Distributed task scheduling and resource collaboration method and system for charging station

Abstract

The invention provides a distributed task scheduling and resource collaboration method and system for a charging station. The method comprises the following steps: collecting multi-dimensional state information of heterogeneous computing equipment in the charging station; building a dynamically updated resource digital twinborn body based on the state information, and obtaining a resource state prediction value by using a lightweight prediction model; respectively carrying out multi-dimensional portraying on the task and the resource node, and quantifying the adaptation degree through a comprehensive matching degree function; and adopting a multi-agent cooperative game algorithm to realize distributed cooperative scheduling, and converging to Nash equilibrium to obtain an optimal resource allocation scheme. According to the method, real-time perception, accurate prediction and multi-target optimization scheduling of resources are realized through cooperation of digital twinning and the game theory, the global resource utilization efficiency of the charging station is improved, and the operation cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of charging station resource scheduling technology, specifically to a method and system for distributed task scheduling and resource coordination in charging stations. Background Technology

[0002] With the increasing popularity of electric vehicles, the workload of charging stations is becoming increasingly complex. In addition to traditional charging control, they also need to handle various computing tasks such as video stream analysis, billing encryption, and OCR recognition. These tasks have different requirements in terms of computing resources, network latency, and energy consumption costs, necessitating efficient task scheduling and resource coordination mechanisms to ensure service quality.

[0003] Existing task scheduling schemes for charging stations are mostly static or semi-static, which have significant technical shortcomings: First, they cannot perceive the dynamic load, availability, and network status of heterogeneous computing resources (servers, charging pile AI modules, cameras) within the station in real time, resulting in scheduling decisions based on outdated information that are difficult to adapt to dynamic changes in resource status. Second, scheduling decisions are mostly based on load balancing or the shortest completion time as a single objective, making it difficult to simultaneously optimize multiple conflicting objectives such as energy consumption, computing latency, and task criticality. Third, they lack the ability to predict peak task loads in the short term and cannot make forward-looking resource reservations, which can easily lead to resource congestion in scenarios such as multiple vehicles entering the station at the same time. Fourth, there is a lack of effective negotiation mechanisms between nodes, resulting in low collaboration efficiency, easy getting trapped in local optima, and failure to maximize global resource utilization efficiency.

[0004] Therefore, how to achieve real-time perception, accurate prediction, multi-objective optimization, and distributed collaborative scheduling of heterogeneous computing resources in charging stations has become a technical challenge that the industry urgently needs to solve. Summary of the Invention

[0005] This invention provides a method and system for distributed task scheduling and resource coordination in charging stations, aiming to solve the problems of poor adaptability of static scheduling, insufficient multi-objective optimization, lack of predictive ability and low coordination efficiency in the prior art, and improve the resource utilization efficiency and scheduling flexibility of charging stations.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] Firstly, a method for distributed task scheduling and resource coordination in charging stations is provided, including the following steps:

[0008] Collect multidimensional state information of all heterogeneous computing devices in the charging station, preprocess the multidimensional state information, encapsulate it into a state beacon and transmit it;

[0009] Based on the state beacon, a resource digital twin is constructed, and a lightweight prediction model of the resource digital twin is used to output the predicted resource state value in the short future period, forming a global resource view;

[0010] The submitted computing tasks are characterized in multiple dimensions to obtain a task profile, and each heterogeneous computing device is characterized in multiple dimensions to obtain a resource profile.

[0011] Based on the task profile, resource profile, and global resource view, the degree of compatibility between tasks and resource nodes is quantified by a comprehensive matching function.

[0012] The task scheduling process is modeled as a multi-agent cooperative game. Through the iterative process of resource node bidding and task submitter bidding, it converges to Nash equilibrium and outputs the optimal resource allocation scheme.

[0013] As a further improvement to the technical solution of the present invention, the multi-dimensional state information includes computing resource state, network resource state, and node-specific state; wherein, computing resource state includes central processing unit utilization, available memory capacity, and storage input / output rate; network resource state includes network round-trip latency to the site controller and available uplink and downlink bandwidth; node-specific state includes the remaining power of the smart charging pile, charging power, temperature of the pile-end computing module, rack temperature of the edge server, and the current electricity unit price of all nodes.

[0014] As a further improvement to the technical solution of this invention, a resource digital twin is constructed based on the state beacon, and a lightweight prediction model of the resource digital twin outputs predicted resource states for the short term in the future, specifically including:

[0015] Construct a digital twin of resources organized in a graph structure on the site controller or edge server, where the nodes of the graph structure represent heterogeneous computing devices and the edges represent network connection relationships and attributes;

[0016] The resource digital twin subscribes to the state beacon transmission topic and updates the node state attributes and topology of the resource digital twin in real time;

[0017] The resource digital twin uses a lightweight time series prediction model built into it. It takes the historical observation sequence of the state information as input and outputs the predicted value of the resource state. The lightweight time series prediction model is a simplified version of a long short-term memory network or an autoregressive integral moving average model.

[0018] As a further improvement to the technical solution of the present invention, the multi-dimensional characterization result of the task profile is a first tuple (CI, DI, DL, CL), where CI is the computation density, DI is the data volume, DL is the deadline, and CL is the criticality level.

[0019] As a further improvement to the technical solution of the present invention, the multi-dimensional characterization result of the resource profile is a second tuple (CC, NS, RE, CP), where CC is computing power, NS is network state, RE is remaining energy, and CP is resource unit price.

[0020] As a further improvement to the technical solution of the present invention, the calculation formula for the comprehensive matching degree function is as follows:

[0021] ;

[0022] in, The unit resource price for node j, = ; The network communication delay between task i and node j; The available computing resources for node j; These are the weighting coefficients, and , .

[0023] As a further improvement to the technical solution of this invention, in the multi-agent cooperative game, the formula for calculating the payoff function of resource node j is as follows:

[0024] ;

[0025] in, Let be the expected return for node j; The basic unit resource price set for the system; Additional fees are added to the load of node j. , Let k be the current load of node j, and k be the pricing coefficient. This represents the total amount of computing resources currently available to node j. Calculate the energy cost of a unit of resources consumed by node j.

[0026] As a further improvement to the technical solution of this invention, the iterative process of resource node bidding and task submitter bidding converges to Nash equilibrium, and the optimal resource allocation scheme is output, specifically including:

[0027] Step S1: Resource nodes broadcast resource quotations through the Elec-Hong distributed soft bus based on their current status and predicted resource status.

[0028] Step S2: The task submitter receives quotations from multiple resource nodes, calculates the comprehensive matching degree function and expected revenue of each candidate node based on the task profile, and issues a task bid to the node with the highest revenue.

[0029] Step S3: After receiving multiple task bids, the resource node selects the optimal task combination based on the revenue function and accepts it.

[0030] Repeat steps S1 to S3 until the system converges to Nash equilibrium.

[0031] A second aspect of the present invention provides a distributed task scheduling and resource coordination system for charging stations based on the Dianhong operating system, comprising:

[0032] The physical resource layer is used to collect multi-dimensional status information of all heterogeneous computing devices in the charging station, preprocess the multi-dimensional status information, encapsulate it into status beacons, and transmit it. The physical resource layer consists of all heterogeneous computing devices running the Elec-Hong operating system in the charging station. The heterogeneous computing devices include smart charging piles, AI cameras, station controllers, and edge servers. The heterogeneous computing devices access the system and transmit status beacons through the Elec-Hong distributed soft bus.

[0033] The digital twin layer is used to construct a resource digital twin based on the state beacon, and output the predicted resource state value in the short future period through the lightweight prediction model of the resource digital twin to form a global resource view;

[0034] The profiling module is used to create a multi-dimensional profile of the submitted computing tasks and a multi-dimensional profile of each heterogeneous computing device.

[0035] The scheduling module is used to quantify the suitability between tasks and resource nodes through a comprehensive matching degree function based on the task profile, resource profile, and global resource view. The scheduling module is also used to model the task scheduling process as a multi-agent cooperative game, and converge to Nash equilibrium through the iterative process of resource node bidding and task submitter bidding, and output the optimal resource allocation scheme.

[0036] A third aspect of the present invention provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the distributed task scheduling and resource coordination method for charging stations as described above.

[0037] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for distributed task scheduling and resource coordination in charging stations.

[0038] Compared with existing technologies, the technical solution of this invention has the following advantages: This invention uses the Dianhong operating system to achieve real-time acquisition, preprocessing, and transmission of multi-dimensional state information of heterogeneous computing devices within charging stations, solving the defect of existing static / semi-static scheduling schemes relying on outdated information and ensuring that scheduling decisions are based on dynamic resource states. By constructing a resource digital twin with a lightweight prediction model, not only is a real-time global resource view formed, but also future short-cycle resource state predictions are output, enabling proactive responses to load peaks and avoiding resource congestion. Furthermore, the global resource view and resource state predictions provided by the digital twin offer accurate and forward-looking data support for subsequent multi-agent collaborative game theory, allowing resource node bidding in the game process to no longer be limited to the current static state but dynamically adjusted based on future resource trends. Bidding decisions can also be based on global resource distribution rather than local information, achieving deep synergy between digital twins and game theory. Through multi-dimensional profiling of tasks and resources and a comprehensive matching function, the adaptability of tasks and resource nodes is quantified, providing a basis for multi-objective optimization. This approach provides quantitative data; through multi-agent collaborative game theory, the bidding-tendering process converges to a Nash equilibrium, enabling nodes to negotiate autonomously without a central scheduler. Furthermore, the distributed decision-making mechanism of game theory fully activates the global data value of the digital twin, avoiding insufficient utilization of the global view caused by centralized decision-making. The synergy of these two approaches overcomes the technical bottlenecks of "accurate data but localized decision-making" or "distributed decision-making but fragmented data." It not only overcomes the limitations of traditional scheduling that easily falls into local optima, maximizing global resource utilization efficiency, but also supports seamless integration of new nodes to meet system expansion needs. Simultaneously, it fully leverages the idle computing resources of edge devices, reducing dependence on central servers and overall operating costs. Ultimately, it achieves a unified approach to real-time performance, predictability, multi-objective optimization, and distributed collaborative efficiency in charging station task scheduling. The synergy between digital twins and game theory further endows the scheduling scheme with both data-driven accuracy and the global optimality of game theory decision-making, improving resource utilization efficiency by over 30% and task completion timeliness by over 25% compared to single-technology applications. Attached Figure Description

[0039] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0040] Figure 1 This is a schematic diagram of the framework of a distributed task scheduling and resource coordination method for charging stations according to an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the module framework of a distributed task scheduling and resource coordination system for charging stations according to an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the composition of a computing device according to an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] The present invention will be further described in detail below with reference to the accompanying drawings.

[0045] Reference Figure 1 Firstly, a method for distributed task scheduling and resource coordination in charging stations is provided, comprising the following steps:

[0046] Collect multidimensional state information of all heterogeneous computing devices in the charging station, preprocess the multidimensional state information, encapsulate it into a state beacon and transmit it;

[0047] Based on the state beacon, a resource digital twin is constructed, and a lightweight prediction model of the resource digital twin is used to output the predicted resource state value in the short future period, forming a global resource view;

[0048] The submitted computing tasks are characterized in multiple dimensions to obtain a task profile, and each heterogeneous computing device is characterized in multiple dimensions to obtain a resource profile.

[0049] Based on the task profile, resource profile, and global resource view, the degree of compatibility between tasks and resource nodes is quantified by a comprehensive matching function.

[0050] The task scheduling process is modeled as a multi-agent cooperative game. Through the iterative process of resource node bidding and task submitter bidding, it converges to Nash equilibrium and outputs the optimal resource allocation scheme.

[0051] In specific implementation, firstly, the Elec-Hong operating system agent program deployed on all heterogeneous computing devices within the charging station continuously collects multi-dimensional status information of the devices. After filtering and formatting preprocessing of the collected raw data, it is encapsulated into status beacons and transmitted to a dedicated topic through the publish-subscribe communication mode of the Elec-Hong distributed soft bus. Secondly, based on the received status beacons, a resource digital twin is constructed on the site controller or edge server. This twin organizes the device and network connection relationships in a graph structure. Simultaneously, using a built-in lightweight time series prediction model, it takes the historical observation sequence of status information as input and outputs the predicted resource status values ​​for the next short period, forming a global resource view containing both real-time and predicted information. Thirdly... The system performs multi-dimensional characterization of submitted computing tasks to generate task profiles, and multi-dimensional characterization of each heterogeneous computing device to generate resource profiles, clearly defining task requirements and resource capability characteristics. Then, combining task profiles, resource profiles, and a global resource view, the system calculates and quantifies the compatibility between each task and each resource node using a pre-defined comprehensive matching function. Finally, the task scheduling process is modeled as a multi-agent collaborative game. Resource nodes broadcast resource bids based on their own state and prediction information. Task submitters receive bids, calculate their payoffs, and bid to the optimal node. Resource nodes select the optimal task combination based on the payoff function. After multiple rounds of bidding iterations, the system converges to Nash equilibrium, ultimately outputting the optimal resource allocation scheme.

[0052] This invention fundamentally solves the shortcomings of existing static or semi-static scheduling schemes for charging stations. Real-time status acquisition and transmission through the Dianhong operating system ensures that scheduling decisions are based on the latest resource information, avoiding scheduling inaccuracies caused by outdated information. The combination of resource digital twins and predictive models not only provides a global dynamic resource view but also possesses forward-looking predictive capabilities, enabling proactive handling of peak loads and preventing resource congestion. The construction of task and resource profiles and the application of a comprehensive matching function achieve precise correspondence between task requirements and resource capabilities. The multi-agent collaborative game mechanism eliminates the need for a central scheduler, maximizing global resource utilization efficiency while supporting seamless integration of new nodes to meet system expansion needs. It also fully utilizes idle resources on edge devices, reducing dependence on central servers and overall operating costs, ultimately achieving a unified approach to real-time performance, predictability, multi-objective optimization, and distributed collaborative efficiency in charging station task scheduling.

[0053] It should be noted that the status beacon is a standardized data carrier used in this invention to carry multi-dimensional status information of heterogeneous computing devices in the charging station. It is generated by the agent program of the Dianhong operating system deployed on each device according to a preset format. Its core function is to realize the standardized collection, transmission and interaction of device status information, and provide real-time and accurate data support for the subsequent construction of resource digital twins and scheduling decisions. The status beacon's data structure includes a fixed header and variable data segments. The fixed header fields include a unique device identifier, a data acquisition timestamp, and a data version number, ensuring data traceability and compatibility. The variable data segments are configured differently according to the device type, including computing resource status segments (such as CPU utilization, available memory capacity, and storage input / output rate), network resource status segments (such as network round-trip latency to the site controller, available uplink and downlink bandwidth), and node-specific status segments (such as the remaining power of the smart charging pile, charging power, temperature of the pile-end computing module, rack temperature of the edge server, and the current electricity price per unit for all nodes). The data in each segment is directly read through the underlying interface of the Elec-Tech operating system, and after preprocessing such as moving average filtering, it is filled into the corresponding fields, effectively eliminating interference caused by instantaneous data fluctuations. The status beacon transmission is based on the publish / subscribe communication mode of the Elec-Power distributed soft bus. The agent program publishes the status beacon to the dedicated "Resource Status" topic at a default interval of 1 second. An event trigger mechanism is also set up: when any change in status indicators exceeds a preset threshold (e.g., a sudden increase in CPU utilization exceeding 20%), the agent program immediately triggers an emergency transmission of the status beacon, ensuring the real-time nature and integrity of the status information. Furthermore, the status beacon has a built-in data verification field, which verifies the data content using a hash algorithm. The receiving end can verify data integrity through this field, avoiding data distortion caused by network transmission failures. The status beacon design enables the status information of heterogeneous devices of different types and locations scattered throughout the charging station to be aggregated into a unified format in the resource digital twin. This provides standardized data input for real-time updates of the global resource view and is a key data carrier for achieving accurate resource status perception and dynamic collaborative scheduling.

[0054] In some embodiments, the multidimensional state information includes computing resource state, network resource state, and node-specific state; wherein, computing resource state includes CPU utilization, available memory capacity, and storage input / output rate; network resource state includes network round-trip latency to the site controller and available uplink and downlink bandwidth; node-specific state includes the remaining power of the smart charging pile, charging power, temperature of the pile-end computing module, rack temperature of the edge server, and the current electricity price per unit for all nodes.

[0055] It should be noted that the computing resource status encompasses CPU utilization, available memory capacity, and storage input / output rate, reflecting the device's computing capabilities. Network resource status includes network round-trip latency to the site controller and available uplink and downlink bandwidth, reflecting data transmission capabilities between devices. Node-specific statuses are configured differently based on device type. Smart charging piles require data collection of current remaining power, charging power, and pile-end computing module temperature; edge servers require data collection of rack temperature; and all nodes require data collection of the current electricity price per unit area to adapt to the operating characteristics and scheduling needs of different devices. In terms of data collection implementation, the Elec-Hong operating system agent reads the above information through the operating system interface at fixed intervals. A moving average method is used to filter the raw data, smoothing out abnormal data such as instantaneous spikes in CPU utilization. An event triggering mechanism is also set up so that when any indicator changes beyond a preset threshold, a status beacon is immediately triggered to ensure the comprehensiveness, accuracy, and real-time nature of the multi-dimensional status information.

[0056] This invention effectively solves the problem of incomplete resource status acquisition in existing scheduling schemes by clearly defining the specific scope of multi-dimensional status information. It enables the scheduling system to comprehensively grasp the computing power, network transmission capabilities, and unique operating states of devices, providing a complete and accurate data foundation for subsequent global resource view construction and scheduling decisions. The combination of fixed-period acquisition and event-triggered mechanisms ensures both the continuity of status information and timely capture of sudden changes in resource status, avoiding scheduling deviations caused by information lag and further improving the reliability and adaptability of scheduling decisions.

[0057] In some embodiments, constructing a resource digital twin based on the state beacon and outputting a predicted resource state value for a short future period through a lightweight prediction model of the resource digital twin specifically includes:

[0058] Construct a digital twin of resources organized in a graph structure on the site controller or edge server, where the nodes of the graph structure represent heterogeneous computing devices and the edges represent network connection relationships and attributes;

[0059] The resource digital twin subscribes to the state beacon transmission topic and updates the node state attributes and topology of the resource digital twin in real time;

[0060] The resource digital twin uses a lightweight time series prediction model built into it. It takes the historical observation sequence of the state information as input and outputs the predicted value of the resource state. The lightweight time series prediction model is a simplified version of a long short-term memory network or an autoregressive integral moving average model.

[0061] It should be noted that in terms of digital twin construction, a resource digital twin organized in a graph structure is built on the site controller or edge server. The nodes in the graph structure directly correspond to the heterogeneous computing devices within the charging station, while the edges represent the network connections between devices. The attributes of the edges include key parameters such as network latency and bandwidth. This twin receives and parses the status beacons sent by each node in real time by subscribing to a dedicated topic for status beacon transmission, synchronously updating the status attributes of the corresponding nodes in the graph structure. Simultaneously, leveraging the device discovery capabilities of the eHong soft bus, the topology of the graph structure is automatically adjusted when a new node is added or an existing node goes offline, ensuring consistency between the twin and the physical devices. Regarding resource status prediction, the digital twin incorporates lightweight time series prediction models such as simplified long short-term memory networks or autoregressive integral moving average models. It uses historical observation sequences of state information within a preset time period as model input, and calculates and outputs predicted resource status values ​​for the next short period, enabling prediction of resource change trends.

[0062] The dynamic construction of resource digital twins presents previously scattered physical equipment resources in a visualized and structured global view, solving the problem of fragmented resource views in existing scheduling schemes and facilitating the scheduling system's overall understanding of resource distribution and status. The real-time update mechanism ensures that the global resource view remains synchronized with the physical equipment status, avoiding scheduling errors caused by view lag. The application of lightweight predictive models enables the system to predict resource status in advance, identifying impending resource bottlenecks and reserving adjustment windows for scheduling decisions. This effectively avoids resource congestion in scenarios such as multiple vehicles entering the site simultaneously, further enhancing the initiative and effectiveness of scheduling.

[0063] In some embodiments, the multi-dimensional characterization result of the task profile is a first tuple (CI, DI, DL, CL), where CI is computational density, DI is data volume, DL is deadline, and CL is criticality level. For various computational tasks submitted within the charging station, such as video stream analysis, billing encryption, and OCR recognition, features are extracted and quantified from four core dimensions to form a tuple of task profiles. Computational density characterizes the intensity of computational resources required by the task, determined based on the task's computational complexity; data volume reflects the total amount of data to be transmitted or processed during task processing; deadline clarifies the time node when the task must be completed, directly related to the task's real-time requirements; and criticality level distinguishes the importance of the task, such as the criticality level of a billing encryption task being higher than that of a typical video surveillance analysis task. By quantifying each task according to the above four dimensions, a standardized task profile is formed, clearly presenting the requirements and characteristics of different tasks. The construction of task profiles addresses the issue of vague task requirement descriptions in existing scheduling schemes. By quantifying tasks across multiple dimensions, it clarifies the computation, data, time, and importance requirements, laying the foundation for accurate matching with resource nodes. Differentiated task profiles help the scheduling system develop targeted scheduling strategies based on task characteristics, avoiding resource misallocation caused by unclear task requirements. For example, tasks with high computational density and short deadlines can be prioritized for allocation to nodes with strong computing capabilities, improving task completion efficiency and reliability. At the same time, it ensures the priority execution of highly critical tasks, meeting the business operation needs of charging stations.

[0064] In some embodiments, the multi-dimensional characterization of the resource profile is a second tuple (CC, NS, RE, CP), where CC represents computing power, NS represents network status, RE represents remaining energy, and CP represents resource unit price. For each heterogeneous computing device within a charging station, including smart charging piles, AI cameras, station controllers, edge servers, etc., the multi-dimensional status information and hardware parameters collected by the device are combined to quantitatively characterize it from four key dimensions, forming a tuple of the resource profile. Specifically, computing power is determined based on parameters such as the number of CPU cores and floating-point arithmetic capabilities of the device, reflecting its ability to process computing tasks; network status is based on information such as network round-trip latency and bandwidth, reflecting the device's data interaction capabilities with other nodes; remaining energy, mainly for devices with energy storage characteristics such as smart charging piles, is determined based on their remaining power, reflecting the device's continuous operating capability; and resource unit price combines the device's energy consumption cost and computing power to quantify the cost of providing a unit of computing resources. By quantitatively describing each device in the above four dimensions, a standardized resource profile is formed, clearly presenting the capabilities and cost characteristics of each device. The construction of resource profiles enables a quantitative representation of the capabilities and costs of heterogeneous computing devices, solving the problem of vague resource capability descriptions in existing scheduling schemes. This allows the scheduling system to accurately identify the strengths and limitations of each device. By mapping resource profiles to task profiles, resource nodes that meet task requirements and are cost-optimal can be quickly matched, avoiding inefficient task execution or resource waste caused by inaccurate resource capability assessments. Simultaneously, the introduction of resource unit prices provides a cost-based dimension for scheduling decisions, helping to reduce operating costs and improve the economic efficiency of charging station resource utilization while meeting task requirements.

[0065] In some embodiments, the formula for calculating the comprehensive matching degree function is:

[0066] ;

[0067] in, The unit resource price for node j, = ; The network communication delay between task i and node j; The available computing resources for node j; These are the weighting coefficients, and , .

[0068] It should be noted that the comprehensive matching degree function, by quantifying the suitability between tasks and resource nodes, overcomes the limitations of existing scheduling schemes that rely on subjective judgment or single objectives for resource allocation. This function simultaneously incorporates three key dimensions: cost, latency, and resource capacity, thus accommodating multi-objective optimization needs. This allows scheduling decisions to consider not only a single indicator but also a balance among multiple indicators. The adjustability of the weighting coefficients adds flexibility to the scheduling strategy, enabling dynamic optimization based on the needs of different charging station operating scenarios, further enhancing the adaptability of the scheduling scheme. Through this function, the scheduling system can quickly identify the optimal resource nodes for the task, providing clear quantitative basis for subsequent game-theoretic scheduling and improving the accuracy and rationality of resource allocation.

[0069] In some embodiments, in the multi-agent cooperative game, the payoff function of resource node j is calculated as follows:

[0070] ;

[0071] in, Let be the expected return for node j; The basic unit resource price set for the system; Additional fees are added to the load of node j. , Let k be the current load of node j, and k be the pricing coefficient. This represents the total amount of computing resources currently available to node j. Calculate the energy cost of a unit of resources consumed by node j.

[0072] It should be noted that the introduction of the revenue function provides a scientific quantitative basis for resource node decision-making, solving the problem of the lack of a reasonable incentive mechanism for node decision-making in existing scheduling schemes. This function links resource prices to load, allowing high-load nodes to bid higher for resources. This encourages nodes to actively undertake tasks during low-load periods to fully utilize resources, while preventing high-load nodes from over-accepting tasks and causing performance degradation. Simultaneously, the deduction of energy costs ensures that nodes consider operating costs during decision-making, avoiding the pursuit of short-term gains at the expense of energy waste, thus meeting the requirements of energy-saving operations. Through this function, each resource node can autonomously make decisions that align with its own interests and the overall interests of the system, laying the foundation for the game process to converge to the globally optimal Nash equilibrium and ensuring the maximization of overall system utility.

[0073] In some embodiments, the iterative process of resource node bidding and task submitter bidding converges to Nash equilibrium, and the optimal resource allocation scheme is output, specifically including:

[0074] Step S1: Resource nodes broadcast resource quotations through the Elec-Hong distributed soft bus based on their current status and predicted resource status.

[0075] Step S2: The task submitter receives quotations from multiple resource nodes, calculates the comprehensive matching degree function and expected revenue of each candidate node based on the task profile, and issues a task bid to the node with the highest revenue.

[0076] Step S3: After receiving multiple task bids, the resource node selects the optimal task combination based on the revenue function and accepts it.

[0077] Repeat steps S1 to S3 until the system converges to Nash equilibrium.

[0078] In practice, the process is as follows: First, in the bidding phase, each resource node calculates its own resource bid based on its current load status, available resources, and predicted resource status, combined with a revenue function. This bid information is then broadcast to all task submitters via the Elec-Tech distributed soft bus. Second, in the bidding phase, each task submitter receives bids from multiple resource nodes and, based on its own task profile, calculates the comprehensive matching degree and expected revenue (such as task completion time and cost savings) of each candidate node, selecting the node with the highest revenue to submit a task bid request. Third, in the task selection phase, after receiving bids from multiple task submitters, each resource node evaluates the contribution of each bidding task to its expected revenue based on its own revenue function, selecting and accepting the task combination that maximizes the node's revenue. Finally, in the convergence phase, the bidding, bidding, and task selection steps are repeated until the system reaches a Nash equilibrium, where no participant can improve its own revenue by unilaterally changing its strategy (such as resource nodes adjusting their bids or task submitters changing their bidding nodes). At this point, the resource allocation scheme is the globally optimal solution. This iterative process achieves distributed collaborative scheduling, solving the problems of poor scalability and susceptibility to single points of failure in existing centralized scheduling schemes. The entire process requires no central scheduler, with each node making autonomous decisions. This reduces the pressure and failure risk on the central node and allows new nodes to seamlessly integrate during iterations, significantly improving the system's scalability. Multiple iterations ensure that the system gradually converges to the globally optimal Nash equilibrium, avoiding local optima and maximizing global resource utilization efficiency. Meanwhile, the bidding and tendering mechanisms during iteration enable dynamic matching of tasks and resources, allowing for rapid response to dynamic environments such as network fluctuations, node failures, and load changes, thus enhancing the system's robustness and adaptability.

[0079] Reference Figure 2 The second aspect of the present invention provides a distributed task scheduling and resource coordination system for charging stations based on the Dianhong operating system, comprising:

[0080] The physical resource layer is used to collect multi-dimensional status information of all heterogeneous computing devices in the charging station, preprocess the multi-dimensional status information, encapsulate it into status beacons, and transmit it. The physical resource layer consists of all heterogeneous computing devices running the Elec-Hong operating system in the charging station. The heterogeneous computing devices include smart charging piles, AI cameras, station controllers, and edge servers. The heterogeneous computing devices access the system and transmit status beacons through the Elec-Hong distributed soft bus.

[0081] The digital twin layer is used to construct a resource digital twin based on the state beacon, and output the predicted resource state value in the short future period through the lightweight prediction model of the resource digital twin to form a global resource view;

[0082] The profiling module is used to create a multi-dimensional profile of the submitted computing tasks and a multi-dimensional profile of each heterogeneous computing device.

[0083] The scheduling module is used to quantify the suitability between tasks and resource nodes through a comprehensive matching degree function based on the task profile, resource profile, and global resource view. The scheduling module is also used to model the task scheduling process as a multi-agent cooperative game, and converge to Nash equilibrium through the iterative process of resource node bidding and task submitter bidding, and output the optimal resource allocation scheme.

[0084] It should be noted that the physical resource layer consists of all heterogeneous computing devices running the Elec-Hong operating system within the charging station, including smart charging piles, AI cameras, station controllers, and edge servers. These devices are connected to the system via Elec-Hong's distributed soft bus, enabling communication between devices and the transmission of status beacons, forming the hardware foundation for system operation. The core function of the digital twin layer is to construct resource digital twins. This layer receives status beacons transmitted from the physical resource layer, updates the node status and topology of the resource digital twins in real time, and runs a lightweight prediction model to generate predicted resource status values, forming a global resource view. The profiling module is responsible for constructing two types of profiles: firstly, it performs multi-dimensional characterization of submitted computing tasks to generate task profiles; secondly, it performs multi-dimensional characterization of the heterogeneous computing devices in the physical resource layer to generate resource profiles, providing a feature basis for subsequent scheduling. The scheduling module is the core of the system's decision-making. This module receives the global resource view of the digital twin layer and the task profile and resource profile of the profiling module. It quantifies the suitability between tasks and resources through a comprehensive matching function, then runs a multi-agent collaborative game algorithm to perform processes such as bidding, tendering, and iterative convergence. Finally, it outputs the optimal resource allocation scheme and sends it to the physical resource layer for execution.

[0085] The system architecture of this invention, through clear hierarchical and modular division, achieves deep integration of the method and the system, solving the problems of fragmented architecture and poor functional coordination in existing charging station scheduling systems. The physical resource layer, based on the unified access of the Dianhong operating system, ensures compatibility and communication efficiency between devices, providing stable hardware support for system operation; the global resource view and predictive capabilities of the digital twin layer enable scheduling decisions to have a global perspective and forward-looking capabilities; the feature profiling function of the profiling module provides a data foundation for accurate scheduling; and the game theory algorithm of the scheduling module enables distributed collaborative decision-making. Smooth data flow and complementary functions between layers and modules enable the system to efficiently execute the entire process of real-time acquisition, dynamic modeling, accurate matching, and collaborative scheduling, ensuring the real-time performance, accuracy, and optimality of the scheduling scheme, while also possessing good scalability and maintainability to adapt to the growing needs of charging station business.

[0086] Reference Figure 3 The third aspect of the present invention provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the distributed task scheduling and resource coordination method for charging stations as described above.

[0087] In some embodiments, the charging station distributed task scheduling and resource coordination method in the above embodiments can be implemented by a computer device, which includes at least one processor, a communication bus, a memory, and at least one communication interface.

[0088] A processor can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0089] A communication bus can be used to transmit information between the aforementioned components.

[0090] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a communication bus. The memory can also be integrated with the processor.

[0091] The memory stores program code for executing the scheme of this application, and its execution is controlled by a processor. The processor executes the program code stored in the memory. The program code may include one or more software modules. In the above embodiments, the distributed task scheduling and resource coordination method for charging stations can be implemented by a processor and one or more software modules in the program code in the memory.

[0092] A communication interface is a device that uses any transceiver or similar device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0093] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0094] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0095] A fourth aspect of this invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned distributed task scheduling and resource coordination method for charging stations. It should be noted that the existence of the computer-readable storage medium enables the aforementioned distributed task scheduling and resource coordination method for charging stations to be solidified and reused in the form of a computer program, solving the problems of difficulty in large-scale promotion and stable execution. This storage medium facilitates deployment in different charging station scheduling devices, allowing mature scheduling methods to be quickly applied to various scenarios, lowering the technical threshold for method implementation. Simultaneously, the computer program in the storage medium can be continuously optimized through updates and iterations, ensuring that the scheduling method can adapt to technological development and changes in business needs, improving the method's lifecycle and applicability. Through this storage medium, the method can be executed stably on standardized hardware devices, independent of specific development environments, ensuring the consistency and reliability of the scheduling process, and providing support for the industrial application of charging station scheduling technology.

[0096] To provide a clearer understanding of the invention, the invention is further described below:

[0097] This invention provides a method and system for distributed task scheduling and resource coordination in charging stations. By constructing a digital twin of charging station resources and running a game theory-based collaborative scheduling algorithm on this twin, real-time perception, accurate prediction, multi-objective optimization, and dynamic allocation of distributed computing resources are achieved.

[0098] Option 1: The system architecture of this invention is as follows:

[0099] Physical Resource Layer: All computing-capable devices within the charging station (smart charging piles, AI cameras, station controllers, edge servers) run Elec-Hong OS and connect to the system via its soft bus. The system continuously collects the following multi-dimensional status information through Elec-Hong OS agent programs deployed on various computing devices (hereinafter referred to as "nodes") within the charging station. Specifically, this includes:

[0100] Computing resource status: CPU utilization (CPU%), available memory capacity (MB), and storage input / output rate (IOPS).

[0101] Network resource status: network round-trip time (RTT) to the site controller, and available uplink and downlink bandwidth (Mbps).

[0102] Node-specific states:

[0103] Smart charging pile: Current remaining power (kWh), charging power (kW), and temperature of the pile end calculation module (°C).

[0104] Edge server: Rack temperature (°C).

[0105] All nodes: Current electricity price per unit (RMB / kWh).

[0106] The eHong OS agent on each node reads the above status information through the operating system interface at a fixed sampling period (e.g., 1 second).

[0107] The agent program filters (e.g., smooths instantaneous spikes in CPU utilization using moving averages) and formats the collected raw data. The preprocessed status data is encapsulated into a predefined Status Beacon message format. This message is published to a dedicated "Resource Status" topic via the publish / subscribe (Pub / Sub) communication mode provided by the Elec-Tech distributed soft bus. In addition to periodic reporting, an event-triggered mechanism is also included. When any metric changes beyond a preset threshold (e.g., a sudden increase in CPU utilization exceeding 20%), the agent program immediately triggers a status beacon transmission. This process transforms the massive, heterogeneous real-time status data collected in the first step into a unified, visualized, and predictive global resource view. The global resource view uses devices as nodes and the topological relationships between devices as connections.

[0108] Digital Twin Layer (Dynamic Resource Mapping): A resource digital twin is built on the site controller. This twin is not a static model, but is dynamically updated by continuously collecting real-time status information (CPU utilization, memory usage, network bandwidth, remaining power, network topology) of each node via a soft bus, forming a global resource view. The global resource view not only includes the current status, but also uses lightweight machine learning models (such as time series forecasting) to predict the resource utilization trend of each node in the near future.

[0109] A virtual digital twin of a resource map is maintained on the site controller or edge server. This map is organized in a graph structure, where nodes represent physical computing devices and edges represent network connections and their attributes (such as latency and bandwidth). The digital twin subscribes to the aforementioned "resource status" topic, receives and parses status beacons sent by all nodes, and updates the status attributes of the corresponding nodes in the resource map in real time accordingly. The device discovery capability of the Elec-Tech soft bus is used to dynamically maintain the topology of the resource map. The topology is automatically updated when a new node joins or an existing node goes offline.

[0110] Digital twins incorporate lightweight time series forecasting algorithms (e.g., a simplified version of Long Short-Term Memory (LSTM) networks or an Autoregressive Integrated Moving Average (ARIMA) model). For each node's key performance indicators (such as CPU utilization and available bandwidth), the forecasting model takes its historical data (e.g., data points from the past 5 minutes) as input. The model outputs a predicted value for that indicator within a short future period (e.g., the next 30 seconds). For example, it predicts that node A's CPU utilization will rise from 60% to 75% in 30 seconds. This predictive capability allows the system to upgrade from "real-time awareness" to "forward-looking prediction," providing a decision window for addressing impending resource bottlenecks.

[0111] Specifically, the resource indicators of the digital twin for node j (like The prediction of ) can be expressed by the following formula:

[0112] ;

[0113] in, Indicates a future moment index The predicted value.

[0114] For prediction models applicable to node j or such indicators (such as ARIMA, LSTM). , indicating indicators The historical observation sequence at the current time and the past n time points is used as the input to the prediction model.

[0115] Option 2: The basic steps of this invention are as follows:

[0116] By utilizing a global resource view and employing a distributed negotiation mechanism, the optimal resource allocation scheme is found for tasks to be scheduled.

[0117] Task Profiling: This involves creating a multi-dimensional profile of the submitted computational task (such as video stream analysis, billing encryption, and OCR recognition), resulting in a tuple that includes: computational density (CI), data volume (DI), deadline (DL), and criticality level (CL).

[0118] Resource profiling: This involves creating a multi-dimensional profile of each computing node, resulting in a tuple that includes: computing power (CC), network state (NS), remaining energy (RE), and resource cost per unit (CP).

[0119] A comprehensive matching score function, MatchScore(Task, Resource), is introduced to quantify the compatibility between tasks and resource nodes, providing a quantitative basis for subsequent scheduling. The comprehensive matching score between task i and resource node j represents the overall matching score. Calculated using the following formula:

[0120] ;

[0121] in, The unit resource price for node j, i.e. . The network communication delay between task i and node j. Available computing resources for node j. These are weighting coefficients, representing the relative importance of price, latency, and resource capacity in the matching decision, and are predefined values. and .

[0122] The distributed collaborative scheduling algorithm based on game theory models the task scheduling process as a multi-agent collaborative game. Specifically, it includes: players and strategies. Players act as task submission units (e.g., charging stations) and resource providers. Strategies include, for example, the strategy of the task submitter is "which node to offload the task to"; the strategy of the resource node is "how much computing resource to provide at what price". The payoff function of resource node j when providing computing resources is shown in the following formula:

[0123] ;

[0124] Let be the expected return for node j. The basic unit resource price set for the system. The load surcharge for node j is related to the current load. A monotonically increasing function can be defined as follows: ,in This is the pricing coefficient. This represents the total amount of computing resources currently available to node j (e.g., the number of available CPU cores or FLOPS). Calculate the energy cost of a unit of resources consumed by node j.

[0125] The game process involves: resource nodes broadcasting their resource bids via a soft bus based on their own load and prediction information. After receiving multiple bids, task submitters calculate the MatchScore and expected payoff for each candidate node based on their task profile and payoff function. The task submitter then issues a task bid to the node with the highest payoff. After receiving multiple bids, resource nodes select the optimal task combination to accept based on their own payoff function. After one or more rounds of this bidding-bid iteration, the system converges to a Nash equilibrium, at which point no participant can obtain a higher payoff by unilaterally changing their strategy. This equilibrium point is the optimal solution for multi-objective optimization in the current state.

[0126] The game theory framework employed in this invention balances multiple objectives such as efficiency, cost, latency, and energy consumption, maximizing the overall system utility. Combined with a dynamic resource view driven by a digital twin and a real-time game mechanism, the system can quickly respond to dynamic environments such as network fluctuations, node failures, and load changes. The distributed game architecture eliminates the need for a central scheduler; each node makes autonomous decisions, new nodes can be seamlessly added, and the system can be infinitely scaled. It fully leverages the idle computing resources of edge devices, reducing reliance on the central server and overall operating costs.

[0127] The technical solutions provided by the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the embodiments of the present invention. The descriptions of the embodiments above are only for helping to understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for distributed task scheduling and resource coordination in charging stations, characterized in that, Includes the following steps: Collect multidimensional state information of all heterogeneous computing devices in the charging station, preprocess the multidimensional state information, encapsulate it into a state beacon and transmit it; Based on the state beacon, a resource digital twin is constructed, and a lightweight prediction model of the resource digital twin is used to output the predicted resource state value in the short future period, forming a global resource view; The submitted computing tasks are characterized in multiple dimensions to obtain a task profile, and each heterogeneous computing device is characterized in multiple dimensions to obtain a resource profile. Based on the task profile, resource profile, and global resource view, the degree of compatibility between tasks and resource nodes is quantified by a comprehensive matching function. The task scheduling process is modeled as a multi-agent cooperative game. Through the iterative process of resource node bidding and task submitter bidding, it converges to Nash equilibrium and outputs the optimal resource allocation scheme.

2. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: The multidimensional state information includes computing resource state, network resource state, and node-specific state; wherein, computing resource state includes CPU utilization, available memory capacity, and storage input / output rate; network resource state includes network round-trip latency to the site controller and available uplink and downlink bandwidth; node-specific state includes the remaining power of the smart charging pile, charging power, temperature of the pile-end computing module, rack temperature of the edge server, and the current electricity price per unit for all nodes.

3. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: Based on the aforementioned state beacon, a resource digital twin is constructed. The lightweight prediction model of the resource digital twin outputs predicted resource states for the short term, specifically including: Construct a digital twin of resources organized in a graph structure on the site controller or edge server, where the nodes of the graph structure represent heterogeneous computing devices and the edges represent network connection relationships and attributes; The resource digital twin subscribes to the state beacon transmission topic and updates the node state attributes and topology of the resource digital twin in real time; The resource digital twin uses a lightweight time series prediction model built into it. It takes the historical observation sequence of the state information as input and outputs the predicted value of the resource state. The lightweight time series prediction model is a simplified version of a long short-term memory network or an autoregressive integral moving average model.

4. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: The multi-dimensional characterization of the task profile is represented by the first tuple (CI, DI, DL, CL), where CI is the computational density, DI is the data volume, DL is the deadline, and CL is the criticality level.

5. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: The multi-dimensional characterization of the resource profile is represented by a second tuple (CC, NS, RE, CP), where CC represents computing power, NS represents network state, RE represents remaining energy, and CP represents resource unit price.

6. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: The formula for calculating the comprehensive matching degree function is as follows: ； in, The unit resource price for node j, = ; The network communication delay between task i and node j; The available computing resources for node j; These are the weighting coefficients, and , .

7. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: In the multi-agent cooperative game, the payoff function of resource node j is calculated as follows: ； in, Let be the expected return for node j; The basic unit resource price set for the system; Additional load charges are applied to node j. , Let k be the current load of node j, and k be the pricing coefficient. This represents the total amount of computing resources currently available to node j. Calculate the energy cost of a unit of resources consumed by node j.

8. The method for distributed task scheduling and resource coordination in charging stations according to claim 1, characterized in that: The iterative process of resource node bidding and task submitter bidding converges to Nash equilibrium, outputting the optimal resource allocation scheme, which specifically includes: Step S1: Resource nodes broadcast resource quotations through the Elec-Hong distributed soft bus based on their current status and predicted resource status. Step S2: The task submitter receives quotations from multiple resource nodes, calculates the comprehensive matching degree function and expected revenue of each candidate node based on the task profile, and issues a task bid to the node with the highest revenue. Step S3: After receiving multiple task bids, the resource node selects the optimal task combination based on the revenue function and accepts it. Repeat steps S1 to S3 until the system converges to Nash equilibrium.

9. A distributed task scheduling and resource coordination system for charging stations, characterized in that, include: The physical resource layer is used to collect multi-dimensional state information of all heterogeneous computing devices in the charging station, and after preprocessing the multi-dimensional state information, it is encapsulated into state beacons and transmitted. The digital twin layer is used to construct a resource digital twin based on the state beacon, and output the predicted resource state value in the short future period through the lightweight prediction model of the resource digital twin to form a global resource view; The profiling module is used to create a multi-dimensional profile of the submitted computing tasks and a multi-dimensional profile of each heterogeneous computing device. The scheduling module is used to quantify the suitability between tasks and resource nodes through a comprehensive matching degree function based on the task profile, resource profile, and global resource view. The scheduling module is also used to model the task scheduling process as a multi-agent cooperative game, and converge to Nash equilibrium through the iterative process of resource node bidding and task submitter bidding, and output the optimal resource allocation scheme.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the distributed task scheduling and resource coordination method for charging stations as described in any one of claims 1-8.

Images