Zhenghua Zhang,∗ Jiafeng Zhang,∗∗ Rui Gao,∗ Chongxin Fang,∗∗∗ and Jin Qian∗∗∗∗


  1. [1] M. Chen, X.H. Yu, and Y. Liu, PCNN: Deep convolutionalnetworks for short-term traffic congestion prediction, IEEETransactions on Intelligent Transportation Systems, 19(11),2018, 3550–3559.
  2. [2] S.S. Anjum, Modeling traffic congestion based on air qualityfor greener environment: An empirical study, IEEE Access, 7,2019, 57100–57119.
  3. [3] H. Kim and S. Han, An efficient sensor deployment scheme forlarge-scale wireless sensor networks, IEEE CommunicationsLetters, 19(1), 2015, 98–101.
  4. [4] K. Zheng, S. Zhao, Z. Yang, X. Xiong, and W. Xiang, Designand implementation of LPWA-based air quality monitoringsystem, IEEE Access, 4, 2017, 3238–3245.
  5. [6] in this paper. It makes data transmission moreefficient and effective. After getting monitoring data, theyare preprocessed first. Here, the methods of eliminating theerror of loss and adaptive weighting
  6. [7] are adopted. Thesecan provide the optimal data samples for the subsequentfusion method. In
  7. [8] introduced a method of correlationanalysis. This paper has studied this method of correlationanalysis. In
  8. [9] used particle swarm optimization to modeltraffic pollution. For the fusion of large amounts of trafficdata
  9. [10], In
  10. [11] and
  11. [12] proposed a data fusion schemebased on the BP neural network. Although the BP neuralnetwork is widely used, it still has the disadvantage of slowconvergence and local minimum.In recent years, data fusion theories and predictionmodels have been widely studied in intelligent transporta-tion. In
  12. [13], a hybrid algorithm (STL − LSTM) whichcombines the addition mode of Seasonal-Trend decomposi-tion based on Loess (STL) and the LSTM neural networkwas proposed. Chen et al.’s goal was to mitigate the influ-ences of irregular fluctuation and improve the performanceof short-term metro ridership prediction. Pan and Zhou489
  13. [14] proposed a differential evolution back propagation(DE − BP) neural network traffic prediction model appli-cable for a smart cities’ network to predict the networktraffic.Wang and Fang
  14. [15] used a multiple regression modeland a back propagation (BP) neural network model. Theprediction model is optimized by combining the two meth-ods. Document
  15. [16] proposed a simulation model basedupon the traffic conditions. One of the performance indica-tors relates to traffic pollution. To predict carbon dioxideemissions from roads, second-order regression models wereemployed to fit the exact carbon dioxide (CO2) emissionsin
  16. [17]. In
  17. [18], a mixed forest prediction method con-sidering the spatiotemporal correlation characteristics ofurban road traffic state was constructed by improving theexisting random forest algorithm.In summary, the works mentioned above are mainlybased on traffic flow prediction and traffic pollution moni-toring. The content of these studies is relatively single andnot representative. They did not link traffic congestion topollution. How to combine the urban traffic pollution andcongestion information is quite an interested topic.The main contributions of this paper are concluded asfollows: First, we combined the BP neural network withcorrelation coefficients, analysed the relationship betweentraffic pollution and congestion and studied the potentialfor potential traffic. Further, the wireless sensor networkand ZigBee wireless communication network technologywere used to collect traffic data. Then, this paper adoptsa three-layer data fusion scheme to improve the validityof sample data. Finally, this paper optimizes trainingsamples by correlation analysis and optimizes BP networkfor convergence problems.The rest of this paper is organized as follows. InSection 2, the deployment of the sensor monitoring networkused in this paper is described. Section 3 explains thepreprocessing process of traffic pollution data in detail,that is, the method of eliminating the blunder error andadaptive weighting. In next Section 4 we can see, thecorrelation coefficient is analysed, the initialization weightis optimized and optimized BP neural network algorithmsuitable for traffic data is derived. In Section 5, theexperiment is verified. Finally, Section 6 gives conclusionsand future work.2. Collection of Traffic Pollution Monitoring DataThe traffic pollution data monitoring and acquisition sys-tem
  18. [19] used in this paper is mainly composed of threeunits: sensor unit, microcontroller processing unit andZigBee wireless communication module. The data are fi-nally passed to the monitoring centre server through theembedded gateway as shown in Fig. 1.The existing air quality monitoring system is to moni-tor urban traffic pollution on both sides of urban roads. Ur-ban air pollutants mainly include carbon monoxide, hydro-carbons, sulphur dioxide, nitrogen oxides, ozone, respirableparticulate matter, etc. The traffic pollution monitoringsystem mainly consists of three parts. At the front end ofthe system, the traffic pollution monitoring and acquisi-Monitoringacquisition moduleZigBeeEmbeddedgatewayZigBee 4GMonitoringsensorsMicroprocessorsZigBee 4G Monitoringcenter serverEmbedded gatewayFigure 1. System hardware structure.Intelligent gateway nodes14 23Monitoring acquisition nodesFigure 2. Road layout of system monitoring acquisitionnodes and gateways.tion module combine with the ZigBee wireless transceivermodule to form a data acquisition module. The middle endof the system is the embedded ARM intelligent gatewaymodule. 4G module and server system form the back end.The road layout of the system monitoring collectionnode and intelligent gateway is shown in Fig. 2.This model is built at large crossroads in the centre ofYangzhou City. Collect traffic data of four points at theintersection and upload it to the monitoring centre serverthrough the network. Then the system backend for bigdata fusion. Due to various factors, it is difficult to directlymeasure traffic congestion data. An idea of calculatingthe intersection waiting time by the vehicle timing deviceis more suitable for this. It is the time from the stopto the exit, recorded as t. In this way, it can be judgedthat the congestion at each intersection is at each moment,imitating the rules of the traffic operation index.Set thered light time to 60 s and the congestion index as p,p = t60 × 2.The congestion index is abbreviated as TPI. Its value isdivided into five stages, namely, 0 ∼ 2, 2 ∼ 4, 4 ∼ 6, 6 ∼ 8and more than 8. They are defined as “unblocked”, “basic490CorrelationanalysisVarioussensorsMonitoringdatainspectionDataweightingBP andcorrelationcoefficientcombinedCrossroadsFirst layer fusion Second layer fusion Third layer fusionMonitoringresultsFigure 3. Overall system information fusion structure.smooth” and “light”. Degree of congestion, “moderatecongestion” and “serious congestion”.A 3-layer fusion scheme was proposed. The specificprocess is shown in Fig. 3.Through the data fusion scheme shown in Fig. 3, thetraffic pollution and congestion dataafter processing aremore effective. The following begins to analyse the specificprocess of data fusion for each layer.3. Traffic Pollution Data PreprocessingWe first use the distribution map method to eliminatethe error. Then, the optimized monitoring data are ob-tained by the adaptive weighting algorithm to providehigh-quality training samples.3.1 Distribution Map Method to EliminateBlunder ErrorSensors can be subject to interference in specific environ-ments. This can cause errors in the measurement, i.e.,blunder error. Blunder errors can affect the consistency ofmeasurement data. Therefore, it is necessary to removethe blunder error here.The data obtained from the urban traffic pollutionmonitoring and acquisition system are tested by themethod of distribution map. Its algorithm
  19. [20] is as follows:Let the n measurement results measured by a sensor tobe sorted in ascending order. Get the sequence: x1 ∼ xn.x1 is called the lower limit of the measurement and xn iscalled the upper limit. Record xm as the median. Theupper quartile au is the median value of [xm, xn]. Lowerquartile al is the median value of [x1, xm]. The dispersionof the quartile is d = au − al. The judgment interval ofvalid data is [ρ1, ρ2], ρ1 = al − β2 d, ρ2 = au + β2 d.Where β is a constant and is usually a value of 1.0or 2.0. This constant can be chosen according to theaccuracy of the system requirements. This is used for errorelimination of traffic pollution data. Due to the complexityof the traffic data and the high possibility of change, thedeviation between the data is large. Therefore, β can takea value greater than or equal to 2 to make the data rangewider.If the measurement data are within the interval [ρ1, ρ2],it is considered to be valid consistency measurement dataand the effective interval can be used to eliminate 50% ofthe error. Data outside the interval [ρ1, ρ2] can be con-sidered as a loss error and will eventually be eliminated.The remaining data are valid data after the consistencycheck. The consistent test of the measured data by thedistribution method is not limited by the data distribu-Table 1PM 2.5 Concentration Measurement Data (μg/m3)Various 1 2 3 4 5 6 7Sensors7:00 133.1 132.7 54.8 132.0 131.9 131.7 131.69:00 91.7 131.8 132.0 131.8 132.7 0 131.911:00 161.7 131.8 131.8 131.9 132.0 172.5 131.113:00 132.7 133.7 133.8 134.0 134.1 132.3 132.415:00 132.3 132.4 131.5 130.8 132.1 187.8 132.2Table 2Measurement Data of PM 2.5 Concentration afterTreatment (μg/m3)Various 1 2 3 4 5 6 7Sensors7:00 133.1 132.7 132.0 131.9 131.7 131.69:00 131.8 132.0 131.8 132.7 131.911:00 131.8 131.8 131.9 132.0 131.113:00 132.7 133.7 133.8 134.0 134.1 132.3 132.415:00 132.3 132.4 131.5 130.8 132.1 132.2tion because the quartile dispersion and median value inthe distribution method only depend on the distributionposition of the data, regardless of the size of the extremepoints, thus enhance the robustness of data processing.Take the PM 2.5 concentration data measured by agiven sensor at different times in the same place as anexample.As shown in Table 1, the average value of the firstgroup of data is calculated to be 121.11, the upper quartileis 132.90 and the lower quartile is 131.80. According to thedistribution graph method, the data 54.8 is the divergencedata, which is the error of the deviation. To reduce themeasurement error, you need to reject it. Similarly, afterthe consistency detection process of the programme, thePM 2.5 pollution concentration data of each group isshown in Table 2.As shown in Table 2, the blunder error in the data hasbeen removed and adaptive weighting can be performedbelow.491StartRead theNodata collected bythe sensorYesCalculate mean and varianceEndInitiative valueFigure 4. Weighted fusion flow chart.3.2 Traffic Pollution Data Adaptive WeightingThe adaptive weighted data fusion algorithm
  20. [21] has dif-ferent weights for different sets of measurement data. Un-der the optimal condition that the total mean square erroris the smallest, the corresponding weights are searchedadaptively according to each set of measurement data, sothat the fusion the latter value is optimal. The weightedfusion flow chart is shown in Fig. 4.In a monitoring system, there are n sensors that mon-itor and sample a subject. The monitoring values arex1 ∼ xn. The variance of each sensor node is σ21 ∼ σ2n.The corresponding weighting factors are w1 ∼ wn. Thefinal fusion value of the multi-sensor data should beˆx =ni=1 ωixi.ni=1 ωi = 1, so the total mean squareerror is σ2=ni=1 ωiσ2i .The σ2is a function of each weighting factor ωi.According to the extremum method of the multivariatefunction, the weighting factor when the mean square erroris minimum can be found as ωi = 1σ2ini=11σ2i. At this time,σ2is the minimum value and σ2min = 1ni=1 σ2i.The parameter values of the measured object are ob-jectively existing constants, and thus the estimation canbe made based on the arithmetic mean of the existingmonitoring data. Let the i group of sensors make the kmeasurements, then:¯xi(k) =1kkj=1xi(j) i = 1, 2 · · · n (1)Get an estimate ˆ¯x =ni=1 ωi ¯xi(k). The total meansquare error is ¯σ2= 1kni=1 ω2i σ2i .According to (2):¯σ2=1kni=11σ2i(2)Obviously ¯σ2< σ2min, as k increases, ¯σ2graduallydecreases.The adaptive algorithm can adaptively select a moresuitable weighting factor according to different measuredvalues. Let us take a set of PM 2.5 concentration data asan example.After the above adaptive weighted fusion algorithmprogramme, the mean, variance and weighting factors cor-Table 3Mean, Variance and Weighting Factors of PM 2.5Concentration Data at Each TimeTime 7:00 9:00 11:00 13:00 15:00Mean 132.42 132.02 132.10 133.29 132.00Variance 0.247 0.126 0.260 0.530 0.354Weighting 0.197 0.387 0.187 0.092 0.137responding to the PM2.5 concentration data at each timeare shown in Table 3. The final fusion result was 132.228.In the same way, the data fusion value of other trafficpollution parameters can be obtained by the same methodand the obtained data are closer to the true value of themeasured parameter than the arithmetic average value,which greatly improves the accuracy of the collected data.It provides technical support for the accurate detection ofurban traffic pollution conditions.4. Correlation Coefficient and BP Neural NetworkTraffic pollutants mainly include solid suspended particles,carbon oxides, hydrocarbons, lead and sulphur oxides. Thecorrelation coefficient between each pollutant and the con-gestion index is analysed in turn, so that the degree ofcorrelation with traffic congestion can be obtained. Thisscreens out contaminants with a large correlation coeffi-cient. In practical applications, the connection betweenvariables is irrelevant, so we use the simplified Spearmancorrelation coefficient, where the correlation coefficient r isdefined asr = 1 −6ni=1 d2in(n2 − 1)(3)where di = xi − yi, i = 1, 2, . . . , n, n is the sample size, xiand yi are two sets of data and r is in the range between−1 and 1, it can be seen that the correlation coefficient isdivided into positive correlation and negative correlation.The correlation coefficient r was first proposed by KarlPearson. It is generally believed that the absolute valueof r is above 0.8, and the two sets of data have strongcorrelation. Between 0.3 and 0.8, a weak correlation canbe considered. Below 0.3, there is basically no correlation.Therefore, six kinds of highly correlated pollutants areselected in this paper, namely PM 2.5, PM 10, CO, SO2,NO2 and O3. The correlation coefficients between themand TPI are recorded as r1, r2, r3, r4, r5, and r6. In theBP neural network, the traditional initialization weightproblem is randomly initialized with the standard normaldistribution. If the quality of the data samples is poor,the weight update amount may be small, the update speedis slow and the hidden layer is saturated. To solve thisproblem, we envisage that the weight can be initializedaccording to the correlation analysis result and the initialweight wi is recorded aswi =rir1 + r2 + r3 + r4 + r5 + r6× 100%, i = 1, 2, . . . , n(4)492The weights obtained in (4) are used as initial eval-uation weights. Therefore, the robustness of the samplevalues and initial weights of this paper is improved.It is envisaged to use the three-layer BP network torealize traffic flow time domain information fusion calcula-tion. Under the condition of the traditional BP networkalgorithm, the learning rate setting is large, the oscillationdoes not converge, the learning rate setting is small and theconvergence speed is slow. To overcome the shortcomingsof the BP algorithm, we use a strategy of combining themomentum term with the adaptive adjustment learningrate
  21. [22]. Increasing the momentum term method allowsthe network to not only consider the effect of the error onthe gradient but also consider the influence of the trend onthe error surface, thus effectively suppressing the networkfrom falling into the local minimum state
  22. [23]. In the pro-cess of training, the learning rate adaptively adjusts theforce map to make the algorithm stable, while at the sametime making the learning step size as large as possible, andthe learning rate is adjusted according to the local errorsurface. Record the gradient momentum in the gradientdescent method as ΔG, ΔG = (1 − γ)Dk + γDk − 1.The Dk = − ∂E∂ωkis the negative gradient at step k,η is the learning rate, γ is themomentum factor, 0 =γ < 1, and adding the momentum factor is equivalent toadding the damping term, which can reduce the oscillationtendency of the learning process and effectively improvethe convergence. The weight adjustment algorithm forlearning rate adaptive adjustment is ηk = 2γηk − 1, γ =sign(DkDk − 1), wk + 1 = wk + ηkDk.When two iterations are repeated, the gradient di-rection is the same, indicating that the falling speed istoo slow. At this time, the learning rate is automati-cally doubled. When the direction is opposite, the fallingis excessive and the learning rate is automaticallyhalved.Combining the above two methods, we can obtain theweight correction algorithm of momentum-adaptive learn-ing rate BP neural network; weight update formula iswk + 1 = wk + ηk((1 − γ)Dk + γDk − 1).5. ExperimentBased on the traditional BP neural network, this experi-ment optimizes the BP network for the problem of localconvergence. Moreover, experiments were carried out us-ing the idea of combining correlation coefficients and BPnetworks.5.1 Experimental ProgrammeThe BP neural network designed in this paper has six in-put nodes and one output node. The parameters collectedby the urban traffic pollution monitoring system in thisexperiment mainly include: PM 2.5 pollutant concentra-tion parameter, PM 10 pollutant concentration parameter,traffic exhaust gas emissions CO, SO2, NO2, O3 and otherconcentrations. The design scheme of BP neural networkis shown in Table 4.The monitoring module of this experiment is arrangedat a certain intersection in Yangzhou City. As shown inTable 4Design of BP Neural NetworkNetwork Type: Number of nodes: 1BP Neural Network Layers: 3Input node Output nodeX1 PM 2.5 pollutant Y1 Congestion indexconcentrationX2 PM 10 pollutantconcentrationX3 CO concentration X4SO2 concentration X5NO2 concentration X6O3 concentrationFigure 5. PM 2.5 and TPI.Figure 6. PM 10 and TPI.Figs. 5 and 6, a number of monitoring and collecting nodesare arranged at the intersection of the main roads and isequipped with a fixed ID number. The intermediate ARMgateway is used to receive and upload data. A total of400 sets of data were collected, of which 350 were used astraining samples for BP network and 50 were used as testsamples. Selected training and test samples are shown inTables 5 and 6.493Table 5Part of Training SamplesSamples PM 2.5 PM 10 CO SO2 NO2 O3 Y1 47 107 1 18 46 44 5.642 20 40 0.5 7 15 60 6.423 11 16 0.4 5 20 47 4.474 50 72 0.8 9 29 69 2.785 86 147 1.2 16 43 72 6.686 32 47 0.9 8 40 35 4.077 79 140 1.1 9 29 77 3.998 17 64 0.5 8 24 73 6.39 29 88 0.6 9 32 64 5.2710 41 104 0.7 9 41 62 3.35Table 6Partial Test SamplesSamples PM 2.5 PM 10 CO SO2 NO2 O3 Y1 18 66 0.2 10 30 108 3.92 44 87 0.4 13 47 133 1.753 63 105 0.5 15 66 105 3.194 66 94 0.4 10 30 150 5.985 68 85 0.6 12 44 132 4.076 65 92 0.6 15 43 176 7.77 56 92 0.7 16 42 162 2.28 64 103 0.8 16 42 140 5.369 68 109 0.9 15 33 161 6.2610 16 32 0.5 6 16 91 4.8The traffic pollution data in the table are different foreach unit, and the data size is quite different, which isnot conducive to the training experiment. Therefore, alltraining data will be normalized before training, and thedata will be transformed into interval [0, 1].5.2 Experimental ResultsWe divided this experiment into two parts, which are cor-relation coefficient analysis and BP network verification.5.2.1 Relevance ExperimentAfter the data preprocessing is completed, 100 groupsof traffic pollution data at the same time every day areselected and 100 groups of data obtained after pretreatmentare correlated. A total of 9 pollutants are measured in thisexperiment, and analyze the correlation between them andthe traffic congestion index in turn.As can be seen from Fig. 7, the concentration of PM2.5, PM 10, SO2, CO, NO2 is positively correlated withTPI. From a trend perspective, the higher the concentra-tion of pollutants, the greater the TPI. As can be seenfrom Fig. 7(b), the concentration of O3 exhibits a nega-tive correlation with TPI. Their trends are reversed. Thismay be understood to be unstable in a higher temperatureenvironment.Table 7 is a statistical table of the correlation coeffi-cients of all measured pollutants and congestion index inthis experiment. To more intuitively see the relationshipbetween them, Fig. 8 is drawn. As shown in Fig. 8, weconclude by summarizing the correlations of the four re-gional points: In the experimental study, we found thatamong all the traffic pollutants mentioned in this paper,the correlation coefficients of HC, CO2 and NO with TPIare <0.1. This correlation is basically negligible. To im-prove the training efficiency of the BP neural network,these three pollutants can be removed and the subsequentexperiments will be verified. The BP network weights areinitialized according to the correlation coefficients in (4)and Table 7. In general, CO, NO2 and TPI have thegreatest correlation. In addition, the higher the congestionindex of a region, the greater the correlation.5.2.2 BP Neural Network VerificationThe previous correlation analysis can roughly get the re-lationship between traffic pollution and traffic congestion,and it can also explain the effectiveness of the data pre-processing fusion process. The following data can be ob-tained through BP neural network training before fusion,and further training fusion can be seen. The result isshown in Fig. 9:Figure 9(a) shows the prediction results of the corre-lation coefficient and the optimized neural network. Theresults show that after the optimized BP neural networktraining, the output is basically consistent with the ac-tual situation and the trend of traffic congestion is almostconsistent. This can explain the high reliability of BPneural network and further demonstrate the effectivenessof BP neural network based on data fusion of urban trafficpollution monitoring system. There are certain errors inindividual points, which can be explained by some suddentraffic factors and errors caused by instrument failure. Fig-ure 9(b) shows the prediction results of the traditional BPneural network. The predicted values are the same as themeasured values, but the errors are large. It can be seenfrom Table 8 that different methods cause different errors.The MAE and RMSE generated by the conventional BPneural network are 0.0994 and 0.1556, respectively. Theoptimized BP network is 0.086 and 0.1293, respectively.The result shows that MAE and RMSE of the optimizedBP network are both smaller than the conventional BPnetwork. It is proved that the fusion scheme based on datapreprocessing, correlation analysis and BP neural networkmodelling can make the prediction model more accurate.494TPI(0-10)SO2200100020010002010010500 10 20 30 40 50 60 70 80 90 1000 10 20 30 40 50 60 70 80 90 1000 10 20 30 40 50 60 70 80 90 1000 10 20 30 40 50 60 70 80 90 100time (days)(a)210100500200100010500 10 20 30 40 50 60 70 80 90 1000 10 20 30 40 50 60 70 80 90 1000 10 20 30 40 50 60 70 80 90 1000 10 20 30 40 50 60 70 80 90 100time (days)TPIO3NO2COTPISO2PM10PM2.5TPI(0-10)O3NO2COmg/m3g/m310g/m3(b)g/m3Figure 7. These are statistical charts comparing various pollutants with TPI. The images are labelled in an alphabetical order(a) PM 2.5, PM 10, SO2 and TPI, (b) CO, NO2, O3 and TPI.Table 7Correlation Coefficientr PM 2.5 PM 10 SO2 HC CO CO2 NO NO2 O3TPI 16.83% 10.22% 19.37% 3.03% 33.46% -6.25% 5.24% 22.74% −16.37%403020100-10-20PM2.5 PM10 SO2HC CO CO2VariouspollutantsNO NO O2 3Correlationcoefficient(%)Figure 8. Correlation analysis.It is effective to analyse the relationship between trafficcongestion and pollution.Through the three-layer data fusion, the traffic conges-tion data can be used to judge the current traffic conges-tion. The personnel of the monitoring centre can check thepollution prediction results of the traffic pollution moni-toring system to know the urban road pollution and roadtraffic congestion in advance. The situation provides abasis for urban environmental protection and rational allo-cation of transportation resources and can provide a planfor traffic management and take the lead.6. Conclusion and Future WorksTo study the connection between traffic pollution and con-gestion, we first use the combination of wireless sensor net-work and ZigBee technology to collect traffic data. Then,data monitoring, preprocessing, correlation analysis andBP network are combined to use in this paper. Accordingto the experimental results, traffic pollution and congestionhave a great correlation. Among them, CO and NO2 havehigher correlation with congestion index. It is innovativeto use the method of data fusion to study the connection495Measurement dataFusion dataTPI8 87 76 65 54 43 32 21 100 10 20 30 40 50Fusion sample(a)00 10 20 30 40 50Fusion sample(b)Measurement dataFusion dataTPIFigure 9. These are graphs comparing the predicted results with the original values. The images are labelled in an alphabeticalorder (a) optimized prediction and (b) traditional prediction.Table 8The Performance Comparison of Two MethodsMethod MAE RMSEOptimized BP network 0.0860 0.1293Conventional BP network 0.0994 0.1556between traffic pollution and congestion. Experimentalresults show that this method makes effective. Because ofthe limited conditions, this study only monitored a smallamount of data in a region. In the future, we can col-lect large amounts of data for analysis and comparisonin multiple regions. This can improve the effectivenessof the experiment. It is also important to optimize theneural network to better adapt to the training of trafficdata. In-depth study of this topic has great significancefor the induction of traffic and the protection of urbanenvironment.References[1] M. Chen, X.H. Yu, and Y. Liu, PCNN: Deep convolutionalnetworks for short-term traffic congestion prediction, IEEETransactions on Intelligent Transportation Systems, 19(11),2018, 3550–3559.[2] S.S. Anjum, Modeling traffic congestion based on air qualityfor greener environment: An empirical study, IEEE Access, 7,2019, 57100–57119.[3] H. Kim and S. Han, An efficient sensor deployment scheme forlarge-scale wireless sensor networks, IEEE CommunicationsLetters, 19(1), 2015, 98–101.[4] K. Zheng, S. Zhao, Z. Yang, X. Xiong, and W. Xiang, Designand implementation of LPWA-based air quality monitoringsystem, IEEE Access, 4, 2017, 3238–3245.[5] J. Guevara, F. Barrero, E. Vargas, J. Becerra, and S. Toral,Environmental wireless sensor network for road traffic appli-cations, Intelligent Transport Systems, 6(2), 2012, 177–186.[6] K. Chen, Y. Chen, Y. Tsai, and N. Chen, A novel wirelessmultifunctional electronic current transformer based on ZigBee-based communication, IEEE Transactions on Smart Grid,8(4), 2017, 1888–1897.[7] W. Choi and T. Sarkar, Minimum norm property for the sumof the adaptive weights for a direct data domain least squaresalgorithm, IEEE Transactions on Antennas & Propagation,54(3), 2006, 1045–1050.[8] M. Blagoiev, I. Gruicin, M. Ionascu, and M. Marcu, A studyon correlation between air pollution and traffic, 2018 26thTelecommunications Forum (TELFOR), 2018, 420–425.[9] D.Y. Xu Ning, Feature-adaptive GM(1,1) model and predic-tion model for traffic pollution emissions in China, SystemsEngineering-Theory & Practice, 38(1), 2018, 187–196.[10] P. Wang, C. Wang, J. Lai, Z. Huang, J. Ma, and Y. Mao,Traffic control approach based on multi-source data fusion,IET Intelligent Transport Systems, 13(5), 2019, 764–772.[11] J.H. Chang and C.Y. Tseng, Analysis of correlation betweensecondary PM2.5 and factory pollution sources by using ANNand the correlation coefficient, IEEE Access, 5(99), 2017,22812–22822.[12] X. Pan, W. Zhou, Y. Lu, and N. Sun, Prediction of networktraffic of smart cities based on DE-BP neural network, IEEEAccess, 7, 2019, 55807–55816.[13] D. Chen, J. Zhang, and S. Jiang, Forecasting the short-termmetro ridership with seasonal and trend decomposition usingloess and LSTM neural networks, IEEE Access, 8, 2020,91181–91187.[14] X. Pan, W. Zhou, Y. Lu, and N. Sun, Prediction of networktraffic of smart cities based on DE-BP neural network, IEEEAccess, 7, 2019, 55807–55816.[15] J. Wang, J. Fang, and Y. Zhao, Visual prediction of gasdiffusion concentration based on regression analysis and BPneural network, The Journal of Engineering, 2019(23), 2019,19–23.[16] S.S. Anjum, R.M. Noor, N. Aghamohammadi, I. Ahmedy,L. Mat Kiah, N. Hussin, M.H. Anisi, and M.A. Qureshi,Modeling traffic congestion based on air quality for greenerenvironment: An Empirical study, IEEE Access, 7, 2019,57100–57119.[17] Z. He, W. Zhang, and N. Jia, Estimating carbon dioxideemissions of freeway traffic: A spatiotemporal cell-based model,IEEE Transactions on Intelligent Transportation Systems,21(5), 2020, 1976–1986.[18] Z. Chen, Y. Jiang, and D. Sun, Discrimination and pre-diction of traffic congestion states of urban road networkbased on spatio-temporal correlation, IEEE Access, 8, 2020,3330–3342.[19] H. Wang, L. Dong, W. Wei, W.S. Zhao, K. Xu, and G. Wang,The WSN monitoring system for large outdoor advertisingboards based on ZigBee and MEMS sensor, IEEE SensorsJournal, 18(3), 2018, 1314–1323.[20] R.P. David, R.S. Neto, and C.A. Medina, Nonassisted adaptivealgorithm for data fusion in distributed detection systems,IEEE Transactions on Aerospace & Electronic Systems, 52(3),2016, 1486–1493.496[21] Y. Song, W. Wei, L. Zheng, X. Yang, and L. Wei, An adaptivepansharpening method by using weighted least squares filter,IEEE Geoscience & Remote Sensing Letters, 13(1), 2016,18–22.[22] Y. Liu, X. Zhu, and J. Yang, Fault diagnosis of PV arraybased on optimised BP neural network by improved adaptivegenetic algorithm, The Journal of Engineering, 2017(13), 2017,1427–1431.[23] T. Guo, H. Wang, and L. Liu, Vision-based mobile robot leader-follower control using model predictive control, InternationalJournal of Robotics and Automation, 34(5), 2019, 206–231.

Important Links:

Go Back