Goods distribution method and device and computer program product

Abstract

The invention discloses a cargo distribution method and device and a computer program product. The method comprises the following steps: determining a target cargo hold of a passenger plane loaded with a to-be-distributed cargo set, and obtaining cargo characteristics of each cargo and cargo hold characteristics of the target cargo hold; obtaining a plurality of cargo loading constraint conditions, and constructing a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold features and the cargo features; generating cue words based on the cargo loading constraint conditions, the cargo hold features and the cargo features, and inputting the cue words and the search space into a target model to obtain a reduced search space; randomly extracting N initial cargo distribution schemes from the reduced search space, and screening out a target cargo distribution scheme from the N initial cargo distribution schemes based on a target evaluation function; and distributing the cargo to the target cargo hold based on the target cargo distribution scheme. According to the invention, the problem of low efficiency of distributing the goods to the passenger plane cargo hold in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and in particular, to a cargo distribution method and device and computer program product. BACKGROUND

[0002] The cargo balancing task of a passenger aircraft is a key link to ensure the balance of the aircraft and flight safety in air transportation. It requires accurate distribution of various cargos in the flight task, and needs to consider the distribution of passengers to minimize the center of gravity deviation in the flight process, thereby reducing fuel consumption and improving flight efficiency. The complexity of this task lies in the multi-dimensional constraint conditions it contains, such as weight limits, cargo type matching, dangerous goods isolation rules, cargo compartment structure characteristics and operation rules, and multiple optimization objectives, such as flight efficiency, cost-effectiveness and cargo safety.

[0003] In the related art, heuristic algorithms are used to simulate the optimization search strategy in nature (such as genetic algorithms, simulated annealing algorithms) to quickly find a satisfactory solution to the problem under incomplete information. However, heuristic algorithms also have limitations when dealing with the multi-constraint civil aviation passenger aircraft cargo balancing problem: on the one hand, the complexity and diversity of constraint conditions in real scenarios make scenario modeling extremely difficult, and heuristic algorithms in the related art are difficult to accurately capture and express all constraint conditions, resulting in a decline in the applicability and accuracy of the model; on the other hand, the search space under multiple constraint conditions is extremely large, and heuristic algorithms may consume impractical computing resources and time when searching for the optimal solution, especially in the context of a sharp increase in data volume, which makes this problem more prominent.

[0004] In view of the low efficiency of distributing cargos to the cargo compartment of a passenger aircraft in the related art, no effective solution has been proposed so far. SUMMARY

[0005] The main purpose of the present application is to provide a cargo distribution method, device and computer program product to solve the problem of low efficiency of distributing cargos to the cargo compartment of a passenger aircraft in the related art.

[0006] To achieve the above object, according to an aspect of the present application, a cargo distribution method is provided. The method comprises: determining a target cargo hold of a passenger aircraft for loading a set of cargos to be distributed, obtaining cargo features of each cargo in the set of cargos to be distributed, and obtaining cargo hold features of the target cargo hold; obtaining a plurality of preset cargo loading constraint conditions of the passenger aircraft, constructing a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold features and the cargo features, wherein the search space is used to represent a set of all cargo distribution schemes, each cargo distribution scheme comprising a position of each cargo in the target cargo hold; generating a prompt based on the plurality of cargo loading constraint conditions, the cargo hold features and the cargo features, inputting the prompt and the search space into a target model to obtain a reduced search space, wherein the reduced search space is the search space from which cargo distribution schemes that do not meet the cargo loading constraint conditions are removed; randomly extracting N initial cargo distribution schemes from the reduced search space, and screening a target cargo distribution scheme from the N initial cargo distribution schemes based on a target evaluation function, wherein N is a positive integer; and distributing the cargos in the set of cargos to be distributed to the target cargo hold based on the target cargo distribution scheme.

[0007] Optionally, constructing the search space of the heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold features and the cargo features comprises: extracting a plurality of cargo positions of the target cargo hold from the cargo hold features, determining cargo position features of each cargo position, wherein the cargo position features comprise at least one of the following: a number of the cargo position, position information of the cargo position in the target cargo hold, and a volume upper limit and a weight bearing upper limit of the cargo position; extracting attributes of each cargo from the cargo features, encoding the set of cargos to be distributed into a target tensor based on the attributes of each cargo, wherein the attributes comprise at least one of the following: a cargo number, a weight, a volume, and a cargo type, each row of the target tensor represents a cargo, and each column represents an attribute; constructing a compatibility matrix based on the target tensor and the cargo position features of each cargo position, wherein each row of the compatibility matrix represents a cargo, each column represents a cargo position in the target cargo hold, and a value in the compatibility matrix represents whether the cargo corresponding to the row in which the value is located is placed in the cargo position corresponding to the column in which the value is located in accordance with the cargo loading constraint condition; determining each type of compatibility matrix as a search solution, and constructing the search space from all search solutions, wherein each search solution represents a cargo distribution scheme.

[0008] Optionally, the target model is obtained by the following method: obtaining historical cargo distribution records of the passenger aircraft, extracting historical prompts, initial search spaces, and target search spaces after removing cargo distribution schemes that do not meet the cargo loading constraint conditions corresponding to each cargo distribution from the historical cargo distribution records; determining the historical prompts, the initial search spaces, and the target search spaces of each cargo distribution as a set of training samples, and obtaining a plurality of sets of training samples; training a large language model based on the plurality of sets of training samples to obtain the target model.

[0009] Optionally, the screening of the target goods allocation scheme from the N initial goods allocation schemes based on the target evaluation function comprises: screening M candidate goods allocation schemes from the N initial goods allocation schemes based on spatial continuity, where M is less than N and M is a positive integer; determining the M candidate goods allocation schemes as an initial population, and performing multiple iteration recombinations on the initial population to obtain a population after each iteration, where each population after iteration recombination comprises M individuals, and each individual represents a candidate goods allocation scheme; for each population after iteration, calculating an evaluation value of each individual in the population based on the target evaluation function to obtain M evaluation values, and determining a maximum evaluation value in the M evaluation values; calculating a difference between the maximum evaluation values of adjacent iteration populations, and stopping iteration in a case where the difference is less than a preset threshold, and determining a candidate goods allocation scheme corresponding to the maximum evaluation value in all populations after iteration as the target goods allocation scheme.

[0010] Optionally, the screening of the M candidate goods allocation schemes from the N initial goods allocation schemes based on spatial continuity comprises: for each initial goods allocation scheme, determining all the storage location numbers between the starting storage location and the ending storage location in the preset continuous region to obtain a target storage location number set; determining a target goods set that needs to be stored adjacently, determining the storage location number to which each goods in the target storage location set is allocated in the initial goods allocation scheme, and judging whether the storage location number belongs to the target storage location number set; in a case where there is at least one storage location number of goods that does not belong to the target storage location number set, eliminating the initial goods allocation scheme; in a case where the storage location numbers of all goods in the target goods set belong to the target storage location number set, determining the initial goods allocation scheme as a candidate goods allocation scheme to obtain the M candidate goods allocation schemes.

[0011] Optionally, the performing of multiple iteration recombinations on the initial population to obtain a population after each iteration comprises: for each iteration, sorting a population after last iteration in descending order of evaluation values to obtain a target sequence, and taking the first P individuals in the target sequence as parent individuals, where P is a positive integer and P is less than M; performing cross recombination on the P parent individuals to generate M population individuals in the current iteration round.

[0012] Optionally, the calculating of the evaluation value of each individual in the population based on the target evaluation function comprises: for each individual, calculating a target evaluation value between goods and allocated storage locations in the candidate goods allocation scheme corresponding to the individual, where the target evaluation value comprises at least one of the following: a feature matching evaluation value, a gravity balance evaluation value and a constraint satisfaction evaluation value; and performing weighted summation on the target evaluation value based on a preset weight to obtain the evaluation value of each individual.

[0013] Optionally, the obtaining the plurality of cargo loading constraints preset for the passenger aircraft comprises: determining a type of the passenger aircraft, and extracting the plurality of cargo loading constraints matching the type from a preset constraint database, wherein the cargo loading constraints comprise at least one of a physical constraint, an operation constraint, and a business constraint, the physical constraint is used to limit a load and a volume of each position of the target cargo hold, the operation constraint is used to limit a cargo loading type and a loading priority of each position of the target cargo hold, and the business constraint is used to limit a generation time of the target cargo allocation scheme.

[0014] To achieve the above object, according to another aspect of the present application, a cargo allocation apparatus is provided. The apparatus comprises: an obtaining unit configured to determine a target cargo hold of a passenger aircraft for loading a set of to-be-allocated cargos, obtain a cargo feature of each cargo in the set of to-be-allocated cargos, and obtain a cargo hold feature of the target cargo hold; a constructing unit configured to obtain a plurality of cargo loading constraints preset for the passenger aircraft, and construct a search space of a heuristic algorithm based on the plurality of cargo loading constraints, the cargo hold feature, and the cargo feature, wherein the search space is used to represent a set of all cargo allocation schemes, and each cargo allocation scheme comprises an allocation of each cargo to a position in the target cargo hold; an input unit configured to generate a prompt word based on the plurality of cargo loading constraints, the cargo hold feature, and the cargo feature, input the prompt word and the search space into a target model, and obtain a reduced search space, wherein the reduced search space is the search space from which a cargo allocation scheme that does not conform to the cargo loading constraints is removed; a screening unit configured to randomly extract N initial cargo allocation schemes from the reduced search space, and screen a target cargo allocation scheme from the N initial cargo allocation schemes based on a target evaluation function, wherein N is a positive integer; and an allocating unit configured to allocate the cargos in the set of to-be-allocated cargos to the target cargo hold based on the target cargo allocation scheme.

[0015] To achieve the above object, according to another aspect of the present application, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of the cargo allocation method described in various embodiments of the present application.

[0016] The present application adopts the following steps: determining a target cargo hold of a passenger aircraft for loading a set of cargo to be allocated, obtaining cargo characteristics of each cargo in the set of cargo to be allocated, and obtaining cargo hold characteristics of the target cargo hold; obtaining multiple cargo loading constraints preset for the passenger aircraft, and constructing a search space for a heuristic algorithm based on the multiple cargo loading constraints, cargo hold characteristics, and cargo characteristics, wherein the search space is used to represent a set of all cargo allocation plans, each cargo allocation plan including the position of each cargo allocated in the target cargo hold; generating prompt words based on the multiple cargo loading constraints, cargo hold characteristics, and cargo characteristics, and inputting the prompt words and the search space into a target model to obtain a simplified search space, wherein the simplified search space is a search space that eliminates cargo allocation plans that do not meet the cargo loading constraints; randomly extracting N initial cargo allocation plans from the simplified search space, and screening a target cargo allocation plan from the N initial cargo allocation plans based on a target evaluation function, wherein N is a positive integer; and allocating the cargo in the set of cargo to be allocated to the target cargo hold based on the target cargo allocation plan, thereby solving the problem of low efficiency in allocating cargo to the cargo hold of a passenger aircraft in the related art. By using prompt words to structure descriptions of complex, multi-constrained scenarios, the modeling process is simplified, improving the model's adaptability and the accuracy of the optimization target. Leveraging the target model's intelligent predictive capabilities, invalid cargo allocation solutions can be quickly filtered out, reducing the search space and improving search efficiency. This, in turn, improves the efficiency of cargo allocation to passenger aircraft cargo holds. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0018] Figure 1 is a flow chart of a cargo distribution method provided according to an embodiment of the present application;

[0019] Figure 2 is a schematic diagram of an optional cargo distribution method provided according to an embodiment of the present application;

[0020] Figure 3 is a schematic diagram of an initial reduction process provided according to an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of a large model-driven simplification process provided according to an embodiment of the present application;

[0022] Figure 5 is a schematic diagram of a cargo distribution device provided according to an embodiment of the present application;

[0023] Figure 6is a schematic view of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0025] In order for those skilled in the technical field to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0027] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present disclosure are all information and data authorized by the user or authorized by all parties.

[0028] It should be noted that the collected information is information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards in relevant regions, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for users to choose authorization or refusal.

[0029] The present application will be described below in combination with the preferred implementation steps, Figure 1 is a flowchart of a cargo distribution method according to an embodiment of the present application, as shown in Figure 1 The method comprises the following steps:

[0030] Step S101, determine the target cargo hold of the passenger plane for loading the set of goods to be allocated, obtain the characteristics of each cargo in the set of goods to be allocated, and obtain the characteristics of the target cargo hold.

[0031] In step S101, first, based on the flight task, the model information, and the cargo hold load and space restrictions, the target cargo hold of the passenger plane for loading the set of goods to be allocated is determined. For each cargo in the set of goods to be allocated, its cargo characteristics are extracted, including but not limited to the weight, volume, dangerous goods number, priority of the cargo, etc. The cargo hold characteristics of the target cargo hold can include the maximum carrying weight of the cargo hold, volume restrictions, cargo hold partition information, weight-bearing linkage rules, and restrictions on special cargo positions, etc.

[0032] Step S102, obtain a plurality of preset cargo loading constraint conditions of the passenger plane, and construct a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold characteristics and the cargo characteristics, wherein the search space is used to represent a set of all cargo allocation schemes, each cargo allocation scheme including the allocation of each cargo to a position in the target cargo hold.

[0033] In step S102, a plurality of cargo loading constraint conditions are extracted from the civil aviation passenger plane operation manual and business requirements, which include but are not limited to weight restrictions, center of gravity offset control, cargo type matching, container type compliance, dangerous goods spacing, etc. Based on the constraint conditions, the cargo hold characteristics and the cargo characteristics, the search space of the heuristic algorithm is constructed. The search space comprehensively covers all possible cargo allocation schemes, and each scheme specifically represents the allocation of each cargo to a position in the target cargo hold.

[0034] For example, in the multi-constraint civil aviation passenger plane cargo hold balancing scenario, the multiple constraint conditions and complex cargo data are constructed into standard large model prompts and search spaces of heuristic algorithms. The cargo attributes (weight, volume, dangerous goods number) are encoded into a multi-dimensional feature tensor, and a position-cargo compatibility matrix is generated in combination with the cargo hold information to construct a search space containing cargo hold positions, capacity restrictions and conflict rules, providing a quantifiable discretization basis for solution space exploration. The solution space is optimized by constraints, fast conflict detection is realized based on R*-Tree (a tree data structure for spatial data indexing) spatial indexing, invalid initial schemes are eliminated through dynamic pruning strategies (such as lower bound estimation of cost, early termination), and finally the feasible solutions are expressed as sparse binary vectors, balancing the size of the space and the quality of the solution.

[0035] Step S103, generate prompts based on the plurality of cargo loading constraint conditions, the cargo hold characteristics and the cargo characteristics, input the prompts and the search space into a target model, and obtain a reduced search space, wherein the reduced search space is a search space in which cargo allocation schemes that do not meet the cargo loading constraint conditions are eliminated.

[0036] In step S103, the cargo loading constraint conditions, cargo compartment features and cargo features are coded into a form understandable by an algorithm, and the prompt words can include problem definitions, constraint descriptions and input and output specifications, providing guidance for subsequent search and optimization. The constructed prompt words and search space are input into the target model (such as a large model), and the predictive ability of the model is used to dynamically adjust the search strategy, eliminate cargo allocation schemes that obviously cannot meet the constraint conditions, thereby reducing the search space, reducing the complexity and time cost of the search.

[0037] For example, by traversing all cargo loading constraint conditions, a multi-constraint problem description prompt word is constructed, and three types of constraints, i.e., physical load limit, operation rule and civil aviation business demand of the civil aviation passenger aircraft cargo compartment, are integrated. The cargo compartment partition parameters and the initial allocation scheme are defined by structured JSON (JavaScript Object Notation, a lightweight data exchange format), and a three-part prompt word template containing scene definition, constraint description and input and output specification is generated, providing clear problem boundaries and data interaction standards for subsequent optimization.

[0038] The structured prompt word is constructed by extracting physical, operational and business constraints from the civil aviation passenger aircraft cargo compartment operation manual. The physical constraints include the center of gravity allowable range (such as 15% to 25% MAC (Mean Aerodynamic Chord)), the maximum load of the cargo compartment (such as 45,000 kg) and the partition volume limit (such as A area: 18m 3 ); the operational constraints are defined as dangerous goods isolation rules (such as IATA Class 3 and Class 7 goods need to be separated by 2 cargo positions) and priority loading order (such as medical supplies are loaded first); the business constraints include scheme calculation time threshold (≤10s) and model adaptation parameters (such as B777F cargo compartment door size limit). The final prompt word adopts a three-part structure of role definition + constraint description + input and output specification, and the cargo compartment parameters are clearly defined by Markdown (a language written in easy-to-read and easy-to-write plain text format) table and the output format is defined by JSON Schema (metadata describing JSON data structure). The prompt word and search space are input into the target model, and the search space is reduced under the driving of the large model.

[0039] In step S104, N initial cargo allocation schemes are randomly extracted from the reduced search space, and a target cargo allocation scheme is selected from the N initial cargo allocation schemes based on a target evaluation function, where N is a positive integer.

[0040] In step S104, N initial cargo allocation schemes are randomly extracted from the reduced search space, and a target evaluation function is defined, which comprehensively considers the matching degree of cargo weight and cargo bay position, the center of gravity offset after allocation, whether all constraint conditions are met, and additional business objectives (such as minimizing fuel consumption). Based on this evaluation function, the highest-scoring cargo allocation scheme is selected from the N initial schemes as the target cargo allocation scheme.

[0041] For example, based on the prompt word initialization scheme generation, N initial cargo allocation schemes are generated by chain reasoning using the target model. The N initial cargo allocation schemes are traversed to perform initial cargo allocation scheme reorganization under multiple constraint conditions, and the reorganized scheme set maintains a size of N. A mixed scoring function (i.e., target evaluation function) is designed, and the weight coefficients are dynamically adjusted according to the airline strategy. The scoring results are stored as a floating-point vector (R is the solution space, and N is the dimension), and the scheme metadata (such as calculation time, constraint violation details) are associated. N new schemes and their scores are injected into the large model context window, and a reinforcement learning ranking strategy is adopted: 1) select excellent historical schemes; 2) evaluate each scheme according to the score; finally output m candidate schemes (usually m ≈ n / 5). The failure features of the eliminated schemes will be recorded for enhancing the constraint description of the next round of prompts. The m optimized schemes are input as a new generation, and the reorganization-scoring-screening process is repeated. The search space is dynamically shrunk in each iteration, and finally the highest-scoring scheme is selected from the m candidate schemes.

[0042] In step S105, based on the target cargo allocation scheme, the cargo in the set of cargo to be allocated is allocated to the target cargo bay.

[0043] In step S105, according to the selected target cargo allocation scheme, the cargo loading is implemented to ensure that each piece of cargo is accurately placed in the specified position of the target cargo bay allocated by the target cargo allocation scheme, all constraint conditions are met, and flight efficiency and safety are optimized.

[0044] The cargo distribution method provided in the embodiments of the present application determines a target cargo hold of a passenger plane for loading a set of to-be-distributed cargos, obtains cargo features of each cargo in the set of to-be-distributed cargos, and obtains cargo hold features of the target cargo hold; obtains a plurality of preset cargo loading constraint conditions of the passenger plane, constructs a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold features and the cargo features, wherein the search space is used to represent a set of all cargo distribution schemes, and each cargo distribution scheme includes a position of each cargo in the target cargo hold; generates a prompt word based on the plurality of cargo loading constraint conditions, the cargo hold features and the cargo features, inputs the prompt word and the search space into a target model, and obtains a reduced search space, wherein the reduced search space is a search space from which cargo distribution schemes that do not meet the cargo loading constraint conditions are removed; randomly extracts N initial cargo distribution schemes from the reduced search space, and screens a target cargo distribution scheme from the N initial cargo distribution schemes based on a target evaluation function, wherein N is a positive integer; and distributes the cargos in the set of to-be-distributed cargos to the target cargo hold based on the target cargo distribution scheme, thereby solving the problem of low efficiency in distributing cargos to the cargo hold of the passenger plane in the related art. The complex multi-constraint scenario is structured and described in the form of a prompt word, thereby simplifying the modeling process and improving the adaptability of the model and the accuracy of the optimization target. The intelligent prediction capability of the target model can quickly filter out invalid cargo distribution schemes, reduce the size of the search space, and improve the search efficiency. Thus, the efficiency of distributing cargos to the cargo hold of the passenger plane is improved.

[0045] To improve the efficiency of obtaining a target cargo distribution scheme, a search space needs to be constructed. In the cargo distribution method provided in the embodiments of the present application, constructing a search space of a heuristic algorithm based on a plurality of cargo loading constraint conditions, cargo hold features and cargo features includes: extracting a plurality of cargo holds divided by the target cargo hold from the cargo hold features, and determining cargo hold features of each cargo hold, wherein the cargo hold features include at least one of the following: a number of the cargo hold, position information of the cargo hold in the target cargo hold, and a volume upper limit and a weight bearing upper limit of the cargo hold; extracting attributes of each cargo from the cargo features, and encoding the set of to-be-distributed cargos into a target tensor based on the attributes of each cargo, wherein the attributes include at least one of the following: a cargo number, a weight, a volume and a cargo type, each row of the target tensor represents a cargo, and each column represents an attribute; constructing a compatibility matrix based on the target tensor and the cargo hold features of each cargo hold, wherein each row of the compatibility matrix represents a cargo, each column represents a cargo hold in the target cargo hold, and a value in the compatibility matrix indicates whether the cargo corresponding to the row where the value belongs is placed in the cargo hold corresponding to the column where the value belongs in accordance with the cargo loading constraint condition; and determining each type of compatibility matrix as a search solution, and constructing the search space from all search solutions, wherein each search solution represents a cargo distribution scheme.

[0046] In some embodiments, each stowage location is assigned a unique number while recording its exact location in the cargo hold, including its three-dimensional coordinates (length, width, height) or relative position (e.g., front, middle, or back) and deck information (upper or lower deck) within the cargo hold. The maximum weight and volume each stowage location can hold are determined based on the physical dimensions of the stowage location and the design specifications of the cargo hold. Each piece of cargo is assigned a unique number or identifier. The actual weight and volume of each piece of cargo are measured to ensure that the loading of the cargo does not exceed the weight and volume limits of the stowage locations. The cargo is classified according to its nature (e.g., general cargo, hazardous goods, frozen cargo). Next, based on the stowage location characteristics and cargo attributes, a compatibility matrix is constructed, the main purpose of which is to determine whether each piece of cargo can be placed in each stowage location of the cargo hold without violating the loading constraints. This matrix can be seen as a two-dimensional array, where the rows represent the cargo and the columns indicate the stowage locations, while each element in the matrix indicates whether the cargo can be compatible with a particular stowage location, usually represented by a numerical value of 0 or 1, where 0 indicates incompatibility and 1 indicates compatibility.

[0047] The combination of each row (i.e., each cargo) and each column (i.e., each stowage location) of the compatibility matrix constitutes a solution in the search space. Therefore, the search space is actually a collection of a series of cargo allocation schemes, each of which defines whether the placement of each piece of cargo in a stowage location satisfies all the constraints, such as weight limits, volume limits, and type matching rules. The construction of the search space is the first step in solving the multi-constraint civil aircraft cargo hold balancing problem. Once the search space is established, heuristic algorithms, dynamic programming, or intelligent optimization algorithms (such as genetic algorithms, simulated annealing, ant colony algorithms, etc.) can be applied to find the optimal cargo balancing scheme, i.e., the scheme that can minimize the center of gravity deviation, reduce fuel consumption, and improve flight efficiency.

[0048] For example, multi-dimensional feature data is extracted from the cargo characteristics of each cargo in the cargo manifest, and a cargo feature tensor CargoTensor∈R (N×6) is constructed, where R represents the target tensor, N represents N pieces of cargo, and each row contains attributes of 6 dimensions, including cargo ID, weight, volume, hazardous category, priority (1-5 levels), and loading stage (1-3 stages). The cargo loading constraint conditions that need to be met can include: hard constraints such as cargo hold volume inequality Σv i ≤V zone , v i represents the volume of the ith cargo, V zone represents the upper limit of the cargo hold volume, and the inequality indicates that the volume of all cargoes loaded in each stowage location is less than or equal to the upper limit of the cargo hold volume. The center of gravity formula (Σm i *x i ) / Σm i ∈[x min ,x max], m i represents the mass of the i-th cargo, x i represents the position information of the i-th cargo, [x min , x max ] represents the range of the center of gravity of all cargos loaded by the passenger plane to which the center of gravity needs to be constrained, which can be obtained based on expert experience, and the soft constraint is converted into priority weight w = 1 + 0.2 * (p - 1), w is the loading priority weight of the cargo, and p is the cargo type. A position-cargo compatible matrix PlacementMatrix ∈ 0, 1 (M ×N) (M is the number of cargo holds, and N is the number of cargos), where element a ij = 1 indicates that the cargo i can be placed in the cargo hold j.

[0049] The target model is based on a Transformer (a neural network architecture for natural language processing) decoder that predicts layer by layer based on cargo hold partitioning: first, the selection of cargo hold partitioning is generated, then the specific cargo ID is recursively assigned, and a built-in physical verification module is called to filter obvious violation schemes. Each scheme p i represents a cargo hold assignment tree structure with weights, stored as a JSON object, containing cargo position mapping, real-time center of gravity coordinates, and constraint satisfaction marks (such as "hazard_valid": true). A temperature coefficient τ = 0.7 is used to control randomness during generation to ensure scheme diversity. Each trim result p1, p2, …, p n is composed of a series of cargo numbers and position numbers, representing the arrangement of different cargos in different positions, mathematically represented as:

[0050] p n = {g1, g2, …, g k};

[0051] where g i represents the number of the passenger plane cargo hold position where the i-th cargo is located. Each cargo hold trim result p n is composed of a set of cargo number and position number pairs, and satisfies the following constraints: each cargo number is unique; the weight corresponding to each position number cannot exceed the carrying limit of the cabin. The n initial cargo hold trim solutions are reorganized, and the reorganization formula is as follows: the current cargo hold trim solution is p n = {p1, p2, …, p k}, and each trim solution p k contains: a cargo position allocation matrix (H is the number of bays, and S is the total number of cargo positions), a constraint satisfaction identifier C k ∈ 0, 1 m (m is the total number of constraints, and 1 indicates satisfaction), and a score vector S k ∈ [sk1 s k2 ,…,s kt ](including sub-items such as gravity offset, weight balance, etc.).

[0052] The embodiment forms an ordered rule search space from the huge cargo data. This modeling process can effectively integrate complex constraints and provide accurate data support for the input of heuristic algorithms, ensuring the rationality of the search space and the precise docking of the optimization target, thereby avoiding the difficulties in modeling and calculation of complex problems in traditional methods and significantly improving the efficiency and accuracy of problem solving.

[0053] To improve the efficiency of reducing the search space, the target model needs to be trained. Optionally, in the cargo allocation method provided in the embodiment of the present application, the target model is obtained by the following method: obtaining historical cargo allocation records of the passenger plane, extracting the historical prompt words, the initial search space and the target search space after eliminating the cargo loading constraint conditions corresponding to each cargo allocation from the historical cargo allocation records; determining the historical prompt words, the initial search space and the target search space of each cargo allocation as a group of training samples, obtaining multiple groups of training samples; training the large language model based on the multiple groups of training samples to obtain the target model.

[0054] In some embodiments, in the scene of multi-constraint civil aviation passenger plane cargo compartment balancing, the amount of cargo data is large, and the attributes of the cargo data are relatively complex. Therefore, the search space generated by the data modeling module is large. Traditional heuristic algorithms face the problems of low solving efficiency and high time cost in a large search space. To solve this problem, the embodiment dynamically adjusts the search method according to the complex multi-constraint civil aviation passenger plane cargo compartment balancing scene under the driving of the large model, and reduces the search space. The square arrangement scoring mechanism is introduced to quantify the compliance of each cargo compartment balancing solution to multiple constraints in this scene, thereby deleting the constraints with low compliance, reducing the search space, and improving the search efficiency.

[0055] For example, historical cargo loading data is extracted from the airline or airport operation database, including cargo manifest, loading position, center of gravity position, load distribution, etc. for different flights. For each record, a historical prompt is constructed, which describes the details of the cargo and cargo hold, including the physical characteristics of the cargo position, the attributes of the cargo (such as weight, volume, category), and various hard constraints and soft requirements. At the same time, the initial search space (i.e. all possible cargo allocation schemes) and the target search space (i.e. valid allocation schemes that meet all constraint conditions) of each record are recorded. From the historical data, each set of historical prompts, initial search space and target search space constitutes a training sample. For example, extract a cargo allocation process from historical records, where the historical prompts describe the characteristics of the cargo and cargo hold, the initial search space contains all possible cargo allocation schemes, and the target search space excludes all schemes that violate the cargo loading constraints.

[0056] These training samples are used to train a large language model, taking the historical prompts and initial search space as input and the target search space as output. The learning goal of the model is to understand and generate valid cargo allocation schemes that meet specific constraint conditions. A neural network architecture suitable for processing sequential data, such as Transformer, is selected to process the sequential information of prompts and search spaces. A loss function is defined to measure the difference between the search space generated by the model and the historical target search space, which can be intersection over union, mean absolute error, mean squared error, etc., depending on the representation of the search space. The backpropagation algorithm is used to adjust the model parameters to minimize the loss function. At the same time, reinforcement learning strategies such as policy gradient methods can be used to optimize the decision path of the model and improve its ability to generate high-quality search spaces. The trained target model can generate a search space according to the given prompts (including information about the aircraft, cargo and constraints). This space contains all cargo allocation schemes that meet the constraint conditions. Then, this search space can be further optimized to find the optimal solution with the lowest cost, highest efficiency or other objective functions.

[0057] This embodiment can learn the effective patterns of cargo balancing under multiple constraints by training the target model based on historical data, thereby improving the efficiency and safety of civil aviation passenger aircraft cargo balancing.

[0058] After obtaining N initial cargo allocation schemes, a target cargo allocation scheme is screened through a target evaluation function. Optionally, in the cargo allocation method provided in the embodiments of the present application, screening a target cargo allocation scheme from the N initial cargo allocation schemes based on the target evaluation function includes: screening M candidate cargo allocation schemes from the N initial cargo allocation schemes based on spatial continuity, where M is less than N and M is a positive integer; determining the M candidate cargo allocation schemes as an initial population, and performing multiple iterations of recombination on the initial population to obtain a population after each iteration, where each population after iteration recombination contains M individuals, and each individual represents a candidate cargo allocation scheme; for each population after iteration, calculating the evaluation value of each individual in the population based on the target evaluation function to obtain M evaluation values, and determining the maximum evaluation value among the M evaluation values; calculating the difference between the maximum evaluation values of adjacent iteration populations, and stopping iteration in the case that the difference is less than a preset threshold, and determining the candidate cargo allocation scheme corresponding to the maximum evaluation value in all populations after iteration as the target cargo allocation scheme.

[0059] In some embodiments, the process starts by first applying the principle of spatial continuity. The consideration of spatial continuity involves the physical layout of the cargo within the cargo hold, ensuring that the arrangement of cargo in the scheme complies with practical operation and safety regulations, for example, ensuring that the cargo is loaded continuously without causing unreasonable or unsafe gaps, while complying with specific cargo loading order or position constraints. By evaluating spatial continuity, a large number of unreasonable allocation schemes are excluded, thereby narrowing the search space and screening M candidate cargo allocation schemes, which are determined as the initial population.

[0060] Subsequently, the initial population is subjected to multiple iterations of recombination. In each iteration, the population contains M individuals, each representing a candidate cargo allocation scheme. This iterative recombination process utilizes the idea of genetic algorithms, exploring new schemes through operations such as crossover and mutation, while retaining those with higher evaluation values. For each population after iteration, the evaluation value of each individual (i.e., candidate cargo allocation scheme) in the population needs to be calculated through the target evaluation function. The calculation of the evaluation value can be based on multiple indicators, such as the barycenter offset, load balance, rationality of cargo loading order, and cargo type matching degree, etc. All these indicators are integrated into a comprehensive evaluation function to comprehensively measure the pros and cons of the cargo allocation scheme.

[0061] After each iteration, M evaluation values are obtained, from which the maximum evaluation value is selected to track the best solution in the optimization process. Then, the difference between the maximum evaluation values of the adjacent iteration populations is calculated as an indicator of convergence speed. If the difference is less than the preset threshold, it means that the improvement space of iterative optimization is limited, and it can be considered that the algorithm is close to convergence, at which point the iteration can be stopped. After the entire iterative optimization process is completed, the candidate cargo distribution scheme corresponding to the maximum evaluation value in all iteration populations is the target cargo distribution scheme.

[0062] For example, physical constraints, operational constraints, and business constraints are extracted from the B777F freighter operation manual to construct a three-section prompt word containing cargo compartment parameters and dangerous goods rules. 1000 initial schemes (cargo loading schemes) are randomly generated by the Monte Carlo method, and after filtering, 50 feasible solutions (such as scheme A loading 7.2 tons in ZoneA, center of gravity 19.3%MAC) are retained and stored as a hierarchical JSON structure, including a list of cargo IDs, real-time center of gravity, and constraint satisfaction markers. Inject 1000 new schemes and their scores into the large model context window, and use reinforcement learning ranking strategy: 1) the model compares the features of historical excellent schemes (Top 5%) through cross-attention mechanism; 2) generate compliance comments for each scheme (such as "ZoneC dangerous goods spacing is insufficient by 2 cargo positions"); 3) based on the comments and scores of the trained model, double-filtering is performed: schemes that meet both Score(p)>θ (such as θ=0.7) and model confidence conf>0.8 are retained. Finally, m candidate schemes are output (usually m≈n / 5). The failure features of the eliminated schemes will be recorded to enhance the constraint description of the next round of prompt words.

[0063] Repeat the above process with m candidate schemes as the input for the next generation. The search space is dynamically shrunk at each iteration. The termination conditions for iteration include: 1) the optimal score improvement is less than 1% for 3 consecutive rounds; 2) the standard deviation σ CG <0.1%MAC; 3) manual forced interruption, continue to reduce the search space until the optimal solution is found, which satisfies all constraint conditions and has the best score of the cargo compartment allocation scheme.

[0064] In an optional embodiment, a large language model is used to drive the search space reduction of multi-constrained civil passenger aircraft cargo compartment balancing, and a genetic algorithm is used as the target heuristic algorithm for reducing the search space. In terms of scenario configuration, for the Boeing B777F freighter, the loading scheme output of 20 pieces of cargo needs to be completed within 10 seconds, meeting the hard constraints of center of gravity range 15%~25%MAC, dangerous goods spacing ≥2 positions, and 100% priority loading of medical supplies. The deployment environment is an airport dispatch server (Intel Xeon 8-core / 32GB memory), the initial population size is 50, and the iteration termination condition is that the center of gravity fluctuation is less than 0.1%MAC for 3 consecutive rounds. After the large model receives the structured prompt word, 50 initial schemes are generated (temperature coefficient τ=0.7), for example, scheme X loads MED001 medical supplies (6.8 tons) in Zone A, and HAZ003 (Class 3 dangerous goods) in Zone B in compliance, and the center of gravity is 20.1%MAC.

[0065] Genetic operations are performed on the initial population: select the top 5 elite schemes (such as load utilization rate >90%, center of gravity offset <0.5%MAC), and cross-recombine with the loading cargo compartment of the remaining schemes to generate 50 child schemes; 5% of the cargo is mutated (randomly adjusted to a compliant cargo compartment), and the spatial index is used to detect the dangerous goods spacing (no conflict between 2 cargo compartments). The scoring function calculates the load utilization rate (weight 0.6), the center of gravity offset (0.3), and the priority satisfaction degree (0.1), for example, scheme Y loads 43.5 tons (96.7% utilization rate), the center of gravity is 19.8%MAC, and medical supplies are 100% loaded, with a score of 0.85. After scoring, the large model performs double filtering on the 50 new schemes: retain schemes with a score >0.7 and a model confidence >0.8 (about 10), and eliminate case learning (such as scheme Z is marked as a violation because the spacing between Class 3 and Class 7 is 1.8 meters). The crossover rate (80%→60%) and mutation rate (5%→3%) are adjusted dynamically in each iteration. After 5 iterations (total time 9s), 3 optimal solutions are output: scheme A loads 44.7 tons (center of gravity 20.3%MAC), scheme B has priority for medical supplies, and scheme C has a dangerous goods spacing of 2 positions. The final search space is reduced from 50 schemes to 3 schemes (94% reduction), and the hard constraint violation rate is reduced from 34% to 1.2%.

[0066] This embodiment quantifies the compliance of each cargo compartment balancing solution to multiple constraints in this scenario by introducing a loading scheme scoring mechanism, thereby deleting constraints with low compliance, reducing the search space, and improving search efficiency. It can effectively find the best cargo compartment balancing scheme for the passenger aircraft, improve flight safety and cargo space utilization, and reduce airline operating costs.

[0067] The loading of part of the goods needs to meet the space continuity requirement, and therefore the initial goods allocation scheme is screened through space continuity. Optionally, in the goods allocation method provided in the embodiments of the present application, screening M candidate goods allocation schemes from N initial goods allocation schemes based on space continuity comprises: for each initial goods allocation scheme, determining all the goods location numbers between the starting goods location and the ending goods location in the preset continuous region to obtain a target goods location number set; determining a target goods set that needs to be stored adjacently, determining the goods location number to which each goods in the target goods set is allocated in the initial goods allocation scheme, and judging whether the goods location number belongs to the target goods location number set; in the case that there is at least one goods location number that does not belong to the target goods location number set, the initial goods allocation scheme is removed; in the case that the goods location numbers of all the goods in the target goods set belong to the target goods location number set, the initial goods allocation scheme is determined as a candidate goods allocation scheme, and M candidate goods allocation schemes are obtained.

[0068] In some embodiments, for each initial goods allocation scheme, all the goods location numbers between the starting goods location and the ending goods location in the preset continuous region are determined to construct a target goods location number set. The preset continuous region refers to a continuous storage region defined in the cargo hold according to a specific rule (such as goods type, priority or weight distribution). Then, a target goods set that needs to be stored adjacently is identified. This can be due to the nature of the goods (such as hazardous goods, fresh food or valuable goods) requirements, or in order to optimize the loading process and cargo hold space utilization efficiency. For each goods in the target goods set, its allocated goods location number in the initial goods allocation scheme is determined, and it is checked whether the number belongs to the target goods location number set constructed previously.

[0069] If there is any goods location number that does not belong to the target goods location number set, it means that the goods are not allocated according to the continuous region requirement, and therefore the entire initial goods allocation scheme is removed. On the contrary, if the goods location numbers of all the goods in the target goods set belong to the target goods location number set, it means that the goods are laid out according to the continuity requirement, and therefore the initial goods allocation scheme is determined as a candidate goods allocation scheme. Through the above verification and removal process, M candidate goods allocation schemes that meet the space continuity requirement are collected from N initial goods allocation schemes. M is a positive integer less than N, representing the number of candidate schemes that meet the condition selected from the initial schemes.

[0070] For example, the cargo hold allocation scheme extraction strategy: selecting a goods location region that meets the space continuity requirement:

[0071] M k,i = a h,s |h∈[h start ,h end ],s∈[s start ,send ];

[0072] Screening conditions:

[0073]

[0074] wherein a∈(0, 1) is an elite reservation coefficient, taking 0.7-0.9, M k,i is the k-th region of the i-th candidate cargo distribution scheme, a h,s h∈[h start ,h end ], s∈[s start ,s end ] represents that the cargo distribution of the k-th region is into the continuous cargo space [s start ,s end ] of the continuous cabin [h start ,h end ].

[0075] The embodiment can efficiently eliminate schemes that do not meet the continuity requirement from N initial cargo distribution schemes based on the screening algorithm based on spatial continuity, so as to bring M candidate schemes that are most likely to meet all constraint conditions to the next optimization stage. The practicability of the scheme is improved, and a more accurate starting point is provided for subsequent iterative optimization, which helps to find the best cargo cabin distribution scheme more quickly.

[0076] Optionally, in the cargo distribution method provided in the embodiment of the application, the initial population is subjected to multiple iterations of recombination, and the population after each iteration is obtained as follows: for each iteration, the population of the last iteration is sorted in descending order of evaluation values to obtain a target sequence, and the first P individuals in the target sequence are taken as parent individuals, wherein P is a positive integer, and P is less than M; the P parent individuals are subjected to cross recombination to generate M population individuals of the current iteration round.

[0077] In some embodiments, at the beginning of each iteration, the population of the last iteration (containing M individuals) is first sorted in descending order of evaluation values to generate a target sequence. Individuals with higher evaluation values correspond to better cargo allocation schemes that are closer to optimizing the objective function while satisfying all constraints. The sorting of the target sequence ensures that the best-performing individuals are at the front of the sequence, providing a basis for subsequent elite selection and crossover recombination. The first P individuals are selected from the target sequence as parent individuals for subsequent crossover recombination operations. The P parent individuals are subjected to crossover recombination operations to generate M population individuals for the current iteration round. Crossover recombination is one of the core evolutionary operations in genetic algorithms, which combines the characteristics of two or more parent individuals to produce new individuals with different genetic characteristics from the parents, aiming to explore new solution spaces, increase solution diversity, and introduce high-quality features to improve the overall fitness of the population. After each iteration, a new population is obtained. Subsequently, the evaluation values of each new individual can be calculated, and the sorting, elite preservation, and crossover recombination operations can be continued until the preset number of iterations is reached or the change in evaluation values is below the set threshold, i.e., the algorithm converges. When the algorithm converges, the individual with the highest evaluation value in the population corresponds to the optimized cargo allocation scheme.

[0078] The elite preservation strategy based on evaluation value sorting and subsequent crossover recombination operations in this embodiment can effectively optimize the initial population and gradually explore and converge to a cargo allocation scheme that satisfies the constraints and optimizes the objective function. Through continuous iteration, the algorithm can continuously evolve to improve the adaptability and quality of the solution, ultimately reaching the optimal or near-optimal solution.

[0079] The target evaluation function is used to calculate the evaluation values for screening the target cargo allocation scheme. Optionally, in the cargo allocation method provided in the present application, calculating the evaluation value of each individual in the population based on the target evaluation function includes: for each individual, calculating the target evaluation value between the cargo and the allocated cargo space of the candidate cargo allocation scheme corresponding to the individual, wherein the target evaluation value includes at least one of the following: feature matching evaluation value, center of gravity balance evaluation value, and constraint satisfaction evaluation value; and performing weighted summation on the target evaluation values based on preset weights to obtain the evaluation value of each individual.

[0080] In some embodiments, the degree of matching between the cargo attributes and the hold characteristics is measured by a feature matching evaluation value. For example, whether the weight, volume and type of the cargo are consistent with the load capacity, size and storage regulations of the hold. A high matching degree means that the cargo is reasonably arranged in a hold suitable for its characteristics. The center of gravity balance evaluation value assesses the impact of the cargo allocation scheme on the stability of the aircraft's center of gravity. Too much deviation of the center of gravity can affect flight safety and efficiency. An ideal scheme should minimize the deviation of the center of gravity and keep the aircraft within the safety envelope. The constraint satisfaction evaluation value checks whether the scheme meets all the hard constraint conditions, such as weight limit, cargo type matching, dangerous goods spacing rules, priority loading order, etc. A scheme that fully satisfies the constraint conditions will get a higher evaluation value in this item. In order to comprehensively evaluate the performance of the scheme in all objectives, the above-mentioned objective evaluation values are weighted and summed by using preset weights. The choice of weights reflects the degree of attention to different optimization objectives, which can be dynamically adjusted according to the airline strategy, flight safety standards and operational efficiency requirements. For example, if safety is the primary consideration, the weight of the center of gravity balance evaluation value may be set higher.

[0081] For example, the formula of the objective evaluation function is:

[0082] S(p n )=

[0083] w1·WeightMatch(p n )+w2·CenterOfGravity(p n )+w3·

[0084] ConstraintSatisfaction(p n );

[0085] where w1, w2, w3 are weight coefficients, which are adjusted according to actual needs. The function WeightMatch(p n ) measures the matching degree of cargo weight and cabin position, CenterOfGravity(p n ) measures the impact of the allocation scheme on the balance of the center of gravity, and ConstraintSatisfaction(p n ) measures whether the scheme meets all the constraint conditions.

[0086] The embodiment can effectively quantify the performance of the scheme on multiple optimization targets by calculating the evaluation value of each individual based on the target evaluation function, and provide guidance for the evolution of the population. The calculation of the evaluation value not only considers the direct effect of the cargo allocation scheme (such as feature matching and center of gravity balancing), but also considers the feasibility of the scheme (constraint satisfaction), ensuring the comprehensiveness and scientificity of the optimization process. Quickly eliminate low-quality schemes, retain and optimize better-performing schemes, thereby improving search efficiency, and ultimately find the optimal cargo allocation scheme.

[0087] In the process of loading the goods into the cargo compartment of the passenger plane, a plurality of constraint conditions need to be met. Optionally, in the cargo allocation method provided in the embodiment of the application, obtaining a plurality of cargo loading constraint conditions preset by the passenger plane includes: determining the type of the passenger plane, and extracting a plurality of cargo loading constraint conditions matched with the type from a preset constraint condition database, wherein the cargo loading constraint conditions include at least one of the following: a physical constraint, an operation constraint, and a business constraint, the physical constraint is used to limit the load and volume of each position of the target cargo compartment, the operation constraint is used to limit the cargo loading type and loading priority of each position of the target cargo compartment, and the business constraint is used to limit the generation time of the target cargo allocation scheme.

[0088] In some embodiments, the cargo loading constraint conditions can include: 1) weight limit rule: the total weight of the loaded cargo shall not exceed the maximum weight limit of the civil aviation passenger plane cargo compartment. 2) center of gravity offset control rule: the cargo compartment cargo distribution is calculated in real time to ensure that the aircraft center of gravity is always within the safety envelope. 3) cargo type matching rule: check whether the cargo type meets the preset classification, if the type is a specific code, ensure that the estimated weight and the actual weight are completely consistent. 4) compliance rule of container type: verify whether the container type conforms to the standard, and check whether the weight is within the range of the self-weight and the maximum carrying capacity of the container. 5) container number association rule: the serial number of the container must start with the standard code of the corresponding type, otherwise it is prohibited to load. 6) dangerous goods spacing rule: different dangerous goods need to maintain a specified safety distance, and if the spacing is insufficient, they are prohibited from being stored in the same compartment. 7) cabin load bearing linkage rule: the total weight of the cargo in a single cabin or combined cabin shall not exceed the independent or joint weight limit threshold. 8) continuous loading sequence rule: the specified cargo position needs to be loaded in sequence, and the end cargo position must meet the fixed ending requirement. 9) hierarchical loading priority rule: the bottom container needs to be loaded first, and the upper container cannot be placed before the bottom container.

[0089] 10) Limitation of cargo space activation rules: whether a specific cargo space is allowed to be loaded needs to be dynamically judged according to the loading state of other cargo spaces. 11) Special cargo space exemption rules: specified cargo spaces can break through the partial limitation conditions such as the regular loading sequence and layering. 12) Mixed cargo space isolation rules: in the mixed cargo space, a specific cargo space is only allowed to store a single level of container types. 13) Container quantity control rules: the actual loading quantity of a certain type of container in the cabin needs to meet the preset upper limit. 14) Front and rear cabin balance rules: the front and rear cabin areas are divided, and weight limits are set respectively to ensure the balance of the aircraft body. 15) Cargo space type exclusive rules: a specific cargo space is only allowed to place a specified type of container, and other types are prohibited. 16) Single cargo space weight limit rules: the weight of the goods in each cargo space cannot exceed the maximum carrying capacity of the location. 17) Special goods total quantity rules: the total weight of dangerous goods, living bodies and other special goods in the cabin needs to be strictly controlled within the safety range.

[0090] The embodiment according to the above-mentioned 17 constraint conditions converts the complex position constraints and stowage requirements in the scene into clear and structured prompt information. The huge cargo data forms an ordered rule search space. This modeling process can effectively integrate complex constraint conditions and provide accurate data support for the input of heuristic algorithms, ensure the rationality of the search space and the accurate docking of the optimization target, thereby avoiding the difficulty of modeling and calculation of complex problems in traditional methods, and significantly improving the efficiency and accuracy of problem solving.

[0091] According to another embodiment of the present application, an optional cargo allocation method is also provided, Figure 2 is a schematic diagram of an optional cargo allocation method provided according to an embodiment of the present application. As Figure 2 shown, the method comprises: extracting constraint conditions and cargo data in a multi-constrained civil aviation passenger aircraft load balancing scene, performing scene modeling based on the constraint conditions and cargo data, generating prompt words and search spaces, inputting the prompt words and search spaces into a large model, outputting an initial scheme from the large model by reducing the search space, scoring and recombining the initial scheme, scoring the recombined scheme, driving the large model to continue reducing the search space based on the score, iteratively solving, and outputting a final cargo allocation scheme.

[0092] wherein, Figure 3 is a schematic diagram of an initial reduction process provided according to an embodiment of the present application. As Figure 3 shown, all scene constraints are traversed to construct multi-constraint problem description prompt words; complex cargo feature extraction and search space modeling; fast conflict detection based on R*-Tree spatial indexing. Figure 4 is a schematic diagram of a reduction process driven by a large model according to an embodiment of the present application. As Figure 4As shown, the prompt word and the search space modeling are input into the large model to generate an initial cargo compartment allocation scheme; the variable n allocation schemes are reorganized under multiple constraint conditions; a mixed scoring function is designed to score the generated allocation results; the schemes are reorganized, and the reorganized n new schemes and their scores are injected into the large model context window, and the large model drives the scheme screening; it is judged whether the output solution meets the multiple constraint conditions, if yes, the iteration is ended and the final solution, i.e., the final allocation scheme, is output, if not, the cargo compartment allocation scheme is iterated, and the search space is updated, and the search space is dynamically shrunk in each iteration.

[0093] The embodiment fuses multiple complex conditions in the multi-constraint civil aviation passenger aircraft cargo compartment balancing scene through the optional cargo allocation method, guarantees the accuracy of the solution, and reduces the high time cost of the heuristic algorithm in the multi-constraint civil aviation passenger aircraft cargo compartment balancing scene due to the excessively large search space. The complex and abstract constraint conditions are constructed into a specific heuristic algorithm search space through modeling, so that the large model can quickly and accurately understand the complex constraint conditions under various different aircraft types and conditions of the to-be-loaded cargo, and find the cargo compartment balancing solution. Under the driving of the large model, the search space of the heuristic algorithm is further simplified, and the time cost of the solution is reduced.

[0094] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0095] The embodiment of the present application also provides a cargo allocation device. It should be noted that the cargo allocation device of the embodiment of the present application can be used to execute the cargo allocation method provided by the embodiment of the present application. The cargo allocation device provided by the embodiment of the present application is introduced as follows.

[0096] Figure 5 is a schematic diagram of the cargo allocation device provided by the embodiment of the present application. As shown in the figure, Figure 5 The device comprises:

[0097] The acquisition unit 501 is configured to determine a target cargo compartment of an aircraft for loading a set of to-be-allocated cargo, acquire a cargo feature of each cargo in the set of to-be-allocated cargo, and acquire a cargo compartment feature of the target cargo compartment;

[0098] The construction unit 502 is configured to acquire a plurality of preset cargo loading constraint conditions of the aircraft, and construct a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo compartment feature, and the cargo feature, wherein the search space is used to represent a set of all cargo allocation schemes, and each cargo allocation scheme includes a position of each cargo allocated in the target cargo compartment;

[0099] The input unit 503 is configured to generate a prompt word based on the plurality of cargo loading constraint conditions, the cargo hold feature and the cargo feature, input the prompt word and the search space into a target model, and obtain a reduced search space, where the reduced search space is a search space from which cargo allocation schemes that do not meet the cargo loading constraint conditions are removed.

[0100] The screening unit 504 is configured to randomly extract N initial cargo allocation schemes from the reduced search space, and screen a target cargo allocation scheme from the N initial cargo allocation schemes based on a target evaluation function, where N is a positive integer.

[0101] The allocation unit 505 is configured to allocate the cargos in the set of cargos to be allocated to the target cargo hold based on the target cargo allocation scheme.

[0102] The cargo allocation device provided by the embodiments of the present application comprises the obtaining unit 501, the construction unit 502, the input unit 503, the screening unit 504 and the allocation unit 505. The obtaining unit 501 is configured to determine a target cargo hold of an aircraft for loading a set of cargos to be allocated, obtain a cargo feature of each cargo in the set of cargos to be allocated, and obtain a cargo hold feature of the target cargo hold. The construction unit 502 is configured to obtain a plurality of cargo loading constraint conditions preset by the aircraft, and construct a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold feature and the cargo feature, where the search space is used to represent a set of all cargo allocation schemes, and each cargo allocation scheme comprises a position of each cargo allocated to the target cargo hold. The input unit 503 is configured to generate a prompt word based on the plurality of cargo loading constraint conditions, the cargo hold feature and the cargo feature, input the prompt word and the search space into a target model, and obtain a reduced search space, where the reduced search space is a search space from which cargo allocation schemes that do not meet the cargo loading constraint conditions are removed. The screening unit 504 is configured to randomly extract N initial cargo allocation schemes from the reduced search space, and screen a target cargo allocation scheme from the N initial cargo allocation schemes based on a target evaluation function, where N is a positive integer. The allocation unit 505 is configured to allocate the cargos in the set of cargos to be allocated to the target cargo hold based on the target cargo allocation scheme. The cargo allocation device provided by the embodiments of the present application solves the problem of low efficiency of allocating cargos to the cargo hold of the aircraft in the related art, simplifies the modeling process by structurally describing a complex multi-constraint scenario through a prompt word, and improves the adaptability of the model and the accuracy of the optimization target. The intelligent prediction capability of the target model can quickly filter out invalid cargo allocation schemes, reduce the size of the search space, and improve the search efficiency. Thus, the efficiency of allocating cargos to the cargo hold of the aircraft is improved.

[0103] Optionally, in the cargo distribution apparatus provided by the embodiment of the present application, the constructing unit 502 comprises: a first extraction module configured to extract a plurality of cargo bays divided by the target cargo bay from the cargo bay features, and determine the cargo bay features of each cargo bay, wherein the cargo bay features comprise at least one of the following: the number of the cargo bay, the position information of the cargo bay in the target cargo bay, and the upper limit of the volume and the upper limit of the load capacity of the cargo bay; a second extraction module configured to extract the attributes of each cargo from the cargo features, and encode the set of cargos to be distributed into a target tensor based on the attributes of each cargo, wherein the attributes comprise at least one of the following: the cargo number, the weight, the volume, and the cargo type, each row of the target tensor represents a cargo, and each column represents an attribute; a first constructing module configured to construct a compatibility matrix based on the target tensor and the cargo bay features of each cargo bay, wherein each row of the compatibility matrix represents a cargo, each column represents a cargo bay in the target cargo bay, and the numerical value in the compatibility matrix indicates whether the cargo corresponding to the row of the numerical value is placed in the cargo bay corresponding to the column of the numerical value in accordance with the cargo loading constraint condition; and a first determining module configured to determine each compatibility matrix as a search solution, and determine the search space by all search solutions, wherein each search solution represents a cargo distribution scheme.

[0104] Optionally, in the cargo distribution apparatus provided by the embodiment of the present application, the target model is obtained by training in the following manner: obtaining the historical cargo distribution records of the passenger plane, extracting the historical prompt words, the initial search space, and the target search space after eliminating the search solutions that do not meet the cargo loading constraint condition corresponding to each cargo distribution from the historical cargo distribution records; determining the historical prompt words, the initial search space, and the target search space of each cargo distribution as a set of training samples, and obtaining a plurality of sets of training samples; training the large language model based on the plurality of sets of training samples, and obtaining the target model.

[0105] Optionally, in the cargo distribution apparatus provided by the embodiment of the present application, the screening unit 504 comprises: a first screening module configured to screen M candidate cargo distribution schemes from N initial cargo distribution schemes based on the spatial continuity, wherein M is less than N, and M is a positive integer; a second determining module configured to determine the M candidate cargo distribution schemes as the initial population, and perform multiple iteration recombinations on the initial population to obtain the population after each iteration, wherein each iteration recombinant population comprises M individuals, and each individual represents a candidate cargo distribution scheme; a first calculating module configured to, for each iteration recombinant population, calculate the evaluation value of each individual in the population based on the target evaluation function, obtain M evaluation values, and determine the maximum evaluation value in the M evaluation values; and a second calculating module configured to calculate the difference between the maximum evaluation values of adjacent iteration populations, stop iteration in the case that the difference is less than a preset threshold, and determine the candidate cargo distribution scheme corresponding to the maximum evaluation value in all iteration recombinant populations as the target cargo distribution scheme.

[0106] Optionally, in the goods distribution apparatus provided by the embodiment of the present application, the first screening module comprises: a first determining submodule, configured to determine all the storage locations between the starting storage location and the ending storage location in the preset continuous area for each initial goods distribution scheme, to obtain a target storage location number set; a second determining submodule, configured to determine a target goods set requiring adjacent storage, determine the storage location number of each goods in the target storage location set in the initial goods distribution scheme, and judge whether the storage location number belongs to the target storage location number set; a rejecting submodule, configured to reject the initial goods distribution scheme in the case that the storage location number of at least one goods does not belong to the target storage location number set; and a third determining submodule, configured to determine the initial goods distribution scheme as a candidate goods distribution scheme in the case that the storage location number of all the goods in the target goods set belongs to the target storage location number set, to obtain M candidate goods distribution schemes.

[0107] Optionally, in the goods distribution apparatus provided by the embodiment of the present application, the second determining module comprises: an ordering submodule, configured to order the population of the last iteration in descending order of the evaluation values for each iteration, to obtain a target sequence, and take the first P individuals in the target sequence as the parent individuals, wherein P is a positive integer, and P is less than M; and a generating submodule, configured to perform cross-recombination on the P parent individuals, to generate M population individuals of the current iteration round.

[0108] Optionally, in the goods distribution apparatus provided by the embodiment of the present application, the first calculating module comprises: a first calculating submodule, configured to calculate, for each individual, a target evaluation value between the goods and the allocated storage locations of the candidate goods distribution scheme corresponding to the individual, wherein the target evaluation value comprises at least one of the following: a feature matching evaluation value, a barycenter balance evaluation value and a constraint satisfaction evaluation value; and a summing submodule, configured to perform weighted summation on the target evaluation values based on preset weights, to obtain the evaluation value of each individual.

[0109] Optionally, in the goods distribution apparatus provided by the embodiment of the present application, the construction unit 502 comprises: a third determining module, configured to determine the type of the passenger aircraft, and extract a plurality of goods loading constraint conditions matching the type from a preset constraint condition database, wherein the goods loading constraint condition comprises at least one of the following: a physical constraint, an operation constraint and a business constraint, the physical constraint is used to limit the load and the volume of each position of the target cargo hold, the operation constraint is used to limit the goods loading type and the loading priority of each position of the target cargo hold, and the business constraint is used to limit the generation time of the target goods distribution scheme.

[0110] The cargo distribution device includes a processor and a memory. The acquisition unit 501, construction unit 502, input unit 503, screening unit 504 and distribution unit 505 are all stored in the memory as program units. The processor executes the program units stored in the memory to implement corresponding functions.

[0111] The processor contains a core, which retrieves the corresponding program unit from the memory. There can be one or more cores, and the efficiency of distributing cargo to the cargo hold of passenger aircraft can be improved by adjusting the core parameters.

[0112] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0113] An embodiment of the present invention provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a cargo distribution method is implemented.

[0114] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes a cargo distribution method when it is run.

[0115] Figure 6 Schematic diagram of an electronic device according to an embodiment of the present application. Figure 6 As shown, the electronic device 601 includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are implemented: determining a target cargo hold of a passenger aircraft for loading a set of cargo to be allocated, obtaining cargo characteristics of each cargo in the set of cargo to be allocated, and obtaining cargo hold characteristics of the target cargo hold; obtaining multiple cargo loading constraints preset for the passenger aircraft, and constructing a search space for a heuristic algorithm based on the multiple cargo loading constraints, cargo hold characteristics, and cargo characteristics, wherein the search space is used to represent a set of all cargo allocation plans, each cargo allocation plan including the position of each cargo allocated in the target cargo hold; generating a prompt word based on the multiple cargo loading constraints, cargo hold characteristics, and cargo characteristics, and inputting the prompt word and the search space into a target model to obtain a reduced search space, wherein the reduced search space is a search space that eliminates cargo allocation plans that do not meet the cargo loading constraints; randomly extracting N initial cargo allocation plans from the reduced search space, and screening a target cargo allocation plan from the N initial cargo allocation plans based on a target evaluation function, wherein N is a positive integer; and allocating the cargo in the set of cargo to be allocated to the target cargo hold based on the target cargo allocation plan. The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0116] The application further provides a computer program product, which is suitable for executing the program of the following method steps when executed on a data processing device: determining a target cargo hold of a passenger plane for loading a set of goods to be allocated, obtaining a cargo characteristic of each good in the set of goods to be allocated, and obtaining a cargo hold characteristic of the target cargo hold; obtaining a plurality of preset cargo loading constraint conditions of the passenger plane, constructing a search space of a heuristic algorithm based on the plurality of cargo loading constraint conditions, the cargo hold characteristic, and the cargo characteristic, wherein the search space is used to represent a set of all cargo allocation schemes, and each cargo allocation scheme includes a position of each good allocated into the target cargo hold; generating a prompt word based on the plurality of cargo loading constraint conditions, the cargo hold characteristic, and the cargo characteristic, inputting the prompt word and the search space into a target model, and obtaining a reduced search space, wherein the reduced search space is the search space from which the cargo allocation schemes that do not meet the cargo loading constraint conditions are removed; randomly extracting N initial cargo allocation schemes from the reduced search space, and screening a target cargo allocation scheme from the N initial cargo allocation schemes based on a target evaluation function, wherein N is a positive integer; and allocating the goods in the set of goods to be allocated into the target cargo hold based on the target cargo allocation scheme.

[0117] Those skilled in the art will understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer usable program code.

[0118] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices generate a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0119] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the function(s) specified in the block or blocks.

[0120] These computer program instructions can also be loaded into computer or other programmable data processing devices to cause a series of operational steps to be performed on the computer or other programmable devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable devices provide steps for implementing the functions of the flow Figure 1 one or more processes and / or blocks Figure 1 the function(s) specified in the block or blocks.

[0121] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0122] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) about which the computer stores information such as computer program instructions. Memory is an example of computer readable media.

[0123] Computer readable media includes permanent and non-permanent, moveable and non- moveable media that can be implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memories (CD-ROMs), digital versatile discs (DVDs), or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that is accessible to a computing device. According to the definition provided herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0124] It is also important to note that the terms "comprises", "comprising", or other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0125] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code thereon for use by a computer or computer system.

[0126] The foregoing is merely illustrative of the embodiments of the present application and is not used to limit the present application. The present application can be varied in many ways. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the scope of claims of the present application.

Claims

1. A cargo distribution method, characterized in that: include: Determining a target cargo hold of a passenger aircraft for loading the set of cargo to be allocated, obtaining cargo characteristics of each cargo in the set of cargo to be allocated, and obtaining cargo hold characteristics of the target cargo hold; obtaining a plurality of cargo loading constraints preset for the passenger aircraft, and constructing a search space for a heuristic algorithm based on the plurality of cargo loading constraints, the cargo hold characteristics, and the cargo characteristics, wherein the search space is used to represent a set of all cargo allocation schemes, each cargo allocation scheme including a position of each cargo allocated to the target cargo hold; generating prompt words based on the multiple cargo loading constraints, the cargo hold characteristics, and the cargo characteristics, and inputting the prompt words and the search space into a target model to obtain a reduced search space, wherein the reduced search space is a search space that excludes cargo allocation solutions that do not meet the cargo loading constraints; Randomly extracting N initial cargo allocation plans from the reduced search space, and screening out a target cargo allocation plan from the N initial cargo allocation plans based on a target evaluation function, where N is a positive integer; The cargo in the to-be-allocated cargo set is allocated to the target cargo hold based on the target cargo allocation plan.

2. The method according to claim 1, characterized in that Constructing a search space of a heuristic algorithm based on the plurality of cargo loading constraints, the cargo hold characteristics, and the cargo characteristics includes: Extracting a plurality of cargo spaces divided into the target cargo hold from the cargo hold characteristics, and determining cargo space characteristics of each cargo space, wherein the cargo space characteristics include at least one of the following: a number of the cargo space, location information of the cargo space in the target cargo hold, and a volume upper limit and a weight upper limit of the cargo space; Extracting attributes of each cargo from the cargo features, and encoding the set of cargo to be allocated into a target tensor based on the attributes of each cargo, wherein the attributes include at least one of the following: cargo number, weight, volume, and cargo type, each row of the target tensor represents one cargo, and each column represents one attribute; Constructing a compatibility matrix based on the target tensor and the cargo location characteristics of each cargo location, wherein each row in the compatibility matrix represents a cargo, each column represents a cargo location in the target cargo hold, and a value in the compatibility matrix indicates whether the cargo corresponding to the row to which the value belongs is placed in the cargo location corresponding to the column to which the value belongs meets the cargo loading constraint condition; Each compatibility matrix is ​​determined as a search solution, and all search solutions constitute the search space, wherein each search solution represents a cargo allocation plan.

3. The method according to claim 1, characterized in that The target model is trained in the following way: Obtaining historical cargo allocation records of the passenger aircraft, extracting historical prompt words corresponding to each cargo allocation, an initial search space, and a target search space after eliminating those that do not meet cargo loading constraints from the historical cargo allocation records; The historical prompt words, initial search space and target search space of each cargo allocation are determined as a set of training samples to obtain multiple sets of training samples; A large language model is trained based on the multiple groups of training samples to obtain the target model.

4. The method according to claim 1, wherein Screening out a target cargo allocation solution from the N initial cargo allocation solutions based on the target evaluation function includes: Selecting M candidate cargo allocation plans from the N initial cargo allocation plans based on spatial continuity, where M is less than N and is a positive integer; Determine the M candidate cargo allocation plans as an initial population, perform multiple iterative recombination on the initial population, and obtain a population after each iteration, wherein the population after each iterative recombination includes M individuals, each individual representing a candidate cargo allocation plan; For the population after each iteration, calculating the evaluation value of each individual in the population based on the target evaluation function, obtaining M evaluation values, and determining the maximum evaluation value among the M evaluation values; The difference between the maximum evaluation values ​​of adjacent iterative populations is calculated, and the iteration is stopped when the difference is less than a preset threshold value, and the candidate cargo allocation scheme corresponding to the maximum evaluation value in all iterated populations is determined as the target cargo allocation scheme.

5. The method according to claim 4, characterized in that M candidate cargo allocation plans are selected from the N initial cargo allocation plans based on spatial continuity, including: For each initial cargo allocation plan, determine all cargo location numbers between the starting cargo location and the ending cargo location in the preset continuous area to obtain a target cargo location number set; Determine a target set of goods that need to be stored adjacently, determine the cargo location number assigned to each of the goods in the target cargo location set in the initial cargo allocation plan, and determine whether the cargo location number belongs to the target cargo location number set; If there is at least one cargo location number that does not belong to the target cargo location number set, the initial cargo allocation plan is eliminated; When the cargo location numbers of all cargoes in the target cargo set belong to the target cargo location number set, the initial cargo allocation plan is determined as the candidate cargo allocation plan, and M candidate cargo allocation plans are obtained.

6. The method according to claim 4, characterized in that The initial population is iteratively reorganized multiple times, and the population obtained after each iteration includes: For each iteration, the population of the previous iteration is sorted in descending order of evaluation value to obtain the target sequence, and the first P individuals in the target sequence are used as parent individuals, where P is a positive integer and P is less than M; Perform crossover recombination on P parent individuals to generate M population individuals for the current iteration round.

7. The method according to claim 4, characterized in that Calculating the evaluation value of each individual in the population based on the target evaluation function includes: For each individual, calculating a target evaluation value between the cargo and the allocated cargo location in the candidate cargo allocation solution corresponding to the individual, wherein the target evaluation value includes at least one of the following: a feature matching evaluation value, a center of gravity balance evaluation value, and a constraint satisfaction evaluation value; The target evaluation values ​​are weighted and summed based on preset weights to obtain the evaluation value of each individual.

8. The method according to claim 1, characterized in that Obtaining multiple cargo loading constraints preset for the passenger aircraft includes: Determine the type of the passenger aircraft, and extract multiple cargo loading constraints that match the type from a preset constraint database, wherein the cargo loading constraints include at least one of the following: physical constraints, operational constraints, and business constraints, the physical constraints are used to limit the load and volume of each position in the target cargo hold, the operational constraints are used to limit the cargo loading type and loading priority of each position in the target cargo hold, and the business constraints are used to limit the generation time of the target cargo allocation plan.

9. A cargo distribution device, characterized in that: include: an acquiring unit, configured to determine a target cargo hold of a passenger aircraft for loading the set of cargo to be allocated, acquire cargo characteristics of each cargo in the set of cargo to be allocated, and acquire cargo hold characteristics of the target cargo hold; a construction unit, configured to obtain a plurality of cargo loading constraints preset for the passenger aircraft, and construct a search space for a heuristic algorithm based on the plurality of cargo loading constraints, the cargo hold characteristics, and the cargo characteristics, wherein the search space is used to represent a set of all cargo allocation schemes, each cargo allocation scheme including a position of each cargo allocated to the target cargo hold; an input unit, configured to generate a prompt word based on the multiple cargo loading constraints, the cargo hold characteristics, and the cargo characteristics, and input the prompt word and the search space into a target model to obtain a reduced search space, wherein the reduced search space is a search space that excludes cargo allocation solutions that do not meet the cargo loading constraints; a screening unit, configured to randomly extract N initial cargo allocation solutions from the reduced search space, and screen out a target cargo allocation solution from the N initial cargo allocation solutions based on a target evaluation function, where N is a positive integer; An allocating unit is configured to allocate the cargo in the set of cargo to be allocated to the target cargo hold based on the target cargo allocation plan.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the cargo distribution method according to any one of claims 1 to 8 is implemented.

Images