A power grid data acquisition and analysis system based on big data

Abstract

The application discloses a power grid data acquisition and analysis system based on big data, and particularly relates to the technical field of power network monitoring, and is used for solving the problem of key feature loss caused by the missing of cross-level data semantic association in the existing hierarchical aggregation mechanism; the original data is standardized and filled and time label layered alignment processed through the equipment acquisition module; the feature extraction module filters abnormal waveform fragments and steady-state parameter offset based on the equipment type and real-time fluctuation amplitude; the semantic association module generates space-time association weight in combination with the regional physical topological relationship and the reactive loop current path sensitivity; the feature clustering module reconstructs the regional level feature aggregation package through kernel density estimation and spatial weighted fusion; the cross-layer analysis module dynamically corrects the aggregation weight based on the current flow direction and transient energy distribution; the decision generation module generates the equipment positioning instruction and the load scheduling strategy combination through the alarm template matching, and finally realizes the accurate positioning of the fault equipment and the generation of the scheduling strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power network monitoring, and more particularly, to a power grid data acquisition and analysis system based on big data. BACKGROUND

[0002] Power network operation needs to realize state monitoring and dispatching decision through data collaboration at the device level, the regional level and the whole network level. The existing power grid data acquisition system usually adopts a hierarchical aggregation mechanism to compress the raw data collected by the bottom layer device and transmit it to the upper layer analysis platform, so as to reduce the transmission and storage cost. This mode relies on preset statistical rules (such as mean value calculation and threshold filtering) to reduce the dimension of multi-source data, and forms different levels of aggregated data sets for system calling.

[0003] The existing hierarchical aggregation mechanism lacks adaptive processing of the semantic association of raw data in the cross-level data fusion process, resulting in irreversible loss of key operation characteristics in the hierarchical compression. For example, the high-frequency transient signal at the device level is averaged and hidden in the regional aggregation, and the load trend statistics at the regional level cannot be traced back to the specific device exception, which directly affects the accuracy of fault location and the reliability of dispatching decision. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a power grid data acquisition and analysis system based on big data to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] A power grid data acquisition and analysis system based on big data, comprising:

[0007] A device acquisition module: obtaining raw data at the device end of the power grid and performing standardized filling and time label hierarchical alignment processing to generate standardized device-level operation data;

[0008] A feature extraction module: screening a dynamic feature set from the standardized device-level operation data, the dynamic feature set including abnormal waveform segments and steady-state parameter offsets within a preset period;

[0009] A semantic association module: judging whether the dynamic feature set triggers a cross-device association event, and if so, generating a spatio-temporal association weight of the dynamic feature set based on the regional physical topology relationship and the sensitivity of the reactive loop current path;

[0010] A feature clustering module: according to the spatio-temporal association weight and the device-level feature distribution density within a predefined period, the dynamic feature set within a preset association radius is aggregated and reconstructed into a regional-level feature aggregation package;

[0011] Cross-layer analysis module: based on the current flow direction and transient energy distribution characteristics between the devices in the regional feature aggregation package, the dynamic aggregation weight of the regional feature aggregation package is corrected;

[0012] Decision generation module: according to the corrected dynamic aggregation weight and the regional feature aggregation package, the preset alarm level template is matched, and the device positioning instruction and the regional scheduling strategy combination are output.

[0013] In a preferred embodiment, the original data of the power grid equipment end is obtained and standardized and filled in and time label hierarchical alignment processing is performed to generate standardized equipment level operation data, including:

[0014] Based on the preset filling rule corresponding to the device type, the missing value of the obtained original data of the power grid equipment end is standardized and filled in;

[0015] According to the device type and the signal sampling frequency, the time label hierarchical alignment strategy is determined, the time label of the power frequency device is aligned according to the cycle period, and the time label of the high frequency sampling device is aligned according to the microsecond level time window;

[0016] The filled and aligned data is integrated into standardized equipment level operation data according to a unified time axis.

[0017] In a preferred embodiment, a dynamic feature set is selected from the standardized equipment level operation data, including:

[0018] Based on the sliding window dynamic detection method, the abnormal waveform segment is extracted, the sliding window period is consistent with the preset period corresponding to the device type, and when the data fluctuation amplitude in the window exceeds the dynamic threshold, it is marked as an abnormal waveform segment. The dynamic threshold is adjusted according to the peak-to-valley difference mean value of the normal waveform in the device historical operation data;

[0019] Based on the steady-state baseline calculation method, the steady-state parameter offset is determined, the steady-state baseline value of each time stamp in the preset period is extracted from the standardized equipment level operation data, the steady-state baseline value is the mean value of the historical operation data of the same device in the absence of abnormal period, and the absolute deviation of the current data from the steady-state baseline value is calculated as the offset. When the offset exceeds the preset offset threshold, it is determined as an effective steady-state parameter offset.

[0020] In a preferred embodiment, it is judged whether the dynamic feature set triggers a cross-device associated event, and if so, the time-space correlation weight of the dynamic feature set is generated based on the regional physical topology relationship and the reactive power loop current path sensitivity, including:

[0021] The cross-device associated event triggering condition is determined based on the electrical distance between devices and the time overlap of dynamic characteristics, the electrical distance is calculated according to the shortest connection path length between devices in the regional physical topology relationship, and the time overlap is the time window overlap ratio of the abnormal waveform segment or the steady-state parameter offset;

[0022] If the cross-device associated event is triggered, the space-time correlation weight is adjusted according to the reactive current loop sensitivity, the reactive current loop sensitivity is calculated by the ratio of the inter-device reactive current mutual information entropy to the topological distance, the initial value of the space-time correlation weight is the reciprocal of the electrical distance between devices, and the final space-time correlation weight is the initial value of the space-time correlation weight multiplied by the reactive current loop sensitivity correction coefficient.

[0023] In a preferred embodiment, according to the space-time correlation weight and the device-level characteristic distribution density within a predefined period, the dynamic characteristics within the preset correlation radius are combined and reconstructed into a regional-level characteristic aggregation package, including:

[0024] The device-level characteristic distribution density is calculated based on the kernel density estimation method, and the kernel function bandwidth is adjusted according to the device spatial distribution sparsity, and the device spatial distribution sparsity is determined by the ratio of the average distance between devices to the preset correlation radius;

[0025] The regional characteristic density distribution map is generated according to the device-level characteristic distribution density, and the regional characteristic density distribution map is represented by a heat map grid;

[0026] The high-density area is extracted from the regional characteristic density distribution map based on the preset correlation radius, and the high-density area is an area with continuous grid cell density values exceeding a preset density threshold;

[0027] The dynamic characteristics within the high-density area are weighted and fused according to the space-time correlation weight, and are reconstructed into a regional-level characteristic aggregation package, and the weighted and fused manner is to multiply the dynamic characteristic value by the space-time correlation weight of the corresponding device, then accumulate, and finally divide by the total weight sum.

[0028] In a preferred embodiment, the grid cell density value is the weighted sum of the device-level characteristic distribution density within the corresponding spatial range, the weight is the space-time correlation weight of the device within the corresponding cell, and the preset density threshold is determined according to the statistical distribution of the density value in the historical regional-level characteristic aggregation event.

[0029] In a preferred embodiment, based on the current flow direction and transient energy distribution characteristics between the devices belonging to the regional-level characteristic aggregation package, the dynamic aggregation weight of the regional-level characteristic aggregation package is corrected, including:

[0030] The current flow direction is determined by the phase difference between the currents and the voltage gradient direction between the devices;

[0031] The transient energy distribution feature in the region-level feature aggregation package is aggregated, and the transient energy distribution feature extracts the frequency band energy proportion of each device transient waveform through wavelet packet decomposition. The frequency band energy proportion is the ratio of the energy of a specified frequency band to the total energy.

[0032] The initial allocation proportion of the dynamic aggregation weight between devices is adjusted according to the current flow direction, and the adjustment amplitude is a function of the consistency of the current flow direction and the topology path.

[0033] The adjusted weight is normalized and corrected in combination with the transient energy distribution feature. The correction coefficient is the ratio of the transient energy proportion of the device to the average transient energy proportion of the region. The final dynamic aggregation weight is the product of the adjusted weight and the correction coefficient.

[0034] In a preferred embodiment, the current phase difference is the phase shift angle of the voltage and current waveforms of the device at the same timestamp, and the voltage gradient direction is the change trend of the voltage amplitude along the topology path.

[0035] In a preferred embodiment, according to the corrected dynamic aggregation weight and the region-level feature aggregation package, a preset alarm level template is matched, and a device positioning instruction and a regional scheduling strategy combination are output, including:

[0036] Based on the corrected dynamic aggregation weight and the device spatial distribution density in the region-level feature aggregation package, the corresponding alarm level in the preset alarm level template is matched. The alarm level template is divided by the joint distribution threshold of the dynamic aggregation weight and the device spatial density in the historical fault event.

[0037] According to the alarm level, a device positioning instruction is generated, which contains the spatial coordinates of the high dynamic aggregation weight device and the associated topology path. The spatial coordinates are calculated by the geographic range in the region-level feature aggregation package and the weighted centroid of the device-level feature distribution density.

[0038] Based on the alarm level and the device positioning instruction, a regional scheduling strategy combination is generated.

[0039] In a preferred embodiment, the regional scheduling strategy combination includes a load reduction ratio, a voltage adjustment priority, and a standby line switching sequence. The load reduction ratio is dynamically allocated according to the power proportion of the high dynamic aggregation weight device. The voltage adjustment priority is sorted by the consistency of the device voltage gradient direction and the current flow path.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] 1、Through the deep coupling of multi-level data fusion and power grid physical characteristics, the key feature loss and cross-level semantic distortion problems in traditional hierarchical aggregation mechanism are effectively solved. The standardized filling of the device acquisition module and the time label hierarchical alignment processing ensure the time reference unification and format compatibility of multi-source heterogeneous data, providing high-precision input for subsequent feature extraction; the dynamic feature screening mechanism adaptively extracts abnormal waveform fragments and steady-state offsets based on device type and real-time fluctuation amplitude, avoiding the masking of traditional mean or threshold filtering to high-frequency transient signals. The semantic association module maps discrete device features to the real connection network of the power grid through joint analysis of regional physical topological relationship and reactive loop path sensitivity, establishes cross-device causal logical association, and solves the defect that artificial rules cannot capture implicit electrical interaction. The feature clustering module combines time and space weights and distribution density to retain device-level detail features while focusing on key operating states through high-density area extraction technology, significantly improving the physical interpretability of data aggregation. The cross-layer analysis module further introduces current flow direction and transient energy distribution characteristics, dynamically modifies the weight to match the real-time power grid operating conditions, and ensures the consistency of data features and electrical rules. The final generated dispatching instructions are directly associated with device positioning and regional operating state, forming a closed-loop link from data acquisition to decision execution, greatly improving fault location accuracy and dispatching reliability.

[0042] 2、While improving the consistency of data semantics, the real-time and adaptability of the dynamic operation of the power grid are also taken into account. Standardized device-level operating data are processed through time label alignment and filling rules to lay the foundation for multi-source data fusion, avoiding feature correlation deviation caused by time misalignment in traditional methods; the selection rules of dynamic feature set are based on device historical operating conditions and real-time fluctuation to dynamically adjust the threshold, replacing fixed threshold filtering to make anomaly detection more consistent with the actual operation of the power grid. The generation of time and space correlation weights combines reactive loop path sensitivity and topological physical connection to quantify the implicit electrical influence between devices, making cross-device correlation analysis have both mathematical statistical rigor and power grid physical regularity. The regional-level feature aggregation package combines kernel density estimation and spatial weighted fusion to retain device-level feature details and reflect the overall trend of regional operating state, solving the irreversible loss of features caused by traditional hierarchical compression. The modification of dynamic aggregation weight combines transient energy distribution and flow direction to ensure that the data aggregation result matches the dynamic changes of the power grid in real time, avoiding the decision lag caused by static weight. The matching mechanism of alarm level template and dispatching strategy is based on joint analysis of dynamic weight and spatial density to directly map data features to control instructions, reducing the dependence on manual intervention and achieving automation and precision of fault location and dispatching decision. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 FIG. 1 is a structural schematic diagram of a power grid data acquisition and analysis system based on big data according to the present application. Detailed Implementation

[0044] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0045] Example: Figure 1 A schematic diagram of a power grid data acquisition and analysis system based on big data is provided according to the present invention. The power grid data acquisition and analysis system based on big data includes the following modules:

[0046] Equipment acquisition module: Acquires raw data from power grid equipment and performs standardized filling and time tag layer alignment processing to generate standardized equipment-level operating data;

[0047] Feature extraction module: Filters dynamic feature sets from standardized equipment-level operating data. The dynamic feature sets include abnormal waveform segments and steady-state parameter offsets within a preset period.

[0048] Semantic association module: Determines whether the dynamic feature set triggers a cross-device association event. If it does, it generates the spatiotemporal association weights of the dynamic feature set based on the regional physical topology and reactive power circulation path sensitivity.

[0049] Feature clustering module: Based on the spatiotemporal correlation weights and the device-level feature distribution density within a predefined time period, the dynamic feature set within a preset correlation radius is reconstructed into a regional feature aggregation package;

[0050] Cross-layer parsing module: Based on the current flow direction and transient energy distribution characteristics among the devices in the regional feature aggregation package, the dynamic aggregation weight of the regional feature aggregation package is corrected;

[0051] Decision generation module: Based on the corrected dynamic aggregation weight and regional feature aggregation package, it matches the preset alarm level template and outputs a combination of device positioning instructions and regional scheduling strategies.

[0052] Firstly, the device acquisition module obtains the original data of the power grid device end, generates standardized device-level operation data through standardized filling and time label hierarchical alignment processing, and solves the time reference and format compatibility problems of multi-source heterogeneous data. Subsequently, the feature extraction module filters out the dynamic feature set containing abnormal waveform fragments and steady-state parameter offset from the standardized data based on the device type, preset operating condition rules and real-time fluctuation amplitude, ensuring that the key operating features are not masked by the data dimensionality reduction process. The dynamic feature set is input into the semantic association module, which triggers cross-device association events by judging the current phase difference between devices and the voltage gradient direction, generates time and space association weights based on the regional physical topology relationship and the sensitivity of the reactive current loop path, and associates discrete device features into a topological network with electrical and physical significance. The feature clustering module uses kernel density estimation to generate regional-level feature aggregation packages based on the time and space association weights and the device-level feature distribution density, and reconstructs the device-level dynamic features into structured data reflecting the regional operating state through heat map gridding and high-density area extraction technology. The cross-layer analysis module further normalizes and adjusts the dynamic aggregation weights in the aggregation package based on the current flow direction and transient energy distribution characteristics through the weight correction coefficient, realizing the deep coupling of electrical parameters and data features. Finally, the decision-making module matches the modified weights and aggregation packages with the preset alarm level templates to output combined instructions containing device spatial coordinates, topological paths and load scheduling strategies, forming a closed-loop link from data acquisition to decision execution.

[0053] The system solves the cross-layer semantic distortion problem in the prior art through a multi-level data fusion mechanism. The standardized filling and time alignment processing of the device acquisition module eliminates the time misalignment and format conflict of multi-source data, laying a foundation for subsequent feature association; the feature extraction module dynamically filters abnormal waveforms and steady-state offsets, avoiding the loss of high-frequency transient signals caused by traditional statistical rules; the semantic association module introduces the sensitivity of the reactive current loop path and the regional physical topology relationship to map device-level features to the real physical connection network of the power grid, solving the defect that artificial rules cannot associate cross-device causal logic; the feature clustering module uses kernel density estimation and spatial weighted fusion to retain device-level detailed features while generating reversible regional-level aggregated data; the cross-layer analysis module couples the current flow direction and transient energy distribution to correct the weights to reflect the real-time electrical state, improving the physical consistency of the aggregated data; the decision-making module generates positioning and scheduling instructions based on dynamic weights and alarm templates, directly associating data features and power grid control actions, ultimately achieving dual improvement of fault positioning accuracy and scheduling reliability.

[0054] The original data of the power grid device end is obtained and standardized filling and time label hierarchical alignment processing is performed to generate standardized device-level operation data, which is implemented as follows:

[0055] When the original data at the end of the power grid equipment is filled with missing values, the operation is performed according to the preset filling rule corresponding to the equipment type. The preset filling rule is defined by a mapping relationship table of equipment type and missing value processing logic, for example, the missing values of the current transformer equipment are filled with linear interpolation of adjacent time stamp data, and the missing values of the voltage transformer equipment are filled with mean value of the same device historical data. The specific way of linear interpolation filling is: if the current data of a certain time stamp is missing, the current value of the previous valid time stamp and the current value of the next valid time stamp are taken to perform arithmetic average, and the average value is taken as the filling value of the missing time stamp. The mean value filling method of the same device historical data is: extracting all normal date voltage data in the same period from the historical database of the voltage transformer equipment, calculating the average value of each time stamp voltage value, and taking the average value as the filling value of the missing time stamp. The interval of adjacent time stamps in the preset filling rule is set according to the maximum allowed data interruption time of the equipment type, for example, the maximum allowed data interruption time of the current transformer is 10 seconds, if the interval between the previous and next valid data of the missing time stamp exceeds 10 seconds, it is determined that the filling is invalid and marked as abnormal data.

[0056] The mapping relationship table includes four fields of equipment type number, equipment type name, maximum allowed interruption time and filling method. For example, the filling method corresponding to the current transformer with equipment type number CT001 is linear interpolation of adjacent time stamps, and the maximum allowed interruption time is 10 seconds; the filling method corresponding to the voltage transformer with equipment type number VT002 is historical mean value filling, and the maximum allowed interruption time is 30 seconds.

[0057] When the time tag hierarchical alignment processing is performed on the filled data, the time tag hierarchical alignment strategy is determined according to the device type and the signal sampling frequency. The device type is divided into two categories: power frequency devices and high-frequency sampling devices. The power frequency devices include transformers, circuit breakers and other devices with 50Hz power frequency signals as the reference. The high-frequency sampling devices include synchronous phasor measurement devices, traveling wave distance measurement devices and other devices with a sampling frequency higher than 1kHz. The time tag alignment strategy for power frequency devices is to align according to the cycle period. The specific method is: the original time stamp is rounded according to the 20ms cycle period, and the data in the same cycle period is marked with the same time tag. For example, the data with time stamps of 15ms and 35ms are aligned to the time tags of 0ms and 20ms respectively. The time tag alignment strategy for high-frequency sampling devices is to align according to the microsecond-level time window. The specific method is: taking 1μs as the reference time window, the original time stamp is rounded according to the integer multiple of 1μs. For example, the data with a time stamp of 1234567.89μs is aligned to the time tag of 1234568μs. In the time tag hierarchical alignment strategy, the reference time error of cycle period alignment is not more than ±0.1ms, and the reference time error of microsecond-level time window alignment is not more than ±0.01μs. The error range is set according to the requirements of power system synchronous phasor measurement, which will not be described here.

[0058] When the filled and aligned data is integrated into standardized device-level operation data according to the unified time axis, the time resolution of the unified time axis is 1μs, and the time tags of all device data are converted to absolute time stamps under the time axis. For example, the cycle alignment time tag 0ms of the power frequency device is converted to 0μs of the unified time axis, and the microsecond-level alignment time tag 1234568μs of the high-frequency sampling device is directly retained as 1234568μs of the unified time axis. During the integration process, if there are multiple device data at the same time stamp, they are stored as a multi-dimensional array according to the ascending order arrangement rule of the device number. The ascending order arrangement rule of the device number is: first sorting according to the device type number from small to large, and then sorting the devices of the same type according to the installation position number from small to large. The data format of the standardized device-level operation data is a structured table containing time stamp, device number and data value. The time stamp accuracy is 1μs, the device number is a globally unique identifier, and the data value is a floating point number after standardized filling and alignment processing.

[0059] The dynamic feature set is selected from the standardized device-level operation data, which is implemented by the following steps:

[0060] In the abnormal waveform segment extraction based on the sliding window dynamic detection method, the sliding window period is set according to the preset period corresponding to the device type. The device types include current transformers, voltage transformers, and traveling wave distance measurement devices. The sliding window period of different device types is defined through the device type number and the preset period mapping table. The mapping table includes three fields of device type number, device type name, and preset period. For example, the preset period of the current transformer with device type number CT001 is 1 minute, the preset period of the traveling wave distance measurement device with device type number TW003 is 1 second, and the preset period of the voltage transformer with device type number VT005 is 10 seconds. The specific operation mode of the sliding window dynamic detection method is as follows: taking the sliding window period as the time length, taking half of the window period as the sliding step, sequentially intercepting the continuous segments of the standardized device-level operation data, and calculating the fluctuation amplitude of the data in each window. The calculation method of the fluctuation amplitude is the difference between the maximum value and the minimum value in the window. For example, the maximum current in a certain current transformer window is 100 A, and the minimum value is 80 A, so the fluctuation amplitude is 20 A. The dynamic threshold is adjusted according to the peak-to-valley difference average of the normal waveform in the device historical operation data. If the historical data is less than 30 days, the average is calculated according to the actual available days and multiplied by 1.5 to generate the dynamic threshold. For example, a certain current transformer has only 15 days of historical data, and the peak-to-valley difference average is 12 A, so the dynamic threshold is set to 18 A. When the fluctuation amplitude in the window exceeds the dynamic threshold, the data in the window is marked as an abnormal waveform segment by the abnormal marking unit. The data input of the abnormal marking unit is the window fluctuation amplitude and the dynamic threshold, and the data output is the time range record of the abnormal waveform segment.

[0061] When determining the steady-state parameter deviation based on the steady-state baseline calculation method, the steady-state baseline value is obtained as follows: data of a non-exception period is selected from historical operation data of the same device, and the determination condition of the non-exception period is that the historical data does not contain any abnormal waveform segment and the fluctuation amplitude is lower than 80% of the dynamic threshold. For example, the dynamic threshold of the current transformer is 22.5 A, and the fluctuation amplitude of the non-exception period needs to be lower than 18 A. The historical data of the non-exception period is aligned to the time axis of the standardized device-level operation data according to the time stamp, and the alignment rule is to ignore the year information and only keep the month, day, hour, minute, second and microsecond information. For example, the historical data time stamp is October 1, 2022 08:00:00.000000, and it is aligned to October 1, 2023 08:00:00.000000. The steady-state baseline value corresponding to each time stamp is calculated, and the steady-state baseline value is the average value of the same time stamp in the historical non-exception period. For example, the average value of the current value of the time stamp October 1, 2023 08:00:00.000000 in the historical non-exception period is 95 A, and the steady-state baseline value is 95 A. When calculating the absolute deviation of the current data and the steady-state baseline value, the absolute deviation is calculated as the absolute value of the difference between the current data value and the steady-state baseline value. For example, the current value of the current time stamp is 105 A, and the steady-state baseline value is 95 A, so the absolute deviation is 10 A. The preset deviation threshold is determined by the statistical distribution of the historical deviation of the device, and the specific method is as follows: calculate the 95% quantile of the absolute deviation of each time stamp in the historical non-exception period, and take the 95% quantile as the preset deviation threshold. For example, the 95% quantile of the historical absolute deviation is 8 A, and the preset deviation threshold is set to 8 A. When the absolute deviation exceeds the preset deviation threshold, it is determined that the deviation of the time stamp is an effective steady-state parameter deviation, and the recording method of the effective steady-state parameter deviation is time stamp, current data value, steady-state baseline value and deviation.

[0062] The abnormal waveform segment and the effective steady-state parameter deviation jointly constitute the dynamic feature set. The data structure of the dynamic feature set is a structured table containing time stamp, feature type and feature value, and the feature type is divided into two categories: abnormal waveform segment and steady-state parameter deviation. For example, the feature type of the abnormal waveform segment is marked as "waveform exception", and the feature value is the window start time stamp, end time stamp and fluctuation amplitude; the feature type of the steady-state parameter deviation is marked as "steady-state deviation", and the feature value is the time stamp, current data value, steady-state baseline value and deviation. The time stamp accuracy of the dynamic feature set is consistent with that of the standardized device-level operation data, which is 1 microsecond.

[0063] It is judged whether the dynamic feature set triggers a cross-device association event, and if so, the time-space association weight of the dynamic feature set is generated based on the regional physical topology relationship and the reactive loop current path sensitivity, and the implementation is as follows:

[0064] When determining whether the cross-device associated event triggering condition is met based on the electrical distance between devices and the time overlap degree of dynamic feature sets, the electrical distance is calculated according to the shortest connection path length between devices in the regional physical topology relationship. The regional physical topology relationship is defined by a power grid connection topology graph, and the construction rule of the topology graph is: each power device is mapped as a node in the graph, and the physical connection (such as a transmission line, a switch) between devices is mapped as an undirected edge between nodes, and the edge weight is the physical length or equivalent impedance of the connection path. For example, a transformer device is mapped as node T1, a circuit breaker device is mapped as node B2, and the two are connected by a transmission line, then an edge is added between nodes T1 and B2, and the edge weight is 1 (indicating one connection section). The shortest connection path length is calculated by Dijkstra's algorithm, for example, devices A and B are connected through two switches and a transmission line, and the shortest path length is 3. The time overlap degree is calculated by the time window overlap ratio of the abnormal waveform segment or the steady-state parameter offset, and the time window is the range from the start timestamp to the end timestamp of the abnormal waveform segment or the steady-state parameter offset. For example, the abnormal waveform time window of device A is 08:00:00 to 08:00:10, and the abnormal waveform time window of device B is 08:00:05 to 08:00:15, the overlapping time window is 08:00:05 to 08:00:10, and the overlapping ratio is the overlapping window length (5 seconds) divided by the length of the shorter window (10 seconds) in the two windows, that is, 50%. When the electrical distance is less than the preset electrical distance threshold and the time overlap degree is greater than the preset overlap ratio threshold, it is determined that the cross-device associated event triggering condition is met. The preset electrical distance threshold is determined by the statistical distribution of the electrical distance between devices in the historical associated events, for example, the 75% quantile of the historical electrical distance is taken as the threshold, if the historical electrical distance data is [2, 3, 4, 5], the 75% quantile is 4.25, and the threshold is set to 4. The preset overlap ratio threshold is set by artificial experience, for example, it is set to 30%.

[0065] If a cross-device association event is triggered, the spatio-temporal association weight is adjusted according to the reactive current path sensitivity. The reactive current path sensitivity is calculated by the ratio of the inter-device reactive current mutual information entropy and the topological distance. The inter-device reactive current mutual information entropy is calculated as follows: the inter-device reactive power sequence is extracted from the historical reactive current data, the reactive power value is discretized into multiple intervals according to the preset bin width, the joint occurrence frequency of the reactive power values of two devices in the same interval is counted, and the joint probability distribution is calculated; the independent occurrence frequency of the reactive power values of each device in each interval is counted, and the independent probability distribution is calculated. The logarithmic expectation value of the ratio of the joint probability distribution and the independent probability distribution is calculated as follows: the joint probability value of each interval is multiplied by the natural logarithm of the ratio of the joint probability and the independent probability product of the interval, and the calculation results of all intervals are accumulated to obtain the mutual information entropy. For example, the reactive power sequence of device A and device B is discretized into 10 intervals, the probability distribution is calculated after counting the joint frequency matrix, and the final entropy value is 0.8 bits. The topological distance is the shortest connection path length between devices in the regional physical topological relationship, for example, the topological distance between device A and device B is 3. The reactive current path sensitivity is the mutual information entropy divided by the topological distance, for example, 0.8 / 3≈0.267. The initial value of the spatio-temporal association weight is the reciprocal of the electrical distance between devices, for example, the electrical distance between device A and device B is 3, and the initial weight is 1 / 3≈0.333. The final spatio-temporal association weight is the initial value multiplied by the reactive current path sensitivity correction coefficient, for example, 0.333×0.267≈0.089.

[0066] The calculation result of the spatio-temporal association weight is recorded as a structured table containing device pair number, initial weight, correction coefficient, and final weight. For example, the initial weight of device A and device B with device pair number AB001 is 0.333, the correction coefficient is 0.267, and the final weight is 0.089. The timestamp accuracy of the structured table is consistent with the timestamp of the dynamic feature set, both of which are 1 microsecond. When there is no physical connection between devices, the electrical distance is set to 100000 (representing infinity), and the initial value of the spatio-temporal association weight is set to 0. If there is a physical connection between devices but the topological distance exceeds the preset electrical distance threshold (for example, 100), the initial value of the spatio-temporal association weight is also set to 0. If the time overlap degree is 0, it is directly determined that the cross-device association event is not triggered. The preset bin width is dynamically adjusted according to the historical reactive power range, for example, the historical reactive power range is -10MVar to +10MVar, the bin width is set to 2MVar, and a total of 10 intervals are divided.

[0067] According to the spatio-temporal association weight and the device-level feature distribution density within the predefined period, the dynamic feature set within the preset association radius is combined and reconstructed into a regional-level feature aggregation package, which is implemented as follows:

[0068] When calculating the device-level feature distribution density based on the kernel density estimation method, the bandwidth of the kernel function is adjusted according to the sparsity of the device spatial distribution. The sparsity of the device spatial distribution is calculated by the ratio of the average distance between devices to the preset correlation radius. The average distance between devices is the average of the Euclidean distances between all pairs of devices in the region. For example, if the preset correlation radius is 500 meters and the average distance between devices is 100 meters, the sparsity is 100 / 500 = 0.2. The adjustment rule for the bandwidth of the kernel function is the inverse of the sparsity multiplied by the base bandwidth, which is determined according to the historical data distribution range corresponding to the device type. For example, the base bandwidth of the current transformer is 50 meters, and the sparsity is 0.2, so the adjusted bandwidth is 50 x (1 / 0.2) = 250 meters. The specific calculation method of kernel density estimation is to use a Gaussian kernel function, which has a mathematical form of a negative exponential relationship between height and distance from the device to the center of the grid. For example, if device A is 100 meters away from the center of the grid and the adjusted bandwidth is 250 meters, the Gaussian kernel value is 0.3 x exp(-(100²) / (2 x 250²)), which is approximately 0.3 x 0.939 = 0.282.

[0069] When generating the regional feature density distribution map based on the device-level feature distribution density, the regional feature density distribution map is represented by a heat map grid. The size of the grid cell of the heat map grid is one-tenth of the preset correlation radius. For example, if the preset correlation radius is 500 meters, the size of the grid cell is 50 meters x 50 meters. The data format of the regional feature density distribution map is a two-dimensional matrix. The row index represents the latitude direction coordinate, and the column index represents the longitude direction coordinate. The origin position is the lower left corner of the region, and the grid cells are arranged in the longitude increasing direction with the column index and in the latitude increasing direction with the row index. For example, if the origin coordinate is (100.00° east, 30.00° north) and the size of the grid cell is 50 meters x 50 meters, then an increase of 1 in the row index corresponds to an increase of 50 meters in the north direction, and an increase of 1 in the column index corresponds to an increase of 50 meters in the east direction. The density value of each grid cell is the weighted sum of the kernel density estimation values of all devices within the corresponding spatial range, and the weight is the spatio-temporal correlation weight of the device within the grid cell. For example, a certain grid cell covers device A (weight 0.3) and device B (weight 0.5), and the kernel density estimation value of device A is 0.8 and that of device B is 1.2. Then the grid density value is 0.3 x 0.8 + 0.5 x 1.2 = 0.84.

[0070] When extracting a high-density area from the area feature density distribution map based on a preset correlation radius, the high-density area is defined as an area where the density values of continuous grid cells exceed a preset density threshold. The preset density threshold is determined by the statistical distribution of density values in historical area-level feature aggregation events, for example, taking the 80th percentile of historical density values as the threshold. If the historical density values are [0.5, 0.8, 1.0, 1.2], the 80th percentile is 1.0, and the threshold is set to 1.0. The judgment criterion for continuous grid cells is that the density values of adjacent grids (including up, down, left, right, and diagonal directions) all exceed the threshold. For example, if the density value of a certain grid cell is 1.1, and 5 of the 8 adjacent grids exceed 1.0, it is determined to be part of a high-density area. If there is no grid cell in the area feature density distribution map that exceeds the threshold, it is determined that there is no high-density area in the current period, and the reconstruction step is skipped.

[0071] When the dynamic feature set in the high-density area is fused by weighting according to the spatio-temporal correlation weight, the dynamic feature set includes abnormal waveform segments and steady-state parameter offsets. The weighted fusion method is to multiply the dynamic feature value by the spatio-temporal correlation weight of the corresponding device, then accumulate, and finally divide by the total weight sum. For example, the high-density area contains device A (weight 0.3), device B (weight 0.5), and device C (weight 0.2), the abnormal waveform amplitude of device A is 10, that of device B is 15, and that of device C is 8, then the weighted fusion value is (10x0.3+15x0.5+8x0.2) / (0.3+0.5+0.2)=12.1. The data structure of the reconstructed area-level feature aggregation package is a structured record containing timestamp, geographic range, and weighted fusion feature value. The timestamp accuracy is consistent with the dynamic feature set (1 microsecond), and the geographic range is the minimum bounding rectangle coordinates of the high-density area. The minimum bounding rectangle coordinates of the high-density area are determined by the following steps: traverse the latitude and longitude coordinates of all grid cells in the high-density area, record the minimum latitude, maximum latitude, minimum longitude, and maximum longitude, and the minimum bounding rectangle vertex coordinates are (minimum longitude, minimum latitude) and (maximum longitude, maximum latitude). For example, a high-density area covers longitude 100.00° to 100.05° and latitude 30.00° to 30.03°, then the minimum bounding rectangle vertex is (100.00°, 30.00°) and (100.05°, 30.03°). If the high-density area contains multiple discrete sub-areas, an independent feature aggregation package is generated for each sub-area.

[0072] When the device spatial distribution is sparse to the extent of 0 (i.e., all devices are concentrated at the same point), the kernel function bandwidth is automatically set to 1% of the preset correlation radius (e.g., 500 meters x 1% = 5 meters), avoiding distortion of the density estimation. If the preset correlation radius exceeds the actual device distribution range (e.g., the maximum distance between devices is 300 meters, and the correlation radius is set to 500 meters), the correlation radius is automatically adjusted to 1.2 times the maximum distance between devices (300 x 1.2 = 360 meters). If the total weight sum is 0 in the weighted fusion process (e.g., all device weights are abnormally zeroed), the device number average is used instead of weighted fusion, for example, the feature values of three devices are 10, 15, and 8, and the average is 11. If the minimum bounding rectangle area of the high-density region exceeds 50% of the area covered by the preset correlation radius, the region is split into multiple sub-regions, and the area of each sub-region does not exceed 25% of the area of the correlation radius. For example, the correlation radius of 500 meters corresponds to a coverage area of 0.785 square kilometers, and if the minimum bounding rectangle area is 0.5 square kilometers (more than 50%), it is divided into two sub-regions according to the longitude or latitude interval.

[0073] Based on the current flow direction and transient energy distribution characteristics between the devices in the regional feature aggregation package, the dynamic aggregation weight of the regional feature aggregation package is corrected, which is implemented by the following steps:

[0074] When determining the current flow direction through the phase difference between the currents and the voltage gradient direction, the phase difference between the currents is the phase shift angle of the voltage and current waveforms at the same timestamp. The phase shift angle of the voltage and current waveforms is calculated by the phase angle difference of the fundamental component extracted by Fourier transform, and the sampling frequency of the Fourier transform is 10 kHz and the time window is 1 power frequency cycle (20 ms). For example, the voltage phase angle of device A is 30 degrees and the current phase angle is 45 degrees, so the phase difference is 15 degrees. The voltage gradient direction is determined by the change trend of the voltage amplitude along the topology path between devices, and the topology path is the shortest connection path between devices in the regional physical topology relationship. For example, the topology path from device A to device B passes through two switches, the voltage amplitude of device A is 10 kV, and the voltage amplitude of device B is 9.8 kV, so the voltage gradient direction is from device A to device B. The determination rule of the current flow direction is: if the current phase difference from device A to device B is less than the phase difference from device B to device A, and the voltage gradient direction is consistent with the topology path, then the flow direction is from device A to device B. If the voltage gradient direction conflicts with the topology path (e.g., the voltage of device A is higher than that of device B, but there is no direct connection in the topology path), the flow direction is invalid and no weight adjustment is performed.

[0075] When extracting the transient energy distribution features in the regional level feature aggregation package, the transient energy distribution features are extracted by wavelet packet decomposition to obtain the energy proportion of each device transient waveform in the frequency band. The frequency band division rule of wavelet packet decomposition is dynamically adjusted according to the typical transient frequency range corresponding to the device type, for example, the transient frequency of transformer equipment is concentrated in 0-1000 Hz, and the traveling wave distance measuring device is concentrated in 1000-3000 Hz. For transformer equipment, the transient waveform is equally divided into two sub-frequency bands (0-500 Hz, 500-1000 Hz); for the traveling wave distance measuring device, it is equally divided into three sub-frequency bands (1000-1500 Hz, 1500-2000 Hz, 2000-2500 Hz). The boundary value of the frequency band division is determined by the peak value distribution of the frequency spectrum of the historical transient data of the device, for example, the historical frequency spectrum peak value of a certain transformer is at 300 Hz and 800 Hz, and the frequency band division is 0-300 Hz, 300-800 Hz, and 800-1000 Hz. The calculation method of the frequency band energy proportion is the ratio of the energy value of each sub-frequency band to the total energy value, for example, the energy of device A in the 0-300 Hz frequency band is 80 J, and the total energy is 200 J, so the frequency band energy proportion is 40%. The transient energy distribution features of all devices in the regional level feature aggregation package are stored as a frequency band energy matrix according to the frequency band classification, the matrix row index corresponds to the device number, the column index corresponds to the frequency band number, and the matrix element value is the frequency band energy proportion.

[0076] When adjusting the initial allocation proportion of the dynamic aggregation weight between devices according to the current flow direction, if the current flow direction is from device A to device B, the weight proportion of device A is increased, and the weight proportion of device B is reduced. The adjustment amplitude is a function of the consistency of the current phase difference and the voltage gradient direction angle, and the consistency function is calculated by the cosine value of the current phase difference and the voltage gradient direction angle. For example, the current phase difference from device A to device B is 15 degrees, and the angle between the voltage gradient direction and the topology path is 10 degrees, so the consistency function value is cos(10 degrees)≈0.985, and the phase difference weight coefficient is determined by the regression analysis of the phase difference and the weight adjustment amplitude in the historical current flow event (such as the fitting coefficient 0.1), and the adjustment amplitude is 0.985×0.1=9.85%. The initial allocation proportion is set according to the historical fault contribution degree of the device, for example, the initial weight of device A is 0.3, and the initial weight of device B is 0.5, and after adjustment, the weight of device A is 0.3×1.0985≈0.329, and the weight of device B is 0.5×0.9015≈0.451. If there is no historical fault data of the device, the initial weight is allocated according to the proportion of the device capacity, for example, the capacity of device A is 100 MVA, and the capacity of device B is 150 MVA, then the initial weight is 100 / (100+150)=0.4, and the initial weight of device B is 0.6.

[0077] When the adjusted weight is normalized and corrected according to the transient energy distribution characteristics, the correction coefficient is the ratio of the transient energy proportion of the device to the average transient energy proportion of the region. The average transient energy proportion of the region is calculated by the average of the energy proportions of all devices in the same frequency band in the frequency band energy matrix. For example, in the 0-300Hz frequency band, the proportion of device A is 40%, the proportion of device B is 35%, and the proportion of device C is 45%, and the average proportion of the region is (40+35+45) / 3=40%. The correction coefficient is calculated by dividing the energy proportion of the device by the average proportion of the region. For example, the correction coefficient of device A is 40% / 40%=1.0, and the correction coefficient of device B is 35% / 40%=0.875. The final dynamic aggregation weight is the product of the adjusted weight and the correction coefficient. For example, the final weight of device A is 0.329*1.0=0.329, and the final weight of device B is 0.451*0.875=0.394. The upper and lower limits of the correction coefficient are determined by the 99% confidence interval of the transient energy proportion of the device in the historical data. For example, if the historical proportion distribution is 5%-200%, the upper limit is set to 2.0, and the lower limit is set to 0.5. If the proportion of the device is more than 200%, the correction coefficient is set to 2.0. If it is less than 5%, the correction coefficient is set to 0.5.

[0078] The sum of the dynamic aggregation weight after normalization and correction is forced to be 1. For example, the weight of device A is 0.329, the weight of device B is 0.394, and the weight of other devices is 0.277, and the sum is 1.0. If the energy of a certain frequency band after wavelet packet decomposition is 0 (such as a device without transient waveform), the energy proportion of the frequency band is set to 50% of the average value to avoid abnormal correction coefficient calculation. When the phase difference between the currents of the devices is 0 degrees (i.e., the voltage and current are in phase), the default power flow direction is the voltage gradient direction. If the voltage gradient direction is invalid, the weight is redistributed according to the capacity proportion of the device.

[0079] According to the corrected dynamic aggregation weight and the regional level feature aggregation package, the preset alarm level template is matched, and the device positioning instruction and the regional scheduling strategy combination are output. The specific implementation is as follows:

[0080] When the corrected dynamic aggregation weight matches the preset alarm level template based on the device spatial distribution density in the regional level feature aggregation package, the alarm level template is divided by the joint distribution threshold of the dynamic aggregation weight and the device spatial density in the historical fault event. The dynamic aggregation weight and the device spatial density data of the historical fault event are distributed in a two-dimensional scatter plot, which is divided into three alarm level intervals by the K-means clustering algorithm. The specific implementation of the K-means clustering algorithm is as follows: three initial cluster centers are randomly selected from the historical data, the Euclidean distance of each data point to the center is iteratively calculated and the category is re-assigned until the center point is stable. The first alarm interval corresponds to the dynamic aggregation weight > 0.5 and the device spatial density > 1.2, the second alarm interval is the dynamic aggregation weight 0.3-0.5 and the device spatial density 0.8-1.2, and the third alarm interval is the dynamic aggregation weight < 0.3 or the device spatial density < 0.8. For example, in a certain historical fault event, the device dynamic aggregation weight is 0.6 and the device spatial density is 1.5, which falls into the first alarm interval. When matching in real time, if the corrected dynamic aggregation weight is 0.55 and the device spatial density is 1.3, it is determined to be a first alarm. If the device spatial density data is missing, the average density of the grid cell where the device is located and the adjacent three grids in the regional level feature aggregation package is used to replace it, for example, the device is located in a grid with a density of 1.0, and the adjacent three grids have densities of 1.1, 1.0 and 0.9, then the device spatial density is (1.0+1.1+1.0+0.9) / 4=1.0.

[0081] When generating the device positioning instruction according to the alarm level, the device positioning instruction contains the spatial coordinates of the high-dynamic aggregation weight device and the associated topological path. The spatial coordinates are calculated by the geographic range in the regional level feature aggregation package and the device level feature distribution density weighted centroid. The calculation method of the weighted centroid is: converting the geographic coordinates (longitude, latitude) of the device into decimal system numbers, multiplying them by the corrected dynamic aggregation weight respectively and accumulating, and then dividing by the total weight sum. For example, the coordinates of device A are 100.00 degrees east and 30.00 degrees north, and the weight is 0.3; the coordinates of device B are 100.01 degrees east and 30.01 degrees north, and the weight is 0.5. The total weight sum is 0.8, the weighted centroid longitude is (100.00*0.3+100.01*0.5) / 0.8=100.00625 degrees, and the latitude is (30.00*0.3+30.01*0.5) / 0.8=30.00625 degrees. The conversion rule of the geographic coordinates in the decimal system is: if the original coordinates are in the degree-minute-second format (such as 100 degrees 15 minutes 30 seconds east), it is converted to 100+15 / 60+30 / 3600≈100.2583 degrees. The associated topological path is the shortest path from the weighted centroid to the nearest key node (such as a substation) in the regional physical topological relationship. The shortest path is calculated by Dijkstra algorithm, and the path length is based on the actual physical connection number between devices. For example, the shortest path from the centroid to the substation passes through two switches and a power transmission line, and the path length is 3.

[0082] When generating the regional scheduling strategy combination based on the alarm level and the device positioning instruction, the strategy combination includes the load reduction ratio, the voltage adjustment priority, and the standby line switching sequence. The load reduction ratio is dynamically allocated according to the power proportion of the high-dynamic aggregation weight device, the power proportion being the ratio of the real-time power of the device to the regional total power, and the regional total power being the sum of the real-time powers of all high-dynamic aggregation weight devices. For example, the power of device A is 10 MW, the power of device B is 15 MW, and the regional total power is 25 MW. The proportion of device A is 40% (10 / 25), and the proportion of device B is 60% (15 / 25). If the alarm level is level one, the load reduction ratio is set to 1.5 times the proportion, the reduction ratio of device A is 40% x 1.5 = 60%, and the reduction ratio of device B is 60% x 1.5 = 90%. If the maximum allowed reduction ratio of the device is 80%, the actual execution of device B is 80%. The voltage adjustment priority is sorted by the consistency of the gradient direction of the device voltage and the power flow path, and the consistency is measured by the cosine value of the included angle between the gradient direction of the device voltage and the power flow path. The cosine value is calculated as follows: the dot product of the gradient direction vector of the device and the power flow path vector divided by the product of the lengths of the two vectors. For example, the gradient direction vector of device A is (1, 0), the power flow path vector is (0.98, 0.17), the included angle is 10 degrees, and the cosine value is approximately 0.985. The vector of device B is (0.87, 0.5), the power flow path vector is (0.98, 0.17), the included angle is 30 degrees, and the cosine value is approximately 0.866. Therefore, the priority of device A is higher than that of device B. The standby line switching sequence is sorted according to the electrical distance of the topological path, and the electrical distance is determined by the shortest path length in the regional physical topological relationship. The shorter the distance, the higher the priority. For example, the electrical distance of path A is 3, and the electrical distance of path B is 5. Path A is preferentially switched.

[0083] When there is no matching interval for the alarm level template (e.g., the dynamic aggregation weight is 0.2 and the density is 1.3), the lowest alarm level (level 3 alarm) is triggered by default. If the total weight sum is 0 when calculating the weighted centroid (e.g., all device weights are abnormally zeroed), the centroid coordinates are set as the geographic center of the region. The geographic center of the region is calculated by aggregating the geographic boundaries of all grid cells in the region-level feature package, and the average of the minimum longitude, maximum longitude, minimum latitude, and maximum latitude is taken. For example, the minimum longitude is 100.00 degrees, the maximum longitude is 100.05 degrees, the minimum latitude is 30.00 degrees, and the maximum latitude is 30.03 degrees. The geographic center is at longitude (100.00+100.05) / 2=100.025 degrees and latitude (30.00+30.03) / 2=30.015 degrees. If the load reduction ratio calculation result exceeds the maximum reducible capacity of the device (e.g., device A has a maximum allowed reduction of 50%), the maximum capacity is executed. For example, the calculation result is 60%, and the actual execution is 50%. When calculating the voltage adjustment priority, if the cosine value of all devices is less than 0.5, the voltage gradient amplitude is sorted from high to low in absolute value. For example, the gradient amplitude of device A is 2 kV / m, and the gradient amplitude of device B is 1 kV / m. Device A is adjusted first. If all standby lines are unavailable (e.g., paths A and B are both faulty), the load is distributed to the remaining lines in proportion to their capacities, e.g., line C has a capacity of 50 MW and line D has a capacity of 30 MW. The load distribution ratio is 50:(50+30)=5:8.

[0084] The embodiment realizes the technical effect of cross-level semantic consistency through the deep integration of multi-level data aggregation and power grid physical characteristics. Existing hierarchical aggregation mechanisms usually rely on preset statistical rules and cannot solve the defects of high-frequency transient signal equalization loss and cross-device causal logic rupture. In the data preprocessing stage, the embodiment introduces time label hierarchical alignment and device type differential filling rules to ensure the time reference of multi-source data and the compatibility of device heterogeneity, providing accurate input for subsequent feature correlation. In the dynamic feature screening process, the threshold is dynamically adjusted based on the device historical working condition and real-time fluctuation amplitude, replacing the traditional fixed threshold filtering, so that the extraction of abnormal waveform fragments and steady-state deviations adapts to the power grid dynamic operation scenario. In the semantic correlation link, the reactive loop path sensitivity and regional physical topology relationship are innovatively combined, the electrical interaction strength between devices is quantified through mutual information entropy calculation, and the spatio-temporal correlation weight is generated. In feature clustering and cross-level analysis, the transient energy distribution characteristics and current flow direction are coupled through kernel density estimation, and the aggregation weight is dynamically corrected, so that the regional-level data reconstruction not only retains the device-level details but also reflects the real-time operation state of the power grid. The final decision generation is based on the matching logic of dynamic weight and alarm template, which directly maps data features to dispatching actions to form a closed-loop link from data dimension reduction to control instructions. Through the cross-domain coupling of electrical physical laws and data analysis models, the contradiction between semantic distortion and causal rupture in traditional hierarchical aggregation is solved.

[0085] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and threshold values in the calculations are set by those skilled in the art according to actual conditions.

[0086] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0087] Those skilled in the art can appreciate that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application of the technical solutions and the constraints of the invention. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0088] In addition, the functional modules in each of the embodiments of the present application can be integrated in one processing module, or each module can exist physically independently, or two or more modules can be integrated in one module.

[0089] In the several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the modules is only a logical function division. There can be another division manner in actual implementation, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the modules shown or discussed can be indirect coupling or communication connection through some interface, device or module, and can be electrical, mechanical or other forms.

[0090] The above describes only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0091] Finally, the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A big data based power grid data acquisition and analysis system, characterized in that, Comprise: Device acquisition module: obtain the original data of power grid equipment end and perform standardized filling and time label hierarchical alignment processing to generate standardized device-level operation data; Feature extraction module: screen dynamic feature set from standardized device-level operation data, dynamic feature set includes abnormal waveform segment and steady-state parameter offset in preset period; Semantic association module: determine whether the dynamic feature set triggers a cross-device associated event, if triggered, generate a time-space association weight of the dynamic feature set based on regional physical topology relationship and reactive current loop path sensitivity, including: Determine whether the cross-device associated event triggering condition is met based on the electrical distance between devices and the time overlap degree of dynamic feature set, the electrical distance is calculated according to the shortest connection path length between devices in the regional physical topology relationship, and the time overlap degree is the time window overlap ratio of abnormal waveform segment or steady-state parameter offset; If the cross-device associated event is triggered, adjust the time-space association weight according to the reactive current loop path sensitivity, the reactive current loop path sensitivity is calculated by the ratio of the inter-device reactive current loop mutual information entropy to the topological distance, and the initial value of the time-space association weight is the inverse of the electrical distance between devices, and the final time-space association weight is the initial value of the time-space association weight multiplied by the reactive current loop path sensitivity correction coefficient; Feature clustering module: according to the time-space association weight and the device-level feature distribution density in the predefined period, the dynamic feature set within the preset association radius is combined and reconstructed into a regional-level feature aggregation package; Cross-layer analysis module: based on the current flow direction and transient energy distribution characteristics between the devices belonging to the regional-level feature aggregation package, correct the dynamic aggregation weight of the regional-level feature aggregation package; Decision generation module: according to the corrected dynamic aggregation weight and the regional-level feature aggregation package, match the preset alarm level template and output the device positioning instruction and regional scheduling strategy combination.

2. The big data based power grid data acquisition and analysis system of claim 1, wherein, Obtain the original data of power grid equipment end and perform standardized filling and time label hierarchical alignment processing to generate standardized device-level operation data, including: Based on the preset filling rule corresponding to the device type, the missing value of the obtained original data of power grid equipment end is standardized and filled; According to the device type and signal sampling frequency, determine the time label hierarchical alignment strategy, align the time label according to the cycle period for power frequency devices, and align the time label according to the microsecond level time window for high frequency sampling devices; Integrate the filled and aligned data into standardized device-level operation data according to the unified time axis.

3. The big data based power grid data acquisition and analysis system of claim 1, wherein, Screen dynamic feature set from standardized device-level operation data, including: Extract abnormal waveform segment based on sliding window dynamic detection method, the sliding window period is consistent with the preset period corresponding to the device type, when the data fluctuation amplitude in the window exceeds the dynamic threshold, it is marked as an abnormal waveform segment, the dynamic threshold is adjusted according to the peak-valley difference mean value of normal waveform in device historical operation data; The steady-state parameter offset is determined based on a steady-state baseline calculation method, steady-state baseline values of each timestamp in a preset period are extracted from standardized device-level operation data, the steady-state baseline value is a historical operation data mean value of the same device in an abnormal period, an absolute deviation of current data from the steady-state baseline value is calculated as an offset, and when the offset exceeds a preset offset threshold, it is determined as an effective steady-state parameter offset.

4. The big data based power grid data acquisition and analysis system of claim 1, wherein, According to the space-time correlation weight and the device-level feature distribution density in the predefined period, the dynamic features within the preset correlation radius are combined and reconstructed into a regional-level feature aggregation package, including: The device-level feature distribution density is calculated based on a kernel density estimation method, and the kernel function bandwidth is adjusted according to the device spatial distribution sparsity, which is determined by the ratio of the average distance between devices to the preset correlation radius; A regional feature density distribution map is generated according to the device-level feature distribution density, and the regional feature density distribution map is represented by a heat map grid; A high-density region is extracted from the regional feature density distribution map based on the preset correlation radius, and the high-density region is a region with a continuous grid cell density value exceeding a preset density threshold; The dynamic features within the high-density region are weighted and fused according to the space-time correlation weight, and are reconstructed into a regional-level feature aggregation package, and the weighted fusion method is to multiply the dynamic feature value by the space-time correlation weight of the corresponding device, then accumulate, and finally divide by the total weight.

5. The big data based power grid data acquisition and analysis system of claim 4, wherein, The grid cell density value is the weighted sum of the device-level feature distribution density within the corresponding spatial range, the weight is the space-time correlation weight of the device within the corresponding unit, and the preset density threshold is determined according to the statistical distribution of the density value in the historical regional-level feature aggregation event.

6. The big data based power grid data acquisition and analysis system of claim 1, wherein, Based on the current direction of the current and the transient energy distribution characteristics between the devices in the regional-level feature aggregation package, the dynamic aggregation weight of the regional-level feature aggregation package is corrected, including: The current direction is determined by the phase difference of the current and the voltage gradient direction between the devices; The transient energy distribution characteristics in the regional-level feature aggregation package are extracted, and the transient energy distribution characteristics are extracted by wavelet packet decomposition to obtain the frequency band energy proportion of each device transient waveform, and the frequency band energy proportion is the ratio of the specified frequency band energy to the total energy; The initial allocation proportion of the dynamic aggregation weight between the devices is adjusted according to the current direction, and the adjustment amplitude is a function of the consistency of the current direction and the topological path; The adjusted weight is normalized and corrected in combination with the transient energy distribution characteristics, and the correction coefficient is the ratio of the device transient energy proportion to the regional average transient energy proportion, and the final dynamic aggregation weight is the product of the adjusted weight and the correction coefficient.

7. The big data based power grid data acquisition and analysis system of claim 6, wherein, The current phase difference is the phase shift angle of the voltage and current waveforms of the device at the same timestamp, and the voltage gradient direction is the change trend of the voltage amplitude along the topological path between the devices.

8. The big data based power grid data acquisition and analysis system of claim 1, wherein, According to the corrected dynamic aggregation weight and the regional-level feature aggregation package, a preset alarm level template is matched, and a device positioning instruction and a regional scheduling strategy combination are output, including: The corresponding alarm level in the preset alarm level template is matched based on the corrected dynamic aggregation weight and the device spatial distribution density in the regional-level feature aggregation package, and the alarm level template is divided by the joint distribution threshold of the dynamic aggregation weight and the device spatial density in the historical fault event; The device positioning instruction is generated according to the alarm level, and the device positioning instruction comprises spatial coordinates of a high-dynamic aggregation-weighted device and an associated topological path, and the spatial coordinates are calculated by a geographical range in a regional-level feature aggregation package and a device-level feature distribution density weighted centroid; The regional scheduling strategy combination is generated based on the alarm level and the device positioning instruction.

9. The big data based power grid data acquisition and analysis system of claim 8, wherein, The regional scheduling strategy combination comprises a load reduction ratio, a voltage adjustment priority and a standby line switching sequence, the load reduction ratio is dynamically allocated according to a power proportion of the high-dynamic aggregation-weighted device, and the voltage adjustment priority is sorted by consistency of a device voltage gradient direction and a power flow path.

Images