Carrying equipment management method and system for intelligent transportation

Abstract

The invention relates to the field of intelligent transportation, and discloses a handling equipment management method for intelligent transportation, which comprises the following steps: acquiring an original data stream of handling equipment according to an Internet of Things sensor; performing space-time alignment calibration on the original data stream to obtain a structured operation state data set with equipment space-time relevance; inputting the structured operation state data set into a space-time convolutional network for feature learning; according to the method, the original data stream of the carrying equipment is collected through the Internet of Things sensor, the state data of the equipment can be obtained in real time in combination with the space-time alignment calibration technology, and the space-time relevance of the data is ensured. The accurate real-time monitoring provides a reliable data basis for subsequent decision making, the scheduling problem caused by data delay or inconsistency in a traditional method is avoided, the structural data are input into the space-time convolutional network for feature learning, the operation features of the equipment can be effectively extracted, and the method is suitable for large-scale popularization and application. And accurate input is provided for subsequent dynamic path planning.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a handling equipment management method and system for intelligent transportation. Background Art

[0002] Intelligent transportation's handling equipment management relies primarily on advanced technologies to improve equipment efficiency, flexibility, and adaptability. This management system combines the Internet of Things, digital twin technology, reinforcement learning, and multi-dimensional optimization strategies to monitor equipment status in real time and intelligently schedule and optimize operations.

[0003] Existing equipment suffers from the following shortcomings: Traditional methods often rely on manually collected or infrequent sensor data, resulting in delayed or inconsistent data updates and an inability to provide real-time equipment status monitoring. Traditional methods often rely on fixed rules or simple models for feature extraction, making them difficult to dynamically adapt to the complexity and variability of equipment operations. Traditional scheduling methods often focus on a single objective (such as path planning or task prioritization), while ignoring other key factors such as energy consumption and equipment wear, failing to achieve comprehensive, multi-dimensional optimization. This can result in low equipment efficiency, even increased energy consumption and worsening equipment wear. Traditional methods often lack digital twin technology and two-way feedback mechanisms, making it impossible to obtain real-time equipment response data or adjust scheduling strategies based on actual operating conditions. This makes continuous improvement in equipment management difficult and their responsiveness is limited in complex environments. Traditional methods often rely on manual intervention or preset rules to make scheduling decisions, lacking adaptive optimization capabilities. They are unable to continuously optimize scheduling models through online learning and the accumulation of historical data, as reinforcement learning algorithms do, thereby improving the system's intelligence and adaptability.

[0004] Therefore, the present application now proposes a handling equipment management method and system for intelligent transportation to solve the above-mentioned problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a handling equipment management method and system for intelligent transportation, so as to solve the problem that the above-mentioned traditional methods usually rely on manual collection or low-frequency sensor data, resulting in delayed or inconsistent data updates and inability to provide real-time equipment status monitoring.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for managing handling equipment for intelligent transportation, comprising: Collecting raw data streams of handling equipment using IoT sensors; performing spatiotemporal alignment on the raw data streams to obtain a structured operating status dataset with spatiotemporal correlation of the equipment; The structured operating status dataset is input into a spatiotemporal convolutional network for feature learning. The potential motion trajectory of the equipment in the future time domain is deduced through virtual simulation, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including equipment obstacle avoidance paths, optimal energy consumption solutions, and task priorities. Convert intelligent scheduling strategy data into a sequence of micro-operation instructions executable by the equipment; push instructions to the handling equipment control system in real time via the 5G network, while building a two-way feedback channel between the digital twin and the physical system to collect actual equipment response data and update the digital twin model status; A closed-loop evaluation system for scheduling strategies is constructed to conduct a multi-dimensional comparative analysis between the actual response data of the equipment and the intelligent scheduling strategy data. A reinforcement learning algorithm is used to optimize the scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability, which is then synchronized to the cloud knowledge base to form a continuously evolving equipment management strategy library.

[0007] The original data stream of the handling equipment is collected based on the IoT sensors in the equipment operation area; the original data stream is time-space aligned and calibrated to obtain a structured operating status dataset with time-space correlation of the equipment, including: Build a global perception network based on IoT sensors in the equipment's operating area; perform spatiotemporal synchronization calibration of sensors through edge computing nodes to obtain a raw data stream containing equipment location coordinates, motion posture, load status, and environmental parameters; The original data stream is subjected to outlier removal and noise filtering; the Kalman filter algorithm is used to smooth the equipment motion trajectory to obtain a cleaned basic operating data set; Ultra-wideband positioning technology is used to perform sub-meter corrections on device coordinates. An event-driven timestamp synchronization mechanism is built to map multi-sensor data streams to a unified spatiotemporal reference to obtain a spatiotemporally aligned intermediate calibration dataset. The intermediate calibration dataset is subjected to feature-level fusion based on a deep belief network. The implicit patterns of the device operating status are extracted through an autoencoder network to obtain a set of structured feature vectors containing spatiotemporal semantic features.

[0008] The process includes collecting raw data streams of handling equipment based on IoT sensors in the equipment operation area, performing spatiotemporal alignment on the raw data streams to obtain a structured operating status dataset with spatiotemporal correlation of the equipment, and further comprising: Build a knowledge graph of the equipment's operating status and inject the structured feature vector set into the graph neural network for relational reasoning. Capture the complex associations of the equipment's load environment through node embedding learning to obtain associated knowledge graph data with causal explanations. A three-dimensional convolutional neural network is used to model the spatio-temporal context of the associated knowledge graph data to obtain a context-enhanced data set containing spatio-temporal evolution rules. An improved dynamic mode decomposition algorithm is used to reduce the dimensionality of the context-enhanced data set; combined with the device operation mode recognition result, a lightweight structured data packet retaining key spatio-temporal features is obtained; The generated adversarial network simulates the real data distribution; the Wasserstein distance is used to measure the similarity between the structured data and the simulated data to obtain the final structured operating state data set that has passed the integrity verification.

[0009] Among them, the structured operating state data set is input into the spatio-temporal convolution network for feature learning; the potential motion trajectory of the device in the future time domain is deduced through virtual simulation, and a multi-objective optimization scheduling instruction set is generated combined with the dynamic path planning algorithm to obtain an intelligent scheduling strategy data containing device obstacle avoidance path, energy consumption optimization solution and task priority, including: The structured operating state data set is decomposed into a spatio-temporal tensor; the dynamic time warping algorithm is used to align the device historical trajectory in time sequence to obtain a standardized input data cube with spatio-temporal continuity; According to the standardized input data cube, spatio-temporal feature extraction is performed to obtain a multi-level feature spectrum containing device motion mode, load variation rule, and environmental interference feature; The multi-level feature spectrum is input into the spatio-temporal feature decoder; the time sequence dependence of the device motion is captured through the gated recurrent unit to obtain the potential motion trajectory distribution cloud map in the future time domain; The potential motion trajectory distribution cloud map is combined with the device dynamics model; the Monte Carlo method is used to simulate the energy consumption fluctuation of different trajectories to obtain a three-dimensional energy consumption map containing energy consumption probability distribution.

[0010] Among them, the structured operating state data set is input into the spatio-temporal convolution network for feature learning; the potential motion trajectory of the device in the future time domain is deduced through virtual simulation, and a multi-objective optimization scheduling instruction set is generated combined with the dynamic path planning algorithm to obtain an intelligent scheduling strategy data containing device obstacle avoidance path, energy consumption optimization solution and task priority, also including: The three-dimensional energy consumption map is injected into an improved search algorithm; by introducing the task priority weight factor and real-time obstacle prediction data, a multi-candidate path set that takes into account energy consumption optimization and task timeliness is obtained; The multi-candidate path set is input into the non-dominated sorting genetic algorithm; by setting energy consumption, time, and device wear as optimization objectives, an optimal path solution set on the Pareto frontier is obtained; Convert the optimal path solution set into a micro-operation instruction sequence executable by the device; encapsulate the scheduling instructions through natural language generation technology to obtain an intelligent scheduling strategy data packet containing obstacle avoidance path coordinates, speed curve, and task switching timing; The intelligent scheduling strategy data packet is injected into the virtual handling system for simulation and deduction, and a closed-loop feedback mechanism is constructed to perform online optimization of the scheduling strategy data to obtain the final intelligent scheduling strategy data that has been verified in actual combat.

[0011] The intelligent scheduling strategy data is converted into a sequence of micro-operation instructions that can be executed by the equipment. These instructions are pushed to the handling equipment control system in real time via the 5G network. A two-way feedback channel between the digital twin and the physical system is established to collect the actual response data of the equipment and update the status of the digital twin model, including: The scheduling policy data is deconstructed into three semantic units: device kinematic constraints, task timing logic, and energy consumption boundary conditions. Templated code generation technology is used to convert the semantic units into micro-operation instruction sequences that can be executed by the device, resulting in a standardized instruction package containing a control instruction set, checksums, and timing labels. Dynamically allocate network resources based on the timing tags of instruction packets; use software-defined networking technology to build a dedicated channel for instruction transmission, and ensure low-latency and high-reliability transmission of instruction streams through forward error correction coding and data packet redundant transmission mechanisms to obtain network transmission confirmation credentials; Perform integrity verification and semantic decompilation on the received standardized instruction packets, and map the micro-operation instruction sequence into pulse width modulation signals and digital output instructions that can be recognized by the device controller to obtain the device-level control instruction stream; By deploying fiber Bragg grating sensor arrays in key equipment components, the stress distribution, vibration spectrum and temperature field changes of the actuator are captured in real time. Compressed sensing technology is used to sparsely sample multimodal sensor data to obtain lightweight equipment response raw data packets.

[0012] This involves converting intelligent scheduling strategy data into a sequence of micro-operation instructions executable by the equipment. This process pushes these instructions to the handling equipment control system in real time via the 5G network. A two-way feedback channel between the digital twin and the physical system is established to collect actual equipment response data and update the digital twin model status. This also includes: Align the device response raw data packet with the network transmission confirmation certificate in time and space; establish a two-way mapping channel between the physical device and the virtual model through digital thread technology, and use a dynamic data-driven application system method to inject real-time data into the twin model to obtain the model status update certificate; Generative adversarial networks are used to perform distribution fitting between model predictions and actual response data. Model error is measured using the Wasserstein distance to obtain an enhanced digital twin model validated with real-world data. Feedback the prediction results of the enhanced digital twin model to the instruction compiler; perform online optimization of the micro-operation instruction sequence through a deep deterministic policy gradient algorithm to obtain an adaptive instruction correction package that takes into account the real-time status of the device; The instruction correction package and model status update certificate are written into the distributed ledger; the instruction update process of the device-side gateway is automatically triggered through the smart contract to obtain a full-link closed-loop iterative system of "strategy generation, instruction execution, response feedback and model optimization".

[0013] Among them, a closed-loop evaluation system for scheduling strategies is constructed, and a multi-dimensional comparative analysis is performed between the actual device response data and the intelligent scheduling strategy data. A reinforcement learning algorithm is used to optimize the scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability. This data is then synchronized to the cloud knowledge base to form a continuously evolving device management strategy library, including: A four-dimensional evaluation framework covering task completion rate, energy efficiency, equipment wear, and path compliance was constructed. Actual equipment response data and intelligent scheduling strategy data were injected into a virtual evaluation environment to generate an initial evaluation report including a spatiotemporal deviation heat map. A dynamic time warping algorithm is used to align the device's motion trajectory in time and space. The uncertainty of policy execution is quantified using the information entropy method and combined with a support vector machine classifier to obtain a multi-dimensional comparative analysis matrix of policy execution quality. Inputting the multi-dimensional comparative analysis matrix into a natural language generation model; capturing the significant features of the evaluation conclusions through sentiment analysis technology to obtain a structured evaluation report data package containing improvement suggestions; The structured evaluation report data packet is deconstructed into state-action pair tuples; the evaluation data is mapped to the reinforcement learning observation space through environment variable injection technology to obtain a standardized training environment configuration.

[0014] Among them, a closed-loop evaluation system for scheduling strategies is constructed, and a multi-dimensional comparative analysis is performed between actual device response data and intelligent scheduling strategy data. A reinforcement learning algorithm is used to optimize scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability. This data package is then synchronized to the cloud knowledge base to form a continuously evolving device management strategy library. The system also includes: Convert indicators such as task completion rate and energy efficiency in the evaluation report into differentiable reward components; automatically adjust the weight coefficient of each reward component based on the weak links of the current scheduling strategy; Injecting standardized training environment configurations and reward functions into the Proximal Policy Optimization (PPO) algorithm; accelerating the model training process through asynchronous parallel computing to obtain an improved scheduling policy model with environmental adaptability; Deploy the improved scheduling policy model to the digital twin sandbox; simulate rare abnormal scenarios through generative adversarial networks and use Monte Carlo tree search to explore the robustness boundaries of the policy to obtain adversarially verified policy quality certification labels; The policy model that has passed quality certification is associated with the original evaluation report and stored; a federated learning architecture is used to achieve the coordinated evolution of policies among multiple edge nodes, so as to obtain an adaptive scheduling policy knowledge graph that continuously absorbs field data and complete the closed-loop evolution of the equipment management policy library.

[0015] A handling equipment management system for intelligent transportation, which is applicable to the above-mentioned handling equipment management method for intelligent transportation, is characterized in that it includes: A spatial alignment unit, configured to collect raw data streams of handling equipment based on IoT sensors; and perform spatiotemporal alignment on the raw data streams to obtain a structured operating status dataset with spatiotemporal correlation of the equipment. An optimization instruction unit is configured to input the structured operating status data set into a spatiotemporal convolutional network for feature learning; generate a multi-objective optimization scheduling instruction set by deducing the potential motion trajectory of the device in the future time domain through virtual simulation and combining it with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including device obstacle avoidance paths, optimal energy consumption solutions, and task priorities; A digital twin unit, which converts intelligent scheduling policy data into a sequence of micro-operation instructions executable by the equipment. This unit pushes these instructions to the handling equipment control system in real time via the 5G network. It also establishes a two-way feedback channel between the digital twin and the physical system, collects actual equipment response data, and updates the digital twin model status. A data optimization unit is used to build a closed-loop evaluation system for scheduling strategies, conduct multi-dimensional comparative analysis between the actual response data of the equipment and the intelligent scheduling strategy data; use a reinforcement learning algorithm to optimize the scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability, and synchronize it to the cloud knowledge base to form a continuously evolving device management strategy library.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention uses IoT sensors to collect raw data streams from handling equipment. Combined with spatiotemporal alignment and calibration technology, it can obtain equipment status data in real time and ensure the spatiotemporal relevance of the data. This precise real-time monitoring provides a reliable data foundation for subsequent decision-making, avoiding scheduling issues caused by data delays or inconsistencies in traditional methods.

[0017] (2) By inputting structured data into a spatiotemporal convolutional network for feature learning, the present invention can effectively extract the operational characteristics of the equipment, providing accurate input for subsequent dynamic path planning. Combined with the dynamic path planning algorithm, it can generate a multi-objective optimized scheduling instruction set in real time, ensuring the optimal operation of the equipment in terms of obstacle avoidance, energy consumption, task priority, etc.

[0018] (3) The present invention uses an intelligent scheduling strategy that not only considers the equipment's obstacle avoidance path but also takes into account the optimal energy consumption solution and task priority, providing the most appropriate scheduling solution for the equipment. This multi-dimensional optimized scheduling method significantly improves handling efficiency and resource utilization.

[0019] (4) Through a two-way feedback channel between digital twin technology and the physical system, the present invention can collect the actual response data of the equipment in real time and update the status of the digital twin model. This closed-loop system can continuously evaluate the actual effect of the scheduling strategy and optimize it, thereby ensuring continuous improvement of equipment management.

[0020] (5) This invention uses a reinforcement learning algorithm to optimize the scheduling model online, enabling the scheduling strategy to continuously and adaptively adjust based on actual operating data, thereby improving the long-term intelligence level of the system. Synchronous updates to the cloud-based knowledge base form a continuously evolving device management strategy library, allowing the scheduling strategy to continuously adapt to different production environments and changing demands.

[0021] (6) This invention combines a reinforcement learning algorithm with a dynamic scheduling strategy. This method is able to cope with different environmental changes and task requirements, greatly improving the flexibility and intelligence level of the handling equipment management system. The adaptive ability of equipment management is significantly enhanced, and it can better cope with complex and uncertain practical application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Fig. 1 A schematic diagram of the overall method steps in one embodiment of the present invention; Fig. 2 FIG. 1 is a schematic diagram of the system architecture structure of the overall system in one embodiment of the present invention.

[0023] In the figure: 1. Spatial alignment unit; 2. Optimization instruction unit; 3. Digital twin unit; 4. Data optimization unit. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] Referring to Figs. 1-2 The present application provides a technical solution: a handling equipment management method for intelligent transportation, comprising: Collecting original data streams of the handling equipment according to Internet of Things sensors; performing spatio-temporal alignment calibration on the original data streams to obtain a structured running state data set with device spatio-temporal correlation; Inputting the structured running state data set into a spatio-temporal convolution network for feature learning; generating a multi-objective optimization scheduling instruction set by combining a dynamic path planning algorithm to obtain an intelligent scheduling strategy data containing a device obstacle avoidance path, an energy consumption optimal solution and a task priority by deducing the potential motion trajectory of the device in the future time domain through virtual simulation; Converting the intelligent scheduling strategy data into a micro-operation instruction sequence executable by the device; pushing the instructions to the handling equipment control system in real time through a 5G network, simultaneously constructing a two-way feedback channel between the digital twin and the physical system, collecting actual response data of the device and updating the state of the digital twin model; Constructing a closed-loop evaluation system of the scheduling strategy, performing multi-dimensional comparative analysis on the actual response data of the device and the intelligent scheduling strategy data; using a reinforcement learning algorithm to perform online optimization on the scheduling model parameters to obtain an improved scheduling strategy data packet with enhanced self-adaptive capability, and synchronizing to the cloud knowledge base to form a continuously evolving equipment management strategy library.

[0026] It should be noted that, when operating, IoT sensors are capable of exchanging data with other devices or systems over a network. These sensors collect real-time data on device or environment status, such as temperature, location, and speed, and transmit this data to a central system for processing and analysis. Raw data streams refer to unprocessed data collected from handling equipment by IoT sensors in real time. This data is typically a high-frequency, continuous stream containing information about the equipment's operating status and the environment. Spatiotemporal alignment is the process of preprocessing the raw data stream to ensure accurate alignment in both time and space. Different devices or sensors may experience time delays or positional deviations. Spatiotemporal alignment uses algorithms to synchronize data from different sources, ensuring consistency in time and space. After spatiotemporal alignment, the raw data stream is converted into a structured data format that contains device operating status information and has been processed for spatiotemporal relationships, facilitating subsequent analysis and processing. A spatiotemporal convolutional network is a deep learning model for processing spatiotemporal data. It combines the characteristics of convolutional neural networks (CNNs) and time series data to effectively extract spatial and temporal features. In intelligent transportation systems, ST-CN is used to analyze the spatiotemporal operational status of equipment and learn its operating modes. Virtual simulation, using virtual simulation technology, predicts the future trajectory of equipment in a computer-simulated environment. This simulation technology can infer the equipment's likely behavior in the future time domain based on its current state, which is then used to generate scheduling strategies. Dynamic path planning algorithms calculate and optimize the movement paths of transport equipment in real time based on constantly changing environmental data and target requirements. These algorithms consider factors such as obstacles, target location, and energy consumption, dynamically adjusting the equipment's path during transportation. A multi-objective optimization scheduling instruction set is a set of optimized scheduling instructions that consider multiple objectives (such as obstacle avoidance, energy optimization, and task priority). Using this multi-objective optimization algorithm, the scheduling system can balance these objectives to generate an optimal scheduling solution. Intelligent scheduling strategy data is generated through this optimization process and includes information such as the equipment's obstacle avoidance path, optimal energy consumption solution, and task priority. Micro-operation instruction sequences convert scheduling strategy data into directly executable control commands. These commands are typically expressed as small operation units (micro-operations) to ensure precision and flexibility during equipment execution. 5G networks, the fifth generation of mobile communications (5G), are characterized by high bandwidth, low latency, and massive connectivity. Through 5G networks, commands can be transmitted to equipment control systems in real time without delay, ensuring real-time and efficient systems. A digital twin is a virtual model of a physical object or system that continuously updates its status using real-time data, simulating its behavior and performance. Digital twin technology enables real-time monitoring and prediction of handling equipment, as well as virtual testing and optimization.A bidirectional feedback channel, established between the digital twin and the physical device, collects actual device response data in real time and feeds this data into the digital twin model, continuously updating the device's virtual state and behavior predictions. A closed-loop evaluation system evaluates the effectiveness and accuracy of the scheduling policy by continuously tracking and comparing the differences between the device's actual response data and the intelligent scheduling policy. This closed-loop evaluation system ensures continuous optimization and adjustment of the scheduling policy. Reinforcement learning algorithms are a machine learning method that uses reward and penalty mechanisms to enable intelligent agents to learn how to take actions in their environment to maximize long-term benefits. In device scheduling, reinforcement learning can be used to optimize scheduling models online and enhance their adaptability. The improved scheduling policy data package with enhanced adaptability uses reinforcement learning and closed-loop evaluation to continuously adjust scheduling policies based on actual device response data, generating scheduling solutions that are more adaptable to diverse environments and changing demands. The cloud-based knowledge base centrally stores and manages device management policies and scheduling models, providing knowledge support for the system and enabling the continuous evolution of device management policies. Through synchronized updates on the cloud, data in the knowledge base can be shared and applied across various device management systems.

[0027] In one embodiment, the original data stream of the handling equipment is collected based on the Internet of Things sensors in the equipment operation area; the original data stream is time-space aligned and calibrated to obtain a structured operation status data set with equipment time-space correlation, including: building a global perception network based on the Internet of Things sensors in the equipment operation area; performing time-space synchronization calibration on the sensors through edge computing nodes to obtain an original data stream containing equipment position coordinates, motion posture, load status, and environmental parameters; performing outlier removal and noise filtering on the original data stream; using the Kalman filter algorithm to smooth the equipment motion trajectory to obtain a cleaned basic operation data set; using ultra-wideband positioning technology to perform sub-meter correction on the equipment position coordinates; constructing an event-driven timestamp synchronization mechanism to map multi-sensor data streams to a unified time-space benchmark to obtain a time-space aligned intermediate calibration data set; performing feature-level fusion on the intermediate calibration data set based on a deep belief network; extracting implicit patterns of the equipment operation status through an autoencoder network to obtain a structured feature vector set containing time-space semantic features.

[0028] This design collects raw data streams from handling equipment through IoT sensors and synchronizes this data using spatiotemporal alignment and calibration technology to obtain a structured dataset of the equipment's operational status. First, edge computing synchronizes the sensors in spatiotemporal order, correcting data such as position, motion posture, and load, and performing outlier removal and noise filtering. A Kalman filter algorithm smooths the equipment's motion trajectory, while ultra-wideband positioning technology corrects position accuracy. An event-driven timestamp synchronization mechanism maps data from multiple sensors to a unified spatiotemporal reference. Next, a deep belief network and an autoencoder network are used to fuse features and extract implicit patterns from the data, resulting in a structured feature vector set containing spatiotemporal semantic features. These processes ensure data accuracy and consistency, providing reliable input for subsequent intelligent scheduling.

[0029] In one embodiment, the original data stream of the handling equipment is collected based on the Internet of Things sensors in the equipment operation area; the original data stream is time-space aligned and calibrated to obtain a structured operation status data set with equipment time-space correlation, and the method also includes: constructing a knowledge graph of the equipment operation status, injecting the structured feature vector set into the graph neural network for relational reasoning; capturing the complex correlation of the equipment load environment through node embedding learning to obtain associated knowledge graph data with causal interpretation; using a three-dimensional convolutional neural network to perform time-space context modeling on the associated knowledge graph data to obtain a context-enhanced data set that contains time-space evolution laws; using an improved dynamic pattern decomposition algorithm to perform dimensionality reduction processing on the context-enhanced data set; combining the equipment operation mode recognition results to obtain a lightweight structured data packet that retains key time-space features; simulating the real data distribution according to the generative adversarial network; using the Wasserstein distance to measure the similarity between the structured data and the simulated data to obtain a final structured operation status data set that has been verified for integrity.

[0030] This design collects the device's raw data stream through IoT sensors and performs spatiotemporal alignment to generate a structured operating status dataset. First, a knowledge graph of the device's operating status is constructed, and a set of feature vectors is injected into a graph neural network for relational reasoning, capturing the complex connections between device loads and the environment. This generates knowledge graph data with causal interpretability. Next, a three-dimensional convolutional neural network is used to model the data's spatiotemporal context, extracting spatiotemporal evolution patterns and generating a context-enhanced dataset. Subsequently, dimensionality reduction is performed using an improved dynamic pattern decomposition algorithm, combined with device operating pattern recognition to generate a lightweight structured data packet. Finally, a generative adversarial network is used to simulate the real-world data distribution, and the Wasserstein distance is used to measure the similarity between the data and the simulated data to verify data integrity. This process ensures the accuracy, interpretability, and efficiency of the dataset, providing strong data support for intelligent scheduling and optimization of equipment.

[0031] In one embodiment, a structured operating status data set is input into a spatiotemporal convolutional network for feature learning; the potential motion trajectory of the device in the future time domain is deduced through virtual simulation, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including the device obstacle avoidance path, optimal energy consumption solution and task priority, including: deconstructing the structured operating status data set into a spatiotemporal tensor; performing time alignment on the device historical trajectory through a dynamic time warping algorithm to obtain a standardized input data cube with spatiotemporal continuity; performing spatiotemporal feature extraction based on the standardized input data cube to obtain a multi-level feature map including the device motion mode, load change law, and environmental interference characteristics; inputting the multi-level feature map into a spatiotemporal feature decoder; capturing the temporal dependency of the device motion through a gated recurrent unit to obtain a potential motion trajectory distribution cloud map in the future time domain; combining the potential motion trajectory distribution cloud map with the device dynamics model; and using a Monte Carlo method to simulate energy consumption fluctuations of different trajectories to obtain a three-dimensional energy consumption map including the probability distribution of energy consumption.

[0032] This design uses a spatiotemporal convolutional network to perform feature learning on structured operating status datasets, thereby generating intelligent scheduling strategy data. First, the data is deconstructed into spatiotemporal tensors, and the dynamic time warping algorithm is used to align the historical trajectories of the equipment in time to obtain a standardized input data cube. Then, spatiotemporal feature extraction is performed to generate a multi-level feature map, which includes equipment motion patterns, load changes, and environmental interference characteristics. The temporal dependencies of equipment motion are captured through gated recurrent units to predict the potential motion trajectory distribution in the future time domain. Combining the equipment dynamics model and the Monte Carlo method, the energy consumption fluctuations of different trajectories are simulated to generate a three-dimensional energy consumption map. Finally, combined with the dynamic path planning algorithm, the multi-objective scheduling strategy is optimized, including obstacle avoidance paths, optimal energy consumption solutions, and task priorities. This method realizes the intelligent scheduling of equipment and improves the efficiency and energy efficiency of task execution.

[0033] In one embodiment, a structured operating status dataset is input into a spatiotemporal convolutional network for feature learning; the potential motion trajectory of the equipment in the future time domain is deduced through virtual simulation, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including the equipment obstacle avoidance path, energy consumption optimal solution and task priority. The method also includes: injecting a three-dimensional energy consumption map into an improved search algorithm; introducing task priority weight factors and real-time obstacle prediction data to obtain a set of multiple candidate paths that take into account both energy consumption optimization and task timeliness; inputting the multiple candidate path sets into a non-dominated sorting genetic algorithm; setting energy consumption, time, and equipment wear as optimization targets to obtain an optimal path solution set on the Pareto front; converting the optimal path solution set into a sequence of micro-operation instructions executable by the equipment; encapsulating the scheduling instructions through natural language generation technology to obtain an intelligent scheduling strategy data packet including obstacle avoidance path coordinates, speed curve, and task switching timing; injecting the intelligent scheduling strategy data packet into a virtual handling system for simulation deduction, and constructing a closed-loop feedback mechanism to perform online optimization of the scheduling strategy data to obtain final intelligent scheduling strategy data that has been verified in actual combat.

[0034] This design generates a three-dimensional energy consumption map and feeds it into an improved search algorithm. By introducing task priorities and real-time obstacle prediction data, multiple candidate path sets are obtained. Next, a non-dominated sorting genetic algorithm is used to optimize multi-objective scheduling, setting energy consumption, time, and equipment wear as targets to obtain the optimal path solution set on the Pareto front. The optimal path solution set is converted into a sequence of micro-operation instructions executable by the device and then encapsulated into an intelligent scheduling strategy data package using natural language generation technology, including obstacle avoidance paths, speed curves, and task switching timing. Finally, the data package is input into a virtual handling system for simulation and deduction, and online optimization is performed through a closed-loop feedback mechanism, resulting in a final scheduling strategy that has been verified in actual combat. This method effectively improves equipment scheduling efficiency and task execution accuracy, while taking into account energy efficiency and task timeliness.

[0035] In one embodiment, the intelligent scheduling strategy data is converted into a micro-operation instruction sequence that can be executed by the device; the instructions are pushed to the handling equipment control system in real time through the 5G network, and a two-way feedback channel between the digital twin and the physical system is constructed at the same time to collect the actual response data of the device and update the state of the digital twin model, including: deconstructing the scheduling strategy data into three layers of semantic units: device kinematic constraints, task timing logic, and energy consumption boundary conditions; converting the semantic units into a micro-operation instruction sequence that can be executed by the device through templated code generation technology to obtain a standardized instruction package containing a control instruction set, a check code, and a timing label; dynamically allocating network resources according to the timing label of the instruction package; and using software-defined network technology to construct a dedicated instruction transmission system. The channel ensures low-latency and high-reliability transmission of the instruction stream through forward error correction coding and data packet redundant transmission mechanism to obtain network transmission confirmation credentials; performs integrity verification and semantic decompilation on the received standardized instruction packets, maps the micro-operation instruction sequence into pulse width modulation signals and digital output instructions that can be recognized by the device controller, and obtains the device-level control instruction stream; deploys fiber grating sensor arrays or micro-MEMS sensors (Micro-MEMSSensors) in key components of the equipment to capture the stress distribution, vibration spectrum and temperature field changes of the actuator in real time; uses compressed sensing technology to sparsely sample multimodal sensor data to obtain lightweight device response raw data packets.

[0036] This design pushes executable micro-operation instruction sequences to the control system in real time via the 5G network, enabling bidirectional feedback between the digital twin model and the physical system. First, the intelligent scheduling policy data is decomposed into three semantic units: device kinematic constraints, task timing logic, and energy consumption boundary conditions. Templated code generation technology is used to convert these into standardized instruction packets. Network resources are then dynamically allocated, and dedicated channels are established through software-defined networking to ensure low-latency, highly reliable transmission. The instruction packets undergo integrity verification and semantic decompilation, mapping them into signals recognizable by the device controller. A fiber Bragg grating sensor array is deployed to capture stress, vibration, and temperature changes in key components in real time. Compressed sensing technology is used to sparsely sample multimodal sensor data to generate lightweight device response data packets. This process ensures precise device control, real-time feedback, and adjustments, improving the overall performance and response speed of the system.

[0037] In one embodiment, intelligent scheduling policy data is converted into a micro-operation instruction sequence executable by the device; instructions are pushed to the handling equipment control system in real time via the 5G network, and a bidirectional feedback channel is established between the digital twin and the physical system to collect the actual device response data and update the digital twin model state. The process also includes: temporally and spatially aligning the device response raw data packet with the network transmission confirmation certificate; establishing a bidirectional mapping channel between the physical device and the virtual model via digital thread technology, and injecting real-time data into the twin model using a dynamic data-driven application system method to obtain a model state update certificate; using a generative adversarial network to perform distribution fitting between the model prediction results and the actual response data; measuring the model error using the Wasserstein distance to obtain an enhanced digital twin model verified by actual combat data; feeding the prediction results of the enhanced digital twin model back to the instruction compiler; online optimization of the micro-operation instruction sequence using a deep deterministic policy gradient algorithm to obtain an adaptive instruction correction package that takes into account the real-time status of the device; writing the instruction correction package and the model state update certificate to a distributed ledger; and automatically triggering the instruction update process of the device-side gateway via a smart contract to obtain a full-link closed-loop iterative system of "strategy generation, instruction execution, response feedback, and model optimization."

[0038] Digital twin model update formula: in, Indicates the updated state of the digital twin model, Indicates the current state of the digital twin model, Represents a mapping function, which indicates how to drive the update of model status with real-time data. Indicates the real-time response data obtained from the device. Represents the hyperparameters or weights in the twin model, used to tune the model.

[0039] Wasserstein distance formula: in, Represents two probability distributions and between (Wasserstein distance), Represents the joint distribution of all possible Take the lower bound, that is, find an optimal joint distribution that minimizes the total "transportation cost" (that is, the weighted sum of distances). express is a joint distribution belonging to In the distribution and The joint distribution of pairs, It is defined in two probability spaces The probability measure on , which indicates how to and Pairing of sample points represents the expected value, which represents the joint distribution Next, right All points in space The distance function of Integrate the powers, Indicates the order of distance, that is, the distance measurement method. Common The values ​​are 1 (Manhattan distance) and 2 (Euclidean distance), Controls the degree of penalty for sample points with a larger distance. A value of means that the influence of distant sample pairs will be more emphasized. Represents sample points and The distance function between them is usually Euclidean distance, which represents the difference or distance between two sample points. In the joint distribution Next, the sample pair The joint probability density of .

[0040] This design pushes micro-operation instructions to the handling equipment control system in real time, and uses digital twin technology to achieve two-way feedback between the virtual model and the physical system. First, the device response data and the network transmission confirmation certificate are aligned in time and space, and the real-time data is injected into the digital twin model through digital thread technology to update the model state. Then, a generative adversarial network is used to fit the model prediction results and actual response data, and the error is measured using the Wasserstein distance to generate an enhanced digital twin model. This model is fed back to the instruction compiler, and the micro-operation instruction sequence is optimized using a deep deterministic policy gradient algorithm to obtain an adaptive instruction correction package. Next, the instruction correction package and the model update certificate are written to the distributed ledger, and the instruction update process of the device-side gateway is triggered by a smart contract, forming a full-link closed-loop system and optimizing device control in real time. This method improves the system's response accuracy and adaptability, ensuring the accuracy and real-time performance of device execution.

[0041] In one embodiment, a closed-loop evaluation system for scheduling strategies is constructed, and a multi-dimensional comparative analysis is performed between the actual response data of the equipment and the intelligent scheduling strategy data; a reinforcement learning algorithm is used to perform online optimization of the scheduling model parameters to obtain an improved scheduling strategy data packet with enhanced adaptability, and the data packet is synchronized to the cloud knowledge base to form a continuously evolving equipment management strategy library, including: constructing a four-dimensional evaluation framework covering task completion rate, energy efficiency, equipment wear and path compliance; injecting the actual response data of the equipment and the intelligent scheduling strategy data into a virtual evaluation environment to obtain an initial evaluation report containing a spatiotemporal deviation heat map; using a dynamic time warping algorithm to perform spatiotemporal alignment of the equipment motion trajectory; quantifying the uncertainty of strategy execution through an information entropy method combined with a support vector machine classifier to obtain a multi-dimensional comparative analysis matrix of the strategy execution quality; inputting the multi-dimensional comparative analysis matrix into a natural language generation model; capturing the significant features of the evaluation conclusions through sentiment analysis technology to obtain a structured evaluation report data packet containing improvement suggestions; deconstructing the structured evaluation report data packet into state-action pair tuples; and mapping the evaluation data to the reinforcement learning observation space through environmental variable injection technology to obtain a standardized training environment configuration.

[0042] This design optimizes equipment management strategies by building a closed-loop evaluation system for scheduling policies. First, a four-dimensional evaluation framework is constructed for task completion rate, energy efficiency, equipment wear, and path compliance. Equipment response data and intelligent scheduling policy data are injected into a virtual evaluation environment to generate a preliminary evaluation report. The motion trajectories are then spatiotemporally aligned using a dynamic time warping algorithm. Information entropy and support vector machine methods are used to quantify the uncertainty of policy execution, forming a multi-dimensional comparative analysis matrix. Next, an evaluation report is generated using a natural language generation model, and sentiment analysis techniques are used to extract significant features from the evaluation conclusions and provide improvement suggestions. These improvement suggestions are converted into state-action pair tuples and mapped into the observation space of reinforcement learning, configuring a standardized training environment. Finally, a reinforcement learning algorithm is used to optimize the scheduling model online, resulting in an improved scheduling policy with enhanced adaptability. This policy is then synchronized and updated to the cloud-based knowledge base, forming a continuously evolving equipment management policy library. This approach enables continuous optimization and precise control of scheduling policies.

[0043] In one embodiment, a closed-loop evaluation system for scheduling strategies is constructed, and actual device response data and intelligent scheduling strategy data are compared and analyzed in multiple dimensions. A reinforcement learning algorithm is used to optimize scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability, which is synchronized to a cloud knowledge base to form a continuously evolving device management strategy library. The system also includes: converting indicators such as task completion rate and energy efficiency in the evaluation report into differentiable reward components; automatically adjusting the weight coefficients of each reward component based on the weak links of the current scheduling strategy; injecting standardized training environment configuration and reward function into the proximal policy optimization PPO algorithm; accelerating the model training process through asynchronous parallel computing to obtain an improved scheduling strategy model with environmental adaptability; deploying the improved scheduling strategy model to a digital twin sandbox; simulating rare abnormal scenarios through generative adversarial networks, and exploring the strategy robustness boundary through Monte Carlo tree search to obtain a strategy quality certification label that has been adversarially verified; and associating the quality-certified strategy model with the original evaluation report for storage. A federated learning architecture is used to achieve strategy co-evolution of multiple edge nodes to obtain an adaptive scheduling strategy knowledge graph that continuously absorbs field data, completing the closed-loop evolution of the device management strategy library.

[0044] This design continuously optimizes device management strategies by building a closed-loop evaluation system for scheduling policies. First, the task completion rate, energy efficiency, and other metrics in the evaluation report are converted into differentiable reward components. The weight coefficients of each reward component are dynamically adjusted based on the weaknesses of the current scheduling policy. By injecting the reward function and standardized training environment configuration into the Proximal Policy Optimization (PPO) algorithm and accelerating model training using asynchronous parallel computing, an adaptive scheduling policy model is developed. This model is deployed in a digital twin sandbox, using a generative adversarial network to simulate abnormal scenarios, and Monte Carlo tree search to verify policy robustness. The quality-certified policy model is associated with the original evaluation report and stored, and a federated learning architecture is used to enable policy co-evolution across multiple edge nodes. Finally, an adaptive scheduling policy knowledge graph that continuously incorporates field data is constructed, completing the closed-loop evolution of the device management policy library. This process not only optimizes the scheduling policy but also ensures its ability to cope with complex scenarios and continuously improves system performance.

[0045] A handling equipment management system for intelligent transportation, which is applicable to the above-mentioned handling equipment management method for intelligent transportation, comprises: The spatial alignment unit 1 is used to collect the original data stream of the handling equipment according to the IoT sensor; perform spatiotemporal alignment on the original data stream to obtain a structured operating status data set with spatiotemporal correlation of the equipment; Optimization instruction unit 2 is used to input the structured operating status data set into the spatiotemporal convolutional network for feature learning. Through virtual simulation, the potential motion trajectory of the equipment in the future time domain is deduced, and combined with the dynamic path planning algorithm, a multi-objective optimization scheduling instruction set is generated to obtain intelligent scheduling strategy data including equipment obstacle avoidance path, energy consumption optimization solution and task priority. Digital twin unit 3 is used to convert intelligent scheduling strategy data into a sequence of micro-operation instructions that can be executed by the equipment. This unit pushes these instructions to the handling equipment control system in real time via the 5G network. It also establishes a two-way feedback channel between the digital twin and the physical system, collects the equipment's actual response data, and updates the digital twin model status. Data optimization unit 4 is used to build a closed-loop evaluation system for scheduling strategies, and conduct multi-dimensional comparative analysis between the actual response data of the equipment and the intelligent scheduling strategy data. It uses a reinforcement learning algorithm to optimize the scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability, and synchronizes it to the cloud knowledge base to form a continuously evolving equipment management strategy library.

[0046] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for managing handling equipment for intelligent transportation, characterized in that: include: Collect the original data stream of the handling equipment based on IoT sensors; Performing spatiotemporal alignment calibration on the original data stream to obtain a structured operating status dataset with spatiotemporal correlation of the equipment; Inputting the structured operating status dataset into a spatiotemporal convolutional network for feature learning; Through virtual simulation, the potential motion trajectory of the equipment in the future time domain is deduced, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including equipment obstacle avoidance paths, optimal energy consumption solutions, and task priorities; Convert intelligent scheduling strategy data into a sequence of micro-operation instructions executable by the equipment; push instructions to the handling equipment control system in real time via the 5G network, while building a two-way feedback channel between the digital twin and the physical system to collect actual equipment response data and update the digital twin model status; A closed-loop evaluation system for scheduling strategies is constructed to conduct multi-dimensional comparative analysis between the actual response data of the equipment and the intelligent scheduling strategy data. A reinforcement learning algorithm is used to optimize the scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability, which is then synchronized to the cloud knowledge base to form a continuously evolving equipment management strategy library.

2. A method for managing handling equipment for intelligent transportation according to claim 1, characterized in that: Collect the original data stream of the handling equipment based on the IoT sensors in the equipment operation area; Performing spatiotemporal alignment on the original data stream to obtain a structured operating status dataset with spatiotemporal correlation of the equipment, including: Build a global perception network based on IoT sensors in the equipment's operating area; perform spatiotemporal synchronization calibration of sensors through edge computing nodes to obtain a raw data stream containing equipment location coordinates, motion posture, load status, and environmental parameters; The original data stream is subjected to outlier removal and noise filtering; the Kalman filter algorithm is used to smooth the equipment motion trajectory to obtain a cleaned basic operating data set; Ultra-wideband positioning technology is used to perform sub-meter corrections on device coordinates. An event-driven timestamp synchronization mechanism is built to map multi-sensor data streams to a unified spatiotemporal reference to obtain a spatiotemporally aligned intermediate calibration dataset. The intermediate calibration dataset is subjected to feature-level fusion based on a deep belief network. The implicit patterns of the device operating status are extracted through an autoencoder network to obtain a set of structured feature vectors containing spatiotemporal semantic features.

3. A method for managing handling equipment for intelligent transportation according to claim 2, characterized in that: Collect the original data stream of the handling equipment based on the IoT sensors in the equipment operation area; Performing spatiotemporal alignment on the original data stream to obtain a structured operating status dataset with spatiotemporal correlation of the equipment, further comprising: Build a knowledge graph of the equipment's operating status and inject the structured feature vector set into the graph neural network for relational reasoning. Capture the complex associations of the equipment's load environment through node embedding learning to obtain associated knowledge graph data with causal explanations. Use 3D convolutional neural networks to model the spatiotemporal context of associated knowledge graph data to obtain a context-enhanced dataset that includes spatiotemporal evolution laws; An improved dynamic pattern decomposition algorithm is used to reduce the dimensionality of the context-enhanced dataset. This is combined with the device operation pattern recognition results to obtain a lightweight structured data packet that retains key spatiotemporal features. The real data distribution is simulated based on the generative adversarial network; the Wasserstein distance is used to measure the similarity between structured data and simulated data to obtain the final structured operating status dataset with verified integrity.

4. A method for managing handling equipment for intelligent transportation according to claim 3, characterized in that: Inputting the structured operating status dataset into a spatiotemporal convolutional network for feature learning; Through virtual simulation, the potential motion trajectory of the equipment in the future time domain is deduced, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including equipment obstacle avoidance paths, optimal energy consumption solutions, and task priorities, including: Deconstructing the structured operating status dataset into a spatiotemporal tensor; performing temporal alignment on the historical trajectories of the equipment using a dynamic time warping algorithm to obtain a standardized input data cube with spatiotemporal continuity; Extracting spatiotemporal features based on the standardized input data cube to obtain a multi-level feature map including equipment motion patterns, load variation patterns, and environmental interference features; The multi-level feature maps are input into the spatiotemporal feature decoder. The temporal dependencies of the device motion are captured through the gated recurrent unit to obtain a potential motion trajectory distribution cloud map in the future time domain. The potential motion trajectory distribution cloud map is combined with the equipment dynamics model; the Monte Carlo method is used to simulate the energy consumption fluctuations of different trajectories to obtain a three-dimensional energy consumption map containing the probability distribution of energy consumption.

5. A method for managing handling equipment for intelligent transportation according to claim 4, characterized in that: Inputting the structured operating status dataset into a spatiotemporal convolutional network for feature learning; Through virtual simulation, the potential motion trajectory of the equipment in the future time domain is deduced, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including equipment obstacle avoidance paths, optimal energy consumption solutions, and task priorities. It also includes: The three-dimensional energy consumption map is injected into the improved search algorithm; by introducing task priority weight factors and real-time obstacle prediction data, a set of multiple candidate paths that takes into account both energy optimization and task timeliness is obtained; Inputting the multiple candidate path sets into a non-dominated sorting genetic algorithm; setting energy consumption, time, and equipment wear as optimization targets to obtain an optimal path solution set on the Pareto frontier; Convert the optimal path solution set into a micro-operation instruction sequence executable by the device; encapsulate the scheduling instructions through natural language generation technology to obtain an intelligent scheduling strategy data packet containing obstacle avoidance path coordinates, speed curve, and task switching timing; The intelligent scheduling strategy data packet is injected into the virtual handling system for simulation and deduction, and a closed-loop feedback mechanism is constructed to perform online optimization of the scheduling strategy data to obtain the final intelligent scheduling strategy data that has been verified in actual combat.

6. A method for managing handling equipment for intelligent transportation according to claim 5, characterized in that: Convert intelligent scheduling strategy data into a sequence of micro-operation instructions executable by the equipment; push instructions to the handling equipment control system in real time via the 5G network. Simultaneously, establish a two-way feedback channel between the digital twin and the physical system, collect actual equipment response data, and update the digital twin model status, including: The scheduling policy data is deconstructed into three semantic units: device kinematic constraints, task timing logic, and energy consumption boundary conditions. Templated code generation technology is used to convert the semantic units into micro-operation instruction sequences that can be executed by the device, resulting in a standardized instruction package containing a control instruction set, checksums, and timing labels. Dynamically allocate network resources based on the timing tags of instruction packets; use software-defined networking technology to build a dedicated channel for instruction transmission, and ensure low-latency and high-reliability transmission of instruction streams through forward error correction coding and data packet redundant transmission mechanisms to obtain network transmission confirmation credentials; Perform integrity verification and semantic decompilation on the received standardized instruction packets, and map the micro-operation instruction sequence into pulse width modulation signals and digital output instructions that can be recognized by the device controller to obtain the device-level control instruction stream; By deploying fiber Bragg grating sensor arrays in key equipment components, the stress distribution, vibration spectrum and temperature field changes of the actuator are captured in real time. Compressed sensing technology is used to sparsely sample multimodal sensor data to obtain lightweight equipment response raw data packets.

7. A method for managing handling equipment for intelligent transportation according to claim 6, characterized in that: Convert intelligent scheduling strategy data into a sequence of micro-operation instructions executable by the equipment; push these instructions to the handling equipment control system in real time via the 5G network. Simultaneously, a two-way feedback channel between the digital twin and the physical system is established to collect actual equipment response data and update the digital twin model status. This also includes: Align the device response raw data packet with the network transmission confirmation certificate in time and space; establish a two-way mapping channel between the physical device and the virtual model through digital thread technology, and use a dynamic data-driven application system method to inject real-time data into the twin model to obtain the model status update certificate; Generative adversarial networks are used to perform distribution fitting between model predictions and actual response data. Model error is measured using the Wasserstein distance to obtain an enhanced digital twin model validated with real-world data. Feedback the prediction results of the enhanced digital twin model to the instruction compiler; perform online optimization of the micro-operation instruction sequence through a deep deterministic policy gradient algorithm to obtain an adaptive instruction correction package that takes into account the real-time status of the device; The instruction correction package and model status update certificate are written into the distributed ledger; the instruction update process of the device-side gateway is automatically triggered through the smart contract to obtain a full-link closed-loop iterative system of "strategy generation, instruction execution, response feedback and model optimization".

8. A method for managing handling equipment for intelligent transportation according to claim 7, characterized in that: A closed-loop evaluation system for scheduling strategies is constructed, and a multi-dimensional comparative analysis is performed between actual device response data and intelligent scheduling strategy data. A reinforcement learning algorithm is used to optimize scheduling model parameters online to obtain an improved scheduling strategy data package with enhanced adaptability. This data is then synchronized to the cloud knowledge base to form a continuously evolving device management strategy library, including: A four-dimensional evaluation framework covering task completion rate, energy efficiency, equipment wear, and path compliance was constructed. Actual equipment response data and intelligent scheduling strategy data were injected into a virtual evaluation environment to generate an initial evaluation report including a spatiotemporal deviation heat map. A dynamic time warping algorithm is used to align the device's motion trajectory in time and space. The uncertainty of policy execution is quantified using the information entropy method and combined with a support vector machine classifier to obtain a multi-dimensional comparative analysis matrix of policy execution quality. Inputting the multi-dimensional comparative analysis matrix into a natural language generation model; capturing the significant features of the evaluation conclusions through sentiment analysis technology to obtain a structured evaluation report data package containing improvement suggestions; The structured evaluation report data packet is deconstructed into state-action pair tuples; the evaluation data is mapped to the reinforcement learning observation space through environment variable injection technology to obtain a standardized training environment configuration.

9. A method for managing handling equipment for intelligent transportation according to claim 8, characterized in that: A closed-loop evaluation system for scheduling strategies is built, performing multi-dimensional comparative analysis between actual device response data and intelligent scheduling strategy data. A reinforcement learning algorithm is used to optimize scheduling model parameters online to generate improved scheduling strategy data packages with enhanced adaptability. These data packages are then synchronized to the cloud knowledge base to form a continuously evolving device management strategy library. This also includes: Convert indicators such as task completion rate and energy efficiency in the evaluation report into differentiable reward components; automatically adjust the weight coefficient of each reward component based on the weak links of the current scheduling strategy; Injecting standardized training environment configurations and reward functions into the Proximal Policy Optimization (PPO) algorithm; accelerating the model training process through asynchronous parallel computing to obtain an improved scheduling policy model with environmental adaptability; Deploy the improved scheduling policy model to the digital twin sandbox; simulate rare abnormal scenarios through generative adversarial networks and use Monte Carlo tree search to explore the robustness boundaries of the policy to obtain adversarially verified policy quality certification labels; The policy model that has passed quality certification is associated with the original evaluation report and stored; a federated learning architecture is used to achieve the coordinated evolution of policies among multiple edge nodes, so as to obtain an adaptive scheduling policy knowledge graph that continuously absorbs field data and complete the closed-loop evolution of the equipment management policy library.

10. A handling equipment management system for intelligent transportation, which is applicable to a handling equipment management method for intelligent transportation according to any one of claims 1 to 9, characterized in that: include, A spatial alignment unit (1), the spatial alignment unit (1) is used to collect the original data stream of the handling equipment according to the Internet of Things sensor; perform spatiotemporal alignment calibration on the original data stream to obtain a structured operating status data set with spatiotemporal correlation of the equipment; An optimization instruction unit (2), the optimization instruction unit (2) is used to input the structured operating state data set into a spatiotemporal convolutional network for feature learning; Through virtual simulation, the potential motion trajectory of the equipment in the future time domain is deduced, and a multi-objective optimization scheduling instruction set is generated in combination with a dynamic path planning algorithm to obtain intelligent scheduling strategy data including equipment obstacle avoidance paths, optimal energy consumption solutions, and task priorities; A digital twin unit (3), the digital twin unit (3) is used to convert the intelligent scheduling strategy data into a micro-operation instruction sequence executable by the device; push the instructions to the handling equipment control system in real time through the 5G network, and at the same time build a two-way feedback channel between the digital twin and the physical system, collect the actual response data of the device and update the digital twin model state; A data optimization unit (4) is used to construct a closed-loop evaluation system for the scheduling strategy, and to perform multi-dimensional comparative analysis on the actual response data of the equipment and the intelligent scheduling strategy data; a reinforcement learning algorithm is used to perform online optimization on the scheduling model parameters to obtain an improved scheduling strategy data packet with enhanced self-adaptation capability, and the data packet is synchronized to a cloud knowledge base to form a continuously evolving device management strategy library.

Images