的优势。以上海地区天气预报作为实例,建立了基于改进粒子群算法的多模型模糊神经网络预报模型,试验结果表明该方法对于短时天气预报具有较好的准确度,得到了上海中心气象台有关专家的肯定。     

Rule-based Mamdani-type fuzzy modeling of skin permeability

Two Mamdani type fuzzy models (three inputs–one output and two inputs–one output) were developed to predict the permeability of compounds through human skin. The models were derived from multiple data sources including laboratory data, published data bases, published statistical models, and expert opinion. The inputs to the model include information about the compound (molecular weight and octonal–H₂O partition coefficient) and the application temperature. One model included all three parameters as inputs and the other model only included information about the compound. The values for mole molecular weight ranged from 30 to 600 Da. The values for the log of the octonal–H₂O partition coefficient ranged from −3.1 to 4.34. The values for the application temperature ranged from 22 to 39 °C. The predicted values of the log of permeability coefficient ranged from −5.5 to −0.08.Each model was a collection of rules that express the relationship of each input to the permeability of the compound through human skin. The quality of the model was determined by comparing predicted and actual fuzzy classification and defuzzification of the predicted outputs to get crisp values for correlating estimates with published values. A modified form of the Hamming distance measure is proposed to compare predicted and actual fuzzy classification. An entropy measure is used to describe the ambiguity associated with the predicted fuzzy outputs.The three input model predicted over 70% of the test data within one-half of a fuzzy class of the published data. The two input model predicted over 40% of the test data within one-half of a fuzzy class of the published data. Comparison of the models show that the three input model exhibited less entropy than the two input model.     

Selecting an optimal model for forecasting the volumes of railway goods transportation

Consideration was given to selection of an optimal model of short-term forecasting of the volumes of railway transport from the historical and exogenous time series. The historical data carry information about the transportation volumes of various goods between pairs of stations. It was assumed that the result of selecting an optimal model depends on the level of aggregation in the types of goods, departure and destination points, and time. Considered were the models of vector autoregression, integrated model of the autoregressive moving average, and a nonparametric model of histogram forecasting. Criteria for comparison of the forecasts on the basis of distances between the errors of model forecasts were proposed. They are used to analyze the models with the aim of determining the admissible requests for forecast, the actual forecast depth included.     

Predicting daily ozone concentration maxima using fuzzy time series based on a two-stage linguistic partition method

Air pollution is a result of global warming, greenhouse effects, and acid rain. Especially in highly industrialization areas, air pollution has become a major environmental issue. Poor air quality has both acute and chronic effects on human health. The detrimental effects of ambient ozone on human health and the Earth’s ecosystem continue to be a national concern in Taiwan. The pollutant standard index (PSI) has been adopted to assess the degree of air pollution in Taiwan. The standardized daily air quality report provides a simple number on a scale of 0 to 500 related to the health effects of air quality levels. The report focuses on health and the current PSI subindices to reflect measured ozone (O₃) concentrations. Therefore, this study uses the O₃ attribute to evaluate air quality. In an effort to forecast daily maximum ozone concentrations, many researchers have developed daily ozone forecasting models. However, this continuing worldwide environmental problem suggests the need for more accurate models. This paper proposes two new fuzzy time series based on a two-stage linguistic partition method to predict air quality with daily maximum O₃ concentration: Stage 1, use the fuzzy time series based on the cumulative probability distribution approach (CPDA) to partition the universe of discourse into seven intervals; Stage 2, use two linguistic partition methods, the CPDA and the uniform discretion method (UDM), to repartition each interval into three subintervals. To verify the forecasting performance of the proposed methods in detail, the practical collected data is used as and evaluating dataset; five other methodologies (AR, MA, ARMA, Chen’s and Yu’s) are used as comparison models. The proposed methods both show a greatly improved performance in daily maximal ozone concentration prediction accuracy compared with the other models.     

On the feasibility of temporally upscaling instantaneous evapotranspiration using weather forecast information

Quantification of land-surface evapotranspiration (ET) is highly significant in water resources management, climate change studies, and numerical weather prediction. The constant reference evaporative fraction method (EF_r, the ratio of the actual to reference ET), which assumes that the daily EF_r is equal to that at the satellite overpass time, is a scheme that has been widely applied to upscale remotely sensed instantaneous ET to daily ET. To overcome the difficulties encountered in the acquisition of tower-based meteorological variables, this study investigates the feasibility of using publicly available weather forecast information to estimate the daily reference ET using the constant EF_r method. A two-source energy balance model is adopted to compute the instantaneous ET using Moderate-Resolution Imaging Spectroradiometer (MODIS) remote-sensing data acquired between January 2011 and October 2012 at the Yucheng Comprehensive Experimental Station in the North China Plain. The results show that the daily maximum and minimum air temperatures from weather forecast information are consistent with the corresponding ground-based measurements, with a bias of 0.8 K and a root mean square error (RMSE) of <2.0 K. The daily global solar radiation and daily wind speed were poorly forecast when compared with the ground-based measurements. Using the meteorological variables from the daily weather forecast information produced a small bias of 0.1 mm day^–1 and an RMSE of 0.6 mm day^–1 when the estimated daily reference ET was compared with that derived using the ground-based meteorological measurements. When the remotely sensed instantaneous ET and half-hourly reference ET were as accurate as the ground-based measurements, the upscaling method produced the daily ET, using the meteorological variables from the weather forecast information, with a bias of 0.1 mm day^–1 and an RMSE of 0.7 mm day^–1.     

An optimization platform for Poland's power sector considering air pollution and health effects

The article demonstrates the features and applicability of the πESA platform designed for optimization of the Poland's power sector considering air pollution and health effects. πESA is comprised of: a bottom-up energy-economic model TIMES-PL, an air quality modelling system Polyphemus and a module for assessment of environmental and health impacts MAEH. It has been designed as a web application employing computational resources of the ZEUS cluster of the PL-Grid infrastructure. The results show, that the impact of carbon prices on the fuel and technological power generation structure is much stronger as compared to impact of fuel prices. Future PM emissions from the centralized power and heat generation sector do not differ much irrespective of energy scenario considered. For analysed cases, the statistical life expectancy in Poland due to long-term exposure to PM_2.5 air pollution is reduced on average by approx. 183 days. That gives over 12 million years lost for all cohorts included in the analysis.     

Predicting weather events using fuzzy rule based system

Discovering and understanding the dynamic phenomena of weather to accurately predict different weather events has been an integral component of scientific investigations worldwide. The weather data, being inherently fuzzy in nature, requires highly complex processing based on human observations, satellite photography, or radar followed by computer simulations. This is further combined with an understanding of the principles of global and local weather dynamics. This paper attempts to solve weather event prediction for Lahore by implementing a fuzzy rule based system. The difficult problem of weather event prediction has been dealt in this paper through two separate experimental settings. In the first experimental setting a smaller dataset consisting of 365 instances with 4 inputs and 8 weather events has been used to develop a fuzzy inference system. In the second experimental setting the developed fuzzy system has been enhanced for a larger dataset consisting of over 2500 data points, having 17 inputs, and 10 weather events. For the later experiments the results of the fuzzy system have been compared with two other models i.e., decision tree (DT) based model and partial least square based regression (PLSR) model. It has been observed in the present study that the performance of the fuzzy system is sensitive to bootstrapping sampling technique that has been used for generating training and test samples for developing the fuzzy, DT and PLSR models. Further the models under consideration have been less sensitive to principal component analysis based dimensionality reduction method.     

Fuzzy robust ν-support vector machine with penalizing hybrid noises on symmetric triangular fuzzy number space

In view of the shortage of ε-insensitive loss function for hybrid noises such as singularity points, biggish magnitude noises and Gaussian noises, this paper presents a new version of fuzzy support vector machine (SVM) which can penalize those hybrid noises to forecast fuzzy nonlinear system. Since there exist some problems of hybrid noises and uncertain data in many actual forecasting problem, the input variables are described as fuzzy numbers by fuzzy comprehensive evaluation. Then by the integration of the triangular fuzzy theory, ν-SVM and loss function theory, the fuzzy robust ν-SVM with robust loss function (FRν-SVM) which can penalize those hybrid noises is proposed. To seek the optimal parameters of FRν-SVM, particle swarm optimization is also proposed to optimize the unknown parameters of FRν-SVM. The results of the application in fuzzy sale system forecasts confirm the feasibility and the validity of the FRν-SVM model. Compared with the traditional model and other SVM methods, FRν-SVM method requires fewer samples and has better generalization capability for Gaussian noise.     

基于随机森林和气象参数的PM2.5浓度等级预测

空气污染不仅危害人类的身心健康,而且还会制约城市的经济发展,其中PM2.5带来的影响尤为突出。为了方便准确地预测出空气中的PM2.5浓度等级,提出了一种基于随机森林的PM2.5浓度等级预测方法,特征因子采用太原市2013年-2017年的气象数据、预测站点的PM2.5浓度变化的时间规律以及与周围站点的时空关联性。该方法首先利用K-Means算法对原始气象数据聚类,降低不同分类器之间的相关性,然后利用欠采样方法对数据进行平衡采样,减少类不平衡对分类器性能的影响,最后利用泛化能力好的随机森林构建预测模型。经过真实数据验证,该方法对PM2.5浓度等级预测具有较好的精确度、召回率与[F]值。     

Prediction of air pollutant concentrations in Tokushima Prefecture,Japan

This paper deals with the prediction problem of air pollutant concentrations over the industrial area of Tokushima Prefecture, Japan. The mathematical model used for the prediction of air pollution, which describes the transport and diffusion of pollutants from stack emissions into the atmosphere, is expressed by the three-dimensional partial differential equation known as the advection diffusion model with initial and boundary conditions. Diffusion coefficients characterizing this model are determined effectively from the statistical processing of the available measured data obtained at two monitoring stations. The measured data of sulpher dioxide concentrations are classified by season, weather and wind speed, and then the identification of these coefficients for each class is carried out by using a line search method because of its numerical stability. The estimation theory is extensively applied to this model for obtaining the useful estimates of the spatial and temporal concentration distributions on the basis of actual measured data, in which the analytical solution of this model is given by using the Green's function under some reasonable assumptions and the suitable transformation. The effectiveness of the proposed approach for the prediction of air pollutant is indicated in simulation results.