River-sea intermodal transportation scheduling optimization method and system based on traffic volume prediction

Abstract

The application provides a river-sea combined transportation scheduling optimization method and system based on traffic volume prediction, obtains river-sea combined transportation historical operation information, performs combined transportation traffic volume feature extraction operation on the river-sea combined transportation historical operation information, and obtains combined transportation traffic volume influence features; performs traffic volume trend analysis according to the combined transportation traffic volume influence features, and generates combined transportation section traffic volume prediction features; constructs ship transfer coordination relations based on the combined transportation section traffic volume prediction features; and formulates a ship cross-section scheduling scheme according to the ship transfer coordination relations, which is used for coordinating the transfer scheduling operation of the ship between the river transportation section and the sea transportation section. The application avoids the problem that the traffic volume prediction is disconnected from the transfer scheduling in the traditional scheduling, realizes the direct correlation between the traffic volume prediction result and the ship transfer arrangement, can better coordinate the transfer scheduling operation of the ship between the river transportation section and the sea transportation section, and improves the rationality and effectiveness of the ship cross-section scheduling.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intermodal dispatching, in particular to a river-sea intermodal dispatching optimization method and system based on traffic volume prediction. BACKGROUND

[0002] Currently, common river-sea intermodal dispatching is based on historical transportation volume statistics or single-section operation records for dispatching arrangement. Specifically, historical sailing time data of ships and cargo handling records of ports are collected, future traffic volume is predicted through manual experience or simple statistical models, and then fixed ship dispatching plans are formulated according to the prediction results. However, these methods often focus on the optimization of transportation efficiency in a single section. In actual application, due to the large difference in geographical environment between river transport sections and sea transport sections, there are significant differences in ship sailing trajectory characteristics, port operation timing rules and cargo transfer connection. It is difficult to fully reflect the complex dynamic changes in the river-sea intermodal process by relying only on historical statistical data or single-section operation records, resulting in deviations between the generated traffic volume prediction results and actual transportation demand, and further making it difficult to accurately match the actual traffic volume demand of river transport and sea transport by establishing ship transfer relationships, which easily leads to poor transfer connection between sections, affecting the overall efficiency and coordination of ship cross-section dispatching. SUMMARY

[0003] The present application provides a river-sea intermodal dispatching optimization method and system based on traffic volume prediction.

[0004] In a first aspect, the present application provides a river-sea intermodal dispatching optimization method based on traffic volume prediction, which includes: obtaining river-sea intermodal historical operation information, the river-sea intermodal historical operation information including ship sailing trajectory information, port operation timing information and cargo transfer connection information; performing intermodal traffic volume feature extraction operation on the river-sea intermodal historical operation information to obtain intermodal traffic volume influence features; conducting traffic volume trend analysis based on the intermodal traffic volume influence features to generate intermodal section traffic volume prediction features; constructing ship transfer coordination relationships based on the intermodal section traffic volume prediction features; and formulating a ship cross-section dispatching scheme according to the ship transfer coordination relationships, the ship cross-section dispatching scheme being used to coordinate the transfer dispatching operation of ships between river transport sections and sea transport sections.

[0005] In a second aspect, the present application provides a computer system, which includes: a memory having a computer program stored therein; and a processor for loading the computer program to realize the river-sea intermodal dispatching optimization method based on traffic volume prediction as described above.

[0006] The application provides a river-sea intermodal transportation scheduling optimization method based on transportation volume prediction. The river-sea intermodal transportation scheduling optimization method based on transportation volume prediction can comprehensively reflect the trajectory form law of ship navigation, the time sequence law of port operation and the connection law of cargo transfer, and can generate intermodal section transportation volume prediction characteristics that are more in line with actual transportation demand by carrying out transportation trend analysis according to the intermodal transportation volume influence characteristics. The intermodal section transportation volume prediction characteristics not only contain the change trend characteristics of the transportation volume, but also fuse the multidimensional influence factors of ship navigation, port operation and cargo transfer, so that the ship transfer coordination relationship established based on the characteristics can accurately reflect the transportation volume matching demand of different sections. The intermodal section transportation volume prediction characteristics are directly converted into the ship transfer coordination relationship, avoiding the problem that the transportation volume prediction is disconnected from the transfer scheduling in the traditional scheduling, realizing the direct association of the transportation volume prediction result and the ship transfer arrangement, and the ship cross-section scheduling scheme formulated according to the ship transfer coordination relationship can better coordinate the transfer scheduling operation of the ship between the river transportation section and the sea transportation section, and improve the rationality and effectiveness of the ship cross-section scheduling. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a flowchart of a river-sea intermodal transportation scheduling optimization method based on transportation volume prediction provided by the embodiment of the application.

[0008] Figure 2 is a composition schematic diagram of a computer system provided by the embodiment of the application. DETAILED DESCRIPTION

[0009] Please refer to Figure 1 , Figure 1 is a flowchart of a river-sea intermodal transportation scheduling optimization method based on transportation volume prediction provided by the embodiment of the application. The method can be executed by a computer system and includes the following steps:

[0010] Step S100: Obtain river-sea intermodal transportation historical operation information, which includes ship navigation trajectory information, port operation time sequence information and cargo transfer connection information.

[0011] The historical operation information of river-sea intermodal transportation refers to a collection of various types of information related to transportation generated in a period of time in the process of river-sea intermodal transportation. The ship navigation trajectory information is data describing the navigation path of a ship in the river transportation and sea transportation sections, recording the position of the ship at different times, and reflecting the navigation route and dynamics of the ship. The port operation timing information is data about the time arrangement and sequence of port cargo handling operations, including the start time, end time, and equipment used for cargo handling, and reflects the process and efficiency of port operations. The cargo transfer connection information is data describing the conversion and connection of cargo between different transportation sections, including the arrival time, departure time, and transfer time interval of cargo at the transfer node, and reflects the continuity and smoothness of the cargo transportation process. The ship navigation trajectory information can be obtained from the positioning system or shipping management platform of the ship. The port operation timing information can be extracted from the operation management system of the port. The cargo transfer connection information can be collected through the logistics information system or cargo tracking equipment.

[0012] Step S200: performing intermodal transportation volume feature extraction operation on the river-sea intermodal transportation historical operation information to obtain intermodal transportation volume influence feature.

[0013] As an implementation manner, step S200 can be specifically implemented as steps S210-S240:

[0014] Step S210: performing trajectory pattern analysis on the river section trajectory data in the ship navigation trajectory information, discretizing the continuous navigation trajectory of the ship in the river section into a coordinate point sequence according to a preset spatial resolution, calculating a sequence of direction deflection angles and a sequence of distance changes between adjacent coordinate points, the sequence of direction deflection angles representing the change of the navigation direction of the ship, and the sequence of distance changes representing the change of the navigation speed of the ship.

[0015] The trajectory pattern analysis is a process of in-depth analysis of ship navigation trajectory data to mine features and rules therein. The river section trajectory data is the navigation trajectory information formed by the ship in the river transportation process. The preset spatial resolution is a spatial interval for discretizing the continuous trajectory, which discretizes the continuous navigation trajectory of the ship in the river section into a coordinate point sequence, i.e., a series of coordinate points are selected from the continuous trajectory according to the preset spatial resolution, and these coordinate points arranged in time sequence constitute the coordinate point sequence. The sequence of direction deflection angles is a group of data obtained by calculating the deflection angle of the direction vector between adjacent coordinate points, reflecting the change of the direction of the ship in the navigation process.

[0016] As an implementation manner, step S210 can be specifically implemented as steps S211-S216:

[0017] Step S211: sampling coordinate points in the river section trajectory data in the ship navigation trajectory information according to a preset spatial resolution, selecting uniformly spaced coordinate points from the continuous navigation trajectory of the ship in the river section to obtain a coordinate point sequence.

[0018] The preset spatial resolution is a predetermined distance value for determining the interval of sampling coordinate points on the continuous navigation trajectory. The river section trajectory data in the ship navigation trajectory information is continuous position information recorded by the ship during river transportation, which can be obtained from the positioning system of the ship. By sampling coordinate points according to the preset spatial resolution, uniformly spaced coordinate points are selected from the continuous position information, and the coordinate points are arranged in time sequence, thereby obtaining the coordinate point sequence.

[0019] Step S212: calculating the direction vectors between adjacent coordinate points in the coordinate point sequence, calculating the deflection angle between adjacent direction vectors by an inverse tangent function to obtain a direction deflection angle sequence.

[0020] The coordinate point sequence is obtained by arranging the uniformly spaced coordinate points obtained in step S211 in time sequence. The direction vector between adjacent coordinate points is obtained by calculating the coordinate difference between two adjacent coordinate points, which represents the direction of the ship from one coordinate point to the next coordinate point. For example, for the i-th coordinate point (x i ,y i ) and the i+1-th coordinate point (x i+1 ,y i+1 ) in the coordinate point sequence, the direction vector therebetween can be represented as (x i+1 -x i ,y i+1 -y i ). The inverse tangent function is used to calculate the inverse function of the tangent value of an angle. The deflection angle between adjacent direction vectors can be calculated by the inverse tangent function. Specifically, for two adjacent direction vectors, the tangent value of the included angle is first calculated, and then the inverse tangent function is used to obtain the angle value of the included angle. The deflection angle of the direction vector between every two adjacent coordinate points in the coordinate point sequence is calculated in sequence, and arranged in time sequence, thereby obtaining the direction deflection angle sequence.

[0021] Step S213: calculating the ratio of the straight line distance between adjacent coordinate points in the coordinate point sequence to the corresponding time interval to obtain a distance change sequence.

[0022] The coordinate point sequence is the coordinate point set arranged in time sequence obtained in step S211. The straight-line distance between adjacent coordinate points can be calculated by the distance formula between two points, and the corresponding time interval refers to the time difference corresponding to two adjacent coordinate points. The straight-line distance between adjacent coordinate points is divided by the corresponding time interval, and the distance traveled by the ship per unit time is obtained, that is, the average speed of the ship. The ratio of the straight-line distance between each two adjacent coordinate points in the coordinate point sequence to the corresponding time interval is calculated in sequence, and arranged in time sequence, that is, the distance change sequence is formed. This sequence reflects the change of the speed of the ship during navigation.

[0023] Step S214: sliding window filtering is performed on the direction deflection angle sequence, and mean value calculation is performed on the continuous direction deflection angle values through a sliding window with a preset length, so as to eliminate instantaneous angle mutation interference.

[0024] The direction deflection angle sequence is a group of data reflecting the change of the navigation direction of the ship obtained in step S212. The sliding window filtering is used to smooth the data sequence and reduce the noise and mutation in the data. The sliding window with a preset length is a window with a fixed size preset in advance, which is used to slide on the direction deflection angle sequence. For example, if the preset length is 5, then 5 continuous direction deflection angle values are selected in sequence on the direction deflection angle sequence. A new value is obtained by performing mean value calculation on the 5 continuous direction deflection angle values, and the new value is used as the filtered value of the center position of the window. Then, the sliding window is moved to the right by one position, and the continuous 5 direction deflection angle values are selected to perform mean value calculation, and the process is repeated until the entire direction deflection angle sequence is traversed. In this way, the filtered direction deflection angle sequence is obtained. The instantaneous angle mutation interference refers to the angle value with a sudden change in the direction deflection angle sequence. These values may be caused by measurement error, temporary adjustment of the ship, etc., and will affect the analysis of the change rule of the navigation direction of the ship.

[0025] Step S215: sliding window filtering is performed on the distance change sequence, and mean value calculation is performed on the continuous distance change values through a sliding window with the same preset length, so as to eliminate instantaneous speed fluctuation interference.

[0026] The distance change sequence is a set of data reflecting the change of the ship's sailing speed obtained in step S213. The sliding window filtering is a method for smoothing the data sequence and reducing noise interference. The sliding window of the same preset length is a fixed size window with the same length as the window length used for filtering the direction deflection angle sequence in step S214. For example, if the preset length is 5 in step S214, a sliding window with a length of 5 is also used in this step. On the distance change sequence, 5 consecutive distance change values are selected in turn, and the average value of the 5 values is calculated as the filtered value of the window center position. Then the sliding window is moved one position to the right, and the next 5 consecutive distance change values are selected for mean value calculation, and the process is repeated until the entire distance change sequence is traversed. In this way, the filtered distance change sequence is obtained. The instantaneous speed fluctuation interference refers to the values with sudden large changes in the distance change sequence, which may be caused by temporary acceleration, deceleration or measurement error of the ship, and will affect the analysis of the change rule of the ship's sailing speed.

[0027] Step S216: integrate the filtered direction deflection angle sequence and the filtered distance change sequence into the trajectory shape feature of the ship in the river transportation section.

[0028] The filtered direction deflection angle sequence is a direction deflection angle data sequence after step S214, which eliminates the instantaneous angle mutation interference and more smoothly reflects the change of the ship's sailing direction. The filtered distance change sequence is a distance change data sequence after step S215, which eliminates the instantaneous speed fluctuation interference and more accurately reflects the change of the ship's sailing speed. The trajectory shape feature of the ship in the river transportation section is a comprehensive feature representation, which combines the change of the ship's sailing direction and speed, and can more comprehensively describe the ship's sailing trajectory feature in the river transportation section. The filtered direction deflection angle sequence and the filtered distance change sequence can be integrated together in various ways, such as combining them as a two-dimensional feature vector, or pairing them in time sequence to form a new sequence.

[0029] Step S220: perform time sequence pattern recognition on the cargo loading and unloading records in the port operation time sequence information, extract the cargo loading and unloading order feature and the equipment use alternation feature, the cargo loading and unloading order feature represents the processing order of different cargo types in the port, and the equipment use alternation feature represents the switching order of different operation equipment.

[0030] Cargo loading and unloading records in port operation time-series information are detailed records of the time and operational information of cargo loading and unloading operations at the port, including information such as cargo type, start time, end time, and equipment used. Time-series pattern recognition is used to discover potential patterns and regularities from time-series data. By performing time-series pattern recognition on cargo loading and unloading records, hidden information about the order of cargo loading and unloading and the order of equipment use can be extracted. The cargo loading and unloading sequence characteristics at the port describe the order in which different types of cargo are loaded and unloaded at the port. The equipment use alternation characteristics describe the order in which different operating equipment is activated and switched during port operations. By analyzing cargo loading and unloading records, the loading and unloading sequences of different cargo types and the activation and switching sequences of different operating equipment can be identified. By organizing and summarizing this information, the cargo loading and unloading sequence characteristics and equipment use alternation characteristics at the port can be obtained.

[0031] As one implementation method, step S220 can be specifically implemented as the following steps S221~S224:

[0032] Step S221: Arrange the cargo loading and unloading records in the port operation time sequence information into an operation event sequence according to the time sequence. Each operation event includes a cargo type identifier and an operation timestamp.

[0033] The cargo loading and unloading records in port operation sequence information contain detailed information recorded by the port during cargo loading and unloading operations, including cargo type, start time, and end time. Arranging these records chronologically forms a sequence of operation events. Each operation event includes a cargo type identifier and an operation timestamp. The cargo type identifier distinguishes different types of cargo; for example, letters or numbers can be used to represent different cargo types, such as "Class A cargo," "Class B cargo," etc. The operation timestamp records the specific start or end time of the cargo loading and unloading operation, typically accurate to the second.

[0034] Step S222: Perform sequence pattern mining on the operation event sequence, identify the combination sequence of goods types that occurs more frequently than the preset frequency, and statistically analyze the position distribution characteristics and transition probability characteristics of different goods types in the combination sequence of goods types. The position distribution characteristics represent the occurrence position pattern of goods types in the sequence, and the transition probability characteristics represent the probability distribution of one goods type following another.

[0035] The job event sequence is the set of job events obtained in step S221, arranged in chronological order. Each job event includes a cargo type identifier and an operation timestamp. Sequence pattern mining is a data mining technique used to discover frequently occurring patterns in sequence data. Preset frequency is a pre-defined threshold used to determine whether cargo type combination sequences occur frequently. Positional distribution characteristics refer to the positional patterns of different cargo types in cargo type combination sequences. For example, some cargo types may always appear at the beginning of the sequence, while other cargo types may tend to appear in the middle or at the end. Transition probability characteristics refer to the probability distribution of one cargo type following another.

[0036] Step S223: Perform state transition analysis on the equipment operation records in the port operation sequence information to identify the activation sequence characteristics and deactivation sequence characteristics of different types of equipment. The activation sequence characteristics represent the order in which different types of equipment start operations, and the deactivation sequence characteristics represent the order in which different types of equipment finish operations.

[0037] The equipment operation records in port operation sequence information are detailed records of the operation of port equipment during use, including equipment activation time, shutdown time, and equipment type. State transition analysis, by analyzing the state transitions of the equipment operation records, can uncover the sequential patterns in the activation and shutdown processes of different types of equipment. The activation sequence characteristics of different types of equipment describe the order in which different types of equipment begin operation. By analyzing the equipment operation records and sorting them according to activation and shutdown times, the activation and shutdown sequences of different types of equipment can be identified. Organizing and summarizing this sequence information yields the activation and shutdown sequence characteristics of different types of equipment.

[0038] Step S224: Integrate the location distribution features, transfer probability features, activation sequence features, and deactivation sequence features to obtain the cargo loading and unloading sequence features and equipment usage alternation features at the port.

[0039] The location distribution feature, obtained in step S222, describes the positional regularity of different cargo types in a cargo type combination sequence, reflecting the sequential positional relationship of cargo during loading and unloading. The transition probability feature, also obtained in step S222, describes the probability distribution of one cargo type following another, reflecting the probabilistic regularity of cargo loading and unloading sequence. The activation sequence feature, obtained in step S223, describes the sequence in which different operating equipment begins operation, reflecting the sequential relationship of equipment activation. The deactivation sequence feature, obtained in step S223, describes the sequence in which different operating equipment ends operation, reflecting the sequential regularity of equipment deactivation. These features can be integrated in various ways. For example, the location distribution feature and transition probability feature can be used as feature dimensions of cargo loading and unloading sequence, while the activation sequence feature and deactivation sequence feature can be used as feature dimensions of equipment alternation, combined to form a comprehensive feature vector.

[0040] Step S230: Perform node association analysis on the transfer node data in the cargo transfer connection information to obtain the characteristics of cargo dwell time at transfer nodes and the characteristics of smooth transfer connection. The dwell time characteristic represents the processing time of cargo at transfer nodes, and the smooth transfer connection characteristic represents the degree of connection between different transfer nodes.

[0041] The transit node data in cargo transshipment information contains detailed information about each node through which cargo passes during transit, including arrival and departure times, and transit time intervals. Node correlation analysis is used to study the relationships and interactions between different nodes. By performing node correlation analysis on the transit node data, hidden information about cargo dwell time and the smoothness of transit connections can be extracted. The dwell time characteristic of cargo at transit nodes describes the time spent on processing (such as loading, unloading, and storage) at the transit nodes.

[0042] As one implementation method, step S230 can be specifically implemented as the following steps S231~S234:

[0043] Step S231: Extract the arrival time and departure time records of the goods at each transfer node from the goods transfer connection information, calculate the dwell time sequence of the goods at each transfer node, and obtain the dwell time sequence by the difference between the departure time record and the arrival time record.

[0044] Cargo transshipment connection information describes the transfer and connection of goods between different transportation segments, including information such as the arrival and departure times of goods at each transshipment node. From this information, the arrival and departure time records of goods at each transshipment node are extracted; these records can be data accurate to specific time points. By calculating the difference between the departure and arrival time records, the dwell time of the goods at each transshipment node can be obtained.

[0045] Step S232: Perform distribution feature analysis on the dwell time sequence to identify the dwell time distribution pattern and dwell time fluctuation range of different cargo types at the transfer node. The dwell time distribution pattern represents the probability distribution shape of the dwell time, and the dwell time fluctuation range represents the numerical variation range of the dwell time.

[0046] The dwell time sequence is a sequence composed of the dwell times of goods obtained in step S231 at each transshipment node. Distribution characteristic analysis is a data analysis method used to study the distribution patterns and characteristics of data. The distribution pattern of dwell time at transshipment nodes for different types of goods refers to the probability distribution shape of the dwell time of different types of goods at transshipment nodes. For example, the dwell time of some types of goods may exhibit a normal distribution, that is, most of the dwell time is concentrated around the average value; while the dwell time of other types of goods may exhibit a skewed distribution, that is, the distribution of dwell time is biased to one side. The dwell time fluctuation range refers to the range of numerical changes in dwell time, reflecting the stability of dwell time. For example, if the dwell time fluctuation range is small, it indicates that the dwell time of goods at transshipment nodes is relatively stable; conversely, if the fluctuation range is large, it indicates that the dwell time is affected by multiple factors and varies greatly. By performing distribution characteristic analysis on the dwell time sequence, statistical methods, such as calculating the mean, variance, and standard deviation, can be used to describe the distribution of dwell time. At the same time, tools such as histograms and probability density functions can be used to visualize the distribution pattern of dwell time. The range of fluctuations in dwell time can be determined by calculating the difference between the maximum and minimum values.

[0047] Step S233: Perform sequence analysis on the transfer time interval records between adjacent transfer nodes in the cargo transfer connection information, identify abnormal fluctuation points and stable change segments in the transfer time interval. Abnormal fluctuation points represent abnormal values ​​that deviate from the normal transfer time interval, and stable change segments represent continuous periods in which the time interval remains stable.

[0048] The transit time interval records between adjacent transit nodes in cargo transit connection information are records of the time spent transporting goods between two adjacent transit nodes. These records reflect the continuity and smoothness of the cargo transportation process. By performing sequence analysis on the transit time interval records, hidden patterns and characteristics can be uncovered. Abnormal fluctuation points refer to outliers in the transit time interval records that deviate from the normal transit time intervals. By performing sequence analysis on the transit time interval records, various methods can be used to identify abnormal fluctuation points and stable variation segments. For example, statistical methods can be used to calculate the mean and standard deviation of the transit time intervals, and values ​​that deviate from the mean by more than a certain multiple of the standard deviation can be identified as abnormal fluctuation points; by observing the changing trends of the time intervals, continuous periods in which the time intervals remain relatively stable can be identified as stable variation segments.

[0049] Step S234: Integrate the dwell time distribution pattern and dwell time fluctuation range into the dwell time characteristics of goods at the transfer node, and integrate abnormal fluctuation points and stable change segments into the characteristics of smooth cargo transfer connection.

[0050] The dwell time distribution pattern is the probability distribution of dwell time for different types of goods at the transfer node, obtained in step S232, reflecting the probabilistic characteristics of dwell time. The dwell time fluctuation range is the range of numerical changes in dwell time obtained in step S232, reflecting the stability of dwell time. These two features can be integrated in various ways, such as combining them as two-dimensional feature vectors or fusing them into a comprehensive feature representation. Abnormal fluctuation points are the outliers identified in step S233 that deviate from the normal transfer time interval, reflecting potential problems during goods transfer. Stable change segments are the continuous periods of stable transfer time intervals identified in step S233, reflecting the continuity and smoothness of goods transfer. Abnormal fluctuation points and stable change segments can also be integrated in various ways, such as combining them as two-dimensional feature vectors or associating them according to certain logic to form a comprehensive feature representation.

[0051] Step S240: The direction deflection angle sequence, distance change sequence, loading and unloading sequence characteristics, equipment usage alternation characteristics, dwell time characteristics, and transfer connection smoothness characteristics are fused to obtain the intermodal transport volume impact characteristics.

[0052] The direction deflection angle sequence, obtained in step S212, is a set of data reflecting changes in the ship's navigation direction. After filtering, it more accurately reflects the changing patterns of the ship's navigation direction. The distance change sequence, obtained in step S213, is a set of data reflecting changes in the ship's navigation speed. After filtering, it more smoothly displays changes in the ship's navigation speed. The loading and unloading sequence feature, extracted in step S220, describes the sequence of loading and unloading operations for different types of cargo at the port, reflecting the process and patterns of cargo loading and unloading at the port. The equipment usage alternation feature, extracted in step S220, describes the activation and switching sequence of different operating equipment during port operations, reflecting the efficiency of port equipment utilization and operational arrangements. The dwell time feature, obtained in step S230, describes the processing time of cargo at transfer nodes, including information such as the distribution pattern and fluctuation range of dwell time. The smooth transfer connection feature, obtained in step S230, describes the continuity and smoothness of cargo transfer between different transfer nodes, including information such as abnormal fluctuation points and stable change segments. Feature fusion is the process of integrating features of different types and dimensions to form a comprehensive feature representation. Features such as direction deflection angle sequence, distance change sequence, loading and unloading sequence features, equipment usage alternation features, dwell time features, and smooth transfer connection features can be fused using various methods. For example, they can be combined as multi-dimensional feature vectors, or feature fusion algorithms in machine learning, such as principal component analysis (PCA) and linear discriminant analysis (LDA), can be used to reduce the dimensionality and fuse these features.

[0053] Step S300: Conduct a volume trend analysis based on the characteristics of intermodal transport volume impact to generate volume prediction characteristics for intermodal transport segments.

[0054] The intermodal transport volume impact characteristics, obtained in step S200, are a comprehensive set of features reflecting the influence of various factors on river-sea intermodal transport volume, including information on ship navigation trajectories, port operations, and cargo transshipment. Volume trend analysis is the process of studying and analyzing the patterns and trends of volume changes over time. By analyzing the intermodal transport volume impact characteristics, information related to volume trends can be extracted.

[0055] As one implementation method, step S300 can be specifically implemented as the following steps S310~S370:

[0056] Step S310: Expand the ship navigation trajectory features in the intermodal transport volume impact characteristics into a time series, and arrange the ship navigation trajectory features into a trajectory time series according to the time order.

[0057] The ship navigation trajectory features in the intermodal transport volume impact characteristics are the features related to the ship navigation trajectory obtained in step S200, including features obtained after processing such as direction deflection angle sequences and distance change sequences. Time series expansion is the process of arranging and expanding feature data that was originally combined in a certain form according to time order. Ship navigation trajectory features usually contain navigation information of the ship at different points in time. By expanding these features into a time series, they can be converted into a trajectory time series arranged in chronological order. The trajectory time series is a sequence formed by arranging ship navigation trajectory features in chronological order, with each point in time corresponding to a corresponding navigation trajectory feature value.

[0058] As one implementation method, step S310 can be specifically implemented as the following steps S311~S314:

[0059] Step S311: Align the ship navigation trajectory features in the intermodal transport volume impact characteristics with the time axis, and unify the ship navigation trajectory features in different time periods into the same time reference system.

[0060] The ship navigation trajectory features within the intermodal transport volume impact characteristics are feature information related to ship navigation trajectories recorded at different time points. Due to differences in data collection time or other reasons, the time reference systems of these features may be inconsistent. Time axis alignment is the process of unifying ship navigation trajectory features from different time periods into the same time reference system. In practice, this can be done by converting the timestamp information in the ship navigation trajectory features into a unified time format. For example, if time records exist in different time zones, they can be converted to Coordinated Universal Time (UTC). Then, based on the unified time reference system, the ship navigation trajectory features are sorted to ensure they are arranged in chronological order.

[0061] Step S312: Decompose the ship's navigation trajectory features into multiple time segments according to the preset time granularity. Each time segment contains the ship's navigation trajectory features within the corresponding time period.

[0062] The preset time granularity is a pre-defined time interval used to divide ship trajectory features according to time. For example, if the preset time granularity is 1 hour, then the ship trajectory features will be decomposed into 1-hour time intervals. Ship trajectory features are those related to the ship's trajectory, aligned to the same time reference frame after alignment with the time axis, including information such as direction deflection angle and distance changes. By decomposing these features according to the preset time granularity, multiple time segments are obtained. Each time segment contains the ship trajectory features within the corresponding time period.

[0063] Step S313: Extract features from the ship's navigation trajectory features within each time segment to obtain the trajectory center point feature and trajectory dispersion feature for each time segment. The trajectory center point feature represents the average position of the ship within the time segment, and the trajectory dispersion feature represents the degree of position dispersion of the ship within the time segment.

[0064] The ship's trajectory features within each time segment are obtained in step S312 according to a preset time granularity. Each time segment includes features related to the ship's trajectory, such as the direction deflection angle sequence and distance change sequence within the corresponding time period. Feature extraction is the process of further mining key features that can reflect the characteristics of the ship's trajectory from these features. The trajectory center point feature refers to the average position of the ship within the time segment, which can be obtained by calculating the average value of all coordinate points within that time segment. The trajectory dispersion feature refers to the degree of dispersion of the ship's position within the time segment, reflecting the dispersion of the ship's trajectory. The trajectory dispersion can be measured by calculating the variance or standard deviation of the coordinate point sequence.

[0065] Step S314: Arrange the trajectory center point features and trajectory dispersion features in chronological order to obtain the trajectory time series.

[0066] The trajectory center point feature is the average position feature of the ship within each time segment obtained in step S313, reflecting the approximate position of the ship in different time segments. The trajectory dispersion feature is the dispersion feature of the ship's position within each time segment obtained in step S313, reflecting the dispersion of the ship's navigation trajectory. Arranging in chronological order means arranging the trajectory center point feature and trajectory dispersion feature according to the order of the time segments. Since each time segment corresponds to a specific time period, arranging these features in chronological order can form a continuous time series. For example, for the first time segment, there are corresponding trajectory center point features and trajectory dispersion features, which are used as the first element; for the second time segment, there are corresponding features as the second element, and so on.

[0067] Step S320: Extract the trend from the trajectory time series and separate the long-term trend component and the short-term fluctuation component using the moving average algorithm. The long-term trend component represents the direction of change of the transport volume over a long period of time, while the short-term fluctuation component represents the fluctuation of the transport volume over a short period of time.

[0068] The trajectory time series, obtained in step S314, is a sequence arranged chronologically, containing features of the trajectory center point and trajectory dispersion, reflecting the changes in ship navigation trajectories over time. Trend extraction is the process of separating long-term trends and short-term fluctuations from time series data. The long-term trend component refers to the direction of change in traffic volume over a long period, such as whether traffic volume shows an upward trend, a downward trend, or remains stable over a span of several years or decades. The short-term fluctuation component refers to the fluctuations in traffic volume over a short period, such as the fluctuations in traffic volume over several months or weeks due to seasonal factors, unexpected events, etc.

[0069] As one implementation method, step S320 can be specifically implemented as the following steps S321~S324:

[0070] Step S321: Set the sliding window size, perform sliding window processing on the trajectory time series, and calculate the average value of trajectory features within each window.

[0071] The sliding window size is a pre-defined integer used to determine the length of the window that slides across the trajectory time series. For example, if the sliding window size is set to 5, then 5 consecutive data points are selected sequentially across the trajectory time series. The trajectory time series is the sequence obtained in step S314, arranged chronologically and containing trajectory center point features and trajectory dispersion features. Sliding window processing is a method for operating on time series data. A set sliding window slides across the trajectory time series, selecting trajectory feature data within the window each time. For the trajectory feature data within each window, their average value is calculated. For example, for a trajectory time series containing trajectory center point coordinates and trajectory dispersion values, the average value of the trajectory center point coordinates and the average value of the trajectory dispersion values ​​are calculated separately within each window. By continuously moving the sliding window, the process of calculating the average value is repeated until the entire trajectory time series has been traversed.

[0072] Step S322: Use the average value sequence of the continuous window as the long-term trend component of the trajectory time series.

[0073] The average value sequence of the continuous window is obtained through sliding window processing in step S321. The average value of the trajectory features within each window is arranged in chronological order to form a sequence. The long-term trend component of the trajectory time series refers to the direction of change of traffic volume over a long period of time, reflecting the trend of change of the trajectory time series at the macro level.

[0074] Step S323: Calculate the difference between the original trajectory time series and the long-term trend component to obtain the short-term fluctuation component of the trajectory time series.

[0075] The original trajectory time series, obtained in step S314, is a sequence arranged chronologically, containing features of the trajectory center point and trajectory dispersion, reflecting the actual changes in the ship's navigation trajectory over time. The long-term trend component is a sequence of continuous window averages obtained in step S322 through sliding window processing, reflecting the changing trend of the trajectory time series over a long period. Difference calculation involves subtracting each data point in the original trajectory time series from the corresponding time point in the long-term trend component. Arranging the differences of all corresponding time points sequentially yields the short-term fluctuation component of the trajectory time series. The short-term fluctuation component reflects the deviation of the trajectory time series from the long-term trend in a short period, which may be caused by seasonal factors, sudden events, etc.

[0076] Step S324: Smooth the short-term fluctuation component by eliminating high-frequency noise interference through Gaussian filtering to obtain the smoothed short-term fluctuation component.

[0077] Step S330: Input the long-term trend component into the trend prediction layer of the traffic volume prediction model to obtain the long-term traffic volume prediction features.

[0078] The long-term trend component, obtained in step S322, is a sequence reflecting the changing trend of the trajectory time series over a long period, embodying the direction of change in freight volume at the macro level. The freight volume forecasting model is used to predict river-sea intermodal freight volume and typically consists of multiple layers. The trend prediction layer is specifically designed to process long-term trend information. The long-term trend component is input into the trend prediction layer of the freight volume forecasting model. The trend prediction layer analyzes and processes the input long-term trend component, extracting information related to freight volume to obtain long-term freight volume forecasting features.

[0079] As one implementation method, step S330 can be specifically implemented as the following steps S331~S335:

[0080] Step S331: Input the long-term trend component into the input layer of the trend prediction layer, and map the long-term trend component into a high-dimensional trend feature vector through a fully connected network.

[0081] The long-term trend component, obtained in step S322, is a sequence reflecting the long-term changing trend of the trajectory time series, containing information on the changes in traffic volume over a long period. The input layer of the trend prediction layer is the first layer in the traffic volume prediction model that receives input data. A fully connected network is a neural network structure in which each neuron is connected to all neurons in the previous layer. After the long-term trend component is input into the input layer through the fully connected network, the neurons in the input layer will pass the information of the long-term trend component to the neurons in the next layer. Each connection in the fully connected network has a corresponding weight. These weights are used to linearly combine the input data, and with the addition of a bias term, the long-term trend component is mapped into a high-dimensional trend feature vector after processing by the activation function. The high-dimensional trend feature vector is a vector with multiple dimensions, each dimension representing a feature of the long-term trend component.

[0082] Step S332: Input the high-dimensional trend feature vector into the LSTM unit of the trend prediction layer, and memorize the time dependency in the long-term trend features through the gating mechanism.

[0083] The high-dimensional trend feature vector is obtained by mapping the long-term trend component through a fully connected network in step S331, and it contains multiple feature information of the long-term trend component. The LSTM unit of the trend prediction layer is a component of the Long Short-Term Memory (LSTM) network. The gating mechanism is the core feature of the LSTM unit, including the input gate, forget gate, and output gate. The input gate determines how much of the current input information can enter the cell state; the forget gate determines which information in the cell state needs to be forgotten; and the output gate determines which information in the cell state can be output. Through the gating mechanism, the LSTM unit can selectively remember and forget the temporal dependencies in the long-term trend features.

[0084] Step S333: Convert the hidden state of the LSTM cell into a long-term freight volume prediction sequence through the output layer of the trend prediction layer.

[0085] The output layer of the trend prediction layer is the last layer in the traffic volume prediction model, used to transform the information obtained from the previous processing into the final prediction result. The hidden state of the LSTM unit is the internal state that the LSTM unit continuously updates and saves during the processing of the input sequence, containing the time dependencies and important information in the long-term trend features. Through the output layer of the trend prediction layer, the hidden states of the LSTM unit undergo linear transformation and activation function processing, transforming them into a long-term traffic volume prediction value sequence. Specifically, the output layer linearly combines the hidden states of the LSTM unit according to pre-trained weights and biases, and then maps the result to an appropriate range through an activation function (such as a linear activation function) to obtain the long-term traffic volume prediction value. Arranging the prediction values ​​at each time step sequentially forms the long-term traffic volume prediction value sequence. This sequence contains the prediction results of traffic volume over a future period, providing an important basis for subsequent transportation scheduling and decision-making.

[0086] Step S334: Extend the long-term freight volume forecast value series in terms of time dimension, and generate freight volume forecast values ​​for multiple future time units according to the preset forecast time length.

[0087] The long-term traffic volume forecast sequence is the traffic volume forecast result obtained in step S333 for a future period of time, and it is a sequence containing forecast values ​​for a finite number of time steps. Time dimension expansion refers to extending the long-term traffic volume forecast sequence along the time dimension to obtain traffic volume forecast values ​​for more time units. The preset forecast time length is a pre-determined range of future time to be predicted. Various methods can be used for time dimension expansion. For example, trend extrapolation can be used to predict traffic volume for future time units based on the trend and changing patterns of the long-term traffic volume forecast sequence. Alternatively, machine learning models, such as regression models or neural network models, can be used to train the long-term traffic volume forecast sequence and then generate traffic volume forecast values ​​for multiple future time units based on the trained model.

[0088] Step S335: Integrate the traffic volume forecast values ​​of multiple future time units into long-term traffic volume forecast features.

[0089] The predicted freight volume for multiple future time units is the result of the freight volume forecast generated in step S334 according to the preset forecast time length, and these predicted values ​​are arranged in chronological order. The long-term freight volume forecast feature is a comprehensive feature representation used to describe the changes in river-sea intermodal freight volume over a long period. Integrating the predicted freight volume for multiple future time units into the long-term freight volume forecast feature can be done in various ways. For example, these predicted freight volume values ​​can be treated as a sequence and directly used as one dimension of the long-term freight volume forecast feature; alternatively, statistical analysis can be performed on these predicted freight volume values ​​to calculate statistical indicators such as the mean, standard deviation, and growth rate, and these indicators can be used as different dimensions of the long-term freight volume forecast feature. Furthermore, the predicted freight volume values ​​can be integrated with other relevant information, such as ship navigation trajectory characteristics and port operation characteristics, to form a more comprehensive long-term freight volume forecast feature.

[0090] Step S340: Input the short-term fluctuation component into the fluctuation adjustment layer of the traffic volume prediction model to obtain the fluctuation impact characteristics.

[0091] The short-term fluctuation component refers to the deviation of the trajectory time series obtained after smoothing in step S324 from the long-term trend within a short period, reflecting the fluctuations in traffic volume over a short time. The fluctuation adjustment layer of the traffic volume forecasting model is a layer specifically designed to process short-term fluctuation information. It can analyze and process the short-term fluctuation component to obtain the fluctuation impact characteristics. The fluctuation impact characteristics are a set of features describing the impact of short-term fluctuations on traffic volume, including information such as the amplitude, frequency, and duration of the fluctuations. When the short-term fluctuation component is input into the fluctuation adjustment layer of the traffic volume forecasting model, the fluctuation adjustment layer performs a series of processes on the input short-term fluctuation component, such as spectrum analysis and feature extraction, to extract information related to traffic volume fluctuations. By analyzing and integrating this information, the fluctuation impact characteristics are obtained.

[0092] As one implementation method, step S340 can be specifically implemented as the following steps S341~S345:

[0093] Step S341: Perform spectral analysis on the short-term fluctuation component and convert the fluctuation characteristics in the time domain into the fluctuation energy distribution in the frequency domain through Fourier transform.

[0094] The short-term fluctuation component is the change in the smoothed trajectory time series obtained in step S324 due to its deviation from the long-term trend within a short period of time. It is a time-domain signal reflecting the fluctuation of traffic volume over a short period. Spectrum analysis is a method that converts a time-domain signal into a frequency-domain signal to analyze the distribution of different frequency components in the signal. Fourier transform is a commonly used spectrum analysis method that can convert the fluctuation characteristics in the time domain into the fluctuation energy distribution in the frequency domain. Through Fourier transform, the short-term fluctuation component is transformed from the time domain to the frequency domain, resulting in a frequency-domain signal. In the frequency domain, each frequency corresponds to an energy value, which represents the contribution of that frequency component to the short-term fluctuation component.

[0095] Step S342: Identify the main frequency components in the wave energy distribution, and extract the amplitude and phase characteristics of each main frequency component. The amplitude characteristics represent the intensity of the wave, and the phase characteristics represent the starting position of the wave.

[0096] The wave energy distribution, obtained through Fourier transform in step S341, represents the energy distribution of different frequency components within the short-term wave component. The dominant frequency components are those with higher energy in the wave energy distribution, contributing significantly to short-term fluctuations. Identifying the dominant frequency components involves setting an energy threshold, designating frequency components with energy above this threshold as dominant frequency components. Amplitude characteristics refer to the intensity of the wave corresponding to each dominant frequency component, reflecting the magnitude of the wave. For example, in a sine wave, a larger amplitude indicates a larger wave amplitude. Analyzing the energy values ​​of the dominant frequency components in the wave energy distribution yields their amplitude characteristics. Phase characteristics refer to the starting position of the wave corresponding to each dominant frequency component, determining the relative position of the wave in time. For example, two sine waves with the same frequency and amplitude will have different starting points in time if their phases differ. Further analysis of the frequency domain signal allows for the extraction of the phase characteristics of each dominant frequency component. Various methods can be used to extract the amplitude and phase characteristics of each dominant frequency component. For amplitude characteristics, the energy values ​​corresponding to the main frequency components can be directly obtained from the wave energy distribution, and then the amplitude can be obtained through appropriate transformation. For phase characteristics, phase spectrum analysis can be used to determine the phase of the main frequency components by calculating the phase information of the frequency domain signal.

[0097] Step S343: Input the amplitude and phase features into the recurrent neural network unit of the wave adjustment layer to learn the periodic variation law of the wave features.

[0098] Amplitude and phase features are key characteristics of each major frequency component extracted in step S342, representing the intensity and starting position of the fluctuation, respectively. The recurrent neural network (RNN) unit in the fluctuation adjustment layer is a crucial component of the transport prediction model. RNNs possess memory capabilities, enabling them to handle temporal dependencies in sequence data and making them suitable for learning the periodic variations of fluctuation features. The amplitude and phase features are arranged chronologically to form a sequence, which is then input into the RNN unit. Each RNN unit contains multiple neurons, each interacting with the input sequence and the hidden state from the previous time step through connection weights. At each time step, the neurons update their hidden states based on the current input and the hidden state from the previous time step, thus achieving the memorization and processing of sequence information. Over time, the RNN unit continuously learns the periodic variations of the fluctuation features.

[0099] Step S344: The learned periodic variation pattern is converted into a series of fluctuation adjustment coefficients through the output layer of the fluctuation adjustment layer.

[0100] The output layer of the fluctuation adjustment layer is the last part of the fluctuation adjustment layer. Its function is to convert the periodic variation patterns of fluctuation characteristics learned by the recurrent neural network units into a specific sequence of fluctuation adjustment coefficients. After the recurrent neural network units have learned the periodic variation patterns of fluctuation characteristics, the hidden states store information about these patterns. The output layer calculates these hidden states and maps them to a sequence of fluctuation adjustment coefficients. Specifically, the output layer contains a set of neurons, each connected to the hidden states of the recurrent neural network units. These connections have corresponding weights. By weighted summing of the hidden states and processing them through an appropriate activation function, the fluctuation adjustment coefficients for each time step are obtained. The fluctuation adjustment coefficients for each time step are arranged sequentially to form a fluctuation adjustment coefficient sequence. Each coefficient in this sequence represents the degree of adjustment of the short-term fluctuation on the traffic volume at the corresponding time step.

[0101] Step S345: Weight the fluctuation adjustment coefficient sequence with the long-term freight volume forecast characteristics to obtain the fluctuation impact characteristics.

[0102] The fluctuation adjustment coefficient sequence, obtained in step S344, reflects the degree to which short-term fluctuations adjust freight volume at different time steps. The long-term freight volume forecast feature, obtained in step S335, describes the trend of river-sea intermodal freight volume changes over a long period. The weighted combination combines the fluctuation adjustment coefficient sequence and the long-term freight volume forecast feature to account for the impact of short-term fluctuations on long-term freight volume forecasts. Specifically, for each time step's freight volume forecast value in the long-term freight volume forecast feature, the corresponding fluctuation adjustment coefficient is multiplied to obtain the adjusted freight volume forecast value. Arranging the adjusted freight volume forecast values ​​for each time step sequentially yields the fluctuation impact feature.

[0103] Step S350: Perform feature fusion on the long-term freight volume forecast features and the fluctuation impact features to obtain preliminary freight volume forecast features.

[0104] The long-term freight volume forecast feature, obtained in step S335, describes the trend of river-sea intermodal freight volume over a long period, reflecting the impact of long-term factors on freight volume. The fluctuation impact feature, obtained in step S345, comprehensively considers the impact of short-term fluctuations on freight volume, reflecting the fluctuations in freight volume over a short period.

[0105] As one implementation method, step S350 can be specifically implemented as the following steps S351~S355:

[0106] Step S351: Convert the long-term freight volume forecast features and fluctuation impact features into feature vectors of the same dimension.

[0107] Long-term freight volume forecasting features and fluctuation impact features may have different dimensions and data formats in their original form. To achieve effective feature fusion, they need to be converted into feature vectors of the same dimension. Dimension refers to the number of elements in a feature vector; feature vectors of the same dimension imply consistency in their data structure, facilitating subsequent calculations and processing. For long-term freight volume forecasting features and fluctuation impact features, appropriate methods for dimension transformation can be selected based on their specific content and properties. For example, if the long-term freight volume forecasting feature is a sequence containing freight volume forecasts for multiple time steps, and the fluctuation impact feature is a similar sequence but of different lengths, their lengths can be made the same through interpolation or truncation, thus converting them into feature vectors of the same dimension. Alternatively, feature extraction and dimensionality reduction methods can be used to convert the original features into low-dimensional feature vectors with the same dimension.

[0108] Step S352: Perform element-wise multiplication on the long-term freight volume prediction feature vector and the fluctuation impact feature vector to obtain the fused feature vector.

[0109] The long-term traffic volume prediction feature vector and the fluctuation impact feature vector are feature vectors converted to the same dimension in step S351, reflecting the long-term trend and short-term fluctuation of traffic volume, respectively. Element-by-element multiplication refers to multiplying corresponding elements of the two feature vectors to obtain a new vector. For example, for the long-term traffic volume prediction feature vector [a1,a2,a3,...,an] and the fluctuation impact feature vector [b1,b2,b3,...,bn], the fused feature vector obtained by element-by-element multiplication is [a1*b1,a2*b2,a3*b3,...,an*bn]. Through element-by-element multiplication, the long-term traffic volume prediction feature and the fluctuation impact feature can be organically combined. The long-term traffic volume prediction feature reflects the basic trend of traffic volume, while the fluctuation impact feature reflects the adjustment effect of short-term fluctuations on traffic volume.

[0110] Step S353: Normalize the fused feature vector to map the feature values ​​to a preset numerical range.

[0111] The fused feature vector is obtained through element-wise multiplication in step S352, integrating information from long-term transport volume prediction features and fluctuation impact features. Normalization maps the data's value range to a specific interval, such as [0,1] or [-1,1]. The preset numerical interval is a pre-defined range of values ​​used to standardize the feature values ​​of the fused feature vector. Normalization eliminates dimensional differences between different features, making feature values ​​comparable, and also helps improve the stability and convergence speed of subsequent algorithms. Various normalization methods can be used for the fused feature vector, such as min-max normalization and z-score normalization.

[0112] Step S354: Weights are assigned to the normalized fused feature vector through an attention mechanism to enhance the feature components that have a significant impact on traffic volume prediction.

[0113] The normalized fused feature vector, obtained through normalization in step S353, has fused long-term transport volume prediction features and fluctuation impact features, mapping the feature values ​​to a preset numerical range. The attention mechanism automatically assigns different weights to different feature components, making the model focus more on feature components that significantly impact transport volume prediction. In the normalized fused feature vector, different feature components may contribute differently to transport volume prediction. The attention mechanism assigns appropriate weights to feature components based on their importance. Specifically, the attention mechanism calculates an attention score for each feature component, representing its importance. Then, the feature components are weighted according to their attention scores, enhancing important feature components and suppressing less important ones.

[0114] Step S355: Use the fused feature vector after weight allocation as the initial traffic volume prediction feature.

[0115] The weighted fusion feature vector is obtained by weighting the normalized fusion feature vector through an attention mechanism in step S354, which enhances the feature components that significantly affect traffic volume prediction. Using this weighted fusion feature vector as the initial traffic volume prediction feature integrates information from long-term traffic volume prediction features and fluctuation impact features. Furthermore, after normalization processing and optimization by the attention mechanism, it can more accurately reflect the actual changes in river-sea intermodal transport volume.

[0116] Step S360: Divide the preliminary freight volume forecast features into segments. According to the geographical boundaries of the river transport segment and the sea transport segment, the overall freight volume forecast features are divided into river transport segment freight volume forecast features and sea transport segment freight volume forecast features.

[0117] The preliminary freight volume forecast feature, obtained in step S355, comprehensively considers both long-term trends and short-term fluctuations, reflecting the overall freight volume changes in river-sea intermodal transport. The river transport segment and the sea transport segment are two distinct transport segments in the river-sea intermodal transport process, possessing different geographical environments, transport conditions, and transport characteristics. Segmenting according to the geographical boundaries of the river transport segment and the sea transport segment involves assigning the freight volume information in the preliminary freight volume forecast feature to the river transport segment and the sea transport segment respectively, based on the actual geographical area and transport route.

[0118] Step S370: Integrate the freight volume prediction features of the river transport segment and the freight volume prediction features of the sea transport segment into the freight volume prediction features of the intermodal transport segment.

[0119] The river transport segment's freight volume prediction features are derived from the preliminary freight volume prediction features in step S360, reflecting the freight volume changes in the river transport segment. Similarly, the sea transport segment's freight volume prediction features are also derived in step S360, reflecting the freight volume change trend in the sea transport segment. Integration involves merging and combining the freight volume prediction features of these two different segments to form a feature representation that comprehensively reflects the overall freight volume of the river-sea intermodal transport segment. By integrating the river transport segment's freight volume prediction features and the sea transport segment's freight volume prediction features, a unified intermodal transport segment freight volume prediction feature can be obtained. This feature includes freight volume information from both the river and sea transport segments, enabling a more accurate description of the overall freight volume changes in river-sea intermodal transport. During the integration process, various methods can be used, such as combining the river transport segment's freight volume prediction features and the sea transport segment's freight volume prediction features as two-dimensional feature vectors, or weighting and summing the two features based on the proportional relationship between river and sea transport in intermodal transport.

[0120] Step S400: Construct ship transshipment coordination relationships based on the predicted characteristics of intermodal transport segment volume.

[0121] The intermodal transport segment's volume forecast characteristics, obtained in step S370, comprehensively reflect the overall volume changes in the river-sea intermodal transport segment, including volume forecast information for both the river and sea transport segments. Vessel transshipment coordination refers to a vessel scheduling and coordination relationship established during river-sea intermodal transport to achieve efficient transshipment between river and sea transport segments. Constructing vessel transshipment coordination relationships based on the intermodal transport segment's volume forecast characteristics involves rationally arranging the transshipment sequence, time, and ports between different transport segments according to volume forecasts, thereby improving transport efficiency and reducing transport costs. By analyzing the intermodal transport segment's volume forecast characteristics, we can understand the volume demand for river and sea transport in different time periods, thus determining vessel transshipment strategies.

[0122] As one implementation method, step S400 can be specifically implemented as the following steps S410~S450:

[0123] Step S410: Time-align the river transport segment freight volume prediction features and the sea transport segment freight volume prediction features in the intermodal transport segment freight volume prediction features so that the prediction features of the two correspond in the same time unit.

[0124] The river transport segment's freight volume forecast features and the sea transport segment's freight volume forecast features are extracted from the preliminary freight volume forecast features in step S360, reflecting the freight volume changes in the two segments. Due to the different transport characteristics and influencing factors of river and sea transport, their freight volume forecast features may differ over time. Time alignment unifies the river and sea transport segment freight volume forecast features in terms of time, ensuring they have corresponding forecast features within the same time unit. This facilitates subsequent comparison and analysis of the freight volumes in the two segments, thereby more accurately constructing ship transshipment coordination relationships. In practice, the time resolution of the river transport segment's freight volume forecast features might be daily, while the time resolution of the sea transport segment's freight volume forecast features might be every two days. Time alignment unifies them to the same time resolution, for example, converting them both to daily forecast features. Interpolation, sampling, and other methods can be used to achieve time alignment.

[0125] As one implementation method, step S410 can be specifically implemented as the following steps S411~S415:

[0126] Step S411: Obtain the timestamp sequence of the freight volume prediction features for the river transport section and the timestamp sequence of the freight volume prediction features for the sea transport section.

[0127] The river transport segment's freight volume forecast features, obtained in step S360, reflect the changes in freight volume within the river transport segment and include freight volume forecast values ​​at different time points. The timestamp sequence is a sequence of time information corresponding to these freight volume forecast values, with each timestamp representing the specific time corresponding to a freight volume forecast value. Similarly, the sea transport segment's freight volume forecast features reflect the changes in freight volume within the sea transport segment and also have corresponding timestamp sequences. Obtaining the timestamp sequences of the river transport segment's freight volume forecast features and the sea transport segment's freight volume forecast features is the first step in time alignment. By obtaining these two timestamp sequences, the temporal distribution of the river and sea transport freight volume forecast features can be understood.

[0128] Step S412: Perform time reference calibration on the two timestamp sequences to unify them into the same time reference system.

[0129] The two timestamp sequences are the timestamp sequences of river transport segment freight volume prediction characteristics and sea transport segment freight volume prediction characteristics obtained in step S411. They may use different time reference systems due to differences in data collection sources, methods, or precision. Time reference calibration unifies these two timestamp sequences to the same time reference system, making them comparable. For example, the timestamp sequence of river transport segment freight volume prediction characteristics may use local time, while the timestamp sequence of sea transport segment freight volume prediction characteristics may use Coordinated Universal Time (UTC). Time reference calibration converts them to UTC or another unified time reference system. Time conversion formulas and algorithms can be used to achieve time reference calibration.

[0130] Step S413: The calibrated timestamp sequence is gridded according to the preset time unit length, dividing the continuous time into multiple time units of equal length.

[0131] The calibrated timestamp sequence is the timestamp sequence of river transport segment freight volume prediction features and sea transport segment freight volume prediction features unified to the same time reference system after time base calibration in step S412. The preset time unit length is a pre-defined time interval used to divide continuous time into multiple time units of equal length. For example, if the preset time unit length is 1 day, then the continuous time covered by the calibrated timestamp sequence is divided into daily intervals. Gridding is a method of discretizing continuous time. By dividing time into multiple time units of equal length, it is easier to analyze and process freight volume prediction features. In practice, the start and end times of the calibrated timestamp sequence are first determined, and then multiple time units are sequentially divided according to the preset time unit length.

[0132] Step S414: Perform feature aggregation on the river transport segment freight volume prediction features and sea transport segment freight volume prediction features within each time unit, and calculate the feature average value within the time unit.

[0133] Each time unit is obtained by gridding the calibrated timestamp sequence according to a preset time unit length in step S413, and each time unit covers a specific time period. The river transport segment's freight volume prediction features and the sea transport segment's freight volume prediction features are the features reflecting the changes in river and sea transport volumes obtained in step S360, respectively. Within each time unit, there may be multiple freight volume prediction values ​​at different time points. Feature aggregation integrates these freight volume prediction values ​​within the same time unit to calculate a feature value representing that time unit. Calculating the average feature value within a time unit is a commonly used feature aggregation method. For example, for a certain time unit, if the river transport segment's freight volume prediction features contain freight volume prediction values ​​at multiple time points, adding these prediction values ​​and then dividing by the number of prediction values ​​yields the average value of the river transport segment's freight volume prediction features within that time unit. The same method can be used to calculate the average value of the sea transport segment's freight volume prediction features within that time unit. By aggregating the predicted freight volume features of the river transport segment and the sea transport segment within each time unit, and calculating the average feature value within the time unit, the noise and fluctuation of the data can be reduced, making the freight volume prediction features smoother and more stable.

[0134] Step S415: Establish a time correspondence between the average value of the predicted freight volume characteristics of the river transport section and the average value of the predicted freight volume characteristics of the sea transport section corresponding to the time unit, to ensure the correspondence of characteristics in the same time unit.

[0135] The average value of the predicted freight volume features for the river transport segment and the sea transport segment corresponding to each time unit are the average values ​​of the predicted freight volume features for river transport and sea transport within each time unit, calculated separately in step S414. Establishing a time correspondence involves associating these average values ​​according to time units, ensuring that the average values ​​of the predicted freight volume features for river transport and sea transport within the same time unit correspond to each other. By establishing a time correspondence, the predicted freight volume features for river transport and sea transport can be aligned in time, forming a unified dataset. In this dataset, each time unit has corresponding average values ​​of the predicted freight volume features for river transport and sea transport, facilitating subsequent matching degree calculations and the construction of ship transshipment coordination relationships.

[0136] Step S420: Calculate the matching degree sequence between the predicted freight volume features of the river transport section and the predicted freight volume features of the sea transport section. The matching degree sequence is obtained by calculating the feature similarity of the corresponding time units.

[0137] The river transport segment's freight volume forecast features and the sea transport segment's freight volume forecast features are those for which a time correspondence was established in step S415. Each time unit has a corresponding average value for both river and sea transport freight volume forecast features. The matching degree sequence is a sequence reflecting the degree of matching between the river and sea transport freight volume forecast features across different time units. The feature similarity of corresponding time units refers to the degree of similarity between the average value of the river transport segment's freight volume forecast features and the average value of the sea transport segment's freight volume forecast features within the same time unit. By calculating the feature similarity for each time unit and arranging these similarities in chronological order, the matching degree sequence is obtained. Various methods can be used to calculate feature similarity, such as cosine similarity and Euclidean distance. By calculating the matching degree sequence of the river and sea transport freight volume forecast features, the matching situation of river and sea transport freight volumes across different time units can be understood. If the matching degree is high, it indicates that the freight volume demand of river and sea transport is relatively consistent, and the transshipment of ships between river and sea transport can be smoother; conversely, if the matching degree is low, corresponding measures may need to be taken to adjust the transshipment strategy of ships to improve transportation efficiency.

[0138] As one implementation method, step S420 can be specifically implemented as the following steps S421~S424:

[0139] Step S421: Represent the predicted freight volume characteristics of the river transport segment and the sea transport segment within each time unit as vectors to obtain the river transport feature vector and the sea transport feature vector.

[0140] The river transport volume prediction features and sea transport volume prediction features within each time unit are features for which a time correspondence was established in step S415, representing the predicted transport volume for river transport and sea transport within that time unit, respectively. Vector representation converts these features into vector form to facilitate subsequent similarity calculations. The river transport feature vector is formed by organizing and combining the river transport volume prediction features within each time unit. For example, if the river transport volume prediction features include multiple aspects of transport volume information, such as cargo quantity and transport time, then this information can be used as different dimensions of the vector to form the river transport feature vector. Similarly, the sea transport feature vector is formed by converting the sea transport volume prediction features within each time unit into vector form. By representing the river and sea transport volume prediction features as vectors, they can be unified into a vector space, and vector operations can be used to calculate their similarity.

[0141] Step S422: Calculate the cosine similarity between the river transport feature vector and the sea transport feature vector to obtain the time unit matching degree value.

[0142] The river transport feature vector and the sea transport feature vector are obtained by vectorizing the river transport segment freight volume prediction features and the sea transport segment freight volume prediction features within each time unit in step S421. Cosine similarity is a commonly used method to measure the similarity between two vectors, which represents their degree of similarity by calculating the cosine value of the angle between the two vectors. The calculated cosine similarity is used as the time unit matching degree value, reflecting the degree of matching between the river transport and sea transport freight volume prediction features within that time unit.

[0143] Step S423: Arrange the matching degree values ​​of multiple time units in chronological order to obtain a matching degree sequence.

[0144] The matching degree values ​​for multiple time units are obtained by calculating the cosine similarity between the river transport feature vector and the sea transport feature vector within each time unit in step S422. Each matching degree value reflects the degree of matching between the river transport and sea transport volume prediction features within that time unit. Arranging these matching degree values ​​in chronological order means arranging them according to the order of the time units to form a sequence.

[0145] Step S424: Smooth the matching degree sequence by eliminating abrupt changes in matching degree between adjacent time units through moving average, and obtain a smoothed matching degree sequence.

[0146] The matching degree sequence is a sequence of matching degree values ​​from multiple time units arranged chronologically in step S423, reflecting the change in the matching degree of the predicted characteristics of river and sea transport volumes over time. However, due to various factors, abrupt changes in matching degree between adjacent time units may occur in the matching degree sequence, which may interfere with the analysis of the matching degree trend. Using a moving average to eliminate abrupt changes in matching degree between adjacent time units makes the matching degree sequence smoother and more clearly shows the changing trend of the matching degree.

[0147] Step S430: Identify the first matching degree time unit and the second matching degree time unit according to the matching degree sequence. The first matching degree time unit represents the time unit in which the similarity between the predicted features of river transport and sea transport volume is greater than the similarity threshold, and the second matching degree time unit represents the time unit in which the similarity between the predicted features of river transport and sea transport volume is less than the similarity threshold.

[0148] The matching degree sequence is the smoothed matching degree sequence obtained in step S424, reflecting the change in the matching degree of the predicted features of river and sea transport volumes over time. The similarity threshold is a pre-set critical value used to distinguish the high and low similarity between the predicted features of river and sea transport volumes. The first and second matching degree time units are identified based on the matching degree sequence; that is, time units are divided into two categories based on the comparison results of the matching degree values ​​in the matching degree sequence with the similarity threshold. The first matching degree time unit refers to the time unit where the similarity between the predicted features of river and sea transport volumes is greater than the similarity threshold. Within these time units, the demand for river and sea transport volumes is relatively consistent, and the transshipment of ships between river and sea can be smoother. The second matching degree time unit refers to the time unit where the similarity between the predicted features of river and sea transport volumes is less than the similarity threshold. Within these time units, the demand for river and sea transport volumes differs significantly, and different ship transshipment strategies may need to be adopted. In actual operation, each matching degree value in the matching degree sequence is traversed and compared with the similarity threshold. If the matching score is greater than the similarity threshold, the time unit corresponding to that matching score is marked as the first matching score time unit; if the matching score is less than the similarity threshold, the corresponding time unit is marked as the second matching score time unit. By identifying the first and second matching score time units, corresponding ship transshipment coordination strategies can be formulated based on different matching situations, thereby improving the transportation efficiency of river-sea intermodal transport.

[0149] As one implementation method, step S430 can be specifically implemented as the following steps S431~S435:

[0150] Step S431: Set a matching degree threshold and mark the time units in the matching degree sequence that are greater than or equal to the matching degree threshold as candidate first matching degree time units.

[0151] The matching degree sequence is the smoothed matching degree sequence obtained in step S424, reflecting the change in the matching degree of the predicted features of river and sea transport volumes over time. The matching degree threshold is a pre-set critical value used to distinguish the high and low similarity between the predicted features of river and sea transport volumes. Setting the matching degree threshold requires comprehensive consideration of the actual situation and transportation demand of river-sea intermodal transport. For example, if it is desired to more strictly filter out time units with high matching degrees between river and sea transport volumes, the matching degree threshold can be set higher; if it is desired to broaden the range of time units with high matching degrees, the matching degree threshold can be set lower. Marking time units in the matching degree sequence that are greater than or equal to the matching degree threshold as candidate first matching degree time units is a preliminary screening of the matching degree sequence based on the matching degree threshold.

[0152] Step S432: Mark the time units in the matching degree sequence that are less than the matching degree threshold as candidate second matching degree time units.

[0153] The matching degree sequence is the smoothed matching degree sequence obtained in step S424, and the matching degree threshold is a pre-set critical value in step S431 used to distinguish the high and low similarity of river transport and sea transport volume prediction features. Marking time units in the matching degree sequence that are less than the matching degree threshold as candidate second matching degree time units is the reverse of the screening in step S431. By marking candidate second matching degree time units, the range of time units with low similarity between river transport and sea transport volume prediction features can be preliminarily determined.

[0154] Step S433: Perform continuity analysis on the candidate first matching degree time units to identify the high matching degree time windows obtained by continuously appearing candidate first matching degree time units.

[0155] The candidate first matching degree time unit is the time unit in the matching degree sequence marked in step S431 that is greater than or equal to the matching degree threshold. Continuity analysis refers to checking whether these candidate first matching degree time units appear consecutively. A high matching degree time window is a time period composed of consecutively appearing candidate first matching degree time units. For example, if the candidate first matching degree time units are time unit 1, time unit 2, time unit 4, time unit 5, and time unit 6, where time unit 1 and time unit 2 are consecutive, and time unit 4, time unit 5, and time unit 6 are consecutive, then two high matching degree time windows can be obtained, namely [time unit 1, time unit 2] and [time unit 4, time unit 5, time unit 6].

[0156] Step S434: Perform continuity analysis on the candidate second matching degree time units to identify the low matching degree time windows obtained by the consecutively appearing candidate second matching degree time units.

[0157] Candidate second-match time units are time units in the match degree sequence marked in step S432 that are less than the match degree threshold. Continuity analysis checks whether these candidate second-match time units appear consecutively. A low-match time window is a time period composed of consecutively appearing candidate second-match time units. For example, if the candidate second-match time units are time units 3, 7, 8, and 9, where time units 7, 8, and 9 are consecutive, then a low-match time window [time unit 7, time unit 8, time unit 9] can be obtained. By performing continuity analysis on the candidate second-match time units and identifying low-match time windows, time periods with low and persistent similarity in river and sea transport volume prediction characteristics can be determined. Within these low-match time windows, the transport volume demand of river and sea transport differs significantly, requiring different ship transshipment strategies. The low-match time windows provide a time range basis for subsequently establishing buffer transshipment relationships for low-match situations.

[0158] Step S435: Determine the time unit within the high matching degree time window as the first matching degree time unit, and determine the time unit within the low matching degree time window as the second matching degree time unit.

[0159] The high-match-degree time window is obtained in step S433 by performing a continuity analysis on the candidate first-match-degree time units, and consists of a time period composed of consecutively occurring candidate first-match-degree time units. The low-match-degree time window is obtained in step S434 by performing a continuity analysis on the candidate second-match-degree time units, and consists of a time period composed of consecutively occurring candidate second-match-degree time units. Determining the time units within the high-match-degree time window as first-match-degree time units means that within these time units, the predicted characteristics of river transport and sea transport volumes are highly similar, and the transshipment of ships between river and sea transport can proceed relatively smoothly. Determining the time units within the low-match-degree time window as second-match-degree time units indicates that within these time units, the predicted characteristics of river transport and sea transport volumes are less similar, requiring special transshipment strategies to address volume differences.

[0160] Step S440: Establish a direct transport relationship for the first matching degree time unit and a buffer transport relationship for the second matching degree time unit.

[0161] The first matching time unit is the time unit where the predicted characteristics of river transport and sea transport volumes are highly similar, as determined in step S435. Within these time units, the demand for river transport and sea transport is relatively consistent, allowing for smoother transshipment between the two modes. Direct transshipment means that within the first matching time unit, ships can directly transship from the river transport segment to the sea transport segment without additional buffering or waiting. Establishing direct transshipment relationships improves transport efficiency and reduces cargo transshipment time and costs. The second matching time unit is the time unit where the predicted characteristics of river transport and sea transport volumes are less similar, as determined in step S435. Within these time units, the demand for river transport and sea transport differs, requiring buffering measures to balance the volumes. Buffer transshipment relationships refer to setting up buffer zones or adopting buffer strategies within the second matching time unit to temporarily store cargo or adjust transshipment times, alleviating the mismatch between river and sea transport volumes. By establishing direct and buffer transshipment relationships for different matching time units, transshipment strategies can be flexibly adjusted based on volume matching, improving the overall transport efficiency of river-sea intermodal transport.

[0162] As one implementation method, step S440 can be specifically implemented as the following steps S441~S444:

[0163] Step S441: Based on the predicted transport volume characteristics of the river transport section and the sea transport section corresponding to the first matching degree time unit, determine the departure port sequence of the vessel in the river transport section and the arrival port sequence of the vessel in the sea transport section.

[0164] The first matching time unit is the time unit with a high similarity between the river transport and sea transport volume prediction features determined in step S435. The corresponding river transport segment volume prediction features and sea transport segment volume prediction features reflect the volume prediction situation of river transport and sea transport in these time units, respectively. Determining the departure port sequence of ships in the river transport segment and the arrival port sequence of ships in the sea transport segment is based on the volume prediction features to rationally arrange the transport routes of ships.

[0165] Step S442: Match the departure port sequence and the arrival port sequence to establish a direct correspondence between the departure ports of river transport and the arrival ports of sea transport.

[0166] The departure port sequence is the sequence of departure ports of ships in the river transport segment determined in step S441 based on the predicted transport volume characteristics of the river transport segment corresponding to the first matching degree time unit. The arrival port sequence is the sequence of arrival ports of ships in the sea transport segment determined based on the predicted transport volume characteristics of the sea transport segment. Port matching involves reasonably pairing the ports in the departure and arrival port sequences to establish a direct correspondence between river transport departure ports and sea transport arrival ports. Multiple factors need to be considered when performing port matching, such as distance, transport time, transport cost, and cargo type. For example, ports with closer distances, shorter transport times, and lower transport costs are prioritized for matching. Algorithms and models, such as the Hungarian algorithm and genetic algorithms, can be used to implement port matching.

[0167] Step S443: Determine the recommended loading capacity characteristics of the vessel in the corresponding time unit based on the volume prediction characteristics of the first matching degree time unit.

[0168] The volume forecast feature for the first matching time unit is the volume forecast corresponding to the time unit with high similarity between the river transport and sea transport volume forecast features determined in step S435. It includes cargo transportation demand information for river transport and sea transport within these time units. The recommended loading capacity feature refers to the appropriate loading capacity determined for the vessel in the corresponding time unit based on the volume forecast feature. For example, based on the river transport segment volume forecast feature, we can know the quantity of cargo exported from a certain river transport departure port within the first matching time unit; based on the sea transport segment volume forecast feature, we can know the cargo receiving capacity of a certain sea transport arrival port. Taking into account this information, combined with factors such as vessel type, carrying capacity, and transportation costs, the recommended loading capacity for the vessel in the corresponding time unit is determined. In practice, a mathematical model can be established to calculate the recommended loading capacity, such as a linear programming model. A linear programming model can maximize transportation efficiency while satisfying various constraints (such as vessel weight limits, transportation time limits, etc.), thereby determining the optimal recommended loading capacity. By determining the recommended loading capacity of a vessel in the corresponding time unit based on the volume prediction characteristics of the first matching degree time unit, the loading capacity of the vessel can be reasonably arranged, transportation efficiency can be improved, and the occurrence of empty or overloaded vessels can be avoided.

[0169] Step S444: Integrate the port correspondence and recommended loading volume characteristics into a direct transshipment relationship.

[0170] The port correspondence is the direct correspondence between the departure ports of river transport and the arrival ports of sea transport established in step S442, clarifying the transshipment routes of vessels between river and sea transport. The recommended loading capacity characteristic is the appropriate loading capacity of the vessel in the corresponding time unit, determined in step S443 based on the transport volume prediction characteristics of the first matching degree time unit. Integration involves merging and unifying the port correspondence and the recommended loading capacity characteristic to form a complete direct transshipment relationship. The direct transshipment relationship includes the transshipment route and loading capacity information of the vessel within the first matching degree time unit, providing specific guidance for vessel scheduling and transportation.

[0171] Step S450: Integrate the direct transshipment relationship and the buffer transshipment relationship into a ship transshipment coordination relationship.

[0172] The direct transshipment relationship, established in step S440 for the first matching degree time unit, clarifies the transshipment routes and loading capacity of vessels within time units where the predicted characteristics of river and sea transport volumes are highly similar. The buffer transshipment relationship, established in step S440 for the second matching degree time unit, is used to balance transport volumes within time units where the predicted characteristics of river and sea transport volumes are less similar by setting buffer zones or adjusting transshipment times. Integration merges and unifies the direct and buffer transshipment relationships to form a complete vessel transshipment coordination relationship. This vessel transshipment coordination relationship comprehensively considers the matching of river and sea transport volumes, enabling the rational arrangement of vessel transshipment based on different time units and transport volume characteristics. For example, for different time units, direct or buffer transshipment methods are adopted depending on whether they belong to the first or second matching degree time unit.

[0173] Step S500: Formulate a vessel cross-segment scheduling plan based on the vessel transfer coordination relationship. The vessel cross-segment scheduling plan is used to coordinate the transfer and scheduling operations of vessels between river transport segments and sea transport segments.

[0174] As one implementation method, step S500 can be specifically implemented as the following steps S510~S550:

[0175] Step S510: Convert the direct transfer relationship in the ship transfer coordination relationship into a ship priority scheduling sequence, which represents the transfer order of ships in the first matching degree time unit.

[0176] The direct transshipment relationship in the vessel transshipment coordination is established in step S440 for the first matching degree time unit, and includes the direct correspondence between the river transport departure port and the sea transport arrival port, as well as information such as the recommended loading capacity of the vessel. The vessel priority scheduling sequence refers to the vessel transshipment order determined according to certain rules within the first matching degree time unit.

[0177] Specifically, the port correspondences in direct transshipment relationships can be sorted according to time unit order to obtain a time unit-port pair sequence. Then, the volume prediction features in each time unit-port pair are compared, and the port pairs are sorted in descending order of volume prediction features. Based on the sorting results, the priority transshipment port pairs for ships in each time unit are determined. Then, the priority transshipment port pairs are arranged according to time unit order to obtain a ship priority scheduling sequence. Conflict detection is performed on the ship priority scheduling sequence to identify port resource conflicts within the same time unit. Conflicting scheduling sequences are adjusted by delaying the transshipment time of some ships to resolve the conflicts, resulting in an adjusted ship priority scheduling sequence.

[0178] Step S520: Convert the buffer transfer relationship in the ship transfer coordination relationship into a ship delay scheduling sequence, which represents the transfer order of ships in the second matching degree time unit.

[0179] The buffer transshipment relationship in the ship transshipment coordination is established in step S440 for the second matching degree time unit. It is used to deal with situations where the similarity of the predicted characteristics of river transport and sea transport volume is low, and to balance the transport volume by setting buffer areas or adjusting transshipment times. The ship delay scheduling sequence refers to the ship transshipment order determined according to certain rules within the second matching degree time unit, where the ship transshipment time will be delayed according to the buffer transshipment relationship.

[0180] For example, the second matching degree time units in the buffer transfer relationship can be arranged in chronological order to obtain a time unit sequence. Then, the difference between the predicted freight volume characteristics of the river transport section and the sea transport section corresponding to each time unit is calculated to obtain a freight volume difference sequence. The freight volume difference sequence is then sorted into time units according to the absolute value of the difference from smallest to largest to obtain a buffer priority sequence. The transfer order of ships in the second matching degree time units is determined based on the buffer priority sequence. Finally, the transfer order is combined with the corresponding time units to obtain a ship delay scheduling sequence.

[0181] Step S530: Integrate the ship priority scheduling sequence and the ship delay scheduling sequence into a comprehensive ship scheduling sequence.

[0182] The vessel priority scheduling sequence, obtained in step S516, is the adjusted transfer order of vessels in the first matching degree time unit. This resolves port resource conflicts and ensures efficient transfer of vessels within time units with high capacity matching degree. The vessel delay scheduling sequence, obtained in step S525, is the delayed transfer arrangement of vessels in the second matching degree time unit. It is used to balance the capacity differences between river transport and sea transport in time units with low capacity matching degree. Integration involves merging and unifying these two sequences to form a complete comprehensive vessel scheduling sequence. The comprehensive vessel scheduling sequence comprehensively considers the capacity matching of river transport and sea transport, covering the transfer order of vessels in different matching degree time units.

[0183] Specifically, the timelines of the ship priority scheduling sequence and the ship delay scheduling sequence can be merged, integrating the time units of the two sequences into a unified timeline according to their chronological order. Then, the transfer operations corresponding to the ship priority scheduling sequence and the ship delay scheduling sequence are marked on the unified timeline. Next, the transfer operations within overlapping time units are prioritized, placing the transfer operations in the ship priority scheduling sequence before those in the ship delay scheduling sequence. A time continuity check is performed on the prioritized transfer operation sequence to ensure that the time interval between adjacent transfer operations meets the ship's sailing time requirements. Transfer operations that do not meet the time interval requirements are adjusted in time, resolving time conflicts by advancing or delaying the transfer operation times, thus obtaining the comprehensive ship scheduling sequence.

[0184] Step S540: Determine the sailing order of ships in the river transport section and the sailing order of ships in the sea transport section based on the integrated ship scheduling sequence.

[0185] The integrated vessel scheduling sequence, obtained in step S535, is a time-adjusted sequence that meets the vessel navigation time requirements, specifying the transfer order of vessels in different time units. The navigation order of vessels in the river transport section and the navigation order of vessels in the sea transport section are determined based on the integrated vessel scheduling sequence.

[0186] Step S550: Integrate the navigation sequence of ships in the river transport section, the navigation sequence of ships in the sea transport section, and the corresponding time units into a cross-section ship scheduling scheme.

[0187] The navigation sequence of vessels in the river transport section is the order in which vessels sail in the river transport section as determined in step S540 based on the comprehensive vessel scheduling sequence. The navigation sequence of vessels in the sea transport section is the order in which vessels sail in the sea transport section, and the corresponding time unit is the time unit corresponding to each navigation task. Integration involves merging and unifying these three elements to form a complete cross-segment vessel scheduling scheme.

[0188] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention, such as Euclidean distance algorithm, cosine distance algorithm, conflict resolution algorithm, etc., can all be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solution of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set the threshold based on historical data, experience or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select the activation function, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0189] Please see Figure 2 , Figure 2 This is a schematic diagram of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 can be connected via a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computing and control core of the computer system, capable of parsing various instructions and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and can be used to send and receive data under the control of the processor 101; the communication interface 102 can also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system used to store programs and data. It is understood that the memory 103 here can include the computer system's built-in memory, or it can include extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system; this invention does not limit this storage space. In one embodiment, the processor 101 executes the river-sea intermodal transport scheduling optimization method based on transport volume prediction provided in the above embodiments of the present invention by running a computer program in the memory 103.

Claims

1. A river-sea intermodal transportation scheduling optimization method based on traffic volume prediction, characterized in that, The method comprises: acquiring river-sea intermodal transportation historical operation information, the river-sea intermodal transportation historical operation information comprising ship navigation track information, port operation time sequence information and cargo transfer connection information; performing intermodal transportation volume feature extraction operation on the river-sea intermodal transportation historical operation information to obtain intermodal transportation volume influence features; carrying out volume trend analysis according to the intermodal transportation volume influence features to generate intermodal transportation section volume prediction features; constructing ship transfer coordination relationship based on the intermodal transportation section volume prediction features; formulating ship cross-section scheduling scheme according to the ship transfer coordination relationship, the ship cross-section scheduling scheme being used for coordinating ship transfer scheduling operation between river transportation section and sea transportation section; the operation of performing intermodal transportation volume feature extraction operation on the river-sea intermodal transportation historical operation information to obtain intermodal transportation volume influence features comprises: performing track pattern analysis on river transportation section track data in the ship navigation track information, and discretizing continuous ship navigation track in the river transportation section into coordinate point sequence according to preset spatial resolution, calculating direction deflection angle sequence and distance change sequence between adjacent coordinate points, the direction deflection angle sequence representing change of ship navigation direction, and the distance change sequence representing change of ship navigation speed; performing time sequence pattern recognition on cargo loading and unloading records in the port operation time sequence information, extracting cargo loading and unloading sequence features and equipment use alternation features, the loading and unloading sequence features representing processing sequence of different cargo types in the port, and the equipment use alternation features representing switching sequence of different operation equipment; performing node association analysis on transfer node data in the cargo transfer connection information to obtain cargo residence time length features and transfer connection smoothness features in the transfer node, the residence time length features representing processing time consumption of the cargo in the transfer node, and the transfer connection smoothness features representing connection closeness between different transfer nodes; performing feature fusion on the direction deflection angle sequence, the distance change sequence, the loading and unloading sequence features, the equipment use alternation features, the residence time length features and the transfer connection smoothness features to obtain intermodal transportation volume influence features; the operation of carrying out volume trend analysis according to the intermodal transportation volume influence features to generate intermodal transportation section volume prediction features comprises: expanding ship navigation track features in the intermodal transportation volume influence features into time sequence according to time sequence, and arranging the ship navigation track features into track time sequence according to time sequence; extracting trend from the track time sequence, and separating long-term trend component and short-term fluctuation component through moving average algorithm, the long-term trend component representing change direction of volume in long time range, and the short-term fluctuation component representing fluctuation change of volume in short time; inputting the long-term trend component into trend prediction layer of volume prediction model to obtain long-term volume prediction features; inputting the short-term fluctuation component into fluctuation adjustment layer of the volume prediction model to obtain fluctuation influence features; performing feature fusion on the long-term volume prediction features and the fluctuation influence features to obtain preliminary volume prediction features; Segmenting the preliminary traffic volume prediction features, dividing the overall traffic volume prediction features into river transportation section traffic volume prediction features and sea transportation section traffic volume prediction features according to geographical boundaries of river transportation sections and sea transportation sections; Integrating the river transportation section traffic volume prediction features and the sea transportation section traffic volume prediction features into intermodal transportation section traffic volume prediction features.

2. The method of claim 1, wherein, The river transportation section trajectory data in the ship navigation trajectory information is analyzed in terms of trajectory patterns, the continuous navigation trajectory of the ship in the river transportation section is discretized into a coordinate point sequence according to a preset spatial resolution, a sequence of direction deflection angles and a sequence of distance changes between adjacent coordinate points are calculated, including: The river transportation section trajectory data in the ship navigation trajectory information is sampled in terms of coordinate points according to a preset spatial resolution, uniform-interval coordinate points are selected from the continuous navigation trajectory of the ship in the river transportation section, and a coordinate point sequence is obtained; The direction vectors between adjacent coordinate points in the coordinate point sequence are calculated, the deflection angles between adjacent direction vectors are calculated through an inverse tangent function, and a sequence of direction deflection angles is obtained; The ratio of the straight-line distance between adjacent coordinate points in the coordinate point sequence to the corresponding time interval is calculated, and a sequence of distance changes is obtained; The sequence of direction deflection angles is filtered through a sliding window, and the mean value of continuous direction deflection angles is calculated through a sliding window of a preset length to eliminate instantaneous angle mutation interference; The sequence of distance changes is filtered through a sliding window, and the mean value of continuous distance changes is calculated through a sliding window of the same preset length to eliminate instantaneous speed fluctuation interference; The filtered sequence of direction deflection angles and the filtered sequence of distance changes are integrated into the trajectory pattern features of the ship in the river transportation section.

3. The method of claim 1, wherein, The cargo loading and unloading records in the port operation time sequence information are subjected to time sequence pattern recognition, the loading and unloading sequence features of the cargo in the port and the equipment use alternation features are extracted, including: The cargo loading and unloading records in the port operation time sequence information are arranged in chronological order into an operation event sequence, each operation event containing a cargo type identifier and an operation timestamp; The operation event sequence is subjected to sequence pattern mining, cargo type combination sequences with an occurrence frequency exceeding a preset frequency are identified, the position distribution features and the transfer probability features of different cargo types in the cargo type combination sequences are counted, the position distribution features represent the occurrence position regularity of the cargo types in the sequence, and the transfer probability features represent the probability distribution of one cargo type followed by another cargo type; The equipment operation records in the port operation time sequence information are subjected to state conversion analysis, the start sequence features and the stop sequence features of different operation equipment are identified, the start sequence features represent the sequence of starting operation of different types of equipment, and the stop sequence features represent the sequence of stopping operation of different types of equipment; The position distribution features, the transfer probability features, the start sequence features, and the stop sequence features are integrated to obtain the loading and unloading sequence features of the cargo in the port and the equipment use alternation features; The node correlation analysis is performed on the transfer node data in the cargo transfer connection information, and a stay duration feature and a transfer connection smoothness feature of the cargo at a transfer node are obtained, including: Arrival time records and departure time records of the cargo at each transfer node are extracted from the cargo transfer connection information, and a stay duration sequence of the cargo at each transfer node is calculated, the stay duration sequence being obtained by a difference between the departure time record and the arrival time record; Distribution feature analysis is performed on the stay duration sequence, and a stay duration distribution mode and a stay duration fluctuation range of different cargo types at the transfer node are identified, the stay duration distribution mode representing a probability distribution form of the stay duration, and the stay duration fluctuation range representing a numerical variation interval of the stay duration; Sequence analysis is performed on transfer time interval records between adjacent transfer nodes in the cargo transfer connection information, and an abnormal fluctuation point and a stable change section in the transfer time interval are identified, the abnormal fluctuation point representing an abnormal value deviating from a conventional transfer time interval, and the stable change section representing a continuous time period in which the time interval remains stable; The stay duration distribution mode and the stay duration fluctuation range are integrated into the stay duration feature of the cargo at the transfer node, and the abnormal fluctuation point and the stable change section are integrated into the transfer connection smoothness feature of the cargo.

4. The method of claim 1, wherein, The ship navigation trajectory feature in the intermodal freight volume influence feature is time series unfolded, and the ship navigation trajectory feature is arranged into a trajectory time series in time sequence, including: The ship navigation trajectory feature in the intermodal freight volume influence feature is time axis aligned, and ship navigation trajectory features in different time periods are unified into the same time reference system; The ship navigation trajectory feature is decomposed into a plurality of time segments according to a preset time granularity, each time segment containing the ship navigation trajectory feature in the corresponding time period; Feature extraction is performed on the ship navigation trajectory feature in each time segment, and a trajectory center point feature and a trajectory dispersion degree feature of each time segment are obtained, the trajectory center point feature representing an average position of the ship in the time segment, and the trajectory dispersion degree feature representing a position dispersion degree of the ship in the time segment; The trajectory center point feature and the trajectory dispersion degree feature are arranged in time sequence, and the trajectory time series is obtained.

5. The method of claim 1, wherein, The trend extraction is performed on the trajectory time series, and a long-term trend component and a short-term fluctuation component are separated by a moving average algorithm, including: A sliding window size is set, and the trajectory time series is processed by a sliding window, and an average value of the trajectory feature in each window is calculated; An average value sequence of the continuous window is taken as a long-term trend component of the trajectory time series; A difference value calculation is performed on the original trajectory time series and the long-term trend component, and a short-term fluctuation component of the trajectory time series is obtained; The short-term fluctuation component is smoothed, and high-frequency noise interference is eliminated by Gaussian filtering, and a smoothed short-term fluctuation component is obtained; The long-term trend component is input into a trend prediction layer of the freight volume prediction model, and a long-term freight volume prediction feature is obtained, including: inputting the long-term trend component into an input layer of a trend prediction layer, mapping the long-term trend component into a high-dimensional trend feature vector through a fully connected network; inputting the high-dimensional trend feature vector into an LSTM unit of the trend prediction layer, memorizing time dependence in the long-term trend feature through a gating mechanism; converting a hidden state of the LSTM unit into a long-term traffic volume prediction value sequence through an output layer of the trend prediction layer; performing time dimension expansion on the long-term traffic volume prediction value sequence, and generating traffic volume prediction values of future multiple time units according to a preset prediction time length; integrating the traffic volume prediction values of the future multiple time units into a long-term traffic volume prediction feature.

6. The method of claim 1, wherein, The inputting of the short-term fluctuation component into a fluctuation adjustment layer of the traffic volume prediction model to obtain a fluctuation influence feature includes: performing frequency spectrum analysis on the short-term fluctuation component, and converting fluctuation features in a time domain into fluctuation energy distribution in a frequency domain through Fourier transform; identifying main frequency components in the fluctuation energy distribution, and extracting amplitude features and phase features of each main frequency component; inputting the amplitude features and the phase features into a recurrent neural network unit of the fluctuation adjustment layer to learn periodic variation laws of the fluctuation features; converting the learned periodic variation laws into a fluctuation adjustment coefficient sequence through an output layer of the fluctuation adjustment layer; weighting and combining the fluctuation adjustment coefficient sequence and the long-term traffic volume prediction feature to obtain the fluctuation influence feature; The feature fusion of the long-term traffic volume prediction feature and the fluctuation influence feature to obtain a preliminary traffic volume prediction feature includes: converting the long-term traffic volume prediction feature and the fluctuation influence feature into feature vectors of the same dimension; performing element-by-element multiplication on the long-term traffic volume prediction feature vector and the fluctuation influence feature vector to obtain a fusion feature vector; performing normalization processing on the fusion feature vector to map feature values into a preset numerical interval; performing weight distribution on the normalized fusion feature vector through an attention mechanism to enhance feature components that have a significant influence on traffic volume prediction; taking the fusion feature vector after the weight distribution as the preliminary traffic volume prediction feature.

7. The method of claim 1, wherein, The construction of a ship transshipment coordination relationship based on the intermodal section traffic volume prediction feature includes: time aligning a river transportation section traffic volume prediction feature and a sea transportation section traffic volume prediction feature in the intermodal section traffic volume prediction feature, so that the prediction features of the two on the same time unit correspond to each other; calculating a matching degree sequence of the river transportation section traffic volume prediction feature and the sea transportation section traffic volume prediction feature, the matching degree sequence being calculated through feature similarity of corresponding time units; identifying a first matching degree time unit and a second matching degree time unit according to the matching degree sequence, the first matching degree time unit representing a time unit in which the similarity of the river transportation and the sea transportation traffic volume prediction features is greater than a similarity threshold, and the second matching degree time unit representing a time unit in which the similarity of the river transportation and the sea transportation traffic volume prediction features is less than the similarity threshold; establishing a direct transshipment relationship for the first matching degree time unit, and establishing a buffer transshipment relationship for the second matching degree time unit; The direct transport relationship and the buffer transport relationship are integrated into a ship transport coordination relationship.

8. A computer system, characterized by Comprise: A memory, in which a computer program is stored; The processor is configured to load the computer program to implement the method of any one of claims The method for optimizing river-sea intermodal dispatching based on traffic volume prediction.

Images