A transformer area demand regulation method and system based on time sequence learning and model prediction

Abstract

The present application relates to a kind of based on timing learning and model prediction transformer area demand regulation method, system, comprising the following steps: S1, historical power consumption state data is collected and predicts the total peak event of electricity in future time period;S2, the total peak event of electricity is decomposed into multiple peak sub-events, the type and intensity of the peak sub-event are determined;S3, according to multiple peak sub-events, establish peak event combination chart;S4, according to the peak event combination chart of structure constraint check, the node and edge in the peak event combination chart are respectively controlled resource reachability matching;S5, the matching result of control resource reachability is written into the peak event combination chart as node attribute and edge constraint, forms resource enhanced peak event combination chart, based on the resource enhanced peak event combination chart, dynamically generate the combination of different control strategies and issue to control resource.

Description

Technical Field

[0001] This invention relates to the field of power dispatching, specifically to a method and system for demand control of distribution transformer areas based on time-series learning and model prediction. Background Technology

[0002] Under current electricity pricing policies, basic electricity charges for large industrial users can be calculated based on either transformer capacity or maximum demand. For production lines with large load fluctuations and short-term peak demand, capacity-based billing results in high unit prices, while demand-based billing incurs punitive charges for exceeding peak limits. To reduce billing costs, existing technologies generally employ optimized dispatching methods, coordinating energy storage discharge, photovoltaic power generation, and regional power dispatching during local short-term peak demand periods to achieve priority consumption of local energy and joint demand reduction.

[0003] In existing distribution area demand optimization scheduling methods, peak loads are treated as a single value or event and controlled using a uniform threshold triggering mechanism. This fails to distinguish the sub-event structures within peak loads, which have different causes, waveforms, durations, and resource adaptability. Consequently, the scheduling of resources simply involves filling gaps by activating energy storage discharge, which is mismatched with the characteristics of peak loads. This results in problems such as untimely response to transient impact peaks, insufficient cross-area transfer for continuous climbing peaks, and chaotic multi-resource coordination for complex peaks, making it impossible to achieve refined and differentiated dynamic scheduling decisions.

[0004] The purpose of this invention is to design a method and system for demand regulation of transformer substations based on time-series learning and model prediction, which addresses the problems existing in the prior art. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for demand regulation of transformer areas based on time series learning and model prediction, which can effectively solve at least one of the problems existing in the prior art.

[0006] The technical solution of this invention is: A method for demand regulation of transformer substations based on time-series learning and model prediction includes the following steps: S1 collects historical electricity consumption data and predicts total peak electricity consumption events in the future time period; S2, extract and analyze the duration, peak intensity, rise rate and over-limit energy area of ​​the total electricity consumption peak event, decompose the total electricity consumption peak event into multiple peak sub-events, determine the type and intensity of the peak sub-events, the type of the peak sub-events includes one or more of the following: transient impact type, continuous climbing type, periodic peak type and compound superposition type; S3, establish a peak event combination graph based on multiple peak sub-events, wherein the graph nodes of the peak event combination graph are peak sub-events, and the edges of the peak event combination graph are the structural relationships between peak sub-events, and perform structural constraint verification on the peak event combination graph; S4. Based on the peak event combination graph of the structural constraint verification, control resource reachability matching is performed on the nodes and edges of the peak event combination graph respectively. The control resources include one or more of the following: energy storage system, flexible interconnection module, distributed power source and power distribution switch. S5, the matching results of control resource reachability are written into the peak event combination graph as node attributes and edge constraints to form a resource-enhanced peak event combination graph. Based on the resource-enhanced peak event combination graph, differentiated control strategy combinations are dynamically generated and distributed to the control resources.

[0007] Further, step S1 includes: S1.1, Collect power consumption status data of distribution substations within a historical time window, as well as control resource operation status data related to flexible interconnection and microgrid coordinated scheduling; S1.2 Based on the electrical topology of the distribution substation, the status of the distribution switch and the connection relationship of the flexible interconnection module, establish the electrical association relationship between the distribution substation, the line, the microgrid and the control resources, and bind the line capacity margin, demand threshold, and control resource availability status with the corresponding power consumption status data to form a power consumption status feature sequence with topology association information; S1.3, Based on the power consumption status feature sequence, combined with ambient temperature, production line scheduling plan and holiday labels, input into a pre-trained long short-term memory network, and output the load prediction sequence and its prediction confidence for the future time period through the long short-term memory network; S1.4 When the average load power of the load prediction sequence within the sliding window is greater than a preset load threshold, the duration is greater than a preset duration threshold, and the prediction confidence is greater than a preset confidence threshold, it is determined that the total electricity consumption peak event exists.

[0008] Further, step S2 includes: S2.1, acquire historical peak event data, extract historical peak features based on the historical peak event data, and construct a peak superposition decomposition model based on the historical peak features; S2.2, Smooth the load prediction sequence corresponding to the total electricity consumption peak event, and identify the load change point. The load change point is one or more of the following: rising inflection point, peak point, falling inflection point, step point, and over-limit state switching point. S2.3, extract multiple load contribution segments based on the load change points and calculate the characteristic parameters of each load contribution segment respectively. Based on the characteristic parameters of the load contribution segments, decompose the total electricity consumption peak event into multiple peak sub-events that are superimposed within the same time window through the peak superposition decomposition model. S2.4, extract the sub-event contribution power of each spike sub-event as the intensity, and determine the type based on the sub-event characteristics of each spike sub-event.

[0009] Furthermore, the type includes: If the sub-event characteristics of the spike sub-event are continuous positive growth, long duration and slow rate of increase, then it is determined to be a continuous climbing type. If the sub-event characteristics of the spike sub-event are periodic repetitive fluctuations, and the similarity with the waveform of the same historical period is greater than or equal to a preset similarity threshold, then it is determined to be a periodic peak type. If the sub-event characteristics of the spike sub-event are high-amplitude short-duration pulses with a fast rise rate, then it is determined to be a transient impact type; If the sub-event characteristics of the peak sub-event are formed by the superposition of multiple load contribution components within the same time window, or if its waveform simultaneously has at least two of the characteristics of transient impact, continuous climbing, and periodic peak type, then it is determined to be a composite superposition type.

[0010] Further, step S3 includes: S3.1, taking each of the spike sub-events as a graph node, configure node attributes for each graph node based on the attributes of the spike sub-events; S3.2, Based on the association relationship and event type combination relationship among the multiple peak sub-events, calculate the structural association degree between any two graph nodes. When the structural association degree between two graph nodes meets the preset association condition, establish an edge between the two graph nodes and confirm the edge attribute to obtain the peak event combination graph. S3.3, perform one or more of the following checks on the peak event combination graph: node integrity check, edge validity check, time overlap consistency check, load contribution conservation check, topology association legality check, event type compatibility check, and graph structure connectivity check. When the peak event combination graph does not meet the check conditions, correct the graph nodes or edges that do not meet the check conditions.

[0011] Further, step S4 includes: S4.1, Based on each graph node and its node attributes in the peak event combination graph, determine the target control resource type that matches the graph node, query the real-time operating status of the control resource within the corresponding control time window, calculate the available capacity, response speed and continuous support capability of the control resource, and form a node-level control resource reachability matching result; S4.2 Based on each edge and its attributes in the peak event combination graph, and combined with the node-level control resource reachability matching results of the graph nodes connected by the edge, determine the resource coordination constraint type corresponding to the edge, and evaluate the coordination feasibility and capacity margin of the control resources associated with the edge when satisfying the constraints corresponding to the edge attributes, and form the edge-level control resource reachability matching results. S4.3, combine the node-level control resource reachability matching results and the edge-level control resource reachability matching results to form a reachability matching scheme.

[0012] Further, step S4.3 includes: The consistency of the node-level control resource reachability matching results and the edge-level control resource reachability matching results is verified. When the control resources required by the node-level control resource reachability matching results and the edge-level control resource reachability matching results conflict in terms of time window, capacity allocation, and response timing, the conflict is resolved according to the preset priority of each sub-event and the edge relationship strength coefficient, and a unified reachability matching scheme is output.

[0013] Further, step S5 includes: S5.1, the node-level control resource reachability matching result is written as a node attribute into the peak event combination graph, and the edge-level control resource reachability matching result is written as an edge attribute into the peak event combination graph to form a resource-enhanced peak event combination graph; S5.2, Traverse the graph nodes in the resource-enhanced peak event combination graph, generate corresponding single resource control instructions based on the event type of each node and its matching control resources, allocated power and execution time window, traverse the edges in the resource-enhanced peak event combination graph, and generate resource coordination control instructions based on the edge type, its carried coordination strategy type and remaining capacity margin. S5.3, combine the single resource control instruction and the resource coordination control instruction into a control strategy combination and decentralize it to the control resource.

[0014] Furthermore, a transformer area demand control system based on time-series learning and model prediction is provided to implement the aforementioned transformer area demand control method based on time-series learning and model prediction, comprising the following modules: The prediction module is used to collect historical electricity consumption data and predict total peak electricity consumption events in the future time period. The decomposition module is used to extract and analyze the duration, peak intensity, rise rate, and over-limit energy area of ​​the total electricity consumption peak event, decompose the total electricity consumption peak event into multiple peak sub-events, and determine the type and intensity of the peak sub-events. The types of peak sub-events include one or more of the following: transient impact type, continuous climbing type, periodic peak type, and compound superposition type. The peak event combination graph generation module is used to establish a peak event combination graph based on multiple peak sub-events, wherein the graph nodes of the peak event combination graph are peak sub-events, the edges of the peak event combination graph are the structural associations between peak sub-events, and the module performs structural constraint verification on the peak event combination graph. The control resource reachability matching module is used to perform control resource reachability matching on the nodes and edges in the peak event combination graph according to the structural constraint verification. The control resources include one or more of the following: energy storage system, flexible interconnection module, distributed power source and power distribution switch. The instruction generation module is used to write the matching results of control resource reachability as node attributes and edge constraints into the peak event combination graph to form a resource-enhanced peak event combination graph. Based on the resource-enhanced peak event combination graph, it dynamically generates differentiated control strategy combinations and sends them to the control resources.

[0015] Therefore, the present invention provides the following effects and / or advantages: This application breaks down the total peak event into peak sub-events with clear physical meanings and establishes a peak event combination diagram to express the complex relationships between peak sub-events. Then, based on the peak sub-events expressed in the combination diagram and the relationships between events, it performs coordinated control resource matching, realizing an automated closed loop from event identification to resource matching to strategy generation. It can match differentiated control resources for different sub-event types, improving the accuracy of demand control.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0017] It should be understood that the above summary and the following detailed description of the invention are exemplary and explanatory, and are intended to provide further explanation of the invention as claimed. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating one embodiment of the present invention. Detailed Implementation

[0019] To facilitate understanding by those skilled in the art, the present invention will now be described in further detail with reference to the embodiments: refer to Figure 1 A method for demand regulation of transformer substations based on time-series learning and model prediction includes the following steps: S1 collects historical electricity consumption data and predicts total peak electricity consumption events in the future time period; S2, extract and analyze the duration, peak intensity, rise rate and over-limit energy area of ​​the total electricity consumption peak event, decompose the total electricity consumption peak event into multiple peak sub-events, determine the type and intensity of the peak sub-events, the type of the peak sub-events includes one or more of the following: transient impact type, continuous climbing type, periodic peak type and compound superposition type; S3, establish a peak event combination graph based on multiple peak sub-events, where the nodes of the peak event combination graph are peak sub-events and the edges of the peak event combination graph are the structural relationships between peak sub-events, and perform structural constraint verification on the peak event combination graph; S4. Based on the peak event combination graph of the structural constraint verification, control resource reachability matching is performed on the nodes and edges of the peak event combination graph respectively. The control resources include one or more of the following: energy storage system, flexible interconnection module, distributed power source and power distribution switch. S5, the matching results of control resource reachability are written into the peak event combination graph as node attributes and edge constraints to form a resource-enhanced peak event combination graph. Based on the resource-enhanced peak event combination graph, differentiated control strategy combinations are dynamically generated and distributed to the control resources.

[0020] This embodiment breaks down the total peak event into peak sub-events with clear physical meanings and establishes a peak event combination diagram to express the complex relationships between peak sub-events, realizing an automated closed loop from event identification to resource matching to strategy generation; it can match differentiated control resources for different sub-event types and improve the accuracy of demand control.

[0021] Further, step S1 includes: S1.1, Collect power consumption status data of distribution substations within a historical time window, as well as control resource operation status data related to flexible interconnection and microgrid coordinated scheduling; In this embodiment, the power consumption status data includes one or more of the following: total active power of the transformer area, total reactive power of the transformer area, active power of each line, reactive power of each line, line voltage, line current, power factor, transformer load rate, historical maximum demand, demand assessment window, line rated capacity, and demand threshold. The control resource operation status data includes one or more of the following: energy storage system state of charge, energy storage system charging and discharging power, energy storage system remaining available capacity, flexible interconnect module port power, flexible interconnect module capacity margin, distributed power historical output, distributed power predicted output, controllable load operation status, distribution switch status, and microgrid grid connection / disconnection status. This step allows us to collect power consumption status data and control resource operation status data, which will serve as the basis for calculations in subsequent steps.

[0022] S1.2 Based on the electrical topology of the distribution substation, the status of the distribution switch and the connection relationship of the flexible interconnection module, establish the electrical association relationship between the distribution substation, the line, the microgrid and the control resources, and bind the line capacity margin, demand threshold, and control resource availability status with the corresponding power consumption status data to form a power consumption status feature sequence with topology association information; The purpose of this step is to simultaneously collect the operational status of flexible interconnect modules, energy storage systems, and distributed power sources. This allows the peak prediction phase to not only determine whether future limitations will be exceeded, but also to anticipate which control resources are available and how these resources can be scheduled. For example, it determines whether the flexible interconnect can transfer power, whether energy storage can still discharge, and whether photovoltaic power is currently generating output. This step transforms simple data collection into data that is understood within the correct electrical relationships, turning the data into data with topological relationships, so that subsequent prediction results can be directly correlated with the accessibility of control resources.

[0023] S1.3, Based on the power consumption status feature sequence, combined with ambient temperature, production line scheduling plan and holiday labels, input into a pre-trained long short-term memory network, and output the predicted load value and its prediction confidence level for the future time period through the long short-term memory network; The purpose of this step is to predict whether the load will continue to rise and form a peak in the future, based on past load changes, topology relationships, and external influencing factors. Long Short-Term Memory (LSTM) networks can train an LSTM model using historical electricity consumption data. The LSTM model receives feature sequences from the past period and outputs a load prediction sequence for the future period. Simultaneously, it predicts and outputs the corresponding peak probabilities as confidence levels based on similar historical scenarios.

[0024] S1.4 When the average load power of the load prediction value within the sliding window is greater than the preset load threshold, the duration is greater than the preset duration threshold, and the prediction confidence is greater than the preset confidence threshold, it is determined that the total electricity consumption peak event exists.

[0025] The purpose of this step is to convert the load forecast values ​​and their prediction confidence levels obtained in S1.3 into explicit event judgment results. By considering three conditions—the average predicted load power exceeding a threshold, the duration exceeding a threshold, and the prediction confidence level exceeding a threshold—a peak electricity consumption event is determined. However, since the LSTM model trained on historical electricity consumption data in step S1.3 can only output the overall waveform and cannot classify or evaluate individual sub-events, further steps are required.

[0026] Step S2 includes: S2.1, acquire historical peak event data, extract historical peak features based on the historical peak event data, and construct a peak superposition decomposition model based on the historical peak features; This step can be implemented as follows: Extract multiple historical total peak power event samples from the historical peak event data, and obtain the historical total load power time series, historical line load power time series, historical microgrid load power time series, historical production line scheduling status, historical equipment start-up and shutdown status, and historical demand threshold corresponding to each historical total peak power event sample. Perform baseline load separation on each of the historical total peak power event samples to obtain historical base load components and historical peak incremental load components. Based on the load change rate, duration, peak intensity, over-limit energy area, periodic fluctuation characteristics, equipment start-up and shutdown status, and production line scheduling status of the historical peak incremental load components, extract the historical peak contribution characteristics corresponding to transient impact type, continuous climbing type, periodic peak type, and composite superposition type, and calculate the historical load contribution weight corresponding to each component.

[0027] Then, based on the historical peak contribution characteristics and historical load contribution weights, a peak contribution component template library is established. The peak contribution component template library includes transient impact templates, continuous climbing templates, periodic peak templates, and composite superposition templates. A peak superposition decomposition model is constructed using the peak contribution component template library, the historical load contribution weights, and preset constraints. The peak superposition decomposition model is used to decompose the total electricity consumption peak event into multiple peak sub-events and their contribution degrees that exist superimposed within the same time window.

[0028] S2.2, Smooth the load prediction sequence corresponding to the total electricity consumption peak event, and identify the load change point. The load change point is one or more of the following: rising inflection point, peak point, falling inflection point, step point, and over-limit state switching point. In this embodiment, smoothing processing can perform noise reduction and smoothing on the load forecast sequence. First-order and second-order differences are calculated, and multiple load change points can be identified by setting thresholds. For example, a change in slope from negative to positive, or from a small positive value to a large positive value, is determined as an upward inflection point; a local maximum value where the slope changes from positive to negative is determined as a peak point; a local minimum value where the slope changes from negative to positive is determined as a downward inflection point; an absolute value of the slope exceeding a preset abrupt change threshold indicates a significant jump in power within a very short time, and is determined as a step point; the moment when the load power value crosses the demand threshold, including the entry point from below the threshold to above the threshold and the exit point from above the threshold to below the threshold, is determined as an over-limit state switching point. This step can divide the continuous total peak event time window into multiple morphologically characteristic sub-intervals.

[0029] S2.3, extract multiple load contribution segments based on the load change points and calculate the characteristic parameters of each load contribution segment respectively. Based on the characteristic parameters of the load contribution segments, decompose the total electricity consumption peak event into multiple peak sub-events that are superimposed within the same time window through the peak superposition decomposition model. In this step, the total peak load is considered to be a superposition of transient impact components, continuous ramp-up components, periodic peak components, and multi-source synchronous superposition components. These components can coexist at the same time. By extracting multiple load contribution segments, several segments that are relatively easy to decompose can be obtained. Then, the characteristic parameters of these segments are statistically analyzed. The load sequence of each load contribution segment is input into the peak superposition decomposition model constructed in S2.1. The peak superposition decomposition model outputs multiple superimposed peak sub-events within the segment. Since the decomposition is performed within a segment, these sub-events coexist and superimpose within the time window of that segment.

[0030] S2.4, extract the sub-event contribution power of each spike sub-event as the intensity, and determine the type based on the sub-event characteristics of each spike sub-event.

[0031] In this step, the total peak load is considered to be the sum of the products of transient impact components, continuous ramp components, periodic peak components, multi-source synchronous superposition components, and their corresponding intensities. After decomposing into multiple sub-events, the contribution ratio of different lines, microgrids, equipment, and load patterns to the total peak can be pre-learned through the peak superposition decomposition model. This serves as an identifier for the contribution of future peak decomposition, thereby determining its intensity. The final output is as follows: Total peak increment = 0.35 × transient impact component + 0.40 × continuous climbing component + 0.15 × periodic peak component + 0.10 × composite superposition component.

[0032] Furthermore, the type includes: If the sub-event characteristics of the spike sub-event are continuous positive growth, long duration and slow rate of increase, then it is determined to be a continuous climbing type. If the sub-event characteristics of the spike sub-event are periodic repetitive fluctuations, and the similarity with the waveform of the same historical period is greater than or equal to a preset similarity threshold, then it is determined to be a periodic peak type. If the sub-event characteristics of the spike sub-event are high-amplitude short-duration pulses with a fast rise rate, then it is determined to be a transient impact type; If the sub-event characteristics of the peak sub-event are formed by the superposition of multiple load contribution components within the same time window, or if its waveform simultaneously has at least two of the characteristics of transient impact, continuous climbing, and periodic peak type, then it is determined to be a composite superposition type.

[0033] This step outputs semantic event types, which are used to provide node attributes for the spike event combination graph in subsequent steps.

[0034] Further, step S3 includes: S3.1, taking each of the spike sub-events as a graph node, configure node attributes for each graph node based on the attributes of the spike sub-events; In this step, the attributes of the spike sub-event can be multiple or all of the following: event type, intensity level, contribution power, superposition weight, duration, start and end time, distribution area, line, microgrid, load source identifier, and prediction confidence. S3.2, Based on the association relationship and event type combination relationship among the multiple peak sub-events, calculate the structural association degree between any two graph nodes. When the structural association degree between two graph nodes meets the preset association condition, establish an edge between the two graph nodes and confirm the edge attribute to obtain the peak event combination graph. The relationships and event type combinations can be multiple or all of the following: time overlap, load contribution, electrical topology, load change synchronization, production line scheduling, and event type combination. Specifically, time overlap indicates that if two peak sub-events occur simultaneously within the same time window, they may overlap to form a total peak; load contribution indicates that the load change trends of two sub-events increase synchronously, or their combined power causes the total load to exceed the demand threshold, indicating a load contribution relationship; electrical topology indicates that two sub-events are located on lines connected by flexible interconnect modules, or on the same microgrid, the same distribution area, or the same bus, indicating a topological coupling relationship; load change synchronization indicates that the load increase, peak occurrence, or periodic fluctuation of two sub-events is synchronous, indicating they may be caused by the same production process or the same operating state; production line scheduling indicates that the load sources corresponding to two sub-events are related to the same production line scheduling plan, the same shift, or the same equipment group, indicating a business-side relationship; and event type combination indicates that two sub-events overlap and combine, potentially forming a compound peak.

[0035] Edge attributes can include temporal sequence edges, overlapping coexistence edges, causal relationship edges, resource competition edges, and distribution area propagation edges. Temporal sequence edges indicate that two peak sub-events occur sequentially in time; overlapping coexistence edges indicate that multiple peak sub-events exist simultaneously within the same time window; causal relationship edges indicate that one sub-event is likely to cause another sub-event to occur; resource competition edges indicate that multiple sub-events compete for the same scarce resource; and distribution area propagation edges indicate that sub-events transfer power between different distribution areas through flexible interconnection modules.

[0036] This step transforms multiple spike sub-events into computable, associative, and verifiable graph nodes, enabling the subsequent expression of the superposition relationship between multiple spike sub-events using a graph structure.

[0037] S3.3, perform one or more of the following checks on the peak event combination graph: node integrity check, edge validity check, time overlap consistency check, load contribution conservation check, topology association legality check, event type compatibility check, and graph structure connectivity check. When the peak event combination graph does not meet the check conditions, correct the graph nodes or edges that do not meet the check conditions.

[0038] This step involves performing structural constraint verification on the peak event combination graph to determine its rationality. For example, if two nodes do not overlap at all but are marked as concurrently superimposed, then that edge is unreasonable; or if a peak sub-event with a high contribution weight is completely isolated, it indicates that its superposition, topological, or synchronization relationships with other events may have been overlooked. When the peak event combination graph fails to meet the verification criteria, the graph nodes or edges that do not meet the verification conditions are corrected to ensure that the peak event combination conforms to time overlap relationships, load contribution relationships, electrical topology relationships, and event type combination relationships, providing a reliable graph structure foundation for subsequent control resource accessibility matching and differentiated control strategy generation.

[0039] Further, step S4 includes: S4.1, Based on each graph node and its node attributes in the peak event combination graph, determine the target control resource type that matches the graph node, query the real-time operating status of the control resource within the corresponding control time window, calculate the available capacity, response speed and continuous support capability of the control resource, and form a node-level control resource reachability matching result; In this step, preliminary matching is first performed based on the type of peak sub-events. For example, for transient impact events, which have extremely short durations, extremely high rise rates, and high power intensity, energy storage systems have millisecond-level response speeds and can release large power in a very short time, matching the instantaneous high power demand of impact events. Flexible interconnect modules, however, have slower response speeds and delays in responding to impact events, serving only as a supplement. For continuous ramp-up events, flexible interconnect modules can operate continuously for tens of minutes, smoothly transferring power from overloaded lines to lightly loaded lines, matching the long-term stability requirements of ramp-up events. Energy storage systems can serve as a supplement when flexible interconnect module capacity is insufficient or fine-tuning is required. For periodic peak events, flexible interconnect modules can pre-transfer part of the load to lightly loaded lines before the predicted peak window opens, achieving cross-cycle load balancing. Energy storage systems can serve as a supplement. For composite superimposed events, which contain multiple sub-event components superimposed and coexisting within the same time window, multiple resources, including energy storage systems, flexible interconnect modules, distributed power sources, and distribution switches, need to be coordinated and dynamically allocated.

[0040] Then, based on the real-time operating status of the control resources within the corresponding control time window, the available capacity, response speed, and continuous support capability are calculated, and an reachability vector containing indicators such as event type, target control resource type, available power, available energy, continuous support capability, response speed, and reachability capability is output as the node-level control resource reachability matching result.

[0041] S4.2 Based on each edge and its attributes in the peak event combination graph, and combined with the node-level control resource reachability matching results of the graph nodes connected by the edge, determine the resource coordination constraint type corresponding to the edge, and evaluate the coordination feasibility and capacity margin of the control resources associated with the edge when satisfying the constraints corresponding to the edge attributes, and form the edge-level control resource reachability matching results. In this step, time-sequential edges generally require the same type of control resources to respond sequentially, and the state of the control resources cannot change abruptly within the time interval. Therefore, energy storage systems and flexible interconnect modules are more suitable. Superimposed coexistence edges require multiple control resources to simultaneously meet the power and energy requirements of all sub-events. Therefore, energy storage systems, flexible interconnect modules, and power distribution switches can be selected. Causal relationship edges need to reserve a portion of available capacity in advance to prevent resources from being exhausted when the result event suddenly occurs. Therefore, energy storage systems and flexible interconnect modules are selected. Resource competition edges need to be allocated according to certain priority rules. The priority is determined based on factors such as the type of sub-event, the severity of exceeding limits, economic efficiency, and the strength of the edge relationship. All control resources can be used. Transformer area propagation edges depend on the currently occupied transmission power. Flexible interconnect modules and power distribution switches can be given priority.

[0042] Next, the reachability vector will be composed of coordination feasibility, capacity margin, and resource coordination constraint type corresponding to the edge, and will serve as the edge-level control resource reachability matching result.

[0043] S4.3, combine the node-level control resource reachability matching results and the edge-level control resource reachability matching results to form a reachability matching scheme.

[0044] Further, step S4.3 includes: The consistency of the node-level control resource reachability matching results and the edge-level control resource reachability matching results is verified. When the control resources required by the node-level control resource reachability matching results and the edge-level control resource reachability matching results conflict in terms of time window, capacity allocation, and response timing, the conflict is resolved according to the preset priority of each sub-event, and a unified reachability matching scheme is output.

[0045] This step requires checking the time windows, capacity allocation, and response timing conflicts of the two matching results. A time-axis-based resource occupancy graph method can be used, dividing each resource into consecutive time periods on the time axis and marking the allocated power, energy consumption, and reservation status within each time period. Then, for each time point, check whether the cumulative sum of all allocation demands exceeds the resource limit.

[0046] If conflicts exist, they can be sorted according to the priority of composite superposition type > transient impact type > continuous climbing type > periodic peak type, so that some of the demand of low priority nodes can be switched to alternative resources, or the power demand of low priority nodes can be reduced proportionally, or some demand can be delayed to non-overlapping periods, and the reservation ratio of causal edge can be reduced.

[0047] Further, step S5 includes: S5.1, the node-level control resource reachability matching result is written as a node attribute into the peak event combination graph, and the edge-level control resource reachability matching result is written as an edge attribute into the peak event combination graph to form a resource-enhanced peak event combination graph; In this step, the peak event combination graph is merged to form an enhanced graph structure that integrates resources and events. This graph not only records the structural relationships between sub-events, but also includes the resource allocation scheme corresponding to each sub-event and the resource coordination constraints required for each edge.

[0048] S5.2, Traverse the graph nodes in the resource-enhanced peak event combination graph, generate corresponding single resource control instructions based on the event type of each node and its matching control resources, allocated power and execution time window, traverse the edges in the resource-enhanced peak event combination graph, and generate resource coordination control instructions based on the edge type, its carried coordination strategy type and remaining capacity margin. In this step, based on the control resources matched to the event types of each node, combined with the allocated power and execution time window, the corresponding single-resource control instructions can be obtained. Then, resource coordination control instructions are generated according to the structural associations corresponding to different edges, specifically: For time-sequence edges, when energy storage continuously serves multiple sub-events, charging instructions are generated during the intervals to ensure that subsequent events have sufficient energy; For overlapping and coexisting edges, resource competition allocation instructions are generated based on the conflict resolution results; For causal edges, generate resource reservation instructions; For the propagation edge of the transformer area, generate flexible interconnection transmission path allocation instructions.

[0049] S5.3, combine the single resource control instruction and the resource coordination control instruction into a control strategy combination and decentralize it to the control resource.

[0050] A transformer area demand control system based on time-series learning and model prediction, used to implement the aforementioned transformer area demand control method based on time-series learning and model prediction, includes the following modules: The prediction module is used to collect historical electricity consumption data and predict total peak electricity consumption events in the future time period. The decomposition module is used to extract and analyze the duration, peak intensity, rise rate, and over-limit energy area of ​​the total electricity consumption peak event, decompose the total electricity consumption peak event into multiple peak sub-events, and determine the type and intensity of the peak sub-events. The types of peak sub-events include one or more of the following: transient impact type, continuous climbing type, periodic peak type, and compound superposition type. The peak event combination graph generation module is used to establish a peak event combination graph based on multiple peak sub-events, wherein the graph nodes of the peak event combination graph are peak sub-events, the edges of the peak event combination graph are the structural associations between peak sub-events, and the module performs structural constraint verification on the peak event combination graph. The control resource reachability matching module is used to perform control resource reachability matching on the nodes and edges in the peak event combination graph according to the structural constraint verification. The control resources include one or more of the following: energy storage system, flexible interconnection module, distributed power source and power distribution switch. The instruction generation module is used to write the matching results of control resource reachability as node attributes and edge constraints into the peak event combination graph to form a resource-enhanced peak event combination graph. Based on the resource-enhanced peak event combination graph, it dynamically generates differentiated control strategy combinations and sends them to the control resources.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0055] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. A method for demand regulation of transformer substations based on time-series learning and model prediction, characterized in that: Includes the following steps: S1 collects historical electricity consumption data and predicts total peak electricity consumption events in the future time period; S2, extract and analyze the duration, peak intensity, rise rate and over-limit energy area of ​​the total electricity consumption peak event, decompose the total electricity consumption peak event into multiple peak sub-events, determine the type and intensity of the peak sub-events, the type of the peak sub-events includes one or more of the following: transient impact type, continuous climbing type, periodic peak type and compound superposition type; S3, establish a peak event combination graph based on multiple peak sub-events, where the graph nodes of the peak event combination graph are peak sub-events, and the edges of the peak event combination graph are the structural relationships between peak sub-events; perform structural constraint verification on the peak event combination graph; step S3 includes: S3.1, taking each of the spike sub-events as a graph node, configure node attributes for each graph node based on the attributes of the spike sub-events; S3.2, Based on the association relationship and event type combination relationship among the multiple peak sub-events, calculate the structural association degree between any two graph nodes. When the structural association degree between two graph nodes meets the preset association condition, establish an edge between the two graph nodes and confirm the edge attribute to obtain the peak event combination graph. S3.3, perform one or more of the following checks on the peak event combination graph: node integrity check, edge validity check, time overlap consistency check, load contribution conservation check, topology association legality check, event type compatibility check, and graph structure connectivity check. When the peak event combination graph does not meet the check conditions, correct the graph nodes or edges that do not meet the check conditions. S4. Based on the peak event combination graph of the structural constraint verification, control resource reachability matching is performed on the nodes and edges of the peak event combination graph respectively. The control resources include one or more of the following: energy storage system, flexible interconnection module, distributed power source and power distribution switch. S5, the matching results of control resource reachability are written into the peak event combination graph as node attributes and edge constraints to form a resource-enhanced peak event combination graph. Based on the resource-enhanced peak event combination graph, differentiated control strategy combinations are dynamically generated and distributed to the control resources.

2. The method for demand regulation of transformer substations based on time-series learning and model prediction according to claim 1, characterized in that: Step S1 includes: S1.1, Collect power consumption status data of distribution substations within a historical time window, as well as control resource operation status data related to flexible interconnection and microgrid coordinated scheduling; S1.2 Based on the electrical topology of the distribution substation, the status of the distribution switch and the connection relationship of the flexible interconnection module, establish the electrical association relationship between the distribution substation, the line, the microgrid and the control resources, and bind the line capacity margin, demand threshold, and control resource availability status with the corresponding power consumption status data to form a power consumption status feature sequence with topology association information; S1.3, Based on the power consumption status feature sequence, combined with ambient temperature, production line scheduling plan and holiday labels, input into a pre-trained long short-term memory network, and output the load prediction sequence and its prediction confidence for the future time period through the long short-term memory network; S1.4 When the average load power of the load prediction sequence within the sliding window is greater than a preset load threshold, the duration is greater than a preset duration threshold, and the prediction confidence is greater than a preset confidence threshold, it is determined that the total electricity consumption peak event exists.

3. The method for demand regulation of transformer substations based on time-series learning and model prediction according to claim 2, characterized in that: Step S2 includes: S2.1, acquire historical peak event data, extract historical peak features based on the historical peak event data, and construct a peak superposition decomposition model based on the historical peak features; S2.2, Smooth the load prediction sequence corresponding to the total electricity consumption peak event, and identify the load change point. The load change point is one or more of the following: rising inflection point, peak point, falling inflection point, step point, and over-limit state switching point. S2.3, extract multiple load contribution segments based on the load change points and calculate the characteristic parameters of each load contribution segment respectively. Based on the characteristic parameters of the load contribution segments, decompose the total electricity consumption peak event into multiple peak sub-events that are superimposed within the same time window through the peak superposition decomposition model. S2.4, extract the sub-event contribution power of each spike sub-event as the intensity, and determine the type based on the sub-event characteristics of each spike sub-event.

4. The method for demand regulation of transformer substations based on time-series learning and model prediction according to claim 3, characterized in that: The types include: If the sub-event characteristics of the spike sub-event are continuous positive growth, long duration and slow rate of increase, then it is determined to be a continuous climbing type. If the sub-event characteristics of the spike sub-event are periodic repetitive fluctuations, and the similarity with the waveform of the same historical period is greater than or equal to a preset similarity threshold, then it is determined to be a periodic peak type. If the sub-event characteristics of the spike sub-event are high-amplitude short-duration pulses with a fast rise rate, then it is determined to be a transient impact type; If the sub-event characteristics of the peak sub-event are formed by the superposition of multiple load contribution components within the same time window, or if its waveform simultaneously has at least two of the characteristics of transient impact, continuous climbing, and periodic peak type, then it is determined to be a composite superposition type.

5. The method for demand regulation of transformer substations based on time-series learning and model prediction according to claim 1, characterized in that: Step S4 includes: S4.1, Based on each graph node and its node attributes in the peak event combination graph, determine the target control resource type that matches the graph node, query the real-time operating status of the control resource within the corresponding control time window, calculate the available capacity, response speed and continuous support capability of the control resource, and form a node-level control resource reachability matching result; S4.2 Based on each edge and its attributes in the peak event combination graph, and combined with the node-level control resource reachability matching results of the graph nodes connected by the edge, determine the resource coordination constraint type corresponding to the edge, and evaluate the coordination feasibility and capacity margin of the control resources associated with the edge when satisfying the constraints corresponding to the edge attributes, and form the edge-level control resource reachability matching results. S4.3, combine the node-level control resource reachability matching results and the edge-level control resource reachability matching results to form a reachability matching scheme.

6. The method for demand regulation of transformer substations based on time-series learning and model prediction according to claim 5, characterized in that: Step S4.3 includes: The consistency of the node-level control resource reachability matching results and the edge-level control resource reachability matching results is verified. When the control resources required by the node-level control resource reachability matching results and the edge-level control resource reachability matching results conflict in terms of time window, capacity allocation, and response timing, the conflict is resolved according to the preset priority of each sub-event and the edge relationship strength coefficient, and a unified reachability matching scheme is output.

7. The method for demand regulation of transformer substations based on time-series learning and model prediction according to claim 6, characterized in that: Step S5 includes: S5.1, the node-level control resource reachability matching result is written as a node attribute into the peak event combination graph, and the edge-level control resource reachability matching result is written as an edge attribute into the peak event combination graph to form a resource-enhanced peak event combination graph; S5.2, Traverse the graph nodes in the resource-enhanced peak event combination graph, generate corresponding single resource control instructions based on the event type of each node and its matching control resources, allocated power and execution time window, traverse the edges in the resource-enhanced peak event combination graph, and generate resource coordination control instructions based on the edge type, its carried coordination strategy type and remaining capacity margin. S5.3, combine the single resource control instruction and the resource coordination control instruction into a control strategy combination and decentralize it to the control resource.

8. A transformer area demand control system based on time-series learning and model prediction, used to implement the transformer area demand control method based on time-series learning and model prediction as described in any one of claims 1-7, characterized in that: Includes the following modules: The prediction module is used to collect historical electricity consumption data and predict total peak electricity consumption events in the future time period. The decomposition module is used to extract and analyze the duration, peak intensity, rise rate, and over-limit energy area of ​​the total electricity consumption peak event, decompose the total electricity consumption peak event into multiple peak sub-events, and determine the type and intensity of the peak sub-events. The types of peak sub-events include one or more of the following: transient impact type, continuous climbing type, periodic peak type, and compound superposition type. The peak event combination graph generation module is used to establish a peak event combination graph based on multiple peak sub-events, wherein the graph nodes of the peak event combination graph are peak sub-events, the edges of the peak event combination graph are the structural associations between peak sub-events, and the module performs structural constraint verification on the peak event combination graph. The control resource reachability matching module is used to perform control resource reachability matching on the nodes and edges in the peak event combination graph according to the structural constraint verification. The control resources include one or more of the following: energy storage system, flexible interconnection module, distributed power source and power distribution switch. The instruction generation module is used to write the matching results of control resource reachability as node attributes and edge constraints into the peak event combination graph to form a resource-enhanced peak event combination graph. Based on the resource-enhanced peak event combination graph, it dynamically generates differentiated control strategy combinations and sends them to the control resources.

Images