Intelligent special vehicle real-time scheduling and operation cooperation method based on open-source gap

Abstract

The invention provides an intelligent special vehicle real-time scheduling and operation collaboration method based on an open source gap. The method comprises the following steps: constructing a working environment dynamic twinborn system, collecting data in real time, updating a three-dimensional digital twinborn model, and predicting an environment change trend; based on twin data, task decomposition, resource conflict detection and dynamic path planning are carried out through a multi-target intelligent decision engine; an adaptive collaborative network is constructed through an open source gap distributed soft bus, instructions are distributed, and autonomous negotiation and fault task redistribution between vehicles are supported; the operation efficiency is monitored in real time, the scheduling strategy is dynamically optimized, and experience is stored in a learning library to continuously improve the model and the algorithm. All the steps form closed-loop control, real-time sensing, intelligent decision making and cooperative execution of special vehicle dispatching and operation full links are achieved, and the operation efficiency, safety and system self-adaptive capacity are improved.

Description

A method for intelligent real-time dispatching and collaborative operation of special vehicles based on open-source HarmonyOS Technical Field

[0001] This invention relates to the field of intelligent transportation and operation scheduling technology, and in particular to an intelligent real-time scheduling and operation collaboration method for special vehicles based on the open-source HarmonyOS. Background Technology

[0002] In complex and dynamic operational environments such as modern logistics hubs, large-scale construction sites, emergency rescue operations, and industrial plants, the efficient dispatching and collaborative operation capabilities of special vehicles directly affect the safety, timeliness, and resource utilization efficiency of the overall task execution. Traditional special vehicle dispatching and management systems mainly rely on the Global Positioning System (GPS) for basic positioning services, combined with wireless communication networks for command transmission, and centralized monitoring and dispatching software for task allocation. While such systems can complete vehicle location tracking and simple command issuance, their technical bottlenecks become increasingly prominent when dealing with unstructured, highly dynamic, and real-time operational scenarios, making it difficult to support the refined management needs of modern special vehicle cluster operations.

[0003] In existing technologies, at the environmental perception and modeling level, systems generally use static electronic maps or Geographic Information System (GIS) data with long update cycles, integrating only limited sensor information from the vehicle itself (such as a single GPS module or low-resolution camera). This isolated perception architecture cannot effectively capture instantaneous dynamic elements of the work site, such as the movement trajectory of moving obstacles, subtle changes in terrain undulations, real-time occupancy status of resource points, and the collaborative behavior of other vehicles. Due to the lack of deep fusion capabilities for multi-source heterogeneous data (including environmental sensor networks, vehicle status monitors, and external meteorological information), the environmental model constructed by the system has significant spatiotemporal deviations from the actual physical scene. More importantly, existing technologies have not established a predictive mechanism for environmental evolution trends, and cannot extrapolate changes in obstacle distribution, fluctuations in resource availability, and traffic flow dynamics in future periods based on historical operation patterns and real-time sensor data. This leads to scheduling decisions relying on outdated or fragmented information, creating potential risks for subsequent operational conflicts.

[0004] At the decision optimization level, most systems employ pre-defined rule bases or simplified optimization algorithms (such as fixed path planning strategies), lacking dynamic adaptability in their decision logic. Faced with real-time changing scenarios such as sudden obstacles, emergency task insertions, unexpected vehicle malfunctions, or resource contention, the system struggles to make complex trade-offs between multiple objectives (including minimizing operation time, controlling energy consumption, ensuring safe distances, and maximizing resource utilization) within millisecond-level time windows. Furthermore, existing decision engines lack the ability to extrapolate in virtual environments, failing to pre-identify potential path intersections, overlapping work areas, or equipment contention risks. This leads to frequent vehicle waiting queues, path congestion, and resource conflicts during actual execution, significantly reducing overall collaborative efficiency.

[0005] In terms of collaborative execution and fault tolerance mechanisms, traditional architectures adopt a centralized control model of "central server unified command - terminal passive execution." Vehicle terminals merely act as instruction receiving units, lacking the ability for autonomous negotiation and dynamic strategy adjustment. When communication links are interrupted, central nodes fail, or individual vehicles malfunction, the system cannot trigger a self-organizing task redistribution process, leading to interruptions in collaborative operations or a sharp drop in efficiency. This rigid collaborative mechanism struggles to meet the high reliability and adaptability requirements of special vehicle clusters in harsh environments, posing significant risks, especially in critical scenarios such as emergency rescue.

[0006] Furthermore, the existing system's perception, decision-making, and execution modules are fragmented, failing to form a data-driven closed-loop optimization chain. Adjustments to scheduling strategies rely on manual offline analysis and experience summarization, unable to automatically feed real-time operational performance data (such as task completion progress and resource consumption indicators) back to the environmental model and decision-making algorithms for online iteration. This lack of a mechanism results in slow improvement in the system's intelligence level, making it difficult to adapt to the evolving needs of increasingly complex operational scenarios. Summary of the Invention

[0007] In view of this, the present invention aims to provide an intelligent real-time scheduling and operation collaboration method for special vehicles based on the open-source HarmonyOS, so as to solve or alleviate the technical problems existing in the prior art.

[0008] The technical solution of this invention is implemented as follows: A method for real-time scheduling and collaborative operation of intelligent special vehicles based on the open-source HarmonyOS framework, comprising: S1. Constructing a dynamic twin system for the operational environment based on the open-source HarmonyOS distributed perception framework, collecting real-time data on special vehicle status, operational environment parameters, and task progress, establishing and updating a three-dimensional digital twin model, and predicting environmental change trends based on historical and real-time data; S2. Based on the environmental data output by the dynamic twin system, decomposing complex operational tasks through a multi-objective intelligent decision engine, generating atomic operation sequences, and performing resource conflict detection and dynamic path planning in real-time; S3. Distributing decision instructions to each special vehicle through an adaptive collaborative network based on the open-source HarmonyOS distributed soft bus, supporting autonomous negotiation between vehicles and dynamic adjustment of operational strategies, and triggering a task reassignment mechanism when a vehicle malfunctions; S4. Real-time monitoring of vehicle operational efficiency and resource utilization, dynamically optimizing scheduling strategies based on evaluation results, and storing optimization experience in an experience learning library for continuous improvement of the digital twin model and decision algorithm; wherein steps S1 to S4 form a closed-loop control, realizing real-time perception, decision optimization, and collaborative execution of the entire special vehicle scheduling and operation chain.

[0009] The embodiments of this invention, employing the above technical solutions, possess the following advantages: By constructing a dynamic twin system for the operational environment, this invention achieves high-fidelity real-time environmental perception and trend prediction, solving the problem of perception lag. Through multi-objective intelligent decision-making and adaptive collaborative networks, it achieves efficient, conflict-free scheduling and rapid fault-tolerant recovery for complex tasks. Utilizing a closed-loop learning optimization mechanism, the system possesses continuous self-improvement capabilities, thereby comprehensively enhancing the efficiency, safety, and reliability of special vehicle cluster operations. Attached Figure Description

[0010] Figure 1 is a flowchart of the present invention; Figure 2 is a system architecture diagram of the present invention; Figure 3 is a detailed architecture diagram of the dynamic twin system of the present invention; Figure 4 is a flowchart of the multi-objective intelligent decision engine of the present invention; Figure 5 is a topology diagram of the adaptive collaborative network of the present invention; Figure 6 is a data flow diagram of the present invention; Figure 7 is a flowchart of the closed-loop learning and optimization of the present invention; Figure 8 is a system deployment and network topology diagram of the present invention. Detailed Implementation

[0011] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0012] It should be noted that terms such as "first," "second," "symmetric," and "array" are used only to distinguish between descriptive and positional descriptions and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first," "symmetric," etc., may explicitly or implicitly include one or more of that feature; similarly, when the quantity of certain features is not limited by words such as "two" or "three," it should be noted that such features also explicitly or implicitly include one or more features. In this invention, unless otherwise explicitly specified and limited, terms such as "installation," "connection," and "fixing" should be interpreted broadly; for example, they can refer to a fixed connection, a detachable connection, or an integral molding; they can refer to a mechanical connection, a direct connection, welding, or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the accompanying drawings and specific circumstances.

[0013] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0014] As shown in Figures 1-8, this invention proposes an intelligent real-time scheduling and operation collaboration method for special vehicles based on the open-source HarmonyOS framework, comprising: S1. Constructing a dynamic twin system for the operational environment based on the open-source HarmonyOS distributed perception framework, collecting real-time data on special vehicle status, operational environment parameters, and task progress, establishing and updating a three-dimensional digital twin model, and predicting environmental change trends based on historical and real-time data; S2. Based on the environmental data output by the dynamic twin system, decomposing complex operational tasks through a multi-objective intelligent decision engine, generating atomic operation sequences, and performing resource conflict detection and dynamic path planning in real time; S3. Distributing decision instructions to each special vehicle through an adaptive collaborative network based on the open-source HarmonyOS distributed soft bus, supporting vehicle... The system involves autonomous negotiation and dynamic adjustment of operational strategies between vehicles, triggering a task reassignment mechanism in case of vehicle malfunction; S4. Real-time monitoring of vehicle operational efficiency and resource utilization, dynamic optimization of scheduling strategies based on evaluation results, and storage of optimization experience in an experience learning library for continuous improvement of the digital twin model and decision-making algorithm; wherein, steps S1 to S4 form a closed-loop control, realizing real-time perception, decision optimization, and collaborative execution of the entire special vehicle scheduling and operation chain. This invention addresses the problems of insufficient real-time and predictive environmental perception, weak dynamic adaptability of decision optimization, lack of flexibility and robustness of collaborative mechanisms, and lack of closed-loop learning optimization capabilities in special vehicle scheduling management, and proposes an intelligent real-time scheduling and operation collaboration method for special vehicles based on the open-source HarmonyOS. The dynamic twin system of the operational environment is a real-time virtual mapping system built by integrating multi-source heterogeneous data based on a distributed architecture. Through a multi-node data synchronization mechanism based on a message queue telemetry transmission protocol, low-latency acquisition and fusion of environmental data are achieved, and a distributed database is used to store time-series data streams. Combined with a lightweight data compression algorithm, the network transmission load is reduced to achieve high-fidelity synchronization between the physical environment and the virtual model. Furthermore, the update mechanism of the 3D digital twin model is designed as a dynamic adjustment process in response to real-time data deviations. In practical applications, model reconstruction can be achieved through a local area refresh strategy based on grid partitioning. For example, when the difference between sensor data and model state exceeds a preset threshold within a specific geographic grid, only the affected grid is incrementally updated, rather than the entire model is reconstructed, thereby ensuring efficient utilization of computing resources. A multi-objective intelligent decision engine is used to decompose complex tasks into executable units. As a preferred implementation, task decomposition can be accomplished through semantic parsing technology based on a rule engine. For example, matching operation instructions according to a predefined task type library generates standardized atomic operation sequences, primarily to support parallel processing and resource allocation of tasks. Resource conflict detection is implemented as a real-time identification process for potential interference between vehicles. Specifically, it can be implemented using a path cross-checking algorithm based on time windows. For example, by discretizing the time axis and calculating the spatial overlap probability of vehicle trajectories at discrete time points, it can determine whether there is a risk of resource contention.The dynamic path planner generates obstacle avoidance trajectories using online optimization methods. In practical applications, path replanning can be achieved through an improved Dijkstra algorithm combined with a heuristic search strategy. For example, weight factors can be dynamically adjusted in a grid map to balance path length and safety, primarily to meet the need for immediate response to environmental changes. The adaptive cooperative network supports instruction distribution and policy negotiation among vehicles. Furthermore, autonomous negotiation among vehicles can be achieved through a distributed decision-making protocol based on a voting mechanism. For instance, each vehicle node anonymously scores the task allocation scheme and aggregates the majority opinion, thereby improving the decentralization of the collaborative process. The task reassignment mechanism is triggered in fault scenarios. Specifically, a priority queue-based polling allocation strategy can be used to transfer tasks. For example, unfinished tasks are sorted by urgency and then allocated to available neighboring vehicles in sequence to ensure the system's continued operation under local anomalies. The experience learning library stores scheduling optimization experience. As a preferred implementation, knowledge accumulation can be achieved through rule extraction methods based on decision trees, such as summarizing condition-action pairs from historical scheduling records and generating reusable policy templates, thereby supporting iterative improvement of digital twin models and decision algorithms. Therefore, this embodiment organically connects real-time perception, intelligent decision-making, flexible collaboration and learning optimization through closed-loop control of steps S1 to S4, enabling special vehicle dispatching and operation to achieve continuous adaptive adjustment in a dynamic environment, thereby effectively improving the system's real-time response capability and overall collaborative efficiency.

[0015] During implementation, a dynamic digital twin system for the operational environment, built upon the open-source HarmonyOS distributed perception framework, was used to collect real-time data on special vehicle status, operational environment parameters, and task progress, thereby establishing and updating a three-dimensional digital twin model. Environmental change trends were predicted based on historical and real-time data, providing a forward-looking basis for subsequent decision-making. A multi-objective intelligent decision engine was applied to decompose complex operational tasks, generating a sequence of atomic operations that could be executed in parallel, and performing resource conflict detection and dynamic path planning in real time. Furthermore, through an adaptive collaborative network based on the open-source HarmonyOS distributed soft bus, decision instructions were distributed to each special vehicle, supporting autonomous negotiation between vehicles and dynamic adjustment of operational strategies. A task reassignment mechanism was triggered in the event of vehicle failure. After real-time monitoring of vehicle operational efficiency and resource utilization, the scheduling strategy was dynamically optimized, and optimization experience was stored in an experience learning library for continuous improvement of the digital twin model and decision-making algorithm. Therefore, steps S1 to S4 form a closed-loop control, realizing real-time perception, decision optimization, and collaborative execution across the entire special vehicle scheduling and operational chain.

[0016] For example, in the special vehicle dispatching scenario of a large logistics park, a dynamic digital twin system is built to collect real-time location information of forklifts and heavy transport vehicles, warehouse rack distribution, and cargo loading and unloading status data. The three-dimensional digital twin model is dynamically updated to reflect the actual working environment and predict congestion trends in cargo handling paths. A multi-objective intelligent decision engine is used to decompose large warehousing tasks into atomic operation sequences such as cargo grabbing, transportation, and stacking, and to detect path intersections and overlaps between vehicles and work areas in real time and generate obstacle avoidance paths online. Decision instructions are distributed to each vehicle through a distributed soft bus. Forklifts and transport workshops autonomously negotiate and adjust the work sequence and speed. If a forklift fails suddenly and fails, the unfinished task is automatically reassigned to other available vehicles. Operational efficiency is monitored in real time, dispatching strategies are dynamically adjusted, and optimization experience is stored for subsequent model iterations.

[0017] This technical solution effectively addresses the issues of insufficient real-time and predictive capabilities in environmental perception through the aforementioned mechanisms, enabling scheduling decisions to be based on a high-fidelity synchronous environmental model. It enhances the dynamic adaptability of decision optimization, achieving online trade-offs between multiple objectives such as time, energy consumption, and safety. It improves the flexibility and robustness of the collaborative mechanism, supporting autonomous vehicle negotiation and rapid fault-tolerant recovery. Furthermore, it achieves continuous optimization of scheduling strategies through closed-loop learning, thereby enhancing the overall efficiency and reliability of special vehicle cluster operations.

[0018] In practical applications, in some embodiments of the present invention, step S1 is proposed to construct a dynamic twin system of the working environment and predict the trend of environmental changes. However, in its implementation, the prediction method lacks a specific implementation mechanism, which makes it impossible to effectively quantify the uncertainty of environmental evolution, resulting in a lack of probabilistic basis for intelligent decision-making and affecting the real-time performance and accuracy of scheduling and coordination.

[0019] To address this, this invention further proposes a method based on the open-source HarmonyOS distributed perception framework to access a multi-source sensor network, enabling real-time acquisition of vehicle positioning information, environmental obstacle distribution, terrain data, and operational resource status. This multi-source data is used to drive a 3D digital twin model to dynamically reconstruct the operational scenario. An environmental change prediction algorithm is employed, fusing historical operational patterns with real-time sensor information to predict obstacle changes, resource occupancy status, and traffic flow evolution trends within the operational area. The environmental change prediction algorithm uses a temporal neural network model, with inputs including historical operational logs, real-time sensor data, and weather influencing factors, and outputting a probability distribution map of the operational environment's state over a future period. The prediction process is expressed as follows:

[0020] Among them, the open-source HarmonyOS distributed perception framework is a distributed data acquisition architecture based on the open-source HarmonyOS operating system, used for multi-node collaborative sensor data aggregation technology. It achieves real-time sharing and low-latency transmission of environmental information across devices. The multi-source sensor network can be understood as a network system composed of heterogeneous sensors, specifically including combinations of devices such as LiDAR arrays, visual camera clusters, and inertial measurement units. It eliminates blind spots of single data sources by providing multi-dimensional environmental perception data. The three-dimensional digital twin model is a virtual mapping entity of the physical operation scene, used for dynamic mesh reconstruction technology or parametric geometric modeling methods based on point cloud data. It ensures that the virtual model is consistent with the geometric structure and state attributes of the physical environment. The environmental change prediction algorithm is a computational logic used to quantify the dynamic evolution of the environment and transform it into a computable probability distribution. It is implemented through a time-series model enhanced by a gated recurrent unit network or attention mechanism. The time-series neural network model can be understood as a deep learning architecture for processing sequential data. Specifically, it can adopt a cascaded structure of convolutional neural networks and recurrent neural networks. It extracts the time-series dependent features of environmental evolution from historical data and integrates external influencing factors.

[0021] Specifically, this invention utilizes the open-source HarmonyOS distributed perception framework to access a multi-source sensor network in real time, acquiring multi-source information such as vehicle positioning, obstacle distribution, and terrain data. This data drives the dynamic reconstruction of a 3D digital twin model, automatically updating the geometric structure and state attributes of the virtual scene as the physical environment changes. Simultaneously, an environmental change prediction algorithm inputs historical operation log feature vectors, real-time sensor data vectors, and weather influencing factor vectors into a temporal neural network model. Through the nonlinear mapping of the neural network, it calculates the probability distribution of future environmental states. This probability distribution quantifies the uncertainty of environmental evolution, such as obstacle movement and resource occupancy, enabling the decision engine to conduct risk assessment and path optimization based on the probability of different states occurring. This forms a complete closed loop from data acquisition and model reconstruction to probability prediction, ensuring that the prediction results provide a reliable probabilistic basis for subsequent multi-objective decision-making.

[0022] As a preferred embodiment, the present invention is implemented as follows: In the scenario of special vehicle scheduling in large ports, a multi-source sensor network composed of vehicle-mounted LiDAR, high-definition cameras, and Beidou positioning modules is accessed based on the open-source HarmonyOS distributed perception framework to obtain real-time information on container truck location, yard obstacle distribution, and ground slope; these data are used to drive a three-dimensional digital twin model, and the virtual mapping of the port operation scenario is dynamically updated using virtual reconstruction technology based on the Unity engine; the environmental change prediction algorithm uses a long short-term memory network as a temporal neural network model, with inputs including historical loading and unloading operation log feature vectors, current sensor data vectors, and weather forecast influencing factor vectors, and outputs a probability distribution map of the state of the yard passage during future operation periods, which is used to predict the distribution trend of temporary obstacles that may appear on the operation path of mobile cranes.

[0023] Through the above-mentioned solution, the present invention can effectively quantify the uncertainty of the dynamic evolution of the working environment, provide a computable probabilistic basis for the intelligent decision engine, thereby improving the real-time response capability and decision accuracy of the special vehicle dispatching system in complex and ever-changing scenarios, and avoiding path conflicts or resource contention caused by environmental prediction deviations.

[0024] In some of the embodiments of the present invention described above, a three-dimensional digital twin model is proposed for real-time synchronization of the working environment. However, in its implementation, the model update lacks a quantitative deviation threshold and a local reconstruction mechanism, resulting in untimely or excessive updates, which affects the synchronization accuracy between the model and the physical environment, and consequently makes scheduling decisions based on inaccurate environmental data.

[0025] To address this, the present invention further proposes an update mechanism for the three-dimensional digital twin model: when the deviation between real-time sensor data and the model state exceeds a preset threshold... When this occurs, a local model reconstruction is triggered; the deviation is calculated as follows:

[0026] Specifically, the update mechanism of the 3D digital twin model is a dynamic update strategy based on a deviation threshold, which is used to compare the model state with sensor data in real time. This reduces computational resource consumption by avoiding unnecessary full model reconstruction; wherein, the preset threshold... This is a configurable threshold value, determined based on empirical values ​​or historical data analysis, used to balance update frequency and resource overhead. In practical applications, deviation calculation... To utilize the Euclidean norm to calculate the overall difference between the model's state vector and the sensor data vector for hardware acceleration of vector operations, it provides comprehensive quantitative indicators of environmental changes. Furthermore, the model's local reconstruction mechanism updates the model only for regions where the deviation exceeds the limit, which is implemented using a spatial partitioning strategy or a dynamic labeling method for key feature points. This reduces computational complexity and ensures the system's real-time performance.

[0027] Specifically, the present invention achieves this through continuous monitoring. Time-Temporal Digital Twin Model State Vector With real-time sensor data vectors Calculate the Euclidean norm deviation between the two. ;when Exceeding the preset threshold In such cases, the system automatically triggers a local model reconstruction operation targeting only the area where the deviation exceeds the limit, rather than performing a full model reconstruction. This mechanism, based on accurate judgment of deviation quantification, ensures timely updates to key areas when significant environmental changes occur, while avoiding frequent updates caused by minor fluctuations. This allows for high-frequency synchronization checks under limited computing resource constraints, achieving high-fidelity dynamic matching between the model and the physical environment, and providing a reliable data foundation for subsequent environmental prediction and scheduling decisions.

[0028] As a specific implementation method, the present invention is implemented in a special vehicle operation scenario as follows: The system divides the operation area into multiple logical grid units, each unit corresponding to a local area of ​​a digital twin model. When the distributed perception framework detects a change in the position of an obstacle or terrain data within a grid, it calculates the Euclidean norm deviation between the model state vector corresponding to that grid and the sensor data vector; if the deviation value exceeds a preset threshold... If the model is reconstructed only for that grid cell, updating the obstacle distribution or terrain parameters, while other grid cells remain unchanged. This process is executed by edge computing nodes in the open-source HarmonyOS distributed perception framework, utilizing hardware acceleration units to perform vector operations, ensuring low-latency response for deviation calculation and local reconstruction.

[0029] Through the above solution, the present invention achieves accurate on-demand updates of the three-dimensional digital twin model, effectively avoiding environmental perception lag caused by untimely updates and waste of computing resources caused by excessive updates, ensuring high-fidelity synchronization between the digital twin model and the physical operating environment, thereby providing real-time and accurate environmental data support for special vehicle dispatching decisions.

[0030] In some of the embodiments of the present invention, a multi-objective intelligent decision engine is proposed for resource conflict detection and dynamic path planning. However, in its implementation, the conflict detection mechanism lacks the ability to accurately quantify vehicle trajectories in the spatiotemporal dimension, and cannot capture potential path intersections, overlapping work areas, and equipment contention risks between vehicles in the dynamic environment in real time. This results in blind spots in decision inference based on the digital twin model, which may lead to collision hazards or reduced collaborative efficiency in actual operations.

[0031] To address this, the present invention further proposes a multi-objective intelligent decision engine in step S2, comprising: a task decomposer, used to parse upper-level operation instructions into a sequence of atomic operations that can be executed in parallel; a resource conflict detection model, based on a digital twin model, to determine in real time the path intersections, overlapping work areas, and equipment contention between vehicles; a dynamic path planner, which combines environmental prediction results and conflict detection information to generate obstacle avoidance paths and multi-vehicle collaborative operation trajectories online; the resource conflict detection model makes judgments by calculating the spatial-temporal intersection of the vehicle's predicted trajectories, for any two vehicles... and During the time period The conditions for a memory conflict are:

[0032]

[0033] The task decomposer is a processing module that breaks down complex task instructions into basic executable units. It is implemented based on rule engines or semantic parsing technology and supports parallel processing and conflict detection by ensuring fine task granularity. The resource conflict detection model is an algorithmic component for real-time risk assessment based on a digital twin environment. It may include geometric spatial analysis or probabilistic prediction methods and quantifies the potential intersection of vehicle trajectories in the spatiotemporal dimensions. The dynamic path planner is a decision-making unit that adjusts the driving path in real time according to environmental changes. It is implemented using heuristic search algorithms or graph theory optimization methods and generates safe and efficient multi-vehicle cooperative trajectories.

[0034] Specifically, this invention uses a task decomposer to parse upper-level job instructions into atomic operation sequences. The resource conflict detection model, based on a digital twin model, calculates the spatial-temporal intersection of the predicted trajectories of each vehicle in real time, and then uses a formula... The system accurately identifies potential conflict points. Subsequently, the dynamic path planner combines environmental prediction results with conflict detection information to generate obstacle avoidance paths and multi-vehicle collaborative operation trajectories online, thus forming a closed-loop decision-making process from task analysis to conflict avoidance to path optimization, ensuring the real-time adaptability and safety of scheduling instructions in dynamic environments.

[0035] As a specific implementation method, the present invention is implemented as follows: The task decomposer, based on the parsing module of the Drools rule engine, decomposes the crane hoisting task into atomic operations such as movement, positioning, and lifting; the resource conflict detection model calculates the real-time position distance between two engineering vehicles in the work area, and determines that a conflict exists when the Euclidean distance is less than the safety threshold; the dynamic path planner uses the Dijkstra algorithm combined with conflict point information to replan the vehicle driving path to avoid potential collision areas.

[0036] Through the above technical solution, the present invention achieves accurate quantitative detection of vehicle trajectory in the spatiotemporal dimension, effectively avoids blind spots in decision-making and deduction, significantly reduces the risk of collision in actual operation, and improves the overall efficiency of multi-vehicle collaborative operation.

[0037] In some of the embodiments of the present invention, a dynamic path planner is proposed to generate obstacle avoidance paths and multi-vehicle collaborative operation trajectories online by combining environmental prediction results and conflict detection information. However, in its implementation, the path planning only relies on basic obstacle avoidance logic and fails to effectively integrate multi-dimensional indicators such as path length, operation time, energy consumption and conflict avoidance for dynamic balancing. This results in the generated path potentially causing redundant detours, extended operation time, excessive energy consumption, or insufficient avoidance of potential conflict risks in actual execution, thereby affecting the overall operation efficiency and safety of the special vehicle cluster.

[0038] To address this, the present invention further proposes a dynamic path planner that employs a reinforcement learning algorithm for online optimization, with its reward function... Taking into account path length, operation time, energy consumption, and conflict avoidance indicators, the specific definition is:

[0039] Among them, reinforcement learning algorithms are machine learning methods that autonomously optimize decision-making strategies through interaction with the environment, and are used in deep learning. Implementation using algorithms such as network, policy gradient, or proximal policy optimization, which enables the path planning process to dynamically adapt to real-time environmental changes; reward function. This can be understood as a comprehensive evaluation index for quantifying the quality of path planning. It forms a unified optimization framework by weighting and combining multi-dimensional objective functions. It transforms discrete indicators such as path length and operation time into calculable continuous optimization objectives. The conflict avoidance reward term, Caution, is a dynamic indicator for evaluating path safety. It is implemented based on the calculation of the spatiotemporal intersection between vehicles or the determination of safe distance thresholds, by actively identifying and avoiding potential collision risks. Weighting coefficients to It can be dynamically adjusted to adapt to the needs of different work scenarios, such as emphasizing time weight configuration in emergency tasks, which achieves flexible priority allocation through multi-objective optimization.

[0040] Specifically, this invention receives environmental prediction data and resource conflict detection results output by a digital twin model through a dynamic path planner, and initiates a reinforcement learning algorithm in the digital twin environment to perform multiple rounds of decision-making simulations; during the simulation process, the reward function... Real-time evaluation of the overall performance of candidate paths, including negative terms. The guiding algorithm minimizes path length, operation time, and energy consumption, while the Cadoid term strengthens safety constraints by rewarding successful conflict avoidance and penalizing conflict occurrence. Based on the reward feedback signal, the reinforcement learning algorithm iteratively optimizes the path strategy parameters, and finally generates a Pareto optimal path set that meets the safety distance threshold and multi-objective constraints. After completing the path feasibility verification in the digital twin environment, the path is output to the execution layer.

[0041] As a specific implementation method, the present invention is implemented as follows: In the scenario of special vehicle dispatching in a large logistics park, the dynamic path planner is implemented using a deep deterministic policy gradient algorithm as a reinforcement learning implementation carrier, and the weight coefficients in the reward function are dynamically configured according to the task type; for example, when handling emergency dispatching tasks for fire trucks, the system automatically increases the weight coefficients. The planner prioritizes timeliness by reducing w3 (energy consumption weight) to ensure timeliness. In a digital twin environment, the planner generates multiple candidate paths based on real-time obstacle distribution data and selects the optimal path that avoids moving obstacles and meets the time window requirements through a reward function, which is then used by special vehicles to perform collaborative operations.

[0042] Through the above solution, the present invention effectively solves the problem of insufficient balancing of multi-dimensional indicators. The generated path significantly reduces the risk of extended operation time and excessive energy consumption while avoiding redundant detours. Furthermore, the conflict avoidance mechanism improves the collaborative safety of special vehicle clusters in complex operating environments, thereby optimizing the overall operational efficiency.

[0043] In complex operational scenarios such as modern logistics, engineering construction, and emergency rescue, the scheduling and collaborative operation of special vehicle clusters face severe challenges. Traditional systems mostly adopt a centralized control mode, with weak autonomous negotiation capabilities between vehicles. The lack of specific implementation mechanisms leads to low negotiation efficiency, untimely fault recovery, and a lack of self-organization capabilities, making it impossible to meet the high reliability and highly adaptive requirements of special vehicle cluster operations.

[0044] In response, this invention further proposes the following specific implementation method for the adaptive cooperative network in step S3: a vehicle group communication channel is constructed through the open-source HarmonyOS distributed soft bus to support low-latency, high-reliability information broadcasting and intra-group negotiation; an adaptive cooperative protocol based on a state machine is designed so that vehicles can autonomously adjust their work sequence, speed, or division of labor strategy according to real-time environmental information and group status; when a vehicle malfunctions or communication is interrupted, other nodes in the cooperative network are notified through an anomaly propagation mechanism, and the decision engine reassigns the unfinished tasks of the vehicle.

[0045] In practical applications, the open-source HarmonyOS distributed soft bus is a distributed device interconnection framework provided by the open-source HarmonyOS operating system. It is used for message queue-based communication mechanisms or point-to-point network topologies. By establishing a decentralized communication foundation, it avoids global paralysis caused by single-point failures and ensures the communication reliability of vehicle groups in complex electromagnetic environments. The state machine-based adaptive coordination protocol can be understood as a logical model defining vehicle behavior states and transition conditions. It is implemented using finite state machines combined with event-driven mechanisms or hierarchical state tree structures, enabling vehicles to break free from preset rules and make autonomous decisions based on real-time dynamic data. Specifically, the anomaly propagation mechanism is a distributed diffusion mechanism for fault information, implemented using heartbeat detection combined with event broadcast protocols or distributed consensus algorithms. It avoids delays caused by manual intervention by rapidly triggering networked fault responses.

[0046] Specifically, this invention first constructs a vehicle group communication channel using the open-source HarmonyOS distributed soft bus. This channel serves as the underlying support layer, leveraging the characteristics of a distributed architecture to achieve low-latency information synchronization among multiple nodes. Building upon this, an adaptive coordination protocol based on a state machine serves as the decision execution layer. This layer receives environmental perception data and group state information in real time, dynamically calculates the optimal operating strategy, and drives vehicle behavior adjustments. When the anomaly propagation mechanism detects a vehicle malfunction or communication interruption, it immediately activates a networked notification process, broadcasting the fault information to all nodes within the coordination network, while simultaneously triggering the decision engine to initiate task redistribution logic. These three layers form a closed-loop workflow: the communication channel ensures the real-time and reliable flow of information, the coordination protocol enables dynamic strategy optimization, and the anomaly mechanism ensures rapid system recovery in fault scenarios. The synergistic effect of these three mechanisms endows the special vehicle cluster with decentralized self-organizing capabilities, effectively addressing uncertainties in dynamic operating environments.

[0047] As a specific implementation method, the present invention is implemented as follows: A distributed soft bus instance is deployed on the open-source HarmonyOS operating system. The vehicle terminal establishes a group communication channel through this instance to support the reliable transmission of broadcast messages. The vehicle state machine is preset with four core states: "standby", "operating", "negotiating", and "fault". The state transition conditions are dynamically determined based on the real-time received obstacle distribution data and the location information of nearby vehicles. When a vehicle's heartbeat signal times out three times in a row, its fault status is broadcast to the network through the soft bus. After receiving the broadcast, nearby vehicles actively request the decision engine to reassign tasks. The task reassignment process is completed by the decision engine based on the vehicle's current load and location information.

[0048] Through the above technical solutions, the present invention significantly improves the negotiation efficiency between vehicles, shortens the response time for adjusting the operation strategy to the second level, and realizes seamless task switching in fault scenarios, avoiding operation interruption caused by single point of failure, and ensuring the continuous and reliable operation capability of special vehicle clusters in complex dynamic environments.

[0049] Specifically, in some of the embodiments of the present invention, an adaptive collaborative protocol is proposed to realize autonomous negotiation and dynamic adjustment of operation strategies among vehicles. However, in the implementation process, the dynamic allocation of vehicle roles lacks a quantitative evaluation mechanism, which may lead to task allocation decisions relying only on simple rules or local information, failing to comprehensively consider vehicle capability matching degree and spatial distance factors, resulting in reduced task execution efficiency, delayed resource response and insufficient overall collaborative operation efficiency.

[0050] In response, this invention further proposes that in the aforementioned adaptive cooperative protocol, the dynamic allocation of vehicle roles is achieved through a benefit function. The function calculates the expected benefits of vehicle k assuming role ri:

[0051] Among them, the benefit function To quantify the comprehensive benefits of a vehicle in a specific role, a linear weighted combination method is used. By integrating capability matching degree and spatial distance factors, a unified decision-making basis is formed. It provides an objective and quantitative evaluation standard for role allocation, avoiding allocation bias caused by subjective judgment. It measures the degree of matching between vehicle capabilities and role requirements, and is used for similarity calculation based on feature vectors. For example, by comparing the vector inner product or cosine similarity between vehicle technical parameters and role task requirements, it improves the reliability of task execution by ensuring that tasks are assigned to vehicles with the corresponding execution capabilities. This is the spatial distance between the vehicle's current location and the character's target location, used for Euclidean distance or path distance calculations. It reflects the spatial cost of the vehicle's response to the task; the closer the distance, the faster the response. The maximum allowable allocation distance threshold set for the system is dynamically adjusted based on the scope of the work area or the urgency of the task. By limiting the geographical range of task allocation, it avoids response delays caused by long-distance scheduling. α and β are the weight coefficients of each item in the benefit function, used for strategies that are fixed or dynamically adjusted according to the work scenario. By balancing the importance of capability matching degree and spatial distance factors, it adapts to the priority changes of different task requirements.

[0052] Specifically, this invention constructs a benefit function. As a core decision-making mechanism, vehicle capability matching degree With normalized distance factor ( The system performs weighted fusion to form a comprehensive benefit evaluation index. During the dynamic allocation of vehicle roles, the system calculates the expected benefit value of each available vehicle for each role in real time, and compares the results. The system automatically selects the vehicle with the best overall efficiency to assume the corresponding role based on the size of the vehicle. This quantitative evaluation mechanism enables task allocation decisions to simultaneously consider vehicle execution capabilities and spatial response efficiency, avoiding the limitations of traditional allocation methods that rely solely on a single factor or empirical rules. By dynamically adjusting the weighting coefficients α and β, the system can flexibly change the priority of capability matching and distance factors according to changes in the operational scenario, ensuring adaptive optimal matching of resources and tasks under different working conditions, thereby effectively improving the real-time performance and resource utilization of special vehicle cluster collaborative operations.

[0053] As a preferred embodiment, the present invention is implemented as follows: In a large port container loading and unloading operation scenario, when a new lifting task role needs to be assigned, the system assigns roles based on a benefit function. Assume there are currently three available vehicles: one with high load capacity but a long distance, one with moderate capacity and a moderate distance, and one with low capacity but the closest distance. The system calculates the benefit value of each vehicle for the lifting role: for the high load capacity vehicle, High value but ( Vehicles with low (low) values; vehicles with moderate capabilities have both indicators at a moderate level; vehicles with very low capabilities Low value but ( The α value is high. Based on current operational requirements, if task reliability is prioritized (larger α), a vehicle with high load capacity might be selected; if response speed is prioritized (larger β), a vehicle with lower capacity might be selected; in normal mode (α and β balanced), the system might select a vehicle with moderate capacity as the allocation scheme with the best overall benefit. This example demonstrates how the benefit function dynamically weighs different factors according to the actual scenario to achieve a scientific and reasonable task allocation.

[0054] Through the above solution, the present invention effectively solves the problem of blindness in the dynamic allocation of vehicle roles, realizes the scientific and quantitative nature of task allocation decisions, significantly improves the task execution efficiency of special vehicle clusters, reduces resource response delay, and optimizes the overall efficiency of collaborative operations.

[0055] In some embodiments of the present invention, step S4 is proposed to monitor vehicle operation efficiency and resource utilization in real time and dynamically optimize scheduling strategies. However, in the implementation process, efficiency monitoring lacks real-time quantification of fine-grained indicators for each vehicle, dynamic optimization fails to effectively integrate multi-objective trade-offs and real-time constraints, and the experience learning mechanism only realizes data storage without actively mining excellent cases for algorithm iteration. This results in scheduling strategy adjustments lagging behind environmental changes, difficulty in balancing multi-objective conflicts, and insufficient closed-loop optimization capabilities of the system, making it impossible to continuously improve the collaborative efficiency and robustness of special vehicle cluster operations.

[0056] To address this, the present invention further proposes step S4, which includes: calculating the task completion rate, resource utilization efficiency, and energy consumption indicators of each vehicle in real time through the task performance monitoring module; adjusting the vehicle scheduling frequency, path weight, or task allocation strategy based on the performance evaluation results and the current environmental state using a dynamic scheduling optimizer; recording each scheduling decision and its execution effect in an experience learning base; extracting excellent task cases through data mining for iterative optimization of the path planning algorithm and conflict detection model; the dynamic scheduling optimizer employs a multi-objective evolutionary algorithm, whose optimization objective function F is:

[0057]

[0058] In practical applications, the operation performance monitoring module is a computing unit used to quantify the operational performance of each special vehicle in real time. Implemented using a distributed data processing architecture, it performs calculations based on real-time vehicle status data streams obtained through the open-source HarmonyOS distributed perception framework. By achieving fine-grained monitoring of individual operation performance, it avoids the shortcomings of traditional monitoring where individual problems are masked by group data. The dynamic scheduling optimizer is a decision engine that dynamically adjusts scheduling strategies based on real-time performance evaluation and environmental conditions. It combines multi-objective optimization algorithms with environmental prediction models, overcoming the response lag problem in fixed-rule optimization by ensuring that scheduling strategy adjustments closely align with the dynamic needs of the current operational scenario. Specifically, the multi-objective evolutionary algorithm is an intelligent optimization method for handling vectorized optimization objectives and constraints. Implemented using classic multi-objective evolutionary algorithms such as NSGA-II or MOEA / D, it automatically balances conflicting objectives such as time and energy consumption by generating a Pareto optimal solution set in parallel while meeting real-time constraints. In addition, the experience learning library is a historical database that stores the scheduling decision chain and its execution effect. It is used to implement a case-based reasoning system. By actively identifying excellent job cases with high completion rate and low energy consumption through data mining technology, it achieves a closed-loop improvement from historical experience to algorithm performance.

[0059] This invention uses a work performance monitoring module to acquire real-time work completion rate, resource utilization efficiency, and energy consumption indicators for each vehicle. These fine-grained quantifiable data, along with environmental conditions predicted by a digital twin model, serve as input to a dynamic scheduling optimizer. Based on a multi-objective evolutionary algorithm, the dynamic scheduling optimizer vectorizes scheduling schemes and, under constraints such as resource capacity and time windows, simultaneously optimizes multiple objective functions, including total time and total energy consumption, generating a Pareto optimal scheme set. The decision-making system selects the most suitable scheduling strategy for the current scenario from this set for execution. Simultaneously, each scheduling decision and its execution effect are fully recorded in an experience learning database. Data mining techniques are used to extract feature patterns from excellent cases for iterative optimization of the path planning algorithm and conflict detection model. This closed-loop mechanism enables the system to dynamically generate multi-objective balanced scheduling strategies based on real-time performance feedback and environmental changes, and continuously enhances decision-making intelligence through experience-driven algorithm iteration.

[0060] As a specific implementation method, in the special vehicle scheduling scenario of a large logistics park, the operation efficiency monitoring module continuously receives real-time status data from each forklift and heavy truck, calculating the cargo loading and unloading completion rate, battery consumption efficiency, and workload per unit time for each vehicle. When the completion rate of a forklift is consistently below a threshold, the dynamic scheduling optimizer, combined with the current cargo distribution in the warehouse and aisle congestion predictions, adjusts the scheduling frequency and task allocation strategy for that forklift. The optimizer employs a multi-objective evolutionary algorithm to balance path length and energy consumption targets while ensuring on-time delivery of goods, generating multiple feasible solutions. After the system selects the optimal solution for execution, the decision parameters and actual execution results are stored in an experience learning library. In subsequent similar scenarios, the system quickly invokes validated scheduling strategies by matching historical best practices, significantly improving overall operational efficiency.

[0061] Through the above technical solutions, this invention achieves fine-grained real-time monitoring of the operational efficiency of special vehicles, enabling accurate identification of vehicles with performance bottlenecks. The dynamic scheduling optimizer generates multi-objective balanced scheduling strategies under real-time constraints, effectively solving the problem of balancing conflicting multi-objectives. The experience learning library actively mines excellent cases to achieve iterative algorithm optimization, enhancing the system's closed-loop learning capability. This solution significantly improves the collaborative efficiency and operational robustness of special vehicle clusters in dynamic and complex operating environments.

[0062] In practical applications, experience learning libraries are used to store scheduling decisions and execution results. However, experience learning libraries lack an efficient case matching mechanism, which makes it impossible to quickly retrieve historically excellent scheduling strategies similar to the current scenario, affecting the optimization speed and accuracy of real-time decisions.

[0063] To address this, the present invention further proposes an experience learning base that employs a case-based reasoning method, with its case matching similarity... The calculation formula is:

[0064] Among them, the case-based reasoning method is a technical strategy for solving new problems by retrieving and reusing historical cases. It is used for the process implementation of case storage, similarity calculation, and strategy adjustment, improving decision generation efficiency by avoiding redundant optimization calculations. The case matching similarity formula is a mathematical model used to quantify the similarity between query cases and historical cases. It can be implemented based on multi-feature weighted aggregation, comprehensively evaluating the matching level of features across different dimensions. Specifically, the weights... Parameters reflecting the relative importance of each feature in scheduling decisions can be set based on expert experience or historical data analysis results, focusing the matching process on key influencing factors. Local similarity function. This is a similarity measurement mechanism designed for specific data types. It is used to implement methods such as Euclidean distance transformation, cosine similarity, or category matching. By adapting to the characteristics of data with different features, such as numerical and categorical types, it ensures the accuracy and adaptability of similarity calculation.

[0065] Specifically, when the system needs to generate scheduling decisions, the experience learning base first obtains the feature data of the current query case, and then, for each case in the historical case base, calculates its case matching similarity with the query case. This calculation process is performed independently for each feature dimension: selecting the corresponding local similarity function based on the feature data type. Calculate the values ​​of the query case and historical cases on this feature. and The similarity between them, multiplied by a weight. Perform weighted summation of all features to obtain the overall similarity. The system sorts historical cases based on similarity scores, selects the most similar case as a reference, and directly reuses its scheduling strategy or makes fine adjustments, thereby avoiding the redundant process of optimizing from scratch and realizing rapid decision generation.

[0066] As a specific implementation method, in the scenario of special vehicle dispatching in a large logistics park, when a fire truck needs to respond to a new fire alarm, the system can extract the current task characteristics, including fire alarm type, location coordinates, available resource status, and environmental parameters, to form a query case. The experience learning base can traverse the historical case library, and for each historical case... Calculate case matching similarity For fire alarm type features, a categorical local similarity function is used; for location coordinate features, a local similarity function based on Euclidean distance transformation is used. Based on the calculation results, the system can retrieve similar historical cases, such as the dispatch strategy for a warehouse fire, and directly apply its vehicle routing and resource allocation schemes to quickly generate execution instructions for the current task.

[0067] Through the above solution, the present invention can achieve efficient matching of historical cases in the experience learning base, improve the speed and accuracy of scheduling strategy retrieval, and ensure that special vehicles can obtain optimized decision support in a timely manner in a dynamically changing operating environment.

[0068] Specifically, in some of the embodiments of the present invention, an adaptive collaborative network based on the open-source HarmonyOS distributed soft bus is proposed to realize autonomous negotiation and task redistribution among vehicles. However, in this process, when the communication link is interrupted or a local node fails, the system cannot maintain data synchronization and decision continuity, resulting in the failure of scheduling instructions, loss of monitoring capabilities, and lack of a mechanism for continuous operation based on local data under adverse network conditions, thereby triggering the risk of work process interruption and collaborative collapse.

[0069] In response, this invention further proposes that the method also includes: deploying a lightweight management terminal on the open-source HarmonyOS operating system, supporting job instruction issuance, real-time monitoring interface display, and manual intervention interface; all scheduling decisions, collaborative messages, and performance data are uniformly stored and synchronized through the open-source HarmonyOS distributed data management module, ensuring that the system can still degrade to run based on the latest local data when the network is down or there is a partial failure, and its data synchronization consistency protocol satisfies the eventual consistency model.

[0070] Among them, the lightweight management terminal is a terminal implementation with low resource consumption and the ability to run independently on edge devices. It is used for embedded operating systems or simplified application frameworks. By reducing dependence on centralized computing resources, it ensures that basic operational capabilities are maintained in communication-constrained environments. The open-source HarmonyOS distributed data management module is a data management framework built on the open-source HarmonyOS distributed capabilities. It is used for distributed key-value storage or event tracing mechanisms. It supports cross-device data synchronization by uniformly managing scheduling decision and performance data scattered across various nodes. The eventual consistency model is a distributed system data consistency protocol used for conflict resolution strategies based on vector clocks or operation logs. It allows for temporary data differences during network fluctuations, but ensures that the data state eventually converges after the system recovers.

[0071] Specifically, this invention enables critical operational capabilities to be deployed at the edge through the independent deployment of lightweight management terminals. When network anomalies occur, operators can directly issue commands and intervene manually through local terminals. Simultaneously, the distributed data management module continuously maintains the data synchronization status and automatically switches to the latest locally cached data version when a communication interruption is detected, ensuring that scheduling decisions are based on the latest environmental information. The data synchronization mechanism adopts an eventual consistency model, which automatically coordinates data differences between nodes after system recovery. This maintains real-time responsiveness while avoiding performance bottlenecks caused by strong consistency requirements, enabling the system to operate elastically in complex network environments.

[0072] As a preferred embodiment, the present invention is implemented as follows: The lightweight management terminal is deployed on an industrial-grade mobile terminal running the open-source HarmonyOS operating system. This terminal provides a touchscreen interface for inputting work instructions and visually monitoring vehicle status. The distributed data management module is specifically implemented as a distributed data service component of the open-source HarmonyOS, responsible for synchronizing task allocation records and performance indicator data between the vehicle terminal and the management terminal. When the wireless network at the construction site is temporarily interrupted, the system automatically switches to the latest scheduling data stored locally. Operators can manually adjust the vehicle task order on the mobile terminal. After the network is restored, the system automatically merges the local operation records with the cloud data to ensure the continuity of operations.

[0073] Through the above solution, the present invention can still maintain basic scheduling functions when the communication link is interrupted or a local node fails, avoiding interruption of the operation process and ensuring the continuity and reliability of special vehicle collaborative operation.

[0074] The present invention also provides an embodiment of a practical application of the method of the present invention.

[0075] I. Application Scenario: A large automated container port terminal is selected as the application scenario. This terminal deploys a cluster of three types of special vehicles: automated guided vehicles (AGVs), rail-mounted gantry cranes (RMGs), and automated straddle carriers (ASCs), which collaboratively complete the transfer and storage of containers from quay cranes to the container yard. Traditional scheduling systems employ centralized control and static path planning, which generally suffer from problems such as response delays (>30 seconds), frequent path conflicts (>15 times per day), and uneven resource utilization (some AGVs have an idle rate >20%) when ships arrive in large numbers and operational intensity increases dramatically.

[0076] This embodiment will fully demonstrate the application and data simulation of this method in a typical operation task: "transferring container CTN20240001 from quay crane A03 to yard B07-05-01".

[0077] II. Specific Implementation and Data Extrapolation at Each Stage: S1: Construction and Trend Prediction of Dynamic Twin System for Working Environment: S1.1. Multi-source Sensor Network Access and Initial Modeling: Based on the open-source HarmonyOS distributed sensing framework, a dock-level sensor network is constructed: Vehicle Terminal: 30 AGVs, 8 RMGs, and 12 ASCs report GNSS / IMU fusion positioning data (accuracy ±2cm), battery level, load status, and robotic arm status in real time through HarmonyOS terminals.

[0078] Environmental perception network: a quay crane perception system (4 sets) that integrates lidar and vision to identify container grabbing / placing status and truck queue length in real time; and a cluster of 24 high-definition cameras in the yard to monitor container occupancy status and mobile device location.

[0079] Task system interface: Real-time access to task data such as ship stowage charts, container bill of lading numbers, and target container locations.

[0080] Based on the initial data, a three-dimensional digital twin model covering the entire terminal area is constructed, including quay cranes, yard blocks, road network, and real-time virtual mapping of all vehicles and containers.

[0081] S1.2. Real-time operation of the environmental change prediction algorithm: Before the task is triggered, the system makes trend predictions based on historical and real-time data.

[0082] Input feature vector construction: For the B07 area passage of the stockyard, extract the feature vector from the historical operation log. The average number of vehicles passing through this lane in the past hour (12 vehicles / hour) and the average percentage of time spent in congestion (15%).

[0083] Real-time sensor data vector Current vehicle density (4 vehicles) and average speed (3m / s) in the channel.

[0084] Vector of weather influencing factors Current wind speed (level 5, approximately 10 m / s, affecting the stability of the RMG truck's movement).

[0085] Predicted output: Pre-trained temporal neural network model The model is running. It predicts the "congestion index" (environmental status) of the B07 zone passage within the next 10 minutes. The probability that one dimension exceeds the threshold of 0.7 is:

[0086] The probability distribution map is updated in real time to the digital twin model and visualized in the form of a heat map, indicating that there is a high probability that channel B07 will become congested.

[0087] S2: Multi-objective intelligent decision engine performs task decomposition and conflict resolution: S2.1. Task decomposition: The upper-level system issues the instruction: "Transfer container CTN20240001 from quay crane A03 to yard B07-05-01".

[0088] The task decomposer parses it into a sequence of atomic operations that can be executed in parallel: RMG-A03: grab CTN20240001 and place it at the AGV docking point.

[0089] AGV-XX: Carries containers from quay crane A03 to the handover point B07 in the yard.

[0090] ASC-YY: Grab the container at the B07 handover point and store it in the target container location B07-05-01.

[0091] S2.2. Resource Conflict Detection and Dynamic Path Planning Conflict Detection: The system assigns AGV-11 to perform a transportation task. The resource conflict detection model is based on a digital twin model, calculating the predetermined trajectory of AGV-11 and the trajectory of AGV-08, which is currently performing another task. Conflicts are detected in... At the time, at the road intersection J15:

[0092] Less than the safe distance threshold Meters, a potential conflict is determined, and the following conditions are met:

[0093]

[0094] Path replanning: The reinforcement learning algorithm of the dynamic path planner is triggered. Based on the current job mode (balancing efficiency and safety), the reward function... The weights are set as follows: (Path length) (time), (Energy consumption) (Conflict avoidance).

[0095] The algorithm is simulated multiple times in a digital twin environment. Assume the original path... The parameters are: (Unit energy consumption) (A conflict occurred). New path bypass The parameters are: (Conflict was successfully avoided).

[0096] Calculate the reward value:

[0097] Although The value is lower, but it successfully avoided the conflict. (Positive). In actual decision-making, the system will re-evaluate based on the security veto principle or after adjusting the weights, and ultimately choose the option that avoids conflict. As the execution path.

[0098] S3: Adaptive Cooperative Execution Based on Distributed Soft Bus: S3.1. Instruction Distribution and Autonomous Negotiation: Decision instructions (atomic operation sequences and AGV-11 planning paths) are broadcast to the relevant ASCs of AGV-11, RMG-A03 and yard B07 area with low latency (<50ms) through the open-source HarmonyOS distributed soft bus.

[0099] When AGV-11 is expected to arrive at handover point B07, ASC-05 and ASC-08 in that area autonomously negotiate via the soft bus to determine who will perform the grabbing operation. The benefit function is dynamically allocated based on the role.

[0100] Assume the current strategy focuses on capability matching ( ), .

[0101] ASC-05: Capability Matching (Optimal boom position), distance .

[0102]

[0103] ASC-08: Capability Matching ,distance .

[0104]

[0105] The system confirms that... ASC-05 with a higher value performs the grabbing job.

[0106] S3.2. Anomaly Handling (Simulated Fault): Assume that ASC-05 experiences a sudden mechanical failure while en route to the handover point, resulting in the loss of the communication heartbeat signal.

[0107] The anomaly propagation mechanism immediately broadcasts the "ASC-05 fault" status to all nodes in the cooperative network via the soft bus.

[0108] The decision engine triggers task reassignment. Based on the latest data, the benefit function is recalculated, and the role of "Catch CTN20240001" is quickly reassigned to ASC-08, which has the second-best benefit value, and the instructions are updated. The entire process is completed within 10 seconds.

[0109] S4: Performance monitoring, optimization and experience learning: S4.1. Performance monitoring: The operation performance monitoring module calculates the various indicators of this task chain in real time: AGV-11 actual path energy consumption: 15% higher than expected (88 units), at 101.2 units (due to detour).

[0110] Overall task completion time: 12% longer than the ideal value (simulation optimal value).

[0111] The actual congestion index of the B07 corridor reached 0.72, verifying the accuracy of the prediction in the S1 phase.

[0112] S4.2. Dynamic Scheduling Optimization: Based on the above performance evaluation and the current environment state, the dynamic scheduling optimizer initiates a multi-objective evolutionary algorithm. The optimization objective function... To minimize the total task time, total energy consumption, and average congestion index of the vector task The constraints include the vehicle's maximum load and battery threshold.

[0113] The algorithm generates a set of Pareto optimal solutions. The decision system selects one of these solutions, which suggests: fine-tuning the path planning weights for subsequent similar tasks; and increasing the conflict avoidance weights. Increased from 0.1 to 0.15.

[0114] To address congestion, temporarily add one ASC (ASC-12) to the B07 area's scheduling priority.

[0115] 4.3. Learning from Experience: This complete scheduling process was structured into a case study. The data is stored in the experience learning library. Case features include: task type (quay crane to yard), predicted congestion probability (0.68), conflict type (path intersection), resolution strategy (detour), and final performance indicators.

[0116] A week later, the system encountered another query case: "Transportation task from the quay crane to area B07, and the predicted congestion probability in this area is >0.65". The case matching similarity algorithm is activated:

[0117] Through calculation, this example is quickly retrieved as a highly similar historical case. Its optimized weight parameters ( The strategies of "pre-allocating resources for predicted congestion areas" and "pre-allocating resources for predicted congestion areas" are given priority to the decision engine to accelerate the generation of new decisions.

[0118] III. Quantitative Comparison of Implementation Results: Table 1: Comparison of Key Performance Indicators (Before and After Implementation of this Method)

[0119] IV. Conclusion: This embodiment, through simulating a typical complex and dynamic operational scenario of a large port container terminal, fully demonstrates the end-to-end management process based on open-source HarmonyOS, encompassing "dynamic twin perception, intelligent decision-making and planning, flexible collaborative execution, and closed-loop learning optimization." Practical data shows that this method, through the deep integration of the distributed capabilities of open-source HarmonyOS with a series of intelligent algorithms, effectively solves core challenges in special vehicle cluster scheduling, such as insufficient real-time and predictive environmental perception, multi-objective dynamic decision-making conflicts, poor robustness of collaborative mechanisms, and a lack of continuous system evolution capabilities. It significantly improves overall operational efficiency, safety, and system adaptability, verifying the advanced nature and practicality of this technical solution.

[0120] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for real-time scheduling and collaborative operation of intelligent special vehicles based on the open-source HarmonyOS, characterized in that: include: S1. A dynamic digital twin system for the operational environment is constructed based on the open-source HarmonyOS distributed perception framework. This system collects real-time data on special vehicle status, operational environment parameters, and task progress, establishes and updates a three-dimensional digital twin model, and predicts environmental change trends based on historical and real-time data. S2. Based on the environmental data output by the dynamic digital twin system, a multi-objective intelligent decision engine decomposes complex operational tasks, generates atomic operation sequences, and performs resource conflict detection and dynamic path planning in real-time. S3. Through an adaptive collaborative network based on the open-source HarmonyOS distributed soft bus, decision instructions are distributed to each special vehicle, supporting autonomous negotiation between vehicles and dynamic adjustment of operational strategies. A task reassignment mechanism is triggered when a vehicle malfunctions. S4. Monitor vehicle operation efficiency and resource utilization in real time, dynamically optimize scheduling strategies based on evaluation results, and store optimization experience in an experience learning library for continuous improvement of digital twin models and decision-making algorithms; wherein, steps S1 to S4 form a closed-loop control to realize real-time perception, decision optimization and collaborative execution of the entire chain of special vehicle scheduling and operation.

2. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS as described in claim 1, characterized in that... Step S1 specifically includes: accessing a multi-source sensor network based on the open-source HarmonyOS distributed perception framework to acquire vehicle positioning information, environmental obstacle distribution, terrain data, and operational resource status in real time; using the multi-source data to drive a three-dimensional digital twin model to dynamically reconstruct the operational scene, and employing an environmental change prediction algorithm to integrate historical operational patterns and real-time sensor information to predict obstacle changes, resource occupancy status, and traffic flow evolution trends within the operational area; the environmental change prediction algorithm uses a temporal neural network model, whose inputs include historical operational logs, real-time sensor data, and weather influencing factors, and whose output is a probability distribution map of the operational environment's state over a future period; its prediction process is expressed as follows: ;in, Indicates at time status Vehicle motion set and environmental context Under the conditions, at time Environmental conditions The probability distribution; This represents the temporal neural network model; The feature vector of historical job logs; This is a real-time sensor data vector; This is a vector of weather-influencing factors; this formula is used to quantify the uncertainty of environmental evolution and provide a probabilistic basis for intelligent decision-making.

3. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS as described in claim 2, characterized in that... The update mechanism of the three-dimensional digital twin model is as follows: when the deviation between the real-time sensor data and the model state exceeds a preset threshold... When this occurs, a local model reconstruction is triggered; the deviation is calculated as follows: ;in, Indicates the deviation value. express State vector in a time-series digital twin model express Real-time sensor data vector at any given moment. It represents the Euclidean norm; this mechanism ensures high-fidelity synchronization between the model and the physical environment.

4. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS as described in claim 1, characterized in that... The multi-objective intelligent decision engine in step S2 includes: a task decomposer, used to parse upper-level operation instructions into a sequence of atomic operations that can be executed in parallel; a resource conflict detection model, which judges the path intersection, work area overlap, and equipment contention between vehicles in real time based on a digital twin model; and a dynamic path planner, which generates obstacle avoidance paths and multi-vehicle collaborative operation trajectories online by combining environmental prediction results and conflict detection information. The resource conflict detection model makes judgments by calculating the spatial-temporal intersection of the vehicle's expected trajectory. For any two vehicles... and During the time period The conditions for a memory conflict are: ； ;in, and Representing vehicles and At any moment The predicted location coordinates, This is a preset safe distance threshold; the formula is used to accurately identify potential collision or interference risks.

5. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS according to claim 4, characterized in that... The dynamic path planner is optimized online using a reinforcement learning algorithm, and its reward function... Taking into account path length, operation time, energy consumption, and conflict avoidance indicators, the specific definition is: ;in, Indicates the length of the planned path. Indicates the estimated operation time. Indicates the estimated energy consumption. Conflict avoidance rewards (positive rewards for successfully avoiding conflict, and negative penalties for causing conflict); Let be the weight coefficients of each item, satisfying This reward function is used to perform multiple rounds of decision-making in a digital twin environment to guide the generation of a comprehensively optimal set of paths.

6. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS according to claim 1, characterized in that... The specific implementation of the adaptive collaborative network in step S3 is as follows: a vehicle group communication channel is constructed through the open-source HarmonyOS distributed soft bus to support low-latency, high-reliability information broadcasting and intra-group negotiation; an adaptive collaborative protocol based on a state machine is designed so that vehicles can autonomously adjust their work sequence, speed, or division of labor strategy according to real-time environmental information and group status; when a vehicle malfunctions or communication is interrupted, other nodes in the collaborative network are notified through an anomaly propagation mechanism, and the decision engine reassigns the unfinished tasks of the vehicle.

7. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS as described in claim 6, characterized in that... In the aforementioned adaptive cooperative protocol, vehicle roles are dynamically allocated through a benefit function. The function calculates the vehicle. Take on a role Expected benefits: ;in, Indicates vehicle Abilities and Roles Required matching degree Indicates vehicle To the role The current distance to the target location. To the maximum allowed distance, and To adjust the weights, the system assigns roles to... The formula identifies the available vehicle with the highest value, achieving an adaptive optimal match between resources and tasks.

8. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS according to claim 1, characterized in that: Step S4 includes: calculating the task completion rate, resource utilization efficiency, and energy consumption indicators of each vehicle in real time through the task performance monitoring module; adjusting the vehicle scheduling frequency, path weight, or task allocation strategy based on the performance evaluation results and the current environmental state using a dynamic scheduling optimizer; recording each scheduling decision and its execution effect in an experience learning base; extracting excellent task cases through data mining for iterative optimization of the path planning algorithm and conflict detection model; the dynamic scheduling optimizer adopts a multi-objective evolutionary algorithm, whose optimization objective function is... for: ； ;in, Represents a scheduling scheme vector. Representing the One optimization objective (such as total time consumption, total energy consumption). For the target quantity, Representing the The optimizer generates a set of Pareto optimal solutions in parallel, subject to real-time constraints, for decision-making.

9. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS according to claim 8, characterized in that, The experience learning base employs a case-based reasoning method, and its case matching similarity... The calculation formula is: ;in, For the current query case, These are cases from the historical case library. The total number of case features and The query cases and historical cases are respectively in the 1st Values ​​on each feature The weight of this feature. It is aimed at the first The local similarity function of each feature; this formula is used to quickly retrieve the best scheduling strategy that is most similar to the current scenario from the history.

10. The intelligent special vehicle real-time scheduling and operation collaboration method based on open-source HarmonyOS according to claim 1, characterized in that... The method also includes: deploying a lightweight management terminal on the open-source HarmonyOS operating system, supporting job instruction issuance, real-time monitoring interface display, and manual intervention interface; all scheduling decisions, collaborative messages, and performance data are uniformly stored and synchronized through the open-source HarmonyOS distributed data management module, ensuring that the system can still operate based on the latest local data in the event of network outage or partial failure, and its data synchronization consistency protocol satisfies the eventual consistency model.

Images