Energy digitization platform resource scheduling method based on cloud edge cooperative computing

Abstract

The invention provides an energy digitization platform resource scheduling method based on cloud edge cooperative computing, which comprises the following steps: acquiring real-time supply and demand data, an energy price signal and network topology information from a distributed energy management system, and preprocessing to obtain a structured dynamic supply and demand scene data set meeting a unified format requirement; aiming at a dynamic supply and demand scene data set, respectively detecting the fluctuation frequency and amplitude of an energy price on different time scales by adopting a time sequence analysis method, detecting the change condition of a network topology structure in real time by adopting a network analysis technology, and extracting key parameters reflecting scene dynamic characteristics from the change condition; and extracting a scheduling demand of cross-regional energy flow from the adjusted edge node permission configuration, and optimizing a cross-regional energy flow path in combination with real-time inter-regional supply and demand difference data and network state evaluation to obtain a globally optimized cross-regional energy scheduling scheme.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a resource scheduling method for an energy digital platform based on cloud-edge collaborative computing. Background Technology

[0002] As a key technology for achieving efficient energy utilization and green transformation, energy digitalization platforms play an irreplaceable role in addressing the global energy crisis and environmental challenges. By integrating distributed energy management systems, these platforms can optimize energy production, distribution, and consumption, promoting the intelligent and sustainable development of energy systems. However, existing solutions exhibit significant shortcomings when dealing with complex energy scenarios, particularly when regional energy supply and demand are dynamically changing. These solutions suffer from slow response times, low decision-making efficiency, and an inability to effectively adapt to rapidly changing energy network topologies, price fluctuations, and cross-regional energy flow patterns. These limitations make it difficult to balance timeliness and economy in energy dispatching, restricting the widespread application of distributed energy management systems. In distributed energy management, cloud-edge collaborative decision-making frameworks have become a core technological direction for solving dynamic energy scenarios, but the technical challenges they face cannot be ignored. The allocation of autonomous decision-making authority for edge computing nodes is a primary challenge. Due to frequent changes in regional energy supply and demand structures, such as local network topology reconfiguration or multi-frequency fluctuations in energy price signals, edge nodes need to make rapid business decisions with limited computing resources. However, the current framework's division of decision-making authority between the edge and the cloud is relatively fixed, making it difficult to flexibly adjust according to dynamic scenarios. This directly results in the inability of edge nodes to adapt their autonomous decision-making scope to changing energy flow patterns. Especially in cross-regional energy dispatch, unreasonable allocation of permissions further exacerbates decision-making delays and resource waste, impacting the overall system's economy and response efficiency. Therefore, designing an adaptive cloud-edge collaborative decision-making mechanism to dynamically adjust the autonomous decision-making scope of edge computing nodes when regional energy supply and demand structures change dynamically, while ensuring a balance between energy dispatch response time and overall system economy, has become a critical issue that distributed energy management systems urgently need to address. Summary of the Invention

[0003] This invention provides a resource scheduling method for an energy digitalization platform based on cloud-edge collaborative computing, mainly comprising:

[0004] After obtaining real-time supply and demand data, energy price signals, and network topology information from the distributed energy management system, preprocessing is performed to obtain a structured dynamic supply and demand scenario dataset that meets the requirements of a unified format.

[0005] For dynamic supply and demand scenario datasets, time series analysis methods are used to detect the frequency and magnitude of energy price fluctuations at different time scales, and network analysis technology is used to detect changes in network topology in real time, thereby extracting key parameters that reflect the dynamic characteristics of the scenario.

[0006] Based on the characteristic parameters of scene changes, a pre-established cloud-edge collaborative decision-making model is adopted. Taking the current supply and demand status, price fluctuation characteristics, and network topology characteristics as inputs, the scope of authority for each edge node to make autonomous energy scheduling decisions in the current scene is calculated, forming a dynamic authority allocation scheme.

[0007] If the autonomous decision-making scope of the edge node in the dynamic permission allocation scheme exceeds the preset computing resource threshold, resulting in the required computing resources exceeding the node's capacity, then the permission scope is compressed to obtain the adjusted edge node permission configuration. If it does not exceed the threshold, then the original dynamic permission allocation scheme is directly adopted.

[0008] The scheduling requirements for cross-regional energy flow are extracted from the adjusted edge node permission configuration. Combined with real-time inter-regional supply and demand difference data and network status assessment, the cross-regional energy flow path is optimized to obtain a globally optimized cross-regional energy scheduling scheme.

[0009] The performance of the optimized scheduling scheme in terms of response latency and resource utilization is evaluated. If the timeliness index of the scheduling scheme is lower than the preset response time threshold, the cause of the problem is determined. If it is a local factor, the scheduling scheme is locally optimized within the node's permission range. If it is a global factor, the cloud-edge collaborative decision-making model is adjusted to generate optimized energy scheduling instructions.

[0010] The optimized energy dispatch instructions are sent to edge nodes and cloud control units to obtain energy flow status data and extract key energy dispatch indicators for each node. When the latency exceeds a preset threshold or the storage utilization rate is lower than a preset threshold, the model is updated with the dispatch effect evaluation indicators as the optimization target to obtain optimized decision framework parameters.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses a resource scheduling method for an energy digitization platform based on cloud-edge collaborative computing. Addressing the energy scheduling problem in dynamic supply and demand scenarios, this invention first acquires real-time data from a distributed energy management system and uses preprocessing techniques to obtain a structured dynamic supply and demand scenario dataset. Then, dynamic feature parameters of the scenario are extracted through time series and network analysis, and the autonomous scheduling permission range of edge nodes is calculated using a cloud-edge collaborative decision model. If the permission range exceeds the computing resource threshold, it is compressed by the cloud. Based on the adjusted permission configuration, this invention optimizes cross-regional energy flow paths and generates a global scheduling scheme. To assess the performance of the scheduling scheme, this invention allows for local adjustments via edge nodes or returns adjustments to the cloud-edge collaborative model. Finally, optimized scheduling instructions are issued and execution data is acquired, triggering a feedback learning process in the decision model. This invention achieves efficient collaborative scheduling of distributed energy systems, improving system response speed and resource utilization. Attached Figure Description

[0013] Figure 1 This is a flowchart of a resource scheduling method for an energy digital platform based on cloud-edge collaborative computing, according to the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0015] like Figure 1 This embodiment of a resource scheduling method for an energy digitalization platform based on cloud-edge collaborative computing may specifically include:

[0016] S101. After obtaining real-time supply and demand data, energy price signals and network topology information from the distributed energy management system, preprocessing is performed to obtain a structured dynamic supply and demand scenario dataset that meets the requirements of a unified format.

[0017] The system acquires energy supply, consumption, and corresponding price signal data for each node at the current moment from monitoring terminals of the distributed energy management system. Simultaneously, it reads the physical connections and transmission capacity constraints between nodes in the network topology. Based on preset data quality inspection rules, outlier identification and missing value marking are performed on the collected raw data, resulting in a preliminary dataset with data quality labels. Outliers in the preliminary dataset with data quality labels are corrected using linear interpolation. Missing values ​​are filled in using historical data from adjacent time periods. Data format conversion rules are used to unify units and normalize numerical ranges for energy data from different sources, resulting in standardized energy data conforming to a unified format. The difference between the supply and consumption of each node in the standardized energy data is calculated as the supply-demand balance value. The Pearson correlation coefficient is used to calculate the correlation between the price signal and the supply-demand balance value. The energy transmission paths and capacity constraints between nodes are determined through the connections in the network topology, resulting in scenario data containing the supply-demand balance value, price correlation coefficient, and transmission constraint parameters. A relational database is used to build a multi-dimensional index for the scenario data according to node number, timestamp, and supply and demand type. An adjacency table structure is established based on the connection relationship between nodes to enable fast querying. The scenario data in the database is updated and version managed at preset time intervals to obtain a structured dynamic supply and demand scenario dataset that supports real-time querying and dynamic updating.

[0018] Specifically, in a distributed energy management system, each monitoring terminal undertakes the crucial task of data collection.

[0019] Specifically, these monitoring terminals typically include devices such as smart meters, power sensors, and temperature sensors, distributed across various nodes in the energy network. Energy supply data reflects the output power of power generation equipment such as photovoltaic panels and wind turbines at a specific moment, while consumption data records the real-time power consumption of various energy-consuming devices, including industrial loads, commercial buildings, and residential electricity consumption. Price signal data is updated in real time through market trading platforms, reflecting the current energy supply and demand relationship and market value. The network topology describes the physical connection of energy transmission lines, including key parameters such as transformer capacity and rated power of transmission lines.

[0020] It's important to note that raw data often contains various quality issues. Outlier identification employs a statistical distribution-based approach; when the power data of a node exceeds three standard deviations of its historical mean, the system marks it as an anomaly. Missing value labeling addresses data gaps caused by sensor malfunctions or communication interruptions, identified through timestamp continuity checks. This data quality identification mechanism provides crucial information for subsequent data processing, ensuring the system can distinguish between normal data and problematic data requiring special handling.

[0021] In one possible implementation, linear interpolation handles outliers by finding normal data points before and after the outlier and calculating a reasonable replacement value through linear relationships. For example, if a photovoltaic power station's output power abnormally reaches 500kW at 12:00, while the normal values ​​at 11:00 and 13:00 are 300kW and 280kW respectively, the system will calculate a corrected value of approximately 290kW for 12:00. Historical data completion utilizes the periodic characteristics of energy consumption. For instance, office building electricity consumption patterns are similar on weekdays; if data for Tuesday at 14:00 is missing, it can be filled by referring to the average value of the same time on previous Tuesdays. Data format conversion rules address the issue of consistency for multi-source heterogeneous data. Equipment from different manufacturers may use different data units, such as kW for power and MW for power, requiring the system to convert them to a standard unit. Normalization maps data of different magnitudes to a range of 0 to 1, facilitating subsequent correlation analysis and pattern recognition. This standardization process eliminates the impact of data source differences and improves the accuracy of data analysis. The calculation of the supply-demand balance directly reflects the operational status of the energy system. When the supply at a node exceeds its consumption, a positive balance value is generated, indicating that the node can supply energy; conversely, it needs to import energy from other nodes. The application of the Pearson correlation coefficient reveals the intrinsic link between price and supply and demand. A correlation coefficient close to 1 indicates a strong positive correlation between price and the supply-demand gap, meaning prices rise when demand exceeds supply; a coefficient close to -1 indicates a negative correlation. This correlation analysis provides a quantitative basis for energy dispatching and price forecasting.

[0022] Preferably, the multi-dimensional index design of the relational database considers both query efficiency and the rationality of data organization. Node number indexes support quick location of data for specific devices, timestamp indexes facilitate time-series analysis and historical backtracking, and supply-demand type indexes distinguish between different categories of data such as power generation, energy storage, and load. The adjacency list structure stores the connection relationships and transmission capacity constraints between nodes. When calculating the energy transmission path from node A to node B, the system can quickly traverse relevant connections, find feasible paths, and evaluate transmission constraints. The data update mechanism ensures the real-time nature and accuracy of system information. Preset time intervals are typically determined based on the dynamic characteristics of the energy system; for example, for highly volatile renewable energy sources, updates may occur every 5 minutes; for relatively stable basic loads, updates may occur every 15 minutes. Version management functionality retains historical data snapshots, supporting data rollback and audit trails, which is invaluable for analyzing the evolution trends of the energy system and tracing the origins of abnormal events. Through this structured data organization, the system can efficiently support the data needs of various energy management applications.

[0023] S102. For dynamic supply and demand scenario datasets, time series analysis methods are used to detect the frequency and magnitude of energy price fluctuations at different time scales, and network analysis technology is used to detect changes in network topology in real time, thereby extracting key parameters that reflect the dynamic characteristics of the scenario.

[0024] Energy price sequence data at various time points are read from a dynamic supply and demand scenario dataset. A sliding time window method is used to segment the price data. Fourier transform is used to identify the main frequency components within each time window, and the amplitude values ​​corresponding to each frequency component are calculated, resulting in price fluctuation frequency spectra and corresponding amplitude distribution data at different time scales. Based on the frequency components in the price fluctuation frequency spectrum whose amplitude exceeds a preset threshold, the dominant frequency is determined, and its reciprocal is calculated as the fluctuation period. The difference between the maximum and minimum prices within each time window is counted as the fluctuation amplitude, yielding the price fluctuation period value and fluctuation amplitude value. Based on the temporal distribution characteristics of the price fluctuation period value and fluctuation amplitude value, the network topology adjacency matrix for the corresponding time period is extracted from the dynamic supply and demand scenario dataset. Topological changes between adjacent time points are identified by calculating the difference between matrix elements. The number of topological changes per unit time is counted to obtain the reconstruction frequency, and the change in the number of connections of each node is calculated to obtain the connectivity change sequence. The average rate of change of node connectivity is calculated based on the connectivity change sequence. Combining the price fluctuation period value, fluctuation amplitude value, and topology reconstruction frequency, a dynamic feature parameter set for the scenario is constructed, resulting in a parameter set reflecting the key characteristics of the dynamic supply and demand scenario.

[0025] Specifically, the energy price series data in the dynamic supply and demand scenario dataset reflects the real-time changes in market supply and demand.

[0026] Specifically, these price data are arranged chronologically, recording the energy trading price at each sampling point. The core of the sliding window method lies in choosing an appropriate window size, such as choosing a 24-hour window to capture intraday price change patterns, or choosing a 7-day window to identify periodic fluctuation patterns. The window slides forward with a fixed step size, analyzing and processing the data within the window with each slide. The Fourier transform plays a crucial role in price analysis. This transform can decompose the time-domain price signal into a superposition of different frequency components, each with a corresponding amplitude value.

[0027] For example, when analyzing a month's worth of price data, a Fourier transform might reveal a 24-hour periodic frequency component, whose amplitude reflects the intensity of intraday price fluctuations; it might also reveal a 7-day periodic frequency component, corresponding to price differences between weekdays and weekends. Frequency components with larger amplitudes indicate that this periodic pattern dominates price changes.

[0028] In one possible implementation, the identification of the dominant frequency relies on the setting of a preset threshold. This threshold is typically determined based on the statistical characteristics of historical data, such as being set to 1.5 times the average amplitude. When the amplitude of a frequency component exceeds this threshold, it is identified as the dominant frequency. The reciprocal of the frequency is the period, because frequency represents the number of cycles per unit time, while the period represents the time required to complete one cycle. If the dominant frequency is 0.0417 Hz, the corresponding fluctuation period is 24 hours, indicating a clear daily cyclical characteristic in the price. The calculation of the fluctuation amplitude reflects the severity of price changes. The range of price fluctuations can be quantified by the difference between the maximum and minimum values ​​within a statistical time window. This combination of amplitude and period information can comprehensively characterize the dynamic characteristics of prices.

[0029] For example, energy prices in a certain region may fluctuate significantly on weekdays, while the fluctuations are relatively mild on weekends. This characteristic can be accurately captured using period and amplitude parameters.

[0030] It's important to note the inherent link between price fluctuations and network topology changes. Drastic price fluctuations are often accompanied by adjustments to energy transmission paths. The network adjacency matrix is ​​a mathematical tool describing node connectivity; its elements represent the existence of direct connections between nodes and the capacity of those connections. By comparing the adjacency matrices at adjacent time points, we can identify which connections have been established or broken, reflecting the dynamic reconfiguration process of the energy network. Calculating the difference between matrix elements provides a method for quantifying topology changes. When the corresponding elements in the adjacency matrices differ between two time points, it indicates a change in the connection state. Counting the number of these changes yields the topology reconfiguration frequency. Node connectivity is obtained by calculating the number of non-zero connections for each node; changes in connectivity reflect the dynamic evolution of a node's position within the network.

[0031] Preferably, the construction of the scenario dynamic feature parameter set comprehensively considers information from both price and topology dimensions. Short-term price cycles may correspond to intraday supply and demand fluctuations, while long-term cycles may reflect seasonal changes; average amplitude and maximum amplitude characterize the general level and extreme cases of price fluctuations, respectively; reconstruction frequency reflects the stability of the network structure; and the connectivity change rate characterizes the dynamic evolution of node importance. These parameters together constitute a multi-dimensional feature description system. By extracting these key parameters, quantitative basis can be provided for subsequent energy dispatch optimization, price prediction, and network planning. The parameter set not only reflects the current operating status but also implies the system's evolutionary trend, which is of great value for formulating forward-looking energy management strategies.

[0032] S103. Based on the characteristic parameters of the changing scenario, a pre-established cloud-edge collaborative decision-making model is adopted. Taking the current supply and demand status, price fluctuation characteristics, and network topology characteristics as inputs, the scope of authority for each edge node to make autonomous energy scheduling decisions in the current scenario is calculated, forming a dynamic authority allocation scheme.

[0033] Based on the characteristic parameters of scene changes, including supply and demand status values, price fluctuation characteristics, and network topology characteristics, these parameters are input into a pre-established cloud-edge collaborative decision-making model. This model is trained on historical data using a random forest algorithm. Through decision tree ensemble operations within the model, the scheduling authority score corresponding to each edge node is obtained. Based on the scheduling authority score of each edge node, combined with the node's current total energy reserves and equipment rated capacity parameters, the types of energy that each node can schedule are determined, including specific categories of electrical energy, thermal energy, and energy storage resources. Simultaneously, the upper limit of scheduling volume for each type of energy is calculated based on the product of the authority score and the reserves. If a node's scheduling authority score exceeds a preset threshold, its scheduling volume upper limit is adjusted according to the coefficient corresponding to the score level. At the same time, the node's neighbor node list and connection capacity information are extracted from the network topology data to obtain the node's scheduling capacity parameters and the set of tradable nodes. Based on the node's scheduling capacity parameters and the set of tradable nodes, a list of trading objects for each node is generated by comparing the energy type matching relationship and transmission capacity constraints between nodes. The list of tradable energy types, scheduling limits for each type of energy, and the list of nodes allowed to trade are integrated to obtain a dynamic permission allocation scheme containing complete permission information.

[0034] Specifically, the characteristic parameters of scene changes serve as the core input for cloud-edge collaborative decision-making, containing multi-dimensional information about the operation of the energy system.

[0035] Specifically, supply and demand status values ​​reflect the real-time energy surplus or deficit of each node. For example, a node in an industrial park might show a supply-demand difference of -500kW during peak electricity consumption, indicating a need to import energy from outside. Price fluctuation characteristics characterize market dynamics through period and amplitude parameters; for example, a 24-hour period combined with a 30% fluctuation amplitude indicates a significant peak-valley price difference. Network topology characteristics record the connection status and transmission capacity changes between nodes. The core of the cloud-edge collaborative decision-making model is the random forest algorithm, an ensemble learning method that improves decision accuracy by constructing multiple decision trees and combining their prediction results.

[0036] In one possible implementation, the model training process utilizes a large amount of historical scene data, with each training sample containing feature parameters at a specific time and the corresponding optimal permission allocation result. During the decision tree ensemble operation, each tree independently judges based on the input feature parameters, deriving a permission score through voting or averaging.

[0037] For example, when the input shows that a node is in a state of oversupply and prices are at a low point, the model may output a higher scheduling authority score, allowing the node to output energy more flexibly.

[0038] It's important to note that the scheduling authority score is a comprehensive indicator that quantifies a node's ability to make autonomous decisions in the current scenario. A higher score indicates a greater potential contribution from the node in maintaining system stability and optimizing resource allocation. This scoring mechanism considers the node's geographical advantages, historical scheduling performance, and current supply and demand status, enabling dynamic allocation of authority. Determining the type of energy involves assessing the capabilities of the node's equipment. Electricity typically comes from photovoltaic power generation, wind power generation, or the traditional power grid; heat may come from combined heat and power (CHP) equipment or dedicated heating facilities; energy storage resources include battery storage, pumped hydro storage, and other forms. Calculating the upper limit of the scheduling capacity requires considering multiple factors: the node's total energy reserves provide the physical upper limit, the rated capacity of the equipment determines the instantaneous output capacity, and the authority score serves as an adjustment coefficient.

[0039] For example, a node with 1000kWh of battery energy storage and a rated output power of 200kW may have its power dispatch limit set to 800kWh or 160kW instantaneous power when its authority score is 0.8.

[0040] Preferably, the introduction of a threshold mechanism increases the system's flexibility. When a node's scheduling authority score exceeds a preset threshold, such as 0.7, the system will increase the scheduling capacity limit based on the coefficient corresponding to the score level. This design allows for full utilization of the regulatory role of high-authority nodes in emergency situations. The score levels may be divided into multiple tiers, such as 0.7-0.8 corresponding to a 1.2x coefficient, and 0.8-0.9 corresponding to a 1.5x coefficient. Each level is verified using historical data to ensure that scheduling needs are met without exceeding safety limits. The list of adjacent nodes and connection capacity information in the network topology data are crucial for determining the transaction scope. Adjacent nodes refer to other nodes with a direct physical connection to the target node; this connection could be a power transmission line, a heat pipe, or other energy transmission channel. Connection capacity information limits the maximum rate of energy transmission between nodes; for example, a power transmission line between two nodes may have a transmission limit of 500kW. The comparison of energy type matching relationships ensures the feasibility of the transaction.

[0041] For example, a photovoltaic node with only electrical output capability cannot directly supply energy to a node that requires thermal energy unless an electrothermal conversion device is present. Transmission capacity constraints prevent overload; when multiple nodes simultaneously supply energy to a single demanding node, the total transmission volume cannot exceed the capacity of the connecting lines. The final dynamic permission allocation scheme integrates information from all dimensions. Each node receives a detailed permission list, clearly defining which types of energy can be dispatched in the current scenario, the maximum dispatch volume for each type of energy, which nodes it can trade with, and the constraints of those trades. This scheme updates in real time as scenario characteristic parameters change, ensuring the energy system can flexibly respond to various operating conditions.

[0042] S104. If the autonomous decision-making scope of the edge nodes in the dynamic permission allocation scheme exceeds the preset computing resource threshold, causing the required computing resources to exceed the node's capacity, then the permission scope is compressed to obtain the adjusted edge node permission configuration. If it does not exceed the threshold, then the original dynamic permission allocation scheme is directly adopted.

[0043] The decision parameters for each edge node are extracted from the dynamic permission allocation scheme, including the number of schedulable energy types, the maximum scheduling volume of each energy type, and the number of nodes allowed to trade. Based on the mapping relationship between the complexity of the decision parameters and the node hardware specifications, the required processor utilization and memory usage are calculated to obtain the computational resource requirements for each edge node to execute the current permission configuration. Based on the computational resource requirements of each edge node, a comparison is made with the node's preset computational resource threshold. If the requirement exceeds the threshold, the ratio of the excess to the threshold is used as the resource over-limit rate. The permission compression ratio is determined based on the over-limit rate, resulting in the node identifiers that require permission adjustments and their corresponding compression ratio parameters. For the node identifiers and compression ratio parameters requiring permission adjustments, the cloud-based system removes low-priority energy types according to their priority order, proportionally reduces the scheduling volume limit of each energy type according to the compression ratio, and eliminates distant trading partners based on the distance between nodes, resulting in a compressed permission configuration that matches the node's computational capabilities. Update the permission information of the corresponding node according to the compressed permission configuration. If the computing resource requirement of the node does not exceed the threshold, maintain the configuration in the original dynamic permission allocation scheme, merge the permission configuration information of all nodes, and obtain the complete edge node permission configuration including the compressed and adjusted nodes and the unadjusted nodes.

[0044] Specifically, the decision parameters included in the dynamic permission allocation scheme directly affect the computational load of edge nodes.

[0045] Specifically, the number of schedulable energy types determines the scale of state variables that a node needs to maintain. Each additional energy type requires extra storage space to record its real-time status, historical data, and scheduling rules. The maximum scheduling capacity for each energy type affects the complexity of optimization calculations; a larger scheduling capacity results in a wider search space for finding the optimal allocation scheme. Furthermore, the number of nodes allowed to trade directly determines the communication overhead and the scale of negotiation calculations.

[0046] In one possible implementation, there is a clear mapping between the complexity of decision parameters and hardware resource requirements. Processor utilization calculations take into account the time complexity of the decision-making algorithm; for example, when a node needs to optimize allocation among 10 energy types and 20 transaction objects, the computational load increases exponentially. Memory usage primarily depends on the size of the data structures that need to be maintained simultaneously, including the energy state matrix, transaction history, and intermediate optimization results.

[0047] For example, an edge node that manages five energy types and transacts with 15 nodes may require 60% of the processor's computing power and 2GB of memory.

[0048] It should be noted that the setting of the computing resource threshold fully considers the hardware limitations of edge nodes and the system stability requirements. The threshold is typically set at 70% to 80% of the hardware capacity, reserving space for sudden computing demands and system maintenance tasks. When the computing resource demand exceeds this threshold, nodes may experience response delays, decreased decision-making quality, or even system crashes. The resource over-limit rate, by quantifying the ratio of the excess portion to the threshold, provides a quantitative basis for subsequent permission compression. The determination of the permission compression ratio follows the principle of ensuring the basic functions of the system.

[0049] For example, when the resource over-limit rate is 1.3, it means that the demand exceeds the threshold by 30%, and the corresponding compression ratio may be set to 0.7, which means compressing the permission range to 70% of the original. This compression is not a simple proportional reduction, but needs to take into account the importance and mutual influence of different permission dimensions.

[0050] Preferably, the priority ranking of energy types is based on a comprehensive evaluation of multiple factors. Renewable energy sources such as photovoltaic and wind power typically have higher priority because their dispatch has a significant impact on the system's clean energy utilization rate; energy storage resources are secondary, serving as a buffer to balance supply and demand; traditional controllable power sources such as thermal power have lower priority. During the compression process, the system will prioritize retaining dispatch rights for high-priority energy sources and gradually eliminate low-priority types until the computational resource demand drops below the threshold. The proportional reduction of the dispatch limit ensures the relative proportional relationship between different energy types. If the original scheme had a dispatch limit of 1000kW for electricity and 500kW for thermal energy, after adjusting by a compression ratio of 0.7, they become 700kW and 350kW respectively. This approach maintains the balance of the system's energy structure and avoids excessive restriction of any type of energy due to compression. The elimination of trading partners based on the distance between nodes is of practical significance. Distance refers not only to physical distance but also to network distance, i.e., the number of relay nodes traversed. More distant trading partners mean higher transmission losses and more complex coordination mechanisms. By prioritizing the retention of close-range trading relationships, the system can maintain high energy transmission efficiency while reducing computational complexity.

[0051] For example, when reducing the number of transaction objects from 20 to 14, the system will remove nodes with a network distance exceeding 3 hops or a physical distance exceeding 10 kilometers. The complete edge node permission configuration reflects the concept of differentiated management. For nodes with strong computing power, their complete permission configuration is maintained to fully leverage their regulatory role in the system; for nodes with limited computing power, reasonable compression ensures stable operation. This configuration scheme achieves a balance between overall system performance and individual node capabilities, preventing some nodes from failing due to excessive load while maximizing the system's scheduling flexibility. Permission configurations are distributed to each edge node through a secure channel, guiding their autonomous decision-making behavior in actual operation.

[0052] S105. Extract the scheduling requirements for cross-regional energy flow from the adjusted edge node permission configuration, combine real-time inter-regional supply and demand difference data with network status assessment, optimize the cross-regional energy flow path, and obtain a globally optimized cross-regional energy scheduling scheme.

[0053] The geographical location information and jurisdiction of each node are extracted from the adjusted edge node permission configuration. Nodes are categorized into corresponding regions according to preset regional division rules. The total energy dispatch permissions of all nodes in each region are calculated. Node pairs with overlapping transaction permissions at adjacent region boundaries are identified. The potential energy exchange volume between regions is calculated based on the types of energy allowed to be traded between node pairs and their respective dispatch volume limits, resulting in cross-regional energy flow demand data labeled with exchange volume values ​​and expected exchange periods. Based on the exchange volume values ​​in the cross-regional energy flow demand data, the total energy supply and consumption of all nodes in each region at the current time are obtained. The net supply-demand difference for each region is calculated. Simultaneously, the real-time current and voltage parameters of the inter-regional backbone transmission lines are read. The remaining available transmission capacity is determined based on the difference between the line's rated capacity and the current load, resulting in inter-regional energy flow constraint parameters that include the net supply-demand difference for each region and the remaining capacity of each transmission channel. Based on the energy flow constraints between regions, a directed weighted graph is constructed with regions as nodes and transmission channels as edges. The Dijkstra algorithm is used to calculate the shortest transmission path from energy-surplus regions to energy-deficient regions. If the remaining capacity of any channel on the calculated path is less than the demand flow, the channel is removed and an alternative path is recalculated. This process is repeated until a feasible combination of paths that meets the capacity constraints is found, resulting in a list of transmission paths between each pair of regions and their corresponding flow allocation values. For the transmission path list and corresponding flow allocation values, the transmission cost per unit flow for each path is calculated based on path length, channel loss rate, and current load rate. A linear programming problem is constructed with the objective of minimizing the total transmission cost and constraints of regional supply-demand balance and channel capacity. The optimized flow values ​​for each path are obtained by solving the simplex method. By integrating path information, optimized flow values, and execution priorities determined based on flow size, a globally optimized cross-regional energy dispatch scheme is obtained.

[0054] Specifically, the adjusted edge node permission configuration includes rich node attribute information, among which the geographical location information is usually stored in the form of latitude and longitude coordinates, and the jurisdiction defines the boundary of the power supply area that the node is responsible for.

[0055] Specifically, the regional division rules are formulated based on factors such as administrative divisions, power grid zones, or energy management areas. For example, nodes within the same industrial park may be classified as one region, or regions may be defined according to the feeder range of the distribution network. This division method ensures that energy dispatch within the region has strong correlation and synergy.

[0056] In one possible implementation, the regional aggregation process of node permissions involves multi-dimensional statistical calculations. The sum of energy dispatch permissions for each region is not a simple numerical accumulation, but needs to take into account the synergistic effects between different nodes.

[0057] For example, if there are three photovoltaic (PV) power generation nodes in a region, each with a dispatch limit of 500kW, but due to similar lighting conditions, they are unlikely to reach peak output simultaneously, the total regional dispatch limit might be set at 1200kW instead of 1500kW. Identifying node pairs at adjacent regional boundaries relies on a comprehensive judgment of network topology information and geographical proximity. Calculating potential energy exchange volume requires analyzing the overlap of permissions between node pairs. When nodes at the boundary of region A and the boundary of region B both have electricity trading permissions, and the transmission channel allows bidirectional flow, the potential exchange volume depends on the smaller of the two node dispatch limit values ​​and historical trading patterns. Expected exchange periods are typically determined based on the energy consumption characteristics of each region; for example, commercial areas have high demand during the day, while industrial areas may require more energy during peak nighttime production periods.

[0058] It should be noted that the calculation of the net supply-demand difference provides a fundamental basis for cross-regional dispatch. A positive net supply-demand difference for a region indicates an energy surplus, which can be exported; a negative difference indicates an energy shortage, requiring energy to be transferred from other regions. Real-time monitoring of the current and voltage parameters of transmission lines is a key means of assessing the channel status. By measuring the real-time current of the line and comparing it with the rated current, the load level of the line can be accurately determined. The calculation of the remaining available transmission capacity also needs to consider a safety margin, typically reserving a 20% margin to cope with emergencies. The construction of a directed weighted graph abstracts the complex energy network into a mathematical model. The nodes in the graph represent various regions, and the edges represent transmission channels between regions. The weight of the edges can be indicators such as transmission distance, transmission cost, or transmission time. Dijkstra's algorithm, as a classic shortest path algorithm, can efficiently find the optimal path from the source node to the target node. In energy dispatch scenarios, the meaning of "shortest" can be flexibly defined according to the optimization objective, such as minimum transmission loss, minimum transmission cost, or fastest arrival time.

[0059] Preferably, the path recalculation mechanism when encountering capacity constraints ensures the feasibility of the solution.

[0060] For example, the shortest path from region A to region C originally passed through region B, but the remaining capacity of the transmission channel in the A-B segment is only 300MW, while the demand is 500MW. The system will remove the A-B channel and recalculate, potentially finding an alternative path A-D-C. This process may require multiple iterations, with each found path occupying corresponding channel capacity, until all demands are met or no feasible solution is confirmed. The multi-factor calculation of transmission costs reflects economic considerations. Path length directly affects line loss, potentially resulting in a 0.5% energy loss per kilometer; channel loss rate is related to the technical parameters of the line, with older lines having significantly higher loss rates than newly built lines; the current load rate affects pricing through the marginal cost principle, with higher load rates leading to greater marginal costs for new transmission. These factors combined form the actual transmission cost of each path. The application of linear programming achieves global optimization. The objective function is to minimize the total transmission cost of all paths, with constraints including that supply and demand in each region must be balanced and the flow of each channel cannot exceed its capacity limit. The simplex method, as a standard algorithm for solving linear programming problems, iteratively finds the vertices of the feasible region, ultimately finding the solution that optimizes the objective function. Determining execution priorities considers the urgency and impact of scheduling; high-volume, long-distance tasks are typically executed first to reduce the complexity of system adjustments.

[0061] Extract the inter-regional energy exchange volume and transmission time window from the edge node permission configuration, filter candidate transmission paths in combination with inter-regional transmission constraints, establish a multi-objective optimization model based on the topology of the candidate transmission paths, the model input includes supply and demand difference gradient data and historical path load balancing coefficient, iteratively calculate the multi-objective optimization model, and output the globally optimal scheduling sequence that satisfies the transmission time window.

[0062] Energy trading permission data for each regional boundary node is read from the edge node permission configuration. The energy exchange volume values ​​between regional pairs and their corresponding transmission start and end times are extracted. The transmission load factor is obtained by multiplying the exchange volume value by the transmission distance. The reciprocal of the time window length is used as the urgency coefficient. The transmission load factor and urgency coefficient are multiplied to obtain the routing priority weight, constructing a dynamic routing weight matrix with regions as nodes and weights as boundary values. Based on the regional connectivity in the dynamic routing weight matrix, the rated capacity, rated voltage, and thermal stability limit of each transmission line are obtained as transmission constraints. By comparing the actual transmission demand of candidate paths with the rated capacity of the lines and the matching degree between the voltage at both ends of the path and the rated voltage, paths with transmission demand exceeding the rated capacity or voltage mismatch are eliminated, resulting in a set of candidate transmission paths that satisfy all physical constraints. Based on the network topology of the candidate transmission path set, a transmission cost function, a load variance function, and a timeliness deviation function are constructed as three optimization objectives. The difference in supply and demand between adjacent regions is calculated as supply and demand gradient data, and the average load ratio of each path during historical operation is statistically analyzed as the path load balancing reference value. The supply and demand gradient data, the path load balancing reference value, and the candidate path parameters are integrated into the input dataset of the multi-objective optimization model. A genetic algorithm is used to process the input dataset of the multi-objective optimization model. The weighted sum of the three objective functions is used as the individual fitness evaluation criterion. New transmission scheme populations are generated through selection, crossover, and mutation operations. If the completion time of all transmission tasks in the new scheme is within the corresponding timeliness window, the scheme is retained. The iteration continues until the optimal solution no longer improves after several consecutive generations. The output is a globally optimal scheduling sequence that satisfies the timeliness window constraint and includes transmission path selection and traffic allocation for each time period.

[0063] Specifically, the energy trading information contained in the edge node permission configuration constitutes the basic data for cross-regional scheduling.

[0064] Specifically, each boundary node records the types of energy it can trade with neighboring areas, the maximum trading volume, and the allowed trading periods. The transmission time window setting reflects the urgency of energy dispatch needs; for example, an industrial area's electricity demand during peak production periods must be responded to within 30 minutes, while a residential area's heating demand may have a 2-hour buffer period.

[0065] In one possible implementation, the transmission load factor is calculated by comprehensively considering two key factors: energy quantity and distance. When 500MW of power needs to be transmitted over a distance of 50 kilometers, the transmission load factor is 25000MW·km, reflecting the degree to which the transmission task occupies network resources. The urgency coefficient is quantified by the reciprocal of the time window length; a 30-minute time window corresponds to an urgency coefficient of 2, while a 2-hour window corresponds to 0.5. The routing priority weight, as the product of the two, directly determines the priority level of the transmission task in resource contention. The construction of a dynamic routing weight matrix enables real-time reflection of the network status. Each element in the matrix represents the transmission priority between two areas; a larger weight value indicates that the path requires higher priority. This dynamic characteristic is reflected in the fact that the weight automatically increases as the time window approaches, ensuring that urgent tasks can obtain the necessary transmission resources.

[0066] It should be noted that the multi-dimensional consideration of transmission constraints ensures the safe and stable operation of the system. Rated capacity limits the maximum transmission power of the line; for example, the rated capacity of a 500kV transmission line may be 1500MW. Rated voltage requirements ensure power quality; lines of different voltage levels need to be matched using transformers. Thermal stability limits prevent equipment damage caused by line overload and are typically set at 90% of the rated capacity. The candidate path selection process employs a rigorous constraint verification mechanism.

[0067] For example, if a route needs to transmit 800MW of power, but the line's rated capacity is only 600MW, that route will be directly rejected. Voltage matching checks ensure that 220kV areas will not directly supply power to 110kV areas unless the route contains corresponding substation facilities. This screening mechanism effectively avoids physically infeasible dispatching schemes.

[0068] Preferably, the construction of the multi-objective optimization model reflects the complexity of energy dispatch. The transmission cost function considers line losses and operation and maintenance costs, with a potential 0.5% energy loss per kilometer of transmission line. The load variance function assesses the system's balance by calculating the dispersion of load rates on each path; a smaller variance indicates a more uniform load distribution. The timeliness deviation function measures the difference between the actual completion time and the required time, ensuring timely dispatch. The calculation of supply and demand gradient data reveals the spatial imbalance of energy distribution. The difference in supply and demand between adjacent regions reflects the natural trend of energy flow. For example, if region A has a surplus of 1000MW while adjacent region B has a deficit of 800MW, the supply and demand gradient is 1800MW, indicating a strong demand for energy flow. The path load balancing reference value is derived from historical operating data statistics, reflecting the typical utilization level of each path. The genetic algorithm demonstrates unique advantages in solving multi-objective optimization problems. Each individual in the population represents a possible dispatch scheme, including the selected transmission path and allocated flow for each time period. Fitness evaluation synthesizes three objective functions through a weighted summation, with weight coefficients adjustable according to actual needs; for example, the transmission cost function has a larger weight in cost-sensitive scenarios. Selection, crossover, and mutation operations simulate natural evolution. Selection retains highly fit solutions; crossover generates new solutions by combining the advantages of two parent solutions; and mutation introduces randomness to prevent the algorithm from getting trapped in local optima. A time window constraint check ensures the feasibility of the generated solutions; only solutions where all transmission tasks can be completed within the specified time are retained. The algorithm's convergence is determined by the criterion that the optimal solution remains unchanged across consecutive algebras. After several generations, if the fitness value of the optimal solution no longer significantly improves, it indicates that an approximate global optimum has been found. The output scheduling sequence details which transmission paths should be activated in each time period and how much traffic each path carries, forming directly executable scheduling instructions.

[0069] S106. Evaluate the performance of the optimized scheduling scheme in terms of response latency and resource utilization. If the timeliness index of the scheduling scheme is lower than the preset response time threshold, determine the cause of the problem. If it is a local factor, optimize the scheduling scheme locally within the node's permission scope. If it is a global factor, adjust the cloud-edge collaborative decision-making model to generate optimized energy scheduling instructions.

[0070] For the optimized energy dispatch scheme, the sending time of each instruction and the actual execution time of the edge node feedback are extracted from the dispatch execution monitoring module. The difference between the two is calculated as the response latency. The resource utilization rate is obtained by statistically analyzing the ratio of the actual traffic to the rated capacity of each transmission path. The response latency is compared with a preset response time threshold to obtain timeliness assessment data that includes latency exceedance indicators and resource utilization rate values. Based on the latency exceedance indicators in the timeliness assessment data, fault codes in the node device status register, network link connectivity test results, deviations between regional load monitoring values ​​and historical averages, and energy device operation flags are obtained. If the fault code is non-zero, connectivity is interrupted, load deviation exceeds a preset range, or the operation flag indicates shutdown, it is determined to be a local factor; otherwise, it is determined to be a global factor, and a problem diagnosis result with specific abnormal location and type is obtained. Based on the local factor determination and anomaly location information in the problem diagnosis results, the local decision-making function of the corresponding edge node is activated. Within the range of energy types and scheduling quantities allowed by the node's permission configuration, a greedy algorithm is used to redistribute the affected energy traffic to other transmission paths connected to the node with sufficient capacity. The supply and demand matching parameters of the relevant nodes are adjusted to obtain a local optimization scheme that includes new path selection and traffic allocation. For the local optimization scheme or the global factor determination, if it is a local optimization scheme, it is issued to the relevant edge nodes through the control channel for execution and the execution effect is monitored. If it is a global factor, the timeliness assessment data, problem diagnosis results, and current network status parameters are uploaded to the cloud. The latency constraint parameters and path selection priority in the cloud-edge collaborative decision-making model are modified, and the model calculation is rerun to obtain a new scheduling instruction set adapted to the current fault state.

[0071] Specifically, the scheduling execution monitoring module, as a key component of the energy management system, is responsible for tracking the lifecycle of each scheduling instruction in real time.

[0072] Specifically, when a scheduling command is issued from the cloud, the monitoring module records the timestamp of the command's transmission. Upon receiving the command, the edge node reports the actual start time of execution via a feedback mechanism. The response latency is the difference between these two timestamps, directly reflecting the system's real-time performance.

[0073] For example, if a scheduling instruction is sent at 10:00:00 and the edge node starts executing it at 10:00:05, the response delay is 5 seconds.

[0074] In one possible implementation, resource utilization calculation involves a precise assessment of the load status of the transmission channel. Actual flow is obtained in real-time through power sensors installed on the transmission line, while the rated capacity is the maximum safe transmission capacity determined during line design. When a 500kV transmission line has a rated capacity of 1500MW and an actual transmission power of 1200MW, the resource utilization rate is 80%. This metric helps determine whether there is resource waste or overload risk in the system. The setting of response time thresholds needs to comprehensively consider the needs of different business scenarios. For power dispatching of industrial production lines, a response time of 3 seconds may be required to avoid production interruptions; while for routine power adjustments in residential areas, a 10-second response time may be sufficient. The timeliness assessment data not only includes a binary judgment of whether or not the standard is exceeded, but also records the specific degree of exceedance, providing a quantitative basis for subsequent problem diagnosis.

[0075] It should be noted that the device status register is an important component of the edge node hardware; it records the device's operating status using a specific encoding method. Fault codes employ a standardized encoding system, such as...

[0076] "E001" may indicate over-temperature protection triggered, and "E002" indicates over-current protection activated. Network link connectivity testing is performed by periodically sending heartbeat packets; if three consecutive heartbeat packets are not responded to, the link is considered interrupted. Anomaly detection for regional load is based on statistical principles. The system maintains the historical load mean and standard deviation for each region at different times. When the real-time load value deviates from the historical mean by more than three times the standard deviation, the system considers it an anomaly of sudden load increase or decrease. The energy equipment operation flag is a simple binary identifier: "1" indicates normal operation, and "0" indicates shutdown.

[0077] Preferably, the logic for determining the cause of a problem reflects the idea of ​​hierarchical diagnosis. Local factors typically manifest as problems with a limited scope of influence that can be resolved through local adjustments.

[0078] For example, a single transformer failure only affects users within its power supply area and can be resolved by switching to a backup transformer. Global factors, however, involve systemic problems, such as large-scale power grid failures or supply-demand imbalances in multiple regions caused by extreme weather, requiring a global re-optimization of the scheduling scheme. The application of greedy algorithms in local optimization follows the principle of "local optimality leading to global betterness." When a transmission path becomes unusable due to a failure, the algorithm selects the path with the lowest transmission cost or highest transmission efficiency from the remaining available paths. This selection is an optimal decision based on local information at the current moment.

[0079] For example, if a plan to transmit 500MW of power via path A is disrupted by a fault and requires reallocation, the algorithm will evaluate the remaining capacity and transmission costs of paths B, C, and D, selecting the optimal combination of paths based on comprehensive metrics. Edge node permission configuration plays a constraining role in local decision-making. Each node has clearly defined scheduling permission boundaries, including the types of energy that can be scheduled, the maximum scheduling volume, and the allowed trading objects. This permission restriction ensures that local decisions do not negatively impact the overall system. The local optimization scheme details which energy flows need to be rerouted, what the new transmission paths are, and the traffic capacity of each path. Parameter adjustment in the cloud-edge collaborative decision-making model is a key means of addressing global problems. Modifying latency constraint parameters affects the model's emphasis on timeliness; when the system detects widespread response delays, it will appropriately relax latency constraints to obtain more feasible solutions. Adjusting path selection priorities is based on real-time fault information, lowering the priority of faulty paths to ensure that newly generated scheduling schemes avoid problem areas. The generation process of the new scheduling instruction set fully utilizes the computing resource advantages of the cloud. The cloud can integrate real-time data from the entire network and use more complex optimization algorithms to generate globally optimal solutions adapted to the current fault state. This dynamic adjustment mechanism ensures that the energy system can maintain stable and efficient operation in the face of various abnormal situations.

[0080] S107. The optimized energy dispatch instructions are sent to edge nodes and cloud control units to obtain energy flow status data and extract key energy dispatch indicators for each node. When the latency exceeds the preset threshold or the storage utilization rate is lower than the preset threshold, the model is updated with the dispatch effect evaluation indicators as the optimization target to obtain the optimized decision framework parameters.

[0081] Optimized energy dispatch instructions are simultaneously sent to each edge node and the cloud control center via the communication network. Upon receiving the instructions, the edge nodes execute the corresponding energy dispatch operations and report their execution status. The cloud records the instruction issuance time and the execution feedback time of each node. Simultaneously, real-time data on energy flow direction, flow rate, and energy exchange between nodes are collected from each node, resulting in a real-time monitoring dataset containing timestamps and energy flow parameters. Based on the timestamp information and flow parameters in the real-time monitoring dataset, the difference between the execution feedback time and the instruction issuance time for each node is calculated as the execution latency. The ratio of the current storage capacity to the rated capacity of the energy storage device is calculated to obtain the storage utilization rate. The consumption ratio is calculated based on renewable energy generation and actual usage, resulting in key performance indicator data including the latency value, utilization rate value, and consumption ratio value of each node. Based on the key performance indicator data, if the execution latency of any node exceeds a preset latency threshold or the storage utilization rate falls below a preset utilization rate threshold, the cloud-based model optimization process is triggered. The current dispatch instruction parameters, execution results, and performance indicators are combined into a state-action-reward triplet. A reinforcement learning algorithm is used to adjust the decision parameters based on the reward value, resulting in an updated set of model parameters. Based on the updated model parameter set, the corresponding parameter values ​​in the cloud-edge collaborative decision-making model are replaced, including decision thresholds, weight coefficients, and constraint boundary values. The decision-making effect after parameter updates is verified through historical data. After confirming the performance improvement, the new parameter configuration is saved to obtain optimized decision framework parameters that adapt to the current operating state.

[0082] Specifically, the optimized energy dispatch instructions are transmitted through the communication network of the distributed energy management system, which is typically built using industrial Ethernet or dedicated power line carrier communication technology.

[0083] Specifically, the scheduling instructions include detailed execution parameters, such as the output power settings of each node, energy storage charging and discharging strategies, and energy trading arrangements with other nodes. Once the cloud control center generates these instructions, it will simultaneously send them to all relevant edge nodes through an encrypted channel to ensure the synchronization and security of the instructions.

[0084] In one possible implementation, edge nodes immediately parse and execute commands upon receiving them. The execution process includes adjusting the output power of local power generation equipment, controlling the charging and discharging status of the energy storage system, and opening or closing energy transmission channels with other nodes. Each node is equipped with an execution feedback mechanism; upon completing the required operations, it generates feedback information including the completion time, actual execution parameters, and execution result status code, and reports this to the cloud control center. Real-time monitoring data acquisition relies on sensors deployed at various key locations within the energy system. These sensors include current transformers, voltage transformers, and power analyzers, capable of accurately measuring the direction and flow of energy.

[0085] For example, at the junction of two regions, a bidirectional power meter can record the direction and magnitude of energy flow in real time; a positive value indicates flow from region A to region B, and a negative value indicates the opposite. Timestamp information is provided by a unified system clock service, ensuring all data has accurate time stamps.

[0086] It's important to note that calculating execution latency requires a precise time synchronization mechanism. The system typically uses a network time protocol to ensure clock synchronization across all nodes, with errors controlled to the millisecond level. Execution latency equals the node's reported completion time minus the cloud-recorded instruction issuance time. This metric directly reflects the system's response speed and is particularly critical for scenarios requiring rapid adjustments, such as frequency regulation and emergency power support. Calculating energy storage utilization involves real-time monitoring of the energy storage device's status. Each energy storage unit has a rated capacity; for example, a battery energy storage system might have a rated capacity of 1000 kWh. The battery management system can obtain the current actual storage capacity. If 600 kWh of energy is currently stored, the utilization rate is 60%. This metric reflects the efficiency of energy storage resource utilization; excessively low utilization means that energy storage investment is not being fully utilized.

[0087] Preferably, the calculation of the renewable energy consumption ratio needs to distinguish between different types of renewable energy. Photovoltaic power generation and wind power generation often face consumption difficulties due to their intermittent nature. The system records the total renewable energy generation in each time period, as well as the energy actually consumed or stored by the load. If a photovoltaic power generation is 1000kW in a certain time period, of which 800kW is directly used, 150kW is stored in energy storage, and 50kW is restricted due to lack of consumption, then the consumption ratio is 95%. The threshold setting for triggering cloud model optimization needs to comprehensively consider system performance requirements and operational economics. The latency threshold is usually set according to different business scenarios; for example, general load regulation can accept a 5-second delay, while grid frequency regulation may require a response within 1 second. The energy storage utilization rate threshold considers the economics of energy storage; excessively low utilization rates, such as below 30%, mean that energy storage devices are idle most of the time, requiring optimized scheduling strategies. The state-action-reward triplet in reinforcement learning algorithms forms the basis of learning. The state includes the current supply and demand situation, network topology, and equipment status; the action is the specific scheduling instruction; and the reward is calculated based on the execution effect, such as lower latency, higher energy storage utilization, and a higher proportion of renewable energy consumption, resulting in a higher reward value. The algorithm learns the optimal scheduling strategy to be adopted under various states by continuously trying different actions. Updating the model parameters is a gradual process. Decision thresholds, such as the supply-demand deviation threshold that triggers emergency scheduling, are dynamically adjusted based on the actual performance of the system; weight coefficients determine the relative importance of different optimization objectives, such as increasing the weight of the consumption ratio during periods of renewable energy surplus; constraint boundary values, such as the maximum allowable transmission power, are appropriately adjusted based on factors such as equipment aging and ambient temperature. This adaptive parameter optimization mechanism enables the system to continuously improve its decision-making performance.

[0088] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A resource scheduling method for an energy digitalization platform based on cloud-edge collaborative computing, characterized in that, The method includes: Real-time supply and demand data, energy price signals, and network topology information are obtained from the distributed energy management system to generate a structured dynamic supply and demand scenario dataset; Based on the structured dynamic supply and demand scenario dataset, time series analysis is used to extract the energy price fluctuation cycle and amplitude, and network analysis methods are used to extract the network topology reconstruction frequency and node connectivity changes, generating a dynamic feature parameter set. Based on the dynamic feature parameter set, input the pre-trained cloud-edge collaborative decision-making model to generate the autonomous scheduling permission configuration for each edge node; Based on the autonomous scheduling permission configuration, the comparison results between the computing resource requirements of each edge node and the preset threshold are detected, and the adjusted edge node permission configuration is generated. Extract the cross-regional energy exchange volume and exchange time period from the adjusted edge node permission configuration, combine the supply and demand difference between regions and network status, optimize the cross-regional energy flow path, and generate a globally optimized cross-regional energy dispatch scheme. Based on the globally optimized cross-regional energy dispatch scheme, the response delay and resource utilization rate are calculated, the local dispatch scheme is adjusted based on local factors or the cloud-edge collaborative decision-making model is updated based on global factors, and the optimized dispatch instructions are generated. The optimized scheduling instructions are sent to the edge nodes and the cloud, and the parameters of the cloud-edge collaborative decision-making model are updated according to the energy flow status data to generate optimized decision framework parameters.

2. The resource scheduling method for an energy digital platform based on cloud-edge collaborative computing as described in claim 1, characterized in that, The process of acquiring real-time supply and demand data, energy price signals, and network topology information from a distributed energy management system to generate a structured dynamic supply and demand scenario dataset includes: Collect energy supply, consumption, price signals, and network topology connections and transmission capacity of each node to generate a preliminary dataset; The preliminary dataset is subjected to outlier correction and missing value completion, and standardized units and values ​​are normalized to generate standardized energy data. Based on the standardized energy data, the supply and demand balance value of each node is calculated, the correlation between the price signal and the supply and demand balance value is calculated using the Pearson correlation coefficient, and the transmission path and capacity constraints are generated by combining the connection relationship of the network topology to generate scenario data. The scenario data is indexed in multiple dimensions according to node number, timestamp, and supply and demand type. An adjacency table structure is generated based on the connection relationship between nodes. The data version is updated regularly to generate a structured dynamic supply and demand scenario dataset that supports real-time querying.

3. The resource scheduling method for an energy digital platform based on cloud-edge collaborative computing according to claim 1, characterized in that, Based on the structured dynamic supply and demand scenario dataset, time series analysis is used to extract the energy price fluctuation cycle and amplitude, and network analysis methods are used to extract the frequency of network topology reconstruction and changes in node connectivity, generating a dynamic feature parameter set, including: Energy price sequences are extracted from the structured dynamic supply and demand scenario dataset, segmented using a sliding time window, and the frequency components and amplitudes of each time window are calculated using Fourier transform to generate a price fluctuation frequency spectrum and amplitude distribution. Based on the price fluctuation frequency spectrum, the dominant frequency is extracted to calculate the fluctuation cycle, and the difference between the maximum and minimum price values ​​in each time window is statistically analyzed to generate the fluctuation amplitude. The network topology adjacency matrix is ​​extracted from the structured dynamic supply and demand scenario dataset. The matrix difference between adjacent times is calculated to generate the topology reconstruction frequency. The changes in the number of node connections are statistically analyzed to generate a connectivity change sequence. The average rate of change is calculated based on the connectivity change sequence, and the fluctuation period, fluctuation amplitude, topology reconstruction frequency and connectivity change rate are integrated to generate a dynamic feature parameter set.

4. The resource scheduling method for an energy digital platform based on cloud-edge collaborative computing as described in claim 1, characterized in that, The step of generating autonomous scheduling permission configurations for each edge node by inputting the pre-trained cloud-edge collaborative decision-making model based on the dynamic feature parameter set includes: The dynamic feature parameter set is input into a pre-trained cloud-edge collaborative decision-making model, and the scheduling permission score of each edge node is generated through decision tree ensemble operation. Based on the scheduling authority score and the total energy reserves of the nodes, the types of energy that can be scheduled and the upper limit of the scheduling amount are determined. Based on the scheduling permission score, extract the list of adjacent nodes and connection capacity of the network topology, and generate a set of tradable nodes; Based on the schedulable energy types, scheduling capacity limits, and tradable node sets, and matching energy type and transmission capacity constraints, an autonomous scheduling permission configuration is generated, which includes energy types, scheduling capacity limits, and a list of trading objects.

5. The resource scheduling method for an energy digital platform based on cloud-edge collaborative computing according to claim 1, characterized in that, The step of detecting the comparison results between the computing resource requirements of each edge node and a preset threshold based on the autonomous scheduling permission configuration, and generating an adjusted edge node permission configuration, includes: Extract the number of schedulable energy types, scheduling volume, and number of trading objects from the autonomous scheduling permission configuration, calculate the processor utilization and memory usage, and generate computing resource requirements. The computational resource requirement value is compared with a preset threshold to calculate the resource over-limit rate exceeding the threshold and generate a compression ratio parameter. Based on the compression ratio parameter, low-priority energy types are removed, the upper limit of scheduling volume is reduced, and a compressed permission configuration is generated. The compressed permission configuration and the permission configuration that does not exceed the threshold are integrated to generate the adjusted permission configuration for edge nodes.

6. The resource scheduling method for an energy digital platform based on cloud-edge collaborative computing according to claim 1, characterized in that, The process involves extracting cross-regional energy exchange volume and time periods from the adjusted edge node permission configuration, combining this with inter-regional supply-demand differences and network status, optimizing cross-regional energy flow paths, and generating a globally optimized cross-regional energy scheduling scheme, including: Extract the geographic location and jurisdiction of the nodes from the adjusted edge node permission configuration, classify the regions, calculate the total scheduling permissions within the region, and generate cross-regional energy exchange volume and exchange time period; Based on the cross-regional energy exchange volume, calculate the regional net supply and demand difference, and combine it with the remaining capacity of transmission lines to generate inter-regional energy flow constraint parameters. Based on the energy flow constraint parameters between the regions, a directed weighted graph is constructed, and the shortest transmission path is calculated using the Dijkstra algorithm. Paths with insufficient capacity are eliminated, and a list of transmission paths and flow allocation values ​​are generated. Based on the transmission path list, the unit traffic transmission cost is calculated, the simplex method is used to optimize the traffic allocation value, and a globally optimized cross-regional energy dispatch scheme is generated.

7. The method according to claim 1, characterized in that, The process involves extracting cross-regional energy exchange volume and time periods from the adjusted edge node permission configuration, combining this with inter-regional supply-demand differences and network status, optimizing cross-regional energy flow paths, and generating a globally optimized cross-regional energy scheduling scheme, including: Extract the transaction permissions of the regional boundary nodes from the adjusted edge node permission configuration, calculate the energy exchange volume and transmission time period, and generate a dynamic routing weight matrix; Based on the dynamic routing weight matrix, and combined with the capacity and voltage constraints of the transmission lines, candidate transmission paths are selected. Based on the candidate transmission paths, a multi-objective optimization model is constructed. The supply and demand gradient data and historical load balancing coefficients are input, and a genetic algorithm is used for iterative calculation to generate a globally optimized cross-regional energy dispatch scheme.

8. The resource scheduling method for an energy digital platform based on cloud-edge collaborative computing according to claim 1, characterized in that, The process of calculating response latency and resource utilization based on the globally optimized cross-regional energy dispatch scheme, adjusting the local dispatch scheme based on local factors or updating the cloud-edge collaborative decision-making model based on global factors, and generating optimized dispatch instructions includes: Extract the instruction sending time and execution time from the globally optimized cross-regional energy dispatch scheme, calculate the response delay time, and generate the resource utilization rate by calculating the ratio of transmission path traffic to rated capacity. Based on the response delay time, detect node fault codes, network connectivity, load deviation and equipment operating status, and generate local or global factor diagnosis results; Based on the local factor diagnosis results, a greedy algorithm is used to redistribute energy flow and generate a local optimization scheme. Based on the global factor diagnosis results, the latency constraints and path priorities of the cloud-edge collaborative decision-making model are modified to generate optimized scheduling instructions.

9. The resource scheduling method for an energy digitalization platform based on cloud-edge collaborative computing according to claim 1, characterized in that, The optimized scheduling instructions are then sent to the edge nodes and the cloud. The parameters of the cloud-edge collaborative decision-making model are updated based on energy flow status data to generate optimized decision framework parameters, including: The optimized scheduling instructions are sent to edge nodes and the cloud to collect energy flow direction and flow data, and generate a real-time monitoring dataset. Based on the real-time monitoring dataset, the execution latency, storage utilization rate, and renewable energy consumption ratio are calculated to generate key performance indicator data. Based on the aforementioned key performance indicator data, the model parameters are updated using a reinforcement learning algorithm; Based on the updated model parameters, the decision-making effect is verified, and optimized decision framework parameters are generated.

Images