Negative air ions are natural components of the air we breathe Forests are the main continuous natural source of negative air ions （NAI）. The spatio-temporal patterns of negative air ions were explored in Shanghai, ...Negative air ions are natural components of the air we breathe Forests are the main continuous natural source of negative air ions （NAI）. The spatio-temporal patterns of negative air ions were explored in Shanghai, based on monthly monitoring in 15 parks from March 2009 to February 2010. In each park, sampling sites were selected in forests and open spaces. The annual variation in negative air ion concentrations （NAIC） showed peak values from June to October and minimum values from December to January. NAIC were highest in summer and autumn, intermediate in spring, and lowest in winter. During spring and summer, NAIC in open spaces were significantly higher in rural areas than those in suburban areas. However, there were no significant differences in NAIC at forest sites among seasons. For open spaces, total suspended particles （TSP） were the dominant determining factor of NAIC in sum- mer, and air temperature and air humidity were the dominant determining factors of NAIC in spring, which were tightly correlated with Shanghai＇s ongoing urbanization and its impacts on the environment. R is suggested that urbanization could induce variation in NAIC along the urban-rural gradient, but that may not change the temporal variation pattern. Fur- thermore, the effects of urbanization on NAIC were limited in non-vegetated or less-vegetated sites, such as open spaces, but not in well-vegetated areas, such as urban forests. Therefore, we suggest that urban greening, especially urban forest, has significant resistance to theeffect of urbanization on NAIC.展开更多
The spatial-temporal relationship between high-quality source rocks and reservoirs is a key factor when evaluating the formation,occurrence,and prospectivity of tight oil and gas reservoirs.In this study,we analyze th...The spatial-temporal relationship between high-quality source rocks and reservoirs is a key factor when evaluating the formation,occurrence,and prospectivity of tight oil and gas reservoirs.In this study,we analyze the fundamental oil and gas accumulation processes occurring in the Songliao Basin,contrasting tight oil sand reservoirs in the south with tight gas sand reservoirs in the north.This is done using geochemical data,constant-rate and conventional mercury injection experiments,and fluid inclusion analyses.Our results demonstrate that as far as fluid mobility is concerned,the expulsion center coincides with the overpressure zone,and its boundary limits the occurrence of tight oil and gas accumulations.In addition,the lower permeability limit of high-quality reservoirs,controlled by pore-throat structures,is 0.1×10^-3μm^2 in the fourth member of the Lower Cretaceous Quantou Formation(K1q^4)in the southern Songliao Basin,and 0.05×10^-3μm^2 in the Lower Cretaceous Shahezi Formation(K1sh)in the northern Songliao Basin.Furthermore,the results indicate that the formation of tight oil and gas reservoirs requires the densification of reservoirs prior to the main phase of hydrocarbon expulsion from the source rocks.Reservoir“sweet spots”develop at the intersection of high-quality source rocks(with high pore pressure)and reservoirs(with high permeability).展开更多
Custom designed and built meso shear test equipment was used to examine the shear crack propagation in gassy coal under different gas pressures.The spatial-temporal evolution of gas migration pathways in the coal duri...Custom designed and built meso shear test equipment was used to examine the shear crack propagation in gassy coal under different gas pressures.The spatial-temporal evolution of gas migration pathways in the coal during shear loading was also researched.The results show that gas pressure can hasten crack growth at the shear fracture surface,can reduce the shear strength of gassy coal,and can accelerate the shear failure process.Shear failure in gassy coal exhibits five stages:the pre-crack stage;the stable crack growth stage;the unsteady crack growth stage;the fracture stage;and,finally,the friction crack stage.The shear breaking creates two kinds of crack,shear cracks and tensile cracks.Cracks first appear in the shear plane at both ends and then extend toward the center until a shear fracture surface forms.The direction of shear crack propagation diverges from the predetermined shear plane by an angle of about 5°-10°.展开更多
Forest soil carbon is a major carbon pool of terrestrial ecosystems,and accurate estimation of soil organic carbon(SOC)stocks in forest ecosystems is rather challenging.This study compared the prediction performance o...Forest soil carbon is a major carbon pool of terrestrial ecosystems,and accurate estimation of soil organic carbon(SOC)stocks in forest ecosystems is rather challenging.This study compared the prediction performance of three empirical model approaches namely,regression kriging(RK),multiple stepwise regression(MSR),random forest(RF),and boosted regression trees(BRT)to predict SOC stocks in Northeast China for 1990 and 2015.Furthermore,the spatial variation of SOC stocks and the main controlling environmental factors during the past 25 years were identified.A total of 82(in 1990)and 157(in 2015)topsoil(0–20 cm)samples with 12 environmental factors(soil property,climate,topography and biology)were selected for model construction.Randomly selected80%of the soil sample data were used to train the models and the other 20%data for model verification using mean absolute error,root mean square error,coefficient of determination and Lin's consistency correlation coefficient indices.We found BRT model as the best prediction model and it could explain 67%and 60%spatial variation of SOC stocks,in 1990,and 2015,respectively.Predicted maps of all models in both periods showed similar spatial distribution characteristics,with the lower SOC in northeast and higher SOC in southwest.Mean annual temperature and elevation were the key environmental factors influencing the spatial variation of SOC stock in both periods.SOC stocks were mainly stored under Cambosols,Gleyosols and Isohumosols,accounting for 95.6%(1990)and 95.9%(2015).Overall,SOC stocks increased by 471 Tg C during the past 25 years.Our study found that the BRT model employing common environmental factors was the most robust method for forest topsoil SOC stocks inventories.The spatial resolution of BRT model enabled us to pinpoint in which areas of Northeast China that new forest tree planting would be most effective for enhancing forest C stocks.Overall,our approach is likely to be useful in forestry management and ecological restoration at and beyond the regional scale.展开更多
The forest stand database of Bilahe Forestry Bureau, Inner Mongolia of China was taken as an example to demonstrate the whole process of building a temporal geodatabase by means of reengineering. The process was compo...The forest stand database of Bilahe Forestry Bureau, Inner Mongolia of China was taken as an example to demonstrate the whole process of building a temporal geodatabase by means of reengineering. The process was composed of establishing a conceptual data model from the initial database, constructing a logical database by means of mapping, and building a temporal geodatabase with the help of Computer-Aided Software Engineering （CASE） tool and Unified Markup Language （UML）. The results showed that as the reengineered forest stand geodatabase was dynamic, it could easily store the historical data and answer time related questions by Structured Query Language （SQL）, meanwhile, it maintains the integrity of database and eliminates the redundancy.展开更多
The volume of trajectory data has become tremendously huge in recent years. How to effectively and efficiently maintain and compute such trajectory data has become a challenging task. In this paper, we propose a traje...The volume of trajectory data has become tremendously huge in recent years. How to effectively and efficiently maintain and compute such trajectory data has become a challenging task. In this paper, we propose a trajectory spatial and temporal compression framework, namely CLEAN. The key of spatial compression is to mine meaningful trajectory frequent patterns on road network. By treating the mined patterns as dictionary items, the long trajectories have the chance to be encoded by shorter paths, thus leading to smaller space cost. And an error-bounded temporal compression is carefully designed on top of the identified spatial patterns for much low space cost. Meanwhile, the patterns are also utilized to improve the performance of two trajectory applications, range query and clustering, without decompression overhead. Extensive experiments on real trajectory datasets validate that CLEAN significantly outperforms existing state-of-art approaches in terms of spatial-temporal compression and trajectory applications.展开更多
Many scholars have focused on applying machine learning models in bottom hole pressure (BHP) prediction. However, the complex and uncertain conditions in deep wells make it difficult to capture spatial and temporal co...Many scholars have focused on applying machine learning models in bottom hole pressure (BHP) prediction. However, the complex and uncertain conditions in deep wells make it difficult to capture spatial and temporal correlations of measurement while drilling (MWD) data with traditional intelligent models. In this work, we develop a novel hybrid neural network, which integrates the Convolution Neural Network (CNN) and the Gate Recurrent Unit (GRU) for predicting BHP fluctuations more accurately. The CNN structure is used to analyze spatial local dependency patterns and the GRU structure is used to discover depth variation trends of MWD data. To further improve the prediction accuracy, we explore two types of GRU-based structure: skip-GRU and attention-GRU, which can capture more long-term potential periodic correlation in drilling data. Then, the different model structures tuned by the Bayesian optimization (BO) algorithm are compared and analyzed. Results indicate that the hybrid models can extract spatial-temporal information of data effectively and predict more accurately than random forests, extreme gradient boosting, back propagation neural network, CNN and GRU. The CNN-attention-GRU model with BO algorithm shows great superiority in prediction accuracy and robustness due to the hybrid network structure and attention mechanism, having the lowest mean absolute percentage error of 0.025%. This study provides a reference for solving the problem of extracting spatial and temporal characteristics and guidance for managed pressure drilling in complex formations.展开更多
The plasma shielding effect is one of the major weaknesses of laser-induced breakdown spectroscopy(LIBS)as it causes non-linearity in signal strength.Although LIBS is typically carried out in constant laser energy,thi...The plasma shielding effect is one of the major weaknesses of laser-induced breakdown spectroscopy(LIBS)as it causes non-linearity in signal strength.Although LIBS is typically carried out in constant laser energy,this non-linearity causes a reduction in sensitivity.In this work,we systematically examine laser-induced plasma,formed by two different excitation source modes,i.e.single pulse(SP)-excitation and single-beam-splitting double-pulse(SBS-DP)-excitation over Zr-2.5%Nb alloy.The two most important plasma parameters influencing the emission line intensity,plasma temperature(Te)and electron density(Ne)were studied and compared for both modes of laser excitation.Comparison of the results conclusively demonstrates that due to the splitting of the laser energy in the SBS-DP mode,the plasma shielding effect is significantly reduced.The reduced plasma shielding translates to an increased laser–sample coupling under SBS-DP mode.Temporal imaging of the total intensity of the laser-induced plasma in both excitation modes was also studied.The study shows how the plasma shielding effect can be reduced to improve the analytical quality of the LIBS methodology.展开更多
We systematically explore near equilibrium, flow-driven, and flow-activity coupled dynamics of polar active liquid crystals using a continuum model. Firstly, we re-derive the hydrodynamic model to ensure the thermodyn...We systematically explore near equilibrium, flow-driven, and flow-activity coupled dynamics of polar active liquid crystals using a continuum model. Firstly, we re-derive the hydrodynamic model to ensure the thermodynamic laws are obeyed and elastic stresses and forces are consistently accounted. We then carry out a linear stability analysis about constant steady states to study near equilibrium dynamics around the steady states, revealing long-wave instability inherent in this model system and how active parameters in the model affect the instability. We then study model predictions for one- dimensional （1D） spatial-temporal structures of active liquid crystals in a channel subject to physical boundary conditions. We discuss the model prediction in two selected regimes, one is the viscous stress dominated regime, also known as the flow-driven regime, while the other is the full regime, in which all active mechanisms are included. In the viscous stress dominated regime, the polarity vector is driven by the prescribed flow field. Dynamics depend sensitively on the physical boundary condition and the type of the driven flow field. Bulk-dominated temporal periodic states and spatially homogeneous states are possible under weak anchoring conditions while spatially inhomogeneous states exist under strong anchoring conditions. In the full model, flow-orientation interaction generates a host of planar as well as out-of-plane spatial-temporal structures related to the spontaneous flows due to the molecular self-propelled motion. These results provide contact with the recent literature on active nematic suspensions. In addition, symmetry breaking pattems emerge as the additional active viscous stress due to the polarity vector is included in the force balance. The inertia effect is found to limit the long-time survival of spatial structures to those with small wave numbers, i.e., an asymptotic coarsening to long wave structures. A rich set of mechanisms for generating and limiting the flow structures as well as the spatial-temporal structures predicted by the model are displayed.展开更多
Accurate traffic pattern prediction in largescale networks is of great importance for intelligent system management and automatic resource allocation.System-level mobile traffic forecasting has significant challenges ...Accurate traffic pattern prediction in largescale networks is of great importance for intelligent system management and automatic resource allocation.System-level mobile traffic forecasting has significant challenges due to the tremendous temporal and spatial dynamics introduced by diverse Internet user behaviors and frequent traffic migration.Spatialtemporal graph modeling is an efficient approach for analyzing the spatial relations and temporal trends of mobile traffic in a large system.Previous research may not reflect the optimal dependency by ignoring inter-base station dependency or pre-determining the explicit geological distance as the interrelationship of base stations.To overcome the limitations of graph structure,this study proposes an adaptive graph convolutional network(AGCN)that captures the latent spatial dependency by developing self-adaptive dependency matrices and acquires temporal dependency using recurrent neural networks.Evaluated on two mobile network datasets,the experimental results demonstrate that this method outperforms other baselines and reduces the mean absolute error by 3.7%and 5.6%compared to time-series based approaches.展开更多
基金supported by the National Natural Science Foundation of China(No.40971041)
文摘Negative air ions are natural components of the air we breathe Forests are the main continuous natural source of negative air ions （NAI）. The spatio-temporal patterns of negative air ions were explored in Shanghai, based on monthly monitoring in 15 parks from March 2009 to February 2010. In each park, sampling sites were selected in forests and open spaces. The annual variation in negative air ion concentrations （NAIC） showed peak values from June to October and minimum values from December to January. NAIC were highest in summer and autumn, intermediate in spring, and lowest in winter. During spring and summer, NAIC in open spaces were significantly higher in rural areas than those in suburban areas. However, there were no significant differences in NAIC at forest sites among seasons. For open spaces, total suspended particles （TSP） were the dominant determining factor of NAIC in sum- mer, and air temperature and air humidity were the dominant determining factors of NAIC in spring, which were tightly correlated with Shanghai＇s ongoing urbanization and its impacts on the environment. R is suggested that urbanization could induce variation in NAIC along the urban-rural gradient, but that may not change the temporal variation pattern. Fur- thermore, the effects of urbanization on NAIC were limited in non-vegetated or less-vegetated sites, such as open spaces, but not in well-vegetated areas, such as urban forests. Therefore, we suggest that urban greening, especially urban forest, has significant resistance to theeffect of urbanization on NAIC.
基金supported by the National Natural Science Foundation of China (Nos. 41210005 and 41776081)the National Oil and Gas Major Project of China (No. 2011ZX05007-001)the Applied Basic Research Program of Qingdao (No. 2016239)
文摘The spatial-temporal relationship between high-quality source rocks and reservoirs is a key factor when evaluating the formation,occurrence,and prospectivity of tight oil and gas reservoirs.In this study,we analyze the fundamental oil and gas accumulation processes occurring in the Songliao Basin,contrasting tight oil sand reservoirs in the south with tight gas sand reservoirs in the north.This is done using geochemical data,constant-rate and conventional mercury injection experiments,and fluid inclusion analyses.Our results demonstrate that as far as fluid mobility is concerned,the expulsion center coincides with the overpressure zone,and its boundary limits the occurrence of tight oil and gas accumulations.In addition,the lower permeability limit of high-quality reservoirs,controlled by pore-throat structures,is 0.1×10^-3μm^2 in the fourth member of the Lower Cretaceous Quantou Formation(K1q^4)in the southern Songliao Basin,and 0.05×10^-3μm^2 in the Lower Cretaceous Shahezi Formation(K1sh)in the northern Songliao Basin.Furthermore,the results indicate that the formation of tight oil and gas reservoirs requires the densification of reservoirs prior to the main phase of hydrocarbon expulsion from the source rocks.Reservoir“sweet spots”develop at the intersection of high-quality source rocks(with high pore pressure)and reservoirs(with high permeability).
基金supported in part by the State Key Basic Research Program of China(No.2011CB201203)in part by the General Project of the National Natural Science Foundation of China(No.50974141)the Fundamental Research Funds for the Central Universities(No.CDJZR12240055)
文摘Custom designed and built meso shear test equipment was used to examine the shear crack propagation in gassy coal under different gas pressures.The spatial-temporal evolution of gas migration pathways in the coal during shear loading was also researched.The results show that gas pressure can hasten crack growth at the shear fracture surface,can reduce the shear strength of gassy coal,and can accelerate the shear failure process.Shear failure in gassy coal exhibits five stages:the pre-crack stage;the stable crack growth stage;the unsteady crack growth stage;the fracture stage;and,finally,the friction crack stage.The shear breaking creates two kinds of crack,shear cracks and tensile cracks.Cracks first appear in the shear plane at both ends and then extend toward the center until a shear fracture surface forms.The direction of shear crack propagation diverges from the predetermined shear plane by an angle of about 5°-10°.
基金funded by the National Key R&D Program of China(Grant No.2021YFD1500200)National Natural Science Foundation of China(Grant No.42077149)+4 种基金China Postdoctoral Science Foundation(Grant No.2019M660782)National Science and Technology Basic Resources Survey Program of China(Grant No.2019FY101300)Doctoral research start-up fund project of Liaoning Provincial Department of Science and Technology(Grant No.2021-BS-136)China Scholarship Council(201908210132)Young Scientific and Technological Talents Project of Liaoning Province(Grant Nos.LSNQN201910 and LSNQN201914)。
文摘Forest soil carbon is a major carbon pool of terrestrial ecosystems,and accurate estimation of soil organic carbon(SOC)stocks in forest ecosystems is rather challenging.This study compared the prediction performance of three empirical model approaches namely,regression kriging(RK),multiple stepwise regression(MSR),random forest(RF),and boosted regression trees(BRT)to predict SOC stocks in Northeast China for 1990 and 2015.Furthermore,the spatial variation of SOC stocks and the main controlling environmental factors during the past 25 years were identified.A total of 82(in 1990)and 157(in 2015)topsoil(0–20 cm)samples with 12 environmental factors(soil property,climate,topography and biology)were selected for model construction.Randomly selected80%of the soil sample data were used to train the models and the other 20%data for model verification using mean absolute error,root mean square error,coefficient of determination and Lin's consistency correlation coefficient indices.We found BRT model as the best prediction model and it could explain 67%and 60%spatial variation of SOC stocks,in 1990,and 2015,respectively.Predicted maps of all models in both periods showed similar spatial distribution characteristics,with the lower SOC in northeast and higher SOC in southwest.Mean annual temperature and elevation were the key environmental factors influencing the spatial variation of SOC stock in both periods.SOC stocks were mainly stored under Cambosols,Gleyosols and Isohumosols,accounting for 95.6%(1990)and 95.9%(2015).Overall,SOC stocks increased by 471 Tg C during the past 25 years.Our study found that the BRT model employing common environmental factors was the most robust method for forest topsoil SOC stocks inventories.The spatial resolution of BRT model enabled us to pinpoint in which areas of Northeast China that new forest tree planting would be most effective for enhancing forest C stocks.Overall,our approach is likely to be useful in forestry management and ecological restoration at and beyond the regional scale.
基金China Scholarship (2004836052)‘Voyage Project’ of Jiangxi Province(2007)
文摘The forest stand database of Bilahe Forestry Bureau, Inner Mongolia of China was taken as an example to demonstrate the whole process of building a temporal geodatabase by means of reengineering. The process was composed of establishing a conceptual data model from the initial database, constructing a logical database by means of mapping, and building a temporal geodatabase with the help of Computer-Aided Software Engineering （CASE） tool and Unified Markup Language （UML）. The results showed that as the reengineered forest stand geodatabase was dynamic, it could easily store the historical data and answer time related questions by Structured Query Language （SQL）, meanwhile, it maintains the integrity of database and eliminates the redundancy.
基金National Natural Science Foundation of China (Grant No. 61772371,No. 61972286)
文摘The volume of trajectory data has become tremendously huge in recent years. How to effectively and efficiently maintain and compute such trajectory data has become a challenging task. In this paper, we propose a trajectory spatial and temporal compression framework, namely CLEAN. The key of spatial compression is to mine meaningful trajectory frequent patterns on road network. By treating the mined patterns as dictionary items, the long trajectories have the chance to be encoded by shorter paths, thus leading to smaller space cost. And an error-bounded temporal compression is carefully designed on top of the identified spatial patterns for much low space cost. Meanwhile, the patterns are also utilized to improve the performance of two trajectory applications, range query and clustering, without decompression overhead. Extensive experiments on real trajectory datasets validate that CLEAN significantly outperforms existing state-of-art approaches in terms of spatial-temporal compression and trajectory applications.
基金The authors express their appreciation to National Key Research and Development Project“Key Scientific Issues of Revolutionary Technology”(2019YFA0708300)Strategic Cooperation Technology Projects of CNPC and CUPB(ZLZX2020-03)+1 种基金Distinguished Young Foundation of National Natural Science Foundation of China(52125401)Science Foundation of China University of Petroleum,Beijing(2462022SZBH002).
文摘Many scholars have focused on applying machine learning models in bottom hole pressure (BHP) prediction. However, the complex and uncertain conditions in deep wells make it difficult to capture spatial and temporal correlations of measurement while drilling (MWD) data with traditional intelligent models. In this work, we develop a novel hybrid neural network, which integrates the Convolution Neural Network (CNN) and the Gate Recurrent Unit (GRU) for predicting BHP fluctuations more accurately. The CNN structure is used to analyze spatial local dependency patterns and the GRU structure is used to discover depth variation trends of MWD data. To further improve the prediction accuracy, we explore two types of GRU-based structure: skip-GRU and attention-GRU, which can capture more long-term potential periodic correlation in drilling data. Then, the different model structures tuned by the Bayesian optimization (BO) algorithm are compared and analyzed. Results indicate that the hybrid models can extract spatial-temporal information of data effectively and predict more accurately than random forests, extreme gradient boosting, back propagation neural network, CNN and GRU. The CNN-attention-GRU model with BO algorithm shows great superiority in prediction accuracy and robustness due to the hybrid network structure and attention mechanism, having the lowest mean absolute percentage error of 0.025%. This study provides a reference for solving the problem of extracting spatial and temporal characteristics and guidance for managed pressure drilling in complex formations.
文摘The plasma shielding effect is one of the major weaknesses of laser-induced breakdown spectroscopy(LIBS)as it causes non-linearity in signal strength.Although LIBS is typically carried out in constant laser energy,this non-linearity causes a reduction in sensitivity.In this work,we systematically examine laser-induced plasma,formed by two different excitation source modes,i.e.single pulse(SP)-excitation and single-beam-splitting double-pulse(SBS-DP)-excitation over Zr-2.5%Nb alloy.The two most important plasma parameters influencing the emission line intensity,plasma temperature(Te)and electron density(Ne)were studied and compared for both modes of laser excitation.Comparison of the results conclusively demonstrates that due to the splitting of the laser energy in the SBS-DP mode,the plasma shielding effect is significantly reduced.The reduced plasma shielding translates to an increased laser–sample coupling under SBS-DP mode.Temporal imaging of the total intensity of the laser-induced plasma in both excitation modes was also studied.The study shows how the plasma shielding effect can be reduced to improve the analytical quality of the LIBS methodology.
基金supported by the National Natural Science Foundation of China(Grant Nos.DMS-1200487,DMR-1122483,and NIH 2R01GM078994-05A1)the Air Force Office of Scientific Research(AFOSR)(Grant No.FA9550-12-1-0178)the Army Research Office(Grant Nos.ARO-12-60317-MS and SC EPSCoR GEAR(CI and CRP))
文摘We systematically explore near equilibrium, flow-driven, and flow-activity coupled dynamics of polar active liquid crystals using a continuum model. Firstly, we re-derive the hydrodynamic model to ensure the thermodynamic laws are obeyed and elastic stresses and forces are consistently accounted. We then carry out a linear stability analysis about constant steady states to study near equilibrium dynamics around the steady states, revealing long-wave instability inherent in this model system and how active parameters in the model affect the instability. We then study model predictions for one- dimensional （1D） spatial-temporal structures of active liquid crystals in a channel subject to physical boundary conditions. We discuss the model prediction in two selected regimes, one is the viscous stress dominated regime, also known as the flow-driven regime, while the other is the full regime, in which all active mechanisms are included. In the viscous stress dominated regime, the polarity vector is driven by the prescribed flow field. Dynamics depend sensitively on the physical boundary condition and the type of the driven flow field. Bulk-dominated temporal periodic states and spatially homogeneous states are possible under weak anchoring conditions while spatially inhomogeneous states exist under strong anchoring conditions. In the full model, flow-orientation interaction generates a host of planar as well as out-of-plane spatial-temporal structures related to the spontaneous flows due to the molecular self-propelled motion. These results provide contact with the recent literature on active nematic suspensions. In addition, symmetry breaking pattems emerge as the additional active viscous stress due to the polarity vector is included in the force balance. The inertia effect is found to limit the long-time survival of spatial structures to those with small wave numbers, i.e., an asymptotic coarsening to long wave structures. A rich set of mechanisms for generating and limiting the flow structures as well as the spatial-temporal structures predicted by the model are displayed.
基金supported by the National Natural Science Foundation of China(61975020,62171053)。
文摘Accurate traffic pattern prediction in largescale networks is of great importance for intelligent system management and automatic resource allocation.System-level mobile traffic forecasting has significant challenges due to the tremendous temporal and spatial dynamics introduced by diverse Internet user behaviors and frequent traffic migration.Spatialtemporal graph modeling is an efficient approach for analyzing the spatial relations and temporal trends of mobile traffic in a large system.Previous research may not reflect the optimal dependency by ignoring inter-base station dependency or pre-determining the explicit geological distance as the interrelationship of base stations.To overcome the limitations of graph structure,this study proposes an adaptive graph convolutional network(AGCN)that captures the latent spatial dependency by developing self-adaptive dependency matrices and acquires temporal dependency using recurrent neural networks.Evaluated on two mobile network datasets,the experimental results demonstrate that this method outperforms other baselines and reduces the mean absolute error by 3.7%and 5.6%compared to time-series based approaches.
