A warehouse logistics system resource planning method and system based on deep learning

Abstract

This invention belongs to the field of resource planning and scheduling, and provides a resource planning method and system for warehousing and logistics systems based on deep learning. The method includes: acquiring full-process operational data of the warehousing system under the resource planning scenario; standardizing the full-process operational data from the perspectives of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor; and performing resource planning using a pre-trained spatiotemporal attention resource planning network model based on the multi-channel three-dimensional input feature tensor to obtain a resource allocation scheme. This invention, by introducing deep learning technology, achieves a paradigm shift from experience-driven to data-driven approaches, enabling adaptive planning and allocation of various resource inputs in the warehousing and logistics system, responding to dynamic changes in system design and operational phases, and improving computational and decision-making efficiency.

Description

Technical Field

[0001] This invention belongs to the field of resource planning and scheduling technology, specifically relating to a resource planning method and system for warehousing and logistics systems based on deep learning. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Warehousing and logistics systems are a crucial link in modern logistics systems. These systems process goods according to specific operational procedures, based on required tasks and requirements, to ensure the smooth flow and timely delivery of goods. As the core hub of modern logistics, the operational efficiency of the warehousing and logistics system directly impacts the responsiveness and cost control of the entire supply chain. This system integrates various resources such as manpower, equipment, and space, handling various tasks according to pre-set operational processes to ensure efficient circulation and timely delivery of goods within the warehouse. Therefore, scientifically and rationally planning the allocation of these resources is key to improving the efficiency of the warehousing and logistics system and reducing operating costs, significantly impacting its operational efficiency.

[0004] However, in current scheme planning and operation practices, the planning of resource input quantities largely relies on the personal experience and rough rules of engineering designers or operations managers. This method is highly subjective, difficult to quantify, and heavily dependent on experience accumulated in specific scenarios, resulting in generally poor adaptability. When the system design scheme changes, the original experience rules may immediately become invalid; at the same time, the dynamic situations faced by the system in actual operation, such as fluctuations in workload, changes in order structure, and changes in response time requirements, also make this method based on fixed experience difficult to effectively cope with, thus affecting the stable and efficient operation of the system.

[0005] To overcome the drawbacks of relying on experience, the industry has mainly turned to the following two types of technical solutions:

[0006] The first type of approach is based on operations research optimization models and algorithms. This method abstracts the resource planning problem into a mathematical model and seeks the optimal solution by solving the objective function under specific constraints, making it more scientifically rigorous in theory. However, it has significant limitations in practical applications: First, in order to construct a solvable mathematical model, a large number of simplifications and assumptions must be made regarding the complex actual warehousing scenario, making it difficult for the model to accurately reflect the full picture of the real system; second, as the problem scale increases, the computational complexity of the model grows exponentially, and the solution efficiency drops sharply, making it difficult to meet the needs of real-time and dynamic decision-making.

[0007] The second type of approach is based on simulation methods. System simulation can effectively handle the complex dynamic characteristics and randomness of warehousing and logistics systems, evaluating performance under different resource allocations by simulating system operation. However, simulation methods are essentially "analysis" tools rather than "design" tools; they cannot automatically generate optimized decision-making solutions and still require manual adjustments based on simulation results. Furthermore, the construction of simulation models is extremely complex, highly dependent on professional simulation engineers and specific simulation tools. Modifying and adjusting the models involves a huge workload, and its adaptability and agility are poor when facing frequent system changes.

[0008] Furthermore, some existing technical solutions focus on using simulation models to describe the system and assist in optimization, but their core technology remains at the simulation analysis level. Other solutions emphasize using deep learning technology to solve inventory optimization or prediction problems of specific parameters, but their technical goals and application scenarios do not directly address the core need of planning the input of multi-dimensional resources within the warehousing system.

[0009] In conclusion, whether relying on subjective experience, operations research optimization, or simulation analysis, existing technologies struggle to provide a resource planning solution that can simultaneously achieve high accuracy, high efficiency, and strong adaptability. Summary of the Invention

[0010] To address the aforementioned issues, this invention proposes a resource planning method and system for warehousing and logistics systems based on deep learning. This invention enables adaptive planning and allocation of various resource inputs within the warehousing and logistics system, responding to dynamic changes in system design and operational phases, and improving computational and decision-making efficiency.

[0011] According to some embodiments, the first solution of the present invention provides a resource planning method for a warehousing and logistics system based on deep learning, which adopts the following technical solution:

[0012] A deep learning-based resource planning method for warehousing and logistics systems includes:

[0013] Acquire full-process operational data of warehousing in the context of resource planning;

[0014] Based on the operational data of the entire warehousing process, standardization processing is performed from the dimensions of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor;

[0015] Based on multi-channel three-dimensional input feature tensors, a pre-trained spatiotemporal attention resource planning network model is used to perform resource planning and obtain a resource allocation scheme.

[0016] Furthermore, the warehousing operation data includes the actual amount of resources invested, the entire system operation process, the average processing time of each process in different time periods, the number of tasks and their processing flow in different time periods, and the target time for the system to complete all tasks.

[0017] Furthermore, based on the operational data of the entire warehousing process, standardization processing is performed from the dimensions of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor, including:

[0018] Based on a complete operational cycle, it is divided into fixed time intervals. A continuous time segment is used to construct the time dimension of the entire warehouse operation data;

[0019] Arrange each step of the entire warehousing process in sequence to form the warehousing and logistics process dimension;

[0020] Describing the system state under time and process from different perspectives, as a feature channel dimension;

[0021] We construct an initial multi-channel three-dimensional input feature tensor using the time dimension as the row dimension and the warehousing and logistics process dimension as the column dimension, and the feature channel dimension as the depth dimension.

[0022] Each feature channel of the initial multi-channel 3D input feature tensor is subjected to data standardization to obtain the multi-channel 3D input feature tensor.

[0023] Furthermore, the feature channels include at least three types: task quantity channel, average process processing time channel, and the number of times the process uses a certain resource.

[0024] Furthermore, the elements of the task quantity channel Used to indicate in the first Arrival at the warehousing and logistics system within a specific time frame requires the execution of a process. The number of tasks;

[0025] The element of the average processing time channel of the process Used to indicate in the first Within each time segment, each task follows the execution process. Average processing time per hour;

[0026] The process refers to the element of the usage quantity channel for a certain resource. Used to indicate in the first Within the time segment, used to process the first Resources deployed in each process quantity , It is a feature channel.

[0027] Furthermore, a pre-trained spatiotemporal attention resource planning network model is used for resource planning to obtain a resource allocation scheme, including:

[0028] High-dimensional mapping and feature dimensionality reduction are performed on the multi-channel 3D input feature tensor to transform it from the original feature space to the implicit feature space, thus obtaining the 3D embedded feature tensor.

[0029] The three-dimensional embedded feature tensor is processed to generate a time subtensor. The time subtensor is then used to mine the dynamic pattern of the state change over time within a single process to obtain temporal features.

[0030] The process sub-tensor is generated by processing the 3D embedded feature tensor. The process sub-tensor is then used to mine the correlation and dependency between different processes at the same point in time to obtain process features.

[0031] The fused features are obtained by fusing time-series features and process features;

[0032] The fusion characteristics are reduced in dimensionality and transformed nonlinearly to obtain the suggested input amount for each resource, which serves as a resource planning and allocation scheme.

[0033] Furthermore, the 3D embedded feature tensor is processed to generate a temporal subtensor, including:

[0034] In the 3D embedded feature tensor of the warehousing and logistics process dimension Slice the data into segments and process each step. Extract all its time segments and all feature channels The data above forms a time subtensor .

[0035] Furthermore, the process of generating sub-tensors from the 3D embedded feature tensor includes:

[0036] In the time dimension of the 3D embedded feature tensor Slice the data into segments, dividing each time period into segments. Extract all its processes and all feature channels The data above forms a process sub-tensor. .

[0037] According to some embodiments, the second aspect of the present invention provides a resource planning system for warehousing and logistics systems based on deep learning, employing the following technical solution:

[0038] A deep learning-based resource planning system for warehousing and logistics systems includes:

[0039] The data acquisition module is configured to acquire full-process operational data of the warehouse under the resource planning scenario;

[0040] The multi-channel 3D input feature tensor construction module is configured to standardize data based on the entire warehousing process operation data from the time dimension, warehousing and logistics process dimension, and feature channel dimension to construct a multi-channel 3D input feature tensor.

[0041] The resource planning module is configured to perform resource planning based on a multi-channel 3D input feature tensor and a pre-trained spatiotemporal attention resource planning network model to obtain a resource planning allocation scheme.

[0042] According to some embodiments, a third aspect of the present invention provides a computer device.

[0043] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the deep learning-based resource planning method for a warehousing and logistics system as described in the first embodiment above.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] This invention effectively improves the objectivity and accuracy of resource planning. By training with historical system operation data and simulation data, the deep learning model can deeply explore the complex nonlinear relationships between task volume, processes, and resource consumption, thereby achieving the quantification and prediction of resource input. This process minimizes the reliance on human experience in traditional methods, ensuring that planning decisions are based on objective data patterns, significantly improving the scientific rigor and accuracy of the results. Secondly, it possesses excellent dynamic adaptive and iterative optimization capabilities. The model can continuously learn new data as the system operates, achieving iterative optimization. Whether it's adjusting the initial planning and design scheme or facing dynamic scenarios such as order fluctuations and process changes in actual operation, this method can quickly adapt and provide matching resource planning solutions, overcoming the rigidity of traditional operations research optimization models and the difficulty in adjusting simulation models, ensuring the timeliness and applicability of the strategy. Finally, it greatly improves computational and decision-making efficiency and directly empowers cost optimization. Based on a maturely trained neural network model for inference and prediction, it can quickly output planning results within seconds or even milliseconds, with computational efficiency far exceeding that of solving complex optimization models or running multiple simulation experiments. This efficiency enables managers to respond quickly to system changes, thereby accurately allocating resources such as manpower and equipment, effectively avoiding resource redundancy or insufficiency caused by planning delays or inaccuracies, and reducing system operating costs from the core aspects. Attached Figure Description

[0046] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0047] Figure 1 This is the training process of a deep learning-based resource planning method for warehousing and logistics systems in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the multi-channel three-dimensional input feature tensor in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the resource requirement output target vector in an embodiment of the present invention;

[0050] Figure 4 This is a diagram of the spatiotemporal attention resource planning network model architecture in an embodiment of the present invention. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0052] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0053] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0054] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0055] Example 1

[0056] This embodiment provides a resource planning method for a warehousing and logistics system based on deep learning. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to terminals, and can also be applied to systems including terminals, servers, and other components, and implemented through interaction between the terminal and the server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. In this embodiment, the method includes the following steps:

[0057] Acquire full-process operational data of warehousing in the context of resource planning;

[0058] Based on the operational data of the entire warehousing process, standardization processing is performed from the dimensions of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor;

[0059] Based on multi-channel three-dimensional input feature tensors, a pre-trained spatiotemporal attention resource planning network model is used to perform resource planning and obtain a resource allocation scheme.

[0060] Based on the operational data of the entire warehousing process, standardization processing is performed from the dimensions of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor, including:

[0061] Based on a complete operational cycle, it is divided into fixed time intervals. A continuous time segment is used to construct the time dimension of the entire warehouse operation data;

[0062] Arrange each step of the entire warehousing process in sequence to form the warehousing and logistics process dimension;

[0063] Describing the system state under time and process from different perspectives, as a feature channel dimension;

[0064] We construct an initial multi-channel three-dimensional input feature tensor using the time dimension as the row dimension and the warehousing and logistics process dimension as the column dimension, and the feature channel dimension as the depth dimension.

[0065] Each feature channel of the initial multi-channel 3D input feature tensor is subjected to data standardization to obtain the multi-channel 3D input feature tensor.

[0066] Resource planning is performed using a pre-trained spatiotemporal attention resource planning network model to obtain resource allocation schemes, including:

[0067] High-dimensional mapping and feature dimensionality reduction are performed on the multi-channel 3D input feature tensor to transform it from the original feature space to the implicit feature space, thus obtaining the 3D embedded feature tensor.

[0068] The three-dimensional embedded feature tensor is processed to generate a time subtensor. The time subtensor is then used to mine the dynamic pattern of the state change over time within a single process to obtain temporal features.

[0069] The process sub-tensor is generated by processing the 3D embedded feature tensor. The process sub-tensor is then used to mine the correlation and dependency between different processes at the same point in time to obtain process features.

[0070] The fused features are obtained by fusing time-series features and process features;

[0071] The fusion characteristics are reduced in dimensionality and transformed nonlinearly to obtain the suggested input amount for each resource, which serves as a resource planning and allocation scheme.

[0072] The process of generating a temporal subtensor from a 3D embedded feature tensor includes:

[0073] In the 3D embedded feature tensor of the warehousing and logistics process dimension Slice the data into segments and process each step. Extract all its time segments and all feature channels The data above forms a time subtensor .

[0074] The process of generating sub-tensors from 3D embedded feature tensors includes:

[0075] In the time dimension of the 3D embedded feature tensor Slice the data into segments, dividing each time period into segments. Extract all its processes and all feature channels The data above forms a process sub-tensor. .

[0076] This embodiment provides a resource planning method for warehousing and logistics systems based on deep learning, and the training process of such a method is as follows: Figure 1 As shown, it includes:

[0077] Step (1): Data collection and integration of the warehousing and logistics system;

[0078] Step (2): Constructing the multi-channel 3D input feature tensor and output target vector;

[0079] Step (3): Construction and training of the spatiotemporal attention resource planning network model;

[0080] Step (4): Use the trained model to perform resource planning for the warehousing and logistics system;

[0081] Step (5): Continuously iterate and optimize the model.

[0082] In a specific embodiment, step (1) describes the data collection and integration process of the warehousing and logistics system, as follows:

[0083] Collect real operational data for the entire warehousing process for resource planning of the warehousing and logistics system, including: the actual amount of resources invested, the entire system operation process, the average processing time of each process in different time periods, the number of tasks and their processing flow in different time periods, and the target time for the system to complete all tasks.

[0084] Since training neural networks requires a large amount of data, and real-world operational data may suffer from insufficient sample size or incomplete scenario coverage, a process-based simulation method is needed to quickly generate supplementary data. This involves constructing a process-based simulation model of the warehousing and logistics system to simulate various operational scenarios (such as process and resource changes), outputting simulation data in a format consistent with real data. It is understood that this process-based simulation model of the warehousing and logistics system can be implemented using existing technologies.

[0085] Finally, real-world warehouse operation data and simulation data are standardized and integrated to construct a comprehensive dataset for training a spatiotemporal attention resource planning network model, providing data support for the model to learn the correlation between task characteristics and resource requirements.

[0086] In a further embodiment, step (2) is the construction of the multi-channel three-dimensional input feature tensor and the output target vector, and the specific process is as follows:

[0087] A multi-channel three-dimensional input feature tensor is used to comprehensively characterize the state of a warehousing and logistics system within a specific historical period. The three dimensions of the multi-channel three-dimensional input feature tensor are defined as the time dimension. Warehousing and logistics process dimensions and feature channel dimension Therefore, the shape of the multi-channel three-dimensional input feature tensor can be represented as: .

[0088] The time dimension The row dimension that constitutes the multi-channel three-dimensional input feature tensor.

[0089] Time dimension Based on a complete operational cycle of the system (e.g., one day, from 08:00:00 to 20:00:00), at fixed time intervals (e.g., Hours) are divided into Each time segment is a consecutive time segment. Each time segment uses an index. express, It refers to the total time period; for example: It indicates 08:00:00 to 09:00:00. This indicates the time from 09:00:00 to 10:00:00, and so on.

[0090] The warehousing and logistics process dimensions The column dimensions that constitute the multi-channel three-dimensional input feature tensor.

[0091] Warehousing and logistics process dimensions This covers all core processes in the actual operation of the warehousing and logistics system. These processes are arranged sequentially to form a process set, with each process using an index. For example: This indicates the receiving and warehousing process. This indicates the picking process, etc.

[0092] The feature channel dimension The depth dimension that constitutes the multi-channel three-dimensional input feature tensor.

[0093] Each channel carries a specific type of feature information, describing it from different perspectives. The system state under a specified time and a specific process. The characteristic channels include at least three types: task quantity channel, average process processing time channel, and the number of times the process uses a certain resource.

[0094] The task quantity channel, the elements of this channel Used to indicate in the first Arrival at the warehousing and logistics system within a specific time frame requires the execution of a process. The number of tasks.

[0095] The average processing time channel of the process, the elements of this channel Used to indicate in the first Within each time segment, each task follows the execution process. The average processing time per hour. The channel data is derived from step (1);

[0096] The process uses a channel for the usage quantity of a certain resource, and the elements of that channel... Used to indicate in the first Within the time segment, used to process the first Resources deployed in each process quantity , It is a feature channel.

[0097] We construct an initial multi-channel three-dimensional input feature tensor using the time dimension as the row dimension and the warehousing and logistics process dimension as the column dimension, and the feature channel dimension as the depth dimension.

[0098] All indivual The two-dimensional matrices are stacked in the depth direction to form the initial multi-channel three-dimensional input feature tensor.

[0099] Each feature channel of the initial multi-channel 3D input feature tensor is standardized to obtain a multi-channel 3D input feature tensor. Before inputting the initial multi-channel 3D input feature tensor into the spatiotemporal attention resource planning network model, each feature channel needs to be standardized, for example, by using Z-score standardization or maximum-minimum normalization, to eliminate the adverse effects of different dimensions and orders of magnitude on model training.

[0100] Based on the process of multi-channel three-dimensional input feature tensor, the real data and simulation data in step (1) are used to construct a multi-channel three-dimensional input feature tensor. This ensures that the data format is compatible with the training requirements of the subsequent spatiotemporal attention resource planning network model. For example... Figure 2 As shown.

[0101] For the spatiotemporal attention resource planning network model, its output target vector is A one-dimensional vector, the vector Each dimension requires planning within the warehousing and logistics system. The quantity of each type of resource corresponds to the planned investment amount of that resource in the corresponding task scenario, directly reflecting the resource demand target. For example... Figure 3 As shown.

[0102] In a further embodiment, step (3), the construction and training of the spatiotemporal attention resource planning network model, is as follows:

[0103] In terms of time, the state of a warehousing and logistics system is a dynamic evolution process; in terms of process, it consists of multiple physically or logically interconnected process steps, which do not operate independently. Furthermore, the dependency between time and process dimensions is not isolated but highly cross-coupled. A change in the state of one process task at a specific point in time may affect the future state of one or more other processes. For example, A batch of goods may arrive at the "receiving area" at any time. This constantly leads to an increase in the resource usage of the "shelf storage area," and ultimately... This constantly affects the resource usage in the "picking area." This spatiotemporal delay transmission effect requires the model to have joint reasoning capabilities.

[0104] Therefore, the spatiotemporal attention resource planning network model is used to analyze the complex spatiotemporal dependencies in warehousing and logistics systems, thereby improving the accuracy of resource planning. This network consists of six parts: an input layer, a feature embedding layer, a temporal feature extraction branch, a process feature extraction branch, an attention fusion layer, and an output layer. Its structure is as follows: Figure 4 As shown.

[0105] The input layer of the spatiotemporal attention resource planning network model is used to receive the multi-channel three-dimensional input feature tensor constructed in step (2). This tensor is a comprehensive digital representation of the state of the warehousing and logistics system within a complete planning cycle.

[0106] The feature embedding layer of the spatiotemporal attention resource planning network model consists of two-dimensional convolutional layers with kernel sizes of 1×1 or 3×3. It is used to perform high-dimensional mapping and feature dimensionality reduction on the input tensor, transforming the original feature space into an implicit feature space to obtain a three-dimensional embedded feature tensor.

[0107] The temporal feature extraction branch of the spatiotemporal attention resource planning network model is used to mine the dynamic patterns of state changes over time within a single process. This branch consists of a set of parallel one-dimensional convolutional neural networks (1D-CNN).

[0108] Specifically, first, process dimension slicing is performed, and then the process dimension of the multi-channel three-dimensional input feature tensor is sliced. Slice the data into segments and process each step. Extract all its time segments and all feature channels The data above forms a time subtensor The data is then fed into a separate 1D-CNN module to mine the dynamic patterns of state changes over time within a single process, thus obtaining temporal features.

[0109] The process feature extraction branch of the spatiotemporal attention resource planning network model is used to discover the correlations and dependencies between different processes at the same point in time. This branch is also composed of a set of parallel one-dimensional convolutional neural network (1D-CNN) modules.

[0110] Unlike the temporal feature extraction branch, it first performs temporal slicing, and then inputs the feature tensor in the multi-channel three-dimensional time dimension. Slice the data into segments, dividing each time period into segments. Extract all its processes and all feature channels The data above forms a process sub-tensor. The data is then fed into a separate 1D-CNN module to mine the correlations and dependencies between different processes at the same time point, thereby obtaining process features.

[0111] The attention fusion layer of the spatiotemporal attention resource planning network model is used to receive and fuse the output features of the temporal and spatial branches, that is, to fuse temporal features and process features to obtain fused features, thereby improving the performance and interpretability of the spatiotemporal attention resource planning network model.

[0112] This layer introduces a multi-head attention mechanism to improve the model's ability to dynamically focus on key information.

[0113] This layer concatenates and transforms the fusion time-series features and process features of the two branches to generate the initial fusion features;

[0114] The initial fused features are processed by the multi-head attention module, and an attention weight matrix is ​​generated through query, key-value operations.

[0115] The initial fused features are weighted and summed using the attention weight matrix, and finally a refined weighted feature representation that incorporates global spatiotemporal context information is output, namely the fused feature.

[0116] The output layer of the spatiotemporal attention resource planning network model is used to perform dimensionality reduction and nonlinear transformation on the attention-weighted fused features, mapping them to the final prediction space. The output layer typically consists of one-dimensional or multi-dimensional fully connected layers. The output layer ultimately outputs the dimension described in step (2). The resource planning vector is a resource planning allocation scheme, where each element in the resource planning vector corresponds to a suggested input amount of a resource.

[0117] The training process of the spatiotemporal attention resource planning network model is as follows:

[0118] A sample set consisting of a multi-channel 3D input feature tensor constructed from historical data and the corresponding real resource vectors is divided into a training set, a validation set, and a test set according to a preset ratio (e.g., 7:2:1). An iteration cycle is set, and multiple iterations are performed based on the training set data. The weights of network neurons are dynamically adjusted through the backpropagation mechanism, enabling the model to learn the dynamic correlation between task features and resource requirements. During the training process, the loss value of the validation set is monitored in real time, and the network parameters corresponding to the minimum loss of the validation set are retained to ensure the model's generalization ability.

[0119] In a specific embodiment, step (4) involves using the trained model to perform resource planning for the warehousing and logistics system, as follows:

[0120] After deploying the trained spatiotemporal attention resource planning network model to the warehousing and logistics system, the entire warehousing process operation data under the resource planning scenario is collected. The warehousing process operation data is standardized according to the initial multi-channel three-dimensional input feature tensor format defined in step (2). The processed multi-channel three-dimensional input feature tensor is then input into the trained spatiotemporal attention resource planning network model. Based on the learned task features and resource demand correlation rules, the trained spatiotemporal attention resource planning network model outputs the planning results of various resource input quantities in the target vector format, which serve as a resource planning and allocation scheme to provide a decision-making basis for resource allocation.

[0121] In a specific embodiment, step (5) further describes the process of continuous iterative optimization of the model, including:

[0122] The spatiotemporal attention resource planning network model deployed in step (4) is incorporated into the operation of the warehousing and logistics system. Real warehousing operation data is collected from the system at preset intervals (e.g., daily / weekly) to construct an incremental dataset. The construction method and data format of the incremental dataset are the same as those defined in steps (1) and (2). The spatiotemporal attention resource planning network model is updated using the incremental dataset training method to avoid the resource consumption of full retraining. Through this optimization mechanism, it is ensured that the spatiotemporal attention resource planning network model can dynamically adapt to system task fluctuations, business process changes, etc., and continuously maintain the accuracy and timeliness of resource planning decisions.

[0123] Example 2

[0124] This embodiment provides a deep learning-based resource planning system for warehousing and logistics systems, including:

[0125] The data acquisition module is configured to acquire full-process operational data of the warehouse under the resource planning scenario;

[0126] The multi-channel 3D input feature tensor construction module is configured to standardize data based on the entire warehousing process operation data from the time dimension, warehousing and logistics process dimension, and feature channel dimension to construct a multi-channel 3D input feature tensor.

[0127] The resource planning module is configured to perform resource planning based on a multi-channel 3D input feature tensor and a pre-trained spatiotemporal attention resource planning network model to obtain a resource planning allocation scheme.

[0128] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0129] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0130] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0131] Example 3

[0132] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the resource planning method for a deep learning-based warehousing and logistics system as described in Embodiment 1 above.

[0133] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0134] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0138] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A resource planning method for warehousing and logistics systems based on deep learning, characterized in that, include: Acquire full-process operational data of warehousing in the context of resource planning; Based on the operational data of the entire warehousing process, standardization processing is performed from the dimensions of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor; The feature channels include at least three types: task quantity channel, average process processing time channel, and the number of resources used by the process for a certain resource. Based on multi-channel 3D input feature tensors, a pre-trained spatiotemporal attention resource planning network model is used for resource planning to obtain a resource allocation scheme, including: High-dimensional mapping and feature dimensionality reduction are performed on the multi-channel 3D input feature tensor to transform it from the original feature space to the implicit feature space, thus obtaining the 3D embedded feature tensor. The process involves processing the 3D embedded feature tensor to generate a temporal subtensor, and then mining the dynamic patterns of state changes over time within a single process to obtain temporal features. This process includes processing the 3D embedded feature tensor to generate the temporal subtensor along the warehousing and logistics process dimension of the 3D embedded feature tensor. Slice the data into segments and process each step. Extract all its time segments and all feature channels The data above forms a time subtensor ; The process involves processing the 3D embedded feature tensor to generate process sub-tensors, then mining the correlations and dependencies between different processes at the same time point to obtain process features. This process of generating process sub-tensors includes processing the 3D embedded feature tensor along its time dimension. Slice the data into segments, dividing each time period into segments. Extract all its processes and all feature channels The data above forms a process sub-tensor. ; The fused features are obtained by fusing time-series features and process features; The fusion characteristics are reduced in dimensionality and transformed nonlinearly to obtain the suggested input amount for each resource, which serves as a resource planning and allocation scheme.

2. The resource planning method for a warehousing and logistics system based on deep learning as described in claim 1, characterized in that, The full-process operation data of the warehousing includes the actual amount of resources invested, the entire system operation process, the average processing time of each process in different time periods, the number of tasks in different time periods and their processing flow, and the target time for the system to complete all tasks.

3. The resource planning method for a warehousing and logistics system based on deep learning as described in claim 1, characterized in that, Based on the operational data of the entire warehousing process, standardization processing is performed from the dimensions of time, warehousing and logistics process, and feature channels to construct a multi-channel three-dimensional input feature tensor, including: Based on a complete operational cycle, it is divided into fixed time intervals. A continuous time segment is used to construct the time dimension of the entire warehouse operation data; Arrange each step of the entire warehousing process in sequence to form the warehousing and logistics process dimension; Describing the system state under time and process from different perspectives, as a feature channel dimension; We construct an initial multi-channel three-dimensional input feature tensor using the time dimension as the row dimension and the warehousing and logistics process dimension as the column dimension, and the feature channel dimension as the depth dimension. Each feature channel of the initial multi-channel 3D input feature tensor is subjected to data standardization to obtain the multi-channel 3D input feature tensor.

4. The resource planning method for a warehousing and logistics system based on deep learning as described in claim 1, characterized in that, The elements of the task quantity channel Used to indicate in the first Arrival at the warehousing and logistics system within a specific time frame requires the execution of a process. The number of tasks; The element of the average processing time channel of the process Used to indicate in the first Within each time segment, each task follows the execution process. Average processing time per hour; The process refers to the element of the usage quantity channel for a certain resource. Used to indicate in the first Within the time segment, used to process the first Resources deployed in each process quantity , It is a feature channel.

5. A deep learning-based resource planning system for warehousing and logistics systems, employing the deep learning-based resource planning method for warehousing and logistics systems as described in any one of claims 1-4, characterized in that, include: The data acquisition module is configured to acquire full-process operational data of the warehouse under the resource planning scenario; The multi-channel 3D input feature tensor construction module is configured to standardize data based on the entire warehousing process operation data from the time dimension, warehousing and logistics process dimension, and feature channel dimension to construct a multi-channel 3D input feature tensor. The resource planning module is configured to perform resource planning based on a multi-channel 3D input feature tensor and a pre-trained spatiotemporal attention resource planning network model to obtain a resource planning allocation scheme.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the resource planning method for a deep learning-based warehousing and logistics system as described in any one of claims 1-4.

Images