Scheduling decision-making method and system based on comprehensive effectiveness evaluation of logistics equipment

Abstract

The invention relates to the related field of logistics data fusion, discloses a scheduling decision-making method and system based on comprehensive effectiveness evaluation of logistics equipment, and solves the problems of insufficient data fusion, low intelligent level, low scheduling efficiency and the like in existing logistics equipment management. An efficient, dynamic and intelligent solution is provided for comprehensive efficiency evaluation and scheduling decision-making of logistics equipment, deep interaction of vision, text and time series data is achieved through high-order tensor decomposition and recombination, fusion of structured data and unstructured data is supported, and different from a mode that a traditional logistics system depends on a single data source, the system has the advantages of being high in efficiency and high in reliability. And the limitation that different modal data are independently processed and the cross-modal interaction capability is lacked is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of logistics data fusion, and particularly to a scheduling decision method and system based on comprehensive efficiency evaluation of logistics equipment. BACKGROUND

[0002] Logistics management is a very important overall process optimization in the logistics field, and logistics equipment management is an important management process for ensuring the stability and efficiency of logistics operations. Traditional logistics equipment management is limited to homogeneous data and relies on a single data source (such as equipment operating parameters or static business data), lacks the ability to integrate multi-modal data (video, audio, text, sensor data), and is difficult to handle deep interaction of heterogeneous modalities. For example, real-time images captured by a camera and text data from equipment maintenance logs cannot be analyzed collaboratively, resulting in insufficient information utilization.

[0003] Existing evaluation methods only focus on single indicators such as equipment utilization or failure rate, lack multi-dimensional quantitative analysis of comprehensive efficiency (energy consumption, maintenance cost, environmental adaptability), and are often inadequate when faced with complex and changing scenarios, making it difficult to meet real-time and dynamic adjustment needs, resulting in low vehicle utilization, low transportation efficiency, and unnecessary increase in transportation costs. SUMMARY

[0004] The present application aims to provide a scheduling decision method and system based on comprehensive efficiency evaluation of logistics equipment to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] The scheduling decision method based on comprehensive efficiency evaluation of logistics equipment comprises:

[0007] Periodically traversing demand information through a terminal device group and a historical database to update and obtain multi-modal data, the demand information including operating parameters of logistics equipment, environmental information, and video image business data;

[0008] Data preprocessing is performed on the multi-modal data to obtain structured data and unstructured data with high signal-to-noise ratio, the data preprocessing including data cleaning, data normalization, and data feature extraction;

[0009] Based on multi-modal Tucker fusion, structured data and unstructured data are fused to establish deep interaction and feature expression of heterogeneous data;

[0010] Based on the fused data, an equipment efficiency evaluation model is constructed, and equipment index information is output, the equipment index information including equipment utilization, equipment failure rate, and equipment energy efficiency;

[0011] Synchronous real-time service requirements, and scheduling management according to device index information, generating an optimized scheduling scheme, outputting and executing the optimized scheduling scheme, and realizing intelligent scheduling of the logistics equipment.

[0012] As a further scheme of the present application: the step of data preprocessing on the multi-modal data specifically comprises:

[0013] The acquired multi-modal data is data cleaned, noise data and abnormal values are removed through a sliding window mean method to eliminate sensor signal jitter, missing values are filled, and the accuracy and integrity of the data are ensured;

[0014] Through eigenvalue scaling, logarithmic change and encoding processing, the data of different modalities are converted to consistent scales and formats;

[0015] Align the multi-source data acquisition frequency based on the equipment operation cycle, and ensure the synchronization of the data through spatial mapping or feature alignment;

[0016] Feature information related to equipment efficiency is extracted from each modal data, including statistical features, image texture features and text features, and a comprehensive feature vector is generated through data fusion for efficient data representation for subsequent analysis.

[0017] As a further scheme of the present application: the structured data is used to represent data with fixed format that can be directly stored by database or table, including operation data, business operation data, maintenance management data and environment detection data of logistics equipment;

[0018] The unstructured data is used to represent data without fixed format that cannot be directly stored by database, including text, image, video and audio data that need to be processed by natural language processing.

[0019] As a further scheme of the present application: the multi-modal Tucker fusion specifically comprises the steps of:

[0020] Based on the preprocessed data, a multi-modal feature tensor is constructed, and feature vectors of different modalities are combined into a high-order tensor;

[0021] The constructed multi-modal feature tensor is Tucker decomposed into a core tensor and several modal matrices, represented as:

[0022] T=G×1A×2B×3C, wherein T represents the multi-modal feature tensor, A, B and C are modal matrices, and G is the core tensor;

[0023] The decomposed modal matrices and core tensor are recombined to obtain the fused feature representation, represented as:

[0024] y=Gx1Ax2Bx3c, wherein y is the fused feature vector;

[0025] Using the fused feature vector y as input, training the downstream task model, optimizing the parameters of Tucker decomposition through back propagation to improve the fusion effect.

[0026] As a further scheme of the application: the step of combining the feature vectors of different modalities into a high-order tensor

[0027] In particular, it is characterized by:

[0028] T is in R Dv×Dt×Ds ,

[0029] Wherein, the Dv, Dt, Ds respectively represent the feature dimensions of visual, text, time sequence modal.

[0030] The embodiment of the application aims to provide a scheduling decision system based on comprehensive efficiency evaluation of logistics equipment, comprising:

[0031] The data synchronization module is used for periodic traversal of demand information through the terminal device group and the historical database to update and obtain multi-modal data, wherein the demand information includes operation parameters, environmental information and video image service data of the logistics equipment.

[0032] The data preprocessing module is used for data preprocessing of the multi-modal data to obtain high signal-to-noise ratio structured data and unstructured data, wherein the data preprocessing includes data cleaning, data normalization and data feature extraction.

[0033] The data fusion module is used for large model data fusion of structured data and unstructured data based on multi-modal Tucker fusion to establish deep interaction and feature expression of heterogeneous data.

[0034] The evaluation building module is used for building an equipment efficiency evaluation model based on the fused data and outputting equipment index information, wherein the equipment index information includes equipment utilization rate, equipment failure rate and equipment energy efficiency.

[0035] The scheduling management module is used for synchronizing real-time business demand and performing scheduling management according to the equipment index information to generate an optimized scheduling scheme, output and execute the optimized scheduling scheme, and realize intelligent scheduling of the logistics equipment.

[0036] As a further scheme of the application: the data preprocessing module comprises:

[0037] The data cleaning unit is used for data cleaning of the obtained multi-modal data, removing noise data and abnormal values by sliding window mean method to eliminate sensor signal jitter, filling missing values, and ensuring the accuracy and integrity of the data.

[0038] a data normalization unit, configured to convert data of different modalities to a consistent scale and format through eigenvalue scaling, logarithmic transformation and encoding processing;

[0039] a data synchronization unit, configured to align the data acquisition frequency of multiple sources based on the device operation cycle, and ensure the synchronization of the data through spatial mapping or feature alignment;

[0040] a feature fusion unit, configured to extract feature information related to the device performance from the data of different modalities, including statistical features, image texture features and text features, and generate a comprehensive feature vector through data fusion, to provide efficient data representation for subsequent analysis.

[0041] As a further scheme of the present application, the structured data is used to represent data with a fixed format that can be directly stored in a database or a table, including operation data of a logistics device, business operation data, maintenance management data and environmental detection data.

[0042] The unstructured data is used to represent data without a fixed format that cannot be directly stored in a database, including text, image, video and audio data that need to be processed through natural language processing.

[0043] As a further scheme of the present application, the data fusion module comprises:

[0044] a tensor construction unit, configured to construct a multi-modal feature tensor based on the preprocessed data, and combine feature vectors of different modalities into a high-order tensor;

[0045] a tensor decomposition unit, configured to perform Tucker decomposition on the constructed multi-modal feature tensor, to decompose the multi-modal feature tensor into a core tensor and a plurality of modal matrices, represented as:

[0046] T = G × 1A × 2B × 3C, wherein T represents the multi-modal feature tensor, A, B and C are the modal matrices, and G is the core tensor.

[0047] a fusion representation unit, configured to recombine the decomposed modal matrices and the core tensor, to obtain a fused feature representation, represented as:

[0048] y = G × 1A × 2B × 3c, wherein y is the fused feature vector.

[0049] a fusion training unit, configured to use the fused feature vector y as input to train a downstream task model, and to optimize the parameters of the Tucker decomposition through back propagation, to improve the fusion effect.

[0050] As a further scheme of the present application, in the tensor construction unit,

[0051] The way to combine the eigenvectors of different modes into a high-order tensor is represented as:

[0052] t∈R Dv×Dt×Ds ,

[0053] Among them, Dv, Dt, and Ds are used to represent the feature dimensions of visual, textual, and temporal modalities, respectively.

[0054] Compared with the existing technology, the beneficial effects of the present invention are: it solves the problems of insufficient data fusion, low intelligence level and low scheduling efficiency in the existing logistics equipment management, and provides an efficient, dynamic and intelligent solution for the comprehensive performance evaluation and scheduling decision-making of logistics equipment. Through high-order tensor decomposition and reorganization, it realizes deep interaction of visual, textual and time series data, and supports the fusion of structured data and unstructured data. It is different from the limitations of traditional logistics systems that rely on a single data source, and independently process different modal data, and lack cross-modal interaction capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the scheduling decision-making method based on comprehensive performance evaluation of logistics equipment.

[0056] Figure 2 This is a flow chart of the preprocessing steps in the scheduling decision-making method based on comprehensive performance evaluation of logistics equipment.

[0057] Figure 3 It is a general architecture of multimodal large models in scheduling decision-making methods based on comprehensive performance evaluation of logistics equipment.

[0058] Figure 4 This is a block diagram of the scheduling decision system based on comprehensive performance evaluation of logistics equipment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0061] like Figure 1 、 Figure 3 The scheduling decision-making method based on comprehensive efficiency evaluation of logistics equipment provided in one embodiment of the present invention includes the following steps:

[0062] S1, periodically traversing demand information through a terminal device group and a historical database to update and obtain multi-modal data, the demand information including operation parameters of a logistics device, environmental information, and video image service data;

[0063] S2, data preprocessing is performed on the multi-modal data to obtain structured data and unstructured data with high signal-to-noise ratio, the data preprocessing including data cleaning, data normalization, and data feature extraction;

[0064] S3, large model data fusion is performed on the structured data and the unstructured data based on multi-modal Tucker fusion to establish deep interaction and feature expression of heterogeneous data;

[0065] S4, based on the fused data, a device performance evaluation model is constructed, and device index information is output, the device index information including device utilization, device failure rate, and device energy efficiency;

[0066] S5, synchronizing real-time service demand, and performing scheduling management according to the device index information to generate an optimized scheduling scheme, outputting and executing the optimized scheduling scheme to realize intelligent scheduling of the logistics device.

[0067] In the embodiment, the scheduling decision method based on comprehensive performance evaluation of the logistics device solves the problems of insufficient data fusion, low intelligent level, and low scheduling efficiency in existing logistics device management, and provides an efficient, dynamic, and intelligent solution for comprehensive performance evaluation and scheduling decision of the logistics device. Through high-order tensor decomposition and reorganization, deep interaction of visual, text, and time series data is realized, and fusion of structured data (device operation parameters) and unstructured data (video, log text) is supported. Unlike traditional logistics systems that rely on a single data source and process different modal data independently, the method has the limitation of lacking cross-modal interaction capability.

[0068] Traditional methods rely on rule engines or shallow machine learning models (such as decision trees and SVM), which cannot dynamically adapt to complex scenarios (such as device sudden failure, environmental mutation, and order surge). Moreover, scheduling decisions are mostly static planning, lacking real-time feedback mechanisms, resulting in low device utilization, redundant paths, and high energy consumption. The parameter quantity of multi-modal large models (such as GPT-4 and CLIP) is large, making it difficult to deploy on edge devices with limited hardware resources, resulting in high inference delay and difficulty in meeting real-time scheduling requirements. Existing deployment solutions (such as LLM.int8 quantization) lack support for multi-modal models, and mixed precision calculation has high complexity, resulting in a decrease in inference speed. Traditional scheduling systems rely on static rules or offline optimization models, which cannot respond to environmental changes (such as device failure and order surge) in real time, resulting in high scheduling delay. Traditional evaluation methods focus on a single indicator, lacking comprehensive quantitative analysis of energy consumption, maintenance cost, and environmental adaptability.

[0069] Based on the above problems, the solution of the embodiment is to realize the intelligent upgrading of logistics equipment management through the technical chain of multi-source data collection, cross-modal fusion, performance modeling, and dynamic scheduling. The multi-modal Tucker fusion method is adopted to realize the deep interaction of visual, textual, and time series data through high-order tensor decomposition and reorganization. The fusion of structured data (device operation parameters) and unstructured data (video, log text) is supported, and the information utilization rate is improved from 65% in the prior art to 92%. Based on the GPT-4V architecture, a scheduling model is constructed, and combined with real-time data input, a preliminary scheduling scheme is generated, and the response time is improved to milliseconds. Cloud-edge collaboration is adopted, and complex computing tasks are processed by the cloud, and real-time tasks are executed on the edge, and the communication delay is significantly reduced. A multi-dimensional evaluation model is adopted to synchronously output device utilization rate, failure probability, energy efficiency, and other indicators. The core architecture of the embodiment is:

[0070] Multi-modal perception layer: integrates sensors, IoT devices, cameras, and other multi-source data, covering device operation parameters, environmental status, video images, and business information. The logistics equipment here includes but is not limited to: transportation equipment, warehousing equipment, packaging equipment, loading and unloading equipment, flow processing equipment, information collection and processing equipment;

[0071] Fusion analysis layer: based on multi-modal Tucker fusion technology, realizes deep interaction and feature expression of heterogeneous data;

[0072] Performance evaluation layer: constructs a multi-task machine learning model to output device utilization rate, failure rate, energy efficiency, and other composite indicators;

[0073] Dynamic scheduling layer: combines large models and optimization algorithms to generate real-time scheduling schemes, supporting edge-cloud collaborative execution.

[0074] In steps S4 and S5, the performance evaluation model is trained based on machine learning algorithms, and the scheduling decision adopts large model technology. A multi-modal input interface is constructed based on GPT-4V, supporting joint reasoning of images, texts, and data. After inputting performance evaluation results, real-time order demand, traffic condition images, and device status text reports, a preliminary scheduling scheme is output, and optimization algorithms are used for scheme optimization. Compared with the prior art, the advantages of the embodiment are:

[0075] Cross-modal interaction capability: Tucker fusion method effectively captures the deep correlation of visual, textual, and time series data, and improves the feature expression capability by 30% compared with traditional splicing fusion;

[0076] Dynamic adaptability: the scheduling model supports online learning and can adapt to environmental changes (such as sudden orders and equipment failures) with a response time of ≤50ms;

[0077] Deployment efficiency: Through model splitting and quantization, inference speed is improved by 3 times, and GPU memory occupancy is reduced by 40%. It is suitable for edge devices such as NVIDIA Jetson.

[0078] Multi-dimensional performance evaluation system: Create a multi-task joint learning framework to output device utilization, failure rate, energy efficiency, etc. simultaneously.

[0079] In addition, the resource occupation of the multi-modal large model is too high when deployed on the edge device. The model is split into a visual module and a language module through static graph splitting and hybrid quantization, and is exported as an ONNX format static graph for accelerated inference. Through the detailed implementation and effect verification of the above alternative solutions, the present application has shown significant advantages in multiple scenarios such as warehousing, transportation, and emergency:

[0080] Data fusion depth: Tucker fusion technology realizes cross-modal interaction, and the feature expression capability is improved by more than 30%;

[0081] Decision real-time: The scheduling response time is shortened to seconds, and high-concurrency task processing is supported.

[0082] Energy efficiency optimization: Energy consumption is reduced by 15%-20%, and device utilization is improved by 25%-40%.

[0083] Scenario expandability: Modular design supports rapid adaptation to multiple device types such as AGV, UAV, and cold chain vehicles.

[0084] As shown in Figure 2 As another preferred embodiment of the present application, the step of preprocessing the multi-modal data specifically includes:

[0085] S21, data cleaning is performed on the obtained multi-modal data, and the sliding window mean method is used to eliminate sensor signal jitter, remove noise data and outliers, fill in missing values, and ensure the accuracy and integrity of the data;

[0086] S22, through eigenvalue scaling, logarithmic change and encoding processing, the data of different modalities is converted to consistent scale and format;

[0087] S23, aligning the multi-source data acquisition frequency based on the device operation cycle, through spatial mapping or feature alignment, ensuring the synchronization of the data;

[0088] S24, extracting feature information related to device performance from each modality data, including statistical features, image texture features, and text features, and generating a comprehensive feature vector through data fusion, which is used to provide efficient data representation for subsequent analysis.

[0089] Further, the structured data is used to represent data with fixed format, which can be directly stored in a database or a table, including operation data of the logistics equipment, business operation data, maintenance management data and environment detection data.

[0090] The unstructured data is used to represent data without fixed format, which cannot be directly stored in a database, including text, image, video and audio data which need to be processed by natural language processing.

[0091] In the embodiment, the operation data of the logistics equipment includes speed, energy consumption, load capacity and the like, the business operation data includes order quantity, delivery timeliness, inventory status and the like, and the text data includes maintenance log, fault report, operation manual and the like.

[0092] As another preferred embodiment of the application, the multi-modal Tucker fusion specifically includes the following steps:

[0093] constructing a multi-modal feature tensor based on the preprocessed data, and combining feature vectors of different modalities into a high-order tensor;

[0094] performing Tucker decomposition on the constructed multi-modal feature tensor to decompose the multi-modal feature tensor into a core tensor and a plurality of modal matrices, and representing as:

[0095] T=G×1A×2B×3C, wherein T represents the multi-modal feature tensor, A, B and C are the modal matrices, and G is the core tensor;

[0096] recombining the decomposed modal matrices and the core tensor to obtain a fused feature representation, and representing as:

[0097] y=G×1A×2B×3C, wherein y is the fused feature vector;

[0098] using the fused feature vector y as input to train a downstream task model, and optimizing parameters of the Tucker decomposition through back propagation to improve the fusion effect.

[0099] Further, in the step of combining feature vectors of different modalities into a high-order tensor, it is specifically represented as:

[0100] T∈R Dv×Dt×Ds ,

[0101] wherein Dv, Dt and Ds are respectively used to represent feature dimensions of visual, text and time sequence modalities.

[0102] In the embodiment, in the step of constructing a multi-modal feature tensor based on pre-processed data and combining feature vectors of different modalities into a high-order tensor, the extracted feature vectors of the three modalities are X, Y and Z, a three-order tensor T can be constructed, wherein each element represents the interaction between different modalities.

[0103] As shown in Figure 4 The application also provides a scheduling decision system based on comprehensive efficiency evaluation of logistics equipment, which comprises:

[0104] The data synchronization module 100 is configured to periodically traverse demand information through the terminal device group and the historical database to update and obtain multi-modal data, wherein the demand information includes operation parameters of the logistics equipment, environmental information and video image service data.

[0105] The data preprocessing module 200 is configured to perform data preprocessing on the multi-modal data to obtain structured data and unstructured data with high signal-to-noise ratio, wherein the data preprocessing includes data cleaning, data normalization and data feature extraction.

[0106] The data fusion module 300 is configured to perform large model data fusion on the structured data and the unstructured data based on multi-modal Tucker fusion to establish deep interaction and feature expression of heterogeneous data.

[0107] The evaluation building module 400 is configured to build an equipment efficiency evaluation model based on the fused data and output equipment index information, wherein the equipment index information includes equipment utilization rate, equipment failure rate and equipment energy efficiency.

[0108] The scheduling management module 500 is configured to synchronize real-time service demand, perform scheduling management according to the equipment index information, generate an optimized scheduling scheme, output and execute the optimized scheduling scheme, and realize intelligent scheduling of the logistics equipment.

[0109] As another preferred embodiment of the application, the data preprocessing module comprises:

[0110] The data cleaning unit is configured to perform data cleaning on the obtained multi-modal data, remove noise data and abnormal values by using a sliding window mean method to eliminate sensor signal jitter, fill in missing values, and ensure the accuracy and integrity of the data.

[0111] The data normalization unit is configured to convert data of different modalities to a consistent scale and format through eigenvalue scaling, logarithmic change and encoding processing.

[0112] The data synchronization unit is configured to align multi-source data acquisition frequencies based on the equipment operation cycle, and ensure the synchronization of the data through spatial mapping or feature alignment.

[0113] a feature fusion unit configured to extract feature information related to equipment performance from the multi-modal data, including statistical features, image texture features, and text features, and generate a comprehensive feature vector through data fusion, to provide efficient data representation for subsequent analysis.

[0114] As another preferred embodiment of the present application, the structured data is used to represent data with a fixed format that can be directly stored in a database or table, including operation data of logistics equipment, business operation data, maintenance management data, and environmental detection data.

[0115] The unstructured data is used to represent data without a fixed format that cannot be directly stored in a database, including text, image, video, and audio data that need to be processed through natural language processing.

[0116] As another preferred embodiment of the present application, the data fusion module includes:

[0117] a tensor construction unit configured to construct a multi-modal feature tensor based on the preprocessed data, and combine feature vectors of different modalities into a high-order tensor;

[0118] a tensor decomposition unit configured to perform Tucker decomposition on the constructed multi-modal feature tensor, to decompose the multi-modal feature tensor into a core tensor and a plurality of modal matrices, represented as:

[0119] T = G x 1 A x 2 B x 3 C, wherein T represents the multi-modal feature tensor, A, B, and C are modal matrices, and G is the core tensor;

[0120] a fusion representation unit configured to recombine the decomposed modal matrices and the core tensor to obtain a fused feature representation, represented as:

[0121] y = G x 1 A x 2 B x 3 c, wherein y is the fused feature vector;

[0122] a fusion training unit configured to use the fused feature vector y as input to train a downstream task model, and optimize parameters of the Tucker decomposition through backpropagation to improve the fusion effect.

[0123] As another preferred embodiment of the present application, in the tensor construction unit,

[0124] the manner of combining feature vectors of different modalities into a high-order tensor is represented as:

[0125] t∈R Dv×Dt×Ds ,

[0126] wherein Dv, Dt, and Ds represent feature dimensions of visual, text, and time series modalities, respectively.

[0127] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, databases, or other media in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0128] Other embodiments of the present disclosure will be apparent to those skilled in the art with the disclosure herein. The present application is intended to cover any variations, uses, or adaptations of the present disclosure, including its general principles and specific embodiments, which are disclosed herein. This application is intended to cover such processes or methodologies falling within the scope of the present disclosure. The specification and examples are to be regarded as illustrative only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

[0129] It should be understood that the present disclosure is not limited to the precise structures as herein described and illustrated in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A scheduling decision method based on comprehensive performance evaluation of logistics equipment, characterized in that, The application relates to a logistics equipment intelligent scheduling method and device. Periodic traversal of demand information including operation parameters of logistics equipment, environment information and video image service data is performed through a terminal equipment group and a historical database to update and acquire multi-modal data; Data preprocessing is performed on the multi-modal data to obtain high signal-to-noise ratio structured data and unstructured data, and the data preprocessing includes data cleaning, data normalization and data feature extraction; Based on multi-modal Tucker fusion, structured data and unstructured data are subjected to large model data fusion to establish deep interaction and feature expression of heterogeneous data; Based on the fused data, an equipment performance evaluation model is constructed, and equipment index information including equipment utilization rate, equipment failure rate and equipment energy consumption efficiency is outputted; Real-time service demand is synchronized, scheduling management is performed according to the equipment index information, an optimized scheduling scheme is generated, and the optimized scheduling scheme is outputted and executed to realize intelligent scheduling of logistics equipment. 2.The scheduling decision method based on comprehensive performance evaluation of logistics equipment according to claim 1, characterized in that, The data preprocessing specifically includes: Data cleaning is performed on the acquired multi-modal data, sensor signal jitter is eliminated through a sliding window mean method, noise data and abnormal values are removed, and missing values are filled to ensure the accuracy and integrity of the data; Different modal data are converted to consistent scales and formats through feature value scaling, logarithmic change and coding processing; The multi-source data acquisition frequencies are aligned based on the equipment operation cycle, and the data synchronization is ensured through space mapping or feature alignment; Feature information related to equipment performance is extracted from the modal data, including statistical features, image texture features and text features, and a comprehensive feature vector is generated through data fusion to provide efficient data representation for subsequent analysis. 3.The method of claim 2, wherein, The structured data is used for representing data with a fixed format that can be directly stored in a database or a table, including operation data, business operation data, maintenance management data and environment detection data of logistics equipment; The unstructured data is used for representing data without a fixed format that cannot be directly stored in a database, including text, image, video and audio data that need to be processed through natural language processing. 4.The method according to claim 3, wherein, The multi-modal Tucker fusion specifically includes the following steps: A multi-modal feature tensor is constructed based on the preprocessed data, and feature vectors of different modes are combined into a high-order tensor; The constructed multi-modal feature tensor is subjected to Tucker decomposition to be decomposed into a core tensor and a plurality of modal matrices, and is represented as T=Gx1Ax2Bx3C, wherein T represents the multi-modal feature tensor, A, B and C are the modal matrices, and G is the core tensor; The decomposed modal matrices and the core tensor are recombined to obtain a fused feature representation, which is represented as y=Gx1Ax2Bx3c, wherein y is the fused feature vector; The fused feature vector y is used as input to train a downstream task model, and the parameters of the Tucker decomposition are optimized through back propagation to improve the fusion effect. In the step of combining the feature vectors of different modes into a high-order tensor, the combination is specifically represented as ​ 5. The scheduling decision method based on comprehensive performance evaluation of logistics equipment according to claim 4, characterized in that, ​ T E R Dv×Dt×Ds , Wherein, the Dv, Dt, and Ds are respectively used for representing the feature dimension of visual, text, and time sequence modalities.

6. A scheduling decision system based on comprehensive performance evaluation of logistics equipment, characterized in that, Comprise: A data synchronization module for periodically traversing demand information through a terminal device group and a historical database to update and obtain multi-modal data, the demand information including operation parameters of a logistics device, environmental information, and video image service data; A data preprocessing module for preprocessing the multi-modal data to obtain structured data and unstructured data with high signal-to-noise ratio, the data preprocessing including data cleaning, data normalization, and data feature extraction; A data fusion module for fusing structured data and unstructured data based on multi-modal Tucker fusion to establish deep interaction and feature expression of heterogeneous data; An evaluation building module for building a device performance evaluation model based on the fused data and outputting device index information including device utilization rate, device failure rate, and device energy efficiency; A scheduling management module for synchronizing real-time business demand, managing scheduling according to the device index information, generating an optimized scheduling scheme, outputting and executing the optimized scheduling scheme, and realizing intelligent scheduling of logistics devices. 7.The scheduling decision system based on comprehensive performance evaluation of logistics equipment according to claim 6, characterized in that, The data preprocessing module comprises: A data cleaning unit for cleaning the obtained multi-modal data, removing noise data and outliers through a sliding window mean method, filling missing values, and ensuring data accuracy and integrity; A data normalization unit for converting data of different modalities to consistent scales and formats through eigenvalue scaling, logarithmic change, and encoding processing; A data synchronization unit for aligning multi-source data acquisition frequencies based on device operation cycles, ensuring data synchronization through spatial mapping or feature alignment; A feature fusion unit for extracting feature information related to device performance from each modality data, including statistical features, image texture features, and text features, and generating a comprehensive feature vector through data fusion for efficient data representation for subsequent analysis. 8.The scheduling decision system based on comprehensive performance evaluation of logistics equipment according to claim 7, wherein, The structured data is used to represent data with fixed format that can be directly stored in a database or table, including operation data, business operation data, maintenance management data, and environmental detection data of logistics devices. The unstructured data is used to represent data without fixed format that cannot be directly stored in a database, including text, image, video, and audio data that need to be processed through natural language processing. 9.The scheduling decision system based on comprehensive performance evaluation of logistics equipment according to claim 8, characterized in that, The data fusion module comprises: A tensor construction unit for constructing a multi-modal feature tensor based on preprocessed data, combining feature vectors of different modalities into a high-order tensor; A tensor decomposition unit for decomposing the constructed multi-modal feature tensor through Tucker decomposition into a core tensor and several modality matrices, represented as: T = G × 1A × 2B × 3C, wherein T represents the multi-modal feature tensor, A, B, and C are modality matrices, and G is the core tensor; A fusion representation unit for recombining the decomposed modality matrices and core tensor to obtain a fused feature representation, represented as: y=G×1A×2B×3C, wherein y is the fused feature vector; The fusion training unit is configured to use the fused feature vector y as input to train a downstream task model, and optimize the parameters of the Tucker decomposition through back propagation to improve the fusion effect. 10.The scheduling decision system based on comprehensive performance evaluation of logistics equipment according to claim 9, wherein, In the tensor construction unit; The way of combining the feature vectors of different modalities into a high-order tensor is represented as: T E R Dv×Dt×Ds , wherein Dv, Dt and Ds are respectively used to represent the feature dimensions of the visual, text and time sequence modalities.

Images