Computing power resource scheduling method, device, system and equipment of low-voltage side power distribution network and medium

Abstract

The invention discloses a computing power resource scheduling method, device, system and equipment of a low-voltage side power distribution network and a medium, and belongs to the field of resource scheduling, and the method comprises the steps: obtaining resource information and service information of the low-voltage side power distribution network; constructing a communication curve of the data volume based on the resource information, and determining the arrival delay of the processing service information according to the communication curve; when the arrival time delay is increased, generating a resource scheduling decision according to the resource information and the service information by using a preset machine learning algorithm; and performing resource scheduling processing on the low-voltage side power distribution network according to the resource scheduling decision. According to the invention, the delay of the business task can be determined according to the real-time resource and business processing related information, when the delay of the processing task is not satisfied, the resource scheduling decision is generated according to the real-time state, and then the resource scheduling processing is carried out according to the resource scheduling decision, so that the resource scheduling can fit the actual state, and the flexibility of the resource scheduling is improved; and the task processing efficiency can be improved, and the processing time is shortened.

Description

Technical Field

[0001] This invention relates to the technical field of resource scheduling, and in particular to a method, apparatus, system, equipment and medium for scheduling computing resources in a low-voltage distribution network. Background Technology

[0002] With the rapid development of new power systems, diverse services such as distributed generation grid connection, V2G (vehicle-to-grid) interaction, flexible load response, and low-voltage power quality monitoring and management place higher demands on the communication transmission and computing power capabilities of distribution networks. However, existing distribution network communication and computing resources often operate independently, lacking effective coordination mechanisms. This leads to problems such as communication congestion, idle or insufficient computing power, severely restricting the real-time processing efficiency of distribution network services and the overall stability and economy of the system. Especially on the low-voltage side, with diverse equipment, varied scenarios, massive amounts of data, and high real-time requirements, traditional isolated communication and computing power management models are no longer sufficient to meet the development needs of new power distribution systems.

[0003] To meet the development needs of new power distribution systems, a commonly used computing power scheduling method is a data-driven resource scheduling approach that combines satellite-ground converged networks and service models, specifically optimized for the transmission and computing requirements of power services. By allocating appropriate application devices to different types of service data (transmission-sensitive and computation-sensitive), the computing resources of these devices can be utilized for service processing, thereby effectively achieving real-time scheduling of system resources and meeting application requirements.

[0004] However, the above method has the following technical problems: Although resource allocation according to different types can meet application requirements, in the low-voltage distribution network environment with diverse equipment, diverse scenarios, massive data and high real-time requirements, the communication and computing resources of each device will change constantly. Changes in various factors may cause the computing resources to be mismatched with the task or difficult to meet the task requirements. Continuing to run the equipment resources independently for task processing will reduce processing efficiency and reduce the flexibility of resource scheduling. Summary of the Invention

[0005] This invention provides a method, apparatus, system, equipment, and medium for scheduling computing resources in a low-voltage distribution network, which can solve the technical problems of low resource scheduling flexibility and reduced task processing efficiency in existing technologies.

[0006] A first aspect of this invention provides a method for scheduling computing resources in a low-voltage distribution network, the method comprising: Obtain resource information and service information of the low-voltage side distribution network, wherein the resource information is information on communication resources and computing resources within a preset time interval; Based on the resource information, a communication curve for the data volume is constructed, and the arrival delay for processing the service information is determined according to the communication curve, wherein the communication curve is a curve for processing data with computing resources; When it is determined that the arrival delay is increasing, a resource scheduling decision is generated based on the resource information and the business information using a preset machine learning algorithm. Resource scheduling is performed on the low-voltage side distribution network based on the resource scheduling decision.

[0007] This invention can determine the delay of business tasks based on real-time resource and business processing information. When the delay of the processing task is not met, a resource scheduling decision is generated based on the real-time status, and then resource scheduling is carried out based on the resource scheduling decision. Not only can the resource scheduling be aligned with the actual status and improve the flexibility of resource scheduling, but it can also improve the processing efficiency of tasks and shorten the processing time.

[0008] A second aspect of the present invention provides a computing resource scheduling device for a low-voltage distribution network, the device comprising: The acquisition module is used to acquire resource information and service information of the low-voltage side distribution network, wherein the resource information is information on communication resources and computing resources within a preset time interval; The determination module is used to construct a communication curve of the data volume based on the resource information, and determine the arrival delay of processing the service information according to the communication curve, wherein the communication curve is a curve of computing power resources processing data; The decision module is used to generate a resource scheduling decision based on the resource information and the business information using a preset machine learning algorithm when it is determined that the arrival delay is increasing. The scheduling module is used to perform resource scheduling processing on the low-voltage side distribution network according to the resource scheduling decision.

[0009] A third aspect of this invention provides a computing resource scheduling system for a low-voltage distribution network. The system is applicable to the computing resource scheduling method for the low-voltage distribution network described above. The system includes: a communication queue module, a computing queue module, and an interactive response module. The interactive response module is communicatively connected to both the communication queue module and the computing queue module, and the communication queue module and the computing queue module are communicatively connected. The communication queue module is used to collect data packets from various communication links of the low-voltage side distribution network and transmit the data packets to the target computing node of the computing queue module. The data packets are business data to be processed. The computing queue module is used to perform resource processing on the data packet using the target computing node. The resource processing includes data parsing, protocol conversion, edge analysis, model inference, and control command generation. The interactive response module is used to provide feedback adjustments to the target computing node transmitted by the communication queue module based on the resource processing results of the computing queue module.

[0010] Compared to existing technologies, the present invention provides a method, apparatus, system, equipment, and medium for scheduling computing resources in a low-voltage distribution network. Its advantages include: acquiring resource and service information of the low-voltage distribution network; constructing a communication curve for the data volume based on the resource information; determining the arrival delay for processing the service information based on the communication curve; generating resource scheduling decisions based on the resource and service information using a preset machine learning algorithm when the arrival delay is increasing; and performing resource scheduling processing on the low-voltage distribution network based on the resource scheduling decisions. The present invention can determine the delay of service tasks based on real-time resource and service processing information. When the delay of the processing task is not met, a resource scheduling decision is generated based on the real-time status, and then resource scheduling processing is performed based on the resource scheduling decision. This not only allows resource scheduling to closely match the actual situation and improve the flexibility of resource scheduling, but also improves task processing efficiency and shortens processing time. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a method for scheduling computing resources in a low-voltage distribution network according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a computing resource scheduling device for a low-voltage distribution network according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computing resource scheduling system for a low-voltage distribution network according to an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] With the rapid development of new power systems, diverse services such as distributed generation grid connection, V2G (vehicle-to-grid) interaction, flexible load response, and low-voltage power quality monitoring and management place higher demands on the communication transmission and computing power capabilities of distribution networks. However, existing distribution network communication and computing resources often operate independently, lacking effective coordination mechanisms. This leads to problems such as communication congestion, idle or insufficient computing power, severely restricting the real-time processing efficiency of distribution network services and the overall stability and economy of the system. Especially on the low-voltage side, with diverse equipment, varied scenarios, massive amounts of data, and high real-time requirements, traditional isolated communication and computing power management models are no longer sufficient to meet the development needs of new power distribution systems.

[0014] To meet the development needs of new power distribution systems, a commonly used computing power scheduling method is a data-driven resource scheduling approach that combines satellite-ground converged networks and service models, specifically optimized for the transmission and computing requirements of power services. By allocating appropriate application devices to different types of service data (transmission-sensitive and computation-sensitive), the computing resources of these devices can be utilized for service processing, thereby effectively achieving real-time scheduling of system resources and meeting application requirements.

[0015] However, the above method has the following technical problems: Although resource allocation according to different types can meet application requirements, in the low-voltage distribution network environment with diverse equipment, diverse scenarios, massive data and high real-time requirements, the communication and computing resources of each device will change constantly. Changes in various factors may cause the computing resources to be mismatched with the task or difficult to meet the task requirements. Continuing to run the equipment resources independently for task processing will reduce processing efficiency and reduce the flexibility of resource scheduling.

[0016] To address the aforementioned issues, the following specific embodiments will provide a detailed description and explanation of a computing resource scheduling method, apparatus, system, equipment, and medium for a low-voltage distribution network provided in this application.

[0017] To address the technical problem of low resource scheduling flexibility and reduced task processing efficiency in existing technologies, referencing Figure 1 The diagram shows a flowchart of a computing resource scheduling method for a low-voltage distribution network according to an embodiment of the present invention.

[0018] As an example, the computing resource scheduling method for the low-voltage side distribution network may include: S11. Obtain resource information and service information of the low-voltage side distribution network, wherein the resource information is information on communication resources and computing resources within a preset time interval.

[0019] To address the diverse equipment and complex services deployed in low-voltage distribution networks, sophisticated communication resource queue models and computing resource queue models can be pre-constructed. Information about communication resources is obtained through the communication resource queue model, and information about computing resources is obtained through the computing resource queue model.

[0020] For the communication resource queue model, the heterogeneous characteristics of various communication media, such as power distribution fiber optics, low-voltage carrier source-load-storage sensing, and power wireless sensing technology, are considered. Each communication link... It can be modeled as a queuing system with available bandwidth. Real-time delay and packet arrival rate Dynamically changing. Data packets can be defined on the link. The queue length is Specifically, low-voltage carrier communication needs to consider channel attenuation and noise interference, while power wireless sensing needs to consider signal strength and multipath effects. Various sensor data, such as distributed fiber optic temperature measurement, low-voltage carrier node power data, and wireless sensor network (WSN) status information, will be encapsulated according to a unified data format, such as using the data packet structure of IEC 61850 or other lightweight IoT protocols (such as MQTT), including metadata such as data type, source node ID, timestamp, service priority, and quality of service (QoS) requirements, for differentiation and scheduling in the communication queue. Specifically, communication resource information can include dynamically changing data such as bandwidth, real-time latency, channel status, and packet arrival rate for different communication media such as power distribution fiber optics, low-voltage carriers, and wireless sensing.

[0021] For the computing resource queue model, the computing units on the low-voltage side include smart switches, edge RTUs, local microcontrollers, and small edge computing servers. Each computing node... Available computing power (For example, MIPS or FLOPS can be used), current task processing queue length and task arrival rate All are dynamic parameters. Each computational task Its computing power requirements can be defined. (Periodic or sudden), data volume and the maximum tolerable latency For example, V2G control commands may require extremely low latency and high reliability, while historical data analysis tasks can tolerate higher latency. Computational resource information may include: available computing power, task arrival rate, task processing queue length, and task computation requirements (such as latency requirements and computational complexity) for each computing node (such as smart switches, edge RTUs, local microcontrollers, etc.).

[0022] The acquired business information may include: data from different services (such as distributed fiber optic temperature measurement, load data, power quality analysis, etc.) also need to be collected and scheduled according to the service priority, service quality requirements, etc.

[0023] The collection of these data is based on monitoring and analyzing the data flow of power distribution network communication, computing resources and various services. The model optimizes scheduling decisions through the feedback of this real-time data.

[0024] Utilizing models to collect diverse data effectively supports resource scheduling optimization in low-voltage distribution networks. Data acquisition on the communication and computing sides primarily relies on standard protocols and lightweight message formats, such as IEC 61850 and MQTT, ensuring consistency and interoperability of data during acquisition, transmission, and scheduling. Key communication-side data (such as bandwidth, latency, and packet arrival rate) and computing-side resource information (such as computing power, task queue length, and load) can be periodically collected and reported through lightweight agents on edge devices, guaranteeing the system's responsiveness under high real-time requirements.

[0025] It should be noted that the time synchronization of data acquisition can also be effectively guaranteed through industry-standard time synchronization protocols (such as PTP and NTP), ensuring the timeliness and accuracy of various types of data, which is crucial for subsequent queuing delay calculation and resource scheduling.

[0026] Furthermore, the constructed model, through refined queue modeling and deep coupling of communication and computing resources, can accurately characterize the latency and resource utilization of data packets in the low-voltage distribution network. Utilizing arrival-departure curve theory, it monitors the load status of communication links and computing nodes in real time, thereby identifying bottlenecks and dynamically adjusting resource scheduling strategies. The core function of the model is to optimize the coordinated scheduling of communication and computing resources. Through interactive scheduling algorithms and feedback mechanisms, it achieves efficient resource allocation, avoids resource waste, and improves the overall system response speed and processing efficiency.

[0027] S12. Construct a communication curve for the data volume based on the resource information, and determine the arrival delay for processing the service information according to the communication curve, wherein the communication curve is a curve for processing data with computing power resources.

[0028] Resource information can include dynamically changing data such as bandwidth, real-time latency, channel status, and packet arrival rate of different communication media, including power distribution fiber optics, low-voltage carrier waves, and wireless sensors. It also includes information such as available computing power, task arrival rate, task processing queue length, and task computation requirements (such as latency requirements and computational complexity) for each computing node (e.g., smart switches, edge RTUs, local microcontrollers). After obtaining the resource information, a communication curve can be constructed, which represents the amount of data processed by each node when processing business data.

[0029] Subsequently, the arrival delay of the processed service information is determined based on the communication curve. This arrival delay refers to the time it takes for the data packets of the service data to reach their computing nodes. Total delay.

[0030] This invention quantifies the latency of data packets using arrival-departure curve theory. Combined with real-time data acquisition, it can accurately identify bottlenecks in communication links and computing nodes, thereby enabling effective resource scheduling.

[0031] In an optional embodiment, constructing a communication curve for the data volume based on the resource information and determining the arrival delay for processing the service information according to the communication curve may include the following sub-steps: S121. Extract data volume parameters and time parameters from the resource information, and construct a communication curve of data volume using the data volume parameters and time parameters, wherein the data volume parameters include: departing data volume and arriving data volume.

[0032] S122. Based on the communication curve, determine the transmission delay and the waiting delay of the communication link, wherein the transmission delay is the time for transmitting the data packet of the service information, and the waiting delay is the queuing time of the data packet in the communication link.

[0033] S123. The arrival delay is calculated using the transmission delay and the waiting delay.

[0034] The communication curve can be an Arrival-Departure Curve (ADC), which can accurately characterize the waiting latency of data packets in the communication queue and the computation queue.

[0035] Specifically, constructing an Arrival-Departure Curve (ADC) involves monitoring the arrival and departure data volumes. Arrival data volume records the amount of data received by each communication link or computing node within a time interval; this data includes the number and size of data packets transmitted from upstream to the node, typically acquired through a real-time data acquisition system. Departure data volume records the amount of data that has been processed and left the communication link or computing node within the same time interval, representing the amount of data for which transmission or computation has been completed.

[0036] Based on the monitored arrival and departure data volumes and the detection time, an arrival-departure curve (ADC) is generated. This curve shows the arrival and departure of data packets in the queue at different time points. By accumulating and statistically analyzing the data at different time points, curves depicting system load and latency changes can be plotted, aiding in further analysis of queuing and waiting latency patterns.

[0037] The transmission delay and the waiting delay of the communication link can then be determined based on the communication curve. The transmission delay is the time it takes to transmit the data packet containing the service information, and the waiting delay is the queuing time of the data packet in the communication link.

[0038] Finally, the arrival delay can be calculated using transmission delay and waiting delay. Specifically, the network delay of the low-voltage distribution network can be obtained, and the sum of the network delay, transmission delay, and waiting delay can be calculated to obtain the arrival delay. Specifically, the arrival delay can be calculated as follows: = + +Network latency; In the above formula, It is a task Arrival delay; Transmission delay refers to the time it takes for a data packet to travel through the communication link. It is the waiting delay of the communication link, reflecting the queuing time of data packets in the communication link.

[0039] As an example, determining the transmission delay and the waiting delay of the communication link based on the communication curve may include the following sub-steps: S1221. Calculate the service rate and the current queued data volume using the communication curve, and calculate the ratio of the queued data volume to the service rate to obtain the waiting delay.

[0040] S1222. Calculate the transmission delay of the communication link using the communication curve.

[0041] In one embodiment, the service rate and the current queued data volume can be calculated using the communication curve, and the transmission delay can be obtained by calculating the ratio of the queued data volume to the service rate. The waiting delay refers to the time required for a data packet to go from arriving in the queue to leaving the queue.

[0042] Specifically, the waiting delay can be calculated as follows: ; In the above formula, It represents the current waiting time, indicating the average waiting time for a data packet in the queue. It represents the amount of queued data at the current moment. It is the service rate, representing the node's performance over time. The amount of data that can be processed internally.

[0043] In one embodiment, the processing operation of determining the amount of queued data may include the following sub-steps: S21. Extract the arrival data volume and departure data volume from the communication curve.

[0044] S22. Calculate the queued data volume using the arrival data volume and the departure data volume.

[0045] In one embodiment, the queued data volume refers to the amount of data currently pending processing in the communication link or computing node, reflecting the current load of the system. The queued data volume can typically be simplified and calculated using the following formula: ; Specifically, the amount of queued data can be calculated as follows: It is the current queue size, that is, the amount of data in the queue over time. At any given moment, the amount of data waiting to be processed in the queue. It is a calculation function corresponding to the amount of data arriving; it represents the time... At any given moment, the rate or quantity of data arrival. It is the calculation function corresponding to the amount of data, which represents the time... At any given moment, the amount of data that has been processed and left the queue.

[0046] The integral formula means that from time 0 to time... The current queued data volume is obtained by integrating the difference between the arriving and departing data volumes. This can also be understood as the "cumulative net arrivals".

[0047] Based on the Arrival-Departure (ADC) curve, this curve shows the arrival and departure of data packets in the queue at different points in time. By accumulating and statistically analyzing the data at different time points, curves showing changes in system load and latency can be plotted, helping to further analyze queuing and waiting latency patterns.

[0048] For a certain communication link or computing node During the time interval The amount of data arriving within and the amount of data leaving It can be monitored in real time.

[0049] Furthermore, for a more precise representation of latency, convolution and deconvolution operations on the arrival and departure curves are required. This is achieved by configuring data packets for different services within their respective queues. and Real-time tracking and prediction can accurately depict the real-time performance bottlenecks of different business types (such as distribution network automation control, distributed power source data uploading, V2G charging scheduling, power quality analysis, etc.).

[0050] This invention, through the analysis and modeling of real-time data streams, reveals the deep bidirectional coupling between communication and computing resources. This coupling is not merely a simple sequential dependency, but a complex interplay of mutual influence and constraints.

[0051] S13. When it is determined that the arrival delay is increasing, a resource scheduling decision is generated based on the resource information and the business information using a preset machine learning algorithm.

[0052] Arrival delays can be used to determine the impact of communication on computation. When a low-voltage communication link (e.g., a power wireless channel used for intensive sensor data backhaul) becomes congested, packet transmission delays increase dramatically. Even if the target edge computing node has sufficient computing power, this power remains idle or inefficient because data cannot arrive in time, leading to an increase in overall task processing latency.

[0053] Specifically, this impact can be reflected in the aforementioned formula for calculating arrival delay. If the task... In communication link The uptransmission latency is Then it reaches the computing node. Total delay Will be significantly affected Impact. Due to limitations in communication resources, such as bandwidth and link congestion, transmission delay and waiting delay both affect the arrival time of tasks.

[0054] When the processing power of a computing node is limited, communication link latency and data packet backlog may prevent the full utilization of idle computing resources, thereby increasing the overall task processing latency. Therefore, the above equation can be used to couple computing resources with communication resources.

[0055] Similarly, congestion in communication links or insufficient computing resources increase the time data packets spend waiting in the transmission queue, thus leading to latency. This delay directly affects the arrival time of tasks, resulting in an increase in overall task processing latency.

[0056] The deep coupling between communication and computing resources means that the load on communication links affects the utilization efficiency of computing resources. If a busy communication link causes delayed arrival of task data packets, the computing node may be idle, but the lack of timely data input will exacerbate the processing latency of the computing task.

[0057] It should be noted that the increase in task processing latency is not solely due to bottlenecks in single computing or communication resources, but rather to the coupling between communication and computing resources. Delays and queuing in communication links affect task arrival times, while insufficient computing resources can lead to idle or delayed task processing.

[0058] When a low-voltage edge computing node becomes overloaded due to handling complex tasks (e.g., edge AI analysis or fault diagnosis of hundreds of sensors on a power distribution line), its data ingestion and output rates decrease. This, in turn, affects the communication layer's data transmission strategy for that node, potentially leading to data backlog in the upstream communication queue and creating a "back pressure" effect. Furthermore, some computationally intensive tasks may generate a large amount of intermediate results or log data during processing; the backhaul of this data consumes additional communication bandwidth, exacerbating the burden on the communication link. Interoperability of these mutually influential data formats is crucial. For example, when a computing node reports its load status, it can use predefined JSON or XML format messages containing information such as node ID, current CPU utilization, task queue length, and a list of processable task types, ensuring that the communication scheduler can accurately understand and respond. This information transmission processing logic reflects the communication layer's ability to perceive the computing layer's state. This invention provides crucial insights for subsequent interactive scheduling by monitoring these coupling effects in real time.

[0059] As the above analysis shows, if the data transmission speed in the network slows down (e.g., bandwidth decreases, communication link congestion), then the transmission delay... This will significantly increase, leading to increased task processing latency.

[0060] Therefore, after calculating the arrival delay, it can be determined whether the arrival delay is greater than the preset delay. If so, it is determined that the task processing time has increased, and resource scheduling is required to improve the task processing efficiency. Optionally, the transmission delay at the current moment can be compared with the transmission delay at the previous moment. This method can capture the trend of transmission delay changes through "time-series dynamic comparison" and adapt to the dynamic changes of communication resources in the low-voltage distribution network and the needs of diverse business scenarios.

[0061] To improve the accuracy and efficiency of resource scheduling, pre-set machine learning algorithms can be used to generate resource scheduling decisions based on resource and business information.

[0062] Specifically, pre-defined machine learning methods (such as LSTM networks) can be used to make short-term predictions of resource and business information, forecasting resource demand over a future period. Based on resource demand, resource information, and business information, optimal or suboptimal scheduling decisions can be generated using advanced algorithms such as heuristics, reinforcement learning (such as DQN or A2C), or constrained programming.

[0063] In one embodiment, generating resource scheduling decisions based on the resource information and the business information using a preset machine learning algorithm may include the following sub-steps: S131. Preprocess the resource information and the business information to obtain processed resource information and processed business information. The preprocessing includes noise reduction processing, time synchronization processing and normalization processing.

[0064] S132. Construct a multi-objective optimization objective function based on the processing resource information and the processing business information using a preset machine learning algorithm.

[0065] S133. Solve the multi-objective optimization objective function based on the preset optimization objective to obtain the resource scheduling decision. The preset optimization objective includes: minimizing end-to-end latency, maximizing resource utilization, and timely processing of high-priority tasks.

[0066] In one embodiment, a lightweight scheduling agent runs on the low-voltage side edge computing node, which is responsible for real-time status reporting and preliminary decision-making of local resources; while a more complex global optimization algorithm runs in the regional scheduling center.

[0067] In one practical operation, after resource information (e.g., communication links (bandwidth, latency, packet arrival rate), computing nodes (CPU utilization, task queue length, available computing power)) and business information (business task (priority, data volume, maximum tolerable latency) related data) are processed, the resource information and business information can be preprocessed. Preprocessing can include noise reduction and outlier removal, time synchronization to ensure timing consistency, and normalization to unify parameter range, providing high-quality data support for subsequent processes.

[0068] Next, machine learning models are used to make short-term predictions on the preprocessed data. Communication traffic prediction uses time series models to analyze the trend of link load changes, and computational load prediction combines historical task data to predict computing power requirements. Finally, a current status assessment report and a bottleneck warning list are generated to identify potential communication or computing resource shortages in advance.

[0069] The core idea of ​​this pre-defined machine learning algorithm is to construct a multi-objective optimization problem, aiming to minimize end-to-end latency, maximize resource utilization, and ensure QoS for high-priority services. The decision variables of the scheduling algorithm include, but are not limited to, packet transmission priority. It is a business In the link Transfer priority. Task offloading decision. It is a business From the current node Unload to node Task parallelism It is a business Whether to perform parallel processing.

[0070] Specifically, construct a multi-objective optimization objective function. It can be represented as follows: ; In the above formula, For business The total end-to-end latency (including communication and computation latency). For communication links utilization rate For computing nodes Utilization rate. For business Penalties for violating its QoS (such as maximum tolerable latency). This is a weighting coefficient that can be dynamically adjusted according to actual business needs.

[0071] To achieve optimal resource allocation, the system adjusts its resources based on different optimization objectives. Common optimization objectives include minimizing end-to-end latency, maximizing resource utilization, and ensuring the timely processing of high-priority tasks. For high-priority tasks, the system prioritizes the allocation of communication and computing resources to ensure their completion within the specified timeframe.

[0072] This analysis uses reinforcement learning as an example: reinforcement learning automatically adjusts resource scheduling strategies through the interaction between the environment and the agent. Reinforcement learning is particularly suitable for making scheduling decisions in complex and dynamically changing systems.

[0073] When solving the above objective function, the state can be defined based on resource and business information. Specifically, defining the state can include: the current state of the system, including information such as communication links, computing node load, and task priorities. Communication links: bandwidth, latency, current load. Computing nodes: CPU utilization, task queue length, memory load. Task information: task priority, data volume, latency requirements, etc.

[0074] Next, the specific actions of the scheduling strategy can be defined, such as: Task scheduling: selecting which computing node to schedule the task to; Route adjustment: selecting different communication links to avoid network bottlenecks; Task offloading: offloading non-real-time tasks to the cloud or computing nodes with lower load.

[0075] Finally, the reward is defined. The reward function evaluates the effectiveness of scheduling decisions based on the overall performance of the system. Common reward criteria include: Latency: A reward is given if latency decreases; a penalty is given if latency increases. Resource utilization: A reward increases if resources are used effectively; a penalty increases if resources are idle or overloaded. Task completion: A reward is given if the task is completed within the specified time limit; otherwise, a penalty is given.

[0076] In the multi-objective optimization objective function construction phase, three main objectives—minimizing end-to-end latency, maximizing resource utilization, and ensuring QoS for high-priority services—are integrated to construct a weighted summation objective function. The weight coefficients in the function are dynamically adjusted according to service priority, and the parameters cover end-to-end latency, communication link utilization, computing node utilization, and QoS violation penalties, transforming the multi-objective requirements into a quantifiable mathematical model.

[0077] In the objective function solution phase, reinforcement learning combined with heuristic algorithms is used. First, the reinforcement learning environment is initialized, mapping the state, action, and reward space. Then, the agent performs a cyclical optimization process of "execute action - obtain reward - update policy," combined with heuristic pruning to reduce invalid actions. Finally, for multiple optimal solutions, the final optimal action combination is selected based on business QoS constraints.

[0078] In the solution verification phase, the optimal action combination is pre-verified through simulation testing or small-scale pilot tests. The focus is on checking whether the service latency meets the tolerance threshold and whether the resource utilization rate is within a reasonable range (communication link ≤ 90%, computing node ≤ 85%). If the verification fails, the prediction model or objective function parameters are adjusted backtracking to ensure the validity of the results.

[0079] The solution results are presented in a specific set of scheduling instructions, and an adaptation solution is output for different scenarios: in communication bottleneck scenarios, the service route is adjusted to an idle link and the QoS level is improved; in computing overload scenarios, non-real-time tasks are offloaded to the cloud or idle nodes; in mixed bottleneck scenarios, routing adjustment and task scheduling are combined to balance latency and resource utilization requirements.

[0080] S14. Perform resource scheduling processing on the low-voltage side distribution network according to the resource scheduling decision.

[0081] In one practical operation, scheduling instructions can be generated based on resource scheduling decisions, and these instructions (such as data packet priority adjustment, task unloading instructions, and route switching instructions) can be sent to the communication equipment and computing nodes on the low-voltage side for execution.

[0082] For example, when a sudden increase in the amount of power quality data of a low-voltage power distribution line is detected and a bottleneck occurs in the reporting link, the algorithm will determine the priority of the service. If it is high, the routing may be adjusted and the QoS level may be improved. If the corresponding edge computing node is overloaded, some non-real-time analysis tasks may be offloaded to the cloud or nearby idle nodes.

[0083] During the scheduling process, the scheduling effect (such as actual latency and resource utilization) can be monitored in real time, and the evaluation results can be used as feedback information to input into the next round of scheduling decisions, forming a closed-loop optimization.

[0084] This interactive scheduling enables intelligent, dynamic, and collaborative scheduling of communication and computing resources on the low-voltage side of the distribution network, thereby effectively improving the real-time processing efficiency of diverse services and the overall system performance.

[0085] In this embodiment, the present invention provides a method for scheduling computing resources in a low-voltage distribution network. Its advantages are as follows: the present invention can acquire resource information and service information of the low-voltage distribution network; construct a communication curve for the data volume based on the resource information, and determine the arrival delay for processing the service information according to the communication curve; when it is determined that the arrival delay is increasing, a preset machine learning algorithm is used to generate a resource scheduling decision based on the resource information and the service information; and resource scheduling processing is performed on the low-voltage distribution network according to the resource scheduling decision. The present invention can determine the delay of service tasks based on real-time resource and service processing related information. When the delay of the processing task is not met, a resource scheduling decision is generated based on the real-time status, and then resource scheduling processing is performed according to the resource scheduling decision. This not only allows resource scheduling to closely match the actual state and improve the flexibility of resource scheduling, but also improves the processing efficiency of tasks and shortens the processing time.

[0086] This invention also provides a computing resource scheduling device for low-voltage distribution networks, see [link to relevant documentation]. Figure 2 The diagram shows a schematic of the structure of a computing resource scheduling device for a low-voltage distribution network according to an embodiment of the present invention.

[0087] As an example, the computing resource scheduling device for the low-voltage side distribution network may include: The acquisition module 201 is used to acquire resource information and service information of the low-voltage side distribution network, wherein the resource information is information on communication resources and computing resources within a preset time interval; The determining module 202 is used to construct a communication curve of the data volume based on the resource information, and determine the arrival delay of processing the service information according to the communication curve, wherein the communication curve is a curve of computing power resources processing data; Decision module 203 is used to generate resource scheduling decisions based on the resource information and the business information using a preset machine learning algorithm when it is determined that the arrival delay is increasing; The scheduling module 204 is used to perform resource scheduling processing on the low-voltage side distribution network according to the resource scheduling decision.

[0088] Optionally, the step of constructing a communication curve for the data volume based on the resource information and determining the arrival delay for processing the service information according to the communication curve includes: Extract data volume parameters and time parameters from the resource information, and construct a communication curve of the data volume using the data volume parameters and time parameters, wherein the data volume parameters include: departing data volume and arriving data volume; Based on the communication curve, the transmission delay and the waiting delay of the communication link are determined respectively, wherein the transmission delay is the time for transmitting the data packet of the service information, and the waiting delay is the queuing time of the data packet in the communication link; The arrival delay is calculated using the transmission delay and the waiting delay.

[0089] Optionally, determining the transmission delay and the waiting delay of the communication link based on the communication curve includes: The service rate and the current queued data volume are calculated using the communication curve, and the waiting delay is obtained by calculating the ratio of the queued data volume to the service rate. The transmission delay of the communication link is calculated using the communication curve.

[0090] Optionally, the processing operation for determining the queued data volume includes: Extract the arriving and departing data volumes from the communication curve; The queuing data volume is calculated using the arrival data volume and the departure data volume.

[0091] Optionally, the step of generating resource scheduling decisions based on the resource information and the business information using a preset machine learning algorithm includes: The resource information and the business information are preprocessed to obtain processed resource information and processed business information. The preprocessing includes noise reduction processing, time synchronization processing and normalization processing. Based on a preset machine learning algorithm, a multi-objective optimization objective function is constructed according to the processing resource information and the processing business information; The multi-objective optimization objective function is solved based on the preset optimization objectives to obtain the resource scheduling decision. The preset optimization objectives include: minimizing end-to-end latency, maximizing resource utilization, and timely processing of high-priority tasks.

[0092] This invention also provides a computing resource scheduling system for low-voltage distribution networks, see [link to relevant documentation]. Figure 3 The diagram shows a schematic of the structure of a computing resource scheduling system for a low-voltage distribution network according to an embodiment of the present invention.

[0093] The system is applicable to the computing resource scheduling method for low-voltage distribution networks as described in the above embodiments. As an example, the computing resource scheduling system for low-voltage distribution networks may include: The module consists of a communication queue module, a computation queue module, and an interactive response module. The interactive response module is communicatively connected to both the communication queue module and the computing queue module, and the communication queue module and the computing queue module are communicatively connected. The communication queue module is used to collect data packets from various communication links of the low-voltage side distribution network and transmit the data packets to the target computing node of the computing queue module. The data packets are business data to be processed. The computing queue module is used to perform resource processing on the data packet using the target computing node. The resource processing includes data parsing, protocol conversion, edge analysis, model inference, and control command generation. The interactive response module is used to provide feedback adjustments to the target computing node transmitted by the communication queue module based on the resource processing results of the computing queue module.

[0094] Specifically, the Communication Queue module serves as the first level, handling the transmission and queuing of data on various low-voltage communication links (such as power distribution fiber optics, low-voltage carrier waves, and power line wireless) before it reaches edge computing nodes or centralized processing units. Data packets (e.g., electricity data collected from smart meters, real-time status of V2G charging piles, and output information from distributed power sources) first enter this queue to await transmission resources. This invention models the communication queue considering factors such as transmission bandwidth, channel conditions, real-time congestion, and multi-hop delay. For example, the queuing delay of data packets... Depends on current link load and service rate The specific formula is as follows: ; in, This represents the length of the communication queue.

[0095] The computation queue module serves as the second level. After a data packet successfully passes through the communication queue and is transmitted to the target computing node, it enters this level queue to await processing by computing resources. This includes data parsing, protocol conversion, edge analysis, model inference (such as load prediction or fault diagnosis based on deep learning), and control command generation. This invention models the computation queue considering factors such as the computing node's computing capacity, current task load, processor utilization, and task processing time. Similarly, the queuing latency of data packets in the computation queue... It depends on the current compute node's load and processing power. The specific formula is as follows: ; in, To calculate the queue length.

[0096] The communication queue module and the computing queue module do not flow in one direction, but form a closed loop through the feedback link and the interactive response module.

[0097] The feedback link is used to transmit real-time status information of the computation queue (such as current computing power utilization, task queue length, expected task completion time, and even the health status of compute nodes) back to the scheduler of the communication queue in real time. The data format of this feedback information can be structured lightweight messages to ensure low-latency transmission.

[0098] The interactive response module, acting as an intelligent decision-making unit, is deployed at the interface between the communication and computing layers. It is responsible for parsing feedback information from the computing queue and adjusting the communication queue's scheduling strategy accordingly. For example, if the computing queue reports that an edge computing node is overloaded, the interactive response module will instruct the communication scheduler to reduce the amount of data sent to that node or route new computing task data to other idle nodes. Conversely, if the computing queue reports that computing resources are idle, it can request the communication layer to accelerate data transmission. This mechanism ensures that the communication scheduler can "sense" the status of computing resources and make optimization decisions accordingly.

[0099] The interactive response module can invoke the communication scheduler to make scheduling decisions. When making these decisions, it is no longer limited to the load of the communication link itself, but can also acquire and respond to the availability and demand of backend computing resources in real time. This enables efficient air interface resource scheduling with computing power awareness and significantly reduces the average end-to-end packet latency. This "computing power awareness" means that the scheduler can accurately understand the current processing capacity, task queue length, and expected processing latency of the target computing node. For example, when a high-priority distribution network fault indication signal data packet arrives, the scheduler will not only select the best communication path but also actively query the real-time computing power status of all computing nodes at the end of that path that may process the signal, ensuring that the data can be received and processed quickly.

[0100] Based on this comprehensive perception capability, the scheduling algorithm can dynamically and intelligently optimize scheduling strategies according to the coupling state of communication queues and computing queues, as well as the priorities of various services and strict QoS requirements. This includes reserving resources and adjusting priorities for critical low-latency services (such as relay protection control commands and trip signals), and ensuring the highest priority for their transmission and processing by adjusting the service order and weight in the communication queues and computing queues.

[0101] The resource scheduling steps performed by the first-level communication queue module are as follows: First, the real-time monitoring module identifies key low-latency services (such as relay protection control commands and trip signals). Based on the priority tags and QoS latency requirements in the service metadata, dedicated communication resources are reserved for these services. For example, fixed bandwidth slices are allocated (e.g., 20% of the link bandwidth is reserved for trip signals) or dedicated channels are assigned to avoid competing for resources with ordinary services. Next, the service order of the communication queue is adjusted, placing high-priority service data packets at the top and prioritizing their entry into the transmission scheduling process. At the same time, a queue weight of 2-3 times higher than that of ordinary services is set for them to ensure that they can still obtain transmission resources with priority when the link load fluctuates. Then, the status of the communication queue is tracked in real time. If the transmission latency of high-priority services is detected to be close to the QoS threshold (e.g., the original latency is 10ms, and it is close to the threshold of 15ms), an early warning is immediately sent to the scheduling center through the feedback link. The dedicated bandwidth is dynamically expanded or the transmission ratio of low-priority services is temporarily reduced (e.g., non-real-time data upload is suspended). Finally, after the high-priority service transmission is completed, the reserved resources are released to the public resource pool to avoid long-term idle resources, forming a closed loop of "identification-reservation-scheduling-feedback-release" to ensure its transmission priority.

[0102] The resource scheduling steps of the second-level computing queue module are as follows: After receiving data packets from the communication queue, the system first parses the task metadata, extracts high-priority service identifiers, and, combined with the real-time computing power status of the computing nodes (such as available MIPS and task queue length), reserves dedicated computing power for these tasks (e.g., locking one CPU core and reserving 512MB of memory). Then, it adjusts the processing order of computing tasks, extracting high-priority tasks from the ordinary queue and inserting them at the front of the processing sequence, prioritizing the data parsing and computation process, and setting a processing weight three times higher than ordinary tasks to ensure priority allocation of computing resources. Furthermore, based on the "high-priority data packet transmission progress" feedback from the communication queue, it initiates computing environment warm-up 1-2 data acquisition cycles in advance (e.g., loading relay protection control algorithms and initializing data processing models) to avoid waiting for computing power after tasks arrive. If the processing latency of a high-priority task is detected to be close to the tolerance threshold (e.g., threshold 20ms, current 18ms), some low-priority tasks (e.g., historical load statistics) are immediately paused, releasing computing power to supplement high-priority tasks. Finally, the processing progress is fed back to the scheduling center in real time. If insufficient computing power occurs, a subsequent task offloading mechanism is triggered to ensure the processing priority of high-priority services.

[0103] Meanwhile, if a bottleneck is detected in a specific communication path or target computing node, the algorithm can immediately trigger dynamic route switching.

[0104] First, key parameters of each communication path are collected in real time through a distributed monitoring unit, including transmission latency, bandwidth utilization, packet loss rate, and queue length. The collection period is set to 50-100ms to ensure data timeliness. Next, based on preset bottleneck judgment rules (such as transmission latency exceeding 1.2 times the service QoS requirement, bandwidth utilization exceeding 90% for three consecutive cycles, and queue length exceeding 80% of maximum capacity), the status of each path is evaluated. If a path meets the bottleneck conditions for three consecutive cycles, it is determined to be congested, and dynamic route switching is immediately triggered. Subsequently, the dispatch center calls the backup link database to select backup links that meet the criteria of "current bandwidth utilization below 50%, transmission latency ≤ service QoS threshold, consistent with the original path endpoint, and no fault records," and sorts them through multi-objective ranking (prioritizing latency). The optimal backup link is determined by selecting the link with the lowest load. Then, a route update command is generated, containing new link path information, data forwarding rules, and a switching time window (selecting a low-traffic period, such as a period where the number of packets per second is less than 30% of the peak). This command is sent to the relevant communication devices (gateways, routers) via the lightweight MQTT protocol. Upon receiving the command, the communication devices update their routing tables within the specified time window, first switching high-priority service data to the backup link, and then gradually diverting low-priority data to avoid data loss. After the switch is complete, the status of the backup link (latency, packet loss rate) is monitored in real time. If it remains stable for five consecutive cycles, the results are reported back to the scheduling center for archiving. If an anomaly occurs (such as a sudden increase in backup link load), the backup link is immediately re-selected, and a second switch is initiated, forming a closed-loop optimization.

[0105] Data is diverted to idle backup links, and offloadable computational tasks (such as training some non-real-time load prediction models and historical data mining) are offloaded.

[0106] Reference Figure 3 In one embodiment, the system further includes a task unloading module, which is connected to the computing queue module; The task unloading module is used to generate a list of tasks to be unloaded based on the resource processing results of the computing queue module, and to adjust the real-time business tasks of the target computing nodes of the computing queue module based on the list of tasks to be unloaded.

[0107] Specifically, the scheduling operations based on the task unloading module are as follows: The task unloading module identifies unloadable tasks and generates a list of tasks to be unloaded based on task attributes (non-real-time, latency tolerance ≥10s, such as load prediction model training, historical data mining) and edge node status (overloaded node CPU utilization exceeds 85%, task queue length exceeds threshold). Next, the scheduling center queries the low-voltage side edge node resource library, filtering for target nodes that are "≤5km from overloaded nodes (reducing transmission latency), current CPU utilization ≤60%, and available computing power ≥1.2 times the task requirement," prioritizing nearby idle nodes. Then, the overloaded node establishes a secure communication connection with the target node using a lightweight protocol. The task data is compressed in Buffers format (e.g., compression rate of 50%) and the task file (including task parameters, data samples, and execution requirements) is transmitted via TCP protocol. A breakpoint resumption mechanism is enabled to prevent data loss due to transmission interruptions. Upon receiving the data, the target node verifies its integrity. If correct, it allocates computing resources and initiates the task execution process. During execution, the target node provides real-time progress updates (completion percentage, computing power utilization) to the scheduling center, while overloaded nodes synchronously monitor their own load changes (e.g., CPU utilization dropping below 70%). After task completion, the target node sends the results (e.g., model files, analysis reports) back to the original overloaded node. The original node verifies the validity of the results; if they meet the requirements, the unloading process terminates. If verification fails, the original node requests the target node to re-execute or replace the target node. Finally, the unloading effect is evaluated (e.g., the load reduction of the overloaded node, total task latency), and feedback is sent to the scheduling center to optimize subsequent unloading strategies.

[0108] It can be offloaded from overloaded edge nodes to other low-pressure edge nodes with lower load, or even uploaded to cloud data centers with stronger computing power for processing.

[0109] First, the task unloading module identifies tasks to be uploaded to the cloud (non-real-time tasks requiring ultra-low voltage edge node computing power, such as large-scale historical electricity consumption data analysis). Combined with the overloaded edge node status (CPU utilization exceeding 90%, no nearby idle edge nodes), a cloud upload task list is generated. Next, the scheduling center assesses the cloud data center status, queries the available computing power (remaining FLOPS) of each cloud node, and the transmission latency with the edge nodes (obtained through Ping tests, prioritizing cloud nodes with latency ≤500ms). It calculates the combined value of "transmission latency + computing latency" and selects the optimal cloud node. Then, the edge node establishes an encrypted communication connection with the target cloud (using TLS 1.3 protocol), preprocesses the task data (removing duplicate and redundant data, converting to a cloud-compatible format (such as Parquet), and compressing data volume) to reduce transmission volume and time consumption. After preprocessing, the edge node uploads the task data in batches, receiving a confirmation signal from the cloud after each batch is uploaded. If a transmission interruption occurs, it triggers a resume function. After receiving the data, the cloud allocates dedicated computing power (such as GPU clusters or high-memory CPU nodes) through the resource scheduling module and loads the task execution environment (such as Python). In a TensorFlow environment, the task computation is initiated. During execution, the cloud sends progress feedback (such as data processing volume and computing power utilization) to the edge nodes every 5 minutes, and the edge nodes monitor the transmission link status (such as bandwidth stability). After the task is completed, the cloud compresses the result file (such as ZIP format) and sends it back to the edge nodes. After receiving the result, the edge nodes verify its integrity and correctness. If it meets the requirements, the results are stored and the process is terminated. If there is an anomaly, the cloud is requested to re-execute or investigate the problem (such as data format error). Finally, the upload effect (total latency and edge node load reduction) is fed back to the scheduling center to provide a basis for subsequent cloud upload decisions.

[0110] This decision-making process fully considers the comprehensive optimization of data transmission latency and computation processing latency. Furthermore, for certain non-real-time but large-volume services, such as batch log uploads or periodic data inspections, the algorithm can schedule operations when communication and computing resources are relatively idle, or implement traffic shaping to smooth data transmission and computational load, avoiding the impact of instantaneous peaks on real-time services.

[0111] Through this interactive scheduling mechanism based on collaborative computation, this invention effectively avoids the resource waste and inefficiency problems of "smooth communication but computational congestion" or "idle computation but unreachable data" in traditional modes. The average latency of data packets in the entire transmission-processing chain is significantly reduced, and the overall end-to-end latency is also significantly reduced. It can be represented as: in For transmission delay, To address latency, this mechanism is crucial for ensuring the performance of high real-time services such as distribution network automation control, distributed power source optimization, and V2G. Ultimately, it ensures the efficient and coordinated use of communication and computing resources, significantly improving the overall response speed and operational efficiency of the entire power distribution system in the face of increasingly complex business demands.

[0112] By employing a tiered sampling and data aggregation strategy, the system can control bandwidth and computational load while ensuring real-time performance and accuracy, avoiding excessive overhead. Furthermore, edge computing and feedback mechanisms can dynamically sense resource load, optimize task offloading and parallelism, and ensure efficient system operation. In summary, using this model to collect data is not only technically feasible but also provides strong data support and an optimization foundation for resource scheduling in low-voltage distribution networks.

[0113] Those skilled in the art will understand that, for ease of description and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0114] Furthermore, this application also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the computing power resource scheduling method for the low-voltage side distribution network as described in the above embodiments.

[0115] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer-executable program, which is used to cause a computer to execute the computing resource scheduling method for the low-voltage side distribution network as described in the above embodiments.

[0116] In the description of the embodiments of the present invention, it should be noted that the terms "above," "below," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. When an element such as a layer, region, or substrate is referred to as being "above" or "on top of" another element, it may be directly on the other element, or there may be an intermediate element. Conversely, when an element is referred to as being "directly on" or "above" another element, there is no intermediate element. It should also be understood that when an element is referred to as being "below" or "under" another element, it may be directly below or under the other element, or there may be an intermediate element. Conversely, when an element is referred to as being "directly below" or "under" another element, there is no intermediate element. Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0117] Those skilled in the art will understand that embodiments of this application may also include computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), devices, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for scheduling computing resources in a low-voltage distribution network, characterized in that, The method includes: Obtain resource information and service information of the low-voltage side distribution network, wherein the resource information is information on communication resources and computing resources within a preset time interval; Based on the resource information, a communication curve for the data volume is constructed, and the arrival delay for processing the service information is determined according to the communication curve, wherein the communication curve is a curve for processing data with computing resources; When it is determined that the arrival delay is increasing, a resource scheduling decision is generated based on the resource information and the business information using a preset machine learning algorithm. Resource scheduling is performed on the low-voltage side distribution network based on the resource scheduling decision.

2. The computing resource scheduling method for low-voltage distribution networks according to claim 1, characterized in that, The step of constructing a communication curve for the data volume based on the resource information, and determining the arrival delay for processing the service information based on the communication curve, includes: Extract data volume parameters and time parameters from the resource information, and construct a communication curve of the data volume using the data volume parameters and time parameters, wherein the data volume parameters include: departing data volume and arriving data volume; Based on the communication curve, the transmission delay and the waiting delay of the communication link are determined respectively, wherein the transmission delay is the time for transmitting the data packet of the service information, and the waiting delay is the queuing time of the data packet in the communication link; The arrival delay is calculated using the transmission delay and the waiting delay.

3. The computing resource scheduling method for low-voltage side distribution networks according to claim 2, characterized in that, The determination of transmission delay and communication link waiting delay based on the communication curve includes: The service rate and the current queued data volume are calculated using the communication curve, and the waiting delay is obtained by calculating the ratio of the queued data volume to the service rate. The transmission delay of the communication link is calculated using the communication curve.

4. The computing resource scheduling method for low-voltage side distribution networks according to claim 2, characterized in that, The processing operation for determining the amount of queued data includes: Extract the arriving and departing data volumes from the communication curve; The queuing data volume is calculated using the arrival data volume and the departure data volume.

5. The computing resource scheduling method for low-voltage distribution networks according to claim 1, characterized in that, The step of generating resource scheduling decisions based on the resource information and the business information using a preset machine learning algorithm includes: The resource information and the business information are preprocessed to obtain processed resource information and processed business information. The preprocessing includes noise reduction processing, time synchronization processing and normalization processing. Based on a preset machine learning algorithm, a multi-objective optimization objective function is constructed according to the processing resource information and the processing business information; The multi-objective optimization objective function is solved based on the preset optimization objectives to obtain the resource scheduling decision. The preset optimization objectives include: minimizing end-to-end latency, maximizing resource utilization, and timely processing of high-priority tasks.

6. A computing resource scheduling device for a low-voltage distribution network, characterized in that, The device includes: The acquisition module is used to acquire resource information and service information of the low-voltage side distribution network, wherein the resource information is information on communication resources and computing resources within a preset time interval; The determination module is used to construct a communication curve of the data volume based on the resource information, and determine the arrival delay of processing the service information according to the communication curve, wherein the communication curve is a curve of computing power resources processing data; The decision module is used to generate a resource scheduling decision based on the resource information and the business information using a preset machine learning algorithm when it is determined that the arrival delay is increasing. The scheduling module is used to perform resource scheduling processing on the low-voltage side distribution network according to the resource scheduling decision.

7. A computing resource scheduling system for a low-voltage distribution network, characterized in that, The system is applicable to the computing power resource scheduling method for low-voltage side distribution networks as described in any one of claims 1-5, and the system includes: a communication queue module, a computing queue module, and an interactive response module; The interactive response module is communicatively connected to both the communication queue module and the computing queue module, and the communication queue module and the computing queue module are communicatively connected. The communication queue module is used to collect data packets from various communication links of the low-voltage side distribution network and transmit the data packets to the target computing node of the computing queue module. The data packets are business data to be processed. The computing queue module is used to perform resource processing on the data packet using the target computing node. The resource processing includes data parsing, protocol conversion, edge analysis, model inference, and control command generation. The interactive response module is used to provide feedback adjustments to the target computing node transmitted by the communication queue module based on the resource processing results of the computing queue module.

8. The computing resource scheduling system for low-voltage distribution networks according to claim 7, characterized in that, The system further includes: a task unloading module, which is connected to the computing queue module; The task unloading module is used to generate a list of tasks to be unloaded based on the resource processing results of the computing queue module, and to adjust the real-time business tasks of the target computing nodes of the computing queue module based on the list of tasks to be unloaded.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the computational resource scheduling method for a low-voltage distribution network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which are used to cause a computer to execute the computing resource scheduling method for a low-voltage distribution network as described in any one of claims 1-7.

Images