当前，世界百年未有之大变局加速演进，国际权力结构正发生深刻重组，新的变革动荡期已经到来，地缘政治风险也随之成为影响全球发展的5大风险之一。面对“一带一路”沿线复杂且严峻的地缘政治风险问题，强化风险评估与防控已成为“一带一路”高质量发展的重要政策导向。然而，当前地缘政治风险研究中关于通过大数据开展的量化评估与监测预警研究并不充分。如何对具有非线性特征和复杂演化特征的地缘政治风险进行定量模拟与科学评价？如何分国别实现地缘政治风险的动态连续监测？在动态监测的同时能否实现及时预警？这些已成为面向地缘政治风险研究亟待解决的问题。对地缘政治风险进行不同尺度下的量化与评估，揭示地缘政治风险演化过程与特征，实现分国别动态监测地缘政治风险并及时预警，能够为准确判断沿线国家的地缘政治安全坏境提供数据支撑，为完善安全风险防范体系提供科学参考，顺应“一带一路”高质量发展的政策需要。

本研究以“一带一路”沿线国家为目标区域，遵循“构建不同尺度下的地缘政治风险量化指标→分析与评估沿线国家的地缘政治安全环境→探索分国别动态监测与及时预警地缘政治安全态势的实现路径”的研究思路，在地理学研究方法不断革新的背景下，丰富与拓展关于地缘政治风险的研究方法，旨在探究利用大数据对“一带一路”沿线国家的地缘政治风险进行合理量化、科学评估以及监测预警。通过比较与分析数据类型与数据库的可靠性，选择全球事件、语言与语调数据库（Global Database of Events, Language and Tone，GDELT）作为数据来源。本研究首先从地缘环境的本底要素出发，梳理“一带一路”沿线国家的地缘环境与地缘战略价值；其次，针对不同空间尺度提出地缘政治风险定量表达的方法，包括地区尺度下的国家间地缘政治风险指数（B_GRI）和国家尺度下的国家地缘政治风险指数（C_GRI）及其2个子类（国内地缘政治风险指数C_GRIdom和国际地缘政治风险指数C_GRIint），分析“一带一路”沿线国家间地缘政治风险网络，对分国别地缘政治风险进行等级划分与评估，总结需要重点关注的沿线国家；再次，基于复杂科学中的随机分形理论，提出可以实现动态监测地缘政治风险的方法与指数（M_GRI），对需要重点关注的沿线国家开展动态监测研究；最后，建立并训练极度梯度提升树（XGBoost）和长短时记忆神经网络（LSTM）机器学习算法模型，确定可用于短期预警地缘政治风险的模型，并进行预测。主要研究结论如下：

（1）“一带一路”沿线国家的地缘环境特征及地缘战略价值在自然环境与资源、政治与战略、经济与贸易以及人口与宗教4个方面均存在较大差异，这种差异也从深层次影响着沿线国家的地缘政治格局，在带来地缘战略价值的同时，也会引发一定的地缘政治风险。

（2）在地区尺度下，本研究提出的B_GRI指数能够对“一带一路”沿线国家间地缘政治风险强度进行有效刻画，通过分析2013至2022年沿线国家间地缘政治风险网络发现：①网络规模整体上未发生较大的变化，逐渐显现小世界性特征，并服从无标度特征；②地理空间格局演化具有显著的空间异质性与地理邻近性，强度较高的风险联系主要产生于西亚北非区内部沿线国家之间以及俄罗斯与中东欧区沿线国家之间；③社团分组数量逐渐增加，印证了世界正逐渐走向“分裂”的趋势，社团划分情况与地理分区的划分大致吻合；④沿线国家间地缘政治风险网络不具有核心-边缘结构；⑤节点两极分化严重，且差异程度有所增加，“地缘政治风险重心国”包括俄罗斯、叙利亚、以色列、土耳其和伊朗，其影响广度和深度在网络中都有着举足轻重的作用；⑥与中国产生地缘政治风险联系的沿线国家主要分布在中国周边地区，地缘政治风险强度始终处于一个合理可控的范围内。

（3）在国家尺度下，本研究提出的C_GRI、C_GRIdom与C_GRIint指数能够对“一带一路”沿线国家分国别的地缘政治风险强度进行有效刻画，通过评估2013至2022年“一带一路”沿线国家地缘政治风险发现：①从整体特征来看，地缘政治风险强度整体不高但级别差异性显著，多数沿线国家仍面临着难以忽视的国内安全困境，而在国际交往中仍然是以友好合作为主线；②从时间演化特征来看，C_GRI、C_GRIdom和C_GRIint强度在不同分区之间呈现差异性，其中C_GRIint的差异性显著，C_GRIdom的差异性较低；③从空间演化特征来看，C_GRI与C_GRIdom均呈现“高风险集聚性”与“南高北低”的空间分异特征，高风险集聚区主要分布在西亚北非区、南亚区以及东南亚区，其中，C_GRI的高风险集聚区在西亚北非区内“东进”趋势明显，C_GRIdom的高风险集聚区范围较大且“北进”至中东欧区的趋势明显，此外，C_GRIint空间变化显著，但仍呈现“高风险集聚性”，高风险集聚区范围较小且沿着南亚区和西亚北非区→西亚北非区→西亚北非区和俄罗斯、乌克兰的方向演进；④叙利亚、伊拉克、也门、巴勒斯坦、以色列、阿富汗、黎巴嫩、缅甸、印度、巴基斯坦与乌克兰的地缘政治风险级别常年处于高风险和中高风险，是影响国家尺度下“一带一路”沿线地缘政治安全与形势的重要国家。

（4）本研究提出的分国别动态监测地缘政治风险的方法与M_GRI指数能够进行科学、有效地动态监测，基于地区尺度和国家尺度下“一带一路”沿线国家地缘政治风险的研究结果，聚焦俄罗斯、土耳其、伊朗、叙利亚、伊拉克、也门、巴勒斯坦、以色列、阿富汗、黎巴嫩、缅甸、印度、巴基斯坦与乌克兰共14个沿线国家，运用M_GRI指数开展动态监测研究发现：①2021年至2022年14个重点沿线国家均面临着具有持续性的地缘政治风险；②M_GRI趋势显示在一些重大风险事件发生前已有预兆，例如2022年2月俄乌冲突之前，俄罗斯与乌克兰的M_GRI数值自2021年底开始便呈现着显著上升的趋势，证明了M_GRI时间序列的趋势变化具有一定的有效性和重要现实意义；③选择阿富汗与缅甸作为案例分析，发现M_GRI时间序列在更长时间尺度下也可以有效监测与识别重大风险事件，趋势变化能与现实充分对应，进一步证明了使用上述方法进行分国别动态监测地缘政治风险的可靠性与现实意义。

（5）本研究建立的分国别短期预警地缘政治风险的方法，能够在动态监测地缘政治风险的同时及时预警，对地缘政治风险态势做出合理的预判，研究发现：①俄罗斯、土耳其、叙利亚、也门、以色列、阿富汗、黎巴嫩、缅甸和印度在未来有重大风险事件发生的可能性增加，需要加以关注；伊朗、伊拉克、巴勒斯坦、巴基斯坦和乌克兰发生负面事件的连续性减弱，处于“调整期”；②选择阿富汗与缅甸作为案例分析，发现其预测结果与现实相对应，进一步证明了本研究所构建的预测模型可以有效地及时预警地缘政治风险。

At present, the world is undergoing a profound reorganization of the international power structure, and the world is entering a new period of change and turbulence. As an initiative to guide new changes in the world and an international open platform to benefit the world, the Belt and Road Initiative (BRI) will trigger the global geo-plate linkage resonance. As geopolitical risks have become one of the top 5 risks affecting global development, in the face of complex geopolitical security issues along the Belt and Road, strengthening risk assessment and prevention have become an important policy direction for the high-quality development of BRI. However, researches about assessment and monitoring of geopolitical risk based on big-data were not enough. How to model and evaluate geopolitical risks with non-linear characteristics and complex evolutionary features? How to achieve monitoring of geopolitical risks by country? Can monitoring be accompanied by predictive analysis of future geopolitical risks in the short term? These have become urgent issues for geopolitical risk research. Quantification and assessment of geopolitical risk at different scales, revealing the evolutionary process and characteristics of geopolitical risks, realizing dynamic monitoring of geopolitical risks by country and timely warning, could provide data support to judge the geopolitical security of countries along the Belt and Road (BRI countries), scientific reference to improve the security risk prevention system, and meet the policy needs of the high-quality development of BRI.

This study took the BRI countries as research region, based on the line of “using big-data to quantify indicators of geopolitical risks at different scales→assessing the geopolitical security of BRI countries→exploring the realization path of dynamic monitoring and timely early warning of geopolitical risks”. This study selected the Global Database of Events, Language and Tone (GDELT) as the data resource. First, according to the fundamental elements of geo-setting, this study sorted out the geo-setting and geo-strategic value of BRI countries. Second, this study provided Bilateral Geopolitical Risk Index (B_GRI) at the regional scale, National Geopolitical Risk Index (C_GRI) at the national scale and its 2 subcategories, National Geopolitical Risk Index from Domestic perspective (C_GRIdom) and National Geopolitical Risk Index from international perspective(C_GRIint). This study further analyzed interstate geopolitical risk networks of BRI countries, assessed the national geopolitical risks, and concluded the BRI countries on which needed to be focused. Third, according to the fractal theory, this study provided the Geopolitical Risk Index for Monitoring (M_GRI), in order to monitor the important BRI countries. Finally, this study built and trained XGBoost and LSTM to determine the best model to achieve the early warning of geopolitical risks. The main findings of the study are as follows:

(1) There existed big differences among the geo-setting and geo-strategic values of BRI countries, in terms of natural environment and resources, politics and strategy, economy and trade, population and religion. Differences affected the geopolitical structure along the Belt and Road, which would deliver geo-strategic values and raise geopolitical risks.

(2) B_GRI could measure the geopolitical risks between different BRI countries. This study analyzed the interstate geopolitical risk networks of BRI countries in the period of 2013 to 2022, and found: ①The network size did not change significantly, had small-world characters and scale-free features. ②The evolution of geographic space pattern had the features of spatial heterogeneity and geographical proximity. The high B_GRI existed between BRI countries in West Asia and North Africa, and Russia and BRI countries in Central and Eastern Europe. ③The number of network communities increased, which proofed the divisive trend of the world. And the community detection roughly coincided with the geographical divisions. ④The network did not have core-periphery structure. ⑤The nodes of BRI countries were heavily polarized, with the increasing degree of variation. BRI countries with ‘the Gravity of Geopolitical Risks’ included Russia, Syria, Israel, Turkey and Iran, which played an important role in the network. ⑥BRI countries which had geopolitical risks with China existed around surrounding areas, and the intensity of geopolitical risks between those remained in a manageable range.

(3) C_GRI、C_GRIdom与C_GRIint could measure the geopolitical risks for each BRI countries. This study assessed the geopolitical risks of BRI countries in the period of 2013 to 2022, and found: ①From overall perspective, Geopolitical risk intensity was low, but the variability was significant. This indicated that most of BRI countries still faced a domestic security dilemma, and remained a friendship line among the international communications. ②From the perspective of time evolution, C_GRI, C_GRIdom and C_GRIint had divergences among different BRI regions. C_GRIint had differences with significance, C_GRIdom had differences with no significance. ③From the perspective of spatial evolution, C_GRI and C_GRIdom had ‘high geopolitical risk clustering’ and spatial heterogeneity of ‘high in the south and low in the north’. The clustering existed in West Asia and North Africa, South Asia, Southeast Asia. The clustering of C_GRI in West Asia and North Africa had a trend towards east, and the clustering of C_GRIdom had large scale and a trend toward to Central and East Europe. Besides, C_GRIint changed significantly from spatial revolution, and still had ‘high geopolitical risk clustering’. But, its clustering had small scale and a trend from West Asia and North Africa to Russia and Ukraine. ④There were several BRI countries which had high and very high geopolitical risks, including Syria, Iraq, Yemen, Palestine, Israel, Afghanistan, Lebanon, Myanmar, India, Pakistan and Ukraine. These BRI countries played an important role in geo-security along the Belt and Road.

(4) M_GRI could monitor the changes of geopolitical risk of BRI countries reasonably. Based on the findings, this study focused on Russia, Turkey, Iran, Syria, Iraq, Yemen, Palestine, Israel, Afghanistan, Lebanon, Myanmar, India, Pakistan and Ukraine, analyzed the M_GRI, and found: ①These BRI countries had geopolitical risks with continuity in the period of 2021 and 2022. ②The trend of M_GRI had omen before some key events, such as the stage before Russia-Ukraine Conflict happened, M_GRI of Russia and Ukraine had significant increased trend from the bottom of 2021. This point could prove the validity and practical significance of M_GRI. ③This study selected Afghanistan and Myanmar as case studies, which also proofed the reliability of M_GRI on a longer time scale.

(5) The predictive model which this study built and trained could realize the early warning while monitoring the geopolitical risks, and found: ①Russia, Turkey, Syria, Yemen, Israel, Afghanistan, Lebanon, Myanmar and India had possibilities of key events happening, which needed to be focused on. Iran, Iraq, Palestine, Pakistan and Ukraine had decreased continuity of key events, which were in an adjustment period. ②This study selected Afghanistan and Myanmar as case studies, which indicated that the predictive value could correspond to the reality. This could proof the validity of predictive model to make an early warning.