A perceptual deep learning resource virtualization energy efficiency optimization system

Abstract

This invention provides a perceptual deep learning resource virtualization energy efficiency optimization system, relating to the field of energy efficiency optimization technology. It includes a centralized control platform responsible for overall reactive power coordination and control optimization, and issuing reactive power control commands to subordinate substations; a field automatic controller for receiving SVG signals and commands from the centralized control center in three operating modes; a field device layer including an FC automatic switching electrical cabinet that receives commands from the field automatic controller and controls the on / off switching of internal contactors, thereby achieving automatic capacitor switching; the centralized control platform also includes a dot matrix mapping module; the dot matrix model automatically generates the topology, requiring only the addition of corresponding feature nodes when adding / modifying distribution network equipment, shortening the graphic update time; maintenance personnel can quickly grasp the distribution network status without viewing backend data; the dot matrix graphic links historical data with real-time control strategies, assisting maintenance personnel in accurately judging FC switching needs, avoiding blind operation, and reducing the number of voltage fluctuations caused by improper control.

Description

Technical Field

[0001] This invention relates to the field of energy efficiency optimization technology, specifically to a perceptual deep learning resource virtualization energy efficiency optimization system. Background Technology

[0002] The existing FC capacity in the substation is insufficient and slow to respond. Although SVG has better performance, it is expensive. There is an urgent need for coordinated control between the two. Adjusting the distribution network topology requires manual redrawing, which is time-consuming and prone to errors. It is impossible to synchronize the FC switching status, SVG compensation capacity, node voltage and other operating data in real time. The graphics are out of sync with the actual control status. The distribution network graphics are only used as a static topology display and cannot be linked with the SVG-FC coordinated control strategy (such as the inability to intuitively locate nodes with insufficient reactive power compensation and the inability to visualize the FC switching sequence). As a result, it is difficult for operation and maintenance personnel to quickly judge the control effect and fault location.

[0003] In existing technologies, distribution network mapping and reactive power compensation control are independent: the control system only outputs data and instructions, and the graphical system only statically displays the topology. The two lack data communication and functional linkage, which restricts the efficiency of distribution network operation and maintenance and the intuitiveness of control decisions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a perceptual deep learning resource virtualization energy efficiency optimization system.

[0005] A perceptual deep learning resource virtualization energy efficiency optimization system includes a centralized control platform, which is responsible for the overall reactive power coordination and control optimization and issues reactive power control commands to subordinate substations. The system also integrates a deep learning perceptual optimization module, a resource virtualization module, and a dot matrix mapping module.

[0006] The deep learning perception optimization module is used to provide the centralized control platform with multi-source data processing, intelligent perception and global control strategy optimization.

[0007] The resource virtualization module is used to map the physical devices of the distribution network into virtual resource units to achieve global scheduling.

[0008] The field automatic controller has three operating modes: receiving SVG signals and commands from the centralized control center.

[0009] The field equipment layer includes the FC automatic switching electrical cabinet, which receives instructions from the field automatic controller and controls the on / off state of the internal contactors, thereby realizing the automatic switching of capacitors.

[0010] The centralized control platform also includes a dot matrix mapping module, the specific working steps of which are as follows:

[0011] Step S1: Construct the distribution network matrix model:

[0012] A two-dimensional grid is established based on the physical space of the distribution network. The coordinate system is the substation relative coordinates with the center point of the substation as the origin (0,0), the X-axis as the east-west direction and the Y-axis as the north-south direction, or the absolute coordinates of latitude and longitude.

[0013] The key elements of the distribution network are mapped to feature nodes in the dot matrix and assigned unique IDs and attribute identifiers. The key elements of the distribution network include distribution network bus, FC equipment, SVG equipment, distributed photovoltaic access points and line branch points. The corresponding feature node types are bus node, FC node, SVG node, photovoltaic node and topology node, respectively. The attribute identifier of each feature node contains control association data.

[0014] The distribution network topology is automatically identified and a matrix topology is generated by determining the adjacency relationship of the matrix nodes.

[0015] Step S2: Collect distribution network data uploaded by the field automatic controller in real time through the Ethernet interface. The distribution network data includes voltage, current, FC switching status and SVG compensation capacity.

[0016] Map the distribution network data to the corresponding matrix nodes according to the following rules:

[0017] Continuous data is displayed next to feature nodes in the form of node labels, discrete states are distinguished by node color and icon, and FC switching instructions issued by the central control platform are marked in the dot matrix as dynamic arrows with the arrows pointing from the central control node to the target FC node.

[0018] Step S3: Two graphics generation modes are built-in: topology simplification mode and panoramic detail mode. The topology simplification mode retains the bus nodes, FC nodes, SVG nodes and their connection relationships and deletes redundant grids to generate the control core topology map. The panoramic detail mode retains all feature nodes and overlays real-time node data annotations to generate the full distribution network status map.

[0019] Step S4: Set up two update triggering mechanisms: event-triggered update and timed-triggered update. The event-triggered update is when the field automatic controller detects a key state change and sends an update signal to the central control platform. After receiving the signal, the dot matrix mapping module updates the color, label or icon of the corresponding node. The key state changes include FC switching, SVG fault and voltage exceeding the threshold.

[0020] Step S5: When maintenance personnel click on the target node in the dot matrix graphic, the system automatically pops up the historical operating data and current control strategy of the node. After issuing the FC switching command, the switching process is displayed through a dynamic progress bar, and the node color changes synchronously after the switching is completed. When the node voltage exceeds the threshold, the corresponding bus node in the dot matrix flashes yellow and automatically marks the FC / SVG device associated with the node.

[0021] Preferably, the dot matrix mapping module further includes intelligent layout and rendering of device elements:

[0022] Step T1: Construct a dedicated element library for distribution network equipment. Associate the busbar segment elements corresponding to the double-horizontal-line structure of the core distribution network equipment busbar nodes with characteristic node types, and the capacitor elements with plate symbols corresponding to FC nodes. Simultaneously, add feeder segment elements, distribution transformer elements, combined switch elements, circuit breaker elements, tower elements, fault indicator elements, and voltage transformer elements. The element styles are designed based on the actual physical structure of the equipment, and the element size is positively correlated with the rated capacity of the equipment; the larger the rated capacity of the equipment, the larger the element size should be to ensure visual recognizability.

[0023] Step T2: Based on the topological relationship between physical connection point I and physical connection point J in the device model, optimize the adjacency relationship recognition rules of the dot matrix topology matrix;

[0024] When two device nodes have a corresponding association between their physical connection point I and physical connection point J, they are determined to be directly connected, thus achieving topology self-adaptive layout, specifically including:

[0025] Single feeder scenario layout: Taking the actual power supply path of a single feeder as the core, the nodes are arranged in the following order: substation outgoing end, overhead line or cable, ring network cabinet, switch station, distribution station, box-type substation, cable branch box, distribution transformer, and line end switch. The towers are retained and the nodes are set according to the actual segment spacing of the line. The fault indicator is retained and the nodes are set at the branch positions of the line.

[0026] Station room layout: The physical space boundary of the independent station building is used to divide the grid area. The nodes are arranged in the order of station busbar, circuit breaker, load switch, voltage transformer, station transformer and bay outgoing terminal. Each node is marked with a unique number for physical connection point I and physical connection point J.

[0027] Connection scenario layout: Integrate two or more feeder nodes with electrical connections into the same lattice space, set exclusive marks at the nodes where the connection switches are located and distinguish them by thick borders, mark the boundaries of different feeders with dashed lines, and adjust the style of the dashed lines according to the topology differentiation requirements, and retain the main trunk equipment and branch key equipment of all feeders.

[0028] Step T3: Use a three-layer rendering logic: The bottom layer renders the device connection lines according to the scene type, using solid lines for single feeder scenes and dashed lines for feeder boundaries in interconnected scenes. The line width is set according to visual clarity requirements, and virtual resource scheduling channel identifiers are overlaid; the middle layer loads the dedicated primitives constructed in step T1, with the center point of the primitive precisely aligned with the node coordinates of the dot matrix model; the top layer labels the device name and device abbreviation, and also labels the physical connection point I and physical connection point J numbers and the core attributes of the virtual resource unit.

[0029] Preferably, the dot matrix mapping module further includes:

[0030] Equipment model data classification and linkage: Connect to the list of power distribution network equipment models to enable filtering and displaying matrix nodes by equipment type. When a type of equipment is selected, only the matrix nodes corresponding to that type of equipment are displayed. Clicking on any node allows you to view the core attributes of the equipment. The attribute data and the attribute identifiers of the feature nodes are synchronized in real time.

[0031] It supports exporting device lists. The list of disconnectors includes disconnector name, voltage level, physical connection point I, physical connection point J, and installation location. The list of distribution transformers includes transformer name, rated capacity, voltage level, physical connection point I, and physical connection point J. The list data is dynamically matched with the attributes of the dot matrix nodes.

[0032] Single feeder interaction: Supports filtering and displaying dispatch-managed equipment, retaining only the matrix nodes corresponding to substation outgoing lines, circuit breakers, load switches, distribution transformers, and fault indicators, and hiding user-side equipment nodes not under dispatch management; when clicking on a tower node, the equipment association relationship of the line segments on both sides of the tower is automatically displayed, and the association relationship is generated based on the matrix topology matrix;

[0033] Station Interaction: Supports switching between stations. Stations are divided into zones according to the wiring functions within the station. After switching, the dot matrix automatically focuses on the equipment nodes of the corresponding zone and uses a light-colored background to distinguish the zone range. At the same time, the wiring logic within the zone is marked. The wiring logic is generated based on the association between physical connection point I and physical connection point J.

[0034] Interactive communication: Supports viewing feeder associations. When a communication switch node is selected, all device nodes of the two feeders associated with that switch are automatically highlighted. The highlighting logic is based on the adjacency relationship of the topology matrix. At the same time, a communication relationship table is generated, which contains the fields of feeder number, communication switch name, associated device type, physical connection point I, and physical connection point J.

[0035] Model data synchronization and export: When the graphics are updated, the changes in the equipment model data are automatically synchronized, including adjustments to the equipment voltage level and modifications to the association between physical connection point I and physical connection point J;

[0036] When a fault signal is detected, the corresponding fault indicator node in the dot matrix starts flashing mode, and the flashing frequency is set according to the fault warning requirements; at the same time, based on the dot matrix topology matrix, the upstream and downstream device nodes associated with the fault indicator are automatically traced to generate a fault impact range map.

[0037] The diagram showing the scope of the fault indicates the physical connection points I and J that need to be investigated.

[0038] Preferably, the centralized control platform specifically performs coordinated control of SVG and FC across the entire domain based on the reactive power control requirements, and calculates the specific reactive power target parameters of the subordinate stations and transmits them to the field automatic controller.

[0039] The field automatic controller intelligently identifies the device's operating mode through port identification. When the network-wide coordinated control command is detected, the field automatic controller automatically enters the first mode and receives instructions from the centralized control platform to automatically switch FC.

[0040] When only the SVG signal within the station is detected, the system enters the second mode and performs automatic FC switching using the built-in time gradient algorithm.

[0041] When the above-mentioned network-wide and station-specific commands cannot be identified, the field automatic controller automatically enters the third mode, and detects voltage and current through communication with the original FC automatic switching device or the station-specific measurement cabinet, and automatically performs reactive power compensation.

[0042] In the first mode:

[0043] Step A1: The centralized management platform uses a neural network algorithm to optimize the reactive power demand of each node in the network, calculates the reactive power compensation demand of each node, and obtains the corresponding number of FC switching groups by matching the remaining number of FC groups with the remaining capacity of SVG, and issues switching instructions for FC groups to each node.

[0044] Step A2: The FC automatic switching field controller is installed in the monitoring room or control room of the substation. When it receives the FC switching command issued by the centralized management platform under intelligent mode 1, it sends the switching signal to the FC automatic switching device according to the switching method of time gradient, or coordinates the SVG to perform dynamic compensation in real time according to the remaining reactive power demand after FC switching.

[0045] In step A3, the contact disconnect switch in the FC field electrical cabinet controls the switching of the contactor by switching the contactor according to the switching command.

[0046] Preferably, in the second mode:

[0047] Step B1: In-station SVG and FC coordinated control, the field automatic controller receives the power factor setting information transmitted by the in-station SVG and calculates the switching timing of FC through the built-in time gradient algorithm.

[0048] Step B2: The field automatic controller sends the control output to the contactor in the field electrical cabinet based on the calculated timing sequence.

[0049] Step B3: The contactor receives the instruction and performs an on / off operation to control the automatic switching of the FC.

[0050] In the third mode:

[0051] Step C1: When the on-site automatic controller does not detect the station SVG and the network-wide coordination command, it automatically enters the self-adjustment mode.

[0052] Step C2: When the field automatic controller is connected to the original FC automatic device, read the status parameters of the field automatic controller, including the FC status and bus voltage and current parameters, and calculate the switching sequence of the capacitor through the built-in reactive power compensation algorithm.

[0053] Step C3: When the original FC does not have an automatic switching system, the field automatic controller is connected to the bus CT in the station. The capacitor switching sequence is calculated by the built-in reactive power compensation algorithm based on the detected voltage and current.

[0054] Step C4: The field automatic controller controls the on / off of the contactors in the field electrical cabinet connected to the capacitors by calculating the switching sequence, thereby realizing automatic switching of the capacitors.

[0055] When the coordination and control system switches between the first mode, the second mode, and the third mode, the dot matrix mapping module automatically switches to the corresponding display mode.

[0056] Preferably, step A1 in the first mode specifically includes:

[0057] The centralized control center uses a neural network global optimization algorithm to calculate the reactive power flow of each node in the power grid. It allocates the reactive power compensation demand of each node based on the principles of minimizing network loss and maximizing reactive power compensation efficiency, i.e., minimizing the number of FC switching operations.

[0058] Using the reactive power demand of each node obtained by the global optimization algorithm, the central control center calculates the number of FC groups that need to be switched at each node by combining the corresponding FC switching information and the compensation capacity of the remaining SVG, and transmits the signal to the field automatic controller of the FC automatic switching device via Ethernet.

[0059] Preferably, the specific steps of A3 are as follows:

[0060] After receiving the switching command from the field automatic controller, the contact disconnecting switch combination device in the field electrical cabinet performs corresponding actions to connect or disconnect the internal coil, thereby executing the switching control of FC.

[0061] The specific steps of B1 are as follows: when no instruction is detected from the centralized management platform, it automatically connects to the SVG in the station for communication, forming a coordinated control mode between the SVG and FC in the station. The field automatic controller uses a time-segmentation algorithm to automatically control the FC.

[0062] Time gradient algorithm: A reactive power compensation curve is calculated using the target power factor, predicted active power data, and predicted reactive power data. The existing fixed reactive power switching capacity is divided, and the entry and exit points are determined based on whether the fixed FC capacity intersects with the reactive power curve to be compensated.

[0063] Preferably, the specific steps of C2 are as follows:

[0064] The reactive power compensation algorithm is as follows:

[0065]

[0066] in, The required compensation capacity; This represents the measured active power. This is the actual power factor; The target power factor.

[0067] The timing sequence for the switching is as follows:

[0068] when Then all FCs in the station will be deployed;

[0069] when If there are packets in the FC within the station, then the packets are deployed to satisfy the condition. and Approximate equality is acceptable; if the FC within the station is not grouped, when If the condition is met, do not allocate any FC (Functional Control Unit); otherwise, allocate all FCs.

[0070] in, The total FC capacity that can be put into operation within the station; the capacitor switching time must meet the requirement that the interval between two switching operations is greater than the capacitor discharge time.

[0071] Preferably, the grid accuracy configuration range in S1 is 1m×1m~5m×5m to adapt to different substation scales;

[0072] The attribute identifiers of each feature node are as follows: the attribute identifiers of bus nodes include node coordinates, voltage value, current value and the line number to which they belong; the attribute identifiers of FC nodes include node coordinates, FC switching status, number of remaining operable groups and capacity; the attribute identifiers of SVG nodes include node coordinates, real-time compensation capacity, response status and power factor setting value; the attribute identifiers of photovoltaic nodes include node coordinates, output power, connection time and voltage fluctuation value; and the attribute identifiers of topology nodes include node coordinates, connection line number and power flow direction.

[0073] Preferably, the period for timed updates in S4 can be adjusted according to the distribution network operation and maintenance needs; the specific implementation of the graphical auxiliary control in S5 is as follows: the historical operating data includes the number of switching operations and voltage fluctuation curves in the past 24 hours, and the display logic of the dynamic progress bar is that the progress corresponding to the number of FC switching groups increases from 0% to 100%, and the marking content when the node voltage exceeds the threshold includes the node number, overvoltage type, and the recommended FC / SVG devices and groups to be switched.

[0074] This invention provides a perceptual deep learning resource virtualization energy efficiency optimization system. It has the following beneficial effects:

[0075] By introducing an intelligent coordinated control system for SVG and FC in distribution networks, existing outdoor fixed capacitor banks (FC) in distribution networks can be quickly newly built or upgraded to form an intelligent coordinated control system with SVG systems. Through coordinated intelligent operation across the entire network and within substations, the overall performance and stability of the power grid can be significantly improved. This system can intelligently identify operating modes and dynamically adjust the switching operations of FCs through built-in algorithms, thereby achieving efficient collaborative work with SVG.

[0076] By optimizing the combined use of FC and SVG, the configuration capacity of SVG can be reduced, saving investment costs and maintenance expenses, thereby achieving a dual improvement in economic efficiency and technical performance. Ultimately, this will provide a strong guarantee for the safe, stable, and economical operation of modern power grids.

[0077] No manual drawing of distribution network diagrams is required; the dot matrix model automatically generates the topology. When adding or modifying distribution network equipment, only the corresponding feature nodes need to be added, shortening the graphic update time. Through node colors, labels, and dynamic arrows, the operating status of FC / SVG and the flow of control commands are intuitively displayed, allowing maintenance personnel to quickly grasp the distribution network status without having to check the backend data. The dot matrix graphic links historical data and real-time control strategies, assisting maintenance personnel in accurately judging FC switching needs, avoiding blind operation, and reducing the number of voltage fluctuations caused by improper control.

[0078] The dot matrix layout supports substations of different sizes, the grid accuracy can be flexibly configured, and it also supports integration with existing SCADA systems, reducing retrofit costs. Attached Figure Description

[0079] Figure 1 This is a flowchart of the method of the present invention;

[0080] Figure 2 This is a flowchart illustrating a specific implementation of the present invention. Detailed Implementation

[0081] 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.

[0082] like Figure 1 As shown, this invention proposes a perceptual deep learning resource virtualization energy efficiency optimization system, including a centralized control platform. The centralized control platform is responsible for the overall reactive power coordination and control optimization, and issues reactive power control commands to subordinate substations. It also integrates a deep learning perceptual optimization module, a resource virtualization module, and a dot matrix mapping module.

[0083] The deep learning perception optimization module is used to provide the centralized control platform with multi-source data processing, intelligent perception and global control strategy optimization.

[0084] The resource virtualization module is used to map the physical devices of the distribution network into virtual resource units to achieve global scheduling.

[0085] The field automatic controller has three operating modes: receiving SVG signals and commands from the centralized control center.

[0086] The field equipment layer includes the FC automatic switching electrical cabinet, which receives instructions from the field automatic controller and controls the on / off state of the internal contactors, thereby realizing the automatic switching of capacitors.

[0087] The centralized control platform also includes a dot matrix mapping module, the specific working steps of which are as follows:

[0088] Step S1: Construct the distribution network matrix model:

[0089] A two-dimensional grid is established based on the physical space of the distribution network. The coordinate system is the substation relative coordinates with the center point of the substation as the origin (0,0), the X-axis as the east-west direction and the Y-axis as the north-south direction, or the absolute coordinates of latitude and longitude.

[0090] The key elements of the distribution network are mapped to feature nodes in the dot matrix and assigned unique IDs and attribute identifiers. The key elements of the distribution network include distribution network bus, FC equipment, SVG equipment, distributed photovoltaic access points and line branch points. The corresponding feature node types are bus node, FC node, SVG node, photovoltaic node and topology node, respectively. The attribute identifier of each feature node contains control association data.

[0091] The distribution network topology is automatically identified and a matrix topology is generated by determining the adjacency relationship of the matrix nodes.

[0092] The adjacency relationship is determined as a direct connection when the grid distance between the bus node and the FC node is ≤2. The elements of the lattice topology matrix are node IDs, with a value of 1 indicating an adjacent connection and a value of 0 indicating no connection.

[0093] Step S2: Collect distribution network data uploaded by the field automatic controller in real time through the Ethernet interface. The distribution network data includes voltage, current, FC switching status and SVG compensation capacity.

[0094] Map the distribution network data to the corresponding matrix nodes according to the following rules:

[0095] Continuous data is displayed next to feature nodes in the form of node labels, discrete states are distinguished by node color and icon, and FC switching instructions issued by the central control platform are marked in the dot matrix as dynamic arrows with the arrows pointing from the central control node to the target FC node.

[0096] Step S3: Two graphics generation modes are built-in: topology simplification mode and panoramic detail mode. The topology simplification mode retains the bus nodes, FC nodes, SVG nodes and their connection relationships and deletes redundant grids to generate the control core topology map. The panoramic detail mode retains all feature nodes and overlays real-time node data annotations to generate the full distribution network status map.

[0097] Step S4: Set up two update triggering mechanisms: event-triggered update and timed-triggered update. The event-triggered update is when the field automatic controller detects a key state change and sends an update signal to the central control platform. After receiving the signal, the dot matrix mapping module updates the color, label or icon of the corresponding node. The key state changes include FC switching, SVG fault and voltage exceeding the threshold.

[0098] The timed trigger update is to automatically collect the running data of all nodes every 30 seconds by default and can be configured to update the data of nodes with no event changes.

[0099] Step S5: When maintenance personnel click on the target node in the dot matrix graphic, the system automatically pops up the historical operating data and current control strategy of the node. After issuing the FC switching command, the switching process is displayed through a dynamic progress bar, and the node color changes synchronously after the switching is completed. When the node voltage exceeds the threshold, the corresponding bus node in the dot matrix flashes yellow and automatically marks the FC / SVG device associated with the node.

[0100] As an optional embodiment: the dot matrix mapping module further includes intelligent layout and rendering of device primitives:

[0101] Step T1: Construct a dedicated element library for distribution network equipment. Associate the busbar segment elements corresponding to the double-horizontal-line structure of the core distribution network equipment busbar nodes with characteristic node types, and the capacitor elements with plate symbols corresponding to FC nodes. Simultaneously, add feeder segment elements, distribution transformer elements, combined switch elements, circuit breaker elements, tower elements, fault indicator elements, and voltage transformer elements. The element styles are designed based on the actual physical structure of the equipment, and the element size is positively correlated with the rated capacity of the equipment; the larger the rated capacity of the equipment, the larger the element size should be to ensure visual recognizability.

[0102] The color of the graphic element is determined by the voltage level in the attribute identifier of the feature node, realizing the accurate association between the device model and the graphic element, and matching the feature node attribute identifier to contain control association data.

[0103] Step T2: Based on the topological relationship between physical connection point I and physical connection point J in the device model, optimize the adjacency relationship recognition rules of the dot matrix topology matrix;

[0104] When two device nodes have a corresponding association between their physical connection point I and physical connection point J, they are determined to be directly connected, thus achieving topology self-adaptive layout, specifically including:

[0105] Single feeder scenario layout: Taking the actual power supply path of a single feeder as the core, the nodes are arranged in the following order: substation outgoing end, overhead line or cable, ring network cabinet, switch station, distribution station, box-type substation, cable branch box, distribution transformer, and line end switch. The towers are retained and the nodes are set according to the actual segment spacing of the line. The fault indicator is retained and the nodes are set at the branch positions of the line.

[0106] Station room layout: The physical space boundary of the independent station building is used to divide the grid area. The nodes are arranged in the order of station busbar, circuit breaker, load switch, voltage transformer, station transformer and bay outgoing terminal. Each node is marked with a unique number for physical connection point I and physical connection point J.

[0107] Connection scenario layout: Integrate two or more feeder nodes with electrical connections into the same lattice space, set exclusive marks at the nodes where the connection switches are located and distinguish them by thick borders, mark the boundaries of different feeders with dashed lines, and adjust the style of the dashed lines according to the topology differentiation requirements, and retain the main trunk equipment and branch key equipment of all feeders.

[0108] Step T3: Use a three-layer rendering logic: The bottom layer renders the device connection lines according to the scene type, using solid lines for single feeder scenes and dashed lines for feeder boundaries in interconnected scenes. The line width is set according to visual clarity requirements, and virtual resource scheduling channel identifiers are overlaid; the middle layer loads the dedicated primitives constructed in step T1, with the center point of the primitive precisely aligned with the node coordinates of the dot matrix model; the top layer labels the device name and device abbreviation, and also labels the physical connection point I and physical connection point J numbers and the core attributes of the virtual resource unit.

[0109] Feeder segment elements (straight structures with routing indicators), distribution transformer elements (double-winding symbol structures), combination switch elements (rectangular structures with on / off indicators), circuit breaker elements (icon structures with tripping mechanisms), pole / tower elements (symbol structures with insulators), fault indicator elements (triangular warning structures), voltage transformer elements (circular coil structures), main line equipment (substation outgoing lines, main line circuit breakers), and branch line key equipment (branch distribution transformers, branch fault indicators).

[0110] As an optional embodiment: the dot matrix mapping module further includes:

[0111] Equipment model data classification and linkage: Connect to the list of power distribution network equipment models to enable filtering and displaying matrix nodes by equipment type. When a type of equipment is selected, only the matrix nodes corresponding to that type of equipment are displayed. Clicking on any node allows you to view the core attributes of the equipment. The attribute data and the attribute identifiers of the feature nodes are synchronized in real time.

[0112] It supports exporting device lists. The list of disconnectors includes disconnector name, voltage level, physical connection point I, physical connection point J, and installation location. The list of distribution transformers includes transformer name, rated capacity, voltage level, physical connection point I, and physical connection point J. The list data is dynamically matched with the attributes of the dot matrix nodes.

[0113] Single feeder interaction: Supports filtering and displaying dispatch-managed equipment, retaining only the matrix nodes corresponding to substation outgoing lines, circuit breakers, load switches, distribution transformers, and fault indicators, and hiding user-side equipment nodes not under dispatch management; when clicking on a tower node, the equipment association relationship of the line segments on both sides of the tower is automatically displayed, and the association relationship is generated based on the matrix topology matrix;

[0114] Station Interaction: Supports switching between stations. Stations are divided into zones according to the wiring functions within the station. After switching, the dot matrix automatically focuses on the equipment nodes of the corresponding zone and uses a light-colored background to distinguish the zone range. At the same time, the wiring logic within the zone is marked. The wiring logic is generated based on the association between physical connection point I and physical connection point J.

[0115] Interactive communication: Supports viewing feeder associations. When a communication switch node is selected, all device nodes of the two feeders associated with that switch are automatically highlighted. The highlighting logic is based on the adjacency relationship of the topology matrix. At the same time, a communication relationship table is generated, which contains the fields of feeder number, communication switch name, associated device type, physical connection point I, and physical connection point J.

[0116] Model data synchronization and export: When the graphics are updated, the changes in the equipment model data are automatically synchronized, including adjustments to the equipment voltage level and modifications to the association between physical connection point I and physical connection point J;

[0117] When a fault signal is detected, the corresponding fault indicator node in the dot matrix starts flashing mode, and the flashing frequency is set according to the fault warning requirements; at the same time, based on the dot matrix topology matrix, the upstream and downstream device nodes associated with the fault indicator are automatically traced to generate a fault impact range map.

[0118] The fault impact range diagram marks the physical connection points I and J that need to be investigated, providing precise guidance for operation and maintenance.

[0119] As an optional embodiment: the centralized control platform specifically performs coordinated control of SVG and FC across the entire domain based on the reactive power control requirements, and calculates the specific reactive power target parameters of the subordinate stations and transmits them to the field automatic controller.

[0120] The field automatic controller intelligently identifies the device's operating mode through port identification. When the network-wide coordinated control command is detected, the field automatic controller automatically enters the first mode and receives instructions from the centralized control platform to automatically switch FC.

[0121] When only the SVG signal within the station is detected, the system enters the second mode and performs automatic FC switching using the built-in time gradient algorithm.

[0122] When the above-mentioned network-wide and station-specific commands cannot be identified, the field automatic controller automatically enters the third mode, and detects voltage and current through communication with the original FC automatic switching device or the station-specific measurement cabinet, and automatically performs reactive power compensation.

[0123] In the first mode:

[0124] Step A1: The centralized management platform uses a neural network algorithm to optimize the reactive power demand of each node in the network, calculates the reactive power compensation demand of each node, and obtains the corresponding number of FC switching groups by matching the remaining number of FC groups with the remaining capacity of SVG, and issues switching instructions for FC groups to each node.

[0125] Step A2: The FC automatic switching field controller is installed in the monitoring room or control room of the substation. When it receives the FC switching command issued by the centralized management platform under intelligent mode 1, it sends the switching signal to the FC automatic switching device according to the switching method of time gradient, or coordinates the SVG to perform dynamic compensation in real time according to the remaining reactive power demand after FC switching.

[0126] In step A3, the contact disconnect switch in the FC field electrical cabinet realizes the on / off switching of the contactor according to the switching command, thereby controlling the switching of the FC and using fast SVG to compensate for the remaining reactive power.

[0127] As an optional embodiment: in the second mode:

[0128] Step B1: In-station SVG and FC coordinated control, the field automatic controller receives the power factor setting information transmitted by the in-station SVG and calculates the switching timing of FC through the built-in time gradient algorithm.

[0129] Step B2: The field automatic controller sends the control output to the contactor in the field electrical cabinet based on the calculated timing sequence.

[0130] Step B3: The contactor receives the instruction and performs an on / off operation to control the automatic switching of the FC.

[0131] In the third mode:

[0132] Step C1: When the on-site automatic controller does not detect the station SVG and the network-wide coordination command, it automatically enters the self-adjustment mode.

[0133] Step C2: When the field automatic controller is connected to the original FC automatic device, read the status parameters of the field automatic controller, including the FC status and bus voltage and current parameters, and calculate the switching sequence of the capacitor through the built-in reactive power compensation algorithm.

[0134] Step C3: When the original FC does not have an automatic switching system, the field automatic controller is connected to the bus CT in the station. The capacitor switching sequence is calculated by the built-in reactive power compensation algorithm based on the detected voltage and current.

[0135] Step C4: The field automatic controller controls the on / off of the contactors in the field electrical cabinet connected to the capacitors by calculating the switching sequence, thereby realizing automatic switching of the capacitors.

[0136] When the coordination and control system switches between the first mode, the second mode, and the third mode, the dot matrix mapping module automatically switches to the corresponding display mode.

[0137] As an optional embodiment: In the first mode, step A1 specifically includes:

[0138] The centralized control center uses a neural network global optimization algorithm to calculate the reactive power flow of each node in the power grid. It allocates the reactive power compensation demand of each node based on the principles of minimizing network loss and maximizing reactive power compensation efficiency, i.e., minimizing the number of FC switching operations.

[0139] Using the reactive power demand of each node obtained by the global optimization algorithm, the central control center calculates the number of FC groups that need to be switched at each node by combining the corresponding FC switching information and the compensation capacity of the remaining SVG, and transmits the signal to the field automatic controller of the FC automatic switching device via Ethernet.

[0140] Specifically, neural network modeling and input feature definition:

[0141] Suppose the power grid contains N nodes, and the input feature vector... Including: node voltage amplitude Active power at nodes Node reactive power Node FC can be switched capacity .

[0142] The output layer is designed with two branches:

[0143] Reactive power demand allocation vector:

[0144]

[0145] This represents the reactive power that each node needs to compensate for. The reactive power to be compensated for at each node, where the value of i ranges from 1, 2, ..., N;

[0146] FC (Functionally Connected) Throwing Decision Vector:

[0147]

[0148] For nodes The number of FC throwing groups.

[0149] The multi-objective loss function is:

[0150] loss function Minimize overall network loss and maximize reactive power compensation efficiency:

[0151] (3)

[0152] in, , , These are the percentage coefficients for each item. For network loss items, For efficiency, This is a constraint term. Network loss term. The active power loss of the entire network is calculated based on the power flow equation, using the following formula:

[0153]

[0154] For line conductance, The term represents the phase angle difference of the node voltage, indicated by the subscript. Indicates the current node, Indicates and Other connected nodes.

[0155] Efficiency Item To maximize the compensation synergy between SVG and FC, the formula is as follows:

[0156]

[0157] in, For the first The capacity of SVG compensation at each node. For the first The capacity of FC compensation at each node.

[0158] Constraints The penalty for violating physical constraints is calculated using the following formula:

[0159]

[0160] in, This is the upper limit of voltage. This represents the maximum allowable compensation capacity for a node. The first penalty coefficient, This is the second penalty coefficient.

[0161] Gradient backpropagation and policy optimization:

[0162] Update network parameters using deterministic policy gradient (DPG) :

[0163]

[0164] in, This indicates the calculation of the expected value. For loss function, Indicates the parameters of the neural network Find the gradient. For the loss function on the neural network parameters The gradient guides the direction of parameter updates, and the same applies to the rest.

[0165] To handle discrete switching decisions Introducing Gumbel-Softmax reparameterization:

[0166]

[0167] in, As an activation function, it normalizes a numerical vector into a probability distribution vector, with the sum of all probabilities being 1; Used to determine the category with the highest probability; For hidden layer features, node state information is encoded; The temperature coefficient controls the degree of discretization. For Gumbel noise, randomness is introduced to explore the optimal solution. The weight matrix of the fully connected layer is used to linearly map the "hidden layer features" to the "output dimension required for discrete decision-making" (for example, if there are N discrete actions, they are mapped to an N-dimensional vector).

[0168] Physical constraint embedding and coupling with power flow equations:

[0169] In the forward propagation of the neural network, the power flow equation is embedded through implicit layers:

[0170]

[0171] in, This is the line susceptance.

[0172] By integrating the above equations as differentiable operators into the computational graph, the physical laws impose hard constraints on the network output, ensuring the reactive power distribution of the algorithm output. And the decision to cut Strictly meet the constraints of power grid operation.

[0173] Algorithm training process:

[0174] Offline pre-training, based on historical data ( Minimize prediction error:

[0175]

[0176] in, It is the amount of reactive power that each node actually needs to compensate for in historical data. , Calculation for the 2-norm. This formula refers to the calculation of neural network parameters. Optimize to make the loss function minimize.

[0177] Online reinforcement learning, interactive optimization in a simulation environment, through a reward function Update the policy network.

[0178] Following the above process, the centralized management platform utilizes a neural network to process the parameters input to each node of the power grid (node ​​voltage amplitude). Active power at nodes Node reactive power Node FC can be switched capacity Through continuous iterative optimization training, the reactive power demand allocation vector of the power grid is finally obtained. and FC switching decision vector This includes the reactive power demand allocation of each node and the signal instructions for the number of FC switching groups.

[0179] As an optional embodiment: the specific steps of A3 are as follows:

[0180] After receiving the switching command from the field automatic controller, the contact disconnecting switch combination device in the field electrical cabinet performs corresponding actions to connect or disconnect the internal coil, thereby executing the switching control of FC.

[0181] The specific steps of B1 are as follows: when no instruction is detected from the centralized management platform, it automatically connects to the SVG in the station for communication, forming a coordinated control mode between the SVG and FC in the station. The field automatic controller uses a time-segmentation algorithm to automatically control the FC.

[0182] Time gradient algorithm: A reactive power compensation curve is calculated using the target power factor, predicted active power data, and predicted reactive power data. The existing fixed reactive power switching capacity is divided, and the entry and exit points are determined based on whether the fixed FC capacity intersects with the reactive power curve to be compensated.

[0183] As an optional embodiment: the specific steps of C2 are as follows:

[0184] The reactive power compensation algorithm is as follows:

[0185]

[0186] in, The required compensation capacity; This represents the measured active power. This is the actual power factor; The target power factor.

[0187] The timing sequence for the switching is as follows:

[0188] when Then all FCs in the station will be deployed;

[0189] when If there are packets in the FC within the station, then the packets are deployed to satisfy the condition. and Approximate equality is acceptable; if the FC within the station is not grouped, when If the condition is met, do not allocate any FC (Functional Control Unit); otherwise, allocate all FCs.

[0190] in, The total FC capacity that can be put into operation within the station; the capacitor switching time must meet the requirement that the interval between two switching operations is greater than the capacitor discharge time.

[0191] As an optional embodiment: the grid accuracy configuration range in S1 is 1m×1m~5m×5m to adapt to different substation sizes;

[0192] The attribute identifiers of each feature node are as follows: the attribute identifiers of bus nodes include node coordinates, voltage value, current value and the line number to which they belong; the attribute identifiers of FC nodes include node coordinates, FC switching status, number of remaining operable groups and capacity; the attribute identifiers of SVG nodes include node coordinates, real-time compensation capacity, response status and power factor setting value; the attribute identifiers of photovoltaic nodes include node coordinates, output power, connection time and voltage fluctuation value; and the attribute identifiers of topology nodes include node coordinates, connection line number and power flow direction.

[0193] As an optional embodiment: the period of timed triggering update in S4 can be adjusted according to the distribution network operation and maintenance needs; the specific implementation of the graphical auxiliary control in S5 is as follows: the historical operation data includes the number of switching and voltage fluctuation curves in the past 24 hours, the display logic of the dynamic progress bar is that the progress corresponding to the number of FC switching groups increases from 0% to 100%, and the marking content when the node voltage exceeds the threshold includes the node number, overvoltage type and the recommended FC / SVG device and group to be switched.

[0194] Working principle:

[0195] The system consists of three layers: the first layer is the centralized control platform, responsible for the overall reactive power coordination and optimization, and issuing reactive power control commands to subordinate substations; the second layer is the field automatic controller, which receives SVG signals and commands from the centralized control center, and automatically controls the existing FC equipment in the distribution network, realizing global coordinated control of SVG and FC across the entire network and local coordinated control of SVG and FC within substations; the third layer is the field equipment layer, including the FC automatic switching electrical cabinet, which receives commands from the field automatic controller and controls the on / off switching of internal contactors, thereby realizing automatic switching of capacitors. The entire system features modular devices and standardized interfaces, enabling quick and convenient coordinated connection between SVG and FC, and effectively achieving coordinated intelligent control of SVG and FC.

[0196] This centralized coordination and control platform, based on the overall reactive power control requirements, uses a neural network optimization algorithm to perform coordinated control of SVG and FC across the entire region. It calculates the specific reactive power target parameters for subordinate stations and transmits them to the field automatic controllers. The field automatic controllers intelligently identify the device's operating mode through port recognition. When a network-wide coordinated control command is detected, the field automatic controller automatically enters Intelligent Mode 1—Network-wide Coordinated Control Mode—to receive commands from the centralized control platform and automatically switch FC. When only an intra-station SVG signal is detected, it enters Intelligent Mode 2—Intra-station SVG and FC Coordinated Control Mode—to automatically switch FC using a built-in time gradient algorithm. When neither the network-wide nor intra-station commands are detected, the field automatic controller automatically enters Intelligent Mode 3—Self-adjustment Mode—to detect voltage and current and automatically perform reactive power compensation through communication with the original FC automatic switching device or the intra-station measurement cabinet.

[0197] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A perceptual deep learning resource virtualization energy efficiency optimization system, characterized in that, It includes a centralized control platform, which is responsible for the overall reactive power coordination and control optimization, and issues reactive power control commands to subordinate substations. It also integrates a deep learning perception optimization module, a resource virtualization module, and a dot matrix mapping module. The deep learning perception optimization module is used to provide the centralized control platform with multi-source data processing, intelligent perception and global control strategy optimization. The resource virtualization module is used to map the physical devices of the distribution network into virtual resource units to achieve global scheduling. The field automatic controller has three operating modes: receiving SVG signals and commands from the centralized control center. The field equipment layer includes the FC automatic switching electrical cabinet, which receives instructions from the field automatic controller and controls the on / off state of the internal contactors, thereby realizing the automatic switching of capacitors. The specific working steps of the dot matrix mapping module are as follows: Step S1: Construct the distribution network matrix model: A two-dimensional grid is established based on the physical space of the distribution network. The coordinate system is the substation relative coordinates with the center point of the substation as the origin (0,0), the X-axis as the east-west direction and the Y-axis as the north-south direction, or the absolute coordinates of latitude and longitude. The key elements of the distribution network are mapped to feature nodes in the dot matrix and assigned unique IDs and attribute identifiers. The key elements of the distribution network include distribution network bus, FC equipment, SVG equipment, distributed photovoltaic access points and line branch points. The corresponding feature node types are bus node, FC node, SVG node, photovoltaic node and topology node, respectively. The attribute identifier of each feature node contains control association data. The distribution network topology is automatically identified and a matrix topology is generated by determining the adjacency relationship of the matrix nodes. Step S2: Collect distribution network data uploaded by the field automatic controller in real time through the Ethernet interface. The distribution network data includes voltage, current, FC switching status and SVG compensation capacity. Map the distribution network data to the corresponding matrix nodes according to the following rules: Continuous data is displayed next to feature nodes in the form of node labels, and FC switching commands issued by the central control platform are marked in the dot matrix as dynamic arrows with the arrows pointing from the central control node to the target FC node. Step S3: Two graphics generation modes are built-in: topology simplification mode and panoramic detail mode. The topology simplification mode retains the bus nodes, FC nodes, SVG nodes and their connection relationships and deletes redundant grids to generate the control core topology map. The panoramic detail mode retains all feature nodes and overlays real-time node data annotations to generate the full distribution network status map. Step S4: Set up two update triggering mechanisms: event-triggered update and timed-triggered update. The event-triggered update is when the field automatic controller detects a key state change and sends an update signal to the central control platform. After receiving the signal, the dot matrix mapping module updates the color, label or icon of the corresponding node. The key state changes include FC switching, SVG fault and voltage exceeding the threshold. Step S5: When maintenance personnel click on the target node in the dot matrix graphic, the system automatically pops up the historical operating data and current control strategy of the node. After issuing the FC switching command, the switching process is displayed through a dynamic progress bar, and the node color changes synchronously after the switching is completed. When the node voltage exceeds the threshold, the corresponding bus node in the dot matrix flashes yellow and automatically marks the FC / SVG device associated with the node.

2. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 1, characterized in that, The dot matrix mapping module also includes intelligent layout and rendering of device primitives: Step T1: Construct a dedicated graphic element library for distribution network equipment, associate the bus segment graphic elements corresponding to the double horizontal structure of the core equipment bus nodes of the distribution network according to the feature node type, and the capacitor graphic elements with plate symbols corresponding to the FC nodes. At the same time, add feeder segment graphic elements, distribution transformer graphic elements, combination switch graphic elements, circuit breaker graphic elements, pole tower graphic elements, fault indicator graphic elements, and voltage transformer graphic elements. The graphic element style is designed based on the actual physical structure of the equipment. The graphic element size is positively correlated with the rated capacity of the equipment. The larger the rated capacity of the equipment, the larger the graphic element size will be to ensure visual recognition. Step T2: Based on the topological relationship between physical connection point I and physical connection point J in the device model, optimize the adjacency relationship recognition rules of the dot matrix topology matrix; When two device nodes have a corresponding association between their physical connection point I and physical connection point J, they are determined to be directly connected, thus achieving topology self-adaptive layout, specifically including: Single feeder scenario layout: Taking the actual power supply path of a single feeder as the core, the nodes are arranged in the following order: substation outgoing end, overhead line or cable, ring network cabinet, switch station, distribution station, box-type substation, cable branch box, distribution transformer, and line end switch. The towers are retained and the nodes are set according to the actual segment spacing of the line. The fault indicator is retained and the nodes are set at the branch positions of the line. Station room layout: The physical space boundary of the independent station building is used to divide the grid area. The nodes are arranged in the order of station busbar, circuit breaker, load switch, voltage transformer, station transformer and bay outgoing terminal. Each node is marked with a unique number for physical connection point I and physical connection point J. Connection scenario layout: Integrate two or more feeder nodes with electrical connections into the same lattice space, set exclusive marks at the nodes where the connection switches are located and distinguish them by thick borders, mark the boundaries of different feeders with dashed lines, and adjust the style of the dashed lines according to the topology differentiation requirements, and retain the main trunk equipment and branch key equipment of all feeders. Step T3: Use a three-layer rendering logic: The bottom layer renders the device connection lines according to the scene type, using solid lines for single feeder scenes and dashed lines for feeder boundaries in interconnected scenes. The line width is set according to visual clarity requirements, and virtual resource scheduling channel identifiers are overlaid; the middle layer loads the dedicated primitives constructed in step T1, with the center point of the primitive precisely aligned with the node coordinates of the dot matrix model; the top layer labels the device name and device abbreviation, and also labels the physical connection point I and physical connection point J numbers and the core attributes of the virtual resource unit.

3. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 1, characterized in that: The dot matrix mapping module also includes: Equipment model data classification and linkage: Connect to the list of power distribution network equipment models to enable filtering and displaying matrix nodes by equipment type. When a type of equipment is selected, only the matrix nodes corresponding to that type of equipment are displayed. Clicking on any node allows you to view the core attributes of the equipment. The attribute data and the attribute identifiers of the feature nodes are synchronized in real time. It supports exporting device lists. The list of disconnectors includes disconnector name, voltage level, physical connection point I, physical connection point J, and installation location. The list of distribution transformers includes transformer name, rated capacity, voltage level, physical connection point I, and physical connection point J. The list data is dynamically matched with the attributes of the dot matrix nodes. Single feeder interaction: Supports filtering and displaying dispatch-managed equipment, retaining only the matrix nodes corresponding to substation outgoing lines, circuit breakers, load switches, distribution transformers, and fault indicators, and hiding user-side equipment nodes not under dispatch management; when clicking on a tower node, the equipment association relationship of the line segments on both sides of the tower is automatically displayed, and the association relationship is generated based on the matrix topology matrix; Station Interaction: Supports switching between stations. Stations are divided into zones according to the wiring functions within the station. After switching, the dot matrix automatically focuses on the equipment nodes of the corresponding zone and uses a light-colored background to distinguish the zone range. At the same time, the wiring logic within the zone is marked. The wiring logic is generated based on the association between physical connection point I and physical connection point J. Interactive communication: Supports viewing feeder associations. When a communication switch node is selected, all device nodes of the two feeders associated with the switch are automatically highlighted. The highlighting logic is based on the adjacency relationship of the topology matrix. At the same time, a communication relationship table is generated, which includes the fields of feeder number, communication switch name, associated device type, physical connection point I, and physical connection point J. Model data synchronization and export: When the graphics are updated, the changes in the equipment model data are automatically synchronized, including adjustments to the equipment voltage level and modifications to the association between physical connection point I and physical connection point J; When a fault signal is detected, the corresponding fault indicator node in the dot matrix starts flashing mode, and the flashing frequency is set according to the fault warning requirements; at the same time, based on the dot matrix topology matrix, the upstream and downstream device nodes associated with the fault indicator are automatically traced to generate a fault impact range map. The diagram showing the scope of the fault indicates the physical connection points I and J that need to be investigated.

4. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 1, characterized in that: The centralized control platform specifically calculates the specific reactive power target parameters of subordinate stations and transmits them to the field automatic controller based on the overall reactive power control requirements through a neural network optimization algorithm. The field automatic controller intelligently identifies the device's operating mode through port identification. When the network-wide coordinated control command is detected, the field automatic controller automatically enters the first mode and receives instructions from the centralized control platform to automatically switch FC. When only the SVG signal within the station is detected, the system enters the second mode and performs automatic FC switching using the built-in time gradient algorithm. When the network-wide command and station-specific command cannot be identified, the field automatic controller automatically enters the third mode. It detects voltage and current and automatically performs reactive power compensation by communicating with the original FC automatic switching device or the station-specific measurement cabinet. In the first mode: Step A1: The centralized management platform uses a neural network algorithm to optimize the reactive power demand of each node in the network, calculates the reactive power compensation demand of each node, and obtains the corresponding number of FC switching groups by matching the remaining number of FC groups with the remaining capacity of SVG, and issues switching instructions for FC groups to each node. Step A2: The FC automatic switching field controller is installed in the monitoring room or control room of the substation. When it receives the FC switching command issued by the centralized management platform under intelligent mode 1, it sends the switching signal to the FC automatic switching device according to the switching method of time gradient, or coordinates the SVG to perform dynamic compensation in real time according to the remaining reactive power demand after FC switching. In step A3, the contact disconnect switch in the FC field electrical cabinet controls the switching of the contactor by switching the contactor according to the switching command.

5. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 4, characterized in that: In the second mode: Step B1: In-station SVG and FC coordinated control, the field automatic controller receives the power factor setting information transmitted by the in-station SVG and calculates the switching timing of FC through the built-in time gradient algorithm. Step B2: The field automatic controller sends the control output to the contactor in the field electrical cabinet based on the calculated timing sequence. Step B3: The contactor receives the instruction and performs an on / off operation to control the automatic switching of the FC. In the third mode: Step C1: When the on-site automatic controller does not detect the station SVG and the network-wide coordination command, it automatically enters the self-adjustment mode. Step C2: When the field automatic controller is connected to the original FC automatic device, read the status parameters of the field automatic controller, including the FC status and bus voltage and current parameters, and calculate the switching sequence of the capacitor through the built-in reactive power compensation algorithm. Step C3: When the original FC does not have an automatic switching system, the field automatic controller is connected to the bus CT in the station. The capacitor switching sequence is calculated by the built-in reactive power compensation algorithm based on the detected voltage and current. Step C4: The field automatic controller controls the on / off of the contactors in the field electrical cabinet connected to the capacitors by calculating the switching sequence, thereby realizing automatic switching of the capacitors. When the coordination and control system switches between the first mode, the second mode, and the third mode, the dot matrix mapping module automatically switches to the corresponding display mode.

6. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 4, characterized in that: In the first mode, step A1 specifically involves: The centralized control center uses a neural network global optimization algorithm to calculate the reactive power flow of each node in the power grid. It allocates the reactive power compensation demand of each node based on the principles of minimizing network loss and maximizing reactive power compensation efficiency, i.e., minimizing the number of FC switching operations. Using the reactive power demand of each node obtained by the global optimization algorithm, the central control center calculates the number of FC groups that need to be switched at each node by combining the corresponding FC switching information and the compensation capacity of the remaining SVG, and transmits the signal to the field automatic controller of the FC automatic switching device via Ethernet.

7. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 5, characterized in that: The specific steps for A3 are as follows: After receiving the switching command from the field automatic controller, the contact disconnecting switch combination device in the field electrical cabinet performs corresponding actions to connect or disconnect the internal coil, thereby executing the switching control of FC. The specific steps of B1 are as follows: when no instruction is detected from the centralized management platform, it automatically connects to the SVG in the station for communication, forming a coordinated control mode between the SVG and FC in the station. The field automatic controller uses a time-segmentation algorithm to automatically control the FC. Time gradient algorithm: The reactive power capacity compensation curve is calculated by combining the target power factor, the predicted active power data, and the predicted reactive power data. The existing fixed reactive power switching capacity is divided, and the entry and exit of the time gradient is formed according to whether the fixed FC capacity intersects with the reactive power curve to be compensated.

8. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 5, characterized in that: The specific steps of C2 are as follows: The reactive power compensation algorithm is as follows: in, The required compensation capacity; This represents the measured active power. This is the actual power factor; The target power factor; The timing sequence for the switching is as follows: when Then all FCs in the station will be deployed; when If there are packets in the FC within the station, then the packets are deployed to satisfy the condition. and Approximate equality is acceptable; if the FC within the station is not grouped, when If the condition is not met, do not allocate any FC (Final Fantasy); otherwise, allocate all FCs. in, The total FC capacity that can be put into operation within the station; the capacitor switching time must meet the requirement that the interval between two switching operations is greater than the capacitor discharge time.

9. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 1, characterized in that: The grid accuracy configuration range in S1 is 1m×1m~5m×5m to adapt to different substation sizes; The attribute identifiers of each feature node are as follows: the attribute identifiers of bus nodes include node coordinates, voltage value, current value and the line number to which they belong; the attribute identifiers of FC nodes include node coordinates, FC switching status, number of remaining operable groups and capacity; the attribute identifiers of SVG nodes include node coordinates, real-time compensation capacity, response status and power factor setting value; the attribute identifiers of photovoltaic nodes include node coordinates, output power, connection time and voltage fluctuation value; and the attribute identifiers of topology nodes include node coordinates, connection line number and power flow direction.

10. The perceptual deep learning resource virtualization energy efficiency optimization system according to claim 9, characterized in that: The period for timed updates in S4 can be adjusted according to the needs of distribution network operation and maintenance; the specific implementation of graphical auxiliary control in S5 is as follows: historical operation data includes the number of switching and voltage fluctuation curves in the past 24 hours, and the display logic of the dynamic progress bar is that the progress corresponding to the number of FC switching groups increases from 0% to 100%. When the node voltage exceeds the threshold, the marking content includes the node number, overvoltage type, and the number of FC / SVG devices and groups switched.

Images