An edge-computing-based power grid side non-intrusive power load identification method

Abstract

The present application relates to the technical field of power system electricity monitoring, in particular to a power grid side non-intrusive electricity load identification method based on edge computing; original electrical quantity data of a target power supply area is continuously collected and cyclically retained, the original electrical quantity data is time-aggregated, and power supply topology relationship and phase attribution relationship are obtained; a load change event is determined according to the aggregated electrical quantity data, power change characteristics and phase current change characteristics are extracted, and a candidate load state is generated; power balance deviation and phase matching deviation are calculated based on the power supply topology relationship and the phase attribution relationship, an identification matching value is generated, and a target load state is determined; when the identification matching value is greater than a preset matching threshold, the original electrical quantity data of the corresponding time period and the preceding and subsequent time periods of the load change event is retrieved for re-identification; when the identification matching value is not greater than the preset matching threshold, a load category, an operating state and an electricity power are output, and load composition information of the target power supply area is updated.

Description

Technical Field

[0001] This invention relates to the field of power system electricity consumption monitoring technology, specifically a non-intrusive power load identification method on the grid side based on edge computing. Background Technology

[0002] With the continuous increase in the types and number of electrical loads in low-voltage distribution networks, the low-voltage side and feeder side of distribution transformers need to continuously monitor the load composition, operating status, and power consumption within the target power supply area to support distribution operation analysis, load dispatching, and power supply management. Non-intrusive load identification collects active power, reactive power, and phase current data at the power supply inlet to analyze changes in electrical quantities caused by load switching. It eliminates the need for independent monitoring equipment at each load location, making it suitable for centralized monitoring of low-voltage distribution networks.

[0003] Existing non-intrusive load identification methods typically match load status based on the power change characteristics of load change events. In distribution networks containing multi-level power supply branches and three-phase power supply lines, loads connected to different power supply branches and different power supply phases may produce similar power change characteristics. It is difficult to determine the load status based solely on power change characteristics, combined with the connection relationship of power supply branches and the connection relationship of power supply phases, which can easily affect the accuracy of load composition information in the target power supply area.

[0004] Furthermore, edge computing nodes located on the low-voltage side and feeder side of distribution transformers are limited by computing power and storage capacity. Continuously processing high-temporal-resolution raw electrical quantity data increases edge computing resource consumption, while using time-aggregated electrical quantity data may result in the loss of load change details. Existing technologies struggle to accurately identify load states with similar power change characteristics in multi-branch, three-phase power supply scenarios under conditions of limited edge computing resources.

[0005] To address this, a non-intrusive power load identification method based on edge computing is proposed for the power grid side. Summary of the Invention

[0006] This invention aims to provide a non-intrusive power load identification method based on edge computing on the power grid side. It performs power balance and phase matching verification on candidate load status through power supply topology and phase affiliation relationships. When the identification matching value does not meet the conditions, it calls the original electrical quantity data for re-identification to output load category, operating status and power consumption, and update the load composition information of the target power supply area.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A non-intrusive power load identification method based on edge computing on the power grid side includes: The original electrical quantity data of the target power supply area is continuously collected and retained in a cyclical overlay manner. The original electrical quantity data is aggregated over time to obtain aggregated electrical quantity data, and the power supply topology and phase attribution relationships are obtained. Based on aggregated electrical quantity data, load change events are identified, power change characteristics and phase current change characteristics are extracted, and at least one candidate load state is generated. Based on the power supply topology and phase assignment, the power balance deviation and phase matching deviation of each candidate load state are calculated, and the identification matching value of each candidate load state is generated. The candidate load state with the smallest identification matching value is determined as the target load state. When the identification matching value corresponding to the target load state is greater than the preset matching threshold, the original electrical quantity data of the time period corresponding to the retained load change event and the time periods before and after are retrieved, the power change characteristics and phase current change characteristics are extracted again, the candidate load state and identification matching value are generated, and the target load state is determined. When the identification matching value corresponding to the target load status is not greater than the preset matching threshold, the load category, operating status and power consumption of the target load status are output and the load composition information of the target power supply area is updated.

[0008] Preferably, the continuous acquisition and cyclical retention of the original electrical quantity data of the target power supply area, the time aggregation of the original electrical quantity data to obtain aggregated electrical quantity data, and the acquisition of power supply topology and phase attribution relationships include: continuously acquiring active power data, reactive power data, and effective value data of phase current of each power supply phase through an edge computing node set at one of the low-voltage side and feeder side of the distribution transformer to form original electrical quantity data; aligning the original electrical quantity data according to the sampling time to obtain the original synchronous electrical quantity sequence; cyclically retaining the original synchronous electrical quantity sequence and performing time aggregation on the original synchronous electrical quantity sequence to obtain aggregated electrical quantity data; generating power supply topology relationships based on the power supply branch connection information and acquisition point location information pre-stored by the edge computing node; and generating phase attribution relationships based on the load identifier, load access branch information, and corresponding power supply information pre-stored by the edge computing node.

[0009] Preferably, determining the load change event based on the aggregated electrical quantity data includes: calculating the active power difference and reactive power difference between adjacent sampling times in the aggregated electrical quantity data; determining the sum of the absolute values ​​of the active power difference and the reactive power difference as the event detection quantity; determining the continuous sampling interval where the event detection quantity is greater than the event judgment threshold as the event interval; and extracting the sampling data before the event interval, the sampling data within the event interval, and the sampling data after the event interval to obtain the load change event data segment.

[0010] Preferably, the step of extracting power change features and phase current change features to generate at least one candidate load state includes: calculating the average active power, average reactive power, and average effective value of phase current of each power supply phase before and after the event interval in the load change event data segment; subtracting the average before the event interval from the average after the event interval to obtain the change in active power, the change in reactive power, and the change in effective value of phase current of each power supply phase; generating power change features based on the change in active power and the change in reactive power; generating phase current change features based on the change in effective value of phase current of each power supply phase; comparing the change in active power, the change in reactive power, and the change in effective value of phase current of the power supply phase corresponding to each load state with the active power change template, the reactive power change template, and the effective value of phase current change template in the load state feature library, respectively; and determining the load state in which the absolute value of the difference between each change and the corresponding change template is within the corresponding allowable deviation range as the candidate load state.

[0011] Preferably, the calculation of the power balance deviation for each candidate load state includes: reading the active power change template, reactive power change template, and power supply branch identifier for each candidate load state from the load state feature library; determining the active power change template and reactive power change template as the predicted active power change and predicted reactive power change corresponding to the location of the collection point based on the upstream and downstream relationship between the power supply branch and the collection point location; and calculating the sum of the absolute values ​​of the differences between the predicted active power change and the predicted reactive power change to obtain the power balance deviation.

[0012] Preferably, the calculation of the phase matching deviation of each candidate load state includes: reading the load identifier and phase current effective value change template of each candidate load state from the load state feature library; determining the power supply phase corresponding to each candidate load state according to the load identifier and phase attribution relationship, configuring the phase current effective value change template to the corresponding power supply phase, and obtaining the predicted phase current change characteristics; calculating the sum of the absolute values ​​of the difference between the predicted phase current effective value change and the phase current effective value change of the same power supply phase, and obtaining the phase matching deviation.

[0013] Preferably, the step of generating identification matching values ​​for each candidate load state and determining the candidate load state with the smallest identification matching value as the target load state includes: normalizing the power balance deviation and phase matching deviation based on the power deviation benchmark value and phase deviation benchmark value pre-stored by the edge computing node; performing weighted calculation on the normalized power balance deviation and phase matching deviation according to preset power weight and preset phase weight to obtain the identification matching value for each candidate load state; determining the candidate load state with the smallest identification matching value as the target load state, and comparing the identification matching value corresponding to the target load state with a preset matching threshold.

[0014] Preferably, when the identification matching value corresponding to the target load state is greater than a preset matching threshold, the target load state is re-determined, including: extracting the original electrical quantity data of the corresponding time period and the time periods before and after the event interval from the original synchronous electrical quantity sequence retained in a cyclic manner according to the sampling time of the event interval; re-extracting the power change characteristics and phase current change characteristics according to the original electrical quantity data of the corresponding time period and the time periods before and after the event interval, generating candidate load states, calculating the power balance deviation and phase matching deviation, generating an identification matching value and determining the target load state; when the identification matching value corresponding to the re-determined target load state is not greater than the preset matching threshold, reading the load category, operating status and status power value from the load status feature library, determining the status power value as the power consumption, outputting the load category, operating status and power consumption, and updating the load composition information of the target power supply area.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention, based on the power supply topology of the target power supply area, associates the power supply branch identifiers corresponding to candidate load states with the upstream and downstream connection paths between collection points. It also reads active power change templates and reactive power change templates formed from historically labeled events at the same collection point, generating predicted active power change and predicted reactive power change at the collection point. Furthermore, it calculates the power balance deviation between the predicted power change and the actual power change. This process ensures that the power change templates, power supply branches, and collection point locations corresponding to candidate load states remain aligned, providing power deviation data for subsequent generation of identification matching values ​​based on phase attribution relationships.

[0016] 2. This invention, based on the phase assignment relationship of the target power supply area, associates the load identifier and phase current change template corresponding to the candidate load state with the power supply phase, and configures the phase current change template to the corresponding power supply phase to form predicted phase current change characteristics. Then, it calculates the phase matching deviation between the predicted phase current change characteristics and the measured phase current change characteristics. This process can verify the consistency between the load state identification result and the actual power supply phase change situation. Compared with judging solely by total active power or total reactive power, it can reduce the confusion of load assignment between different power supply phases under three-phase power supply conditions.

[0017] 3. This invention performs cyclic retention and temporal aggregation on continuously collected raw electrical quantity data, and uses the aggregated electrical quantity data to complete load change event detection, feature extraction, and routine identification. When the aggregated electrical quantity data has identified a load change event and the corresponding identification matching value is greater than a preset matching threshold, the raw electrical quantity data of the corresponding time period and the time periods before and after the load change event are retrieved to re-extract features and re-identify, thereby reducing the quantization impact of temporal aggregation on the event interval boundaries of the detected event and the calculation of the average electrical quantity before and after. When the identification matching value is not greater than the preset matching threshold, the identification result is directly output. This process can reduce the computational and storage overhead required for edge computing nodes to continuously process high temporal resolution data, and provide a data foundation for further identification of load change events that have been detected but have a low degree of matching. Attached Figure Description

[0018] Figure 1 This is a flowchart of a non-intrusive power load identification method based on edge computing for the power grid side according to the present invention; Figure 2 This is a flowchart of the original electrical quantity data aggregation and load change event data processing of the present invention; Figure 3 This is a diagram illustrating the relationship between candidate load state matching and target load state determination in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figures 1 to 3 This invention provides a non-intrusive power load identification method based on edge computing on the power grid side, referring to... Figure 1 Flowchart Figure 2 A flowchart of the aggregation of raw electrical quantity data and the processing of load change event data, and Figure 3 Relationship diagram for matching candidate load status and determining target load status; Technical solution as follows: The original electrical quantity data of the target power supply area is continuously collected and retained in a cyclical overlay manner. The original electrical quantity data is aggregated over time to obtain aggregated electrical quantity data, and the power supply topology and phase attribution relationships are obtained. Based on aggregated electrical quantity data, load change events are identified, power change characteristics and phase current change characteristics are extracted, and at least one candidate load state is generated. Based on the power supply topology and phase assignment, the power balance deviation and phase matching deviation of each candidate load state are calculated, and the identification matching value of each candidate load state is generated. The candidate load state with the smallest identification matching value is determined as the target load state. When the identification matching value corresponding to the target load state is greater than the preset matching threshold, the original electrical quantity data of the time period corresponding to the retained load change event and the time periods before and after are retrieved, the power change characteristics and phase current change characteristics are extracted again, the candidate load state and identification matching value are generated, and the target load state is determined. When the identification matching value corresponding to the target load status is not greater than the preset matching threshold, the load category, operating status and power consumption of the target load status are output and the load composition information of the target power supply area is updated.

[0021] Example 1: Furthermore, the continuous acquisition and cyclical retention of original electrical quantity data of the target power supply area, the time aggregation of the original electrical quantity data to obtain aggregated electrical quantity data, and the acquisition of power supply topology and phase attribution relationships include: continuously acquiring active power data, reactive power data, and effective value data of phase current of each power supply phase through an edge computing node set at one of the low-voltage side and feeder side of the distribution transformer to form original electrical quantity data; aligning the original electrical quantity data according to the sampling time to obtain the original synchronous electrical quantity sequence; cyclically retaining the original synchronous electrical quantity sequence and performing time aggregation on the original synchronous electrical quantity sequence to obtain aggregated electrical quantity data; generating power supply topology relationships based on the power supply branch connection information and acquisition point location information pre-stored by the edge computing node; and generating phase attribution relationships based on the load identifier, load access branch information, and corresponding power supply information pre-stored by the edge computing node.

[0022] Specifically, taking the example of an edge computing node located on the low-voltage side of a distribution transformer, it continuously acquires active power data, reactive power data, and effective current data of each phase of the power supply area. The sampling time is then matched with the electrical quantities obtained at the same sampling time, forming raw electrical quantity data arranged according to sampling time. In this embodiment, the target power supply area is determined based on the observable power supply range of the acquisition point where the edge computing node is located. When the edge computing node is located on the low-voltage side of the distribution transformer, the target power supply area is the power supply range covered by each low-voltage power supply branch downstream of the acquisition point; when the edge computing node is located on the feeder side, the target power supply area is the power supply range downstream of the feeder acquisition point. The load status feature library includes load identifiers, load statuses, power supply branch identifiers, and corresponding change templates located within the target power supply area.

[0023] Using the sampling time as an index, active power data, reactive power data, and effective value data of phase current of each power supply phase with the same sampling time are combined into a synchronous electrical quantity record. The continuously formed synchronous electrical quantity records constitute the original synchronous electrical quantity sequence, so that each electrical quantity in the same record corresponds to the same operating time of the target power supply area.

[0024] The original synchronous electrical quantity sequence is cyclically retained according to the sampling time order. The retention duration corresponding to the cyclic retention of the original synchronous electrical quantity sequence is determined jointly based on the total time span before, within, and after the event interval in the load change event data segment, and the longest processing time required for the edge computing node to complete the identification and matching value comparison corresponding to the target load state from the formation of the load change event data segment, and the retention duration is not less than the sum of the total time span and the longest processing time. The longest processing time is determined based on the maximum processing time recorded by the edge computing node when processing historical load change events. Based on the sampling frequency of the original electrical quantity data and the retention duration, the number of synchronous electrical quantity records that can be stored within the data range used to retain the original synchronous electrical quantity sequence is determined. When the data range is full, the earliest synchronous electrical quantity record is overwritten by the newly generated synchronous electrical quantity record. When the number of pending events and resource occupancy status of the edge computing node do not exceed the historical processing load range corresponding to the longest processing time, the original electrical quantity data of the time period corresponding to the load change event and the time periods before and after the completion of the identification matching value comparison are still in a reserved state; when the actual processing time exceeds the reserved time and the original electrical quantity data corresponding to the load change event has been covered, the re-identification based on the covered data is not performed, the current load status identification result is not output, and the load composition information of the target power supply area is not updated.

[0025] Time merging is performed on continuous raw synchronous electrical quantity records, and the average values ​​of active power data, reactive power data, and effective values ​​of phase current of each power supply phase are determined within the same merging time range to form aggregated electrical quantity records. Aggregated electrical quantity data is composed of aggregated electrical quantity records arranged in chronological order. Using the sampling period of the raw electrical quantity data as the basic time unit, within a range not exceeding the shortest duration of load change events in historically marked events, different integer multiples of the sampling period are used as candidate time aggregation spans, and time aggregation is performed on the raw synchronous electrical quantity sequences corresponding to historically marked events. Among the candidate time aggregation spans that can still determine the event interval corresponding to the known load switching period according to the event judgment threshold after aggregation, the candidate time aggregation span with the largest value is selected as the time aggregation span.

[0026] Based on the branch identifier and connection direction between branches in the power supply branch connection information, the upstream and downstream relationships between each power supply branch are determined, and the collection points are configured to the corresponding power supply branches according to the collection point location information, so as to obtain the power supply topology relationship that represents the collection points, power supply branches and upstream and downstream connection relationships.

[0027] Using the load identifier as the associated object, the corresponding load access branch information and the corresponding power supply information are associated to obtain the phase association relationship between each load and the access branch, as well as one or more power supply phases, so as to calculate the power balance deviation and phase matching deviation according to the branch and each power supply phase corresponding to the candidate load status.

[0028] Through the above steps, the original electrical quantity data can be synchronized, retained, and aggregated over time, forming power supply topology relationships and phase attribution relationships, providing a data foundation for subsequent load change event identification and candidate load status matching.

[0029] Further, determining load change events based on aggregated electrical quantity data includes: calculating the active power difference and reactive power difference between adjacent sampling times in the aggregated electrical quantity data; determining the sum of the absolute values ​​of the active power difference and reactive power difference as the event detection quantity; determining the continuous sampling interval where the event detection quantity is greater than the event judgment threshold as the event interval; and extracting the sampled data before the event interval, the sampled data within the event interval, and the sampled data after the event interval to obtain the load change event data segment.

[0030] Specifically, the aggregated electrical quantity data is read in the order of sampling time, and the active power and reactive power at the current sampling time are subtracted from the corresponding data at the previous sampling time to obtain the active power difference and reactive power difference at the current sampling time.

[0031] The absolute values ​​of the active power difference and reactive power difference are taken separately. The active power difference is expressed in watts or kilowatts, and the reactive power difference is expressed in volt-amperes or kilovolt-amperes, with both using the same decimal conversion factor. The numerical values ​​of the active power difference and reactive power difference are extracted separately, and the sum of the absolute values ​​of the two values ​​is determined as the event detection quantity. The event detection quantity is a composite numerical index used to characterize the combined degree of change in active power and reactive power, and is not assigned a single physical unit corresponding to watts, volt-amperes, or volt-amperes, nor is it considered as actual active power, reactive power, or apparent power output. The event judgment threshold uses the same numerical scale as the event detection quantity.

[0032] The event determination threshold is determined based on the fluctuation range of event detection quantities in the target power supply area when no load switching occurs, as well as the magnitude of load changes to be identified. The event determination threshold uses the same numerical scale as the event detection quantities, and its value is set to be higher than the normal fluctuation range of event detection quantities under stable operating conditions, while still retaining the changes caused by the load switching to be identified. For example, if the maximum event detection quantity corresponding to historical stable operating data is 6.0, the event determination threshold is set to 6.0.

[0033] The event detection values ​​are compared sequentially with the event determination threshold based on the sampling time. The sampling time when the value first exceeds the event determination threshold is taken as the starting position of the event interval. Subsequent sampling times that are all greater than the event determination threshold are grouped into the same event interval, until the event detection value corresponding to a subsequent sampling time is no greater than the event determination threshold. For example, if the event determination threshold is 6.0, and the event detection values ​​corresponding to three consecutive sampling times are 8.2, 10.1, and 7.4, respectively, then all three event detection values ​​are greater than the event determination threshold, and these three sampling times are collectively defined as one event interval.

[0034] Sampling data within an event interval is extracted from aggregated electrical quantity data. Continuous sampling data adjacent to the event interval time is extracted both before and after the event interval. The sampling data before and after the event interval each cover a time span equal to the duration of the event interval, using sampling records where the event detection quantity does not exceed the event determination threshold. If the continuous sampling data before or after the event interval is insufficient to cover the stated time span, no corresponding load change event data segment is formed. The sampling data before, within, and after the event interval are combined in chronological order to form a load change event data segment, which serves as input for subsequent calculation of the average electrical quantity before and after the event interval and for extracting load change characteristics.

[0035] Through the above steps, the event interval can be determined based on the power changes in the aggregated electrical quantity data, and a load change event data segment containing sampled data before, during, and after the event can be formed.

[0036] Further, the step of extracting power change features and phase current change features to generate at least one candidate load state includes: calculating the average active power, average reactive power, and average effective value of phase current of each power supply phase before and after the event interval in the load change event data segment; subtracting the average before the event interval from the average after the event interval to obtain the change in active power, the change in reactive power, and the change in effective value of phase current of each power supply phase; generating power change features based on the change in active power and the change in reactive power; generating phase current change features based on the change in effective value of phase current of each power supply phase; comparing the change in active power, the change in reactive power, and the change in effective value of phase current of the power supply phase corresponding to each load state with the active power change template, the reactive power change template, and the effective value of phase current change template in the load state feature library, respectively; and determining the load state where the absolute value of the difference between each change and the corresponding change template is within the corresponding allowable deviation range as the candidate load state.

[0037] Specifically, the sampled data before and after the event interval are read from the load change event data segment, and the mean values ​​of active power, reactive power and effective values ​​of phase current of each power supply phase are calculated in the two time periods. The obtained mean values ​​represent the relatively stable power consumption state before and after the load change.

[0038] Subtracting the corresponding mean before the event interval from the mean of each item after the event interval yields the changes in active power, reactive power, and the effective value of the phase current of each power supply phase. Positive values ​​indicate an increase in the corresponding electrical quantity, while negative values ​​indicate a decrease in the corresponding electrical quantity. For example, when the mean active power before and after the event interval is 420 watts and 1180 watts respectively, the change in active power is 760 watts.

[0039] The load status feature database is established based on historical labeled events of the target power supply area. Historical labeled events are obtained by performing controlled switching and synchronous recording of the corresponding loads within the target power supply area, given the load identifier, access branch, and power supply phase. During controlled switching, the load identifier, operating status before switching, operating status after switching, and switching period are recorded. Based on the load access branch information and corresponding power supply phase information pre-stored by the edge computing node, the corresponding access branch and power supply phase are read. The labeled content is then associated with the original synchronous electrical quantity sequence formed by the acquisition point where the edge computing node is located, according to the switching period, to form historical labeled events. The active power change template, reactive power change template, and phase current RMS value change template calculated from historical labeled events all represent the equivalent electrical quantity changes formed by the corresponding load status at the acquisition point, rather than the nameplate rated values ​​of the load terminals or load-end measurements not transmitted by the line. Only historical labeled events with complete original synchronous electrical quantity records before, within, and after the event interval, and without other registered load status changes occurring during the controlled switching period, are used to establish the load status feature database. If there are missing sampling records, the labeled content does not correspond to the switching time period, or other load status changes occur during the controlled switching period, the corresponding events will not be included in the calculation of the change template, state power value, and allowable deviation range.

[0040] Historically labeled events are grouped according to load identification and post-switching operating status. For each historically labeled event, the changes in active power, reactive power, and the effective value changes of phase current for each power supply phase are calculated using the aforementioned feature extraction method. The arithmetic mean of the changes in active power, reactive power, and the effective value changes of phase current for each power supply phase are calculated for each historically labeled event in the same group. These average values ​​are then used as the active power change template, reactive power change template, and phase current effective value change template for the corresponding load status, retaining the positive and negative directions of each change. After a new confirmed historically labeled event is added, it is assigned to the corresponding group, and the change templates are recalculated. For loads corresponding to multiple power supply phases, the phase current effective value change template is calculated and stored separately for each power supply phase. For loads with different specifications and switching characteristics, load identification is configured, and corresponding load status templates are established.

[0041] The state power values ​​in the load state feature library are determined based on the operating states corresponding to historically labeled events. The state power value corresponding to a stopped operating state is set to 0. For other operating states, the state power corresponding to each historically labeled event is obtained by summing the known state power value of the previous operating state with the change in active power when the load switches from the previous operating state to the current operating state. The arithmetic mean of the state power values ​​corresponding to the same load identifier and the same operating state is then determined as the state power value for that operating state. The state power values ​​are associated with and stored in the load state feature library along with the corresponding load identifier, load category, and operating state.

[0042] The candidate screening criteria consist of the allowable deviation range corresponding to each change template. For historical switching data under the same load condition, the absolute values ​​of the differences between the historical active power change, historical reactive power change, and historical phase current RMS value change of each power supply phase and the corresponding change template are calculated. The maximum value among the absolute values ​​of each type of difference is determined as the corresponding upper limit of the allowable deviation, and the allowable deviation range is formed from 0 to the upper limit of the allowable deviation.

[0043] For each load state, the power change characteristics are compared with the corresponding active power change template and reactive power change template, respectively. The phase current change characteristics of the power supply phase associated with the corresponding load identifier in the phase attribution relationship are compared with the corresponding phase current RMS value change template. Power supply phases not associated with the corresponding load identifier are not included in the candidate selection of the load state. Load states whose power change characteristics and the compared phase current change characteristics are both within the corresponding allowable deviation range are determined as candidate load states. The load identifier, operating status, and various change templates corresponding to the candidate load states are retained for subsequent calculation of power balance deviation and phase matching deviation. When no load state meets the candidate selection criteria, no identification matching value is generated for this load change event, and no load category, operating status, or power consumption is output, nor is the load composition information of the target power supply area updated.

[0044] Through the above steps, corresponding power change characteristics and phase current change characteristics can be formed from the load change event data segment, and candidate load states that match the change characteristics can be screened out.

[0045] Further, the calculation of the power balance deviation for each candidate load state includes: reading the active power change template, reactive power change template, and power supply branch identifier for each candidate load state from the load state feature library; determining the active power change template and reactive power change template as the predicted active power change and predicted reactive power change corresponding to the location of the collection point based on the upstream and downstream relationship between the power supply branch and the collection point location; and calculating the sum of the absolute values ​​of the differences between the predicted active power change and the predicted reactive power change to obtain the power balance deviation.

[0046] Specifically, based on the candidate load status, the corresponding active power change template, reactive power change template, and power supply branch identifier are read from the load status feature library, and each change template is associated with the power supply branch identifier to determine the power change corresponding to the candidate load status and its power supply branch.

[0047] The connection path between the power supply branch corresponding to the candidate load state and the location of the collection point is queried based on the power supply topology. Based on this connection path, the power supply branch identifier corresponding to the candidate load state is associated with the location of the collection point. The active power change template and reactive power change template formed from the historical events marked at the collection point are determined as the predicted active power change and predicted reactive power change of the candidate load state at the collection point, respectively. These change templates represent the combined impact of voltage drop and line loss on power changes at the collection point under historical marked operating conditions. They are empirical equivalent power changes under historical operating conditions and are not used as the real-time calculation result of line loss at the current operating moment. When the difference between the current operating condition and the historical marked operating condition increases, the corresponding power balance deviation may increase, and the process will either involve re-identification or no output after comparing the identification matching value.

[0048] The active power change template mapped to the location of the acquisition point is determined as the predicted active power change, and the reactive power change template is determined as the predicted reactive power change. The positive and negative directions of the change templates are retained to represent the power changes formed at the acquisition point location when the load is added and removed, respectively.

[0049] The predicted change in active power is compared with the change in active power extracted from the load change event data segment, and the predicted change in reactive power is compared with the change in reactive power extracted from the load change event data segment. The absolute value of the difference between the two comparison results is obtained, and the two absolute values ​​of the difference are added together to obtain the power balance deviation of the corresponding candidate load state.

[0050] For example, if the predicted active power change and predicted reactive power change mapped to the acquisition point location for a candidate load state are 760 watts and 120 varawatts respectively, and the actual extracted active power change and reactive power change are 740 watts and 135 varawatts respectively, then the values ​​of active power change deviation and reactive power change deviation are 20 and 15 respectively. In this embodiment, the active power change deviation and reactive power change deviation are combined with equally weighted values, and the combination coefficient of both deviation components is 1. Therefore, the sum of their values, 35, is determined as the power balance deviation of the candidate load state. The power balance deviation is a composite distance index composed of the active power change deviation component and the reactive power change deviation component, used for comparison between candidate load states, and is not considered as actual active power, reactive power, or apparent power output. The power balance deviation referred to in this application is a composite deviation index used to evaluate the consistency between the predicted power change and the actual power change at the acquisition point, under the condition that the power supply branch where the candidate load state is located is within the downstream power supply range of the acquisition point. It does not represent the node power flow balance residual or real-time line loss calculated based on line impedance parameters.

[0051] By combining the above steps, the upstream and downstream relationship between the power supply branch and the location of the collection point can be used to form the predicted power change of the candidate load state and obtain the corresponding power balance deviation.

[0052] Further, the calculation of the phase matching deviation of each candidate load state includes: reading the load identifier and phase current effective value change template of each candidate load state from the load state feature library; determining the power supply phase corresponding to each candidate load state according to the load identifier and phase attribution relationship, configuring the phase current effective value change template to the corresponding power supply phase, and obtaining the predicted phase current change characteristics; calculating the sum of the absolute values ​​of the difference between the predicted phase current effective value change and the phase current effective value change of the same power supply phase, and obtaining the phase matching deviation.

[0053] Specifically, based on the candidate load status, the corresponding load identifier and phase current effective value change template are read from the load status feature library, and the load identifier is used as the associated information for querying phase attribution relationships.

[0054] Based on the load identifier, one or more corresponding power supply phases are read from the phase assignment relationship. The load state feature library stores the corresponding phase current RMS value change templates according to the power supply phase. For a load corresponding to one power supply phase, the change in the phase current RMS value change template is configured to the corresponding power supply phase. For a load corresponding to multiple power supply phases, the phase current RMS value change templates of each power supply phase are configured to the corresponding power supply phases according to the corresponding power supply phase information. The configured phase current RMS value changes of each power supply phase are arranged in the order of the power supply phases to form predicted phase current change characteristics, where the positive and negative directions of each phase current RMS value change template correspond to the increase and decrease of the phase current of the corresponding power supply phase, respectively.

[0055] For each candidate load state, the predicted phase current change characteristics are matched with the phase current change characteristics extracted from the load change event data segment according to the power supply phase. Only for the power supply phases associated with the load identifier of the candidate load state in the phase attribution relationship, the absolute value of the difference between the predicted effective value change of the phase current and the actual effective value change of the phase current is calculated. Power supply phases not associated with the load identifier do not participate in the phase matching deviation calculation of the candidate load state.

[0056] The phase matching deviation of the candidate load state is obtained by summing the absolute values ​​of the differences between the corresponding power supply phases. For example, if the predicted effective changes in the phase current of three power supply phases corresponding to a candidate load state are 3.0 A, 2.8 A, and 3.1 A, and the actual effective changes in the phase current are 2.9 A, 2.7 A, and 3.0 A, the resulting phase matching deviation is 0.3 A. This phase matching deviation is used to evaluate the degree of characteristic similarity between the historical effective changes in phase current on the same power supply phase and the current measured effective changes in phase current. Here, "phase" refers to the power supply phase to which the load belongs, not the voltage phase angle or current phase angle. This phase matching deviation is not used to calculate the phasor sum of currents in different branches, nor is it used as the current balance residual corresponding to Kirchhoff's Current Law.

[0057] Through the above steps, the predicted phase current variation characteristics can be formed based on the power supply correspondence of the candidate load state, and the phase matching deviation between the same power supply phases can be obtained.

[0058] Further, the step of generating identification matching values ​​for each candidate load state and determining the candidate load state with the smallest identification matching value as the target load state includes: normalizing the power balance deviation and phase matching deviation based on the power deviation benchmark value and phase deviation benchmark value pre-stored by the edge computing node; performing weighted calculation on the normalized power balance deviation and phase matching deviation according to preset power weight and preset phase weight to obtain the identification matching value for each candidate load state; determining the candidate load state with the smallest identification matching value as the target load state, and comparing the identification matching value corresponding to the target load state with a preset matching threshold.

[0059] Specifically, the power balance deviation and phase matching deviation of each candidate load state are read, and the power deviation benchmark value and phase deviation benchmark value pre-stored by the edge computing node are also read. For each historical marked event in the target power supply area, based on the actual load state recorded in the historical marked event, the corresponding active power change template, reactive power change template, and effective value change template of each power supply phase current are read. According to the aforementioned calculation method for power balance deviation, the power balance deviation corresponding to the actual load state in each historical marked event is calculated, and the arithmetic mean of the obtained power balance deviation is calculated. The arithmetic mean is determined as the power deviation benchmark value. According to the aforementioned calculation method for phase matching deviation, the phase matching deviation corresponding to the actual load state in each historical marked event is calculated, and the arithmetic mean of the obtained phase matching deviation is calculated. The arithmetic mean is determined as the phase deviation benchmark value. The power deviation benchmark value and the power balance deviation use the same composite numerical scale. The phase deviation benchmark value and the phase matching deviation are both expressed in amperes. The two arithmetic means obtained represent the average change scale of the corresponding deviation in the historical marked events.

[0060] When both the power deviation reference value and the phase deviation reference value are greater than 0, the power balance deviation is compared with the power deviation reference value to obtain the normalized power balance deviation, and the phase matching deviation is compared with the phase deviation reference value to obtain the normalized phase matching deviation. This converts the two deviations of different dimensions into values ​​that can participate in subsequent weighted calculations. When either the power deviation reference value or the phase deviation reference value is not greater than 0, no ratio processing is performed, no identification matching value corresponding to the current load change event is generated, no load category, operating status, or power consumption is output, and the load composition information of the target power supply area is not updated.

[0061] In this embodiment, both the preset power weight and the preset phase weight are dimensionless values ​​that are not less than 0 and not greater than 1, and their sum is 1. The number of decimal places retained in the weight data is the fixed storage precision of the weight data field in the edge computing node, which is determined before generating candidate weight combinations and remains unchanged during the weight search process in the same target power supply area. The edge computing node reads the number of decimal places retained in the weight data, determines the smallest decimal increment corresponding to the number of decimal places as the weight search step size, generates power weight values ​​sequentially according to the weight search step size with 0 as the starting value and 1 as the ending value, and sets the phase weight corresponding to each power weight to 1 minus the power weight; for each weight group, load status identification is performed using historical labeled events in the target power supply area, and the number of events in which the identified load identifier and operating status are the same as the load identifier and post-switching operating status recorded in the corresponding historical labeled events is counted, and the weight combination with the most consistent events is determined as the candidate weight combination; when there are multiple candidate weight combinations, the weight combination with the smallest standard deviation of the identification matching value corresponding to the correctly identified event is selected to determine the preset power weight and the preset phase weight.

[0062] The normalized power balance deviation and normalized phase matching deviation are weighted according to preset power weights and preset phase weights, respectively, and the two weighted results are added together to obtain the identification matching value for each candidate load state. For example, if the normalized power balance deviation and normalized phase matching deviation of a candidate load state are 0.20 and 0.10, respectively, and the preset power weight and preset phase weight are 0.6 and 0.4, respectively, the identification matching value is 0.20×0.6+0.10×0.4=0.16; the normalized power balance deviation, normalized phase matching deviation, preset power weight, preset phase weight, and identification matching value are all dimensionless values.

[0063] The identification matching values ​​of each candidate load state are compared. These values ​​characterize the overall deviation between the predicted power and phase current changes corresponding to the candidate load state and the actual power and phase current changes. A smaller identification matching value indicates a smaller overall deviation. The candidate load state with the smallest identification matching value is determined as the target load state. When two or more candidate load states have the same minimum identification matching value and cannot be further distinguished based on the corresponding power supply phase and change template, the load state identification result for this instance is not output, the load composition information of the target power supply area is not updated, and the current load state identification process ends.

[0064] The preset matching threshold is determined based on the initial identification results of historically labeled events. The identification matching values ​​corresponding to each historically labeled event are sequentially used as candidate matching thresholds. For each candidate matching threshold, the number of events with correct identification results and identification matching values ​​not exceeding the candidate matching threshold, and the number of events with incorrect identification results and identification matching values ​​exceeding the candidate matching threshold, are counted. The candidate matching threshold with the largest sum of these two types of event counts is determined as the preset matching threshold. If multiple candidate matching thresholds have the same sum of the two types of event counts, the candidate matching threshold with the smaller value is selected to reduce the direct output of erroneous identification results. The identification matching value corresponding to the target load state is compared with the preset matching threshold to form a judgment result for either directly outputting the identification result or re-identifying by calling the original electrical quantity data.

[0065] Through the above steps, power balance deviation and phase matching deviation can be evaluated in a unified manner, and the target load state with the smallest identification matching value can be determined from the candidate load states.

[0066] Further, when the identification matching value corresponding to the target load state is greater than the preset matching threshold, the target load state is re-determined, including: extracting the original electrical quantity data of the corresponding time period and the time periods before and after the event interval from the original synchronous electrical quantity sequence retained in a cyclic manner according to the sampling time of the event interval; re-extracting the power change characteristics and phase current change characteristics according to the original electrical quantity data of the corresponding time period and the time periods before and after the event interval, generating candidate load states, calculating the power balance deviation and phase matching deviation, generating the identification matching value and determining the target load state; when the identification matching value corresponding to the re-determined target load state is not greater than the preset matching threshold, reading the load category, operating status and status power value from the load status feature library, determining the status power value as the power consumption, outputting the load category, operating status and power consumption, and updating the load composition information of the target power supply area.

[0067] Specifically, when the identification matching value corresponding to the target load state determined based on the aggregated electrical quantity data is greater than the preset matching threshold, the start sampling time and end sampling time of the event interval corresponding to the load change event are read, and the sampling time is used as the time index for querying the original synchronous electrical quantity sequence.

[0068] Based on the time index, active power data, reactive power data, and effective value data of phase current of each power supply phase are read from the original synchronous electrical quantity sequence that is cyclically retained. At the same time, the original electrical quantity data of adjacent times before and after the event interval are also read to form the original load change event data segment arranged according to the sampling time.

[0069] Based on the active power, reactive power, and effective values ​​of phase current of each power supply phase before and after the event interval in the original load change event data segment, the changes in active power, reactive power, and effective values ​​of phase current of each power supply phase are re-determined, and power change characteristics and phase current change characteristics are formed respectively.

[0070] The re-extracted power change features and phase current change features are compared with the corresponding change templates in the load state feature library to generate candidate load states. The power balance deviation and phase matching deviation of each candidate load state are then recalculated according to the power supply topology and phase assignment relationships.

[0071] The recalculated power balance deviation and phase matching deviation are normalized and weighted to generate identification matching values ​​for each candidate load state. The candidate load state with the smallest identification matching value is then redefined as the target load state.

[0072] When the identification matching value corresponding to the redefined target load state is not greater than the preset matching threshold, the corresponding load category, operating status and status power value are read from the load state feature library according to the target load state. For example, when the regenerated identification matching value is 0.18 and the preset matching threshold is 0.25, the corresponding load state is determined as the target load state that meets the matching conditions.

[0073] The read state power value is determined as the power consumption of the target load state, and the load category, operating status, and power consumption are output. Using the load identifier corresponding to the target load state as an index, the load composition information of the target power supply area is queried. If the load identifier already exists in the load composition information, the original operating status and state power value of the load identifier are replaced with the operating status and state power value obtained in this identification. If the load identifier does not exist in the load composition information, a new record of the load category, operating status, and state power value corresponding to the load identifier is added. The state power value corresponding to the stopped operating state is 0, and historical state power values ​​for the same load identifier are not repeatedly accumulated. The power consumption represents the representative active power of the corresponding load state formed based on historical annotation events, used to record the power composition corresponding to each load state in the target power supply area, and does not replace the total active power of the target power supply area collected by the edge computing node. When the identification matching value corresponding to the re-determined target load state is still greater than the preset matching threshold, the load category, operating status, and power consumption are not output, the load composition information of the target power supply area is not updated, and the current load state identification ends. If multiple loads switch on and off simultaneously within the same event interval, and the extracted comprehensive power change characteristics and phase current change characteristics cannot match the single load state in the load state feature library, no identification results will be output in the manner described above.

[0074] By following the steps above, if the initial identification result does not meet the matching conditions, the retained original electrical quantity data can be used for re-identification, and the load status information can be output when the re-identification result meets the matching conditions.

[0075] First, the raw electrical quantity data of the target power supply area is continuously collected, cyclically retained, and aggregated over time. Then, load change events are determined based on the aggregated electrical quantity data, power change characteristics and phase current change characteristics are extracted, and candidate load states are generated. Subsequently, power balance and phase matching verifications are performed on each candidate load state in conjunction with the power supply topology and phase attribution relationships, and the target load state is determined by identifying the matching value. When the aggregated electrical quantity data has determined the load change event and the identified matching value is greater than the preset matching threshold, the raw electrical quantity data of the corresponding time period and the time periods before and after the load change event are further retrieved. The average electrical quantity and load change characteristics before and after the event interval are re-determined to reduce the quantitative impact of time aggregation on the event interval boundary and average value calculation of the detected events, while avoiding continuous processing of high-temporal-resolution raw electrical quantity data.

[0076] Example 2: Furthermore, before calculating the active power difference and reactive power difference between adjacent sampling times in the aggregated electrical quantity data, the effective value difference of the bus voltage at adjacent sampling times is obtained, and the effective value difference of the bus voltage is compared with the effective value of the reference voltage to obtain the voltage fluctuation rate.

[0077] Specifically, this embodiment is applicable to scenarios where the aggregated loads in the target power supply area mainly exhibit constant impedance characteristics during bus voltage fluctuations, and the effective value of the bus voltage used can characterize the overall voltage change of each power supply phase in the target power supply area. If any applicable condition is not met, the voltage fluctuation compensation of this embodiment will not be activated. The load characteristics are confirmed based on the nameplate parameters of known loads within the target power supply area. If the nameplate parameters are insufficient for confirmation, the adjacent sampling times where the absolute value of the voltage fluctuation rate is greater than the voltage warning threshold during historical stable operation are read. The active and reactive power at the previous sampling time are calculated by squared the ratio of the effective value of the bus voltage at the later sampling time to the effective value of the bus voltage at the previous sampling time, resulting in the predicted active and reactive power at the later sampling time. The actual active and reactive power at the later sampling time are subtracted from the corresponding predicted active and reactive power to obtain the corrected active power difference and reactive power difference, and the corrected event detection quantity is obtained according to the aforementioned calculation method for event detection quantities. When the corrected event detection quantity corresponding to each adjacent sampling time within the historical stable operating period is not greater than the event judgment threshold, it is confirmed that the aggregated load of the target power supply area mainly exhibits constant impedance characteristics during bus voltage fluctuations; when there is a corrected event detection quantity greater than the event judgment threshold, it is not confirmed that the aggregated load mainly exhibits constant impedance characteristics. When the above applicable conditions are met, when the absolute value of the voltage fluctuation rate is not greater than the voltage warning threshold, the active power difference and reactive power difference between adjacent sampling times are directly calculated; when the absolute value of the voltage fluctuation rate is greater than the voltage warning threshold, and the effective value of the bus voltage at the previous sampling time is greater than 0, the ratio of the effective value of the bus voltage at the current sampling time to the effective value of the bus voltage at the previous sampling time is calculated, and the square of the obtained ratio is determined as the voltage conversion ratio; the active power at the previous sampling time is multiplied by the voltage conversion ratio to obtain the predicted active power corresponding to the current sampling time, and the reactive power at the previous sampling time is multiplied by the voltage conversion ratio to obtain the predicted reactive power corresponding to the current sampling time. If the above applicable conditions are not met, the voltage fluctuation compensation of this embodiment will not be enabled, and the active power difference and reactive power difference between adjacent sampling times will be calculated directly.

[0078] The corrected active power difference is obtained by subtracting the predicted active power from the actual active power at the current sampling time; the corrected reactive power difference is obtained by subtracting the predicted reactive power from the actual reactive power at the current sampling time. These corrected active power difference and reactive power difference are then used for event detection quantity calculation. Let the effective values ​​of the bus voltage at the previous sampling time and the current sampling time be respectively... and The active power and reactive power at the previous sampling time were respectively and The predicted active power is The predicted reactive power is The voltage warning threshold is determined based on the voltage fluctuation range during historically stable operating periods in the target power supply area.

[0079] By taking the above steps, the interference of power changes caused by bus voltage dips or rises on load change event detection can be reduced, and the possibility of false events occurring when the load operating status has not changed can be reduced.

[0080] Furthermore, before performing weighted calculations on the normalized power balance deviation and phase matching deviation according to preset power weights and preset phase weights, the method further includes: calculating the average value of the effective value of the phase current of each power supply phase in the sampled data before the event interval, determining the difference between the maximum and minimum values ​​of the corresponding average values ​​of each power supply phase as the range, and determining the ratio of the range to the arithmetic mean of the corresponding average values ​​of each power supply phase as the three-phase imbalance degree. In this embodiment, the three-phase imbalance degree refers to the phase current imbalance index constructed by the ratio of the range to the arithmetic mean of the average values ​​of the effective values ​​of the phase currents of each power supply phase, and does not represent the standard three-phase imbalance degree calculated based on negative sequence current, negative sequence voltage, or maximum phase deviation.

[0081] Based on the three-phase imbalance degree corresponding to each time range during the historical stable operation period of the target power supply area, the maximum value is determined as the imbalance threshold. When the current three-phase imbalance degree is greater than the imbalance threshold, the ratio of the imbalance threshold to the current three-phase imbalance degree is determined as the weight adjustment coefficient. The product of the preset phase weight and the weight adjustment coefficient is determined as the adjusted phase weight, and the reduction in phase weight is added to the preset power weight, so that the sum of the adjusted power weight and the phase weight remains unchanged. When the current three-phase imbalance degree is not greater than the imbalance threshold, the preset power weight and the preset phase weight remain unchanged. When the arithmetic mean of the corresponding average values ​​of each power supply phase is not greater than 0, the three-phase imbalance degree is not calculated, and the preset power weight and the preset phase weight remain unchanged.

[0082] By following the steps above, the weight of phase matching deviation in the identification matching value can be reduced according to the degree of three-phase imbalance before the event occurs, and the weight of power balance deviation can be increased accordingly, thereby reducing the impact of background phase current imbalance on the phase matching result.

[0083] Furthermore, before retrieving the original electrical quantity data, the current memory occupancy rate of the edge computing node is detected. The edge computing node predetermines an upper limit of memory capacity allocated to this method, and the ratio of the current memory capacity occupied by this method to the upper limit is determined as the current memory occupancy rate. Based on the additional memory occupancy generated when re-identifying according to the original sampling frequency based on historical load change events, the maximum value is determined as the memory capacity required for full sampling re-identification, and the ratio of the remaining capacity after deducting the memory capacity required for full sampling re-identification from the upper limit to the upper limit is determined as the load threshold.

[0084] The integer step size used for downsampling is determined based on historically labeled events. For each integer step size that allows sampling records to be retained before, during, and after an event interval, continuous original synchronous electrical quantity records are divided into continuous data groups according to the corresponding integer step size within the data range before, during, and after the event interval. The average values ​​of active power, reactive power, and effective phase current of each power supply phase are calculated within each continuous data group to form downsampled synchronous electrical quantity records. Power change characteristics and phase current change characteristics are then re-extracted from these records. Integer step sizes that retain the redefined target load state consistent with the redefined target load state according to the original sampling frequency, and whose corresponding matching value is not greater than a preset matching threshold, are retained. The largest integer step size among these is determined as the downsampling step size. If no integer step size greater than 1 and satisfying the above conditions exists, the downsampling process of this embodiment is not used. The downsampling step size is only used for load change events whose duration is not shorter than the shortest duration of the historically labeled event used to determine the downsampling step size. When the duration of the current load change event is shorter than the shortest duration of the historical labeled event, downsampling is not used; when the current memory usage is insufficient to support re-identification at the original sampling frequency, this re-identification is not performed, the load status identification result is not output, and the load composition information of the target power supply area is not updated.

[0085] When the current memory occupancy rate is greater than the load threshold and there is a downsampling step size greater than 1, the corresponding continuous original synchronous electrical quantity records are divided into continuous data groups according to the downsampling step size before, within, and after the event interval of the load change event. The average value of active power, reactive power, and the effective value of phase current of each power supply phase are calculated in each continuous data group to form downsampled electrical quantity data. The power change characteristics and phase current change characteristics are then re-extracted using the downsampled electrical quantity data. When the current memory occupancy rate is not greater than the load threshold, the original electrical quantity data is retrieved according to the original sampling frequency.

[0086] The above steps can reduce the amount of data that needs to be processed for re-identification when the memory load of the edge computing node is high. By grouping and averaging the continuous electrical quantity records, the impact of short-term fluctuation information folding caused by direct interval extraction can be reduced, providing downsampled data for subsequent re-extraction of power change characteristics and phase current change characteristics.

[0087] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A non-intrusive power load identification method based on edge computing on the power grid side, characterized in that, include: The original electrical quantity data of the target power supply area is continuously collected and retained in a cyclical overlay manner. The original electrical quantity data is aggregated over time to obtain aggregated electrical quantity data, and the power supply topology and phase attribution relationships are obtained. Based on aggregated electrical quantity data, load change events are identified, power change characteristics and phase current change characteristics are extracted, and at least one candidate load state is generated. Based on the power supply topology and phase assignment, the power balance deviation and phase matching deviation of each candidate load state are calculated, and the identification matching value of each candidate load state is generated. The candidate load state with the smallest identification matching value is determined as the target load state. When the identification matching value corresponding to the target load state is greater than the preset matching threshold, the original electrical quantity data of the time period corresponding to the retained load change event and the time periods before and after are retrieved, the power change characteristics and phase current change characteristics are extracted again, the candidate load state and identification matching value are generated, and the target load state is determined. When the identification matching value corresponding to the target load status is not greater than the preset matching threshold, the load category, operating status and power consumption of the target load status are output and the load composition information of the target power supply area is updated.

2. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 1, characterized in that, The process of continuously collecting and retaining the original electrical quantity data of the target power supply area in a cyclical coverage manner, performing time aggregation on the original electrical quantity data to obtain aggregated electrical quantity data, and obtaining power supply topology and phase attribution relationships includes: continuously collecting active power data, reactive power data, and effective value data of phase current of each power supply phase through an edge computing node set at one of the low-voltage side and feeder side of the distribution transformer to form original electrical quantity data; aligning the original electrical quantity data according to the sampling time to obtain the original synchronous electrical quantity sequence; cyclically retaining the original synchronous electrical quantity sequence and performing time aggregation on the original synchronous electrical quantity sequence to obtain aggregated electrical quantity data; generating power supply topology relationships based on the power supply branch connection information and collection point location information pre-stored by the edge computing node; and generating phase attribution relationships based on the load identifier, load access branch information, and corresponding power supply information pre-stored by the edge computing node.

3. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 2, characterized in that, The step of determining load change events based on aggregated electrical quantity data includes: calculating the active power difference and reactive power difference between adjacent sampling times in the aggregated electrical quantity data; determining the sum of the absolute values ​​of the active power difference and the reactive power difference as the event detection quantity; determining the continuous sampling interval where the event detection quantity is greater than the event judgment threshold as the event interval; and extracting the sampled data before the event interval, the sampled data within the event interval, and the sampled data after the event interval to obtain the load change event data segment.

4. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 3, characterized in that, The extraction of power change characteristics and phase current change characteristics to generate at least one candidate load state includes: calculating the average active power, average reactive power, and average effective value of phase current of each power supply phase before and after the event interval in the load change event data segment; subtracting the average before the event interval from the average after the event interval to obtain the change in active power, the change in reactive power, and the change in effective value of phase current of each power supply phase; generating power change characteristics based on the change in active power and the change in reactive power; generating phase current change characteristics based on the change in effective value of phase current of each power supply phase; comparing the change in active power, the change in reactive power, and the change in effective value of phase current of the power supply phase corresponding to each load state with the active power change template, the reactive power change template, and the effective value of phase current change template in the load state feature library, respectively; and determining the load state where the absolute value of the difference between each change and the corresponding change template is within the corresponding allowable deviation range as the candidate load state.

5. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 4, characterized in that, The calculation of the power balance deviation for each candidate load state includes: reading the active power change template, reactive power change template, and power supply branch identifier for each candidate load state from the load state feature library; determining the active power change template and reactive power change template as the predicted active power change and predicted reactive power change corresponding to the location of the collection point based on the upstream and downstream relationship between the power supply branch and the collection point location; and calculating the sum of the absolute values ​​of the differences between the predicted active power change and the predicted reactive power change to obtain the power balance deviation.

6. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 5, characterized in that, The calculation of the phase matching deviation for each candidate load state includes: reading the load identifier and phase current effective value change template of each candidate load state from the load state feature library; determining the power supply phase corresponding to each candidate load state according to the load identifier and phase attribution relationship, configuring the phase current effective value change template to the corresponding power supply phase, and obtaining the predicted phase current change characteristics; calculating the sum of the absolute values ​​of the difference between the predicted phase current effective value change and the phase current effective value change of the same power supply phase, and obtaining the phase matching deviation.

7. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 6, characterized in that, The process of generating identification matching values ​​for each candidate load state includes: normalizing the power balance deviation and phase matching deviation based on the power deviation benchmark value and phase deviation benchmark value pre-stored by the edge computing node; performing weighted calculation on the normalized power balance deviation and phase matching deviation according to preset power weight and preset phase weight to obtain the identification matching value for each candidate load state; determining the candidate load state with the smallest identification matching value as the target load state, and comparing the identification matching value corresponding to the target load state with a preset matching threshold.

8. The non-intrusive power load identification method based on edge computing on the power grid side according to claim 3, characterized in that, When the identification matching value corresponding to the target load state is greater than the preset matching threshold, the target load state is re-determined, including: extracting the original electrical quantity data of the corresponding time period and the time periods before and after the event interval from the original synchronous electrical quantity sequence retained in a loop according to the sampling time of the event interval; re-extracting the power change characteristics and phase current change characteristics according to the original electrical quantity data of the corresponding time period and the time periods before and after the event interval, generating candidate load states, calculating the power balance deviation and phase matching deviation, generating the identification matching value and determining the target load state; when the identification matching value corresponding to the re-determined target load state is not greater than the preset matching threshold, reading the load category, operating status and status power value from the load status feature library, determining the status power value as the power consumption, outputting the load category, operating status and power consumption, and updating the load composition information of the target power supply area.

Images