Computer system based on electric power Internet of Things cloud edge computing power cooperative computing model

Abstract

The invention discloses a computer system based on an electric power internet of things cloud edge computing power cooperative computing model, which is suitable for monitoring an electric energy quality disturbance event, and comprises an edge side time period determination module, an edge side frequency domain analysis module, a cloud end event classification module, a cloud end source node determination module and an edge side monitoring module, wherein the edge side time period determination module is used for determining a disturbance event time period through a machine learning model; the edge side frequency domain analysis module is used for obtaining abnormal analysis data through calculation; the cloud event classification module is used for determining a disturbance event type through a K-means clustering algorithm; the cloud source node determination module is used for determining an actual source node through a preset graph neural network; and the edge side monitoring module is used for establishing a power grid disturbance event early warning monitoring point position layout for the actual source node. According to the invention, the real-time performance and accuracy of power quality disturbance event monitoring can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, and particularly relates to a computer system based on a cloud-edge computing power collaborative computing model of power internet of things. BACKGROUND

[0002] In the power internet of things environment, the rapid switching of loads, the grid connection of new energy and the large number of access of various nonlinear devices make the power quality indicators such as grid voltage, frequency and harmonic continuously change dynamically. Once the disturbance event (voltage sag, frequency deviation, harmonic exceeding, etc.) is not discovered and traced in time, it may cause sensitive equipment malfunction, production line shutdown, even cause chain failure, expand the power outage range, and cause significant economic loss and social impact. Therefore, it is necessary to monitor the power quality disturbance event in real time and accurately, so as to locate the disturbance source, evaluate the disturbance type and evolution trend in the shortest time, and optimize the early warning distribution and operation control strategy accordingly, and ensure the safe, high-quality and economic operation of the power grid.

[0003] The prior art usually adopts the method of edge device monitoring or cloud deep analysis, but the edge device has limited computing power and can only complete simple threshold exceeding detection or short-time Fourier transform, and it is difficult to complete high-precision and wide-band frequency domain deep analysis locally. Although the cloud has strong computing power, it is limited by transmission delay and bandwidth, and cannot respond to instantaneous disturbance in real time, resulting in missed reports or delay. In addition, the traditional method uses a fixed length analysis window, which cannot be dynamically adjusted according to the disturbance duration, and is easy to truncate the waveform and lose the characteristics. In terms of historical data utilization, only the threshold or waveform similarity is simply compared, and there is a lack of modeling ability for disturbance type evolution and regional migration, and it is difficult to quickly reposition when the disturbance source position changes.

[0004] Therefore, it is urgent to invent a cloud-edge computing power collaborative computing model that fuses the instantaneous characteristics and long-term evolution trend of the disturbance event to realize the rapid positioning and accurate identification of the disturbance source. SUMMARY

[0005] The present application provides a computer system based on a cloud-edge computing power collaborative computing model of power internet of things, which can improve the real-time performance and accuracy of power quality disturbance event monitoring through cloud-edge computing power collaboration.

[0006] In the first aspect, the present application provides a computer system based on a cloud-edge computing power collaborative computing model of power internet of things, which is suitable for power quality disturbance event monitoring, and the computer system is integrated with an edge side time period determination module, an edge side frequency domain analysis module, a cloud end event classification module, a cloud end source node determination module and an edge side monitoring module, wherein, The edge side time period determination module is configured to calculate a real-time disturbance intensity level according to a first voltage signal sequence and a first frequency signal sequence of the target power grid collected in real time by using a preset machine learning model, and determine a disturbance event time period according to the real-time disturbance intensity level when the real-time disturbance intensity level exceeds a preset intensity threshold. The edge side frequency domain analysis module is configured to adjust a width of an analysis window according to the disturbance event time period, intercept a second voltage signal sequence and a second frequency signal sequence by using the analysis window, and perform frequency domain feature extraction according to the second voltage signal sequence and the second frequency signal sequence to obtain abnormal analysis data. The cloud event classification module is configured to calculate a first similarity between the abnormal analysis data and each historical disturbance record in a historical disturbance record set stored in advance, obtain a first similarity, and filter the historical disturbance record set according to the first similarity to obtain a historical disturbance record subset, and determine a disturbance event type according to the historical disturbance record subset and the abnormal analysis data by using a preset K-means clustering algorithm. The cloud source node determination module is configured to obtain a candidate node list of a disturbance source by using a preset graph neural network according to target power grid topology data stored in advance and the disturbance event type, calculate a second similarity between each candidate node and a historical source node stored in advance one by one, and determine an actual source node according to the second similarity, wherein the historical source node and each candidate node are of the same disturbance event type. The edge side monitoring module is configured to perform real-time frequency monitoring on the actual source node to obtain a real-time frequency value sequence, generate a disturbance event identification label according to the real-time frequency value sequence, and establish a power grid disturbance event early warning monitoring point layout according to the disturbance event identification label.

[0007] The embodiment of the present application determines the edge side time period through the edge side time period determination module, collects the voltage and frequency signals in real time, calculates the disturbance intensity level through the machine learning model, quantifies the disturbance intensity through the machine learning model first, filters out slight fluctuations, avoids invalid data processing at the edge, reduces the algorithm consumption, quickly identifies the time window of the disturbance event through the advantage of fast transmission speed of the edge device, limits the effective analysis window for subsequent frequency domain analysis, avoids the distortion of characteristics caused by mixing a large amount of normal data in the frequency domain analysis, and is the time reference for all subsequent analyses; through the edge side frequency domain analysis module, the window width is dynamically adjusted to ensure that the complete disturbance waveform is covered, the problems of "missing disturbance" or "over-sampling noise" of the fixed window are solved, and the time domain signal is converted into quantifiable frequency domain features through frequency domain extraction, the "abnormal analysis data" is generated by combining the abnormal identifier, the feature carrier is provided for cloud classification, and the core value of real-time preliminary analysis at the edge is realized; through the cloud event classification module, the cloud algorithm is used to process massive historical data, the similarity filtering and K-means clustering are used to solve the problem that the edge algorithm is insufficient to perform deep analysis, and the disturbance type is accurately classified to provide a type basis for subsequent source positioning; through the cloud source node determination module, the power grid topology and the disturbance type are combined, the candidate nodes are quickly locked through the graph neural network reverse tracing, the blind investigation of the whole power grid is avoided, and the actual source is accurately selected from the candidates through the second similarity, the positioning difficulty caused by disturbance migration is solved, and the positioning rate of the disturbance source is improved; the source that has been positioned is continuously monitored through the edge side monitoring module, the identification label is generated, the support is provided for the disturbance full information record, the early warning point position covering the source and the propagation path is established according to the label, the closed loop optimization is formed, similar disturbances in the future can be quickly found through the early warning point position, the response speed is improved, and the value of one-time positioning and long-term early warning is realized. Compared with the prior art, the present application can improve the real-time performance and accuracy of power quality disturbance event monitoring through the cloud edge algorithm cooperation.

[0008] Further, the first voltage signal sequence and the first frequency signal sequence of the target power grid are collected in real time, and the real-time disturbance intensity level is calculated through a preset machine learning model, specifically: The first voltage signal sequence and the first frequency signal sequence of a plurality of sampling points in the target power grid are obtained in real time through a preset sliding time window; The difference between the real-time voltage effective value and the rated voltage value of each sampling point is calculated in sequence according to the first voltage signal sequence, and a voltage amplitude deviation sequence is obtained; The difference between the real-time frequency value and the standard frequency value of each sampling point is calculated in sequence according to the first frequency signal sequence, and a frequency offset sequence is obtained; According to the voltage amplitude deviation sequence, the frequency offset sequence, the preset deviation anomaly threshold and the preset offset anomaly threshold, the number of abnormal sampling points is counted to obtain an abnormal statistical result, and a preset neural network model is used to calculate a real-time disturbance intensity level according to the abnormal statistical result.

[0009] The embodiment of the application realizes the "continuous real-time sampling" of the voltage and frequency signals by collecting signals through a sliding time window, ensures that no transient disturbance is missed, and provides a continuous data basis for subsequent deviation calculation; the time domain signals are converted into quantitative deviation indicators by calculating the voltage amplitude deviation sequence and the frequency offset sequence, avoiding qualitative and fuzzy judgment of the disturbance, and providing a quantitative basis for abnormal statistics; the deviation sequence is converted into abnormal degree data by abnormal sampling point statistics, avoiding misjudgment of a single sampling point, and the neural network is good at processing multi-dimensional data, so that the calculated intensity level is more accurate and the mis-triggering of subsequent actions is reduced.

[0010] Further, the real-time disturbance intensity level is used to determine a disturbance event time period, specifically: Difference operation is performed on the voltage amplitude deviation sequence of each sampling point to obtain a deviation change rate, and the sampling point with the deviation change rate exceeding a preset change rate threshold is taken as a mutation point; The voltage amplitude deviation sequence corresponding to the mutation point is used to identify a disturbance starting point and a disturbance ending point, and the maximum value of the voltage amplitude deviation sequence and the peak value of the frequency offset sequence between the disturbance starting point and the disturbance ending point are extracted; The maximum value and the peak value are weighted and summed according to the real-time disturbance intensity level and a preset weight coefficient to obtain a comprehensive disturbance index, and if the comprehensive disturbance index is greater than a preset comprehensive disturbance threshold, the disturbance event time period is determined according to the disturbance starting point and the disturbance ending point.

[0011] The embodiment of the application accurately locates the starting point and the ending point of the disturbance by calculating the deviation change rate, avoids the inclusion of too much normal data in the time period due to inaccurate starting and ending point judgment, reduces the calculation amount while improving the accuracy; the comprehensive disturbance index is obtained by extracting the maximum value of the voltage and the peak value of the frequency, wherein the maximum value and the peak value are the core indicators of the intensity of the disturbance, reflecting the most serious degree of the disturbance, and the weighted sum comprehensively considers the influence of the voltage and the frequency, avoiding the judgment deviation caused by a single indicator, and generating a more comprehensive quantitative value of the severity of the disturbance; the time period is determined according to the comprehensive index exceeding the threshold, further filtering disturbances with insufficient intensity, ensuring that the finally determined time period is a valid disturbance that needs to be processed, avoiding the waste of computing power on invalid data in subsequent frequency domain analysis, and improving the accuracy of disturbance identification.

[0012] Further, the second voltage signal sequence and the second frequency signal sequence are used to perform frequency domain feature extraction to obtain abnormal analysis data, specifically: The second voltage signal sequence is subjected to fast Fourier transform to obtain a frequency domain amplitude spectrum, and each harmonic component amplitude is extracted according to the frequency domain amplitude spectrum, so that the percentage of each harmonic content is calculated according to the each harmonic component amplitude; The second frequency signal sequence is subjected to fast Fourier transform to extract frequency components, and the component with the largest amplitude in the frequency components is taken as a dominant frequency component; The percentage of each harmonic content and the pre-acquired target power grid phase data are used to extract a harmonic feature vector, and the difference between the dominant frequency component and the preset power grid rated frequency is calculated to obtain an actual frequency offset value, so that the harmonic feature vector and the actual frequency offset value are integrated to obtain a frequency domain feature set; If the percentage of harmonic content in the harmonic feature vector exceeds a preset alarm threshold, or the actual frequency offset value exceeds a frequency deviation limit value, it is determined that a frequency abnormal event occurs and an abnormal identifier is generated, and the frequency domain feature set and the abnormal identifier are integrated to obtain abnormal analysis data.

[0013] In the embodiment of the present application, the percentage of harmonic content is extracted by voltage signal FFT (fast Fourier transform), and the dominant frequency component is extracted by frequency signal FFT (fast Fourier transform). FFT (fast Fourier transform) is a core method of frequency domain analysis, which converts time domain signals into frequency domain spectrum, and extracts harmonic content to determine whether the disturbance is harmonic pollution. The dominant frequency component reflects the main source of frequency offset, and the percentage of harmonic content and the dominant frequency are the core frequency domain features of the disturbance. Different disturbance types correspond to different harmonic combinations, which provide differentiated features for cloud classification. By constructing a harmonic feature vector and calculating an actual frequency offset value, discrete harmonic parameters and frequency offset are quantized into a structured feature vector, which facilitates similarity comparison and clustering analysis in the cloud. By setting a threshold to generate an abnormal identifier and integrating the abnormal analysis data, it is determined whether it is a frequency abnormal event, so that the cloud classification module can directly focus on abnormal data without the need to judge data validity again, improving the efficiency of cloud processing. The integrated abnormal analysis data contains frequency domain features and abnormal identifiers, which provides a complete feature package for cloud classification.

[0014] Further, the similarity between the abnormal analysis data and each historical disturbance record in the pre-stored historical disturbance record set is calculated to obtain a first similarity, specifically: The similarity between the abnormal analysis data and each historical disturbance record is calculated in sequence by a dynamic time warping algorithm to obtain a first similarity.

[0015] The embodiment of the application calculates the first similarity through a dynamic time warping (DTW) algorithm, the power grid disturbance signal is time series data, and there may be length or rhythm differences, and the DTW is good at processing nonlinear alignment of time series, and can accurately calculate the similarity even if the lengths of two sequences are different, avoids the too-low similarity caused by the length difference, and ensures that the selected historical disturbance record subset is highly related to the current disturbance.

[0016] Further, the disturbance event type is determined through a preset K-means clustering algorithm according to the historical disturbance record subset and the abnormal analysis data, specifically: The historical disturbance record subset is subjected to a difference operation to extract disturbance change rules and disturbance periodicity characteristics. The historical disturbance record subset and the abnormal analysis data are clustered and grouped through a preset K-means clustering algorithm according to the disturbance change rules and the disturbance periodicity characteristics, a grouping type is obtained, and the disturbance event type is determined according to the grouping type and a preset disturbance classification standard.

[0017] The embodiment of the application extracts dynamic characteristics and periodicity characteristics of the disturbance through difference calculation, and these characteristics are key to distinguishing the disturbance type and provide differentiated basis for clustering; similar historical records and current data are grouped through K-means clustering, the natural categories in the data are automatically identified (without manual labeling), the problem of manual classification of massive historical data is solved, and the clustering result is converted into an explicit disturbance type identifier in combination with a preset classification standard, so that the problem that the clustering result cannot correspond to an actual disturbance scene is avoided, and a closed loop from data grouping to type landing is realized.

[0018] Further, the candidate node list of the disturbance source is obtained through a preset graph neural network according to the pre-stored target power grid topology data and the disturbance event type, specifically: The electrical parameters of each node in the target power grid are obtained, and a weighted directed graph containing the electrical parameters is constructed according to the electrical parameters, the pre-stored target power grid topology data and the disturbance event type. The candidate node list of the disturbance source is obtained by taking the corresponding edge side node of the detected disturbance event as a starting point and performing source tracing and propagation on the weighted directed graph through a preset graph neural network.

[0019] The embodiment of the application abstracts the power grid as a weighted graph by integrating power grid topology data, electrical parameters and disturbance types, enables the graph neural network to trace based on electrical rules rather than only based on physical connection, and improves the rationality of tracing; the candidate node list is obtained by reverse tracing from the edge monitoring point, the characteristics of the graph neural network are used to reverse propagate along the electrical connection path, the nodes that may cause disturbance are automatically identified, and the candidate node list is generated, so that the source positioning range is reduced from the whole grid node to several candidate nodes to avoid blind investigation and improve the positioning speed.

[0020] Further, the second similarity between each candidate node and the pre-stored historical source node is calculated, and the actual source node is determined according to the second similarity, specifically: The electrical parameters and load characteristics of each candidate node are extracted from the candidate node list in turn, and the cosine similarity between the electrical parameters and load characteristics of each candidate node and the pre-stored historical source node is calculated to obtain the second similarity. If the second similarity exceeds the preset matching threshold, the corresponding candidate node is determined as the actual source node.

[0021] The embodiment of the application extracts the electrical parameters and load characteristics of the candidate node, calculates the cosine similarity, wherein the electrical parameters and load characteristics are the essential characteristics of the source node, and the cosine similarity is used to compare the feature vectors of the candidate node and the historical source node to quantify the matching degree of the two, so as to avoid subjective judgment of whether the node is the source; the actual source is determined by the similarity exceeding the threshold, and the node with the most matched characteristics with the historical source is accurately selected from the candidate list, so as to avoid misjudging the nodes on the propagation path as the source node, and ensure the accuracy of source positioning.

[0022] Further, the disturbance event identification label is generated according to the real-time frequency value sequence, specifically: The difference between each real-time frequency value and the preset grid rated frequency is calculated according to the real-time frequency value sequence to obtain a real-time frequency offset value sequence; The offset change rate is calculated according to the real-time frequency offset value sequence, and the historical offset change rate of the source node is obtained to calculate the difference between the offset change rate and the historical offset rate; If the difference is greater than the preset threshold, the strong monitoring mode is started, and the sampling frequency is adjusted to a preset multiple of the original frequency to obtain high sampling rate disturbance data in the strong monitoring mode; The disturbance amplitude, duration and harmonic component feature parameters are extracted from the high sampling rate disturbance data to generate the disturbance event identification label.

[0023] The embodiment of the application verifies whether the current disturbance is consistent with the historical disturbance rule by continuously monitoring the frequency state of the source node and comparing the real-time and historical change rate, avoids false source misjudgment caused by monitoring error, and ensures the reliability of source monitoring; the strong monitoring mode is started by the difference value exceeding the threshold value, the sampling rate data is improved, the strong monitoring mode can capture more detailed disturbance characteristics, solves the problem of missing fast transient process of regular sampling rate, provides high-resolution feature data for the label, integrates the source node, the disturbance type and the feature parameter, extracts the feature parameter, generates the label containing the disturbance information, provides structured data support for cloud database update and early warning layout, and avoids fragmentation of disturbance information.

[0024] Further, the disturbance event identification label is used to establish the power grid disturbance event early warning monitoring point layout, specifically: According to the disturbance event identification label, the early warning monitoring level of the source node is determined, and the source node early warning monitoring point is established according to the early warning monitoring level. According to the disturbance event identification label and the target power grid topology data, the disturbance event propagation path is determined, and the early warning monitoring point is respectively established on each node of the disturbance event propagation path.

[0025] The embodiment of the application determines the early warning level according to the disturbance severity, deploys monitoring equipment on high-level source nodes, ensures real-time controllability of key sources, avoids waste of resources on low-risk nodes, combines the disturbance propagation path, deploys monitoring equipment on the nodes on the path, forms a full-chain monitoring network of "source + propagation path", and the subsequent disturbance can be quickly found through the point on the propagation path, avoids disturbance diffusion, and improves the response speed.

[0026] The above description is only a summary of the technical scheme of the embodiment of the application, in order to more clearly understand the technical means of the embodiment of the application, the content of the specification can be implemented, and in order to make the above and other purposes, characteristics and advantages of the embodiment of the application more obvious and easy to understand, the specific implementation manner of the application is as follows. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 A computer system structure diagram based on the power Internet of Things cloud edge computing power collaborative computing model is provided for the embodiment of the application. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0029] Example 1: like Figure 1 As shown, a computer system based on a power Internet of Things cloud-edge computing power collaborative computing model provided by an embodiment of the present invention is suitable for monitoring power quality disturbance events. The computer system includes an edge-side time period determination module 101, an edge-side frequency domain analysis module 102, a cloud-side event classification module 103, a cloud-side source node determination module 104, and an edge-side monitoring module 105, wherein: The edge-side time period determination module 101 is configured to calculate a real-time disturbance intensity level based on a first voltage signal sequence and a first frequency signal sequence of a target power grid collected in real time using a preset machine learning model, and determine a disturbance event time period based on the real-time disturbance intensity level when the real-time disturbance intensity level exceeds a preset intensity threshold; In this embodiment, the edge-side time period determination module 101 calculates the real-time disturbance intensity level based on the first voltage signal sequence and the first frequency signal sequence of the target power grid collected in real time through a preset machine learning model. Specifically, the edge-side time period determination module 101 obtains the first voltage signal sequence and the first frequency signal sequence of several sampling points in the target power grid in real time through a preset sliding time window; based on the first voltage signal sequence, sequentially calculates the difference between the real-time voltage effective value and the rated voltage value of each sampling point to obtain a voltage amplitude deviation sequence; based on the first frequency signal sequence, sequentially calculates the difference between the real-time frequency value and the standard frequency value of each sampling point to obtain a frequency offset sequence; based on the voltage amplitude deviation sequence, the frequency offset sequence, the preset deviation abnormality threshold, and the preset offset abnormality threshold, the number of abnormal sampling points is counted to obtain abnormal statistical results, and the real-time disturbance intensity level is calculated based on the abnormal statistical results through a preset neural network model.

[0030] In a specific embodiment, on the edge side, the microprocessor first continuously collects the three-phase instantaneous voltage and frequency of the target power grid through the synchronous phasor measurement unit (PMU) at a fixed sampling rate of 1 kHz, forming a first voltage signal sequence and a first frequency signal sequence within a sliding time window of 512 points (about 0.5s). Subsequently, the following three-step operations are completed on the 32-bit floating-point DSP kernel: first, the effective value of the first voltage signal sequence is calculated point by point, and the difference between the effective value and the rated voltage 220V stored in the EEPROM is obtained to obtain the voltage amplitude deviation sequence AV(n); second, the first frequency signal sequence is subtracted from the standard power frequency 50Hz point by point to obtain the frequency offset sequence Af(n); finally, AV(n) and Af(n) are compared with the pre-set deviation abnormal threshold ±5% and the offset abnormal threshold ±0.2Hz, the number of out-of-limit sampling points in the window is counted, and the abnormal statistical result is obtained. The statistical result is sent to the 3-layer fully connected neural network (input layer 6-dimensional, hidden layer 32-dimensional, output layer 1-dimensional, ReLU activation, Adam optimizer offline training, and loss function Huber) quantified in TensorFlow Lite for inference, and the network output is the real-time 0-1 normalized disturbance intensity level R. If R exceeds the pre-set threshold value 0.6, it is determined that there is a power quality disturbance event in the current window, and the subsequent disturbance event time period determination module is triggered.

[0031] In the present embodiment, the edge side time period determination module 101 determines the disturbance event time period according to the real-time disturbance intensity level, specifically: the edge side time period determination module 101 differentiates the voltage amplitude deviation sequence of each sampling point to obtain the deviation change rate, and takes the sampling point whose deviation change rate exceeds the pre-set change rate threshold as the mutation point; according to the voltage amplitude deviation sequence corresponding to the mutation point, the disturbance starting point and the disturbance ending point are identified, and the maximum value of the voltage amplitude deviation sequence and the peak value of the frequency offset sequence between the disturbance starting point and the disturbance ending point are extracted; according to the real-time disturbance intensity level and the pre-set weight coefficient, the maximum value and the peak value are weighted and summed to obtain a comprehensive disturbance index, and if the comprehensive disturbance index is greater than the pre-set comprehensive disturbance threshold, the disturbance event time period is determined according to the disturbance starting point and the disturbance ending point.

[0032] In a specific embodiment, when the edge-side microprocessor determines that the real-time disturbance intensity level R≥0.6, a disturbance event time period accurate positioning process is immediately started: first, a first-order backward difference operation is performed on the obtained voltage amplitude deviation sequence ΔV(n) to obtain a deviation rate of change sequence dΔV(n) / dt, and the sampling points with a change rate absolute value exceeding a preset change rate threshold 3% / ms are marked as mutation points; then, taking the first mutation point as the starting point, the time axis is traced back to the sampling point where the deviation first falls below the baseline noise band (±0.5%) and is recorded as the disturbance starting point, and then searched back to the sampling point where the deviation again falls into the baseline noise band and lasts for more than 50 ms and is recorded as the disturbance ending point. In the closed interval of the disturbance time point and the disturbance ending point, the maximum value V_max of ΔV(n) and the peak value f_max of the frequency offset sequence Δf(n) are extracted respectively, and the comprehensive disturbance index D is calculated according to the real-time disturbance intensity level R and the preset weight coefficient 0.7, k_f=0.3. If D exceeds the comprehensive disturbance threshold 4.5, the closed interval from the disturbance time point to the disturbance ending point is finally determined as the disturbance event time period of this time, and the time period information is packaged together with the original sampling sequence and uploaded to the cloud event classification module through the MQTT protocol; if D does not meet the standard, it is considered as a transient disturbance, and the data in this section is discarded and the monitoring continues.

[0033] The edge-side frequency domain analysis module 102 is configured to adjust the width of the analysis window according to the disturbance event time period, so as to intercept the second voltage signal sequence and the second frequency signal sequence through the analysis window, and perform frequency domain feature extraction according to the second voltage signal sequence and the second frequency signal sequence to obtain abnormal analysis data. In this embodiment, the edge-side frequency domain analysis module 102 performs frequency domain feature extraction according to the second voltage signal sequence and the second frequency signal sequence to obtain abnormal analysis data, specifically: the edge-side frequency domain analysis module 102 performs fast Fourier transform on the second voltage signal sequence to obtain a frequency domain amplitude spectrum, and performs harmonic component amplitude extraction according to the frequency domain amplitude spectrum to calculate the harmonic content percentage according to the harmonic component amplitudes; performs fast Fourier transform on the second frequency signal sequence to extract frequency components, and takes the component with the maximum amplitude in the frequency components as the dominant frequency component; extracts a harmonic feature vector according to the harmonic content percentage and the pre-acquired target power grid phase data, and calculates the difference between the dominant frequency component and the preset power grid rated frequency to obtain an actual frequency offset value, so as to integrate the harmonic feature vector and the actual frequency offset value to obtain a frequency domain feature set; if there is a harmonic content percentage exceeding a preset alarm threshold in the harmonic feature vector, or the actual frequency offset value exceeds a frequency deviation limit value, it is determined that a frequency abnormal event occurs and an abnormal identifier is generated, and the frequency domain feature set and the abnormal identifier are integrated to obtain abnormal analysis data.

[0034] In a specific embodiment, after obtaining the disturbance event time period on the edge side, the edge computing unit immediately re-acquires the original three-phase voltage and frequency waveforms in this time period at a higher sampling rate of 12.8 kHz, and after pre-processing by an 8th-order Chebyshev type I anti-aliasing filter of 0.1-3 kHz, a second voltage signal sequence and a second frequency signal sequence with a length of N=2 048 are formed. Then the following frequency domain feature extraction steps are performed: First, the second voltage signal sequence is windowed by Blackman-Harris, and a 2 048-point base-2 FFT is performed to obtain the frequency domain amplitude spectrum; the amplitude of each harmonic at the 50 Hz fundamental wave and its 2-50th harmonic position is accurately extracted, and the harmonic content percentage HR_i=H_i / ΣH_j×100 % is calculated to construct a 50-dimensional harmonic feature vector h_vec.

[0035] Further, the second frequency signal sequence is also windowed and FFTed, and the maximum amplitude component in the 45-55 Hz interval is taken as the dominant frequency component f_d, and the actual frequency offset Δf=f_d-50 from the grid rated frequency 50 Hz is calculated.

[0036] Further, combined with the pre-acquired grid phase data, the base phase is extracted by a synchronous phase-locked loop, and the phase-amplitude coupling coefficients of the 5th, 7th, 11th, and 13th characteristic harmonics are calculated and concatenated to h_vec to form a 54-dimensional harmonic feature vector.

[0037] Further, when any HR_i> alarm threshold 5% or |Δf|> frequency deviation limit 0.5 Hz, set the 8-bit abnormality flag Flag=0xFF, otherwise Flag=0x00.

[0038] Finally, the 54-dimensional harmonic feature vector, Δf, and the abnormality flag are integrated into a 56-byte "abnormality analysis data" structure, serialized using Protocol Buffers, and uploaded to the cloud event classification module through a 4G / 5G encrypted link.

[0039] The cloud event classification module 103 is configured to calculate the similarity between the abnormality analysis data and each historical disturbance record in a pre-stored historical disturbance record set to obtain a first similarity, and according to the first similarity, filter the historical disturbance record set to obtain a historical disturbance record subset, and determine the disturbance event type according to the historical disturbance record subset and the abnormality analysis data by a pre-set K-means clustering algorithm. In this embodiment, the cloud event classification module 103 calculates the similarity between the abnormal analysis data and each historical disturbance record in the pre-stored historical disturbance record set to obtain a first similarity. Specifically, the cloud event classification module 103 calculates the similarity between the abnormal analysis data and each historical disturbance record in turn by using a dynamic time warping algorithm to obtain the first similarity.

[0040] In a specific embodiment, after receiving the abnormal analysis data uploaded on the edge side, the cloud event classification module first unpacks the 54-dimensional harmonic feature vector h_vec, the frequency offset Δf, and the abnormal identifier Flag; then the 56-byte data is compared with the “historical disturbance record set” deployed in the cloud time series database (InfluxDB cluster, retaining the last 24 months of records) for similarity calculation. The specific process is as follows: each historical record is also parsed into a 54-dimensional vector h_hist and a scalar Δf_hist, and both are standardized by z-score to have a mean of 0 and a variance of 1; then a dynamic time warping (DTW) algorithm based on Sakoe-Chiba constraint (window width r=5) is used to calculate the shortest warping path distance d_h between the 54-dimensional harmonic vector of the current abnormal analysis data and the corresponding vector of each historical record, and the absolute difference d_f between Δf and Δf_hist; finally, the first similarity S is defined as S=1 / (1+α·d_h+β·d_f), where α=0.85 and β=0.15 are the weighting coefficients obtained by offline Bayesian optimization, and the S value range is 0-1, with a value closer to 1 indicating greater similarity; all historical records are sorted in descending order of S to form a candidate set for subsequent K-means clustering.

[0041] In this embodiment, the cloud event classification module 103 determines the disturbance event type based on the historical disturbance record subset and the abnormal analysis data by using a pre-set K-means clustering algorithm. Specifically, the cloud event classification module 103 performs a difference operation on the historical disturbance record subset to extract disturbance change rules and disturbance periodicity features; performs clustering and grouping on the historical disturbance record subset and the abnormal analysis data based on the disturbance change rules and the disturbance periodicity features by using a pre-set K-means clustering algorithm to obtain a grouping type; and determines the disturbance event type based on the grouping type and a pre-set disturbance classification standard.

[0042] In a specific embodiment, after completing the first similarity calculation in the cloud, the system uses a similarity threshold of 0.65 to extract the top 5% of high-similarity samples from the entire historical disturbance record set to form a “historical disturbance record subset” (usually 300-800 records), and combines the current abnormal analysis data with the subset to form a new clustering sample pool. Then the following K-means clustering process is performed to determine the disturbance event type: (1) Feature reprocessing: Take the first-order difference of the 54-dimensional harmonic feature vector of each record in the sample pool, extract the change rate of adjacent harmonic content, and obtain a 54-dimensional "perturbation change rule" vector. At the same time, perform 256-point FFT on the vector, and take the top 4 frequency points with the largest energy as the "perturbation periodicity feature". Finally, a 54+4=58-dimensional feature vector is sent to the clusterer.

[0043] (2) Clusterer configuration: Use mini-batch K-means, batch size=100, maximum iteration 300 times; K value is determined by offline elbow method as 7 (corresponding to 7 main power quality events specified in national standard GB / T 14549-2008: harmonic distortion, voltage sag, voltage swell, short interruption, flicker, oscillatory transient, frequency deviation).

[0044] (3) Type determination: After clustering, take the centroid record of the cluster to which the current abnormal analysis data belongs, and the event label manually labeled is the type of the current disturbance event; if the current data falls on the boundary of two clusters (contour coefficient <0.25), a secondary SVM fine classifier (RBF kernel, γ=0.8, C=10) is started for secondary judgment to ensure that the classification confidence is ≥0.9.

[0045] (4) Result output: The cloud will write the finally determined disturbance event type (such as "harmonic distortion-5 times dominant" or "voltage sag-depth 35%") together with the cluster number and centroid distance into the Kafka message queue for calling by the downstream source node positioning module.

[0046] The cloud source node determination module 104 is configured to, according to the pre-stored target power grid topology data and the disturbance event type, obtain a candidate node list of a disturbance source through a preset graph neural network, and calculate the second similarity between each candidate node and a pre-stored historical source node one by one, so as to determine an actual source node according to the second similarity; wherein the historical source node and each candidate node are of the same disturbance event type. In the embodiment, the cloud source node determination module 104 obtains a candidate node list of a disturbance source according to the pre-stored target power grid topology data and the disturbance event type through a preset graph neural network, specifically: the cloud source node determination module 104 obtains the electrical parameters of each node in the target power grid, and constructs a weighted directed graph containing the electrical parameters according to the electrical parameters, the pre-stored target power grid topology data and the disturbance event type; taking the corresponding edge side node of the disturbance event as the starting point, the preset graph neural network is used to perform source tracing propagation on the weighted directed graph to obtain the candidate node list of the disturbance source.

[0047] In a specific embodiment, the cloud source node determination module first reads the pre-stored CIM / XML format target grid topology from PostgreSQL after receiving specific disturbance event types such as "harmonic distortion-5th dominant", and abstracts the 10kV and above busbars, main transformers, distributed power sources, SVGs, and load nodes in the entire grid into a weighted directed graph G=(V, E): Node attributes: For each node in V, the fundamental wave voltage, current, active power, reactive power, node short-circuit capacity, harmonic impedance 5-13 components, and load type weight (industrial 0.35, commercial 0.25, residential 0.15, etc.) uploaded in real time by phasor measurement devices are spliced into a 14-dimensional electrical parameter vector; Edge weights: 3-dimensional edge features are constructed based on branch resistance, reactance, power flow direction, and harmonic attenuation coefficients α_h related to event types (e.g. 0.92 / km for 5th harmonic), and normalized to [0, 1]; Graph neural network: A 3-layer GraphSAGE (aggregation function mean-pooling, hidden dimension 64, ReLU, Dropout 0.2) is used, and the supervised training has been completed with 120,000 labeled samples covering all 7 types of disturbances for 18 months in the offline stage. The loss function is negative log likelihood, the optimizer is AdamW, the learning rate is 1e-3, and the early stopping patience is 15. Online inference: The edge-side node that detects the disturbance is taken as the starting point, the G is input into the model solidified to TensorFlow Serving, the forward propagation is performed, the network aggregates the neighborhood features layer by layer and outputs the probability p_i that each node becomes a disturbance source; set the probability threshold to 0.55, sort the nodes with p_i≥0.55 in descending order of probability to form a candidate node list (usually 3-7 in length), and cache the list along with the 14-dimensional features of the nodes to Redis for the next "cosine similarity matching" call.

[0048] In this embodiment, the cloud source node determination module 104 calculates the second similarity of each candidate node with the pre-stored historical source nodes, and determines the actual source node according to the second similarity. Specifically, the cloud source node determination module 104 extracts the electrical parameters and load features of each candidate node from the candidate node list in turn, and performs cosine similarity calculation according to the electrical parameters and load features of each candidate node and the pre-stored electrical parameters and load features of the historical source nodes, to obtain the second similarity. If the second similarity exceeds the preset matching degree threshold, the corresponding candidate node is determined as the actual source node.

[0049] In a specific embodiment, after the cloud candidate node list is generated, the system immediately starts the "historical source node matching" process to accurately locate the actual disturbance source. First, from the historical source node library of PostgreSQL, only records that are completely consistent with the current disturbance event type (such as "5th harmonic dominant" or "voltage sag") and occurred within the past 180 days are retained to form a historical source node subset H. For each node C_i in the candidate node list, the following 10-dimensional feature vector is extracted in real time: fundamental voltage amplitude; 5th, 7th, 11th, and 13th harmonic voltage content rate; node short-circuit capacity S_k; current active load P; current reactive load Q; load power factor λ; load type encoding (industry 1.0, commercial 0.7, residential 0.4); distributed power capacity proportion D_p; historical average harmonic impedance modulus Z_h; load change rate ΔL before and after the event. For each historical source node record in H, a 10-dimensional vector h_j is also generated using the same feature template. Then, cosine similarity is used as the second similarity measure: first, all features are min-max normalized to [0, 1], then sim(C_i, h_j) = (C_i · h_j) / (‖C_i‖‖h_j‖) is calculated. The system calculates sim for C_i and all h_j in H in turn, taking the maximum value sim_max; if sim_max exceeds the preset matching degree threshold 0.82, C_i is immediately identified as the actual source node of the current disturbance event, and its node ID, geographic coordinates, and electrical parameters are written to the Kafka message queue; if all candidate nodes sim_max are below 0.82, the manual review process is triggered, and the one with the highest sim is selected as the temporary source node, with a confidence level of "low" marked for subsequent correction.

[0050] The edge side monitoring module 105 is configured to perform real-time frequency monitoring on the actual source node to obtain a real-time frequency value sequence, generate a disturbance event identification label based on the real-time frequency value sequence, and establish a power grid disturbance event early warning monitoring point layout based on the disturbance event identification label.

[0051] In this embodiment, the edge-side monitoring module 105 generates a disturbance event identification label according to the real-time frequency value sequence. Specifically, the edge-side monitoring module 105 calculates the difference between each real-time frequency value and the preset grid rated frequency according to the real-time frequency value sequence to obtain a real-time frequency offset value sequence; calculates the offset change rate according to the real-time frequency offset value sequence, and obtains the historical offset change rate of the source node to calculate the difference between the offset change rate and the historical offset rate; if the difference is greater than a preset threshold, a strong monitoring mode is started, and the sampling frequency is adjusted to a preset multiple of the original frequency to obtain high sampling rate disturbance data in the strong monitoring mode; and the disturbance amplitude, duration and harmonic component feature parameters are extracted from the high sampling rate disturbance data to generate a disturbance event identification label.

[0052] In a specific embodiment, after confirming the actual source node, the edge-side monitoring module immediately performs real-time collection on the PMU frequency channel of the node at a sampling rate of 6.4 kHz for 10 s to obtain a real-time frequency value sequence f_raw(t) of 64000 sampling points. Then the following steps are performed to generate a disturbance event identification label: (1) Preprocessing: remove direct current and slow drift by using a 5th order zero-phase Butterworth high-pass filter (cutoff 0.1 Hz) to obtain f_hp(t).

[0053] (2) Real-time offset sequence: calculate the difference between f_hp(t) and the grid rated frequency 50 Hz point by point to obtain a real-time frequency offset value sequence Δf(t).

[0054] (3) Offset change rate calculation: perform 80-point (12.5 ms) sliding difference on Δf(t) to obtain an instantaneous offset change rate sequence r(t)=dΔf / dt.

[0055] (4) Historical offset rate comparison: read the historical average offset change rate r_hist and its standard deviation σ of the same event type in the past 30 days of the node from Redis; if |r(t)-r_hist|>1.5σ, the "strong monitoring mode" is triggered.

[0056] (5) Strong monitoring mode: increase the sampling rate to 4 times the original frequency (25.6 kHz), continuously collect for 2 s to obtain high sampling rate disturbance data f_hi(t).

[0057] (6) Feature extraction: apply a 10% overlapping 512-point Kaiser window to f_hi(t), and extract a. disturbance amplitude A=max(|FFT(f_hi)|) peak value in the passband of 0.5-45 Hz; b. Duration T = time between first crossing of ±0.2Hz and last crossing of ±0.05Hz; c. Harmonic component feature parameters: 2-7 integer harmonic energy ratios H2-H7.

[0058] (7) Tag encoding: pack the 8 float parameters of A, T, H2-H7 according to IEEE-754 single precision, then splice the 8-bit event type code (taken from the result of claim 6) and the 4-bit confidence code to form a 40-byte "disturbance event identification tag". The tag is transmitted back to the cloud in real time through TLS encrypted UDP data packets, and is simultaneously written into the local SQLite history table for subsequent early warning and monitoring point layout.

[0059] It should be noted that the present application includes several edge side monitoring modules, and the selection of which edge side monitoring modules to start is determined by the location of the actual disturbance source in the power grid.

[0060] In the present embodiment, the edge side monitoring module 105 establishes a power grid disturbance event early warning and monitoring point layout according to the disturbance event identification tag, specifically: the edge side monitoring module 105 determines the early warning and monitoring level of the source node according to the disturbance event identification tag, so as to establish the early warning and monitoring point of the source node according to the early warning and monitoring level; determines the disturbance event propagation path according to the disturbance event identification tag and the target power grid topology data, and respectively establishes early warning and monitoring points on each node of the disturbance event propagation path.

[0061] In a specific embodiment, after receiving the 40 Byte disturbance event identification tag, the cloud early warning layout engine immediately parses it into 8 characteristics such as event type code, disturbance amplitude A, duration T, and calculates the early warning monitoring level L of the source node according to the built-in "level mapping table": if A≥2 Hz or T≥1.5 s, then L=Ⅳ (highest), if 1 Hz≤A<2 Hz and 0.5 s≤T<1.5 s, then L=Ⅲ, and so on. Subsequently, the engine takes the source node as the root node and performs an improved Dijkstra algorithm on the pre-stored power grid topology directed graph, with edge weights weighted by branch impedance and harmonic attenuation coefficient, to search for all downstream nodes within 3 km of the source node in terms of electrical distance or within the range of 2-level buses, forming a disturbance propagation path set P. For each node in P: (1) calculate the local monitoring level according to L and the node importance coefficient (0-1, scored by short-circuit capacity, load level, and number of sensitive users); (2) if the local monitoring level is greater than or equal to level II, deploy a PMU monitoring unit with 6.4 kHz continuous sampling, FFT real-time analysis, and MQTT return function at the node; if the local monitoring level is greater than or equal to level I and less than level II, add a 1 kHz sampling plug-in to the existing distribution automation terminal; if the local monitoring level is less than level I, only the SCADA slow data interface is retained.

[0062] Finally, the engine generates a "power grid disturbance event early warning monitoring point layout file" in JSON format, containing node ID, latitude and longitude, monitoring level, device model, communication method, and data return period (Ⅳ level 1 s, Ⅲ level 5 s, Ⅱ level 30 s, Ⅰ level 300 s), which is audited by the power grid dispatching EMS and then issued to each edge-side node for execution, completing the closed-loop layout.

[0063] The embodiment of the present application collects voltage and frequency signals in real time through the edge side time period determination module 101, calculates the disturbance intensity level through the machine learning model, quantifies the disturbance intensity through the machine learning model first, filters out slight fluctuations, avoids invalid data processing at the edge, reduces the consumption of computing power, quickly identifies the time window of the disturbance event by taking advantage of the fast transmission speed of the edge device, limits the effective analysis window for subsequent frequency domain analysis, avoids the distortion of features caused by mixing a large amount of normal data in the frequency domain analysis, and is the time reference for all subsequent analyses; through the edge side frequency domain analysis module 102, the window width is dynamically adjusted to ensure that the complete disturbance waveform is covered, the problem of "missing disturbance" or "picking up noise" of the fixed window is solved, and the time domain signal is converted into quantifiable frequency domain features through frequency domain extraction, the "abnormal analysis data" is generated by combining the abnormal identifier, the feature carrier is provided for cloud classification, and the core value of real-time preliminary analysis at the edge is realized; through the cloud event classification module 103, the cloud computing power is used to process massive historical data, the similarity filtering and K-means clustering are used to solve the problem that the edge computing power is insufficient for deep analysis, and the disturbance type is accurately classified to provide a type basis for subsequent source positioning; through the cloud source node determination module 104, the power grid topology and the disturbance type are combined, the graph neural network is used for reverse tracing to quickly lock the candidate node, blind search of the whole power grid is avoided, and the actual source is accurately picked out from the candidate through the second similarity, the positioning difficulty problem caused by disturbance migration is solved, and the positioning rate of the disturbance source is improved; through the edge side monitoring module 105, the positioned source is continuously monitored to generate an identification label, support is provided for disturbance full information recording, and early warning point positions covering the source and the propagation path are established according to the label to form a closed loop optimization, similar disturbances in the future can be quickly found through the early warning point positions, the response speed is improved, and the value of one-time positioning and long-term early warning is realized. Compared with the prior art, the present application can improve the real-time performance and accuracy of power quality disturbance event monitoring through the cooperation of cloud edge computing power.

[0064] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of each method. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), etc.

[0065] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are merely examples of the present application and are not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A computer system based on the power Internet of Things cloud-edge computing power collaborative computing model, characterized in that: Applicable to power quality disturbance event monitoring, the computer system includes an edge-side time period determination module, an edge-side frequency domain analysis module, a cloud-side event classification module, a cloud-side source node determination module and an edge-side monitoring module, wherein, The edge-side time period determination module is configured to calculate a real-time disturbance intensity level based on a first voltage signal sequence and a first frequency signal sequence of a target power grid collected in real time using a preset machine learning model, and determine a disturbance event time period based on the real-time disturbance intensity level when the real-time disturbance intensity level exceeds a preset intensity threshold; The edge-side frequency domain analysis module is configured to adjust a width of an analysis window according to a time period of the disturbance event, so as to intercept a second voltage signal sequence and a second frequency signal sequence through the analysis window, and perform frequency domain feature extraction based on the second voltage signal sequence and the second frequency signal sequence to obtain abnormal analysis data; The cloud event classification module is configured to calculate a similarity between the anomaly analysis data and each historical disturbance record in a pre-stored historical disturbance record set to obtain a first similarity, and filter the historical disturbance record set based on the first similarity to obtain a historical disturbance record subset, so as to determine a disturbance event type based on the historical disturbance record subset and the anomaly analysis data using a preset K-means clustering algorithm; The cloud-based source node determination module is configured to obtain a list of candidate nodes of the disturbance source through a preset graph neural network based on the pre-stored target power grid topology data and the disturbance event type, and calculate a second similarity between each candidate node and the pre-stored historical source nodes one by one, so as to determine the actual source node based on the second similarity; wherein the historical source node and each candidate node are of the same disturbance event type; The edge-side monitoring module is used to perform real-time frequency monitoring on the actual source node to obtain a real-time frequency value sequence, generate a disturbance event identification tag based on the real-time frequency value sequence, and establish a power grid disturbance event early warning monitoring point layout based on the disturbance event identification tag.

2. A computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1, characterized in that: The real-time disturbance intensity level is calculated based on the first voltage signal sequence and the first frequency signal sequence of the target power grid collected in real time by using a preset machine learning model, specifically: Acquire a first voltage signal sequence and a first frequency signal sequence of a plurality of sampling points in the target power grid in real time through a preset sliding time window; According to the first voltage signal sequence, the difference between the real-time voltage effective value and the rated voltage value of each sampling point is calculated in sequence to obtain a voltage amplitude deviation sequence; According to the first frequency signal sequence, the difference between the real-time frequency value of each sampling point and the standard frequency value is calculated in sequence to obtain a frequency offset sequence; According to the voltage amplitude deviation sequence, the frequency offset sequence, the preset deviation abnormality threshold and the preset offset abnormality threshold, the number of abnormal sampling points is counted to obtain abnormal statistical results, and the real-time disturbance intensity level is calculated according to the abnormal statistical results through a preset neural network model.

3. A computer system based on the power Internet of Things cloud-edge computing power collaborative computing model as described in claim 2, characterized in that: The determination of the disturbance event time period according to the real-time disturbance intensity level is specifically as follows: Performing a differential operation on the voltage amplitude deviation sequence at each sampling point to obtain a deviation change rate, and taking a sampling point where the deviation change rate exceeds a preset change rate threshold as a mutation point; Identify the disturbance start point and the disturbance end point according to the voltage amplitude deviation sequence corresponding to the mutation point, and extract the maximum value of the voltage amplitude deviation sequence and the peak value of the frequency offset sequence between the disturbance start point and the disturbance end point; According to the real-time disturbance intensity level and the preset weight coefficient, the maximum value and the peak value are weighted and summed to obtain a comprehensive disturbance index. If the comprehensive disturbance index is greater than the preset comprehensive disturbance threshold, the disturbance event time period is determined according to the disturbance starting point and the disturbance ending point.

4. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1 is characterized in that: The frequency domain feature extraction is performed based on the second voltage signal sequence and the second frequency signal sequence to obtain abnormal analysis data, specifically: Performing a fast Fourier transform on the second voltage signal sequence to obtain a frequency domain amplitude spectrum, and extracting the amplitude of each harmonic component based on the frequency domain amplitude spectrum, so as to calculate the percentage of each harmonic content based on the amplitude of each harmonic component; Performing a fast Fourier transform on the second frequency signal sequence to extract frequency components, and taking the component with the largest amplitude among the frequency components as the dominant frequency component; Extracting harmonic feature vectors based on the percentages of each harmonic content and pre-collected target grid phase data, and calculating the difference between the dominant frequency component and a preset grid rated frequency to obtain an actual frequency offset value, and integrating the harmonic feature vectors and the actual frequency offset value to obtain a frequency domain feature set; If the percentage of harmonic content in the harmonic feature vector exceeds the preset alarm threshold, or the actual frequency offset value exceeds the frequency deviation limit, it is determined to be a frequency abnormality event and an abnormality mark is generated. The frequency domain feature set is integrated with the abnormality mark to obtain abnormal analysis data.

5. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1 is characterized in that: The similarity between the abnormality analysis data and each historical disturbance record in the pre-stored historical disturbance record set is calculated to obtain a first similarity, specifically: The similarity between the abnormal analysis data and each historical disturbance record is calculated in sequence by using a dynamic time warping algorithm to obtain a first similarity.

6. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1, characterized in that: The method of determining the disturbance event type by using a preset K-means clustering algorithm based on the historical disturbance record subset and the abnormality analysis data is as follows: Performing a differential operation on the subset of historical disturbance records to extract disturbance variation patterns and disturbance periodicity characteristics; By using a preset K-means clustering algorithm, the historical disturbance record subset and the abnormal analysis data are clustered and grouped according to the disturbance change law and the disturbance periodicity characteristics to obtain a grouping type, and the disturbance event type is determined according to the grouping type and a preset disturbance classification standard.

7. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1 is characterized in that: According to the pre-stored target power grid topology data and the disturbance event type, a candidate node list of the disturbance source is obtained through a preset graph neural network, specifically: Acquire electrical parameters of each node in the target power grid, and construct a weighted directed graph containing the electrical parameters based on the electrical parameters, pre-stored target power grid topology data, and the disturbance event type; Starting from the corresponding edge node where the disturbance event is detected, the weighted directed graph is traced back through a preset graph neural network to obtain a list of candidate nodes of the disturbance source.

8. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1 is characterized in that: The step of calculating the second similarity between each candidate node and the pre-stored historical source node one by one to determine the actual source node according to the second similarity is specifically as follows: Extracting electrical parameters and load characteristics of each candidate node from the candidate node list in sequence, and performing cosine similarity calculation based on the electrical parameters and load characteristics of each candidate node and the electrical parameters and load characteristics of the pre-stored historical source node to obtain a second similarity; If the second similarity exceeds a preset matching threshold, the corresponding candidate node is determined to be the actual source node.

9. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1, characterized in that: The generating of a disturbance event identification label according to the real-time frequency value sequence is specifically as follows: According to the real-time frequency value sequence, the difference between each real-time frequency value and the preset grid rated frequency is calculated in sequence to obtain a real-time frequency offset value sequence; Calculating an offset change rate based on the real-time frequency offset value sequence, and obtaining a historical offset change rate of the source node to calculate a difference between the offset change rate and the historical offset rate; If the difference is greater than a preset threshold, a strong monitoring mode is turned on, and in the strong monitoring mode, the sampling frequency is adjusted to a preset multiple of the original frequency to obtain high sampling rate disturbance data; Based on the high sampling rate disturbance data, the disturbance amplitude, duration and harmonic component characteristic parameters are extracted to generate disturbance event identification labels.

10. The computer system based on the power Internet of Things cloud-edge computing power collaborative computing model according to claim 1, characterized in that: The establishment of a grid disturbance event early warning monitoring point layout based on the disturbance event identification tag is specifically as follows: Determine the early warning monitoring level of the source node according to the disturbance event identification tag, and establish the early warning monitoring point of the source node according to the early warning monitoring level; A disturbance event propagation path is determined according to the disturbance event identification tag and the target power grid topology data, and an early warning monitoring point is established at each node of the disturbance event propagation path.

Images