CN-122020423-A - Abnormal early warning platform and method based on energy data stream calculation

CN122020423ACN 122020423 ACN122020423 ACN 122020423ACN-122020423-A

Abstract

The invention discloses an anomaly early warning platform and method based on energy data stream calculation, wherein the method comprises the steps of obtaining energy data stream and preprocessing to obtain standardized energy data stream; when monitoring data which is larger than a corresponding early warning threshold exists in the standardized energy data flow and the corresponding duration is larger than a preset duration threshold, early warning reminding is directly carried out, otherwise, a first data flow is determined through the standardized data flow, sub-window selection is carried out on the first data flow, sub-windows of all the monitoring data are determined, analysis is carried out according to the sub-windows of all the monitoring data, one or more overlapping intervals are determined, and total early warning score and corresponding early warning grade are respectively determined according to each fault type, and corresponding early warning reminding is carried out. According to the method, through real-time acquisition, intelligent preprocessing and comprehensive analysis of the energy data, timeliness and accuracy of monitoring of the abnormality of the energy system are improved, and energy waste and operation and maintenance risks are reduced.

Inventors

CAO JINWEI
LIN CUNHUI

Assignees

瑞诺技术(深圳)有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. An anomaly early warning method based on energy data stream calculation is characterized by comprising the following steps: Acquiring a historical monitoring data stream, analyzing according to the historical monitoring data stream, and determining associated risk parameters of various fault types and corresponding influence coefficients; Acquiring energy data streams of different monitoring data sent by a multi-source data acquisition terminal, and preprocessing the energy data streams to obtain a plurality of standardized energy data streams; Analyzing the plurality of standardized energy data streams, and directly carrying out early warning reminding when monitoring data which are larger than corresponding early warning thresholds exist in the standardized energy data streams and the corresponding duration time is larger than a preset duration time threshold; otherwise, frame selection is carried out on the standardized data stream through a preset time window, and a first data stream is determined; selecting a sub-window of the first data stream, and determining the sub-window of each monitoring data; Analyzing according to the sub-window of each monitoring data, and determining one or more overlapped sections; And respectively calculating a first early warning score of the corresponding associated risk parameter in the overlapped interval and a second early warning score of other time intervals according to each fault type, accumulating and standardizing calculation results, determining a total early warning score and a corresponding early warning grade, and carrying out corresponding early warning reminding.
2. The anomaly early warning method based on energy data stream calculation according to claim 1, wherein the acquiring the historical monitoring data stream, analyzing according to the historical monitoring data stream, determining associated risk parameters and corresponding influence coefficients of various fault types, comprises: acquiring historical monitoring data flow of each monitoring parameter when various faults occur; Analyzing each fault type in sequence, carrying out anomaly labeling on historical monitoring data streams of each monitoring parameter when the fault type occurs, determining the monitoring parameters with anomaly values in the same time period as sample association parameters, and calculating average anomaly duration of each sample association parameter, wherein the sample association parameters at least consist of two monitoring parameters; Filtering sample association parameters with average abnormal duration less than a preset duration threshold and with number less than a preset sample number threshold, sorting the rest sample association parameters in descending order according to the duration of the average abnormal duration, and sequentially selecting the preset number of sample association parameters to determine as association risk parameters; calculating the triggering frequency of the associated risk parameters to determine a first influence coefficient of the associated risk parameters; Dividing intervals of the same time period in which abnormal values appear simultaneously in the associated risk parameters according to preset time intervals to obtain a plurality of sample associated parameter sub-data streams; Calculating the average anomaly score of each sample associated parameter sub-data stream; Determining a plurality of average abnormal score distribution intervals according to preset score intervals, and counting the distribution proportion of the sub-data streams of the sample association parameters in each average abnormal score distribution interval to determine a second influence coefficient of the association risk parameters in each average abnormal score distribution interval; And respectively analyzing the single monitoring parameters based on the method, and determining a third influence coefficient of each monitoring parameter and a fourth influence coefficient corresponding to each average abnormal score distribution interval when various faults occur.
3. The anomaly early warning method based on energy data stream calculation according to claim 1, wherein the selecting the sub-window of the first data stream to determine the sub-window of each monitoring data comprises: analyzing based on the first data stream, determining the time corresponding to the maximum data in the first data stream as the initial point of a sub-window, respectively calculating the data fluctuation values of the maximum data and the monitoring data on the left side and the right side, and expanding the sub-window in the direction of small data fluctuation values; In the expansion process, calculating the data score of each monitoring data in the sub window, accumulating the data scores of all the monitoring data, and determining the window score of the sub window; Stopping expanding when the window score of the sub-window is larger than or equal to a preset window score threshold value, determining a final sub-window, and otherwise, continuing expanding the sub-window; in the expansion process, if no data with the data fluctuation value smaller than the preset data fluctuation value exists in the preset distance, determining the time of the last expansion as an expansion end point of the current expansion direction; And continuing to select the time corresponding to the next maximum data to select the sub-window until the monitoring data meeting the sub-window selection condition does not exist in the first data stream.
4. The anomaly early warning method based on energy data stream calculation according to claim 1, wherein the analyzing according to the sub-window of each monitoring data, determining one or more overlapping intervals, comprises: Calculating the mutual overlapping rate of sub-windows of different monitoring data based on the time stamp, determining a group of different monitoring data with overlapping rate larger than a first preset overlapping rate threshold value in the mutual overlapping rate as to-be-determined associated monitoring data, and determining the corresponding sub-window as to-be-determined sub-window group; Based on the to-be-determined sub-window group, determining the overlapping part of the time axis as an initial overlapping section, and respectively expanding the initial overlapping section forwards and backwards according to the mutual overlapping rate between sub-windows of the to-be-determined associated monitoring data according to the time axis to determine a final overlapping section; in the interval expansion process, a sub-window with a longer current expansion direction time axis is defined as a first sub-window, and a sub-window with a shorter current expansion direction time axis is defined as a second sub-window.
5. The anomaly early warning method based on energy data stream computation according to claim 4, wherein the expanding the initial overlapping section forward and backward according to the mutual overlapping rate between sub-windows of the monitor data to be correlated in time axis, determining the final overlapping section, comprises: S501, when the mutual overlapping rates among the sub-windows of the undetermined associated monitoring data are all smaller than a second preset overlapping rate threshold value, determining an initial overlapping interval as a final overlapping interval; S502, when the mutual overlapping rate between the sub-windows of the undetermined associated monitoring data is larger than or equal to a second preset overlapping rate threshold value, determining a first non-overlapping interval according to the non-overlapping parts of the first sub-window and the initial overlapping interval; S503, calculating the maximum data fluctuation value of the monitoring data of the monitoring parameters corresponding to the second sub-window in the first non-overlapping interval; S504, when the maximum data fluctuation value is smaller than a preset fluctuation value threshold, adding the first non-overlapping section into the initial overlapping section to complete section expansion in the current expansion direction; S505, otherwise, traversing the first non-overlapping interval direction based on the time point between the initial overlapping interval and the first non-overlapping interval, and calculating the maximum data fluctuation value of the monitoring data of the monitoring parameters corresponding to the second sub-window in the traversing process; And S506, finishing traversing when the maximum data fluctuation value is smaller than the product of a preset fluctuation value threshold value and a fluctuation value adjustment coefficient, adding a first non-overlapping section in the traversing section into the initial overlapping section, and completing section expansion in the current expansion direction.
6. The anomaly early warning method based on energy data stream computation of claim 5, further comprising: When the overlapping rate of only one sub-window among the sub-windows of the undetermined associated monitoring data is larger than or equal to a second preset overlapping rate threshold value, aiming at the extending direction that the overlapping rate of the second sub-window is larger than or equal to the second preset overlapping rate threshold value, performing interval extending according to the steps S502-S506; For the other expansion direction, when the second sub-window does not have an adjacent sub-window, the interval expansion is carried out according to the steps S502-S506, otherwise, whether the first sub-window has the adjacent sub-window is judged; if so, ending the section expansion in the current expansion direction, and determining a new initial overlapping section based on the first non-overlapping section.
7. The anomaly early warning method based on energy data stream computation of claim 5, further comprising: And dynamically adjusting the fluctuation value adjusting coefficient according to the traversal distance, setting the fluctuation value adjusting coefficient of the first traversal time point as a preset minimum fluctuation value adjusting coefficient in the traversal process towards the first non-overlapping interval direction, setting the fluctuation value adjusting coefficient of the last traversal time point as a preset maximum fluctuation value adjusting coefficient, and enabling the fluctuation value adjusting coefficients of other traversal time points in the first non-overlapping interval to meet the linear trend.
8. The anomaly early warning method based on energy data stream computation according to claim 2, wherein for each fault type, respectively computing a first early warning score of a corresponding associated risk parameter in an overlapping interval and a second early warning score of other time intervals, accumulating the computation results, and determining a total early warning score and a corresponding early warning level, including: Analyzing each fault type in sequence, respectively calculating average abnormal scores of all overlapping intervals of the associated risk parameters in the first data stream, and determining corresponding second influence coefficients; Respectively calculating the product of the time length of each overlapping interval, the corresponding second influence coefficient and the first influence coefficient of the associated risk parameter, accumulating the calculation results, and determining the first early warning score of the total overlapping interval; Calculating second early warning scores of the associated risk parameters in other time intervals except the overlapping interval in the first data stream according to the third influence coefficient corresponding to the fault type in the associated risk parameters, the fourth influence coefficient corresponding to each average abnormal score distribution interval and the time length; and calculating the sum of the first early warning score and the second early warning score, and determining the total early warning score and the corresponding early warning grade.
9. The anomaly early warning method based on energy data stream computation of claim 1, further comprising: Setting an analysis time interval, and updating the frame selection position of the first data stream; judging whether the time corresponding to the maximum data in the updated first data stream changes or not; if the change occurs, all sub-windows are cleared, and sub-window division is carried out again based on the updated first data stream; If no change occurs, judging whether the frame selection ending time before updating is the ending time of the last sub-window; If not, not carrying out sub-window cleaning; If yes, judging whether the maximum data of the last sub-window is larger than the maximum data of the undivided part of the updated first data stream; if the window is larger than the window, removing the sub-window is not performed; If the data is smaller than or equal to the maximum data, clearing the last sub-window, continuing to traverse forward until no adjacent sub-window exists or the maximum data of the adjacent sub-window is larger than the maximum data of the non-divided part of the updated first data stream, ending the traversing, and dividing the sub-window of the non-divided part of the updated first data stream.
10. An anomaly early warning platform based on energy data stream calculation for implementing the anomaly early warning method based on energy data stream calculation as claimed in any one of claims 1 to 9, comprising: The associated risk parameter analysis module is used for acquiring a historical monitoring data stream, analyzing according to the historical monitoring data stream and determining associated risk parameters of various fault types and corresponding influence coefficients; The data acquisition and preprocessing module is used for acquiring energy data streams of different monitoring data sent by the multi-source data acquisition terminal, and preprocessing the energy data streams to obtain a plurality of standardized energy data streams; The first abnormal early warning module is used for analyzing the plurality of standardized energy data streams, and when monitoring data which is larger than a corresponding early warning threshold value exists in the standardized energy data streams and the corresponding duration time is larger than a preset duration time threshold value, early warning reminding is directly carried out; The data stream analysis module is used for carrying out frame selection on the standardized data stream through a preset time window to determine a first data stream, carrying out sub-window selection on the first data stream to determine sub-windows of all monitoring data, analyzing according to the sub-windows of all monitoring data, and determining one or more overlapped sections; The second abnormal early warning module is used for respectively calculating a first early warning score of the corresponding associated risk parameter in the overlapped interval and a second early warning score of other time intervals according to each fault type, accumulating and standardizing calculation results, determining a total early warning score and a corresponding early warning grade, and carrying out corresponding early warning reminding.

Description

Abnormal early warning platform and method based on energy data stream calculation Technical Field The application relates to the technical field of energy data early warning, in particular to an abnormal early warning platform and method based on energy data stream calculation. Background With the deep digital transformation of the energy industry, the large-scale sensor deployment is realized in the scenes of wind power, photovoltaic, power grid, industrial energy consumption and the like, and the high-concurrency, strong time sequence and multi-source heterogeneous energy data stream is formed. The data directly reflects the running state of the full link of energy production, transmission and consumption, timely detects data abnormality and gives early warning, and has important significance for guaranteeing the stable running of an energy system, reducing energy consumption loss and preventing equipment faults. The conventional energy data abnormality early warning method has the defects that firstly, the real-time requirement of streaming data cannot be adapted to a traditional batch processing mode, the end-to-end delay is too high, abnormal early warning is delayed, second-level response scenes such as wind control and equipment operation and maintenance are difficult to meet, secondly, the abnormality detection algorithm adopts a single threshold value or a fixed model, the characteristic that the energy data dynamically changes along with a production plan and environmental parameters (air temperature and humidity) cannot be adapted, the problems of misinformation and misinformation are easy to occur, thirdly, the fusion degree of multi-source energy data (such as electrical parameters, environmental parameters and equipment state data) is insufficient, and under most conditions, the data abnormality of various energy data exists before equipment failure, and the delay exists possibly only through single energy data for early warning. Therefore, the prior art has defects, and improvement is needed. Disclosure of Invention In view of the above problems, the invention aims to provide an anomaly early warning platform and an anomaly early warning method based on energy data stream calculation, which realize real-time acquisition, intelligent preprocessing and comprehensive analysis of multi-energy data of energy data, improve timeliness and accuracy of anomaly monitoring of an energy system and reduce energy waste and operation and maintenance risks. The first aspect of the invention provides an anomaly early warning method based on energy data stream calculation, which comprises the following steps: Acquiring a historical monitoring data stream, analyzing according to the historical monitoring data stream, and determining associated risk parameters of various fault types and corresponding influence coefficients; Acquiring energy data streams of different monitoring data sent by a multi-source data acquisition terminal, and preprocessing the energy data streams to obtain a plurality of standardized energy data streams; Analyzing the plurality of standardized energy data streams, and directly carrying out early warning reminding when monitoring data which are larger than corresponding early warning thresholds exist in the standardized energy data streams and the corresponding duration time is larger than a preset duration time threshold; otherwise, frame selection is carried out on the standardized data stream through a preset time window, and a first data stream is determined; selecting a sub-window of the first data stream, and determining the sub-window of each monitoring data; Analyzing according to the sub-window of each monitoring data, and determining one or more overlapped sections; And respectively calculating a first early warning score of the corresponding associated risk parameter in the overlapped interval and a second early warning score of other time intervals according to each fault type, accumulating and standardizing calculation results, determining a total early warning score and a corresponding early warning grade, and carrying out corresponding early warning reminding. In this scheme, obtain the historical monitoring data stream, analyze according to the historical monitoring data stream, confirm the associated risk parameter and the corresponding influence coefficient of various fault types, include: acquiring historical monitoring data flow of each monitoring parameter when various faults occur; Analyzing each fault type in sequence, carrying out anomaly labeling on historical monitoring data streams of each monitoring parameter when the fault type occurs, determining the monitoring parameters with anomaly values in the same time period as sample association parameters, and calculating average anomaly duration of each sample association parameter, wherein the sample association parameters at least consist of two monitoring parameters; Filtering sample association parameters with