Truck matching method and system based on cargo variety association mining
By building a structured feature definition library and deep learning models to identify similar cargo types and generate the optimal transport capacity recommendation sequence, the problem of insufficient resource utilization in traditional transport capacity matching methods is solved, and the intelligence and transportation efficiency of the logistics system are improved.
Patent Information
- Application Number
- CN202510610144.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-13
- Publication Date
- 2025-09-09
AI Technical Summary
Traditional capacity matching methods fail to fully utilize the capacity resources of similar cargo types, resulting in some freight demands being difficult to match with suitable transport vehicles, affecting logistics efficiency. This is especially difficult when the coverage of historical capacity data is limited or the demand for a certain type of cargo increases dramatically, making it difficult to effectively expand the capacity candidate set.
By building a structured feature definition library, extracting cargo keywords and generating attribute vectors, combining deep learning models to process historical shipping records, identifying similar cargo types, and generating the optimal capacity recommendation sequence through feature fusion and score sorting.
It has achieved effective expansion of the candidate capacity set in complex transportation environments, improved the utilization rate of capacity resources and the intelligence level of the logistics system, and improved the matching efficiency of transportation tasks.
Smart Images
Figure CN120611897A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of transportation technology, and in particular to a truck matching method and system based on cargo variety association mining. Background Art
[0002] As logistics market demands diversify, transportation requirements for different cargo types vary significantly. Traditional capacity matching methods often fail to fully utilize the capacity resources for similar cargo types, making it difficult to match some freight demands with appropriate transport vehicles, impacting overall logistics efficiency. In practice, some cargoes can share some capacity resources due to similar packaging methods, storage and transportation requirements, or loading conditions. However, current capacity matching methods fail to fully account for this characteristic, resulting in limitations in capacity scheduling.
[0003] In recent years, my country has introduced a series of policies for logistics and transportation optimization to enhance the intelligence of its logistics systems. These policies provide support for capacity expansion based on cargo type similarity. However, existing capacity matching technologies primarily rely on fixed-rule screening or direct matching based on historical data, lacking flexibility and failing to meet the demands of complex transportation environments. When historical capacity data coverage is limited or demand for a particular cargo type surges, traditional methods struggle to effectively expand the candidate capacity set, resulting in delays for some transport tasks or idle capacity resources. Summary of the Invention
[0004] In view of this, the object of the present invention is to provide a truck matching method and system based on cargo type association mining to at least solve the above problems.
[0005] The technical solution adopted in the present invention is as follows:
[0006] A first aspect of the present invention provides a truck matching method based on cargo type association mining, the method comprising the following steps:
[0007] Step 1: Obtain the original text from the current cargo basic information and transportation order record data source and pre-process the original text;
[0008] Step 2: Build a structured feature definition library and use it to match the original text to extract the historical transport vehicle structure, historical cargo carrier records, and cargo keywords for the current cargo.
[0009] Step 3: Segment and encode the extracted cargo keywords to generate cargo attribute vectors;
[0010] Step 4: Convert the vehicle structure, cargo attributes, and historical cargo transportation records of the current cargo into a unified vector representation to generate the current cargo feature vector;
[0011] Step 5: Use the current cargo feature vector as a query, retrieve a first set of similar cargo vectors from the historical sample library, and enhance the cargo feature vector by using the first set of similar cargo vectors;
[0012] Step 6: Retrieve a second set of similar cargo vectors by enhancing the cargo feature vectors, retrieve a candidate transport capacity set by using the second set of similar cargo vectors, and score and rank the candidate transport capacity set using a scoring and ranking function.
[0013] Furthermore, the preprocessing of the original text in step 1 is specifically: removing noise words, stop words and repeated words in the original text through text cleaning.
[0014] Furthermore, step 3 is specifically as follows: segment and encode the pre-processed cargo keywords, identify key attribute words through the contextual semantic understanding model and sequence labeling mechanism, integrate the key attribute words, and generate a standardized vector representation of the current cargo attributes:
[0015]
[0016] in, represents the normalized score of the dth key attribute word; R d represents a d-dimensional real vector space; a i Represents the current cargo attribute vector.
[0017] Furthermore, step 4 is specifically as follows:
[0018] The vehicle structure is converted into a low-dimensional vector representation through embedding mapping:
[0019] x i =Embedding(X i )
[0020] l i =Embedding(L i )
[0021] Among them, X i Indicates the vehicle model data of the current historical cargo transportation vehicle, L i Represents the vehicle length data of the current historical cargo transport vehicle; Embedding() represents mapping the input data into a low-dimensional continuous vector space; x i Represents X i The embedding vector representation of is a low-dimensional continuous vector; l i Indicates L i The embedding vector representation of is a low-dimensional continuous vector;
[0022] Construct a time series feature set based on the historical shipping records of the current cargo:
[0023]
[0024] in, represents the structured feature vector proposed in the T-th historical transportation record of the current cargo i; H i Represents a time series feature set;
[0025] The time series feature set of the historical shipping records of the current cargo is processed through the Transformer model to generate a historical semantic feature vector:
[0026] z i =Transformer(H i )
[0027] Among them, Transformer() is a deep learning model; z i Represents historical semantic feature vector
[0028] The low-dimensional vector of the historical transport vehicle structure, the cargo attribute vector, and the historical semantic feature vector are fused to form the current cargo feature vector:
[0029]
[0030] Among them, f i Represents the current cargo feature vector; Indicates d f dimensional real vector space.
[0031] Furthermore, step 5 is specifically as follows:
[0032] Taking the current cargo representation vector as the query, the search enhancement generation mechanism is used to retrieve the semantically closest k cargo types in the historical sample library to form the first similar cargo vector set:
[0033]
[0034] Among them, r i (k) represents the feature vector of the kth historical record sample that is most similar to the current product i. represents the first similar goods vector set;
[0035] The current product feature vector and the first similar product vector set are fused through the fusion function to form the enhanced feature vector of the current product:
[0036]
[0037] in, Represents the current cargo enhancement feature vector, Fusion() represents the feature fusion function, which means f i and Perform feature fusion.
[0038] Furthermore, step 6 is specifically as follows:
[0039] Step S1: Set a similarity threshold, and retrieve a second similar product vector using the similarity threshold and the current product enhanced feature vector:
[0040] Step S2: extracting the historical capacity configuration of each similar cargo type in the second similar cargo vector as a candidate capacity configuration, and matching the candidate capacity configuration with the existing capacity configuration;
[0041] Step S3: If the match fails, the similarity threshold is relaxed and steps 1 and 2 are repeated; if the match succeeds, the extracted candidate capacity configurations are scored and ranked using the scoring ranking function to generate the optimal scheduling recommendation sequence.
[0042] Furthermore, in step 1, a similarity threshold is set, and the formula for retrieving the second similar product vector using the similarity threshold and the current product enhanced feature vector is:
[0043]
[0044] Among them, τ r is the similarity threshold, which is gradually relaxed with the iteration of the rth round; In the semantic vector space, the feature vector is enhanced with the current goods As a benchmark, calculate the cosine similarity between the feature vectors of similar goods types, and select the similarity greater than the threshold τ r The first k types of goods are selected as candidate sets
[0045] In step 2, the historical capacity configuration of each similar cargo type in the second similar cargo vector is extracted as the candidate capacity configuration formula:
[0046]
[0047] in, represents the candidate transport capacity configuration set that the current cargo i matches in round r, v n Represents a specific candidate capacity configuration vector unit, and MatchRule(j) represents the candidate capacity configuration set corresponding to similar cargo type j.
[0048] Furthermore, in step S3, the extracted candidate capacity configurations are scored and ranked using a ranking scoring function to generate the optimal scheduling recommendation sequence:
[0049] Construct the order preference score of candidate capacity configuration v for the current cargo i:
[0050]
[0051] Where pref(v,i) represents the historical order acceptance preference score of the candidate capacity configuration v for the current cargo i;
[0052] The historical order preference scores are combined with the similarity between the candidate capacity configuration and the current cargo feature vector to construct a comprehensive scoring function:
[0053]
[0054] Where score(v,i) represents the score of the candidate transport configuration v for the current cargo i; λ represents the weight ratio, and λ∈[0,1]; Represents the candidate capacity configuration v and the current cargo enhancement feature vector The similarity of f v is the vector representation of the candidate capacity configuration;
[0055] Finally, the recommended sorting sequence is generated:
[0056]
[0057] in, Indicates that the comprehensive score of each candidate capacity configuration is sorted from high to low, Rank(C i ) represents the candidate capacity set C i Recommended scheduling sequence.
[0058] The second aspect of the present invention provides a truck matching system based on cargo variety association mining, the system is used to execute the method described in claims 1-8, the system includes a feature extraction module, a feature representation module, a similarity calculation module, a feature fusion module, a candidate capacity generation module and an optimal capacity generation module, the feature extraction module is used to extract features of cargo data, the feature representation module is used to construct a high-dimensional feature representation of cargo, the similarity calculation module is used to construct a similarity measure between cargo types, the candidate capacity generation module is used to generate a candidate capacity set, and the optimal capacity generation module is used to generate the optimal capacity based on the candidate capacity set.
[0059] Compared with the prior art, the present invention has the following beneficial effects:
[0060] The present invention provides a truck matching method and system based on cargo type association mining. By analyzing the historical correlation between carrier vehicle models of different cargo types and combining the physical and chemical characteristics of the cargo itself, the cargo categories with high similarity to the current cargo to be transported are identified. On this basis, the transportation rule constraints of the two types of cargo are integrated into similar items to achieve effective expansion of the candidate transport capacity candidate set. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0062] Figure 1 This is a flow chart of a truck matching method based on cargo type association mining provided by the present invention. DETAILED DESCRIPTION
[0063] The principles and features of the present invention are described below with reference to the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.
[0064] Reference Figure 1 The embodiment of the present invention provides a truck matching method based on cargo type association mining, the method comprising the following steps:
[0065] Step 1: Obtain the original text from the current cargo basic information and transportation order record data source and preprocess the original text.
[0066] The specific preprocessing of the original text is to remove noise words, stop words and repeated words in the original text through text cleaning.
[0067] Step 2: Build a structured feature definition library and use it to match the original text to extract the historical transport vehicle structure, historical cargo carrier records, and cargo keywords for the current cargo.
[0068] For example, cargo keywords include physical and chemical property keywords such as fragility, mode of transportation, corrosiveness, and explosiveness.
[0069] Step 3: Segment and encode the extracted goods keywords to generate goods attribute vectors; specifically:
[0070] The pre-processed cargo keywords are segmented and encoded, and key attribute words are identified through the contextual semantic understanding model and sequence labeling mechanism. The key attribute words are integrated to generate a standardized vector representation of the current cargo attributes:
[0071]
[0072] in, It represents the normalized score of the d-th key attribute word, which is used to measure the importance of this attribute in the current goods. For example, when the d-th attribute represents "fragility", The higher the value, the more attention should be paid to preventing damage and pressure during transportation of the goods; d represents a d-dimensional real vector space; a i Represents the current cargo attribute vector
[0073] Step 4: Convert the vehicle structure, cargo attributes, and historical cargo transportation records of the current cargo into a unified vector representation to generate the current cargo feature vector, specifically:
[0074] The vehicle structure is converted into a low-dimensional vector representation through embedding mapping:
[0075] x i =Embedding(X i )
[0076] l i =Embedding(L i )
[0077] Among them, X i Indicates the vehicle model data of the current historical cargo transportation vehicle, L i Represents the vehicle length data of the current historical cargo transport vehicle; Embedding() represents mapping the input data into a low-dimensional continuous vector space; x i Represents X i The embedding vector representation of is a low-dimensional continuous vector; l i Indicates L i The embedding vector representation of is a low-dimensional continuous vector;
[0078] Construct a time series feature set based on the historical shipping records of the current cargo:
[0079]
[0080] in, It represents the structured feature vector proposed in the T-th historical transportation record of the current cargo i. This vector includes attribute information such as transportation mode, load, timestamp, etc., and is an encoded representation of a single transportation behavior.
[0081] The time series feature set of the historical shipping records of the current cargo is processed through the Transformer model to generate a historical semantic feature vector:
[0082] z i=Transformer(H i )
[0083] Among them, Transformer() is a deep learning model; z i Represents the historical semantic feature vector.
[0084] The low-dimensional vector of the historical transport vehicle structure, the cargo attribute vector, and the historical semantic feature vector are fused to form the current cargo feature vector:
[0085]
[0086] Among them, f i Represents the current cargo feature vector; Indicates d f dimensional real vector space.
[0087] Step 5: Use the current cargo feature vector as a query to retrieve a first set of similar cargo vectors from the historical sample library. Enhance the cargo feature vector using the first set of similar cargo vectors. Specifically:
[0088] Taking the current cargo representation vector as the query, the search enhancement generation mechanism is used to retrieve the semantically closest k cargo types in the historical sample library to form the first similar cargo vector set:
[0089]
[0090] in, represents the feature vector of the kth historical record sample that is most similar to the current product i. Vector structure and current cargo feature vector f i Similarly, the set is obtained by selecting the top-k samples after sorting by cosine similarity; Represents the first similar goods vector set.
[0091] The current product feature vector and the first similar product vector set are fused through the fusion function to form the semantically enhanced feature vector of the current product:
[0092]
[0093] in, Represents the current cargo enhancement feature vector, Fusion() represents the feature fusion function, which means f i and Perform feature fusion.
[0094] For example, the similarity between any two items is calculated using cosine similarity:
[0095]
[0096] Step 6: Retrieve a second set of similar cargo vectors by enhancing the cargo feature vectors, retrieve a candidate capacity set by using the second set of similar cargo vectors, and rank the candidate capacity set by using the scoring ranking function, specifically:
[0097] Step S1: Set a similarity threshold, and retrieve a second similar product vector using the similarity threshold and the current product enhanced feature vector:
[0098]
[0099] Among them, τ r is the similarity threshold, which is gradually relaxed with the iteration of the rth round; In the semantic vector space, the feature vector is enhanced with the current goods As a benchmark, calculate the cosine similarity between the feature vectors of similar goods types, and select the similarity greater than the threshold τ r The first k types of goods are selected as candidate sets
[0100] Step S2: extracting the historical capacity configuration of each similar cargo type in the second similar cargo vector as a candidate capacity configuration, and matching the candidate capacity configuration with the existing capacity configuration;
[0101]
[0102] in, represents the candidate transport capacity configuration set that the current cargo i matches in round r, v n Represents a specific candidate capacity configuration vector unit, representing a transport vehicle, whose attributes include vehicle type, vehicle length, load capacity and other information; MatchRule(j) represents the candidate capacity configuration set corresponding to similar cargo type j, specifically including matching vehicle type, vehicle length range, load capacity and so on.
[0103] Step S3: If the match fails, the similarity threshold is relaxed and steps 1 and 2 are repeated. If the match succeeds, the extracted candidate capacity configurations are scored and ranked using the scoring ranking function to generate the optimal scheduling recommendation sequence, specifically:
[0104] Construct the order preference score of candidate capacity configuration v for the current cargo i:
[0105]
[0106] Where pref(v,i) represents the historical order acceptance preference score of the candidate transport configuration v for the current cargo i, which can be obtained through historical waybill data statistics. If the candidate transport configuration has never accepted cargo of the current cargo type, this value is the default preference value.
[0107] The historical order preference scores are combined with the similarity between the candidate capacity configuration and the current cargo feature vector to construct a comprehensive scoring function:
[0108]
[0109] Where score(v,i) represents the score of the candidate transport configuration v for the current cargo i; λ represents the weight ratio, and λ∈[0,1]; represents the similarity between the candidate capacity configuration and the current cargo enhancement feature vector; f v It is a vector representation of the candidate capacity configuration, which represents the static attributes of the capacity configuration, such as vehicle type, vehicle length, and vehicle height.
[0110] Finally, the recommended sorting sequence is generated:
[0111]
[0112] in, Indicates that the comprehensive score of each candidate capacity configuration is sorted from high to low. The higher the ranking of the candidate capacity configuration, the more likely it is to be selected by the system to carry the current cargo; Rank (C i ) represents the candidate capacity set C i Recommended scheduling.
[0113] Another embodiment of the present invention provides a truck matching system based on cargo variety association mining, the system is used to execute the method described in claims 1-8, and the system includes a feature extraction module, a feature representation module, a similarity calculation module, a feature fusion module, a candidate capacity generation module and an optimal capacity generation module.
[0114] For example, the feature extraction module is used to extract the physical and chemical properties of goods, including cargo type, transportation method, loading requirements, fragility, corrosivity, and explosiveness. This module uses a structured feature definition library for standardized descriptions and combines it with the text understanding capabilities of a language model pre-trained on large-scale corpus data to extract key attribute words from cargo descriptions and transportation records to form a standardized physical and chemical property vector representation, providing a basis for subsequent similarity calculations.
[0115] The feature representation module is used to construct a high-dimensional feature representation of cargo types. It specifically consists of two parts: embedding and encoding the original characteristics of the cargo (such as vehicle type and length) to eliminate the discreteness of categorical variables; and using a self-attention mechanism to learn the high-order dependencies between features in historical shipping records to form a more representative comprehensive feature representation. The Transformer encoder is introduced to capture long-range dependency features in historical shipping sequences, further enhancing semantic expression capabilities.
[0116] The similarity calculation module is used to construct a similarity metric between cargo types. The core idea is to integrate historical shipping records and cargo physical characteristics to generate a unified feature representation.
[0117] The feature fusion module is used to fuse the features of similar cargo types with the features of the current cargo to achieve a more comprehensive and accurate similarity expression;
[0118] The candidate transport capacity generation module is used to expand the set of candidate transport capacities available for cargo under the constraints of current matching rules (such as vehicle type, vehicle length, etc.);
[0119] After obtaining a set of candidate capacity, the Optimal Capacity Generation Module generates a recommended capacity ranking sequence for the current shipment based on carriers' historical order preferences for different cargo types. If no available transportable resources are found within the current ranking, the system will further expand the candidate cargo type set and regenerate the recommended sequence until a match is found.
[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A truck matching method based on cargo type association mining, characterized in that: The method comprises the following steps: Step 1: Obtain the original text from the current cargo basic information and transportation order record data source and pre-process the original text; Step 2: Build a structured feature definition library and use it to match the original text to extract the historical transport vehicle structure, historical cargo carrier records, and cargo keywords for the current cargo. Step 3: Segment and encode the extracted cargo keywords to generate cargo attribute vectors; Step 4: Convert the vehicle structure, cargo attributes, and historical cargo transportation records of the current cargo into a unified vector representation to generate the current cargo feature vector; Step 5: Use the current cargo feature vector as a query, retrieve a first set of similar cargo vectors from the historical sample library, and enhance the cargo feature vector by using the first set of similar cargo vectors; Step 6: Retrieve a second set of similar cargo vectors by enhancing the cargo feature vectors, retrieve a candidate transport capacity set by using the second set of similar cargo vectors, and score and rank the candidate transport capacity set using a scoring and ranking function.
2. The truck matching method based on cargo type association mining according to claim 1 is characterized in that: The preprocessing of the original text in step 1 is specifically to remove noise words, stop words and repeated words in the original text through text cleaning.
3. The truck matching method based on cargo type association mining according to claim 2 is characterized in that: Step 3 is as follows: segment and encode the pre-processed cargo keywords, identify key attribute words through the contextual semantic understanding model and sequence labeling mechanism, integrate the key attribute words, and generate a standardized vector representation of the current cargo attributes: in, represents the normalized score of the dth key attribute word; R d represents a d-dimensional real vector space; a i Represents the current cargo attribute vector.
4. The truck matching method based on cargo type association mining according to claim 3 is characterized in that: Step 4 is as follows: The vehicle structure is converted into a low-dimensional vector representation through embedding mapping: x i =Embedding(X i ) l i =Embedding(L i ) Among them, X i Indicates the vehicle model data of the current historical cargo transportation vehicle, L i Represents the vehicle length data of the current historical cargo transport vehicle; Embedding() represents mapping the input data into a low-dimensional continuous vector space; x i Represents X i The embedding vector representation of is a low-dimensional continuous vector; l i Indicates L i The embedding vector representation of is a low-dimensional continuous vector; Construct a time series feature set based on the historical shipping records of the current cargo: in, represents the structured feature vector proposed in the T-th historical transportation record of the current cargo i; H i Represents a time series feature set; The time series feature set of the historical shipping records of the current cargo is processed through the Transformer model to generate a historical semantic feature vector: z i =Transformer(H i ) Among them, Transformer() is a deep learning model; z i Represents historical semantic feature vector; The low-dimensional vector of the historical transport vehicle structure, the cargo attribute vector, and the historical semantic feature vector are fused to form the current cargo feature vector: Among them, f i Represents the current cargo feature vector; Indicates d f dimensional real vector space.
5. The truck matching method based on cargo type association mining according to claim 4 is characterized in that: Step 5 is as follows: Taking the current cargo representation vector as the query, the search enhancement generation mechanism is used to retrieve the semantically closest k cargo types in the historical sample library to form the first similar cargo vector set: Among them, r i (k) represents the feature vector of the kth historical record sample that is most similar to the current product i. represents the first similar goods vector set; The current product feature vector and the first similar product vector set are fused through the fusion function to form the enhanced feature vector of the current product: in, Represents the current cargo enhancement feature vector, Fusion() represents the feature fusion function, which means f i and Perform feature fusion.
6. The truck matching method based on cargo type association mining according to claim 5 is characterized in that: Step 6 is as follows: Step S1: Set a similarity threshold, and retrieve a second similar product vector using the similarity threshold and the current product enhanced feature vector: Step S2: extracting the historical capacity configuration of each similar cargo type in the second similar cargo vector as a candidate capacity configuration, and matching the candidate capacity configuration with the existing capacity configuration; Step S3: If the match fails, the similarity threshold is relaxed and steps 1 and 2 are repeated; if the match succeeds, the extracted candidate capacity configurations are scored and ranked using the scoring ranking function to generate the optimal scheduling recommendation sequence.
7. The truck matching method based on cargo type association mining according to claim 6 is characterized in that: In step 1, the similarity threshold is set. The formula for retrieving the second similar product vector by using the similarity threshold and the current product enhanced feature vector is: Among them, τ r is the similarity threshold, which is gradually relaxed with the iteration of the rth round; In the semantic vector space, the feature vector is enhanced with the current goods As a benchmark, calculate the cosine similarity between the feature vectors of similar goods types, and select the similarity greater than the threshold τ r The first k types of goods are selected as candidate sets In step 2, the historical capacity configuration of each similar cargo type in the second similar cargo vector is extracted as the candidate capacity configuration formula: in, represents the candidate transport capacity configuration set that the current cargo i matches in round r, v n Represents a specific candidate capacity configuration vector unit, and MatchRule(j) represents the candidate capacity configuration set corresponding to similar cargo type j.
8. The truck matching method based on cargo type association mining according to claim 7 is characterized in that: In step S3, the extracted candidate capacity configurations are scored and ranked using a ranking scoring function to generate the optimal scheduling recommendation sequence: Construct the order preference score of candidate capacity configuration v for the current cargo i: Where pref(v,i) represents the historical order acceptance preference score of the candidate capacity configuration v for the current cargo i; The historical order preference scores are combined with the similarity between the candidate capacity configuration and the current cargo feature vector to construct a comprehensive scoring function: Where score(v,i) represents the score of the candidate transport configuration v for the current cargo i; λ represents the weight ratio, and λ∈[0,1]; Represents the candidate capacity configuration v and the current cargo enhancement feature vector Similarity; f v is the vector representation of the candidate capacity configuration; Finally, the recommended sorting sequence is generated: in, Indicates that the comprehensive score of each candidate capacity configuration is sorted from high to low, Rank(C i ) represents the candidate capacity set C i Recommended scheduling sequence.
9. A truck matching system based on cargo type association mining, characterized by: The system is used to execute the method described in claims 1-8, and the system includes a feature extraction module, a feature representation module, a similarity calculation module, a feature fusion module, a candidate capacity generation module and an optimal capacity generation module. The feature extraction module is used to extract features of cargo data, the feature representation module is used to construct a high-dimensional feature representation of cargo, the similarity calculation module is used to construct a similarity measure between cargo types, the candidate capacity generation module is used to generate a candidate capacity set, and the optimal capacity generation module is used to generate the optimal capacity based on the candidate capacity set.