Method for port multimodal transport path planning by using AI big language model

Abstract

The invention relates to the technical field of data processing, in particular to a method for port multimodal transport path planning by using an AI big language model. Comprising the following steps: acquiring and preprocessing multi-modal data and historical multi-modal data to obtain preprocessed multi-modal data and preprocessed historical multi-modal data; obtaining historical comprehensive data based on the preprocessed historical multi-modal data, and constructing a path planning model to obtain a path planning result; obtaining real-time comprehensive data based on the preprocessed multi-modal data, and processing the real-time comprehensive data through a path planning model to obtain a preliminary path plan; and introducing multi-dimensional influence factors, and optimizing the preliminary path planning by using a multi-objective optimization algorithm to obtain an optimized path planning. The technical problems that port multimodal transport path planning is poor in self-adaption and low in efficiency and safety are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method for port multimodal transport route planning using an AI large language model. Background Art

[0002] At present, port multimodal transport route planning faces many challenges. Traditional route planning methods mostly rely on static optimization algorithms or manual intervention, and cannot effectively cope with the changing transport environment and real-time data. With the continuous increase of globalization and the volume of goods transported, the transport tasks of ports have become more complex, involving an increasing number of transport modes (such as sea transport, railway, road, etc.), and the characteristics and limiting factors of each transport mode are different, so the difficulty of route planning has also increased. Especially in the port multimodal transport system, not only the timeliness and cost of transportation need to be considered, but also multi-dimensional factors such as weather changes, traffic conditions, port handling capacity, and the availability of transport tools need to be dynamically considered, which makes route planning a complex, multi-objective optimization problem.

[0003] Existing route planning methods usually rely on traditional algorithms, which are mostly used for calculating the shortest path in a static environment and are suitable for relatively simple scenarios. In the complex environment of multimodal transport, these methods often cannot handle dynamic traffic flow, sudden weather conditions, changes in port handling capacity, etc., resulting in low efficiency and accuracy of route planning. In addition, traditional route planning methods lack the ability to efficiently process multi-modal data and cannot fully integrate information from multiple different data sources (such as real-time traffic, weather, port status, etc.), resulting in the inability to comprehensively analyze possible problems in the transportation process, thus affecting the final route planning effect.

[0004] At the same time, the above existing technologies also have technical problems such as poor adaptability, low efficiency and low safety in port multimodal transport route planning. Summary of the Invention

[0005] The present invention provides a method for port multimodal transport route planning using an AI large language model to solve the technical problems of poor adaptability, low efficiency and low safety in port multimodal transport route planning.

[0006] A method for port multimodal transport route planning using an AI large language model according to the present invention specifically includes the following technical solutions:

[0007] A method for port multimodal transport route planning using an AI large language model includes the following steps:

[0008] S1. Obtain and preprocess multi-modal data and historical multi-modal data to get preprocessed multi-modal data and preprocessed historical multi-modal data; Based on the preprocessed historical multi-modal data, obtain historical comprehensive data, and construct a path planning model to get a path planning result;

[0009] S2. Process the preprocessed multi-modal data through a large language model based on AI to get real-time comprehensive data, and at the same time introduce a missing data filling algorithm based on a compensation mechanism to fill in the data missing when using the large language model based on AI to process the preprocessed multi-modal data; The missing data filling algorithm based on the compensation mechanism compensates and fills the missing data, and makes inferences based on the temporal characteristics of the preprocessed historical multi-modal data to minimize the impact brought by the missing data;

[0010] Process the real-time comprehensive data through the path planning model to get a preliminary path planning; Introduce multi-dimensional influencing factors, and use a multi-objective optimization algorithm to optimize the preliminary path planning to get an optimized path planning.

[0011] Preferably, the S1 specifically includes:

[0012] Process the preprocessed historical multi-modal data through a large language model based on AI to get historical comprehensive data; Based on the historical comprehensive data, use a neural network model to construct a path planning model.

[0013] Preferably, the S1 specifically includes:

[0014] During the process of model construction, input the historical comprehensive data into the input layer, and in the input layer, convert the historical comprehensive data into a vector representation through the embedding layer and use it as the output of the input layer; Process the vector-represented historical comprehensive data using an encoder to get the output of the encoder; The decoder processes according to the output of the encoder to get a path planning result, send the path planning result to the output layer, and the output layer generates a path planning suggestion according to the path planning result and uses it as the output of the output layer.

[0015] Preferably, the S2 specifically includes:

[0016] During the implementation process of the missing data filling algorithm based on the compensation mechanism, subtract the predicted value of the missing preprocessed multi-modal data from the actual preprocessed multi-modal data at the previous moment to get a prediction deviation, multiply the prediction deviation by a compensation quantization coefficient to get a corrected deviation, and add the predicted value of the missing preprocessed multi-modal data to the corrected deviation to get the preprocessed multi-modal data after compensation filling.

[0017] Preferably, the S2 specifically includes:

[0018] Process the real-time comprehensive data using the path planning model constructed in step S1 to obtain a preliminary path plan, including: path name, path selection, estimated time, cost estimate, and transportation mode selection.

[0019] Preferably, the S2 specifically includes:

[0020] Introduce multi-dimensional influencing factors, and based on the multi-dimensional influencing factors, introduce a multi-objective optimization algorithm, set optimization objectives, define an objective function according to the optimization objectives, and at the same time set constraint conditions. Then use a heuristic algorithm to combine the multi-dimensional influencing factors in path planning, dynamically evaluate and select the optimal path, and optimize and adjust the preliminary path plan to obtain an optimized path plan.

[0021] The beneficial effects of the technical solution of the present invention are:

[0022] 1. The large language model based on AI can identify and capture complex patterns, temporal dependencies, and interactions between multiple factors in port transportation based on preprocessed historical multi-modal data and real-time multi-modal data. Through learning historical comprehensive data, the large language model based on AI can automatically extract key features in path planning, thus making intelligent decisions between multiple transportation modes. Especially when dealing with complex path planning problems involving multiple intertwined influencing factors, the large language model based on AI has strong generalization ability and flexibility, making path planning more intelligent and automated.

[0023] 2. By introducing multi-dimensional influencing factors to optimize the preliminary path plan, the path selection can be adjusted in real time in a dynamic environment.

[0024] 3. By introducing a missing data filling algorithm based on a compensation mechanism, reasonable missing data filling can be performed according to preprocessed historical multi-modal data and temporal characteristics, ensuring data integrity in the path planning process, reducing wrong decisions caused by data missing, and greatly improving the reliability and robustness of the path planning system in the case of incomplete data. Brief Description of the Drawings

[0025] Figure 1 It is a flowchart of a method for multi-modal transport path planning in a port using an AI large language model according to the present invention. Detailed Embodiments

[0026] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0028] The following specifically describes in conjunction with the accompanying drawings the specific solution of a method for port multimodal transport route planning using an AI large language model provided by the present invention.

[0029] Refer to the attached Figure 1 , which shows a flowchart of a method for port multimodal transport route planning using an AI large language model provided by an embodiment of the present invention. The method includes the following steps:

[0030] S1. Obtain and preprocess multimodal data and historical multimodal data to obtain preprocessed multimodal data and preprocessed historical multimodal data; based on the preprocessed historical multimodal data, obtain historical comprehensive data, and construct a route planning model to obtain a route planning result;

[0031] Obtain multimodal data through means such as sensors, API interfaces, and existing databases. The multimodal data includes the status of transportation tools in the port (such as the status of cargo handling equipment, the stacking of containers, and the operating status of transportation tools (such as trailers, cranes, etc.) (such as idle, working, faulty, etc.)), shipping schedule information (such as the scheduled voyage, speed, arrival time, and expected departure time of each ship), the real-time traffic conditions of railway and highway networks (such as the real-time traffic flow, congestion, and road accident information of trains and highway transportation), weather changes (such as climate data that affects shipping and road transportation, such as sea wind, sea conditions, temperature, precipitation, etc.).

[0032] Preprocess the multimodal data, such as data cleaning (if there are missing or incorrect data in some sensor data, interpolation or other methods are needed to fill the missing data, or eliminate the invalid data), data formatting (unify all data into a format acceptable to the system. For example, convert real-time traffic flow data into a time series format, and merge weather data and vehicle status data into a unified timestamp index), standardization and normalization (since the dimensions of multimodal data are different. For example, the unit of weather data may be degrees Celsius, traffic flow is the number of vehicles per hour, and the vehicle status may be binary values of "0" and "1". It is necessary to standardize or normalize all data to ensure that they are on the same scale for subsequent processing), preliminary feature extraction (use existing feature engineering techniques, such as machine learning methods to extract preliminary features. For example, extract the estimated arrival time from the shipping schedule information and extract the key factors that can affect shipping from the wind speed in the weather data), etc., to obtain the preprocessed multimodal data. The purpose of data formatting in the above preprocessing process is to convert multimodal data from different sources and types into a unified format so that it can be effectively processed by subsequent models and algorithms. The specific formatting process includes unifying timestamps, formatting different types of data, merging different data sources, handling inconsistent units, filling missing data, and outlier processing, etc.; specifically, convert all timestamps to UNIX timestamps (i.e., the number of seconds since January 1, 1970), or convert them to the standard ISO 8601 format (such as "YYYY-MM-DD HH:MM:SS"); use interpolation methods or interpolate data points according to the nearest timestamp to achieve time alignment; the standardization and normalization in the above preprocessing process convert the data into the same scale through methods such as Z-Score standardization, min-max normalization, unit vector normalization, and logarithmic normalization, avoiding the unbalanced impact on subsequent processing caused by the too large or too small dimensions of some features; in practical applications, it is used alone or in combination according to requirements and data characteristics, and determined according to specific scenarios; specifically, the unit of weather data is degrees Celsius, traffic flow is the number of vehicles per hour, and the vehicle status is binary (0 or 1). After Z-score standardization, the mean of all features is 0 and the standard deviation is 1. In this way, even if their original units and scales are different, they will have an equal impact on the model; all the above preprocessing processes adopt technical means well-known to those skilled in the art and will not be elaborated here.

[0033] Obtain historical multi-modal data from existing databases, perform the above-mentioned preprocessing process on the historical multi-modal data to obtain preprocessed historical multi-modal data, and process the preprocessed historical multi-modal data using existing AI-based large language models (such as GPT, BERT, etc.) to obtain historical comprehensive data; the AI-based large language model is used to capture the temporal dependencies and complex patterns in the preprocessed historical multi-modal data. Especially when dealing with the intertwined influence of multiple factors in port transportation, it can effectively extract the interrelationships between each factor and further form highly relevant and representative historical comprehensive data. The specific implementation process of processing the preprocessed historical multi-modal data using existing AI-based large language models (such as GPT, BERT, etc.) is as follows: First, send the preprocessed historical multi-modal data into an AI-based large language model (such as BERT, GPT). For GPT, the input preprocessed historical multi-modal data is processed through an autoregressive generation method. At each step of generation, the content generated in the previous step and the current input context information are considered, and then a comprehensive representation containing all historical information is generated. This enables GPT to effectively summarize the complex spatio-temporal dependencies in the preprocessed historical multi-modal data and further form a high-dimensional "historical comprehensive data"; for BERT or similar models, first process these input preprocessed historical multi-modal data through multiple layers of Transformer networks; the Transformer architecture is good at capturing long-range dependencies in data and is especially suitable for the processing of temporal data and multi-modal data; in this process, the BERT model will perform context modeling on each data unit of the input to identify potential relationships and patterns. For example, in multi-modal data, the relationship between weather conditions and traffic flow, the correlation between port equipment status and cargo handling efficiency, etc. can all be effectively captured through the context modeling ability. For example, the correlation between traffic conditions and transportation efficiency, the impact of weather on transportation route selection, etc. After the above process, historical comprehensive data is finally generated.

[0034] Further, based on the historical comprehensive data, use existing neural network models (such as LSTM, Transformer) to construct a path planning model. Specifically, the historical comprehensive data is fed into the input layer, and in the input layer, the historical comprehensive data is transformed into a vector representation through the embedding layer and used as the output of the input layer; the historical comprehensive data in vector representation is processed by an encoder to learn the dependencies and patterns in the historical comprehensive data, and the output of the encoder is obtained. The encoder consists of multiple attention layers (Self-Attention) and a feed-forward neural network (Feed Forward Network); further, the decoder processes the output of the encoder to obtain the path planning result; the decoder also contains multiple self-attention layers and a feed-forward neural network; the path planning result is sent to the output layer, and the output layer generates suggestions for path planning based on the path planning result and uses it as the output of the output layer. The suggestions for path planning can be expressed by a multi-dimensional vector, including, for example, path selection (the selected shortest path, optimal transportation path), transportation time (estimated transportation time), cost estimation (estimated transportation cost), transportation mode selection (suggestions for transportation modes such as sea, rail, road, etc.).

[0035] S2. Based on the preprocessed multi-modal data, obtain real-time comprehensive data, and process the real-time comprehensive data through the path planning model to obtain a preliminary path planning; introduce multi-dimensional influencing factors, and use a multi-objective optimization algorithm to optimize the preliminary path planning to obtain an optimized path planning.

[0036] Process the preprocessed multi-modal data through existing AI-based large language models (such as GPT, BERT, etc.) to obtain real-time comprehensive data. To avoid inaccurate data processing caused by data loss when using an AI-based large language model to process the preprocessed multi-modal data, introduce a missing data filling algorithm based on a compensation mechanism. The missing data filling algorithm based on the compensation mechanism compensates and fills the missing data and makes inferences based on the temporal characteristics of the preprocessed historical multi-modal data and related modal data to minimize the impact caused by missing data; the specific formula is as follows:

[0037]

[0038] where represents the preprocessed multi-modal data at the th after compensation filling; is at the th The predicted value of the missing preprocessed multimodal data is obtained through a time series prediction model (such as LSTM, Transformer, etc.) based on the preprocessed historical multimodal data; is at time the th actual preprocessed multimodal data; is a regulation coefficient, used to control the adjustment strength of the compensation quantization coefficient, determines the weight of the compensation quantization coefficient, and is determined according to the expert experience method. The reference value range is , and the specific value depends on the application field and data characteristics. For data with large dynamic changes (such as traffic flow), a smaller value will be selected, such as 0.1; while for a stable environment (such as weather data), a larger value will be selected, such as 0.7; is a similarity adjustment factor, used to adjust the influence of the similarity measurement, affects the final compensation effect by changing the contribution of the similarity measurement to the compensation quantization coefficient, and is obtained through the existing regression method. The reference value range is , and a larger value will amplify the change difference between similar moments, thus having a greater impact on the compensation filling. For example, traffic flow has large fluctuations and a strong impact, and the value is taken as 0.8; is a similarity weighting factor, used to control the weighting degree of the similarity between different preprocessed multimodal data when accumulating the similarity, affects the final compensation effect, and is obtained through the existing regression method. The reference value range is , for example, when the correlation is moderate, the value is taken as 0.5; is the total number of preprocessed multimodal data; is at time and the preprocessed multimodal data and the similarity between them, such as calculated using the cosine similarity formula; represents the exponentially weighted term, making the influence of the similarity more prominent; is a normalization term introduced to ensure that the influence of all data is balanced; is the normalized similarity influence; is used to adjust the difference between the predicted value of the missing preprocessed multimodal data and the actual preprocessed historical multimodal data, reflects the similarity and deviation between the predicted data and the actual data, that is, adjusts the influence of the normalized similarity to obtain the compensation quantization coefficient; is used to represent the prediction deviation; represents the corrected deviation; It means that the predicted value of the missing pre - processed multi - modal data is added to the corrected deviation to obtain the compensated and filled pre - processed multi - modal data.

[0039] Furthermore, the path planning model constructed in step S1 is used to process the real - time comprehensive data to obtain a preliminary path plan, including: path name, such as the path from port A to destination B; path selection, the specific path route, including the transportation modes passed through, such as sea transportation, railway, highway, etc.; estimated time, the total estimated transportation time; cost estimation, the costs of each transportation stage and the total transportation cost; transportation mode selection, based on transportation conditions, the optimal transportation mode is selected.

[0040] For example: path name, the path from port A to destination B.

[0041] Path selection: starting point, port A; first stage, sea transportation, from port A → transit port C; second stage, railway transportation, from transit port C → transfer station D; third stage, highway transportation, from transfer station D → destination B;

[0042] Estimated time: sea transportation time, 24 hours; railway transportation time, 12 hours; highway transportation time, 6 hours; total transportation time, 42 hours;

[0043] Cost estimation: sea transportation cost, $1000; railway transportation cost, $500; highway transportation cost, $200; total transportation cost, $1700;

[0044] Transportation mode selection: Since the distance is long and in order to save transportation costs, the relatively cheap sea transportation is selected; in order to achieve the timeliness of transportation, the railway transportation with better timeliness is selected; since the last section is short - distance transportation, the fastest highway transportation is selected.

[0045] Furthermore, multi-dimensional influencing factors are introduced, including: traffic flow, such as obtaining real-time traffic flow, congestion conditions, etc. of highways and railways through a traffic management system; weather changes, such as obtaining meteorological factors affecting the performance and transportation safety of transportation tools, such as wind speed, precipitation, sea conditions, etc. through a meteorological API; port handling capacity, such as obtaining factors affecting the cargo handling efficiency, such as the workload of the port, equipment failures, etc. through a port management system; availability of transportation tools, such as obtaining the number of transportation tools, status (such as whether it is under maintenance) affecting the transportation capacity, etc. through a transportation scheduling system; Based on the multi-dimensional influencing factors, existing multi-objective optimization algorithms are introduced, and optimization objectives are set according to specific application scenarios and requirements. The optimization objectives are such as minimizing transportation costs, minimizing transportation time, maximizing transportation safety, etc. Further, an objective function is defined according to the optimization objective, and at the same time, constraint conditions are set according to the specific scenario. Then, existing heuristic algorithms (such as A* algorithm, Dijkstra algorithm) are used to consider multi-dimensional influencing factors in path planning, dynamically evaluate and select the optimal path, and realize the optimization and adjustment of the preliminary path planning to obtain an optimized path planning. For example, in the sea transportation stage, if the real-time weather forecast shows that the wind speed is too high, the choice of the transportation path will be adjusted in real time, and the sea transportation will be changed to railway transportation; when congestion occurs in the railway stage, the feasibility of other transportation modes (such as highway or sea transportation) will be automatically evaluated and new path suggestions will be given.

[0046] For example: Path name: Optimized path planning from Port A to Destination B.

[0047] Path selection: Starting point: Port A; First stage: Railway transportation, from Port A → Transfer Port C - 18 hours (after path optimization, due to weather reasons, sea transportation is replaced by railway transportation); Second stage: Highway transportation, from Transfer Port C → Transfer Station D - 8 hours (congestion in the railway section, alternative route is selected); Third stage: Railway transportation, from Transfer Station D → Destination B - 10 hours (replacing highway transportation with railway to avoid highway traffic congestion);

[0048] Estimated time: 36 hours (the optimized path selection reduces the transportation time by 6 hours compared to the preliminary path planning);

[0049] Cost estimate: Railway transportation cost: $600 (high weather safety of railway transportation); Highway transportation cost: $300 (the proportion of highway transportation increases after adjustment); Railway transportation cost: $400 (replacing highway transportation due to traffic congestion, increasing the use of railway transportation); Total transportation cost: $1300 (the cost is reduced due to the adjustment of the transportation mode after optimization);

[0050] Transport mode selection: Due to weather conditions, sea transportation is replaced by rail transportation, the proportion of road transportation increases, and the safety of overall transportation is improved. Although the choice of rail transportation is slightly more expensive, it can ensure a smoother and more timely route.

[0051] In summary, a method for port multimodal transport route planning using an AI large language model is completed.

[0052] The sequence of the invention embodiments is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or consecutive order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0053] Each embodiment in this specification is described in a progressive manner. For the same or similar parts between the embodiments, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

[0054] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention and should all be included in the protection scope of the present invention.

Claims

1. A method for port multimodal transport route planning using an AI large language model, characterized in that, It includes the following steps: S1. Obtain and preprocess multi-modal data and historical multi-modal data to obtain preprocessed multi-modal data and preprocessed historical multi-modal data; Based on the preprocessed historical multi-modal data, obtain historical comprehensive data, and construct a path planning model to obtain a path planning result; S2. Process the preprocessed multi-modal data through a large language model based on AI to obtain real-time comprehensive data. At the same time, introduce a missing data filling algorithm based on a compensation mechanism to fill the data missing when using the large language model based on AI to process the preprocessed multi-modal data; The missing data filling algorithm based on the compensation mechanism compensates and fills the missing data, and makes inferences based on the temporal characteristics of the preprocessed historical multi-modal data to minimize the impact caused by the missing data; Process the real-time comprehensive data through the path planning model to obtain a preliminary path planning; introduce multi-dimensional influencing factors, and use a multi-objective optimization algorithm to optimize the preliminary path planning to obtain an optimized path planning.

2. The method for port multimodal transport route planning using an AI large language model according to claim 1, characterized in that, The S1 specifically includes: Process the preprocessed historical multi-modal data using a large language model based on AI to obtain historical comprehensive data; based on the historical comprehensive data, use a neural network model to construct a path planning model.

3. A method for port multimodal transport route planning using an AI large language model according to claim 2, characterized in that, The S1 specifically includes: During the process of model construction, input the historical comprehensive data into the input layer. In the input layer, convert the historical comprehensive data into a vector representation through the embedding layer and use it as the output of the input layer; process the vector-represented historical comprehensive data using an encoder to obtain the output of the encoder; the decoder processes according to the output of the encoder to obtain a path planning result, send the path planning result to the output layer, and the output layer generates a path planning suggestion based on the path planning result and uses it as the output of the output layer.

4. A method for port multimodal transport route planning using an AI large language model according to claim 1, characterized in that, The S2 specifically includes: During the implementation process of the missing data filling algorithm based on the compensation mechanism, subtract the predicted value of the missing preprocessed multi-modal data from the actual preprocessed multi-modal data at the previous moment to obtain a prediction deviation, multiply the prediction deviation by the compensation quantization coefficient to obtain a corrected deviation, and add the predicted value of the missing preprocessed multi-modal data to the corrected deviation to obtain the preprocessed multi-modal data after compensation filling.

5. A method for port multimodal transport route planning using an AI large language model according to claim 1, characterized in that The S2 specifically includes: Use the path planning model constructed in step S1 to process the real-time comprehensive data to obtain a preliminary path planning, including: path name, path selection, estimated time, cost estimation, and transportation mode selection.

6. A method for port multimodal transport route planning using an AI large language model, characterized in that, The S2 specifically includes: Introduce multi-dimensional influencing factors, and based on the multi-dimensional influencing factors, introduce a multi-objective optimization algorithm, set optimization objectives, define an objective function according to the optimization objectives, and at the same time set constraint conditions. Then use a heuristic algorithm to combine the multi-dimensional influencing factors in path planning, dynamically evaluate and select the optimal path, and optimize and adjust the preliminary path planning to obtain an optimized path planning.

Images