我国对于PM_2.5浓度的监测工作起步相对较晚，当前所采用的主要监测手段包括地基观测和卫星遥感监测技术。地面监测设施虽然能够直接测量PM_2.5浓度，但成本昂贵，而卫星遥感反演技术尽管实现了大面积覆盖，却受限于精度不高的问题，对揭示PM_2.5浓度时空变化特征带来了困难。鉴于以上现状，依据GNSS水汽与大气颗粒物的吸湿性的物理关系，本文通过高精度数据解算软件GAMIT/GLOBK解算地基GNSS观测数据获得各测站天顶方向水汽含量，并分析关中城市群PM_2.5浓度和GNSS水汽的相关性，同时分析PM_2.5与其它大气污染物及气象要素的相关性，最后融合上述多源数据构建基于深度学习的关中城市群PM_2.5浓度预测模型。主要得到以下研究结论：

（1）地基GNSS水汽的反演。根据大气颗粒物的吸湿性特点及地基GNSS的反演技术以及GNSS湿延迟受水汽影响的特点，对PM_2.5浓度的变化进行探索与预测研究。首先，通过对关中城市群各GNSS站点观测数据解算，获得对流层天顶总延迟ZTD；随后利用气象资料和Saastamoinen模型计算出天顶干延迟ZHD；通过将天顶总延迟ZTD减去天顶干延迟ZHD，推导出天顶湿延迟ZWD。其次，为提高ZWD向大气水汽转化的精度，构建关中城市群大气加权平均温度模型。最后，结合该模型确定的转换系数与ZWD，计算出地基GNSS水汽值，并通过选取西安和平凉的探空站水汽对计算结果进行了精度评估，证明地基GNSS反演大气水汽具有可靠性。

（2）PM_2.5浓度与GNSS水汽、大气污染物和气象要素的相关性分析。无论采用皮尔逊相关系数还是小波相干谱分析，均显示出地基GNSS水汽与PM_2.5浓度之间在月尺度上存在显著相关性，且小波相干谱分析的结果在展现二者相关性特征及其变化规律方面更为直观且详尽。同时提取发生中度及以上污染天气的PM_2.5浓度和GNSS水汽数据，进行短时间序列上（逐小时）的相关性分析，结果表明，在整个雾霾发生至消散的过程中， GNSS水汽和PM_2.5浓度也呈现一定的相关性。而且，通过比较分析PM_2.5浓度与大气污染物（PM₁₀、SO₂、NO₂、和CO）和气象要素（相对湿度、露点温度）的相关性，发现GNSS水汽和PM_2.5浓度的相关性更显著。

（3）基于深度学习的PM_2.5浓度预测模型构建。结合地基GNSS水汽、大气污染物（PM₁₀、SO₂、NO₂和CO）和气象要素（相对湿度、露点温度）构建了LSTM预测模型和耦合了Transformer与LSTM的T-LSTM预测模型，实现对PM_2.5浓度的预测预报。并将两种模型预测结果进行精度指标评估和对比。结果表明，两种模型对关中城市群的各个地基GNSS测站的PM_2.5浓度均有较好的模拟效果，但相比较于传统LSTM模型，T-LSTM模型在研究区内各个测站的模拟效果更佳，LSTM模型验证集R²在0.73~0.91之间，而T-LSTM模型验证集R²提高到了0.82~0.97之间。

The monitoring of PM_2.5 concentration in China started relatively late, and the main monitoring methods currently used include ground-based observation and satellite remote sensing monitoring technology. Although ground monitoring facilities can directly measure PM_2.5 concentrations, their high cost poses certain challenge, and satellite remote sensing inversion technology, although achieving large-scale coverage, is limited by low accuracy, which poses difficulties in exploring the temporal and spatial characteristics of PM_2.5 concentration changes. Given the above situation, based on the physical relationship between GNSS water vapor and atmospheric particulate matter hygroscopicity, this paper uses high-precision data processing software GAMIT/GLOBK to calculate the ground-based GNSS observation data to obtain the water vapor content in the zenith direction of each station, and analyzes the correlation between PM_2.5 concentration and GNSS water vapor in the Guanzhong urban agglomeration. At the same time, the correlation between PM_2.5 and other atmospheric pollutants and meteorological elements is analyzed. Finally, the above multi-source data is integrated to construct a deep learning based PM_2.5 concentration prediction model for the Guanzhong urban agglomeration. The main research conclusions are as follows:

(1) Inversion of GNSS water vapor in the foundation. Based on the hygroscopicity characteristics of atmospheric particulate matter and the inversion technology of ground GNSS, as well as the influence of water vapor on GNSS wet delay, this study explores and predicts the changes in PM_2.5 concentration. Firstly, by calculating the observation data of various GNSS stations in the Guanzhong urban agglomeration, the total zenith delay ZTD in the troposphere is obtained; Subsequently, meteorological data and the Saastamoinen model were used to calculate the zenith dry delay ZHD; By subtracting the total zenith delay ZTD from the zenith dry delay ZHD, the zenith wet delay ZWD is derived. Secondly, to improve the accuracy of ZWD conversion into atmospheric water vapor, a weighted average temperature model for the Guanzhong urban agglomeration is constructed. Finally, combining the conversion coefficient determined by the model with ZWD, the ground-based GNSS water vapor value was calculated, and the accuracy of the calculation results was evaluated by selecting water vapor from sounding stations in Xi'an and Ping Liang, proving the reliability of ground-based GNSS inversion of atmospheric water vapor.

(2) Correlation analysis between PM_2.5 and GNSS water vapor, atmospheric pollutants, and meteorological elements. Two correlation analysis methods are adopted: one is the traditional Pearson correlation coefficient method, and the other is the wavelet coherence spectrum analysis method with more time-frequency domain correlation characteristics. Comparative studies have shown that at specific time scales, both Pearson correlation coefficient and wavelet coherence spectrum analysis show a significant correlation between the water vapor content and PM_2.5 concentration inverted by ground-based GNSS. Moreover, the results of wavelet coherence spectrum analysis are more intuitive and detailed in demonstrating the correlation characteristics and variation patterns between the two. Simultaneously extracting PM_2.5 concentration and GNSS water vapor data for weather with moderate pollution or above for short-term correlation analysis, the results show that there is a certain correlation between GNSS water vapor and PM_2.5 overall during the entire process of haze occurrence and dissipation. Finally, the correlation between PM_2.5 concentration and atmospheric pollutants and meteorological factors was analyzed, and it was found that there was a significant correlation.

(3) Construction of a PM_2.5 concentration prediction model based on deep learning. One prediction model of LSTM and other prediction model based on coupling LSTM with Transformer named T-LSTM were constructed by combining ground GNSS water vapor, atmospheric pollutants (PM₁₀, SO₂, NO₂, and CO), and meteorological elements (relative humidity, dew point temperature) to achieve the prediction and forecast of PM_2.5 concentration. And evaluate and compare the accuracy indicators of the prediction results of the two models. The results showed that both models had good simulation effects on PM_2.5 concentrations at various GNSS stations in the Guanzhong urban agglomeration. However, compared to the traditional LSTM model, the T-LSTM model had better simulation effects at various stations in the study area. The validation set R² of the LSTM model was between 0.73 and 0.91, while that of the T-LSTM model was increased to between 0.82 and 0.97.

[1]程春英, 尹学博. 雾霾之PM2.5的来源、成分、形成及危害 [J]. 大学化学, 2014, 29(05): 1-6.

[2]云龙龙, 张天舒, 陆钒, 等.无锡市一次霾形成过程大气污染物特征分析[J].大气与环境光学学报, 2015, 10(01): 22-30.

[3]耿迎雪, 胡金朝. 雾霾形成机理研究进展 [J]. 环境与可持续发展, 2017, 42(04): 65-67.

[4]李征航. GPS测量与数据处理 [M]. 武汉：武汉大学出版社, 2013: 154-164.

[5]潘元进. GPS气象学理论方法应用与研究 [D].桂林理工大学, 2012:1-6.

[6]丁金才. GPS气象学及其应用 [M]. 北京：气象出版社, 2009: 14-18.

[7]张洛恺. 地基GNSS反演大气水汽含量方法研究 [D]. 解放军信息工程大学, 2014:1-5.

[8]王勇, 刘严萍, 李江波, 等.水汽和风速对雾霾中PM2.5/PM10变化的影响[J]. 灾害学, 2019,30(1): 3.

[9]卢文, 王红磊, 朱彬, 等. 南京江北2014-2016年PM2.5质量浓度分布特征及气象和传输影响因素分析 [J]. 环境科学学报, 2019, 39(04): 1039-1048.

[10]王鸥, 何秉宇.乌鲁木齐市空气质量5参数对PM2.5的影响及PM2.5浓度时空变化特征分析[J]. 新疆环境保护, 2018, 40(01): 6-11.

[11]李国平.地基GPS遥感大气可降水量及其在气象中的应用研究[D].成都:西南交通大学,2007:1-7.

[12]WILLIAMS S, BOCK Y, FANG P. Integrated satellite interferometry: Tropospheric noise, GPS estimates and implications for interferometric synthetic aperture radar products [J]. Journal of Geophysical Research, 1998, 103(B11): 27051.

[13]BEVIS M, BUSINGER S, HERRING T A, et al. GPS meteorology: Remote sensing of atmospheric water vapor using the global positioning system [J]. Journal of Geophysical Research: Atmospheres, 1992, 97(D14): 15787-15801.

[14]ROCKEN C, HOVE T V, JOHNSON J, et al. GPS/STORM—GPS Sensing of Atmospheric Water Vapor for Meteorology [J]. Journal of Atmospheric Oceanic Technology, 1995, 12(3): 468-478.

[15]MEERTENS C, MANAGER F, UNAVCO. Support of UNAVCO Community and Facility Activities [J]. AGU Fall Meeting Abstracts, 2002: 4-6.

[16]AONASHI, KAZUMASA, IWABUCHI, et al. Statistical Study on Precipitable Water Content Variations Observed with Ground-Based Microwave Radiometers (1.Ground-Based GPS Meteorology) [J]. Journal of the Meteorological Society of Japan Ser II, 2004, 82(1): 269-275.

[17]EMARDSON T R, JOHANSSON J, ELGERED G. The systematic behavior of water vapor estimates using four years of GPS observations [J]. Geoscience Remote Sensing IEEE Transactions on, 2000, 38(1): 324-329.

[18]NIELL A E, SPACE. Improved atmospheric mapping functions for VLBI and GPS [J]. Earth, Planets, 2014: 14-16.

[19]MACIAS-VALADEZ D, COCARD M, SANTERRE R. 3D modeling of the tropospheric refractivity using a permanent GPS network [J]. Geomatica, 2007, 61(4): 445-453.

[20]BOHM J, NIELL A, TREGONING P, et al. The Global Mapping Function (GMF): A new empirical mapping function based on numerical weather model data [J]. Geowissenschaftliche Mitteilungen, 2007, (80):1-2.

[21]HAASE J, GE M, VEDEL H, et al. Accuracy and variability of GPS tropospheric delay measurements of water vapor in the western Mediterranean [J]. Journal of Applied Meteorology, 2009, 42(11): 1547-1568.

[22]ADAMS D K, FERNANDES R M, KURSINSKI E R, et al. A dense GNSS meteorological network for observing deep convection in the Amazon [J]. Atmospheric Science Letters, 2011, 12(2): 207-212.

[23]SHOJI Y, MASHIKO W, YAMAUCHI H, et al. Estimation of Local-Scale Precipitable Water Vapor Distribution Around Each GNSS Station Using Slant Path Delay: Evaluation of a Severe Tornado Case Using High-Resolution NHM [J]. Scientific Online Letters on the Atmosphere Sola, 2015, 11: 31-35.

[24]SHOJI Y, SATO K, YABUKI M, et al. Comparison of shipborne GNSS-derived precipitable water vapor with radiosonde in the western North Pacific and in the seas adjacent to Japan [J]. Earth, Planets Space, 2017, 69: 1-13.

[25]ALSHAWAF F, FERSCH B, HINZ S, et al. Water vapor mapping by fusing InSAR and GNSS remote sensing data and atmospheric simulations [J]. Hydrology Earth System Sciences, 2015, 19(12): 4747-4764.

[26]FERNANDES R M S, VITERBO P, BOS M S, et al. NUVEM-‐New methods to Use gnss water Vapor Estimates for Meteorology of Portugal[C]. proceedings of the AGU Fall Meeting Abstracts, 2014:15-16.

[27]OIGAWA M, REALINI E, TSUDA T. Study of water vapor variations associated with meso-γ scale convection: comparison between GNSS and non-hydrostatic model data [J]. SOLA, 2015, 11: 27-30.

[28]HEUBLEIN M, ALSHAWAF F, ZHU X X, et al. Sparsity-driven tomographic reconstruction of atmospheric water vapor using GNSS and InSAR observations [C]. proceedings of the EGU General Assembly Conference Abstracts, 2016:5-9.

[29]KAKIHARA H, YABUKI M, KITAFUJI F, et al. New calibration technique for water-vapor Raman lidar combined with the GNSS precipitable water vapor and the Meso-Scale Model [C].proceedings of the AGU Fall Meeting Abstracts, 2019:23.

[30]MANANDHAR S, LEE Y H, MENG Y S, et al. A simplified model for the retrieval of precipitable water vapor from GPS signal [J]. IEEE Transactions on Geoscience Remote Sensing, 2017, 55(11): 6245-6253.

[31]PéREZ-JORDáN W, CASTRO-ALMAZAN J, MUNOZ-TUNON C. Precipitable water vapour forecasting: a tool for optimizing IR observations at Roque de los Muchachos Observatory [J]. Monthly Notices of the Royal Astronomical Society, 2018, 477(4): 5477-5485.

[32]李国平, 黄丁发. GPS气象学研究及应用的进展与前景[J].气象科学, 2005, 25(6): 651-661.

[33]陈洪滨 吕. GPS测量中的大气路径延迟订正 [J]. 测绘学报, 1996, (02): 48-53.

[34]李国平, 黄丁发. GPS气象学研究及应用的进展与前景 [J]. 气象科学, 2005, (06): 651-661.

[35]李成才. GPS气象应用:水汽总量绝对值的直接解算 [J]. 气象科技, 1997, (03): 23.

[36]王小亚, 丁金才, 朱文耀, 等. 上海地区GPS/STORM试验与结果.1998年中国地球物理学会第十四届学术年会[C].中国浙江杭州, 1998:8-10.

[37]王小亚, 严豪健,丁金才. 地面GPS观测探测大气可降水汽量的方法和前景 [J]. 天文学进展, 1998, (02): 135-142.

[38]王小亚,严豪健,程宗颐,丁金才. 地面GPS探测大气可降水量的初步结果[J].大气科学,1999, (05): 605-612.

[39]王小亚 ,丁金才,严豪健.上海地区GPS/STORM试验与结果[J].全球定位系统,2000,(03):6-10.

[40]李延兴 , 胡新康, 何平,黄永详.用地基GPS观测站遥测大气含水量和可降雨量的理论基础与试验结果 [J]. 中国科学(A辑), 2000, (S1): 107-110.

[41]姚宜斌 , 郭健健, 张豹, 等. 湿延迟与可降水量转换系数的全球经验模型 [J]. 武汉大学学报(信息科学版), 2016, 41(01): 45-51.

[42]何亚东, 姚宜斌, 许超钤, 等. 我国西南地区PWV与降水量的周期特征及关联性分析 [J].测绘地理信息, 2016, 41(06): 44-48.

[43]ZHAO Q, YAO Y, YAO W, et al. GNSS-derived PWV and comparison with radiosonde and ECMWF ERA-Interim data over mainland China [J]. Journal of Atmospheric and Solar-Terrestrial Physics, 2019:18-22.

[44]姚宜斌, 赵庆志, 何亚东, 等. 基于水汽密度比例因子的三维水汽层析算法 [J]. 测绘学报, 2016, 45(03): 260-266.

[45]周聪林, 徐晟, 杨翼飞, 等.GNSS水汽反演资料在台风分析中的应用[J].南方国土资源,2018, (09): 39-43.

[46]LI M, LI W, SHI C, et al. Assessment of precipitable water vapor derived from ground-based BeiDou observations with Precise Point Positioning approach [J]. Advances in Space Research, 2015, 55(1): 150-162.

[47]朱恩慧, 杨力, 贾鹏志,等.暴雨天气中地基GNSS可降水量时序变化分析[J].导航定位学报, 2018, 6(1): 21-26.

[48]黄良珂, 吴丕团, 王浩宇, 等. 中国西南地区GPS大气水汽转换系数模型精化研究[J].大地测量与地球动力学, 2019, 39(03): 256-261.

[49]姚君. 环境空气中PM2.5自动监测方法比较及应用 [J]. 大科技, 2018, 000(023): 352-353.

[50]BOYNARD A, CLERBAUX C, CLARISSE L, et al. First simultaneous space measurements of atmospheric pollutants in the boundary layer from IASI: A case study in the North China Plain [J]. American Geophysical Union, 2014,41 (2):645-651.

[51]孟晓艳, 王瑞斌, 张欣, 等. 2006-2010年环保重点城市主要污染物浓度变化特征[J].环境科学研究, 2012, 25(6):623-625.

[52]FANG C, WANG Z, XU G J J O G S. Spatial-temporal characteristics of PM 2.5 in China: A city-level perspective analysis [J]. Journal of Geographical Sciences, 2016, 26: 1519-1532.

[53]丁冰, 陈健, 王彬, 等. 城市环境 PM2.5 空间分布监测方法研究进展 [J]. 地球与环境, 2016,44(1): 130-138.

[54]徐祥德, 周秀骥, 施晓晖. 城市群落大气污染源影响的空间结构及尺度特征 [J]. 中国科学: D 辑, 2005, 35(A01): 1-19.

[55]PROVENçAL S, BUCHARD V, DA SILVA A M, et al. Evaluation of PM2.5 surface concentration simulated by Version 1 of the NASA’s MERRA Aerosol Reanalysis over Israel and Taiwan [J]. Aerosol air quality research, 2017, 17(1): 253.

[56]于文金, 于步云, 谢涛, 等.基于MODIS气溶胶光学厚度与气象要素的苏锡常地区PM 2.5 地面浓度分布研究 [J]. 环境科学学报, 2019, 36(10): 3535-3542.

[57]聂晨晖, 潘骁骏, 金洪芳. 杭州地区 2015 年 PM2.5 浓度时空变化特征分析 [J]. 测绘通报, 2019, (11): 75-79.

[58]陶金花, 张美根, 陈良富,等.一种基于卫星遥感AOT估算近地面颗粒物的方法[J].中国科学:地球科学, 2013, 43(01): 143-154.

[59]ZANG Z, WANG W, YOU W, et al. Estimating ground-level PM2. 5 concentrations in Beijing, China using aerosol optical depth and parameters of the temperature inversion layer [J]. Science of the total environment, 2017, 575: 1219-1227.

[60]XIN J, GONG C, LIU Z, 等. The observation‐based relationships between PM2. 5 and AOD over China [J]. Journal of Geophysical Research: Atmospheres, 2016, 121(18): 10701-10716.

[61]薛文博, 武卫玲, 付飞, 等.中国2013年1月PM2.5重污染过程卫星反演研究[J].环境科学, 2015, (3): 7.

[62]RAMOS Y, ST-ONGE B T, BLANCHET J P, et al. Spatio-temporal models to estimate daily concentrations of fine particulate matter in Montreal: Kriging with external drift and inverse distance-weighted approaches [J]. J Expo Sci Environ Epidemiol, 2016:266-267.

[63]程兴宏, 刁志刚, 胡江凯, 等. 基于CMAQ模式和自适应偏最小二乘回归法的中国地区PM2.5浓度动力-统计预报方法研究 [J]. 环境科学学报, 2020:5-8.

[64]薛文博, 武卫玲, 付飞, 等. 中国2013年1月PM2.5重污染过程卫星反演研究 [J]. 环境科学, 2015, 36(03): 794-799.

[65]VILLE-VEIKKOPAUNU, NIKOKARVOSENOJA, KAARLEKUPIAINEN,et al. Validation of PM2.5 Concentrations Based on Finnish Emission—Source-Receptor Scenario Model [M]. Air Pollution Modeling and its Application XXV, 2018:23.

[66]黄仁东, 仝慧贤, 刘抗, 等. 西安市PM2.5污染特征及其估测模型 [J]. 环境工程学报, 2015, 9(06): 2974-2978.

[67]刘严萍, 王勇, 赖迪辉. 基于 PM10 与气态污染物的北京市 PM2. 5 浓度模型研究 [J]. 灾害学, 2016, 31(2): 116-118.

[68]王勖之, 曾沛, 刘永辉.上海市PM2.5浓度的分析与预测 [J]. 数学的实践与认识, 2017, 47(15): 210-217.

[69]朱亚杰, 李琦, 侯俊雄, 等. 运用贝叶斯方法的PM2.5浓度时空建模与预测 [J]. 测绘科学, 2021, 41(2): 44-48.

[70]陈宁, 毛善君, 李德龙, 等. 多基站协同训练神经网络的PM2.5预测模型 [J]. 测绘科学, 2019, 43(7): 87-93.

[71]罗宏远, 王德运, 刘艳玲, 等. 基于二层分解技术和改进极限学习机模型的PM2.5浓度预测研究 [J]. 系统工程理论与实践, 2018, 38(5): 10.

[72]尹建光, 彭飞, 谢连科, 等. 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 [J]. 环境科学学报, 2018, 38(5): 9.

[73]郭建平, 吴业荣, 张小曳, 等. BP网络框架下MODIS气溶胶光学厚度产品估算中国东部PM2.5 [J]. 环境科学, 2013, 34(3): 9.

[74]王勇, 刘严萍, 李江波, 等. GPS和无线电探空的水汽变化与PM2.5/PM10变化的相关性研究 [J]. 武汉大学学报：信息科学版, 2016, 41(12): 6.

[75]王勇, 郝振航, 任栋, 等. 中国大陆地区GNSS对流层延迟与PM2.5浓度的相关性研究 [J]. 灾害学, 2017, 32(4): 5.

[76]姚宜斌.基于小波相干的雾霾与GNSS对流层延迟相关性分析[J].武汉大学学报:信息科学版, 2018, 43(12):8.

[77]谢劭峰, 靳利洋, 张朋飞, 等. 雾霾过程中AQI与ZTD,PM2.5与PWV的时序特征分析 [J]. 科技通报, 2017, 33(4): 4.

[78]曹玉静. 地基GPS技术遥感大气水汽含量及其在气象学中的应用 [D]. 南昌大学, 2008:3.

[79]高扬骏, 杨力, 吕志伟. 雾霾期间PWV对PM2.5的影响趋势分析及预测[C].第九届中国卫星导航学术年会论文集——S01 卫星导航应用技术. 2019:456-458.

[80]李峰. GAMIT/GLOBK与Bernese软件在高精度GPS数据处理中的应用分析 [D]. 山东科技大学, 2011:19.

[81]范朋飞. 高精度GPS站点坐标时间序列分析与应用[D]. 长安大学, 2013:23-25.

[82]鄂栋臣, 詹必伟, 姜卫平, 等. 应用GAMIT/GLOBK软件进行高精度GPS数据处理[J]. 极地研究, 2015, (03): 173-182.

[83]赵庆志. 地基GPS大气可降水量估计及水汽层析研究 [D]. 中国矿业大学, 2014:33.

[84]王勇, 刘严萍. 地基 GPS 气象学原理与应用硏究 [M]. 测绘出版社, 2012:45-47.

[85]刘焱雄, 刘经南. 利用地面气象观测资料确定对流层加权平均温度 [J]. 武汉测绘科技大学学报, 2000, (05): 400-404.

[86]杨露华, 丁金才, 陆艳, 等. GPS/PWV资料在上海地区强对流预报中的应用研究[C].中国气象学会2007年年会, 中国广东广州.2007:236-237.

[87]李建国, 李成才,夏青. 使用全球定位系统遥感水汽分布原理和中国东部地区加权"平均温度"的回归分析[J]. 气象学报, 2019, (03): 28-37.

[88]马下平,李秦政,陈鹏.基于探空数据的对流层延迟模型精度评估[J].大地测量与地球动力学2020,40(2):123-128.

[89]闫小霞,李薇,张宝成,等.基于ECMWF再分析地表资料计算中国区域ZTD的精度分析[J].大地测量与地球动力学, 2019, 39(11):6.

[90]李延兴. GPS测量大气折射模型研究[C].97全国GPS及DGPS应用学术交流会, 西安. 1997:15-17.

[91]孟昊霆,张克非,杨震,等.GPT2/GPT2w+Saastamoinen模型ZTD估计的亚洲地区精度分析[J].测绘科学,2020,45 (8):70-76.

[92]HOCHREITER S, SCHMIDHUBER J. Long Short-Term Memory [J]. Neural Computation, 2018, 9(8): 1735-1780.

[93]FISCHER T, Krauss C. Deep learning with long short-term memory networks for financial market predictions [J]. European Journal of Operational Research, 2017, 270(2):8-12.

[94]J. F-L, K. K, C. M, et al. A deep encoder-decoder for surrogate modelling of liquid moulding of composites[J]. Engineering Applications of Artificial Intelligence, 2023, 120:4-6.