A method and system for optimizing distribution network operation based on load forecasting
By using load forecasting and counterfactual matching, this method identifies load growth segments in the distribution network and generates causal spectrums, eliminates source load growth, and performs spatiotemporal registration of action response images. This solves the problem of difficulty in identifying multiple sources of anomaly propagation in existing technologies, and achieves proactive optimization and enhanced security of distribution network operation.
Patent Information
- Application Number
- CN202611296226.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-25
- Publication Date
- 2026-11-17
AI Technical Summary
Existing power distribution network operation optimization methods are unable to identify the source load growth segments that trigger the continuous propagation of multiple anomalies, resulting in incomplete anomaly elimination or the generation of new operational anomalies, and it is difficult to accurately match the scope and direction of different candidate actions.
By combining load forecasting and counterfactual matching, basic data of the distribution network is collected and cleaned to generate load forecasting sequences and future operating status sequences. Load growth segments are identified and causal spectrums are generated to eliminate source load growth and generate counterfactual load absence images. Spatiotemporal registration of candidate action response images and counterfactual load absence images is performed, and target action response images are selected to generate optimization schemes.
It enables proactive optimization of distribution network operation, improves the accuracy of identifying load growth at the source, enhances the matching degree of abnormal operation events, and ensures the effectiveness and safety of optimization schemes.
Smart Images

Figure CN123121631A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of distribution network operation optimization technology, and in particular to a distribution network operation optimization method and system based on load forecasting. Background Technology
[0002] The existing distribution network operation optimization usually starts by collecting historical load, real-time operating status and topology data, using load forecasting methods to obtain future load changes, and then using power flow calculation to identify node voltage over-limit, line overload or transformer overload, and combines energy storage, reactive power compensation, load transfer and switch regulation to generate an operation plan; However, current methods often select adjustment actions directly based on the location of anomalies or the results of a single power flow. This makes it difficult to identify the source load growth segments that cause multiple anomalies to propagate continuously, and also fails to describe the spatiotemporal changes in operating status when the source load growth is absent. As a result, it is difficult to accurately match the scope, direction and combined response of different candidate actions, which can easily lead to problems such as incomplete anomaly elimination or the generation of new operating anomalies. Summary of the Invention
[0003] One objective of this invention is to propose a method and system for optimizing the operation of a distribution network based on load forecasting. This invention utilizes load forecasting and counterfactual matching to achieve the source tracing and optimization of distribution network anomalies, which is accurate, safe and efficient.
[0004] A distribution network operation optimization method based on load forecasting according to an embodiment of the present invention includes the following steps: Collect basic data of the distribution network and perform data cleaning, time alignment and topology association to generate distribution network operation data; The node load prediction sequence is generated by using distribution network operation data. Combined with distribution network basic data, a future operation simulation scenario is established and continuous power flow calculation is performed to generate a future operation state sequence and a set of future operation abnormal events. Divide the continuous load growth interval, generate a set of predicted load growth segments, establish the causal correspondence between the predicted load growth segments and future abnormal operation events, generate a congestion causal spectrum, and determine the source predicted load growth segment. The load growth amount of the source-predicted load growth segment is eliminated, the continuous power flow calculation is re-performed, a counterfactual operating state sequence is generated, and the difference is processed with the future operating state sequence to generate a counterfactual load absence image. A set of candidate operation actions is generated using adjustable resource data. Each candidate operation action is added to the future operation simulation scenario and continuous power flow calculation is performed to generate a set of action response images. The action response image and the counterfactual load absence image are spatiotemporally registered. The target action response image is selected in sequence and the corresponding candidate operation action is combined to generate a joint operation action combination. The current remaining load absence image is updated and the target action response image is connected to generate an action image mosaic set. Select target motion image mosaics from the motion image mosaic set, and generate power distribution network operation optimization schemes based on the joint operation action combinations corresponding to the target motion image mosaics.
[0005] Optionally, the generation of the distribution network operation data includes: Collect load forecasting data, distribution network status data, distribution network topology data, and adjustable resource data to generate distribution network basic data; Remove duplicate and invalid data from the distribution network basic data, supplement missing data and correct abnormal data to generate cleaned distribution network basic data; The cleaned distribution network basic data is time-aligned according to the collection time. Based on the distribution network topology data, the load forecast basic data, distribution network status data and adjustable resource data are associated with the corresponding distribution nodes and equipment locations to generate distribution network operation data.
[0006] Optionally, the generation of the set of future operational exception events includes: The load of each distribution node at multiple future times is predicted using distribution network operation data. The prediction results are arranged according to the distribution node and the future time to generate a node load prediction sequence. By associating the node load forecast sequence with the basic data of the distribution network, the forecast load at each future time point is configured to the corresponding distribution node, and a future operation simulation scenario is established. In the future operation simulation scenario, continuous power flow calculation is performed according to future time points, and the node voltage, line power flow and transformer load corresponding to each equipment location are arranged to generate a future operation state sequence. Identify abnormal operation events in the future operating state sequence that exceed the corresponding operating limits in the distribution network state data, record the abnormality type, equipment location, occurrence time and abnormality magnitude, and generate a set of future abnormal operation events.
[0007] Optionally, the generation of the congestion causal hierarchy includes: Calculate the load change of each distribution node in adjacent future moments, connect consecutive positive future moments, and generate a set of predicted load growth segments; The load growth of each predicted load growth segment is reduced according to an increasing reduction ratio, while the remaining predicted loads remain unchanged, to generate a gradient load reduction prediction sequence; The gradient reduction load prediction sequence is added to the future operation simulation scenario to perform continuous power flow calculation, generating a set of gradient reduction operation abnormal events. The set of abnormal operation events during gradient reduction is mapped to the set of abnormal operation events in the future. An abnormal reduction response trajectory is generated based on the changes in abnormal amplitude and occurrence time, and a causal correspondence is established between the predicted load growth segment and the abnormal operation events in the future. Connect future operational anomalies caused by the same predicted load growth segment, whose occurrence times are sequentially delayed and whose topological locations are continuous, generate anomaly propagation pairs and splice them together to form a congestion causal spectrum; The congestion causal spectrum is sorted to determine the target congestion causal spectrum and its corresponding source predicted load growth segment.
[0008] Optionally, the generation of the counterfactual load absence image includes: Read the node positions, start and end times, and load growth amounts of the source predicted load growth segments, and arrange them according to future times to generate a source load growth sequence; The predicted load of the corresponding node is subtracted hourly from the corresponding load increase during the start and end times, while keeping the other predicted loads unchanged, to generate a counterfactual node load prediction sequence. The corresponding node load prediction sequence is replaced with the counterfactual node load prediction sequence. The remaining data in the future operation simulation scenario are kept unchanged and the continuous power flow calculation is re-performed to generate the counterfactual operation state sequence. The counterfactual operating state sequence and the future operating state sequence are interpolated according to the equipment location, future time, and operating state type to generate a set of operating state difference elements containing the direction and magnitude of change. Connect the operating state difference elements that have the same operating state type, are located continuously along the distribution network topology, are continuous in the future time and have the same direction of change, and generate a set of load absence response fragments. The load absence response segments are arranged according to the equipment location and future time, and the status channels are divided according to the operating status type. The corresponding change direction and change amplitude are written to generate a counterfactual load absence image.
[0009] Optionally, the generation of the motion response image set includes: Based on the resource type, resource location, adjustable time period, adjustment direction, and adjustment range of adjustable resources, the execution time and adjustment amount are divided to generate a set of candidate running actions; Each candidate action is added to the future simulation scenario in sequence, while keeping the operating state of the remaining adjustable resources unchanged and performing continuous power flow calculations to generate a sequence of candidate action operating states. The candidate action running state sequence and the future running state sequence are processed by difference according to the device location, future time and running state type to generate a set of action response elements containing the direction and magnitude of change; The action response elements are arranged according to the data arrangement method of the counterfactual load absence image, and the corresponding device location, future time and operating status type are written to generate the action response image corresponding to each candidate operating action. The motion response images are associated with the corresponding candidate running actions and aggregated to generate a motion response image set.
[0010] Optionally, the generation of the motion image mosaic set includes: The counterfactual load absence image is used as the initial residual load absence image, and spatiotemporal registration is performed with each action response image. The same-direction coverage and reverse interference are calculated respectively. Action response images with coverage greater than interference in the same direction are identified as the first target action response images. Action selection paths are established, and the initial remaining load absence images are copied as the current remaining load absence images for each path. The corresponding candidate operation actions are used to generate a combination of joint operation actions. The various joint operation actions are combined and added to the future operation simulation scenario for continuous power flow calculation, generating a joint operation state sequence. After difference processing and arrangement, a joint operation response image is generated. The joint action response image is matched with the current remaining load absence image item by item. The difference in change amplitude is calculated for positions with the same change direction, and the change amplitude is superimposed for positions with opposite change direction to update the current remaining load absence image. The unselected action response images and the updated current remaining load absence images are re-registered in time and space. The action response image with the largest difference between the same-direction coverage and the reverse interference is selected. The corresponding candidate operation actions are added to the joint operation action combination. The continuous power flow calculation is re-performed and the current remaining load absence images are updated until there are no unselected action response images with the same-direction coverage greater than the reverse interference. Connect the target motion response images in the selected order to generate motion image mosaic, and associate it with the corresponding joint operation motion combination to generate a motion image mosaic set.
[0011] Optionally, the generation of the power distribution network operation optimization scheme includes: Read the combined operation action combinations corresponding to each motion image mosaic, add them to the future operation simulation scenario to perform continuous power flow calculation, and generate a sequence to verify the future operation state; Identify and verify abnormal operation events in the sequence of future operation states, and generate a set of abnormal operation events to verify the future operation. The set of future operational anomalies will be verified and matched with the set of future operational anomalies according to the anomaly type and device location to determine the operational anomalies that have disappeared and the newly added operational anomalies. The motion image mosaics are selected from those where all previous future operational anomalies have disappeared and no new operational anomalies have been added. A set of qualified motion image mosaics is generated, and the qualified motion image mosaic with the fewest candidate operational actions is determined as the target motion image mosaic. Read the combined operation action combination corresponding to the target action image mosaic, arrange the resource type, resource location, adjustment direction and adjustment amount of the candidate operation actions according to the execution time, and generate a distribution network operation optimization scheme.
[0012] A distribution network operation optimization system based on load forecasting according to an embodiment of the present invention includes: The data processing module is used to collect and process basic data of the distribution network and generate distribution network operation data; The load forecasting module is used to predict the load of each distribution node at multiple future times, and generate a node load forecast sequence and a set of future operational anomalies. The causal analysis module is used to generate a set of predicted load growth segments and a congestion causal spectrum, and to identify the source predicted load growth segments; The counterfactual analysis module is used to eliminate the load growth corresponding to the source predicted load growth segment and generate a counterfactual load absence image. The motion response generation module is used to generate a set of candidate running actions and a set of motion response images corresponding to each candidate running action; The action combination module is used to perform spatiotemporal registration of action response images and counterfactual load absence images to generate joint action combinations and action image mosaic sets. The scheme generation module is used to filter target motion image mosaics that will eliminate future abnormal operation events without generating new abnormal operation events, and generate power distribution network operation optimization schemes.
[0013] The beneficial effects of this invention are: This invention collects load forecasting data, distribution network status data, distribution network topology data, and adjustable resource data, and performs data cleaning, time alignment, and topology correlation. This enables distribution network basic data from different sources and at different collection times to be accurately correlated to the corresponding distribution nodes and equipment locations, providing a consistent data foundation for node load forecasting, future operation simulation, and continuous power flow calculation. By predicting the load of each distribution node at multiple future times and generating a future operating state sequence, it is possible to identify future operational anomalies corresponding to node voltage, line power flow, and transformer load in advance. This transforms distribution network operation optimization from passive adjustment after anomalies occur to proactive optimization before anomalies occur.
[0014] This invention establishes a causal relationship between predicted load growth segments and future operational anomalies by dividing the predicted load growth segments into segments and performing incremental reduction and continuous power flow calculations on the load growth of each segment. By generating a congestion causal spectrum through connecting future operational anomalies with consecutive occurrence times and topological locations, it can identify the source predicted load growth segment from multiple load changes and multiple operational anomalies. This avoids selecting adjustment targets solely based on the location of abnormal equipment or a single power flow result, improves the accuracy of identifying the source of future operational anomalies, and reflects the gradual propagation of operational anomalies along the distribution network topology.
[0015] This invention further eliminates the load growth corresponding to the source-predicted load growth segment. It generates a counterfactual load absence image by using the difference between the counterfactual operating state sequence and the future operating state sequence. This allows for a unified expression of the direction and magnitude of change for each device's location, future time, and operating state type when source load growth does not exist. Simultaneously, the operating state changes generated by each candidate operating action are converted into action response images consistent with the arrangement of the counterfactual load absence image. This enables comparison of candidate operating actions of different resource types, locations, and adjustment directions within the same spatiotemporal data structure.
[0016] This invention utilizes spatiotemporal registration between action response images and current remaining load absence images. It sequentially selects target action response images by comprehensively considering both unidirectional coverage and reverse interference. Furthermore, it updates the actual operational response generated by combined operational actions through continuous power flow calculations. This approach reduces the mutual cancellation or amplification of influence between multiple candidate operational actions. Finally, by filtering target action image mosaics that eliminate existing future operational anomalies without generating new ones, a distribution network operation optimization scheme is generated. This improves the matching degree between adjustable resource combinations and future operational anomalies, enhancing the effectiveness and operational safety of the distribution network operation optimization scheme. Attached Figure Description
[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a distribution network operation optimization method based on load forecasting proposed in this invention; Figure 2 This diagram illustrates the relationship between predicted load growth segments, future abnormal operation events, and congestion causal spectrum in a load forecast-based distribution network operation optimization method proposed in this invention. Detailed Implementation
[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0019] refer to Figures 1-2 A distribution network operation optimization method based on load forecasting includes the following steps: The system collects basic data of the distribution network and performs data cleaning, time alignment, and topology correlation to generate distribution network operation data. The basic data of the distribution network includes load forecasting basic data, distribution network status data, distribution network topology data, and adjustable resource data. The load of each distribution node at multiple future times is predicted using distribution network operation data, generating a node load prediction sequence. A future operation simulation scenario is established by combining distribution network basic data and continuous power flow calculation is performed to generate a future operation state sequence. An abnormal operation event in the future operation state sequence is identified to generate a set of future operation abnormal events. The load growth intervals in the node load prediction sequence are divided to generate a set of predicted load growth segments. In future operation simulation scenarios, the load growth corresponding to each predicted load growth segment is reduced and continuous power flow calculation is performed again. The future operation anomalies before and after reduction are compared to generate the causal correspondence between the predicted load growth segments and the future operation anomalies. The future operation anomalies caused by the same predicted load growth segment and which are continuous at the time of occurrence and topological location are connected to generate a congestion causal spectrum. The source predicted load growth segment is determined from the congestion causal spectrum. The load growth corresponding to the source predicted load growth segment in the node load prediction sequence is eliminated. The continuous power flow calculation is re-performed in the future operation simulation scenario to generate a counterfactual operation state sequence. The counterfactual operation state sequence is then compared with the future operation state sequence, and the comparison results are arranged to generate a counterfactual load absence image. A candidate set of operational actions is generated using adjustable resource data. Each candidate operational action is added to the future operational simulation scenario and continuous power flow calculation is performed again to generate a sequence of candidate operational action operational states. The difference between the candidate operational action operational state sequence and the future operational state sequence is processed, and an action response image set is generated according to the data arrangement of the counterfactual load absence image. The counterfactual load absence image is used as the initial current remaining load absence image. Each action response image is spatiotemporally registered with the current remaining load absence image. Target action response images are selected in sequence. Candidate operation actions corresponding to each target action response image are combined in sequence to generate a joint operation action combination. The current remaining load absence image is updated through continuous power flow calculation. Each target action response image is connected in the selection order to generate an action image mosaic set. Select target motion image mosaics from the motion image mosaic set that will eliminate future operational anomalies without generating new ones, and generate power distribution network operation optimization schemes based on the combined operation actions corresponding to the target motion image mosaics.
[0020] In this embodiment, the generation of distribution network operation data includes: Collect load forecasting data, distribution network status data, distribution network topology data, and adjustable resource data to generate distribution network basic data; Remove duplicate and invalid data from the distribution network basic data, supplement missing data and correct abnormal data to generate cleaned distribution network basic data; The cleaned distribution network basic data is time-aligned according to the collection time. Based on the distribution network topology data, the load forecast basic data, distribution network status data and adjustable resource data are associated with the corresponding distribution nodes and equipment locations to generate distribution network operation data.
[0021] In this embodiment, the generation of the future runtime exception event set includes: The load of each distribution node at multiple future times is predicted using distribution network operation data. The prediction results are arranged according to the distribution node and the future time to generate a node load prediction sequence. By associating the node load forecast sequence with the basic data of the distribution network, the forecast load at each future time point is configured to the corresponding distribution node, and a future operation simulation scenario is established. In the future operation simulation scenario, continuous power flow calculation is performed according to future time points, and the node voltage, line power flow and transformer load corresponding to each equipment location are arranged to generate a future operation state sequence. Identify abnormal operation events in the future operating state sequence that exceed the corresponding operating limits in the distribution network state data, record the abnormality type, equipment location, occurrence time and abnormality magnitude, and generate a set of future abnormal operation events.
[0022] In this embodiment, the generation of the congestion causal spectrum includes: The load change of each distribution node in the load prediction sequence is calculated in the adjacent future time. The future time when the load change is continuously positive is connected, and a set of predicted load growth segments containing node location, start time, end time and load growth amount is generated. Specifically, the load change between adjacent future times of the same distribution node is calculated in the order of future times. The consecutive positive load changes are connected. The previous future time corresponding to the first positive load change is determined as the start time, and the next future time corresponding to the last positive load change is determined as the end time. The increment of the predicted load of each future time from the start time to the end time relative to the predicted load of the start time is calculated as the load growth of the corresponding future time. Select the predicted load growth segments in sequence, reduce the load growth amount at the corresponding future time according to the incremental reduction ratio, keep the remaining predicted load unchanged, and generate the gradient reduction load prediction sequence corresponding to each predicted load growth segment; Specifically, within the range of load growth from non-reduction to complete reduction, multiple incremental reduction ratios are sequentially divided. The load growth amount at each future time within the predicted load growth segment is multiplied by each incremental reduction ratio to obtain the reduction amount at the corresponding future time. The reduction amount is deducted from the predicted load at the corresponding distribution node and the corresponding future time, while keeping the remaining predicted load unchanged. The processed predicted loads are arranged according to the reduction ratios to generate the gradient reduction load prediction sequence corresponding to each predicted load growth segment. The gradient reduction load prediction sequence is added to the future operation simulation scenario for continuous power flow calculation, and the operation anomaly events corresponding to each reduction ratio are identified to generate a gradient reduction operation anomaly event set. The set of gradient reduction operation anomalies and the set of future operation anomalies are mapped according to anomaly type and equipment location. The changes in the anomaly amplitude and occurrence time of each future operation anomaly as the reduction ratio increases are arranged to generate anomaly reduction response trajectory. Future operation anomalies with continuously decreasing anomaly amplitude or continuously delayed occurrence time are associated with the corresponding predicted load growth segments to generate causal correspondence. Specifically, the abnormal operation events identified under each reduction ratio are matched with the set of future abnormal operation events according to the abnormality type and equipment location. The abnormality amplitude and occurrence time of the same future abnormal operation event are arranged in ascending order of the reduction ratio. Abnormal operation events that do not continue to occur are written with zero abnormality amplitude and non-occurrence mark. The abnormality amplitude and occurrence time corresponding to each reduction ratio are connected to generate an abnormality reduction response trajectory. The abnormality reduction response trajectory is associated with the predicted load growth segment that generated the trajectory to generate a causal correspondence. Based on causal correspondence, future operational anomalies that are associated with the same predicted load growth segment, whose occurrence time is sequentially delayed and whose topological positions are continuous are connected as anomaly propagation pairs. The anomaly propagation pairs are spliced together sequentially according to their occurrence time to generate a congestion causal spectrum. Specifically, the future operational anomalies associated with the same predicted load growth segment are arranged according to their occurrence time. The connection order of the equipment locations corresponding to each future operational anomaly in the distribution network topology data is read. Adjacent future operational anomalies whose connection occurrence time is successively delayed and whose equipment locations are connected step by step along the distribution network topology are generated into anomaly propagation pairs. The anomaly propagation pairs containing common future operational anomalies are sequentially connected according to their occurrence time to generate a congestion causal spectrum. Based on the order of the number of future operational anomalies from most to least, the cumulative change of the anomaly mitigation response trajectory from largest to smallest, and the occurrence time of the first future operational anomaly from earliest to latest, the congestion causal spectrum is sorted in sequence. The congestion causal spectrum that ranks first is determined as the target congestion causal spectrum, and the predicted load growth segment corresponding to the target congestion causal spectrum is determined as the source predicted load growth segment. Specifically, the reduction in abnormal amplitude relative to the original abnormal amplitude under each reduction ratio is calculated. The reduction in abnormal amplitude is divided by the operating limit adopted for the corresponding abnormal event to generate a dimensionless abnormal change. The dimensionless abnormal change in the same abnormal reduction response trajectory is accumulated in the order of reduction ratio. The accumulated results of each abnormal reduction response trajectory contained in the same congestion causal spectrum are added together to obtain the cumulative change of the abnormal reduction response trajectory.
[0023] In this embodiment, the generation of the counterfactual load absence image includes: The load growth rate is generated by reading the node position, start time, end time and load growth amount at each future time from the source predicted load growth segment, and arranging the load growth amount according to the future time. Specifically, the node positions, start times, end times, and recorded load growth amounts for each future time are read from the source-predicted load growth segments. The load growth amounts are arranged in chronological order of the future time segments, and the arrangement results are associated with the corresponding node positions to generate a source load growth sequence. The corresponding load growth in the source load growth sequence is subtracted hourly from the predicted load of the corresponding node in the node load prediction sequence from the start time to the end time, while keeping the remaining predicted loads unchanged, to generate the counterfactual node load prediction sequence. Specifically, the source load growth sequence and the node load prediction sequence are matched one by one according to the node location and future time. The predicted load of the corresponding node from the start time to the end time is reduced by the load growth at the same future time, so that the predicted load within this time range is restored to the load level before the source predicted load growth segment is formed. The predicted loads of the remaining distribution nodes and the remaining future time remain unchanged, thus generating the counterfactual node load prediction sequence. The corresponding node load prediction sequence in the future operation simulation scenario is replaced with the counterfactual node load prediction sequence. The rest of the data in the future operation simulation scenario remains unchanged, and the continuous power flow calculation is re-performed to generate the counterfactual operation state sequence. The counterfactual operating state sequence and the future operating state sequence are matched item by item according to equipment location, future time and operating state type. The difference between the corresponding operating state values is calculated to generate a set of operating state difference elements. Each operating state difference element includes equipment location, future time, operating state type, direction of change and magnitude of change. Specifically, a one-to-one correspondence is established between the counterfactual operating state sequence and the future operating state sequence according to the equipment location, future time, and operating state type. The operating state difference at the corresponding location is generated by subtracting the future operating state value from the counterfactual operating state value. The sign of the operating state difference indicates the direction of change, and the absolute value of the operating state difference indicates the magnitude of change. The equipment location, future time, operating state type, direction of change, and magnitude of change are written into the same data record to generate a set of operating state difference elements. Under the same operating state type, the operating state difference elements with continuous connection device locations along the distribution network topology, continuous future time, and consistent change direction are used to generate a set of load absence response fragments. Specifically, the operating status difference elements are grouped according to the type of operating status and the direction of change, and arranged layer by layer according to the order of future times. For the operating status difference elements of the connected equipment location along the distribution network topology at each future time, the connection results of the same equipment location or continuous along the distribution network topology in adjacent future times are connected. The connected operating status difference elements are written into the same load absence response segment and aggregated to generate a load absence response segment set. The load absence response fragment set is arranged according to the equipment location and future time, the status channel is divided according to the operating status type, and the change direction and change amplitude in each status channel are written into the corresponding equipment location and future time to generate a counterfactual load absence image. Specifically, the equipment location arrangement order is determined according to the distribution network topology data, the future time arrangement order is determined according to the node load prediction sequence, and the status channel is divided according to the operation status type in the future operation status sequence. The change direction and change amplitude of each operation status difference element in the load absence response segment are written into the corresponding equipment location, future time and status channel, and the remaining data locations are written with zero change amplitude, thus generating a counterfactual load absence image.
[0024] In this embodiment, the generation of the motion response image set includes: Based on the resource type, resource location, adjustable time period, adjustment direction, and adjustment range in the adjustable resource data, the execution time and adjustment amount are divided respectively. The resource type, resource location, execution time, adjustment direction, and adjustment amount are associated to generate a set of candidate running actions. Specifically, the adjustment range in the adjustable resource data is divided into multiple sequentially arranged adjustment amounts, and the adjustable time period is divided into multiple execution times. For adjustable resources with multiple executable states, the change direction and change amount corresponding to the current state switching to each executable state are calculated, and the change amount is used as the adjustment amount. Resource type, resource location, execution time, adjustment direction and adjustment amount are associated and aggregated to generate a set of candidate running actions. Candidate running actions are selected in sequence and added to the future running simulation scenario. The running state of the other adjustable resources remains unchanged. The continuous power flow is recalculated according to the future time and the corresponding candidate action running state sequence is generated. Specifically, the adjustment direction and adjustment amount in the candidate operation actions are written into the corresponding power distribution node, equipment location or topology connection location in the future operation simulation scenario, while keeping the operation status of the other adjustable resources unchanged. Power flow calculations are performed sequentially according to future time, and the node voltage, line power flow and transformer load corresponding to each equipment location are arranged to generate the corresponding candidate operation operation status sequence. Align the candidate action running state sequence with the future running state sequence item by item according to the device location, future time and running state type, calculate the difference of the corresponding running state value, and generate a set of action response elements containing device location, future time, running state type, change direction and change amplitude based on the positive and negative signs of the difference and the absolute value of the difference; Specifically, a one-to-one correspondence is established between the candidate action running state sequence and the future running state sequence according to the equipment location, future time, and running state type. The action response difference is generated by subtracting the future running state value from the candidate action running state value. The sign of the action response difference indicates the direction of change, and the absolute value of the action response difference indicates the magnitude of change. The equipment location, future time, running state type, direction of change, and magnitude of change are written into the same data record to generate a set of action response elements. According to the data arrangement method of the counterfactual load absence image, the action response element set is arranged according to the equipment location, future time and operating status type, and the change direction and change amplitude are written into the corresponding positions to generate the action response image corresponding to each candidate operating action. Specifically, the device location order, future time order, and status channel order in the counterfactual load absence image are read to establish the data location corresponding to the counterfactual load absence image. The change direction and change amplitude in the action response element are written to the data location corresponding to the same device location, future time, and operating status type, and the remaining data location is written with zero change amplitude to generate the action response image corresponding to each candidate operating action. Each action response image is associated with its corresponding candidate action, and the action response images are aggregated to generate an action response image set.
[0025] In this embodiment, the generation of the motion picture splice set includes: The counterfactual load absence image is used as the initial residual load absence image. Each action response image and the initial residual load absence image are spatiotemporally registered according to the equipment location, future time and operating status type. The same-direction coverage at the location with the same direction of change and the reverse interference at the location with the opposite direction of change are accumulated respectively. Specifically, the action response image and the initial remaining load absence image are matched item by item according to equipment location, future time and operating status type. The change amplitude is divided by the operating limit adopted for the corresponding operating status type to generate a normalized change amplitude. For data locations with the same change direction, the smaller value of the normalized change amplitude corresponding to the two images is taken and accumulated to obtain the same-direction coverage. For data locations with opposite change directions, the normalized change amplitude in the action response image is accumulated to obtain the reverse interference. The action response images with the same-direction coverage greater than the reverse interference are respectively identified as the first target action response images. Action selection paths are established, the initial remaining load absence images are copied as the current remaining load absence images of each action selection path, and the candidate operation actions corresponding to the first target action response image are used to generate a combination of joint operation actions. Specifically, the difference between the same-direction coverage and the reverse interference corresponding to each action response image is calculated, and the action response images are arranged from largest to smallest according to the obtained difference. The action response images with positive difference are determined as the first target action response images, and an action selection path is established for each first target action response image. The initial remaining load absence image is copied to each action selection path, and the candidate operation action corresponding to the first target action response image is used to generate the corresponding joint operation action combination. The combined operation action of each action selection path is added to the future operation simulation scenario for continuous power flow calculation, generating a combined action operation state sequence. The combined action operation state sequence is then processed by difference with the future operation state sequence, and a combined action response image is generated according to the data arrangement of the counterfactual load absence image. Specifically, the combined operation actions in the same action selection path are written into the future operation simulation scenario, and the continuous power flow is recalculated according to the future time to generate a combined operation operation state sequence. The combined operation operation state value is subtracted from the future operation state value to generate a combined operation response difference. The sign of the combined operation response difference indicates the direction of change, and its absolute value indicates the magnitude of change. The combined operation response image is generated according to the data arrangement method of the counterfactual load absence image. The joint action response image is matched with the current remaining load absence image one by one. The difference in change amplitude is calculated at the position where the change direction is consistent. The change direction is maintained or reversed according to the positive or negative difference. The change amplitude is superimposed at the position where the change direction is opposite to update the current remaining load absence image. Specifically, the direction of change in the current remaining load absence image and the joint action response image is converted into a positive or negative sign, and combined with the corresponding change amplitude to form a signed change value. The signed change value of the current remaining load absence image is subtracted from the signed change value of the joint action response image to generate a remaining signed change value. The sign of the remaining signed change value is used to generate the updated direction of change, and its absolute value is used to generate the updated change amplitude. The current remaining load absence image is then updated. The unselected action response images are spatiotemporally registered with the updated current remaining load absence images. The action response image with the largest difference between the same-direction coverage and the reverse interference is selected as the next target action response image. The corresponding candidate operation actions are added to the joint operation action combination. The joint operation action combination is added to the future operation simulation scenario for continuous power flow calculation. The updated joint action response image is generated according to the data arrangement of the counterfactual load absence images. The current remaining load absence image is updated according to the change amplitude processing method in S64 until there are no unselected action response images with the same-direction coverage greater than the reverse interference. Specifically, the candidate operation actions corresponding to the next target action response image are added to the joint operation action combination. The updated joint operation action combination is written into the future operation simulation scenario and the continuous power flow calculation is performed again to generate the updated joint action operation state sequence. The updated joint action response image is generated by subtracting the future operation state value from the updated joint action operation state value. The initial remaining load absence image is used as the calculation benchmark. The signed change value of the initial remaining load absence image is subtracted from the signed change value of the updated joint action response image. The current remaining load absence image is updated with the sign and absolute value of the difference. Connect the target action response images according to the selection order to generate the action image mosaic of the corresponding action selection path. Associate the action image mosaic with the corresponding joint operation action combination, gather all action image mosaics, and generate an action image mosaic set. Specifically, a stitching sequence number is set for each target action response image according to the selection order of the target action response images, and the target action response images are arranged according to the stitching sequence number. Each target action response image is associated with the corresponding candidate running action, and the arranged target action response images are associated with the joint running action combination of the corresponding action selection path to generate an action image stitching body. The action image stitching bodies corresponding to each action selection path are collected to generate an action image stitching body set.
[0026] In this embodiment, the generation of the distribution network operation optimization scheme includes: Read the combined operation action combinations corresponding to each motion image mosaic in the motion image mosaic set, add each combined operation action combination to the future operation simulation scenario and perform continuous power flow calculation to generate the corresponding verification future operation state sequence; Identify and verify abnormal operation events in the sequence of future operating states, record the abnormality type, device location, occurrence time and abnormality magnitude, and generate a corresponding set of abnormal events for verifying future operation. The set of future operational anomalies is matched with the set of future operational anomalies according to the anomaly type and device location. Future operational anomalies that do not have a corresponding operational anomaly in the set of future operational anomalies are identified as disappeared operational anomalies. Operational anomalies that do not have a corresponding future operational anomaly in the set of future operational anomalies are identified as newly added operational anomalies. Specifically, the verification of future operational anomalies and future operational anomalies are grouped according to anomaly type and device location. Operational anomalies within the same group are arranged in chronological order of occurrence and matched one by one according to the order of arrangement. Operational anomalies that do not have a corresponding result in the future operational anomalies are identified as disappeared operational anomalies, and operational anomalies that do not have a corresponding result in the verification of future operational anomalies are identified as newly added operational anomalies. The motion image mosaics that have disappeared and cover the set of future motion anomalies without any new motion anomalies are selected to generate a set of qualified motion image mosaics. The number of candidate actions contained in the joint operation action combination corresponding to each qualified motion image mosaic is counted, and the qualified motion image mosaic with the fewest candidate operation actions is determined as the target motion image mosaic. Read the combined operation action combination corresponding to the target action image mosaic, arrange the resource type, resource location, adjustment direction and adjustment amount of each candidate operation action according to the execution time, and generate a distribution network operation optimization scheme.
[0027] A distribution network operation optimization system based on load forecasting includes: The data processing module is used to collect and process basic data of the distribution network and generate distribution network operation data; The load forecasting module is used to predict the load of each distribution node at multiple future times, and generate a node load forecast sequence and a set of future operational anomalies. The causal analysis module is used to generate a set of predicted load growth segments and a congestion causal spectrum, and to identify the source predicted load growth segments; The counterfactual analysis module is used to eliminate the load growth corresponding to the source predicted load growth segment and generate a counterfactual load absence image. The motion response generation module is used to generate a set of candidate running actions and a set of motion response images corresponding to each candidate running action; The action combination module is used to perform spatiotemporal registration of action response images and counterfactual load absence images to generate joint action combinations and action image mosaic sets. The scheme generation module is used to filter target motion image mosaics that will eliminate future abnormal operation events without generating new abnormal operation events, and generate power distribution network operation optimization schemes.
[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to an operation optimization scenario in a regional power distribution network dispatch center. This power distribution network includes 26 distribution nodes, 30 distribution lines, and 4 distribution transformers, and is connected to 3 energy storage devices, 2 reactive power compensation devices, 2 transferable loads, and 1 tie switch. When the load increases significantly, the voltage at the end distribution nodes will drop first, causing a continuous increase in the load on adjacent lines, upstream lines, and distribution transformers. Existing methods typically arrange adjustment actions separately for each operational anomaly, making it difficult to identify whether multiple operational anomalies are caused by the same load increase process. This easily leads to problems such as duplicate actions, mutual cancellation of effects, or the generation of new operational anomalies.
[0029] In this scenario, load forecasting baseline data, distribution network status data, distribution network topology data, and adjustable resource data are collected. The collected data undergoes deduplication (removing duplicates and invalid data), supplementation of missing data, and correction of abnormal data. The data is then time-aligned according to the collection time, and various data types are associated with corresponding distribution nodes and equipment locations to generate distribution network operation data. Over 30 consecutive days, a total of 2880 sets of aligned data were generated, including supplementation of 27 sets of missing data and correction of 11 sets of abnormal data. The processed load data, operation status data, and adjustable resource data can all be accurately mapped to the corresponding equipment locations.
[0030] The load of each distribution node at multiple future times is predicted using distribution network operation data, generating a node load prediction sequence. The predicted load of a certain terminal distribution node continuously increases from 1.42 MW to 2.36 MW, forming a predicted load growth segment of 0.94 MW. The node load prediction sequence is configured in a future operation simulation scenario and continuous power flow calculation is performed to generate a future operation state sequence.
[0031] By identifying the future operational state sequence, a total of five future operational anomaly events were identified. The voltage at the end distribution node drops to 9.62 kV, below the operational limit of 9.70 kV; the current in the adjacent distribution line increases to 428 A, exceeding the operational limit of 400 A; the current in the upstream distribution line increases to 365 A, exceeding the operational limit of 350 A; the output capacity of the corresponding distribution transformer increases to 6.28 MVA, exceeding the operational limit of 6.00 MVA; and the voltage at the adjacent distribution node drops to 9.66 kV, below the operational limit of 9.70 kV. These operational anomaly events are sequential in their occurrence and in terms of equipment connection location.
[0032] Calculate the load change at adjacent future moments for each distribution node, connect the future moments where the load change is continuously positive, and generate a set of predicted load growth segments; for each predicted load growth segment, reduce the corresponding load growth by increasing reduction ratios of 0.2, 0.4, 0.6 and 0.8, and recalculate the continuous power flow.
[0033] For the predicted load growth segments at the aforementioned terminal distribution nodes, the current exceeding the operating limits of adjacent distribution lines decreased sequentially from 28 amps to 21 amps, 13 amps, 5 amps, and 0 amps, and the output capacity exceeding the operating limits of distribution transformers decreased sequentially from 0.28 MVA to 0.20 MVA, 0.11 MVA, 0.04 MVA, and 0 MVA. As the reduction ratio increased, each abnormal operation event was gradually delayed and eventually disappeared. When the same treatment was applied to other predicted load growth segments, the current changes of the aforementioned distribution lines were all less than 4 amps, and the output capacity changes of the distribution transformers were all less than 0.03 MVA.
[0034] Therefore, the predicted load growth segment of the terminal distribution node is associated with five future operational anomaly events to generate a causal correspondence; then, the future operational anomaly events that occur sequentially and whose equipment locations are connected level by level along the distribution network topology are connected as anomaly transmission pairs, and are spliced together in sequence to generate a congestion causal spectrum.
[0035] The future runtime segment forms three congestion causal profiles. One congestion causal profile contains five future runtime anomaly events, with a cumulative change of 0.46 in the anomaly mitigation response trajectory. The other two congestion causal profiles contain three and two future runtime anomaly events, respectively, with cumulative changes of 0.21 and 0.13. By arranging the congestion causal profiles, the congestion causal profile containing five future runtime anomaly events is identified as the target congestion causal profile, and the corresponding predicted load growth segment is identified as the source predicted load growth segment.
[0036] Read the load growth amount from the source predicted load growth segment to generate the source load growth amount sequence, and subtract the corresponding load growth amount from the node load prediction sequence hourly to generate the counterfactual node load prediction sequence; add the counterfactual node load prediction sequence to the future operation simulation scenario to recalculate the continuous power flow and generate the counterfactual operation state sequence.
[0037] After subtracting the counterfactual operating state sequence from the future operating state sequence, the voltage at the end distribution node increases by 0.19 kV, the current in the adjacent distribution line decreases by 39 A, the current in the upstream distribution line decreases by 26 A, the output capacity of the distribution transformer decreases by 0.41 MVA, and the voltage at the adjacent distribution node increases by 0.17 kV. Arranging the directions and magnitudes of these changes according to equipment location, future time, and operating state type generates a counterfactual load absence image.
[0038] Based on adjustable resource data, 24 candidate operation actions were generated. Each candidate operation action was added to the future operation simulation scenario and continuous power flow calculation was performed to generate 24 action response images. After spatiotemporal registration, the same-direction coverage of energy storage discharge action was 0.38 and the reverse interference was 0.04; the same-direction coverage of load transfer action was 0.29 and the reverse interference was 0.03; and the same-direction coverage of reactive power compensation action was 0.22 and the reverse interference was 0.02.
[0039] Energy storage discharge action, load transfer action, and reactive power compensation action are selected sequentially to generate a combination of joint operation actions. After each candidate operation action is added, the continuous power flow calculation is re-performed, and the current remaining load absence image is updated with the initial remaining load absence image as the calculation benchmark to avoid the candidate operation actions that have been added being calculated repeatedly.
[0040] After the combined operation was executed, the current of the adjacent distribution line decreased from 428 amps to 382 amps, the current of the upstream distribution line decreased from 365 amps to 331 amps, the output capacity of the distribution transformer decreased from 6.28 MVA to 5.72 MVA, the voltage of the terminal distribution node increased from 9.62 kV to 9.92 kV, and the voltage of the adjacent distribution node increased from 9.66 kV to 9.89 kV. All of the above operating status values were restored to the corresponding operating limit range.
[0041] An action image mosaic was generated according to the selection order of the three target action response images, and the corresponding joint operation action combination was added to the future operation simulation scenario for verification. The verification results show that all five original future operation anomalies have disappeared, and no new operation anomalies have been generated. When using the original method of adjusting the position of each anomaly, five candidate operation actions need to be executed, and the current of another power distribution line increases to 414 amps, exceeding the operation limit of 400 amps. After adopting the present invention, only three candidate operation actions need to be executed.
[0042] During a 30-day application period, 40 future operating segments were analyzed, identifying 112 future operational anomalies. After using the distribution network operation optimization scheme generated by this invention, all 112 future operational anomalies disappeared in the verified future operating state sequence. The final output distribution network operation optimization scheme did not generate any new operational anomalies. Each distribution network operation optimization scheme contained an average of 2.8 candidate operating actions, compared to an average of 4.6 candidate operating actions in the original method, representing a reduction of 1.8 candidate operating actions. The above operating records demonstrate that this invention can identify the source predicted load growth segments that cause multiple future operational anomalies and generate distribution network operation optimization schemes with fewer actions that do not cause new operational anomalies.
[0043] Table 1. Comparison of Anomaly Elimination and Action Configuration Performance of Different Distribution Network Operation Optimization Methods
[0044] As shown in Table 1, among the same 112 future operational anomalies, the traditional anomaly-by-anomaly location adjustment method eliminated 103, with an operational anomaly elimination rate of 92.0%; the elimination rates of the sensitivity-based ranking adjustment method and the multi-objective power flow optimization method were 94.6% and 98.2%, respectively; the method of the present invention eliminated 111 future operational anomalies, with an elimination rate of 99.1%.
[0045] Traditional anomaly-by-anomaly location adjustment methods generated 8 new operational anomaly events, resulting in a qualified future runtime rate of only 67.5%. Sensitivity-ranking-based adjustment methods still generated 5 new operational anomaly events, while multi-objective power flow optimization methods generated 2. The method of this invention verifies the combined operational action combinations through continuous power flow calculations, filtering out action image mosaics containing unresolved or newly generated operational anomaly events, ensuring that 39 future runtimes meet the requirements, achieving a qualified future runtime rate of 97.5%.
[0046] Regarding action configuration, the traditional method of adjusting actions based on anomaly location uses 184 candidate actions, averaging 4.6 actions per future runtime segment. The method of this invention uses 112 candidate actions, averaging 2.8 actions per future runtime segment, reducing the number of candidate actions by 72 compared to the traditional method, a reduction rate of 39.1%. This indicates that this invention does not configure actions for each operational anomaly event individually, but rather selects candidate actions that can simultaneously cover multiple device locations and multiple future time-series changes in operational status.
[0047] The performance improvement of this invention lies in the fact that it first establishes a causal correspondence by gradient reduction of predicted load growth segments, and then uses the congestion causal spectrum to determine the source of the predicted load growth segments. Subsequently, it generates counterfactual load absence images and action response images, and combines the same-direction coverage and reverse interference to form a joint operational action combination. This process can reduce the duplication and reverse effects between candidate operational actions, and avoids the generation of new operational anomalies through final verification.
[0048] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for optimizing the operation of a distribution network based on load forecasting, characterized in that, Includes the following steps: Collect basic data of the distribution network and perform data cleaning, time alignment and topology association to generate distribution network operation data; The node load prediction sequence is generated by using distribution network operation data. Combined with distribution network basic data, a future operation simulation scenario is established and continuous power flow calculation is performed to generate a future operation state sequence and a set of future operation abnormal events. Divide the continuous load growth interval, generate a set of predicted load growth segments, establish the causal correspondence between the predicted load growth segments and future abnormal operation events, generate a congestion causal spectrum, and determine the source predicted load growth segment. The load growth amount of the source-predicted load growth segment is eliminated, the continuous power flow calculation is re-performed, a counterfactual operating state sequence is generated, and the difference is processed with the future operating state sequence to generate a counterfactual load absence image. A set of candidate operation actions is generated using adjustable resource data. Each candidate operation action is added to the future operation simulation scenario and continuous power flow calculation is performed to generate a set of action response images. The action response image and the counterfactual load absence image are spatiotemporally registered. The target action response image is selected in sequence and the corresponding candidate operation action is combined to generate a joint operation action combination. The current remaining load absence image is updated and the target action response image is connected to generate an action image mosaic set. Select target motion image mosaics from the motion image mosaic set, and generate power distribution network operation optimization schemes based on the joint operation action combinations corresponding to the target motion image mosaics.
2. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the power distribution network operation data includes: Collect load forecasting data, distribution network status data, distribution network topology data, and adjustable resource data to generate distribution network basic data; Remove duplicate and invalid data from the distribution network basic data, supplement missing data and correct abnormal data to generate cleaned distribution network basic data; The cleaned distribution network basic data is time-aligned according to the collection time. Based on the distribution network topology data, the load forecast basic data, distribution network status data and adjustable resource data are associated with the corresponding distribution nodes and equipment locations to generate distribution network operation data.
3. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the set of future operational anomalies includes: The load of each distribution node at multiple future times is predicted using distribution network operation data. The prediction results are arranged according to the distribution node and the future time to generate a node load prediction sequence. By associating the node load forecast sequence with the basic data of the distribution network, the forecast load at each future time point is configured to the corresponding distribution node, and a future operation simulation scenario is established. In the future operation simulation scenario, continuous power flow calculation is performed according to future time points, and the node voltage, line power flow and transformer load corresponding to each equipment location are arranged to generate a future operation state sequence. Identify abnormal operation events in the future operating state sequence that exceed the corresponding operating limits in the distribution network state data, record the abnormality type, equipment location, occurrence time and abnormality magnitude, and generate a set of future abnormal operation events.
4. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the congestion causal spectrum includes: Calculate the load change of each distribution node in adjacent future moments, connect consecutive positive future moments, and generate a set of predicted load growth segments; The load growth of each predicted load growth segment is reduced according to an increasing reduction ratio, while the remaining predicted loads remain unchanged, to generate a gradient load reduction prediction sequence; The gradient reduction load prediction sequence is added to the future operation simulation scenario to perform continuous power flow calculation, generating a set of gradient reduction operation abnormal events. The set of abnormal operation events during gradient reduction is mapped to the set of abnormal operation events in the future. An abnormal reduction response trajectory is generated based on the changes in abnormal amplitude and occurrence time, and a causal correspondence is established between the predicted load growth segment and the abnormal operation events in the future. Connect future operational anomalies caused by the same predicted load growth segment, whose occurrence times are sequentially delayed and whose topological locations are continuous, generate anomaly propagation pairs and splice them together to form a congestion causal spectrum; The congestion causal spectrum is sorted to determine the target congestion causal spectrum and its corresponding source predicted load growth segment.
5. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the counterfactual load absence image includes: Read the node positions, start and end times, and load growth amounts of the source predicted load growth segments, and arrange them according to future times to generate a source load growth sequence; The predicted load of the corresponding node is subtracted hourly from the corresponding load increase during the start and end times, while keeping the other predicted loads unchanged, to generate a counterfactual node load prediction sequence. The corresponding node load prediction sequence is replaced with the counterfactual node load prediction sequence. The remaining data in the future operation simulation scenario are kept unchanged and the continuous power flow calculation is re-performed to generate the counterfactual operation state sequence. The counterfactual operating state sequence and the future operating state sequence are interpolated according to the equipment location, future time, and operating state type to generate a set of operating state difference elements containing the direction and magnitude of change. Connect the operating state difference elements that have the same operating state type, are located continuously along the distribution network topology, are continuous in the future time and have the same direction of change, and generate a set of load absence response fragments. The load absence response segments are arranged according to the equipment location and future time, and the status channels are divided according to the operating status type. The corresponding change direction and change amplitude are written to generate a counterfactual load absence image.
6. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the motion response image set includes: Based on the resource type, resource location, adjustable time period, adjustment direction, and adjustment range of adjustable resources, the execution time and adjustment amount are divided to generate a set of candidate running actions; Each candidate action is added to the future simulation scenario in sequence, while keeping the operating state of the remaining adjustable resources unchanged and performing continuous power flow calculations to generate a sequence of candidate action operating states. The candidate action running state sequence and the future running state sequence are processed by difference according to the device location, future time and running state type to generate a set of action response elements containing the direction and magnitude of change; The action response elements are arranged according to the data arrangement method of the counterfactual load absence image, and the corresponding device location, future time and operating status type are written to generate the action response image corresponding to each candidate operating action. The motion response images are associated with the corresponding candidate running actions and aggregated to generate a motion response image set.
7. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the motion picture mosaic set includes: The counterfactual load absence image is used as the initial residual load absence image, and spatiotemporal registration is performed with each action response image. The same-direction coverage and reverse interference are calculated respectively. Action response images with coverage greater than interference in the same direction are identified as the first target action response images. Action selection paths are established, and the initial remaining load absence images are copied as the current remaining load absence images for each path. The corresponding candidate operation actions are used to generate a combination of joint operation actions. The various joint operation actions are combined and added to the future operation simulation scenario for continuous power flow calculation, generating a joint operation state sequence. After difference processing and arrangement, a joint operation response image is generated. The joint action response image is matched with the current remaining load absence image item by item. The difference in change amplitude is calculated for positions with the same change direction, and the change amplitude is superimposed for positions with opposite change direction to update the current remaining load absence image. The unselected action response images and the updated current remaining load absence images are re-registered in time and space. The action response image with the largest difference between the same-direction coverage and the reverse interference is selected. The corresponding candidate operation actions are added to the joint operation action combination. The continuous power flow calculation is re-performed and the current remaining load absence images are updated until there are no unselected action response images with the same-direction coverage greater than the reverse interference. Connect the target motion response images in the selected order to generate motion image mosaic, and associate it with the corresponding joint operation motion combination to generate a motion image mosaic set.
8. The method for optimizing distribution network operation based on load forecasting according to claim 1, characterized in that, The generation of the power distribution network operation optimization scheme includes: Read the combined operation action combinations corresponding to each motion image mosaic, add them to the future operation simulation scenario to perform continuous power flow calculation, and generate a sequence to verify the future operation state; Identify and verify abnormal operation events in the sequence of future operation states, and generate a set of abnormal operation events to verify the future operation. The set of future operational anomalies will be verified and matched with the set of future operational anomalies according to the anomaly type and device location to determine the operational anomalies that have disappeared and the newly added operational anomalies. The motion image mosaics are selected from those where all previous future operational anomalies have disappeared and no new operational anomalies have been added. A set of qualified motion image mosaics is generated, and the qualified motion image mosaic with the fewest candidate operational actions is determined as the target motion image mosaic. Read the combined operation action combination corresponding to the target action image mosaic, arrange the resource type, resource location, adjustment direction and adjustment amount of the candidate operation actions according to the execution time, and generate a distribution network operation optimization scheme.
9. A distribution network operation optimization system based on load forecasting, comprising executing the distribution network operation optimization method based on load forecasting as described in any one of claims 1 to 8, characterized in that, include: The data processing module is used to collect and process basic data of the distribution network and generate distribution network operation data; The load forecasting module is used to predict the load of each distribution node at multiple future times, and generate a node load forecast sequence and a set of future operational anomalies. The causal analysis module is used to generate a set of predicted load growth segments and a congestion causal spectrum, and to identify the source predicted load growth segments; The counterfactual analysis module is used to eliminate the load growth corresponding to the source predicted load growth segment and generate a counterfactual load absence image. The motion response generation module is used to generate a set of candidate running actions and a set of motion response images corresponding to each candidate running action; The action combination module is used to perform spatiotemporal registration of action response images and counterfactual load absence images to generate joint action combinations and action image mosaic sets. The scheme generation module is used to filter target motion image mosaics that will eliminate future abnormal operation events without generating new abnormal operation events, and generate power distribution network operation optimization schemes.
