Method and system for predicting loading allowance of aviation container board carrying and related equipment
By comparative learning model generating implicit margin features in the embedded space, the problem of load margin prediction in irregular loading scenarios is solved, accurate margin prediction for irregular loading scenarios is achieved, and the efficiency of air transportation is improved.
Patent Information
- Application Number
- CN202411659076.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-20
- Publication Date
- 2025-03-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to accurately predict the loading margin of aviation container plates in irregular loading scenarios, resulting in inefficient transportation and waste of resources.
By obtaining the current cargo space state on the container board, using the comparative learning model to generate implicit margin features in the embedded space, and predicting the training model parameters, the margin prediction of irregular loading scenarios is achieved.
This method can accurately predict the load margin of irregular loading scenes without clear irregular scene margin labels, thereby improving transportation efficiency by learning the characteristic differences between rules and irregular loading scenes.
Smart Images
Figure CN119599187A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of pallet loading margin prediction, and specifically to a method, system, electronic device and storage medium for predicting the loading margin of aviation pallet transportation. Background Art
[0002] In the field of air transportation, air pallets, as an important carrier for cargo transportation, are widely used in the intensive transportation of bulk goods, express parcels, etc. Due to the limited space on the aircraft, accurately predicting the loading margin of the pallet is crucial to optimizing the loading strategy and improving transportation efficiency. Especially during the loading process, factors such as the placement, shape, and size of the cargo will affect the utilization of the loading space. Therefore, how to effectively and accurately predict the loading margin has become a key issue in air logistics management.
[0003] In the prior art, loading margin prediction usually relies on computing models or visual inspection technologies based on regular loading scenarios. For example, the utilization rate of the loading space is calculated by analyzing the volume, position and other characteristics of the goods on the container. However, these methods have limited performance in irregular loading scenarios. When the goods are irregular in shape, randomly placed or have gaps, it is difficult for traditional methods to accurately estimate the loading margin. In this case, the prediction error of irregular scenarios is large, which affects the overall transportation efficiency and may even lead to overloading or waste of space. It is costly to provide clear loading margin labels for each irregular loading scenario. Therefore, in the absence of a large number of irregular loading scenario margin labels, how to use limited regular loading scenario data to indirectly realize the margin prediction of irregular loading scenarios has become a research difficulty. Summary of the invention
[0004] This embodiment provides a method, system, electronic device and storage medium for predicting the loading margin of aviation pallets, and explores how to solve the problem of how to use limited regular loading scene data to indirectly realize the margin prediction of irregular loading scenes without a large number of irregular loading scene margin labels.
[0005] In a first aspect, the present invention provides a method for predicting the loading margin of an aviation pallet transport.
[0006] Obtain the current cargo space status on the container;
[0007] Receiving the current cargo space state as input of a preset surplus prediction model, and generating implicit surplus features of the current loading scenario in the embedding space;
[0008] The implicit margin feature of the current loading scenario is used as the input of the final output layer, and the predicted value of the loading margin is obtained through the trained model parameters of the margin prediction model.
[0009] The output is the predicted value of the loading margin, which is expressed as: Y = f embed (X c θ * )
[0010] Y represents the predicted value of the loading margin, X c represents the current cargo space state, θ * Represents the trained model parameters of the residual prediction model.
[0011] In some of the embodiments, the method for constructing the surplus prediction model includes:
[0012] S1. Obtain several positive sample pairs and negative sample pairs for residual prediction model training;
[0013] S2, initialize the three-dimensional convolution model to obtain the contrastive learning model;
[0014] S3. Inputting the plurality of positive sample pairs and negative sample pairs into a contrastive learning model for iterative training to obtain the residual prediction model.
[0015] In some embodiments, the step of obtaining a plurality of positive sample pairs and negative sample pairs for residual prediction model training includes:
[0016] S1-1, obtaining a first rule sample of a regular loading scenario on a container board;
[0017] S1-2, performing several minor difference enhancements on the first regular sample to generate several second regular samples;
[0018] S1-3, pairing a number of second regular samples with the first regular samples respectively to construct a number of positive sample pairs;
[0019] S1-4, obtaining several irregular samples of irregular loading scenes on the container board;
[0020] S1-5. Pair several irregular samples with the first regular sample to construct several negative sample pairs.
[0021] In some embodiments, obtaining a first regular sample of a regular loading scenario on a container includes:
[0022] S1-1-1, obtaining three-dimensional point cloud data of goods loaded on a number of pallets;
[0023] S1-1-2, voxelizing the three-dimensional point cloud data into a three-dimensional voxel grid of standard size, and assigning values to the three-dimensional voxel grid based on the cargo occupancy status in the three-dimensional point cloud data; wherein an occupied voxel grid is assigned a value of 1, and an idle voxel grid is assigned a value of 0;
[0024] S1-1-3. Standardize and normalize the assigned three-dimensional grid data to a uniform size to obtain a first regular sample.
[0025] In some embodiments, the initializing the three-dimensional convolutional model to obtain the contrastive learning model includes:
[0026] S2-1, load the pre-trained general 3D convolutional model;
[0027] S2-2, delete the fully connected layer and classification layer of the general 3D convolutional model, and keep the convolutional layer;
[0028] S2-3, add a contrastive learning head between the convolutional layer and the output layer to generate an embedding vector space;
[0029] S2-4, defining a contrastive learning loss function in the embedding vector space by maximizing the similarity of positive samples and minimizing the similarity of negative samples;
[0030] S2-5. Use the three-dimensional convolutional model having the contrastive learning head and the contrastive learning loss function as the initialized contrastive learning model.
[0031] In some embodiments, the iterative training of the margin prediction model includes:
[0032] S3-1, receiving the plurality of positive sample pairs and negative sample pairs as inputs of the contrastive learning model, and generating positive sample pair embedding vectors and negative sample pair embedding vectors through forward propagation of the convolutional layer and the contrastive learning head;
[0033] The expression for generating the embedding vector of the positive sample pair is: p =f(X p ),h q =f(X q );
[0034] X p represents the first regular sample, X q represents the second regular sample, f represents the mapping function after the convolutional layer and the contrastive learning head, and h p represents the embedding vector of the first regular sample, h q The embedding vector representing the second regular sample;
[0035] The expression of the negative sample pair embedding vector is: p =f(X p ),h n =f(X n );
[0036] X n represents irregular samples, h nEmbedding vector representing irregular samples;
[0037] S3-2. In the embedding vector space, the cosine similarity is used to calculate the positive sample similarity between the positive sample pairs and the negative sample similarity between the negative sample pairs;
[0038] The expression of the positive sample similarity is:
[0039] s pq Represents the positive sample similarity, which represents the cosine similarity between the embedding vectors of the positive sample pair;
[0040] The expression of the negative sample similarity is:
[0041] s nq Represents negative sample similarity, which represents the cosine similarity between the embedding vectors of negative sample pairs;
[0042] S3-3, according to the contrastive learning loss function, maximize the similarity of positive sample pairs while minimizing the similarity of negative sample pairs;
[0043] The contrastive learning loss function is:
[0044] represents the contrast loss, N represents the total number of positive and negative sample pairs involved in the loss calculation, represents the similarity of the i-th positive sample pair, represents the similarity of the i-th negative sample pair, and m represents the boundary value of the contrast loss, which is used to control the minimization of the similarity of the negative sample pair; Represents the minimization part of the similarity of negative samples; Represents the part that maximizes the similarity of the positive sample pair.
[0045] In some embodiments, before maximizing the similarity of the positive sample pairs and minimizing the similarity of the negative sample pairs, the method further includes:
[0046] Compare the positive sample similarity with the preset positive sample similarity threshold;
[0047] When the positive sample similarity reaches or exceeds the preset positive sample similarity threshold, the learning rate of the negative sample pair is dynamically adjusted; the expression for dynamically adjusting the learning rate of the negative sample pair is: η neg =η base +k·(s pq -τ) p ;
[0048] η neg Denotes the learning rate of negative samples after dynamic adjustment, ηbase represents the initial learning rate of the negative sample pair, k represents the adjustment coefficient, which is used to control the learning rate increase, and s pq represents the similarity of the positive sample pair, τ represents the positive sample similarity threshold, and p represents the exponential parameter, which is used to control the amplitude of nonlinear adjustment;
[0049] Back propagation is performed using the adjusted learning rate to update the model parameters of the contrastive learning model until the model reaches a convergence condition, and the contrastive learning model of the current model parameters is derived and defined as the residual prediction model.
[0050] The expression of the margin prediction model is: M = f embed (X,θ * ),X=(X p , X q , X n );
[0051] M represents the margin prediction model, f embed represents the training step, and X represents the input sample in the training step.
[0052] Compared with the prior art, the loading margin prediction method for aviation pallets of the present invention can learn the significant feature differences between regular loading scenes and irregular loading scenes through comparative learning of positive sample pairs and negative sample pairs, so that the model can implicitly infer the margin space state of irregular loading scenes even without clear irregular scene margin labels. It uses the contrastive learning loss function to maximize the similarity of positive sample pairs and minimize the similarity of negative sample pairs, so that the model can form a clear distinction between regular and irregular loading features. The model can identify and quantify the loading features of irregular scenes in the application stage, thereby more accurately predicting the margin.
[0053] In a second aspect, the present invention provides a loading margin prediction system for aviation pallet transportation, comprising: a current loading status acquisition module, for acquiring the current cargo space status on the pallet;
[0054] An implicit surplus feature embedding module is used to receive the current cargo space state as an input of a preset surplus prediction model and generate an implicit surplus feature of the current loading scenario in an embedding space;
[0055] The loading margin prediction module is used to use the implicit margin feature of the current loading scenario as the input of the final output layer, and obtain the predicted value of the loading margin through the trained model parameters of the margin prediction model.
[0056] In a third aspect, the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores at least one computer executable instruction, and the processor is configured to execute the computer executable instruction, and when the computer executable instruction is executed by the processor, the loading margin prediction method for aviation pallet transportation described in the first aspect is implemented.
[0057] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the loading margin of air pallet transportation as described in the first aspect.
[0058] Compared with the prior art, the beneficial effects of the system, electronic device and storage medium for predicting the loading margin of an aviation pallet carried by the present invention are the same as the beneficial effects of the method for predicting the loading margin of an aviation pallet carried by the above-mentioned, so they are not repeated here. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] Figure 1 This is an application flow chart of the surplus prediction model of the present invention;
[0060] Figure 2 is a training flow chart of the surplus prediction model of the present invention;
[0061] Figure 3 This is a structural block diagram of the loading margin prediction system for aviation pallet transportation according to the present invention;
[0062] Figure 4 A diagram of an electronic device of the present invention; DETAILED DESCRIPTION
[0063] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.
[0064] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.
[0065] refer to Figures 1 to 4 The present invention discloses:
[0066] Embodiment 1: This embodiment provides a method for predicting the loading margin of an aviation pallet. The process of the method for predicting the loading margin includes:
[0067] The current cargo space status on the container is obtained.
[0068] In this embodiment, the spatial distribution information of the cargo in the current loading state can be collected in real time by the sensing device, and the cargo spatial state includes the three-dimensional position, shape, density and space occupancy of the cargo.
[0069] The current cargo space status is received as the input of the preset surplus prediction model, and the implicit surplus feature of the current loading scenario is generated in the embedding space.
[0070] The collected cargo space status is input into the surplus prediction model, and the model generates corresponding implicit surplus features in the embedding space to reflect the spatial characteristics of the current loading scene.
[0071] The implicit margin feature of the current loading scenario is used as the input of the final output layer, and the predicted value of the loading margin is obtained through the trained model parameters of the margin prediction model.
[0072] The output is the predicted value of the loading margin, which is expressed as: Y = f embed (X c θ * )
[0073] Y represents the predicted value of the loading margin, X c represents the current cargo space state, θ * Represents the trained model parameters of the residual prediction model.
[0074] In this embodiment, implicit margin features are generated through the embedding space of the margin prediction model to achieve prediction of the loading margin in irregular loading scenarios. The prediction process does not require explicit margin labels, but is inferred through the hidden feature representation generated in the embedding space, providing effective prediction of the margin in irregular loading scenarios.
[0075] Embodiment 2: The technical solution of Embodiment 2 is different from that of Embodiment 1 in that a method for constructing the surplus prediction model described in Embodiment 1 is disclosed, and the method for constructing the surplus prediction model includes:
[0076] S1. Obtain several positive sample pairs and negative sample pairs for residual prediction model training.
[0077] Specifically, the positive sample pair contains the cargo space state of the regular loading scenario and its corresponding surplus label. The cargo space state refers to the specific three-dimensional spatial distribution information of the cargo on the pallet, which usually includes the location, shape, density and volume of the cargo. Furthermore, the cargo space state is represented by three-dimensional point cloud data to accurately reflect the location of each cargo and its relative position in the pallet.
[0078] The model will learn the cargo space state in regular loading scenarios and its direct association with the margin label through positive sample pairs. The negative sample pairs contain the cargo space state in irregular loading scenarios but do not contain explicit margin labels. The model will learn the significant differences between the cargo space state in irregular loading scenarios and regular loading scenarios through negative sample pairs.
[0079] Furthermore, the S1 specifically includes:
[0080] S1-1. Obtain a first regular sample of a regular loading scenario on a container.
[0081] Specifically, the first rule sample includes the cargo space status and the corresponding balance label in the regular loading scenario. It can build an application scenario of the regular loading scenario on the container through the on-site field model, so as to obtain the above-mentioned first rule sample based on the constructed application scenario.
[0082] Furthermore, the S1-1 specifically includes:
[0083] S1-1-1. Obtain three-dimensional point cloud data of goods loaded on several pallets.
[0084] S1-1-2, voxelizing the three-dimensional point cloud data into a three-dimensional voxel grid of standard size, and assigning values to the three-dimensional voxel grid based on the cargo occupancy status in the three-dimensional point cloud data, wherein an occupied voxel grid is assigned a value of 1, and an idle voxel grid is assigned a value of 0.
[0085] S1-1-3. Standardize and normalize the assigned three-dimensional grid data to a uniform size to obtain a first regular sample.
[0086] Specifically, slight difference enhancement refers to performing operations that do not significantly enhance the data on the first regular sample, such as small-angle rotation, translation, and occlusion, to obtain a second regular sample based on the first regular sample. The generated second regular sample can ensure that the model recognizes subtle feature changes in the rule-loading scenario and its impact on the margin. The setting of slight difference enhancement makes the positive sample pairs no longer exactly the same samples, but sample pairs with slight changes. This enables the subsequent model to better capture small changes in the rule-loading scenario in the process of learning the positive sample pairs, which helps to enhance the adaptability to the rule-loading scenario.
[0087] In this embodiment, by performing slight difference enhancement on the first regular sample, voxel data of different perspectives or small transformations are generated from the same loading scene of the pallet, thereby generating data based on the original data (first regular sample) and slightly transformed data (second regular sample), thereby forming a similar positive sample pair.
[0088] S1-2. Perform several minor difference enhancements on the first regular samples to generate several second regular samples.
[0089] S1-3. Pair several second regular samples with the first regular samples respectively to construct several positive sample pairs.
[0090] Specifically, the positive sample pairs enable the model to learn the direct relationship between cargo space status and margin in regular loading scenarios.
[0091] S1-4. Obtain several irregular samples of irregular loading scenarios on the container.
[0092] S1-5. Pair several irregular samples with the first regular sample respectively to construct several negative sample pairs.
[0093] Specifically, the data of irregular samples are closer to the actual loading data on the pallet. For example, in the actual pallet loading operation, due to the different shapes, sizes, and quantities of goods, the goods may be unevenly placed, with large gaps and irregular stacking. Compared with regular loading scenarios, the loading density, position distribution, and gap distribution of goods on the pallet in irregular loading scenarios are relatively random, resulting in significant spatial differences in loading states. Therefore, it is very difficult for negative samples to obtain margin labels. In this embodiment, the irregular sample without margin label and the first regular sample are directly paired as negative sample pairs. By comparing with the positive sample pairs, the implicit margin inference capability of the model for irregular loading scenarios is established.
[0094] S2. Initialize the three-dimensional convolutional model to obtain the contrastive learning model.
[0095] The S2 specifically includes:
[0096] S2-1. Load the pre-trained general 3D convolutional model.
[0097] S2-2. Delete the fully connected layer and classification layer of the general 3D convolutional model, and keep the convolutional layer.
[0098] Specifically, by deleting the fully connected layers and classification layers of the general three-dimensional convolutional model, only the convolutional layers used to extract spatial features in the model are retained. This makes the features output by the model no longer labels for specific classification tasks, but retains the embedded representation of the spatial state of the cargo, which is convenient for similarity calculation.
[0099] S2-3. Add a contrastive learning head between the convolutional layer and the output layer to generate an embedding vector space.
[0100] The embedding vector space is a process where the high-dimensional features extracted by the convolutional layer are further processed by the contrastive learning head to map them into low-dimensional features. Low-dimensional features can effectively calculate the similarity between samples in contrastive learning to help the model distinguish the characteristics of regular and irregular loading scenarios.
[0101] Specifically, the role of the contrastive learning head is to compress and normalize high-dimensional features in order to generate an embedding vector in the embedding space that reflects features such as loading density and spatial distribution. The embedding vector will be used to represent the spatial state of irregular loading scenarios in the application phase. After training, the embedding space representation of the model maps negative sample pairs (i.e., irregular loading scenarios) into an implicit representation of the margin, which is distinguished from the margin space state of regular loading scenarios. In the actual prediction process, the embedding vector of the irregular sample can be regarded as an implicit representation of the margin, which can be converted into the margin prediction value of the irregular loading scenario through the feature relationship with the regular margin sample.
[0102] S2-4, defining a contrastive learning loss function in the embedding vector space, by maximizing the similarity of positive samples and minimizing the similarity of negative samples, so that the model can form a clear feature distinction between regular loading scenarios and irregular loading scenarios, providing for subsequent irregular margin inference.
[0103] Specifically, the feature similarity can be calculated using cosine similarity.
[0104] S2-5. Use the three-dimensional convolutional model having the contrastive learning head and the contrastive learning loss function as the initialized contrastive learning model.
[0105] S3. Inputting the plurality of positive sample pairs and negative sample pairs into a contrastive learning model for iterative training to obtain the residual prediction model.
[0106] The S3 specifically includes:
[0107] S3-1. Receive the plurality of positive sample pairs and negative sample pairs as inputs of the contrastive learning model, and generate positive sample pair embedding vectors and negative sample pair embedding vectors through forward propagation of the convolutional layer and the contrastive learning head.
[0108] The expression for generating the embedding vector of the positive sample pair is: p =f(X p ),h q =f(X q ).
[0109] X p represents the first regular sample, X q represents the second regular sample, f represents the mapping function after the convolutional layer and the contrastive learning head, and h p represents the embedding vector of the first regular sample, h q Represents the embedding vector of the second regular sample.
[0110] The expression of the negative sample pair embedding vector is: p =f(X p ),h n =f(X n ).
[0111] X n represents irregular samples, h n Embedding vector representing irregular samples.
[0112] S3-2. In the embedding vector space, the cosine similarity is used to calculate the positive sample similarity between the embedding vectors of the positive sample pairs and the negative sample similarity between the embedding vectors of the negative sample pairs.
[0113] The expression of the positive sample similarity is:
[0114] s pq Represents the positive sample similarity, which represents the cosine similarity between the embedding vectors of the positive sample pair.
[0115] The expression of the negative sample similarity is:
[0116] s nq It represents the negative sample similarity, which represents the cosine similarity between the embedding vectors of negative sample pairs.
[0117] Specifically, the positive sample similarity can be regarded as the relationship between the cargo space state and the margin space state in the regular loading scenario, while the negative sample similarity represents the characteristic difference between the cargo space state in the irregular loading scenario and the regular loading scenario. This characteristic difference can be regarded as the offset of the cargo space state in the irregular loading scenario relative to the margin space state in the regular loading scenario, thereby indirectly expressing the margin state in the irregular loading scenario.
[0118] This offset of the model reflects the model’s understanding of the characteristics of irregular loading scenarios in the embedding space, allowing the model to generate a “relative margin change” based on the margin of regular loading scenarios.
[0119] S3-3, according to the contrastive learning loss function, the similarity of the positive sample pairs is maximized while the similarity of the negative sample pairs is minimized, so that the model generates an embedded feature vector of the irregular loading scene, which can indirectly reflect the implicit margin characteristics of the scene.
[0120] The contrastive learning loss function is:
[0121] represents the contrast loss, N represents the total number of positive and negative sample pairs involved in the loss calculation, represents the similarity of the i-th positive sample pair, represents the similarity of the i-th negative sample pair, and m represents the boundary value of the contrast loss, which is used to control the minimization of the similarity of the negative sample pair. That is, negative sample pairs with a similarity less than m will not incur losses. Represents the minimization part of the similarity of the negative sample pair. That is, only when the similarity of the negative sample pair There is a loss only when it is less than m. It represents the maximized part of the similarity of the positive sample pair, that is, by reducing the gap between the similarity of the positive sample pair and 1, the similarity of the positive sample pair is made as close to 1 as possible.
[0122] The contrastive learning loss aims to maximize the similarity of positive sample pairs. At the same time, minimize the similarity of negative sample pairs By controlling the boundary value m, the upper limit of the similarity of negative sample pairs can be limited. When it is less than m, it will contribute to the loss, and when the similarity of the positive sample pair When it is close to 1, the loss approaches 0, thus achieving the distinction between positive and negative sample pairs.
[0123] Furthermore, before maximizing the similarity of the positive sample pairs and minimizing the similarity of the negative sample pairs, the method further includes:
[0124] Compare the positive sample similarity with the preset positive sample similarity threshold.
[0125] When the positive sample similarity reaches or exceeds the preset positive sample similarity threshold, the learning rate of the negative sample pair is dynamically adjusted. The expression for dynamically adjusting the learning rate of the negative sample pair is: neg =η base +k·(s pq -τ) p .
[0126] η neg Denotes the learning rate of negative samples after dynamic adjustment, η base represents the initial learning rate of the negative sample pair, k represents the adjustment coefficient, which is used to control the learning rate increase, and s pq Represents the similarity of the positive sample pair, τ represents the positive sample similarity threshold, and p represents the exponential parameter, which is used to control the amplitude of nonlinear adjustment.
[0127] Specifically, the purpose of dynamically adjusting the negative sample pairs is to make the model more focused on optimizing the features of irregular loading scenarios, improve the ability to distinguish irregular loading scenarios, and ensure that the model retains its uniqueness in the feature embedding of irregular loading scenarios, thereby establishing the ability to infer the implicit margin space state. After determining the learning rate between the negative sample pairs, use this learning rate to perform back propagation to accelerate the minimization of the similarity of the negative sample pairs.
[0128] Back propagation is performed using the adjusted learning rate to update the model parameters of the contrastive learning model until the model reaches a convergence condition, and the contrastive learning model of the current model parameters is derived and defined as the residual prediction model.
[0129] The expression of the margin prediction model is: M = f embed (X,θ * ),X=(X p , X q , X n ).
[0130] M represents the margin prediction model, f embed represents the training step, and X represents the input sample in the training step.
[0131] Specifically, the convergence condition of the model refers to the rate of change of the contrast loss function being less than a set threshold, indicating that the model has reached stability in distinguishing the features of positive and negative sample pairs. The rate of change of the contrast loss function refers to the degree of change of the loss value of the contrast loss function between consecutive iterations during the training process. Specifically, it is the reduction or change in the loss value of the contrast loss function in each round of training iteration. If this rate of change is lower than a set threshold, it can be considered that the training of the model is gradually stabilizing, indicating that the model is close to convergence and no longer requires large-scale parameter updates.
[0132] The convergence condition of the model can also be that the similarity of the positive and negative sample pairs reaches a preset threshold, indicating that the model's ability to distinguish between positive and negative samples in the embedding vector space meets expectations.
[0133] The training method of the margin prediction model in this embodiment can learn the significant feature differences between regular loading scenarios and irregular loading scenarios through comparative learning of positive sample pairs and negative sample pairs, so that the model can implicitly infer the margin space state of irregular loading scenarios even without clear irregular scene margin labels. In addition, the contrastive learning loss function in its model maximizes the similarity of positive sample pairs and minimizes the similarity of negative sample pairs, so that the model can form a clear distinction between regular and irregular loading features. The model can identify and quantify the loading features of irregular scenarios in the application stage, thereby more accurately predicting the margin.
[0134] Furthermore, by dynamically adjusting the learning rate of negative sample pairs, the model can focus on feature optimization of irregular loading scenarios, ensuring that the model retains its uniqueness when distinguishing irregular scenarios, improving the model's adaptability to irregular scenarios, and further enhancing the accuracy of margin prediction.
[0135] The embodiment of the present invention also provides a loading margin prediction system for aviation pallet transportation, which is used to implement the above method embodiment, which has been described and will not be repeated. The terms "module", "unit", "subunit", etc. used below can implement a combination of software and / or hardware for a predetermined function. Although the system described in the following embodiments is preferably implemented in software, it is also possible and conceivable to implement hardware, or a combination of software and hardware.
[0136] like Figure 3 As shown, Figure 3 The present invention is a structural block diagram of a system for predicting the loading margin of an aviation pallet transport, the system comprising:
[0137] A current loading status acquisition module, used to acquire the current cargo space status on the container board;
[0138] An implicit surplus feature embedding module is used to receive the current cargo space state as an input of a preset surplus prediction model and generate an implicit surplus feature of the current loading scenario in an embedding space;
[0139] The loading margin prediction module is used to use the implicit margin feature of the current loading scenario as the input of the final output layer, and obtain the predicted value of the loading margin through the trained model parameters of the margin prediction model.
[0140] In the above system, the current cargo space status on the container is obtained through the current loading status acquisition module; the implicit margin feature of the current loading scenario is generated through the implicit margin feature embedding module; and the predicted value of the loading margin is obtained through the loading margin prediction module, thereby solving the problem of difficulty in realizing margin prediction in the container loading scenario.
[0141] like Figure 4 As shown, an embodiment of the present invention further provides an electronic device, the electronic device comprising a memory 230 and a processor 210, the memory 230 stores at least one computer executable instruction, the processor 210 is configured to run the computer executable instruction, and when the computer executable instruction is run by the processor 210, the loading margin prediction method for aviation pallet transportation is implemented.
[0142] The electronic device may include a processor 210, a communications interface 220, a memory 230, and a communication bus 240, wherein the processor 210, the communications interface 220, and the memory 230 communicate with each other via the communication bus 240. The processor 210 may call the logic instructions in the memory 230 to execute the loading margin prediction method for air pallet transportation disclosed in this embodiment.
[0143] In addition, the logic instructions in the above-mentioned memory 230 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0144] The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.
[0145] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting the loading margin of air pallet transport, characterized in that: The steps include: Obtain the current cargo space status on the container; Receiving the current cargo space state as input of a preset surplus prediction model, and generating implicit surplus features of the current loading scenario in the embedding space; Taking the implicit margin feature of the current loading scenario as the input of the final output layer, and obtaining the predicted value of the loading margin through the trained model parameters of the margin prediction model; The output is the predicted value of the loading margin, which is expressed as: Y = f embed (X c θ * ) Y represents the predicted value of the loading margin, X c represents the current cargo space state, θ * Represents the trained model parameters of the residual prediction model.
2. The method for predicting the loading margin of air pallet transport according to claim 1, characterized in that: The method for constructing the surplus prediction model includes: S1. Obtain several positive sample pairs and negative sample pairs for residual prediction model training; S2, initialize the three-dimensional convolution model to obtain the contrastive learning model; S3. Inputting the plurality of positive sample pairs and negative sample pairs into a contrastive learning model for iterative training to obtain the residual prediction model.
3. The method for predicting the loading margin of air pallet transport according to claim 2, characterized in that: The obtaining of a plurality of positive sample pairs and negative sample pairs for residual prediction model training includes: S1-1, obtaining a first rule sample of a regular loading scenario on a container board; S1-2, performing several minor difference enhancements on the first regular samples to generate several second regular samples; S1-3, pairing a number of second regular samples with the first regular samples respectively to construct a number of positive sample pairs; S1-4, obtaining several irregular samples of irregular loading scenes on the container board; S1-5. Pair several irregular samples with the first regular sample respectively to construct several negative sample pairs.
4. The method for predicting the loading margin of air pallet transport according to claim 2, characterized in that: The step of obtaining a first rule sample of a regular loading scenario on a container plate comprises: S1-1-1, obtaining three-dimensional point cloud data of goods loaded on a number of pallets; S1-1-2, voxelizing the three-dimensional point cloud data into a three-dimensional voxel grid of standard size, and assigning values to the three-dimensional voxel grid based on the cargo occupancy status in the three-dimensional point cloud data; wherein an occupied voxel grid is assigned a value of 1, and an idle voxel grid is assigned a value of 0; S1-1-3. Standardize and normalize the assigned three-dimensional grid data to a uniform size to obtain a first regular sample.
5. The method for predicting the loading margin of air pallet transport according to claim 2, characterized in that: The initialization of the three-dimensional convolution model includes: S2-1, load the pre-trained general 3D convolutional model; S2-2, delete the fully connected layer and classification layer of the general 3D convolutional model, and keep the convolutional layer; S2-3, add a contrastive learning head between the convolutional layer and the output layer to generate an embedding vector space; S2-4, defining a contrastive learning loss function in the embedding vector space by maximizing the similarity of positive samples and minimizing the similarity of negative samples; S2-5. Use the three-dimensional convolutional model having the contrastive learning head and the contrastive learning loss function as the initialized contrastive learning model.
6. The method for predicting the loading margin of air pallet transport according to claim 2, characterized in that: The iterative training of the margin prediction model comprises: S3-1, receiving the plurality of positive sample pairs and negative sample pairs as inputs of the contrastive learning model, and generating positive sample pair embedding vectors and negative sample pair embedding vectors through forward propagation of the convolutional layer and the contrastive learning head; The expression for generating the embedding vector of the positive sample pair is: p =f(X p ),h q =f(X q ); X p represents the first regular sample, X q represents the second regular sample, f represents the mapping function after the convolutional layer and the contrastive learning head, and h p represents the embedding vector of the first regular sample, h q The embedding vector representing the second regular sample; The expression of the negative sample pair embedding vector is: p =f(X p ),h n =f(X n ); X n represents irregular samples, h n Embedding vector representing irregular samples; S3-2. In the embedding vector space, the cosine similarity is used to calculate the positive sample similarity between the positive sample pairs and the negative sample similarity between the negative sample pairs; The expression of the positive sample similarity is: s pq Represents the positive sample similarity, which represents the cosine similarity between the embedding vectors of the positive sample pair; The expression of the negative sample similarity is: s nq Represents negative sample similarity, which represents the cosine similarity between the embedding vectors of negative sample pairs; S3-3, according to the contrastive learning loss function, maximize the similarity of positive sample pairs while minimizing the similarity of negative sample pairs; The contrastive learning loss function is: represents the contrast loss, N represents the total number of positive and negative sample pairs involved in the loss calculation, represents the similarity of the i-th positive sample pair, represents the similarity of the i-th negative sample pair, and m represents the boundary value of the contrast loss, which is used to control the minimization of the similarity of the negative sample pair; Represents the minimization part of the similarity of negative samples; Represents the part that maximizes the similarity of the positive sample pair.
7. The method for predicting the loading margin of air pallet transport according to claim 2, characterized in that: Before maximizing the similarity of the positive sample pairs and minimizing the similarity of the negative sample pairs, the method further includes: Compare the positive sample similarity with the preset positive sample similarity threshold; When the positive sample similarity reaches or exceeds the preset positive sample similarity threshold, the learning rate of the negative sample pair is dynamically adjusted; the expression for dynamically adjusting the learning rate of the negative sample pair is: η neg =η base +k·(s pq -τ) p ; η neg Denotes the learning rate of negative samples after dynamic adjustment, η base represents the initial learning rate of the negative sample pair, k represents the adjustment coefficient, which is used to control the learning rate increase, and s pq represents the similarity of the positive sample pair, τ represents the positive sample similarity threshold, and p represents the exponential parameter, which is used to control the amplitude of nonlinear adjustment; Performing back propagation using the adjusted learning rate to update the model parameters of the contrastive learning model until the model reaches a convergence condition, deriving a contrastive learning model of current model parameters, which is defined as the margin prediction model; The expression of the margin prediction model is: M = f embed (X,θ * ),X=(X p , X q , X n ); M represents the margin prediction model, f embed represents the training step, and X represents the input sample in the training step.
8. A system for predicting the amount of cargo carried by air pallets, characterized in that: include: A current loading status acquisition module, used to acquire the current cargo space status on the container board; An implicit surplus feature embedding module is used to receive the current cargo space state as an input of a preset surplus prediction model and generate an implicit surplus feature of the current loading scenario in an embedding space; The loading margin prediction module is used to use the implicit margin feature of the current loading scenario as the input of the final output layer, and obtain the predicted value of the loading margin through the trained model parameters of the margin prediction model.
9. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores at least one computer executable instruction, and the processor is configured to execute the computer executable instruction. When the computer executable instruction is executed by the processor, the method for predicting the loading margin carried by an aviation pallet according to any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the method for predicting the loading margin of air pallet transportation according to any one of claims 1 to 7 is implemented.