Distributed energy real-time control method and system based on edge calculation

Abstract

The invention provides a distributed energy real-time control method and system based on edge calculation, and relates to the technical field of energy management, and the method comprises the steps: obtaining the operation data of a distributed energy node power grid, building an electrical coupling relation, carrying out the multi-scale wavelet decomposition and time sequence processing, calculating the supply and demand deviation, and recognizing an imbalance event. And point-to-point communication is established between the edge computing nodes to exchange data, and iterative solution is carried out by taking minimization of energy transmission loss as a target until the boundary power flow deviation converges. The real-time cooperative control of the distributed energy system is realized, the communication overhead is reduced, and the energy scheduling efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, and in particular to a method and system for real-time control of distributed energy based on edge computing. Background Technology

[0002] With the widespread application of renewable energy, distributed energy systems are playing an increasingly important role in the power system. Distributed energy systems include distributed photovoltaic power generation, wind power generation, and energy storage systems. These systems are typically distributed across different locations within the power grid, requiring real-time monitoring and control to ensure stable grid operation. Traditional centralized control methods rely primarily on a central control center to collect, process, and analyze data from the entire network before sending control commands to various execution units. As the scale of distributed energy integration continues to expand and the power grid structure becomes increasingly complex, higher demands are placed on the real-time performance, reliability, and adaptability of the control system.

[0003] Existing technologies still have problems such as the need to transmit a large amount of data to the central control center for processing, causing communication delays and bandwidth pressure, making it difficult to meet the control requirements of rapid response in distributed energy systems, lacking an effective inter-node coordination mechanism, making it difficult to achieve coordinated control among multiple nodes under supply and demand imbalance, easily causing voltage fluctuations in local areas, and insufficient consideration of the output characteristics of distributed energy, making it difficult to accurately predict short-term power fluctuations and make corresponding control adjustments. Summary of the Invention

[0004] This invention provides a method and system for real-time control of distributed energy based on edge computing, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a real-time control method for distributed energy resources based on edge computing, comprising:

[0006] The grid operation data of distributed energy nodes are obtained, and the electrical coupling relationship between nodes is established by combining the geographical location information of the distributed energy nodes to obtain an energy operation dataset. Multi-scale wavelet decomposition is performed on the energy operation dataset to obtain high-frequency components and low-frequency components. Autoregressive moving average processing is performed on the high-frequency components to obtain a short-term power output sequence, and trend fitting is performed on the low-frequency components to obtain the operating boundary range.

[0007] The supply-demand deviation is calculated based on the short-term output sequence and the load demand data in the energy operation dataset. Imbalance events where the supply-demand deviation exceeds a preset threshold are identified, and the set of affected nodes is extracted based on the electrical coupling relationship.

[0008] The affected node set is mapped to the corresponding edge computing node and a point-to-point communication connection is established when the imbalance event is triggered. The short-term output sequence and the operating boundary range are exchanged based on the point-to-point communication connection. Distributed collaborative data is constructed by combining the local load status information of the edge computing node.

[0009] Based on the distributed collaborative data, the active power output adjustment and reactive power compensation are solved with the goal of minimizing energy transmission loss to obtain an initial control scheme and execute it. The boundary power flow deviation caused by the initial control scheme is calculated and sent to the adjacent edge computing nodes. The solution is repeated until the boundary power flow deviation converges to obtain a global control scheme.

[0010] In one alternative implementation,

[0011] The energy operation dataset obtained by acquiring grid operation data from distributed energy nodes and establishing electrical coupling relationships between nodes based on the geographical location information of the distributed energy nodes includes:

[0012] Real-time power generation and voltage amplitude of distributed energy nodes are collected to obtain operating status data; real-time load power demand of the area where the distributed energy nodes are located is collected to obtain load demand data; the connection relationship between distributed energy nodes and transmission capacity limit are read to obtain the grid topology connection relationship; and the grid operation data is obtained by combining the operating status data and the load demand data.

[0013] Based on the real-time power generation in the operating status data and the load demand data, the power surplus or power deficit of each distributed energy node is calculated, and the distributed energy nodes are classified to obtain a set of power surplus nodes and a set of power deficit nodes.

[0014] Calculate the geographical distance between each node in the power surplus node set and each node in the power deficit node set, and mark the node pairs whose geographical distance is less than a preset distance threshold and have a connection relationship in the power grid topology connection relationship as potential coupled node pairs;

[0015] Based on the line parameters of the connection path in the power grid topology connection relationship and the voltage amplitude in the operating status data, the effective transmission capacity corresponding to the potential coupling node pair is calculated. The potential coupling node pair with the effective transmission capacity greater than the preset capacity threshold is taken as the coupling node pair and the electrical coupling relationship is constructed using the effective transmission capacity as a quantitative index.

[0016] The energy operation dataset is obtained by integrating the power grid operation data and the electrical coupling relationships between the nodes.

[0017] In one alternative implementation,

[0018] Multi-scale wavelet decomposition is performed on the energy operation dataset to obtain high-frequency and low-frequency components. Autoregressive moving average processing is applied to the high-frequency components to obtain a short-term output sequence. Trend fitting is performed on the low-frequency components to obtain the operating boundary range, which includes:

[0019] The operating status data and load demand data of each distributed energy node within a preset time window are extracted from the energy operation dataset and time series data is constructed.

[0020] The time series data is subjected to multi-level wavelet decomposition. The high-frequency components obtained from each level of decomposition are merged to obtain the high-frequency components. The low-frequency components obtained from the last level of decomposition are taken as the low-frequency components. The autocorrelation coefficient sequence and partial autocorrelation coefficient sequence of the high-frequency components are calculated, and the order of the autoregressive term and the order of the moving average term are determined.

[0021] Extract the historical values ​​and historical prediction errors corresponding to the high-frequency components, linearly weight the historical values ​​according to the order of the autoregressive term to obtain the autoregressive term, linearly weight the historical prediction errors according to the order of the moving average term to obtain the moving average term, and solve the short-term output sequence based on the autoregressive term and the moving average term.

[0022] Historical values ​​corresponding to the low-frequency components are extracted and polynomial fitting is performed to obtain a low-frequency trend function. The maximum and minimum values ​​of the low-frequency components are calculated. The difference between the maximum value and the low-frequency trend function is taken as the upper boundary deviation, and the difference between the low-frequency trend function and the minimum value is taken as the lower boundary deviation. Based on the upper boundary deviation and the lower boundary deviation, the upper and lower limits of the output of each distributed energy node are determined to obtain the operating boundary range.

[0023] In one alternative implementation,

[0024] The supply-demand deviation is calculated based on the short-term output sequence and the load demand data in the energy operation dataset. Imbalance events where the supply-demand deviation exceeds a preset threshold are identified, and the set of affected nodes is extracted based on the electrical coupling relationship, including:

[0025] Extract the load demand data from the energy operation dataset, and calculate the difference between the output value of each node in the short-term output sequence and the load demand data to obtain the original deviation sequence;

[0026] The original deviation sequence is constructed and singular value decomposition is performed. The maximum singular value of the first preset number and the corresponding left singular vector and right singular vector are extracted and used as the time mode and spatial mode. The dominant deviation component is obtained by performing an outer product on the time mode and spatial mode.

[0027] The deviation rate of change is obtained by time difference of the dominant deviation component, and the deviation gradient is calculated based on the electrical coupling relationship. The supply and demand deviation amount is obtained by solving based on the deviation rate of change and the deviation gradient. The node whose supply and demand deviation amount exceeds the preset deviation threshold is identified as the imbalance source node and the imbalance event is determined.

[0028] Based on the electrical coupling relationship, a Laplace matrix is ​​constructed and the eigenvectors corresponding to the non-zero minimum eigenvalues ​​are extracted as diffusion basis functions. The supply and demand deviation is projected onto the diffusion basis functions to obtain the diffusion coefficient. Based on the diffusion coefficient, the propagation deviation corresponding to the imbalance source node is calculated and accumulated over time to obtain the cumulative propagation deviation.

[0029] Identify neighboring nodes whose cumulative propagation deviation exceeds the propagation threshold and add them to a pre-initialized empty set. Repeat the calculation of the diffusion coefficient and cumulative propagation deviation until no new nodes are added to obtain the set of affected nodes.

[0030] In one alternative implementation,

[0031] The affected node set is mapped to the corresponding edge computing node, and a point-to-point communication connection is established when the imbalance event is triggered. Based on the point-to-point communication connection, the short-term output sequence and the operating boundary range are exchanged. Distributed collaborative data is constructed by combining the local load status information of the edge computing node, including:

[0032] Based on the geographical location information of the nodes in the affected node set, the nodes in the affected node set are mapped to the corresponding edge computing nodes. Based on the electrical coupling relationship, the affected nodes mapped to the same edge computing node are clustered and grouped to obtain the node grouping results.

[0033] When the imbalance event is triggered, a hierarchical coverage network topology is constructed between each edge computing node based on the node grouping results, and link quality is detected to obtain link quality parameters. Based on the link quality parameters, a link cost function is constructed and multi-objective optimization is performed to obtain the optimal path set and establish a point-to-point communication connection.

[0034] The short-term output sequence and the operating boundary range corresponding to each edge computing node are broadcast and transmitted through the point-to-point communication connection, and neighborhood collaborative state data are obtained by combining integrity verification and timing calibration.

[0035] Local load status information of each node in the corresponding node group is collected locally from each edge computing node. The local load status information and the neighborhood collaborative status data are timestamped and a status feature matrix is ​​constructed. The status feature matrix is ​​decomposed into a core tensor and a factor matrix, and tensor reconstruction is performed to obtain a dimension-reduced feature vector. The identification information corresponding to each edge computing node is added to the dimension-reduced feature vector to obtain the distributed collaborative data.

[0036] In one alternative implementation,

[0037] Based on the node grouping results, a hierarchical coverage network topology is constructed among the edge computing nodes, and link quality parameters are obtained through link quality detection. Based on these link quality parameters, a link cost function is constructed and multi-objective optimization is performed to obtain the optimal path set and establish point-to-point communication connections, including:

[0038] Based on the node grouping results, the number of nodes corresponding to each edge computing node and the connection strength of the electrical coupling relationship are counted, and the edge computing nodes are hierarchically divided to obtain core edge computing nodes and peripheral edge computing nodes. Full connectivity is established between the core edge computing nodes, and star connectivity is established between the peripheral edge computing nodes and the core edge computing nodes to obtain a hierarchical coverage network topology.

[0039] For each communication path in the hierarchical coverage network topology, the end-to-end transmission delay is determined by sending probe data packets and receiving response data packets. The number of lost probe data packets is counted and the packet loss rate is calculated. The available bandwidth of the communication path is measured and the link quality parameters are determined by combining the end-to-end transmission delay and the packet loss rate. Weighting coefficients are set for each type of data in the link quality parameters and the link cost function is determined by weighting.

[0040] With minimizing the link cost function as the optimization objective and maintaining connectivity and load balancing of the hierarchical coverage network topology as constraints, a multi-objective optimization solution is obtained to obtain a Pareto optimal solution set. Solutions that satisfy the constraints are selected from the Pareto optimal solution set and summarized to obtain an optimal path set. Based on the optimal path set, communication resources are allocated to edge computing nodes and point-to-point communication connections are established.

[0041] In one alternative implementation,

[0042] Based on the distributed collaborative data, an initial control scheme is obtained by solving for the active power output adjustment and reactive power compensation with the goal of minimizing energy transmission loss. This scheme is then executed. The boundary power flow deviation caused by the initial control scheme is calculated and sent to adjacent edge computing nodes. This process is repeated until the boundary power flow deviation converges, resulting in a global control scheme, including:

[0043] Dimensionally reduced feature vectors and local load status information are extracted from the distributed collaborative data, and regional energy network topology of each edge computing node is constructed. Based on the local load status information, the current active power output and current reactive power output of each node are determined. An optimization function is established with the goal of minimizing energy transmission loss, and the active power output adjustment and reactive power compensation of each node are solved to obtain the initial control scheme.

[0044] The initial control scheme is sent to each node and the control action is executed. The real-time voltage and real-time current of each node are collected after execution, and the power flow value corresponding to the boundary node of the regional energy network topology is calculated. The power flow value is compared with the preset power flow value corresponding to the boundary node to obtain the boundary power flow deviation.

[0045] The boundary power flow deviation is sent to the adjacent edge computing node through the point-to-point communication connection. It is determined whether the boundary power flow deviation is less than the preset convergence threshold. If it is not less than the threshold, the boundary power flow deviation fed back by the adjacent edge computing node is received and used as the boundary constraint condition to update the optimization function. The optimization function is re-solved based on the updated optimization function until the boundary power flow deviation is less than the convergence threshold to obtain the global control scheme. If it is less than the threshold, the current control scheme is determined as the global control scheme and output for execution.

[0046] A second aspect of this invention provides a real-time control system for distributed energy based on edge computing, comprising:

[0047] The preprocessing unit is used to acquire grid operation data of distributed energy nodes, establish electrical coupling relationships between nodes by combining the geographical location information of the distributed energy nodes to obtain an energy operation dataset, perform multi-scale wavelet decomposition on the energy operation dataset to obtain high-frequency components and low-frequency components, perform autoregressive moving average processing on the high-frequency components to obtain a short-term power output sequence, and perform trend fitting on the low-frequency components to obtain the operating boundary range.

[0048] An imbalance identification unit is used to calculate the supply-demand deviation based on the short-term output sequence and the load demand data in the energy operation dataset, identify imbalance events where the supply-demand deviation exceeds a preset threshold, and extract the set of affected nodes based on the electrical coupling relationship.

[0049] The collaborative communication unit is used to map the affected node set to the corresponding edge computing node and establish a point-to-point communication connection when the imbalance event is triggered. Based on the point-to-point communication connection, the short-term output sequence and the operating boundary range are exchanged, and distributed collaborative data is constructed by combining the local load status information of the edge computing node.

[0050] The optimization control unit is used to solve for the active power output adjustment and reactive power compensation based on the distributed collaborative data, with the goal of minimizing energy transmission loss, to obtain an initial control scheme and execute it. It calculates the boundary power flow deviation caused by the initial control scheme and sends it to the adjacent edge computing nodes. The solution is repeated until the boundary power flow deviation converges to obtain a global control scheme.

[0051] A third aspect of the present invention provides an electronic device, comprising:

[0052] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0053] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0054] In this invention, energy operation data is separated into high-frequency and low-frequency components using multi-scale wavelet decomposition technology, and processed using autoregressive moving average and trend fitting methods respectively. This enables more accurate prediction of short-term output sequences and operational boundary ranges, improving the prediction accuracy of energy dispatch. Based on the calculation of supply and demand deviations and the analysis of electrical coupling relationships, imbalance events can be quickly identified and the affected node set can be accurately located, improving the response speed to abnormal situations. By mapping affected nodes to edge computing nodes and establishing point-to-point communication connections, efficient construction of distributed collaborative data is achieved, avoiding the communication bottleneck of centralized control and reducing response delay. With minimizing energy transmission losses as the optimization objective, the active power output adjustment and reactive power compensation are solved, effectively reducing losses in the energy transmission process and improving overall energy efficiency. Through iterative calculation of boundary power flow deviations and collaborative optimization between adjacent nodes, coordination and consistency among distributed nodes are achieved, ensuring that a globally optimal control scheme is achieved while maintaining stability. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the real-time control method for distributed energy based on edge computing according to an embodiment of the present invention.

[0056] Figure 2 This is a flowchart illustrating the hierarchical coverage network topology construction process of the real-time control method for distributed energy based on edge computing, as described in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0058] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0059] Figure 1 This is a flowchart illustrating the real-time control method for distributed energy based on edge computing according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0060] The grid operation data of distributed energy nodes are obtained, and the electrical coupling relationship between nodes is established by combining the geographical location information of the distributed energy nodes to obtain an energy operation dataset. Multi-scale wavelet decomposition is performed on the energy operation dataset to obtain high-frequency components and low-frequency components. Autoregressive moving average processing is performed on the high-frequency components to obtain a short-term power output sequence, and trend fitting is performed on the low-frequency components to obtain the operating boundary range.

[0061] The supply-demand deviation is calculated based on the short-term output sequence and the load demand data in the energy operation dataset. Imbalance events where the supply-demand deviation exceeds a preset threshold are identified, and the set of affected nodes is extracted based on the electrical coupling relationship.

[0062] The affected node set is mapped to the corresponding edge computing node and a point-to-point communication connection is established when the imbalance event is triggered. The short-term output sequence and the operating boundary range are exchanged based on the point-to-point communication connection. Distributed collaborative data is constructed by combining the local load status information of the edge computing node.

[0063] Based on the distributed collaborative data, the active power output adjustment and reactive power compensation are solved with the goal of minimizing energy transmission loss to obtain an initial control scheme and execute it. The boundary power flow deviation caused by the initial control scheme is calculated and sent to the adjacent edge computing nodes. The solution is repeated until the boundary power flow deviation converges to obtain a global control scheme.

[0064] In one alternative implementation,

[0065] The energy operation dataset obtained by acquiring grid operation data from distributed energy nodes and establishing electrical coupling relationships between nodes based on the geographical location information of the distributed energy nodes includes:

[0066] Real-time power generation and voltage amplitude of distributed energy nodes are collected to obtain operating status data; real-time load power demand of the area where the distributed energy nodes are located is collected to obtain load demand data; the connection relationship between distributed energy nodes and transmission capacity limit are read to obtain the grid topology connection relationship; and the grid operation data is obtained by combining the operating status data and the load demand data.

[0067] Based on the real-time power generation in the operating status data and the load demand data, the power surplus or power deficit of each distributed energy node is calculated, and the distributed energy nodes are classified to obtain a set of power surplus nodes and a set of power deficit nodes.

[0068] Calculate the geographical distance between each node in the power surplus node set and each node in the power deficit node set, and mark the node pairs whose geographical distance is less than a preset distance threshold and have a connection relationship in the power grid topology connection relationship as potential coupled node pairs;

[0069] Based on the line parameters of the connection path in the power grid topology connection relationship and the voltage amplitude in the operating status data, the effective transmission capacity corresponding to the potential coupling node pair is calculated. The potential coupling node pair with the effective transmission capacity greater than the preset capacity threshold is taken as the coupling node pair and the electrical coupling relationship is constructed using the effective transmission capacity as a quantitative index.

[0070] The energy operation dataset is obtained by integrating the power grid operation data and the electrical coupling relationships between the nodes.

[0071] Edge computing terminal devices are installed at each distributed energy node to collect real-time power generation and voltage amplitude data. Taking a photovoltaic (PV) power generation node as an example, the edge device collects power generation data every 10 seconds via a communication interface connected to the inverter, with the collected values ​​in kilowatts (kW). It also collects voltage amplitude data via a voltage sensor, with the unit being volts (V). For example, the data collected by a PV node at noon on a sunny day might be: power generation 85.6 kW and voltage amplitude 220.5 volts. For wind power generation nodes, the edge device collects power data via the data interface of the wind turbine control system. For instance, a wind power generation node might have a power generation of 65.3 kW and a voltage amplitude of 219.8 volts at a specific moment.

[0072] Load demand data is collected by deploying smart meters at key load points in the regional power distribution network. Each smart meter collects load power demand data every minute and transmits it to the regional control center via a wireless communication network. For example, at a data collection point in a residential area, the load demand might be 150 kW at 7:00 AM, but could drop to 105 kW at 3:00 PM. Load demand fluctuations in industrial areas may be more pronounced; for instance, a factory area might have a peak production demand of 480 kW, which could decrease to 120 kW during off-peak hours.

[0073] The power grid topology connection data is obtained by reading the static topology database of the distribution network management system. This database stores the physical connections between nodes, including the type, length, impedance parameters, and rated transmission capacity of the connecting lines. For example, the connection between node 1 and node 2 is an overhead line, 2.5 km long, with an impedance of 0.6 ohms and a rated transmission capacity of 200 kW. The connection between node 2 and node 3 is an underground cable, 1.8 km long, with an impedance of 0.4 ohms and a rated transmission capacity of 300 kW.

[0074] Based on real-time power generation and load demand data from the operational status data, the power surplus or deficit of each node is calculated. The power surplus equals the node's power generation minus its local load demand. A positive result indicates a node with a power surplus; a negative result indicates a node with a power deficit. For example, if node A has a power generation of 120 kW and a local load of 80 kW, its power surplus is 40 kW, and node A is classified as a power surplus node. If node B has a power generation of 60 kW and a local load of 90 kW, its power deficit is 30 kW, and node B is classified as a power deficit node. All power surplus nodes are grouped into a power surplus node set, and all power deficit nodes are grouped into a power deficit node set.

[0075] The geographical distance between nodes is calculated based on their geographic coordinates, and the straight-line distance between two nodes is calculated using the spherical distance formula. If the geographical distance is less than a preset distance threshold, such as 5 kilometers, and there is a direct connection or a connection path through no more than two intermediate nodes in the power grid topology, then this pair of nodes is marked as a potential coupled node pair. For example, if the geographical distance between node A and node B is 3.2 kilometers, and they are connected through node C, then node A and node B are marked as a potential coupled node pair.

[0076] Effective transmission capacity calculation is based on line parameters and node voltage amplitudes in the power grid topology, considering voltage drops caused by line impedance and line thermal capacity limitations, to comprehensively calculate the effective transmission capacity between two nodes. For example, if the line impedance between a power surplus node A and a power deficit node B is 0.5 ohms, the voltage at node A is 220 volts, and the rated line capacity is 150 kW, considering voltage stability requirements, the effective transmission capacity between nodes A and B is calculated to be 120 kW. If the effective transmission capacity is greater than a preset capacity threshold, such as 20 kW, then this pair of nodes is determined as a coupled node pair, and the effective transmission capacity value is used as a quantitative indicator to construct the electrical coupling relationship.

[0077] The electrical coupling relationships are constructed using a directed weighted graph structure, where nodes represent distributed energy nodes, edges represent electrical connections between nodes, and the weight of an edge represents its effective transmission capacity. For example, a directed edge from node A to node B with a weight of 120 kilowatts indicates that node A can transmit a maximum of 120 kilowatts of power to node B. By integrating the information of all coupled node pairs, a complete electrical coupling relationship network is formed.

[0078] An energy operation dataset is constructed by integrating grid operation data with the electrical coupling relationships between nodes. This dataset includes each node's identifier, geographical location, real-time power generation, voltage amplitude, local load demand, power surplus or deficit, and its category (surplus or deficit); the connection status between node pairs, geographical distance, effective transmission capacity, and electrical coupling strength. Indexed by timestamps, the energy operation dataset is updated every 5 minutes and stored in a distributed database for use by the energy dispatch and control system.

[0079] In this embodiment, by collecting real-time data on the power generation, voltage amplitude, and regional load demand of distributed energy nodes and integrating it with the grid topology, the system can comprehensively reflect the supply and demand distribution and constraints of the grid under actual operating conditions, improving the completeness and authenticity of the grid's operating status representation. By calculating the power surplus and power deficit of nodes and classifying them, the system can quickly distinguish between energy supply capacity and energy demand, significantly reducing computational complexity and improving matching efficiency. By screening potential coupled node pairs based on geographical distance constraints and topological connectivity, the system effectively avoids the shortcomings of relying solely on electrical connectivity while ignoring spatial feasibility, reducing node combinations that lack practical scheduling value, and improving the rationality of coupling relationship identification and engineering feasibility.

[0080] In one alternative implementation,

[0081] Multi-scale wavelet decomposition is performed on the energy operation dataset to obtain high-frequency and low-frequency components. Autoregressive moving average processing is applied to the high-frequency components to obtain a short-term output sequence. Trend fitting is performed on the low-frequency components to obtain the operating boundary range, which includes:

[0082] The operating status data and load demand data of each distributed energy node within a preset time window are extracted from the energy operation dataset and time series data is constructed.

[0083] The time series data is subjected to multi-level wavelet decomposition. The high-frequency components obtained from each level of decomposition are merged to obtain the high-frequency components. The low-frequency components obtained from the last level of decomposition are taken as the low-frequency components. The autocorrelation coefficient sequence and partial autocorrelation coefficient sequence of the high-frequency components are calculated, and the order of the autoregressive term and the order of the moving average term are determined.

[0084] Extract the historical values ​​and historical prediction errors corresponding to the high-frequency components, linearly weight the historical values ​​according to the order of the autoregressive term to obtain the autoregressive term, linearly weight the historical prediction errors according to the order of the moving average term to obtain the moving average term, and solve the short-term output sequence based on the autoregressive term and the moving average term.

[0085] Historical values ​​corresponding to the low-frequency components are extracted and polynomial fitting is performed to obtain a low-frequency trend function. The maximum and minimum values ​​of the low-frequency components are calculated. The difference between the maximum value and the low-frequency trend function is taken as the upper boundary deviation, and the difference between the low-frequency trend function and the minimum value is taken as the lower boundary deviation. Based on the upper boundary deviation and the lower boundary deviation, the upper and lower limits of the output of each distributed energy node are determined to obtain the operating boundary range.

[0086] The operating status data and load demand data of each distributed energy node within a preset time window are extracted from the energy operation dataset to construct time series data. The preset time window is typically set to 7 days, with a sampling interval of 15 minutes, resulting in time series data with 672 sampling points. For example, for a photovoltaic power generation node, the power generation data every 15 minutes over the past 7 days is extracted to form a power generation time series; the load demand data of the corresponding area of ​​the node is extracted to form a load demand time series. The two time series data are aligned according to timestamps to obtain the final time series data.

[0087] Multi-level wavelet decomposition was performed on the time series data, using the de Beci 5 wavelet basis function. Five levels of wavelet decomposition were applied to the 672-point time series data. Each level of decomposition divided the signal into high-frequency and low-frequency components. The first level of decomposition yielded 336 high-frequency and 336 low-frequency components. The second level further decomposed the low-frequency components from the first level, yielding 168 high-frequency and 168 low-frequency components. This process continued until the fifth level, which yielded 21 high-frequency and 21 low-frequency components. The high-frequency components from the first to the fifth levels were merged to form the high-frequency component set; the low-frequency components from the fifth level were used as the final low-frequency components.

[0088] The autocorrelation coefficient and partial autocorrelation coefficient sequences of high-frequency components are calculated to determine the order of the autoregressive term and the moving average term. Taking the high-frequency components of a photovoltaic node as an example, the autocorrelation coefficients with delays of 1 to 30 time units are calculated, resulting in the following autocorrelation coefficient sequences: 0.82, 0.67, 0.52, 0.40, 0.30, 0.22, 0.16, 0.12, 0.09, 0.07, 0.05, 0.04, 0.03, 0.02, 0.02, 0.01, 0.01, 0.00, etc. The partial autocorrelation coefficient sequences are: 0.82, 0.05, 0.03, 0.02, 0.01, 0.01, 0.00, etc. Based on the truncation characteristics of the autocorrelation coefficient and partial autocorrelation coefficient sequences, the order of the autoregressive term is determined to be 2, and the order of the moving average term is determined to be 1. A second-order autoregressive first-order moving average model is adopted.

[0089] Historical values ​​and prediction errors corresponding to high-frequency components are extracted. The historical values ​​are then linearly weighted according to the order of the autoregressive term to obtain the autoregressive term. For a given moment, the high-frequency component values ​​from the previous two moments are extracted to be 5.2 and 3.8, respectively. Setting the autoregressive coefficients to 0.8 and 0.15, the autoregressive term is calculated as: 0.8 × 5.2 + 0.15 × 3.8 = 4.73. The moving average term is then linearly weighted according to the order of the moving average term to obtain the moving average term. The prediction error from the previous moment is extracted to be -0.5, and the moving average coefficient is set to 0.3, resulting in a moving average term of -0.15. The short-term output sequence is obtained based on the autoregressive and moving average terms. Adding the autoregressive and moving average terms yields the predicted high-frequency component value of 4.58 for the current moment. The high-frequency component values ​​for future time points are predicted sequentially using the same method to obtain the high-frequency component prediction sequence.

[0090] Historical values ​​corresponding to low-frequency components were extracted and polynomial fitting was performed to obtain the low-frequency trend function. Taking the low-frequency component at point 21 as an example, its values ​​are: 45.3, 46.2, 48.7, 52.1, 56.8, 61.2, 65.7, 68.3, 69.5, 68.2, 65.3, 60.5, 55.2, 49.8, 44.3, 40.5, 38.2, 37.3, 37.8, 39.5, 42.3. A third-order polynomial was used for fitting, resulting in the trend function expression: the coefficient of the cubic term is -0.002, the coefficient of the quadratic term is 0.05, the coefficient of the linear term is -0.3, and the constant term is 46.5. The maximum and minimum values ​​of the low-frequency components were calculated. For the aforementioned low-frequency component data, the maximum value is 69.5, and the minimum value is 37.3. The difference between the maximum value and the low-frequency trend function is used as the upper boundary deviation, and the difference between the low-frequency trend function and the minimum value is used as the lower boundary deviation. The trend function value at the time point corresponding to the maximum value is calculated to be 67.2, and the upper boundary deviation is 69.5 minus 67.2, resulting in 2.3. The trend function value at the time point corresponding to the minimum value is calculated to be 39.1, and the lower boundary deviation is 39.1 minus 37.3, resulting in 1.8. Based on the upper and lower boundary deviations, the upper and lower limits of output for each distributed energy node are determined, thus obtaining the operating boundary range. For example, for a future time point, if the calculated trend function value is 50.0, the upper limit of output is 50.0 plus 2.3, i.e., 52.3; the lower limit of output is 50.0 minus 1.8, i.e., 48.2.

[0091] In this embodiment, by constructing time-series data containing operating status and load demand within a preset time window, the output changes of distributed energy nodes can be continuously modeled on a unified time scale, improving the overall grasp of the output evolution law. By performing multi-level wavelet decomposition on the time-series data and modeling high-frequency and low-frequency components separately, the random fluctuation components and long-term variation trends are effectively separated, improving the stability and reliability of the prediction results. By introducing autocorrelation and partial autocorrelation analysis on the high-frequency components to adaptively determine the order, and solving the short-term output sequence based on autoregressive and moving average terms, the short-term output prediction can closely fit the historical fluctuation characteristics, significantly improving the response accuracy to rapid power fluctuations and uncertain disturbances. By performing polynomial fitting on the low-frequency components to extract the long-term variation trend, and combining historical extreme values ​​to construct dynamic output upper and lower limits, the node operating boundary can be adaptively adjusted with changes in the operating environment and trends, improving the rationality and adaptability of the operating constraint description.

[0092] In one alternative implementation,

[0093] The supply-demand deviation is calculated based on the short-term output sequence and the load demand data in the energy operation dataset. Imbalance events where the supply-demand deviation exceeds a preset threshold are identified, and the set of affected nodes is extracted based on the electrical coupling relationship, including:

[0094] Extract the load demand data from the energy operation dataset, and calculate the difference between the output value of each node in the short-term output sequence and the load demand data to obtain the original deviation sequence;

[0095] The original deviation sequence is constructed and singular value decomposition is performed. The maximum singular value of the first preset number and the corresponding left singular vector and right singular vector are extracted and used as the time mode and spatial mode. The dominant deviation component is obtained by performing an outer product on the time mode and spatial mode.

[0096] The deviation rate of change is obtained by time difference of the dominant deviation component, and the deviation gradient is calculated based on the electrical coupling relationship. The supply and demand deviation amount is obtained by solving based on the deviation rate of change and the deviation gradient. The node whose supply and demand deviation amount exceeds the preset deviation threshold is identified as the imbalance source node and the imbalance event is determined.

[0097] Based on the electrical coupling relationship, a Laplace matrix is ​​constructed and the eigenvectors corresponding to the non-zero minimum eigenvalues ​​are extracted as diffusion basis functions. The supply and demand deviation is projected onto the diffusion basis functions to obtain the diffusion coefficient. Based on the diffusion coefficient, the propagation deviation corresponding to the imbalance source node is calculated and accumulated over time to obtain the cumulative propagation deviation.

[0098] Identify neighboring nodes whose cumulative propagation deviation exceeds the propagation threshold and add them to a pre-initialized empty set. Repeat the calculation of the diffusion coefficient and cumulative propagation deviation until no new nodes are added to obtain the set of affected nodes.

[0099] Load demand data is extracted from the energy operation dataset, and the difference between the output value of each node and the load demand data in the short-term output sequence is calculated. Taking a distributed energy network as an example, load demand data for 10 nodes every 15 minutes over the past 7 days is extracted, forming a 10-row, 672-column load demand matrix. Simultaneously, the corresponding short-term output forecast values ​​for each node are extracted, forming a 10-row, 672-column node output matrix. For each time point, the difference between the output value and the load demand value of each node is calculated to obtain the original deviation sequence. For example, if the output forecast value of a photovoltaic node at a certain time is 65.3 kW, and the corresponding load demand is 72.1 kW, then the original deviation value at that time is -6.8 kW. The original deviation values ​​of each node at all time points are organized into a 10-row, 672-column original deviation matrix.

[0100] The original deviation matrix was subjected to singular value decomposition (SVD), and the top three largest singular values ​​and their corresponding left and right singular vectors were extracted. SVD was performed on the 10-row, 672-column original deviation matrix, yielding 10 singular values: 56.8, 32.4, 21.7, 15.3, 10.2, 8.6, 6.5, 5.2, 3.8, and 2.1. The top three largest singular values, 56.8, 32.4, and 21.7, along with their corresponding left and right singular vectors, were extracted. The left singular vector is a 10-dimensional column vector representing the spatial mode; the right singular vector is a 672-dimensional row vector representing the temporal mode. The first left singular vector is [0.32, 0.15, 0.45, 0.28, 0.18, 0.36, 0.42, 0.25, 0.30, 0.35], and the corresponding right singular vector reflects intraday periodicity. For each set of singular values ​​and their corresponding left and right singular vectors, calculate the outer product to obtain a component matrix. Add the three component matrices together to obtain a 10-row, 672-column dominant deviation component matrix.

[0101] The dominant deviation component matrix is ​​subjected to time difference processing to calculate the deviation change at adjacent time points. Taking a certain node as an example, the corresponding dominant deviation component is -6.5 kW and -8.2 kW at two adjacent time points, respectively, so the deviation change rate is -1.7 kW every 15 minutes. The deviation gradient is calculated based on the electrical coupling relationship, which is represented as a 10-row, 10-column adjacency matrix, where non-zero elements indicate a connection between nodes. For example, the dominant deviation components of nodes 1 and 2 are -6.5 kW and -4.2 kW, respectively. If the two nodes are directly connected, the deviation gradient between them is 2.3 kW. For each node, the supply-demand deviation is calculated by comprehensively considering the deviation change rate and the deviation gradient in each direction. Nodes whose supply-demand deviation exceeds a preset threshold of 5 kW are identified as imbalance source nodes. For example, the supply-demand deviation of node 3 is -7.8 kW, which exceeds the preset threshold and is marked as an imbalance source node.

[0102] A Laplace matrix is ​​constructed based on the electrical coupling relationship. The diagonal elements of the Laplace matrix represent the connectivity of the nodes, and the off-diagonal elements represent the negatives of the adjacency matrix. For a network of 10 nodes, a 10x10 Laplace matrix is ​​constructed, and the corresponding eigenvalues ​​and eigenvectors are calculated. The eigenvalues, sorted from smallest to largest, are: 0, 0.42, 0.85, 1.37, 1.92, 2.53, 3.12, 3.76, 4.35, 4.68. The eigenvector [0.25, 0.32, 0.18, 0.36, 0.42, 0.15, 0.28, 0.35, 0.30, 0.45] corresponding to the second smallest eigenvalue, 0.42, is extracted as the diffusion basis function. The supply-demand deviation of the imbalance source node is projected onto the diffusion basis function to obtain the diffusion coefficient. Taking node 3 as an example, its supply and demand deviation is -7.8 kW, the corresponding diffusion basis function component is 0.18, and the calculated diffusion coefficient is -1.404.

[0103] The propagation deviation of the imbalance source node to its neighboring nodes is calculated based on the diffusion coefficient. For example, node 3 is the imbalance source, node 1 is the corresponding neighboring node, and the component of the diffusion basis function for node 1 is 0.25. The calculated propagation deviation is -1.404 multiplied by 0.25 and divided by 0.18, which is -1.95 kW. The propagation deviation is accumulated over time to obtain the cumulative propagation deviation. Assuming an accumulation time of 1 hour (4 time points), the cumulative propagation deviation for node 1 is -7.8 kW. Setting a propagation threshold of 5 kW, if the cumulative propagation deviation of node 1 exceeds the threshold, node 1 is added to the initially empty set of affected nodes. The propagation deviation and cumulative propagation deviation of node 1 to its neighboring nodes are calculated again. If any node exceeds the propagation threshold, it is added to the set of affected nodes. This calculation of propagation deviation and cumulative propagation deviation is repeated until no new nodes are added to the set. The final set of affected nodes is {1, 5, 7, 9}, indicating that these nodes will be significantly affected by the imbalance source node 3.

[0104] In this embodiment, by calculating the difference between the short-term output sequence of each node and the load demand data and constructing the original deviation sequence, the degree of supply and demand mismatch is continuously quantified, improving the perception accuracy of the overall state of system operation deviation. By performing singular value decomposition on the original deviation sequence and extracting the dominant time mode and spatial mode, the main evolution mode in complex deviation data is effectively extracted, which can filter out random noise and minor disturbances, significantly improving the stability and robustness of imbalance feature identification. By performing time difference on the dominant deviation component and combining it with the electrical coupling relationship to calculate the deviation gradient, the supply and demand deviation amount is constructed, realizing a comprehensive evaluation of the deviation intensity and its changing trend, improving the accuracy of imbalance event judgment, and reducing false alarms and missed alarms.

[0105] In one alternative implementation,

[0106] The affected node set is mapped to the corresponding edge computing node, and a point-to-point communication connection is established when the imbalance event is triggered. Based on the point-to-point communication connection, the short-term output sequence and the operating boundary range are exchanged. Distributed collaborative data is constructed by combining the local load status information of the edge computing node, including:

[0107] Based on the geographical location information of the nodes in the affected node set, the nodes in the affected node set are mapped to the corresponding edge computing nodes. Based on the electrical coupling relationship, the affected nodes mapped to the same edge computing node are clustered and grouped to obtain the node grouping results.

[0108] When the imbalance event is triggered, a hierarchical coverage network topology is constructed between each edge computing node based on the node grouping results, and link quality is detected to obtain link quality parameters. Based on the link quality parameters, a link cost function is constructed and multi-objective optimization is performed to obtain the optimal path set and establish a point-to-point communication connection.

[0109] The short-term output sequence and the operating boundary range corresponding to each edge computing node are broadcast and transmitted through the point-to-point communication connection, and neighborhood collaborative state data are obtained by combining integrity verification and timing calibration.

[0110] Local load status information of each node in the corresponding node group is collected locally from each edge computing node. The local load status information and the neighborhood collaborative status data are timestamped and a status feature matrix is ​​constructed. The status feature matrix is ​​decomposed into a core tensor and a factor matrix, and tensor reconstruction is performed to obtain a dimension-reduced feature vector. The identification information corresponding to each edge computing node is added to the dimension-reduced feature vector to obtain the distributed collaborative data.

[0111] Mapping nodes to edge computing devices is performed based on the geographical location information of each node in the affected node set. In the example above, each node in the affected node set {1, 5, 7, 9} has its geographical coordinates, expressed in latitude and longitude. The geographical coordinates of node 1 are (35.12°N, 103.45°E), node 5 is (35.14°N, 103.47°E), node 7 is (35.16°N, 103.38°E), and node 9 is (35.08°N, 103.52°E). Three edge computing nodes are deployed in the area: edge computing node A is located at (35.13°N, 103.46°E), edge computing node B is located at (35.15°N, 103.40°E), and edge computing node C is located at (35.10°N, 103.50°E). Calculate the distance from each affected node to each edge computing node, and map them according to the nearest distance principle. Node 1 is 0.018° from edge computing node A, 0.068° from edge computing node B, and 0.074° from edge computing node C. Therefore, node 1 is mapped to edge computing node A. Similarly, node 5 is mapped to edge computing node A, node 7 to edge computing node B, and node 9 to edge computing node C.

[0112] Based on electrical coupling, affected nodes mapped to the same edge computing node are clustered and grouped. Nodes 1 and 5 are both mapped to edge computing node A. By querying the electrical coupling matrix, a direct connection is found between nodes 1 and 5, with an electrical coupling strength of 0.8. Nodes 1 and 5 are grouped together. Node 7 is mapped to edge computing node B, forming a separate group. Node 9 is mapped to edge computing node C, also forming a separate group. The resulting node groupings are {{1, 5}, {7}, {9}}. Edge computing node A is responsible for processing node group {1, 5}, edge computing node B is responsible for processing node group {7}, and edge computing node C is responsible for processing node group {9}.

[0113] When an imbalance event is triggered, a hierarchical overlay network topology is constructed based on the node grouping results. On the physical network, edge computing nodes are used as nodes in the overlay network, and logical connections are established. Logical connections are established between edge computing node A and edge computing node B, between edge computing node B and edge computing node C, and between edge computing node C and edge computing node A, forming a triangular topology. Link quality probing is performed on these three logical links. Using an active measurement method, 10 probe packets are sent to each link, and the round-trip time and packet loss rate are recorded. The average round-trip time for the link from edge computing node A to edge computing node B is 15 milliseconds, with a packet loss rate of 0.5%; the average round-trip time for the link from edge computing node B to edge computing node C is 25 milliseconds, with a packet loss rate of 1.2%; and the average round-trip time for the link from edge computing node C to edge computing node A is 20 milliseconds, with a packet loss rate of 0.8%.

[0114] A link cost function is constructed based on link quality parameters and optimized. The link cost function comprehensively considers round-trip time, packet loss rate, and link bandwidth. For each link, the cost value is calculated. The link cost value from edge computing node A to edge computing node B is 18.5, the link cost value from edge computing node B to edge computing node C is 32.4, and the link cost value from edge computing node C to edge computing node A is 24.8. For example, taking edge computing node A as the source node, the optimal paths to other edge computing nodes are calculated. The optimal path to edge computing node B is a direct connection with a cost value of 18.5; the optimal path to edge computing node C is also a direct connection with a cost value of 24.8. Similar calculations are performed for each pair of edge computing nodes, ultimately obtaining the optimal path set: {A→B, B→A, B→C, C→B, C→A, A→C}, and establishing point-to-point communication connections.

[0115] The short-term output sequence and operational boundary range of each edge computing node are broadcast via point-to-point communication. Edge computing node A broadcasts information from nodes 1 and 5, including the short-term output sequence and operational boundary range. Node 1's short-term output sequence is a predicted value every 15 minutes over the next 24 hours, such as [45.2, 47.8, 50.3, ..., 42.6] kW, and its operational boundary range is [42.1, 48.3] kW to [38.9, 45.1] kW; similar information for node 5 is also broadcast. Edge computing nodes B and C also broadcast information from their respective management nodes. The receiver performs integrity verification on the received data packets, calculates the checksum and compares it with the check field in the data packet to ensure data integrity. The timestamp is calibrated to correct possible clock deviations between different edge computing nodes, ensuring data timing consistency. Through the above processing, neighborhood collaborative state data is obtained.

[0116] Local load status information of each node within its corresponding node group is collected from each edge computing node. Edge computing node A collects the load status of nodes 1 and 5, including the current actual load value, load change rate, and voltage status. Node 1's current load is 43.8 kW, the load change rate is 0.2 kW / min, and the voltage is 380.5 V; Node 5's current load is 65.2 kW, the load change rate is -0.1 kW / min, and the voltage is 378.9 V. Edge computing nodes B and C also collect the load status information of their respective management nodes. The local load status information and the neighborhood collaborative status data are timestamped to ensure data temporal consistency. The time-aligned data is organized into a status feature matrix, where rows correspond to different nodes and columns correspond to different feature dimensions, forming a multi-dimensional data structure.

[0117] Tensor decomposition is performed on the state feature matrix. The state feature matrix is ​​represented as a three-dimensional tensor with dimensions of the number of nodes, the number of features, and the number of time points. This tensor is then decomposed into a core tensor and three factor matrices. The core tensor has a dimension of 3×3×3, representing the interaction relationship of the three principal components across the three dimensions. The first factor matrix has a dimension of node number × 3, representing the projection of nodes onto the three principal components; the second factor matrix has a dimension of feature number × 3, representing the projection of features onto the three principal components; and the third factor matrix has a dimension of time point × 3, representing the projection of time onto the three principal components. Tensor reconstruction is performed based on the core tensor and factor matrices to obtain the dimensionality-reduced feature vector. For the node group {1, 5} managed by edge computing node A, the dimensionality reduction yields a 12-dimensional feature vector [0.82, 0.65, 0.73, 0.45, 0.91, 0.58, 0.77, 0.63, 0.84, 0.52, 0.68, 0.79]. Add the edge computing node identifier "EC_A" to the dimensionality-reduced feature vector to form complete distributed collaborative data. Similarly, edge computing nodes B and C also generate corresponding distributed collaborative data.

[0118] In this embodiment, affected nodes are mapped to nearby edge computing nodes based on their geographical location. Furthermore, affected nodes within the same edge node are clustered based on their electrical coupling relationships. This avoids communication redundancy and uneven computational load caused by arbitrary or excessive node grouping, improving the targeting and efficiency of local collaborative processing. By constructing a layered edge computing network when an imbalance event is triggered, and performing multi-target path optimization based on real-time link quality detection results, point-to-point communication connections are established. This overcomes the problems of latency and packet loss that easily occur in complex network environments with fixed communication paths or single-index routing, improving the stability and real-time performance of collaborative communication. By broadcasting short-term output sequences and operating boundary ranges on point-to-point communication connections, and combining integrity checks and timing calibration, the consistency and timing accuracy of shared data between different edge computing nodes are ensured, enhancing the credibility of cross-node collaborative sensing results.

[0119] In one alternative implementation,

[0120] Based on the node grouping results, a hierarchical coverage network topology is constructed among the edge computing nodes, and link quality parameters are obtained through link quality detection. Based on these link quality parameters, a link cost function is constructed and multi-objective optimization is performed to obtain the optimal path set and establish point-to-point communication connections, including:

[0121] Based on the node grouping results, the number of nodes corresponding to each edge computing node and the connection strength of the electrical coupling relationship are counted, and the edge computing nodes are hierarchically divided to obtain core edge computing nodes and peripheral edge computing nodes. Full connectivity is established between the core edge computing nodes, and star connectivity is established between the peripheral edge computing nodes and the core edge computing nodes to obtain a hierarchical coverage network topology.

[0122] For each communication path in the hierarchical coverage network topology, the end-to-end transmission delay is determined by sending probe data packets and receiving response data packets. The number of lost probe data packets is counted and the packet loss rate is calculated. The available bandwidth of the communication path is measured and the link quality parameters are determined by combining the end-to-end transmission delay and the packet loss rate. Weighting coefficients are set for each type of data in the link quality parameters and the link cost function is determined by weighting.

[0123] With minimizing the link cost function as the optimization objective and maintaining connectivity and load balancing of the hierarchical coverage network topology as constraints, a multi-objective optimization solution is obtained to obtain a Pareto optimal solution set. Solutions that satisfy the constraints are selected from the Pareto optimal solution set and summarized to obtain an optimal path set. Based on the optimal path set, communication resources are allocated to edge computing nodes and point-to-point communication connections are established.

[0124] Based on the node grouping results, the number of nodes and the connection strength of electrical coupling relationships for each edge computing node are statistically analyzed. In the aforementioned embodiment, edge computing node A is responsible for node group {1, 5}, totaling 2 nodes; edge computing node B is responsible for node group {7}, totaling 1 node; and edge computing node C is responsible for node group {9}, totaling 1 node. The electrical coupling connection strength within edge computing node A is 0.8, representing the coupling strength between node 1 and node 5. Edge computing nodes B and C each have only 1 node, and their internal coupling strength is recorded as 0. The electrical coupling relationship strength between each edge computing node is calculated. Through the electrical connection topology diagram, it is determined that there are 2 electrical connections between edge computing node A and edge computing node B, with an average connection strength of 0.65; there is 1 electrical connection between edge computing node B and edge computing node C, with a connection strength of 0.4; and there are 3 electrical connections between edge computing node C and edge computing node A, with an average connection strength of 0.75.

[0125] Edge computing nodes are hierarchically divided. A hierarchy index is calculated based on the number of nodes managed by each edge computing node and the strength of its electrical coupling with other edge computing nodes. The hierarchy index of edge computing node A is 2 × (0.65 + 0.75) = 2.8; the hierarchy index of edge computing node B is 1 × (0.65 + 0.4) = 1.05; and the hierarchy index of edge computing node C is 1 × (0.4 + 0.75) = 1.15. A threshold is set according to actual needs; in this embodiment, the threshold is set to 1.5. Edge computing nodes with a hierarchy index greater than this threshold are designated as core edge computing nodes; otherwise, they are designated as peripheral edge computing nodes. Therefore, edge computing node A is classified as a core edge computing node, and edge computing nodes B and C are classified as peripheral edge computing nodes.

[0126] A full connectivity relationship is established between core edge computing nodes, and a star-shaped connectivity relationship is established between peripheral edge computing nodes and core edge computing nodes. In this embodiment, only edge computing node A is a core edge computing node, so a full connectivity relationship between core nodes is not required. A bidirectional connection is established between peripheral edge computing node B and core edge computing node A; a bidirectional connection is also established between peripheral edge computing node C and core edge computing node A. This forms a hierarchical coverage network topology, including four communication paths: A→B, B→A, A→C, and C→A.

[0127] Link quality probing was performed on each communication path in the hierarchical overlay network topology. For the connection A → B, 100 probe packets, each 256 bytes in size, were sent, and the round-trip time was recorded. Statistics showed an average end-to-end transmission latency of 12 milliseconds and a standard deviation of 1.5 milliseconds. During the probing, two probe packets were found to have no response, resulting in a packet loss rate of 2%. Using a bandwidth probing tool, the available bandwidth of this communication path was measured to be 50 megabits per second. Similarly, probing the connection B → A yielded an average end-to-end transmission latency of 13 milliseconds, a standard deviation of 1.8 milliseconds, a packet loss rate of 2.5%, and an available bandwidth of 48 megabits per second. Probing the connection A → C yielded an average end-to-end transmission latency of 18 milliseconds, a standard deviation of 2.1 milliseconds, a packet loss rate of 3%, and an available bandwidth of 45 megabits per second. The connection from C to A was probed, and the average end-to-end transmission delay was 17 milliseconds, the standard deviation was 2.0 milliseconds, the packet loss rate was 2.8%, and the available bandwidth was 46 megabits per second.

[0128] Weighting coefficients are assigned to each type of data in the link quality parameters, and the link cost function is determined by weighting these coefficients. For example, the weighting coefficient for end-to-end transmission delay is set to 0.5, the weighting coefficient for packet loss rate is set to 0.3, and the weighting coefficient for available bandwidth is set to 0.2. These weighting coefficients can be set according to task requirements. For available bandwidth, the reciprocal is used to convert it to a cost value; that is, the larger the bandwidth, the lower the cost. The cost of the link connecting A to B is calculated as follows: 0.5×12+0.3×2+0.2×(1 / 50×1000)=6.04; the cost of the link connecting B to A is calculated as follows: 0.5×13+0.3×2.5+0.2×(1 / 48×1000)=6.66; the cost of the link connecting A to C is calculated as follows: 0.5×18+0.3×3+0.2×(1 / 45×1000)=9.44; the cost of the link connecting C to A is calculated as follows: 0.5×17+0.3×2.8+0.2×(1 / 46×1000)=8.84.

[0129] The optimization objective is to minimize the link cost function, with connectivity preservation and load balancing of the hierarchical coverage network topology as constraints. The connectivity preservation constraint requires that any edge computing node can reach any other edge computing node. The load balancing constraint requires that the inbound and outbound connection loads of each edge computing node be relatively balanced, i.e., the inbound / outbound connection ratio does not exceed 3:1. A Pareto optimal solution set is obtained using a multi-objective optimization algorithm. Solutions satisfying the constraints are selected from the Pareto optimal solution set, including four communication paths: A→B, B→A, A→C, and C→A, forming the optimal path set.

[0130] Communication resources are allocated to edge computing nodes and point-to-point communication connections are established based on the optimal path set. For connection A→B, communication port number 6001 is allocated, the transmission buffer size is set to 512 kilobytes, the maximum transmission unit is 1460 bytes, and the initial congestion window value is 4. For connection B→A, communication port number 6002 is allocated, the transmission buffer size is set to 512 kilobytes, the maximum transmission unit is 1460 bytes, and the initial congestion window value is 4. For connection A→C, communication port number 6003 is allocated, the transmission buffer size is set to 512 kilobytes, the maximum transmission unit is 1460 bytes, and the initial congestion window value is 4. For connection C→A, communication port number 6004 is allocated, the transmission buffer size is set to 512 kilobytes, the maximum transmission unit is 1460 bytes, and the initial congestion window value is 4.

[0131] These point-to-point communication connections are established using the Transmission Control Protocol (TCP). Edge computing node A acts as a client connecting A to B and A to C, and simultaneously as a server connecting B to A and C to A, listening on the corresponding ports for connection requests. Edge computing node B acts as a client connecting B to A, and simultaneously as a server connecting A to B. Edge computing node C acts as a client connecting C to A, and simultaneously as a server connecting A to C. Clients initiate connection requests using the server's network address and port number to establish point-to-point communication connections. During transmission, the congestion window size is dynamically adjusted based on network conditions. When packet loss is detected, the congestion window size is halved; when an acknowledgment packet is successfully received, the congestion window size is linearly increased.

[0132] After establishing a point-to-point communication connection, connection status is monitored. A heartbeat packet is sent every 30 seconds. If there is no response after three consecutive heartbeat packets, the connection is considered interrupted, triggering a reconnection mechanism. The reconnection mechanism uses an exponential backoff algorithm, with an initial waiting time of 1 second. The waiting time doubles after each failed reconnection attempt, with a maximum waiting time of 30 seconds.

[0133] In this embodiment, the edge computing nodes are hierarchically divided by comprehensively considering the number of nodes carried by each edge computing node and the connection strength of the electrical coupling relationship. This ensures that the network topology matches the actual collaborative importance, avoiding the problem of network structure being disconnected from service load. It also improves the communication redundancy between key nodes and the fault tolerance of the overall network. By comprehensively measuring the end-to-end latency, packet loss rate, and available bandwidth of the communication paths in the hierarchical coverage network, the link status can be quantified and perceived in real time, providing a more reliable basis for subsequent path selection. By setting weights for different types of link quality parameters and constructing a unified link cost function, multi-dimensional communication performance indicators can be comprehensively weighed under the same optimization framework, improving the flexibility and adaptability of communication path selection.

[0134] Figure 2This is a flowchart illustrating the hierarchical coverage network topology construction process of the real-time control method for distributed energy based on edge computing, as described in an embodiment of the present invention.

[0135] In one alternative implementation,

[0136] Based on the distributed collaborative data, an initial control scheme is obtained by solving for the active power output adjustment and reactive power compensation with the goal of minimizing energy transmission loss. This scheme is then executed. The boundary power flow deviation caused by the initial control scheme is calculated and sent to adjacent edge computing nodes. This process is repeated until the boundary power flow deviation converges, resulting in a global control scheme, including:

[0137] Dimensionally reduced feature vectors and local load status information are extracted from the distributed collaborative data, and regional energy network topology of each edge computing node is constructed. Based on the local load status information, the current active power output and current reactive power output of each node are determined. An optimization function is established with the goal of minimizing energy transmission loss, and the active power output adjustment and reactive power compensation of each node are solved to obtain the initial control scheme.

[0138] The initial control scheme is sent to each node and the control action is executed. The real-time voltage and real-time current of each node are collected after execution, and the power flow value corresponding to the boundary node of the regional energy network topology is calculated. The power flow value is compared with the preset power flow value corresponding to the boundary node to obtain the boundary power flow deviation.

[0139] The boundary power flow deviation is sent to the adjacent edge computing node through the point-to-point communication connection. It is determined whether the boundary power flow deviation is less than the preset convergence threshold. If it is not less than the threshold, the boundary power flow deviation fed back by the adjacent edge computing node is received and used as the boundary constraint condition to update the optimization function. The optimization function is re-solved based on the updated optimization function until the boundary power flow deviation is less than the convergence threshold to obtain the global control scheme. If it is less than the threshold, the current control scheme is determined as the global control scheme and output for execution.

[0140] Dimensionally reduced feature vectors and local load status information are extracted from distributed collaborative data. Taking edge computing node A as an example, dimensionality-reduced feature vectors [0.82, 0.65, 0.73, 0.45, 0.91, 0.58, 0.77, 0.63, 0.84, 0.52, 0.68, 0.79] and local load status information of nodes 1 and 5 are extracted from its distributed collaborative data, including the current actual load, load change rate, and voltage status. The current load of node 1 is 43.8 kW, the load change rate is 0.2 kW / min, and the voltage is 380.5 V; the current load of node 5 is 65.2 kW, the load change rate is -0.1 kW / min, and the voltage is 378.9 V.

[0141] Construct a regional energy network topology for edge computing node A. Based on electrical connections, there is a direct connection between node 1 and node 5, with an electrical coupling strength of 0.8. The regional energy network topology includes nodes 1 and 5 and their corresponding connections. Connections exhibit impedance characteristics, with the real part representing resistance and the imaginary part representing reactance. Similarly, the regional energy network topology constructed by edge computing node B includes node 7, and the regional energy network topology constructed by edge computing node C includes node 9.

[0142] The current active and reactive power outputs of each node are determined based on local load status information. The current active power output of node 1 is 43.8 kW, and the current reactive power output is calculated to be 14.4 kV based on the measured power factor of 0.95. The current active power output of node 5 is 65.2 kW, the power factor is 0.92, and the current reactive power output is 27.3 kV.

[0143] An optimization function is established with the objective of minimizing energy transmission loss, which is proportional to the line resistance and the square of the current. In the connection between node 1 and node 5, the transmission loss is calculated as resistance multiplied by the square of the current. The resistance is 0.05 ohms, and the current is determined by the power transmission between the two nodes. Currently, node 1 transmits 15.6 kW of power to node 5, resulting in a calculated current of 41.3 amperes and a transmission loss of 85.5 watts. The optimization function considers the impact of active power output adjustments and reactive power compensation on transmission loss at each node, while also taking into account the output boundary constraints of each node. The upper and lower limits of active power output for node 1 are [38.9, 48.3] kW, and the upper and lower limits of reactive power output are [10.0, 20.0] kV; the upper and lower limits of active power output for node 5 are [60.1, 70.5] kW, and the upper and lower limits of reactive power output are [20.0, 35.0] kV.

[0144] The gradient descent method was used to solve the optimization problem. The initial point was set to the current output state, the step size was set to 0.01, and the maximum number of iterations was set to 100. After 72 iterations, the algorithm converged. The calculated active power output adjustment for node 1 was +2.1 kW, and the reactive power compensation was -1.2 kV; the active power output adjustment for node 5 was -1.8 kW, and the reactive power compensation was +2.5 kV. Based on this, the initial control scheme was obtained: the active power output of node 1 was adjusted to 45.9 kW, and the reactive power output was adjusted to 13.2 kV; the active power output of node 5 was adjusted to 63.4 kW, and the reactive power output was adjusted to 29.8 kV.

[0145] The initial control scheme is issued to each node and control actions are executed. Control commands are sent to the control devices of nodes 1 and 5 via the Industrial Ethernet protocol. Upon receiving the commands, the control devices execute the corresponding control actions. For node 1, active power output is increased by 2.1 kW, and reactive power output is decreased by 1.2 kV; for node 5, active power output is decreased by 1.8 kW, and reactive power output is increased by 2.5 kV.

[0146] After data collection and execution, the real-time voltage and current of each node were recorded. Using a power quality analyzer, the real-time voltage of node 1 was measured to be 381.2 volts and the real-time current to be 121.3 amperes; the real-time voltage of node 5 was measured to be 379.5 volts and the real-time current to be 167.8 amperes. The power flow values ​​corresponding to the boundary nodes of the regional energy network topology were calculated. Boundary nodes include nodes connected to other regional energy networks. Node 1 is connected to node 7 managed by edge computing node B, and the calculated power flow value between the two nodes is 12.8 kilowatts; node 5 is connected to node 9 managed by edge computing node C, and the calculated power flow value between the two nodes is 18.5 kilowatts.

[0147] The boundary current deviation is obtained by comparing the current flow value with the preset current flow value corresponding to the boundary node. The preset current flow value between node 1 and node 7 is 15.0 kW, the actual current flow value is 12.8 kW, and the deviation is -2.2 kW; the preset current flow value between node 5 and node 9 is 20.0 kW, the actual current flow value is 18.5 kW, and the deviation is -1.5 kW.

[0148] The boundary power flow deviation is sent to the adjacent edge computing nodes via point-to-point communication connections. Edge computing node A sends the boundary power flow deviation of -2.2 kW between node 1 and node 7 to edge computing node B via the aforementioned point-to-point communication connection A→B; and sends the boundary power flow deviation of -1.5 kW between node 5 and node 9 to edge computing node C via the point-to-point communication connection A→C.

[0149] Determine whether the boundary power flow deviation is less than the preset convergence threshold. For example, the convergence threshold is set to 1.0 kW. The boundary power flow deviation between node 1 and node 7 is -2.2 kW, which is greater than the convergence threshold. The boundary power flow deviation between node 5 and node 9 is -1.5 kW, which is also greater than the convergence threshold. Therefore, it is necessary to continue iterating.

[0150] The system receives boundary power flow deviations from adjacent edge computing nodes and uses them as boundary constraint conditions to update the optimization function. Edge computing node A connects to node B via point-to-point communication and receives a boundary power flow deviation of +2.2 kW from node B; it also connects to node C via point-to-point communication and receives a boundary power flow deviation of +1.5 kW from node C. These boundary power flow deviations are used as boundary constraint conditions to update the optimization function. The constraints require an increase of 2.2 kW in the power flow value between nodes 1 and 7, and an increase of 1.5 kW in the power flow value between nodes 5 and 9.

[0151] The solution was recalculated based on the updated optimization function. The same gradient descent method was used, with the initial point set to the current output state, a step size of 0.01, and an upper limit of 100 iterations. After 85 iterations, the algorithm converged. The calculated active power output adjustment for node 1 was +3.5 kW, and the reactive power compensation was -1.8 kV; the active power output adjustment for node 5 was +2.2 kW, and the reactive power compensation was +1.9 kV. Based on this, a new control scheme was obtained: the active power output of node 1 was adjusted to 47.3 kW, and the reactive power output to 12.6 kV; the active power output of node 5 was adjusted to 67.4 kW, and the reactive power output to 29.2 kV.

[0152] The new control scheme is issued to each node and the control actions are executed. The real-time voltage and current of each node are collected again after execution. The real-time voltage of node 1 is 380.8 volts and the real-time current is 125.6 amps; the real-time voltage of node 5 is 379.1 volts and the real-time current is 178.2 amps. The boundary power flow values ​​and deviations are recalculated. The actual power flow value between node 1 and node 7 is 14.6 kW, and the boundary power flow deviation is -0.4 kW; the actual power flow value between node 5 and node 9 is 19.4 kW, and the deviation is -0.6 kW.

[0153] Determine if the new boundary power flow deviation is less than the preset convergence threshold. The boundary power flow deviation between node 1 and node 7 is -0.4 kW, which is less than the convergence threshold of 1.0 kW. The boundary power flow deviation between node 5 and node 9 is -0.6 kW, which is less than the convergence threshold of 1.0 kW. Therefore, the iterative process converges. The current control scheme is determined as the global control scheme and the following outputs are executed: the active power output of node 1 is 47.3 kW, and the reactive power output is 12.6 kV; the active power output of node 5 is 67.4 kW, and the reactive power output is 29.2 kV.

[0154] In this embodiment, by constructing the regional energy network topology based on distributed collaborative data and combining local load status information, the active and reactive power output status of each node is accurately determined. This avoids control deviations caused by global information lag or coarse modeling, improving the rationality and feasibility of the initial control scheme. By simultaneously solving for active power output adjustment and reactive power compensation with the goal of minimizing energy transmission losses, node regulation not only focuses on power balance but also takes into account voltage levels and line losses, significantly improving the operating efficiency and power quality of the regional energy network. By collecting voltage and current data in real time after control execution and calculating the power flow deviation of boundary nodes, the power interaction status between regions can be dynamically perceived, improving the timeliness of response to cross-regional coupling effects.

[0155] A second aspect of this invention provides a real-time control system for distributed energy based on edge computing, comprising:

[0156] The preprocessing unit is used to acquire grid operation data of distributed energy nodes, establish electrical coupling relationships between nodes by combining the geographical location information of the distributed energy nodes to obtain an energy operation dataset, perform multi-scale wavelet decomposition on the energy operation dataset to obtain high-frequency components and low-frequency components, perform autoregressive moving average processing on the high-frequency components to obtain a short-term power output sequence, and perform trend fitting on the low-frequency components to obtain the operating boundary range.

[0157] An imbalance identification unit is used to calculate the supply-demand deviation based on the short-term output sequence and the load demand data in the energy operation dataset, identify imbalance events where the supply-demand deviation exceeds a preset threshold, and extract the set of affected nodes based on the electrical coupling relationship.

[0158] The collaborative communication unit is used to map the affected node set to the corresponding edge computing node and establish a point-to-point communication connection when the imbalance event is triggered. Based on the point-to-point communication connection, the short-term output sequence and the operating boundary range are exchanged, and distributed collaborative data is constructed by combining the local load status information of the edge computing node.

[0159] The optimization control unit is used to solve for the active power output adjustment and reactive power compensation based on the distributed collaborative data, with the goal of minimizing energy transmission loss, to obtain an initial control scheme and execute it. It calculates the boundary power flow deviation caused by the initial control scheme and sends it to the adjacent edge computing nodes. The solution is repeated until the boundary power flow deviation converges to obtain a global control scheme.

[0160] A third aspect of the present invention provides an electronic device, comprising:

[0161] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0162] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0163] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time control method for distributed energy resources based on edge computing, characterized in that, include: The grid operation data of distributed energy nodes are obtained, and the electrical coupling relationship between nodes is established by combining the geographical location information of the distributed energy nodes to obtain an energy operation dataset. Multi-scale wavelet decomposition is performed on the energy operation dataset to obtain high-frequency components and low-frequency components. Autoregressive moving average processing is performed on the high-frequency components to obtain a short-term power output sequence, and trend fitting is performed on the low-frequency components to obtain the operating boundary range. The supply-demand deviation is calculated based on the short-term output sequence and the load demand data in the energy operation dataset. Imbalance events where the supply-demand deviation exceeds a preset threshold are identified, and the set of affected nodes is extracted based on the electrical coupling relationship. The affected node set is mapped to the corresponding edge computing node and a point-to-point communication connection is established when the imbalance event is triggered. The short-term output sequence and the operating boundary range are exchanged based on the point-to-point communication connection. Distributed collaborative data is constructed by combining the local load status information of the edge computing node. Based on the distributed collaborative data, the active power output adjustment and reactive power compensation are solved with the goal of minimizing energy transmission loss to obtain an initial control scheme and execute it. The boundary power flow deviation caused by the initial control scheme is calculated and sent to the adjacent edge computing nodes. The solution is repeated until the boundary power flow deviation converges to obtain a global control scheme.

2. The method according to claim 1, characterized in that, The energy operation dataset obtained by acquiring grid operation data from distributed energy nodes and establishing electrical coupling relationships between nodes based on the geographical location information of the distributed energy nodes includes: Real-time power generation and voltage amplitude of distributed energy nodes are collected to obtain operating status data; real-time load power demand of the area where the distributed energy nodes are located is collected to obtain load demand data; the connection relationship between distributed energy nodes and transmission capacity limit are read to obtain the grid topology connection relationship; and the grid operation data is obtained by combining the operating status data and the load demand data. Based on the real-time power generation in the operating status data and the load demand data, the power surplus or power deficit of each distributed energy node is calculated, and the distributed energy nodes are classified to obtain a set of power surplus nodes and a set of power deficit nodes. Calculate the geographical distance between each node in the power surplus node set and each node in the power deficit node set, and mark the node pairs whose geographical distance is less than a preset distance threshold and have a connection relationship in the power grid topology connection relationship as potential coupled node pairs; Based on the line parameters of the connection path in the power grid topology connection relationship and the voltage amplitude in the operating status data, the effective transmission capacity corresponding to the potential coupling node pair is calculated. The potential coupling node pair with the effective transmission capacity greater than the preset capacity threshold is taken as the coupling node pair and the electrical coupling relationship is constructed using the effective transmission capacity as a quantitative index. The energy operation dataset is obtained by integrating the power grid operation data and the electrical coupling relationships between the nodes.

3. The method according to claim 1, characterized in that, Multi-scale wavelet decomposition is performed on the energy operation dataset to obtain high-frequency and low-frequency components. Autoregressive moving average processing is applied to the high-frequency components to obtain a short-term output sequence. Trend fitting is performed on the low-frequency components to obtain the operating boundary range, which includes: The operating status data and load demand data of each distributed energy node within a preset time window are extracted from the energy operation dataset and time series data is constructed. The time series data is subjected to multi-level wavelet decomposition. The high-frequency components obtained from each level of decomposition are merged to obtain the high-frequency components. The low-frequency components obtained from the last level of decomposition are taken as the low-frequency components. The autocorrelation coefficient sequence and partial autocorrelation coefficient sequence of the high-frequency components are calculated, and the order of the autoregressive term and the order of the moving average term are determined. Extract the historical values ​​and historical prediction errors corresponding to the high-frequency components, linearly weight the historical values ​​according to the order of the autoregressive term to obtain the autoregressive term, linearly weight the historical prediction errors according to the order of the moving average term to obtain the moving average term, and solve the short-term output sequence based on the autoregressive term and the moving average term. Historical values ​​corresponding to the low-frequency components are extracted and polynomial fitting is performed to obtain a low-frequency trend function. The maximum and minimum values ​​of the low-frequency components are calculated. The difference between the maximum value and the low-frequency trend function is taken as the upper boundary deviation, and the difference between the low-frequency trend function and the minimum value is taken as the lower boundary deviation. Based on the upper boundary deviation and the lower boundary deviation, the upper and lower limits of the output of each distributed energy node are determined to obtain the operating boundary range.

4. The method according to claim 1, characterized in that, The supply-demand deviation is calculated based on the short-term output sequence and the load demand data in the energy operation dataset. Imbalance events where the supply-demand deviation exceeds a preset threshold are identified, and the set of affected nodes is extracted based on the electrical coupling relationship, including: Extract the load demand data from the energy operation dataset, and calculate the difference between the output value of each node in the short-term output sequence and the load demand data to obtain the original deviation sequence; The original deviation sequence is constructed and singular value decomposition is performed. The maximum singular value of the first preset number and the corresponding left singular vector and right singular vector are extracted and used as the time mode and spatial mode. The dominant deviation component is obtained by performing an outer product on the time mode and spatial mode. The deviation rate of change is obtained by time difference of the dominant deviation component, and the deviation gradient is calculated based on the electrical coupling relationship. The supply and demand deviation amount is obtained by solving based on the deviation rate of change and the deviation gradient. The node whose supply and demand deviation amount exceeds the preset deviation threshold is identified as the imbalance source node and the imbalance event is determined. Based on the electrical coupling relationship, a Laplace matrix is ​​constructed and the eigenvectors corresponding to the non-zero minimum eigenvalues ​​are extracted as diffusion basis functions. The supply and demand deviation is projected onto the diffusion basis functions to obtain the diffusion coefficient. Based on the diffusion coefficient, the propagation deviation corresponding to the imbalance source node is calculated and accumulated over time to obtain the cumulative propagation deviation. Identify neighboring nodes whose cumulative propagation deviation exceeds the propagation threshold and add them to a pre-initialized empty set. Repeat the calculation of the diffusion coefficient and cumulative propagation deviation until no new nodes are added to obtain the set of affected nodes.

5. The method according to claim 1, characterized in that, The affected node set is mapped to the corresponding edge computing node, and a point-to-point communication connection is established when the imbalance event is triggered. Based on the point-to-point communication connection, the short-term output sequence and the operating boundary range are exchanged. Distributed collaborative data is constructed by combining the local load status information of the edge computing node, including: Based on the geographical location information of the nodes in the affected node set, the nodes in the affected node set are mapped to the corresponding edge computing nodes. Based on the electrical coupling relationship, the affected nodes mapped to the same edge computing node are clustered and grouped to obtain the node grouping results. When the imbalance event is triggered, a hierarchical coverage network topology is constructed between each edge computing node based on the node grouping results, and link quality is detected to obtain link quality parameters. Based on the link quality parameters, a link cost function is constructed and multi-objective optimization is performed to obtain the optimal path set and establish a point-to-point communication connection. The short-term output sequence and the operating boundary range corresponding to each edge computing node are broadcast and transmitted through the point-to-point communication connection, and neighborhood collaborative state data are obtained by combining integrity verification and timing calibration. Local load status information of each node in the corresponding node group is collected locally from each edge computing node. The local load status information and the neighborhood collaborative status data are timestamped and a status feature matrix is ​​constructed. The status feature matrix is ​​decomposed into a core tensor and a factor matrix, and tensor reconstruction is performed to obtain a dimension-reduced feature vector. The identification information corresponding to each edge computing node is added to the dimension-reduced feature vector to obtain the distributed collaborative data.

6. The method according to claim 5, characterized in that, Based on the node grouping results, a hierarchical coverage network topology is constructed among the edge computing nodes, and link quality parameters are obtained through link quality detection. Based on these link quality parameters, a link cost function is constructed and multi-objective optimization is performed to obtain the optimal path set and establish point-to-point communication connections, including: Based on the node grouping results, the number of nodes corresponding to each edge computing node and the connection strength of the electrical coupling relationship are counted, and the edge computing nodes are hierarchically divided to obtain core edge computing nodes and peripheral edge computing nodes. Full connectivity is established between the core edge computing nodes, and star connectivity is established between the peripheral edge computing nodes and the core edge computing nodes to obtain a hierarchical coverage network topology. For each communication path in the hierarchical coverage network topology, the end-to-end transmission delay is determined by sending probe data packets and receiving response data packets. The number of lost probe data packets is counted and the packet loss rate is calculated. The available bandwidth of the communication path is measured and the link quality parameters are determined by combining the end-to-end transmission delay and the packet loss rate. Weighting coefficients are set for each type of data in the link quality parameters and the link cost function is determined by weighting. With minimizing the link cost function as the optimization objective and maintaining connectivity and load balancing of the hierarchical coverage network topology as constraints, a multi-objective optimization solution is obtained to obtain a Pareto optimal solution set. Solutions that satisfy the constraints are selected from the Pareto optimal solution set and summarized to obtain an optimal path set. Based on the optimal path set, communication resources are allocated to edge computing nodes and point-to-point communication connections are established.

7. The method according to claim 1, characterized in that, Based on the distributed collaborative data, an initial control scheme is obtained by solving for the active power output adjustment and reactive power compensation with the goal of minimizing energy transmission loss. This scheme is then executed. The boundary power flow deviation caused by the initial control scheme is calculated and sent to adjacent edge computing nodes. This process is repeated until the boundary power flow deviation converges, resulting in a global control scheme, including: Dimensionally reduced feature vectors and local load status information are extracted from the distributed collaborative data, and regional energy network topology of each edge computing node is constructed. Based on the local load status information, the current active power output and current reactive power output of each node are determined. An optimization function is established with the goal of minimizing energy transmission loss, and the active power output adjustment and reactive power compensation of each node are solved to obtain the initial control scheme. The initial control scheme is sent to each node and the control action is executed. The real-time voltage and real-time current of each node are collected after execution, and the power flow value corresponding to the boundary node of the regional energy network topology is calculated. The power flow value is compared with the preset power flow value corresponding to the boundary node to obtain the boundary power flow deviation. The boundary power flow deviation is sent to the adjacent edge computing node through the point-to-point communication connection. It is determined whether the boundary power flow deviation is less than the preset convergence threshold. If it is not less than the threshold, the boundary power flow deviation fed back by the adjacent edge computing node is received and used as the boundary constraint condition to update the optimization function. The optimization function is re-solved based on the updated optimization function until the boundary power flow deviation is less than the convergence threshold to obtain the global control scheme. If it is less than the threshold, the current control scheme is determined as the global control scheme and output for execution.

8. A distributed energy real-time control system based on edge computing, used to implement the method of any one of claims 1-7, characterized in that, include: The preprocessing unit is used to acquire grid operation data of distributed energy nodes, establish electrical coupling relationships between nodes by combining the geographical location information of the distributed energy nodes to obtain an energy operation dataset, perform multi-scale wavelet decomposition on the energy operation dataset to obtain high-frequency components and low-frequency components, perform autoregressive moving average processing on the high-frequency components to obtain a short-term power output sequence, and perform trend fitting on the low-frequency components to obtain the operating boundary range. An imbalance identification unit is used to calculate the supply-demand deviation based on the short-term output sequence and the load demand data in the energy operation dataset, identify imbalance events where the supply-demand deviation exceeds a preset threshold, and extract the set of affected nodes based on the electrical coupling relationship. The collaborative communication unit is used to map the set of affected nodes to the corresponding edge computing nodes and establish a point-to-point communication connection when the imbalance event is triggered. Based on the point-to-point communication connection, the short-term output sequence and the operating boundary range are exchanged, and distributed collaborative data is constructed by combining the local load status information of the edge computing nodes. The optimization control unit is used to solve for the active power output adjustment and reactive power compensation based on the distributed collaborative data, with the goal of minimizing energy transmission loss, to obtain an initial control scheme and execute it. It calculates the boundary power flow deviation caused by the initial control scheme and sends it to the adjacent edge computing nodes. The solution is repeated until the boundary power flow deviation converges to obtain a global control scheme.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images