Cross-border logistics demand prediction method integrated with geographical risk correction

Abstract

The invention provides a cross-border logistics demand prediction method integrated with geographical risk correction, and belongs to the technical field of logistics information, and the method comprises the steps: according to the prediction period of the cross-border logistics demand of a target port, calculating the total amount of foreign trades and the whole-course generalized trade cost based on the effective economic abdominal region of the target port; calculating a reference logistics demand value through a pre-constructed scene-based trade gravitation reference model; based on the multi-level structured risk quantification system and the sudden geographic risk, obtaining a geographic risk feature vector corresponding to the prediction period; and fusing the reference logistics demand value and the geographical risk feature vector to obtain a combined feature vector, inputting the combined feature vector into a pre-trained cross-border logistics demand fusion prediction model, and outputting a final logistics demand prediction value. According to the method, the problems of low cross-border logistics demand prediction precision, poor robustness and response lag caused by insufficient port applicability, lack of dynamic risk quantification and lack of an automatic response mechanism of a macroscopic trade prediction model in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of logistics information technology, and in particular relates to a cross-border logistics demand forecasting method that incorporates geopolitical risk correction. Background Technology

[0002] As key nodes in international trade and supply chains, accurate forecasting of cross-border logistics demand at ports is crucial for port operations and supply chain management. Existing forecasting methods mainly fall into two categories, but both have significant limitations: one type relies on macroeconomic trade statistics, which can predict bilateral trade flows at the national level but cannot characterize the actual economic hinterland characteristics of a specific port. This leads to a severe disconnect between macroeconomic trade forecasts and actual port operations, resulting in low accuracy and limited guidance value in practical applications. The other type is time-series forecasting models based on the port's own historical data, such as time series analysis or LSTM neural networks. These models extrapolate by learning historical patterns, but their accuracy drops sharply when faced with uncertain geopolitical risks such as sudden trade adjustments that deviate from historical patterns. Furthermore, existing cross-border logistics demand forecasting generally lacks automated real-time response mechanisms for sudden risks, relying mostly on manual assessment and adjustments. This results in slow response times and high subjectivity, making it difficult to quantify and respond to sudden geopolitical risks such as abrupt changes in cross-border trade regulations or channel disruptions. When sudden geopolitical risk events deviate from historical cases, the accuracy of cross-border logistics demand forecasting drops rapidly. Summary of the Invention

[0003] To address the aforementioned shortcomings in existing technologies, this invention provides a cross-border logistics demand forecasting method that incorporates geopolitical risk correction. This method can dynamically perceive and quantify geopolitical risks, enabling accurate forecasting of cross-border logistics demand. It solves the problems of low accuracy, poor robustness, and delayed response in cross-border logistics demand forecasting caused by the insufficient applicability of macro-trade forecasting models to specific ports, the lack of dynamic risk quantification, and the absence of automated response mechanisms in existing technologies.

[0004] To achieve the above objectives, the technical solution adopted by this invention is: a cross-border logistics demand forecasting method incorporating geopolitical risk correction, comprising the following steps: Based on the forecast period of cross-border logistics demand at the target port, and the total foreign trade volume of the effective economic hinterland of the target port and the overall generalized trade cost, the benchmark logistics demand value is calculated through a pre-constructed scenario-based trade gravity benchmark model. Based on the forecast period of cross-border logistics demand at the target port, and using a multi-level structured risk quantification system and sudden geopolitical risks, the geopolitical risk characteristic vector corresponding to the forecast period is obtained. By fusing the baseline logistics demand value and the geopolitical risk feature vector, a combined feature vector is obtained; The combined feature vectors are input into a pre-trained cross-border logistics demand fusion prediction model, which outputs the final logistics demand prediction value and optimizes the pre-trained cross-border logistics demand fusion prediction model.

[0005] Existing cross-border logistics demand forecasting often directly applies macro-level trade gravity models or time-series forecasting models based on historical port data. While trade gravity models can predict the total bilateral trade volume at the national level, they struggle to describe the effective economic hinterland of a specific port's logistics, leading to significant discrepancies between forecast results and actual port operations. This fails to provide practical and effective guidance for trade logistics at specific ports. Time-series forecasting models based on historical port data, while capable of accurately forecasting cross-border logistics demand based on historical data, struggle to quantify sudden geopolitical risk events such as port closures and channel disruptions. When these events disrupt historical patterns, the lack of automated real-time response mechanisms actually reduces forecast accuracy. This invention provides a cross-border logistics demand forecasting method incorporating geopolitical risk correction. By constructing a scenario-based trade gravity benchmark model, the forecasting target is fixed at the effective economic hinterland of the port at the national level. Using the total foreign trade volume and the overall generalized trade cost as core variables, a stable benchmark logistics demand reflecting long-term trends is calculated. Simultaneously, a structured risk quantification system is designed, including node risk, channel risk, and industry dependence risk, to achieve dynamic assessment of normalized risks. A dual-path emergency risk response mechanism, combining rule-based risk matching and historical case similarity matching, is introduced to quantify and correct for sudden geopolitical risk events, updating the risk feature vector and reducing the impact of sudden risk events on cross-border logistics demand forecasting results. Finally, the benchmark demand value is fused with the real-time updated risk feature vector and input into a pre-trained machine learning model for cross-border logistics demand forecasting. This invention achieves high-precision, highly robust, and adaptive cross-border logistics demand forecasting, enhancing the scientific rigor and resilience of port logistics supply decisions.

[0006] Furthermore, the expression for the pre-built scenario-based trade gravity benchmark model is as follows:

[0007]

[0008]

[0009] in, As the baseline logistics demand value, , and All are weighted parameters. This refers to the total trade volume exclusive to a port, which is the sum of the total foreign trade volume of all sub-regions within the effective economic hinterland through that port. The weighted average generalized trade cost is the total value of goods used in the port-specific trade. For the first Foreign trade volume of each sub-region through target ports for The corresponding total generalized trade costs, The total number of subdivided regions within the effective economic hinterland.

[0010] To address the problem that existing trade gravity models, which focus on national GDP and thus fail to effectively describe the logistics demand of a specific port, this invention transforms the prediction target into the effective economic hinterland of the port, using its total foreign trade volume. and weighted average total generalized trade costs Using [variable name] as the core variable, a scenario-based trade gravity benchmark model was constructed, which enabled the prediction of cross-border logistics demand in specific port scenarios and output stable and interpretable logistics benchmark demand prediction results, thereby improving the accuracy of the final logistics demand prediction value.

[0011] Furthermore: the multi-level structured risk quantification system includes: Node risk is used to quantify the operational stability of domestic ports and related overseas ports, as well as the risks associated with the cross-border trade regulatory environment. Corridor risk is used to quantify the reliability of domestic connections between domestic ports and economic hinterlands, as well as cross-border transport corridors. Industry dependence risk is used to quantify the vulnerability of hinterland industries to specific international markets; The underlying quantitative indicators of node risk include: the percentage of annual operational interruption time at domestic and foreign ports, the frequency of changes in cross-border trade regulations in the region where the port is located, and the GPR index based on the Global Peace Index (GPI). The underlying quantitative indicators of the corridor risk include: the number of traffic interruptions per year on key road sections, the number of days the corridor is shut down due to extreme weather, and the failure rate of multimodal transport connections. The underlying quantitative indicators of industry dependence risk include: the proportion of a specific industry's exports to the target market, the industry loss value after a simulated trade barrier is triggered, and the production capacity proportion of alternative suppliers in the target market.

[0012] To address the problem that existing cross-border logistics demand forecasting lacks real-time response capabilities and struggles to quantify complex geopolitical risks related to port logistics, this invention proposes a multi-level structured risk quantification system based on node risk, channel risk, and industry-dependent risk. It designs objectively monitorable underlying quantitative indicators for each level of risk, transforming various logistics risks into calculable numerical features. This solves the problem of difficulty in risk quantification and generates structured risk feature vector inputs that can be directly processed by machine learning models, providing a basis for accurate prediction correction.

[0013] Further: the acquisition of the geopolitical risk feature vector corresponding to the prediction period specifically includes: We weighted and fused the underlying quantitative indicators of node risk, channel risk, and industry dependence risk to obtain the median values ​​of node risk, channel risk, and industry dependence risk. The median values ​​of node risk, channel risk, and industry dependence risk are standardized to obtain standardized node risk values, standardized channel risk values, and standardized industry dependence risk values, respectively. Based on standardized node risk values, standardized channel risk values, and standardized industry dependence risk values, combined with sudden geopolitical risks, a geopolitical risk feature vector is constructed.

[0014] In cross-border logistics and trade at ports, different geopolitical risk events have different indicator dimensions. For example, the time of operational interruption is measured in hours, the frequency of changes in cross-border trade regulations is measured in times, and the GPR index is a dimensionless score. The difference in dimensions makes it impossible to directly integrate multi-dimensional risk events such as node risk, channel risk, and industry dependence risk, making it difficult to form a unified and effective input for risk assessment. This invention transforms the underlying indicators of each dimension into standardized risk values ​​by weighting, integrating, and standardizing them, thereby constructing a feature vector that can reflect complex risk events.

[0015] Furthermore, the aforementioned combination of sudden geopolitical risks specifically includes: When a sudden geopolitical risk event occurs, an update operation is performed based on the risk type of the sudden geopolitical risk event to correct the geopolitical risk feature vector; the risk types include node risk, channel risk, and industry dependence; The update operation includes: matching the risk type with a pre-set emergency risk rule base; if the match is successful, calculating the correction value based on the corresponding quantitative rule, and updating the geopolitical risk feature vector of the corresponding risk type according to the correction value; If a match fails, similar historical emergency risk cases are retrieved from the historical emergency risk event database based on similarity calculation. Based on the risk values ​​of the retrieved cases, an updated suggested value is generated, and the geopolitical risk feature vector of the corresponding risk type is updated.

[0016] Existing time-series forecasting models based on historical data cannot automatically identify and quickly update risk characteristics in the face of sudden risk events, resulting in a large discrepancy between the predicted cross-border logistics demand at ports and the actual logistics demand. This invention proposes a dual-path dynamic response scheme based on matching a preset rule base with historical sudden risk event cases. When a sudden geopolitical risk event occurs, it is first matched with the preset risk rule base. If the match is successful, the corresponding quantification rules are automatically executed to achieve rapid quantification of the sudden risk event and update of its feature vector. For sudden risk events that fail to match, i.e., unknown events not covered by the rule base, the most similar past events in the historical case base are matched using a cosine similarity algorithm, and update suggestions are generated based on their impact values, ensuring that adaptive quantification can also be achieved for unknown sudden risk events. This ensures that the model can respond to drastic changes in the external environment in real time, enhancing the dynamic adaptability and timeliness of the forecast.

[0017] Furthermore, the expression for calculating the similarity is as follows:

[0018] in, For sudden geopolitical risk events, For historical sudden risk events, To assess the similarity between sudden geopolitical risk events and historical sudden risk events, For sudden geopolitical risk events in the first Standardized values ​​of each feature dimension. For historical sudden risk events in the first Standardized values ​​of each feature dimension. This represents the total number of feature dimensions.

[0019] To address the difficulty in quantifying unknown and sudden risk events not covered by the rule base, this invention uses a cosine similarity algorithm to quantitatively compare the feature vectors of the current event with a historical case database. Based on the calculation results, the most similar historical risk event case is selected, and its impact value is referenced to provide a quantitative basis for unknown and sudden risk events, thereby enhancing the adaptability of logistics demand forecasting.

[0020] Furthermore, the pre-trained cross-border logistics demand fusion prediction model specifically includes: Acquire historical time series data, including historical true values ​​of logistics demand, historical benchmark logistics demand values, and historical geopolitical risk feature vectors; The fusion feature sequence, composed of the historical baseline logistics demand value from the previous moment to the current moment and the historical geopolitical risk feature vector, is used as input. The actual historical logistics demand value at the current moment is used as the prediction target. The long short-term memory network model is trained under supervision to obtain a pre-trained cross-border logistics demand fusion prediction model. The optimization of the pre-trained cross-border logistics demand fusion prediction model specifically includes: optimizing and updating the cross-border logistics demand fusion prediction model based on newly added logistics demand data and its corresponding logistics benchmark values ​​and risk feature vectors.

[0021] To address the problem that existing time-series forecasting models, which rely on historical data, cannot adapt to dynamic and unforeseen risk events and market changes in cross-border logistics, this invention proposes a continuous optimization method. By using historical actual logistics demand values, historical benchmark logistics demand values, and historical geopolitical risk feature vectors, the model can learn the complex changes in how risk events affect actual demand. Furthermore, by using verified actual data from historical events that have not yet occurred as new samples, the model undergoes incremental learning, enabling cross-border logistics demand forecasting results to adapt to the dynamic risk event environment and ensuring that forecasting capabilities continuously improve over time.

[0022] The beneficial effects of this invention are: This invention combines a scenario-based trade gravity benchmark model with a machine learning-based predictive model to achieve synergistic prediction of long-term trends and short-term fluctuations. The scenario-based trade gravity benchmark model, based on highly explanatory macroeconomic variables such as hinterland trade volume and broad trade costs, accurately depicts the long-term fundamental trends of port logistics. The machine learning model focuses on dynamic variables such as risk feature vectors, responsible for capturing and correcting short-term market fluctuations and abnormal shocks. This layered architecture retains both macroeconomic explanatory power and trend stability while incorporating the machine learning's ability to respond sensitively to complex short-term factors.

[0023] This invention constructs a systematic emergency risk handling mechanism. This mechanism uses a dual-path approach of rule matching and historical similarity matching to quickly transform abstract emergency risk events into structured, computable risk feature vectors, which are then fed back into the prediction process in real time. This enables the model to rapidly adjust its prediction output when faced with uncertainties such as sudden changes in trade regulations or channel disruptions, forming a closed loop of risk identification, quantification, and response. This enhances the resilience and decision-making timeliness of the prediction system in complex and ever-changing environments. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of a cross-border logistics demand forecasting method that incorporates geopolitical risk correction. Detailed Implementation

[0025] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0026] Example 1 like Figure 1 The diagram shown is a flowchart of a cross-border logistics demand forecasting method incorporating geopolitical risk correction. This invention provides a cross-border logistics demand forecasting method incorporating geopolitical risk correction, comprising the following steps: Based on the forecast period of cross-border logistics demand at the target port, and based on the total foreign trade volume of the effective economic hinterland of the target port and the generalized trade cost throughout the process, a scenario-based trade gravity benchmark model is established, and the benchmark logistics demand value is calculated. Based on the forecast period of cross-border logistics demand at the target port, and using a multi-level structured risk quantification system and sudden geopolitical risks, the geopolitical risk characteristic vector corresponding to the forecast period is obtained. By fusing the baseline logistics demand value and the geopolitical risk feature vector, a combined feature vector is obtained; The combined feature vectors are input into a pre-trained cross-border logistics demand fusion prediction model, which outputs the final logistics demand prediction value.

[0027] In one embodiment of the present invention, in the field of cross-border logistics demand forecasting, existing technologies often employ the classical trade gravity model at the national macro level, using variables such as the GDP of two countries and geographical distance to predict cross-border logistics demand. Its linear form is generally as follows:

[0028] in, for Guohe Bilateral trade flows between countries, including imports and exports in bilateral trade. for The country's gross domestic product, for The country's gross domestic product, for Guohe The geographical distance of the country , , and All are parameters; the classic trade gravity model is often used for trade forecasting at the national macro level. However, when facing micro-level port logistics scenarios, the national GDP cannot accurately reflect the actual cargo flow and trade sources of a specific port. It cannot determine the effective economic hinterland of the port, leading to a large deviation between the logistics demand forecast and the actual throughput of the port, and insufficient explanatory power. Therefore, this invention improves upon it by transforming the forecast target from bilateral trade flow between countries to the total cross-border logistics demand of the port's effective economic hinterland through the target port; it concretizes the national economic scale as the total foreign trade volume of the effective hinterland; and it deepens geographical distance into the generalized trade cost throughout the entire process. The expression of the pre-constructed scenario-based trade gravity benchmark model is as follows:

[0029]

[0030]

[0031] As the baseline logistics demand value, , and All are weighted parameters. This refers to the total trade volume exclusive to a port, which is the sum of the total foreign trade volume of all sub-regions within the effective economic hinterland through that port. The weighted average generalized trade cost is the total value of goods used in the port-specific trade. For the first Foreign trade volume of each sub-region through target ports for The corresponding total generalized trade costs, The total number of subdivided regions within the effective economic hinterland.

[0032] In one embodiment of this invention, existing cross-border logistics demand forecasting also employs time-series forecasting models based on historical data, such as the Autoregressive Integral Moving Average (ARIMA) model or the Long Short-Term Memory (LSTM) network. These models extrapolate forecasts by mining the time-series patterns of historical throughput data from ports, essentially learning and continuing historical patterns. However, cross-border logistics is highly susceptible to sudden and uncertain geopolitical risks, such as temporary closures of key ports, disruptions to transportation routes, and changes in trade agreements. Such events can instantly disrupt the stable patterns inherent in historical data, causing models relying on historical patterns to fail and exhibiting an imbalance between static logistics data and actual dynamic logistics. Existing methods typically treat these risks as unquantifiable noise or simply label them, lacking a systematic dynamic quantification and response mechanism. Therefore, this invention introduces a multi-level structured risk quantification system and sudden geopolitical risks. The multi-level structured risk quantification system performs risk analysis and specific quantification, enabling accurate identification and characterization of complex risks. Sudden geopolitical risks are used to perceive and correct specific events during actual logistics trade, achieving preparedness for uncertain events. This invention obtains a geopolitical risk feature vector corresponding to the forecast period, specifically including: We weighted and fused the underlying quantitative indicators of node risk, channel risk, and industry dependence risk to obtain the median values ​​of node risk, channel risk, and industry dependence risk. The median values ​​of node risk, channel risk, and industry dependence risk are standardized to obtain standardized node risk values, standardized channel risk values, and standardized industry dependence risk values, respectively. Based on standardized node risk values, standardized channel risk values, and standardized industry dependence risk values, combined with sudden geopolitical risks, a geopolitical risk feature vector is constructed.

[0033] In a specific embodiment of the present invention, the multi-level structured risk quantification system classifies the risk factors affecting cross-border logistics into first-level risks with clear logical boundaries, including: Node risk is used to quantify the operational stability of domestic ports and related overseas ports, as well as the risks associated with the cross-border trade regulatory environment. Corridor risk is used to quantify the reliability of domestic connections between domestic ports and economic hinterlands, as well as cross-border transport corridors. Industry dependence risk is used to quantify the vulnerability of hinterland industries to specific international markets; Each primary risk level is further defined with several observable and calculable underlying quantitative indicators. These include: the percentage of annual operational interruptions at domestic and international ports, the frequency of changes in cross-border trade regulations in the port's region, and the GPR index based on the Global Peace Index (GPI). The percentage of annual operational interruptions at domestic and international ports can be calculated from the port operation system logs, by calculating the cumulative duration of complete or partial port closures due to maintenance, accidents, or regulations within a year, divided by the port's planned total annual operating time. The frequency of changes in cross-border trade regulations in the port's region can be calculated from publicly available information sources such as authoritative information platforms, by calculating the number of trade, customs, or regulatory adjustments issued by the target port's country or region within a statistical period that have a substantial impact on cross-border logistics. The GPR index based on the Global Peace Index (GPI) can directly reference the GPI published by authoritative international organizations, with annual detailed scores for the port's country and even specific administrative regions.

[0034] The underlying quantitative indicators of corridor risk include: the number of annual traffic interruptions on key road sections, the number of days of corridor shutdown due to extreme weather, and the failure rate of multimodal transport connections. The number of annual traffic interruptions on key road sections can be obtained by integrating data from traffic management departments, logistics company reports, and news monitoring, and by counting the number of events that occurred within a year on core highways connecting the hinterland and ports and key overseas transport corridors, where traffic flow interruption time exceeded a certain threshold due to accidents, weather, etc. The number of days of corridor shutdown due to extreme weather and the failure rate of multimodal transport connections can be obtained by using historical data from meteorological departments and logistics operation records, and by counting the cumulative number of days within a year when the above-mentioned corridors were completely closed or severely restricted due to extreme weather events such as blizzards, floods, and typhoons.

[0035] The underlying quantitative indicators of industry dependence risk include: the export share of a specific industry to the target market, the simulated industry loss value after a trade barrier is triggered, and the capacity share of alternative suppliers in the target market. The export share of a specific industry to the target market can be obtained from detailed trade data released by statistical departments, which shows the export value of a specific key industry in the hinterland to the target international market, as well as the industry's total global exports. The calculation formula is: Export share = Export value to the target market / Total industry exports, directly reflecting market concentration. The simulated industry loss value after a trade barrier is a calculated value. Based on a historical case database of anti-dumping and tariff increases encountered by similar industries over the past five years, an average demand shock coefficient is derived. For example, a 20% tariff increase might lead to a decrease of approximately 30% in exports to that market. Then, the current export volume of the industry to the target market is multiplied by this demand shock coefficient to obtain a potential loss simulation value. The formula is: Loss simulation value = Current export value to the target market × Historical average demand shock coefficient.

[0036] This invention transforms heterogeneous information from multiple sources, such as news and public opinion, operational data, and trade statistics, into structured and numerical risk feature vectors through a multi-level structured risk quantification system. This transforms qualitative risks, which are difficult to apply to model training, into quantitative features that can be directly identified and learned by machine learning models, thus solving the problem of difficulty in modeling risk factors in logistics and trade.

[0037] In a specific embodiment of the present invention, due to the existence of sudden and high-impact geopolitical risk events in cross-border logistics trade, and the fact that the content of the multi-level structured risk quantification system is mostly routine and can be obtained based on historical data, the forecast results of cross-border logistics demand will decrease when facing sudden geopolitical risk events, leading to trade decision failure and economic losses. Therefore, the present invention proposes the concept of sudden geopolitical risks, specifically including: No action will be taken unless a sudden geopolitical risk event occurs. When a sudden geopolitical risk event occurs, an update operation is performed based on the risk type of the event to correct the geopolitical risk feature vector. Risk types include node risk, channel risk, and industry dependence. Sudden geopolitical risk events can be monitored in real time through the integration of multi-source information streams such as news APIs, official announcement crawlers, and logistics operation alerts. Events that may affect cross-border logistics channels at target ports can be captured, and event information that meets the preset sudden risk keywords and scope of impact can be marked as sudden geopolitical risk events, and the response process can be initiated.

[0038] The update process includes: matching the risk type with a pre-set emergency risk rule base; if a match is successful, calculating a correction value based on the corresponding quantitative rule, and updating the geopolitical risk feature vector of the corresponding risk type according to the correction value; the emergency risk rule base is composed of risk type, trigger condition, and quantitative rule as basic knowledge units. For example, a rule for corridor risk can be defined as: risk type is "corridor risk"; trigger condition is "the X highway connecting port A and a key hub in country B is interrupted due to geological disaster, and the estimated reopening time is >48 hours"; quantitative rule is "temporarily increase the corridor risk value by 0.4". When a match is successful, the pre-set quantitative rule is directly used to obtain the correction value and update the geopolitical risk feature vector. The processing speed is fast and it is applicable to known and predefined typical risk scenarios. For unknown types or novel sudden risk events for which there is no complete match in the sudden risk rule base (i.e., matching fails), similar historical sudden risk cases are retrieved from the historical sudden risk event database based on similarity calculation. Based on the risk values ​​of the retrieved cases, updated suggested values ​​are generated, and the geopolitical risk feature vector of the corresponding risk type is updated. The expression for similarity calculation is as follows:

[0039] in, For sudden geopolitical risk events, For historical sudden risk events, To assess the similarity between sudden geopolitical risk events and historical sudden risk events, For sudden geopolitical risk events in the first Standardized values ​​of each feature dimension. For historical sudden risk events in the first Standardized values ​​of each feature dimension. The total number of feature dimensions is defined here. These feature dimensions include assessment dimensions of the risk event itself, such as the scope of its impact, expected duration, and severity level. Standardized values ​​are obtained using a standardization method to facilitate calculation and comparison between different risk events. This invention can calculate the similarity of all recorded historical sudden risk event cases and select the top three historical sudden risk events with the highest similarity. Updated suggested values ​​are obtained through methods such as average, median, and weighted average to update the geopolitical risk feature vector. Simultaneously, new sudden risk events can be recorded for subsequent expert evaluation, enriching the historical risk event case library and the sudden risk rule library. This invention, combined with sudden geopolitical risks, realizes the processing of sudden risks from perception to quantification to predictive response, improving the resilience, adaptability, and timeliness of decision support in cross-border logistics demand forecasting when facing the impact of uncertain risk events.

[0040] In one embodiment of the present invention, after obtaining the baseline logistics demand value and the geopolitical risk feature vector, the baseline logistics demand value and the geopolitical risk feature vector are fused to obtain a combined feature vector; The combined feature vectors are input into a pre-trained cross-border logistics demand fusion prediction model, which outputs the final logistics demand prediction value and optimizes the pre-trained cross-border logistics demand fusion prediction model. Acquire historical time series data, including historical true values ​​of logistics demand, historical benchmark logistics demand values, and historical geopolitical risk feature vectors; The fused feature sequence, composed of historical baseline logistics demand values ​​from the previous time to the current time and historical geopolitical risk feature vectors, is used as input. The actual historical logistics demand value at the current time is used as the prediction target. The Long Short-Term Memory (LSTM) network model is then trained under supervision to obtain a pre-trained cross-border logistics demand fusion prediction model. Alternatively, other time-series prediction models can be used for supervised training. The pre-trained cross-border logistics demand fusion prediction model is optimized, specifically by updating the model based on newly added logistics demand data and its corresponding logistics benchmark values ​​and risk feature vectors.

[0041] The beneficial effects of this invention are as follows: This invention combines a scenario-based trade gravity benchmark model with a machine learning-based predictive model to achieve synergistic prediction of long-term trends and short-term fluctuations. The scenario-based trade gravity benchmark model, based on highly explanatory macroeconomic variables such as hinterland trade volume and broad trade costs, accurately depicts the long-term fundamental trends of port logistics. The machine learning model focuses on dynamic variables such as risk feature vectors, responsible for capturing and correcting short-term market fluctuations and abnormal shocks. This layered architecture retains the macroeconomic explanatory power and trend stability of economic models while incorporating the sensitive response of machine learning to complex short-term factors.

[0042] This invention constructs a systematic emergency risk handling mechanism. This mechanism uses a dual-path approach of rule matching and historical similarity matching to quickly transform abstract emergency risk events into structured, computable risk feature vectors, which are then fed back into the prediction process in real time. This enables the model to quickly adjust its prediction output when faced with uncertain events such as channel interruption, forming a closed loop of risk identification-quantification-response, thereby enhancing the resilience and decision-making timeliness of the prediction system in complex and ever-changing environments.

Claims

1. A cross-border logistics demand forecasting method incorporating geopolitical risk correction, characterized in that, Includes the following steps: Based on the forecast period of cross-border logistics demand at the target port, and the total foreign trade volume of the effective economic hinterland of the target port and the overall generalized trade cost, the benchmark logistics demand value is calculated through a pre-constructed scenario-based trade gravity benchmark model. Based on the forecast period of cross-border logistics demand at the target port, and using a multi-level structured risk quantification system and sudden geopolitical risks, the geopolitical risk characteristic vector corresponding to the forecast period is obtained. By fusing the baseline logistics demand value and the geopolitical risk feature vector, a combined feature vector is obtained; The combined feature vectors are input into a pre-trained cross-border logistics demand fusion prediction model, which outputs the final logistics demand prediction value and optimizes the pre-trained cross-border logistics demand fusion prediction model.

2. The cross-border logistics demand forecasting method incorporating geopolitical risk correction as described in claim 1, characterized in that, The expression for the pre-built scenario-based trade gravity benchmark model is as follows: in, As the baseline logistics demand value, , and All are weighted parameters. This refers to the total trade volume exclusive to a port, which is the sum of the total foreign trade volume of all sub-regions within the effective economic hinterland through that port. The weighted average generalized trade cost is the total value of goods used in the port-specific trade. For the first Foreign trade volume of each sub-region through target ports for The corresponding total generalized trade costs, The total number of subdivided regions within the effective economic hinterland.

3. The cross-border logistics demand forecasting method incorporating geopolitical risk correction as described in claim 1, characterized in that, The multi-level structured risk quantification system includes: Node risk is used to quantify the operational stability of domestic ports and related overseas ports, as well as the risks associated with the cross-border trade regulatory environment. Corridor risk is used to quantify the reliability of domestic connections between domestic ports and economic hinterlands, as well as cross-border transport corridors. Industry dependence risk is used to quantify the vulnerability of hinterland industries to specific international markets; The underlying quantitative indicators of node risk include: the percentage of annual operational interruption time at domestic and foreign ports, the frequency of changes in cross-border trade regulations in the region where the port is located, and the GPR index based on the Global Peace Index (GPI). The underlying quantitative indicators of the corridor risk include: the number of traffic interruptions per year on key road sections, the number of days the corridor is shut down due to extreme weather, and the failure rate of multimodal transport connections. The underlying quantitative indicators of industry dependence risk include: the proportion of a specific industry's exports to the target market, the industry loss value after a simulated trade barrier is triggered, and the production capacity proportion of alternative suppliers in the target market.

4. The cross-border logistics demand forecasting method incorporating geopolitical risk correction according to claim 3, characterized in that, The acquisition of the geopolitical risk feature vector corresponding to the prediction period specifically includes: We weighted and fused the underlying quantitative indicators of node risk, channel risk, and industry dependence risk to obtain the median values ​​of node risk, channel risk, and industry dependence risk. The median values ​​of node risk, channel risk, and industry dependence risk are standardized to obtain standardized node risk values, standardized channel risk values, and standardized industry dependence risk values, respectively. Based on standardized node risk values, standardized channel risk values, and standardized industry dependence risk values, combined with sudden geopolitical risks, a geopolitical risk feature vector is constructed.

5. The cross-border logistics demand forecasting method incorporating geopolitical risk correction as described in claim 4, characterized in that, The aforementioned combination of sudden geopolitical risks specifically includes: When a sudden geopolitical risk event occurs, an update operation is performed based on the risk type of the sudden geopolitical risk event to correct the geopolitical risk feature vector; the risk types include node risk, channel risk, and industry dependence; The update operation includes: matching the risk type with a pre-set emergency risk rule base; if the match is successful, calculating the correction value based on the corresponding quantitative rule, and updating the geopolitical risk feature vector of the corresponding risk type according to the correction value; If a match fails, similar historical emergency risk cases are retrieved from the historical emergency risk event database based on similarity calculation. Based on the risk values ​​of the retrieved cases, an updated suggested value is generated, and the geopolitical risk feature vector of the corresponding risk type is updated.

6. The cross-border logistics demand forecasting method incorporating geopolitical risk correction as described in claim 5, characterized in that, The expression for calculating the similarity is as follows: in, For sudden geopolitical risk events, For historical sudden risk events, To assess the similarity between sudden geopolitical risk events and historical sudden risk events, For sudden geopolitical risk events in the first Standardized values ​​of each feature dimension. For historical sudden risk events in the first Standardized values ​​of each feature dimension. This represents the total number of feature dimensions.

7. The cross-border logistics demand forecasting method incorporating geopolitical risk correction as described in claim 1, characterized in that, The pre-trained cross-border logistics demand fusion prediction model specifically includes: Acquire historical time series data, including historical true values ​​of logistics demand, historical benchmark logistics demand values, and historical geopolitical risk feature vectors; The fusion feature sequence, composed of the historical baseline logistics demand value from the previous moment to the current moment and the historical geopolitical risk feature vector, is used as input. The actual historical logistics demand value at the current moment is used as the prediction target. The long short-term memory network model is trained under supervision to obtain a pre-trained cross-border logistics demand fusion prediction model. The optimization of the pre-trained cross-border logistics demand fusion prediction model specifically includes: optimizing and updating the cross-border logistics demand fusion prediction model based on newly added logistics demand data and its corresponding logistics benchmark values ​​and risk feature vectors.

Images