A SF6 gas pressure trend early warning method and system based on a grey prediction algorithm

Abstract

The application discloses a kind of SF6 gas pressure trend early warning method and system based on grey prediction algorithm, method includes: the gas pressure historical data of SF6 electrical equipment is collected and pretreated;The gas pressure historical data after pretreatment is trend forecasted using grey prediction model, and the gas pressure prediction data of future preset length is generated;According to the preset multi-stage early warning threshold, the gas pressure prediction data is compared, and the graded early warning information is generated;Graded early warning information is sent to alarm service and is shown and circulates;Gas pressure prediction data is written into sampling library, and the latest prediction value is updated in device monitoring table.The application introduces grey system theory into SF6 pressure monitoring field, and can issue early warning before pressure drops to dangerous line, to provide research and judgment basis for field operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault early warning technology, and in particular to an SF6 gas pressure trend early warning method and system based on a grey prediction algorithm. Background Technology

[0002] SF6 gas, with its excellent insulation and arc-quenching properties, is widely used as an insulation and arc-quenching medium in primary equipment such as high-voltage circuit breakers and GIS switchgear. Once a leak occurs during operation, the internal gas pressure of the equipment continuously decreases. When this pressure reaches a certain level, it directly threatens the insulation strength and breaking capacity, and in severe cases, can lead to breakdown or explosion accidents. Therefore, SF6 pressure changes are one of the key indicators reflecting the health status of equipment, and continuous monitoring and trend analysis are important means of condition-based maintenance.

[0003] Currently, in actual substation operation and maintenance, SF6 pressure monitoring commonly uses a fixed threshold alarm method: the backend system periodically collects pressure telemetry values, and triggers an alarm when the value drops to a pre-set alarm threshold. This method has certain drawbacks: when an alarm occurs, the pressure is already at a low level, and maintenance personnel often need to rush to the site to handle the situation after receiving the notification, lacking buffer time; moreover, the normal pressure range varies for different substations and equipment of different voltage levels, and a uniform threshold is difficult to adapt to all scenarios, resulting in frequent false alarms and missed alarms. More importantly, single-point threshold judgment does not involve trend analysis at all; whether the pressure is decreasing steadily or rapidly, maintenance personnel have no way of knowing, and therefore cannot determine the urgency of the fault.

[0004] Grey system theory has unique advantages in handling uncertainties with small samples and limited information. Its GM(1,1) model requires only a small amount of historical data to establish exponential trend predictions, resulting in low computational overhead. This makes it particularly suitable for online scenarios like SCADA systems that require periodic batch predictions. However, research has revealed that there are currently no reports of engineering integration between grey prediction and SF6 pressure monitoring and early warning, indicating a technological gap in this area. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a method and system for SF6 gas pressure trend analysis and early warning based on a grey prediction algorithm, which solves the problems of existing SF6 pressure monitoring methods lacking trend prediction capabilities, having delayed early warnings, and being unable to support differentiated operation and maintenance decisions, and enables quantitative early warnings to be given in the early stages of pressure decline.

[0006] Technical solution: The SF6 gas pressure trend early warning method based on gray prediction algorithm of the present invention includes:

[0007] Historical gas pressure data of SF6 electrical equipment were collected according to the collection cycle and duration.

[0008] Historical gas pressure data and preprocessing;

[0009] The grey prediction model is used to predict the trend of preprocessed historical gas pressure data and generate gas pressure prediction data for a preset time period in the future.

[0010] The gas pressure prediction data is compared with the preset multi-level early warning thresholds to generate graded early warning information;

[0011] The tiered early warning information is sent to the alarm service for display and circulation; the gas pressure prediction data is written into the sampling database, and the latest prediction value is updated in the equipment monitoring table.

[0012] Furthermore, the preprocessing of historical gas pressure data includes: marking samples with pressure values ​​less than or equal to a preset value or greater than a preset value as invalid values; filling the data with the average of valid values ​​when the proportion of invalid values ​​does not exceed a preset ratio; and checking the most recent preset number of samples when the proportion exceeds a preset ratio. If invalid values ​​are found, the prediction is terminated and an error code is returned; otherwise, all data is filled with the average of the most recent preset number of valid samples.

[0013] Furthermore, the gray prediction model performs trend prediction by performing a cumulative generation, background value construction, least squares parameter identification, and cumulative subtraction and restoration operation on the preprocessed historical gas pressure data, and iteratively producing full gas pressure prediction data in a rolling prediction manner.

[0014] Furthermore, after each data point is generated, the rolling prediction adds it to the end of the historical input gas pressure data and removes an equal number of old points. Through iteration, prediction points with preset values ​​are generated, and negative gas pressure prediction data are reset to zero.

[0015] Furthermore, the multi-level warning thresholds include the over-limit value, the attention 2 value, and the attention 1 value.

[0016] Furthermore, the method for generating graded early warning information is as follows: traverse the gas pressure prediction data, check the position where the first value is below the threshold according to the priority of exceeding the limit value, attention 2 value, and attention 1 value, generate the corresponding level of early warning and attach the expected time of exceeding the limit; at the same time, supplement the current status alarm according to the current actual value.

[0017] Furthermore, the multi-level warning thresholds satisfy the following conditions: Exceeding limit value < Note 2 value < Note 1 value < Initial pressure value.

[0018] Furthermore, the collection period for historical gas pressure data is a preset time interval, and the collection length is a preset duration, totaling a preset number of sample points.

[0019] An SF6 gas pressure trend early warning system based on a grey prediction algorithm includes:

[0020] The data acquisition module obtains historical gas pressure data from the SCADA system;

[0021] The data processing module cleans and replaces anomalies in the raw data;

[0022] The grey prediction engine module generates future pressure prediction data for a preset duration based on the grey prediction model.

[0023] The multi-level early warning analysis module compares the predicted value with multi-level thresholds such as the limit-breaking value and the attention value and outputs graded early warnings; the alarm service interface module pushes the early warning to the alarm service and updates the database status.

[0024] The prediction data write-back module writes the gas pressure prediction data into the SCADA sampling library to support display and traceability.

[0025] Furthermore, the gray prediction engine module includes a data preprocessing submodule to determine the stability of gas pressure prediction data: when the number of samples deviating from the mean by more than a set multiple accounts for more than a preset proportion of the total, the mean is used to replace the entire data; when all data in the gas pressure prediction are identical, the last value is used to fill all prediction points directly.

[0026] Furthermore, the multi-level early warning analysis module automatically sends a recovery notification when an early warning is restored; the alarm service interface module suppresses early warnings of the same level for the same device to avoid duplicate alarms.

[0027] Beneficial effects: Compared with existing technologies, this invention has the following significant advantages: 1. It introduces gray prediction into the field of SF6 pressure monitoring, shifting the early warning window from post-event alarms to pre-event prediction; 2. It establishes a supporting data cleaning process to systematically process invalid values ​​and anomalies in the telemetry link, reducing the false alarm rate; 3. It adopts a three-level threshold hierarchical early warning strategy, allowing maintenance personnel to respond according to the differences in early warning levels, avoiding a "one-size-fits-all" approach; 4. After the gas pressure prediction data is written back to the SCADA library, trend curves can be directly superimposed on the existing monitoring screen without the need for an additional independent display system. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the complete execution flow of the method of the present invention;

[0029] Figure 2 This is a schematic diagram of the module structure and data flow relationship of the system of the present invention;

[0030] Figure 3 This is a schematic diagram of the calculation of the grey prediction GM(1,1) model of the present invention;

[0031] Figure 4 These are screenshots of the system function interface implemented in an embodiment of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0033] like Figure 1 As shown, the SF6 gas pressure trend analysis and early warning method based on gray prediction algorithm of the present invention includes:

[0034] Collect historical telemetry data of gas pressure from SF6 electrical equipment and perform preprocessing;

[0035] Query the historical pressure data of a specified SF6 device through the sampling data service interface of the SCADA system.

[0036] The historical gas pressure data was collected every hour for 42 days, totaling 1008 sample points.

[0037] Data preprocessing includes: first, marking invalid values ​​by assigning a special mark -99999 to samples with pressure values ​​≤0 or >1MPa; counting the number of invalid samples Ninv; if Ninv does not exceed 30% of the total number of samples, filling all invalid positions with the mean of the valid portion; if Ninv exceeds 30%, checking the last 12 samples of the historical gas pressure data, terminating subsequent processing and returning error code -102 if any invalid value is found in any of them, otherwise filling the entire gas pressure prediction data with the mean of the most recent 12 valid samples.

[0038] After handling invalid values, the mean Xavg of historical gas pressure data is calculated, and each sample is checked for deviation from the mean. If more than 20% of the samples deviate by more than 2 times, it indicates a large number of abnormal fluctuations in the historical gas pressure data, and all data are directly replaced with Xavg. If the deviation exceeds 2 times but the number is less than 20%, only these outliers are replaced. Furthermore, when all sample values ​​are found to be completely identical, the last value can be directly output as the full gas pressure prediction data.

[0039] The grey prediction model was used to predict the trend of preprocessed historical gas pressure data, generating pressure prediction data for the next 7 days, totaling 168 points, with a time resolution of 1 hour.

[0040] The grey prediction model GM(1,1) trend prediction includes:

[0041] Step 1 (1-AGO): Processing pre-processed historical gas pressure data Perform an accumulation generation once to obtain the accumulated gas pressure value. The cumulative calculation formula is as follows:

[0042] ;

[0043] Step 2 (Background Value Construction): Calculate the nearest mean of the accumulated historical gas pressure data. :

[0044]

[0045] This forms a data vector. sum coefficient matrix :

[0046] ,

[0047] Step 3 (Parameter Identification): Solve using the least squares method:

[0048]

[0049] in The development coefficient (reflecting the strength of the growth / decline trend in historical gas pressure data). This is the grey action quantity (reflecting the intrinsic driving force of historical gas pressure data). To avoid division-by-zero errors in subsequent calculations, when... season (like =0, then take );

[0050] Step 4 (Time Response and Cumulative Reduction Restoration): Calculate the predicted value of the cumulative sequence using the time response formula. Substitute:

[0051]

[0052] Calculate the predicted value within the cumulative domain, and then:

[0053]

[0054] Restore the original domain value to obtain the predicted value;

[0055] Step 5 (Rolling Iteration): A single call to GM(1,1) produces only a small number of prediction points (2 in this example). The predicted values ​​are appended to the end of the input gas pressure prediction data, and an equal number of old data points are removed, ensuring that the input window always retains the latest 1008 points. After repeating this process 84 times, the full gas pressure prediction data is produced, resulting in 168 complete prediction points, corresponding to the hourly pressure trend over the next 7 days. Any negative predictions generated in any step are corrected to zero.

[0056] The pressure prediction data is compared with the preset multi-level early warning thresholds to generate graded early warning information;

[0057] The multi-level warning thresholds include the over-limit value (default initial pressure -0.05MPa), the attention 2 value (default initial pressure -0.04MPa), and the attention 1 value (default initial pressure -0.02MPa). The three values ​​must satisfy the condition that the over-limit value < the attention 2 value < the attention 1 value < the initial pressure value. Unconfigured thresholds are automatically set to the initial value minus a fixed offset (value -99999) as the default value.

[0058] The warning information is generated as follows: It iterates through the gas pressure prediction data, i.e., 168 prediction points, checking them according to priority: exceeding limits, Attention 2, and Attention 1. If the predicted value is less than or equal to the exceeding limit value, an alarm of the "Exceeding Limit Warning" level is generated, and the expected date and time of the first exceeding limit point are recorded. If there are no exceeding limits throughout the process, the Attention 2 and Attention 1 values ​​are checked, triggering only the highest priority hit. Simultaneously, the same three-level comparison is performed on the latest actual value, generating three real-time status alarms: "Current Exceeding Limit," "Current Attention 2," and "Current Attention 1," which can be used in combination with the prediction warning.

[0059] The tiered early warning information is sent to the alarm service for display and circulation, and the alarm service is responsible for subsequent user-side presentation such as pop-ups, push images, or SMS notifications.

[0060] Finally, the 168 predicted points are written into the SCADA sampling library, and the latest predicted values ​​are updated in the equipment monitoring table. The SCADA front end can then plot the predicted trend curve for operators' reference.

[0061] The above method completes the process within one processing cycle, which is 5 minutes. At the start of the next cycle, the latest data is collected again and the old results are overwritten, achieving dynamic rolling updates to the predictions.

[0062] See Figure 2 , Figure 4 The SF6 gas pressure trend early warning system based on the grey prediction algorithm of the present invention includes:

[0063] Data Acquisition Module: Queries historical pressure telemetry data for a specified SF6 device via the SCADA system's sampling data service interface. The query interval is fixed at 1 hour, requesting data from the most recent 42 days (1008 points) each time. Searches are performed by device ID and time range. Considering potential data gaps due to signal jitter and communication interruptions, the query results are further cleaned by the data processing module before being used directly.

[0064] Data processing module: First, invalid values ​​are marked. Samples with pressure values ​​≤0 or >1MPa are marked with a special flag -99999. The number of invalid samples, Ninv, is counted. If Ninv does not exceed 30% of the total number of samples, all invalid positions are filled with the mean of the valid portion. If Ninv exceeds 30%, the last 12 samples in the sequence are checked. If any invalid value is found in any of them, subsequent processing is terminated and error code -102 is returned. Otherwise, the mean of the last 12 valid samples is filled with the entire historical gas pressure data. After invalid value processing, the mean Xavg of the historical gas pressure data is calculated. The deviation of each sample from the mean is checked one by one: If the deviation exceeds 2 times and accounts for more than 20% of the total, it indicates that there are a large number of abnormal fluctuations in the sequence. All data are directly replaced with Xavg, sacrificing details for model stability. If the deviation exceeds 2 times but the number is less than 20%, only the outlier points are replaced. In addition, when all sample values ​​are found to be completely consistent, the last value can be directly output as the full gas pressure prediction data without any modeling.

[0065] Grey Prediction Engine Module: Preprocessed historical gas pressure data enters the grey prediction engine module, where GM(1,1) modeling and rolling prediction are performed. The process is as follows: Figure 3 As shown:

[0066] Step 1 (1-AGO): Processing pre-processed historical gas pressure data Perform an accumulation generation once to obtain the accumulated gas pressure value. The cumulative calculation formula is as follows:

[0067] ;

[0068] Step 2 (Background Value Construction): Calculate the nearest mean of the accumulated historical gas pressure data. :

[0069]

[0070] This forms a data vector. sum coefficient matrix :

[0071] ,

[0072] Step 3 (Parameter Identification): Solve using the least squares method:

[0073]

[0074] in The development coefficient (reflecting the strength of the growth / decline trend in historical gas pressure data). This is the grey action quantity (reflecting the intrinsic driving force of historical gas pressure data). To avoid division-by-zero errors in subsequent calculations, when... season (like =0, then take );

[0075] Step 4 (Time Response and Cumulative Reduction Restoration): Calculate the predicted value of the cumulative sequence using the time response formula. Substitute:

[0076]

[0077] Calculate the predicted value within the cumulative domain, and then:

[0078]

[0079] Restore to the original field value;

[0080] Step 5 (Rolling Iteration): A single call to GM(1,1) produces only a small number of prediction points (2 in this example). The predicted values ​​are appended to the end of the historical gas pressure data, and an equal number of old data points are removed, ensuring that the input window always retains the latest 1008 points. After repeating this process 84 times, the full gas pressure prediction data is produced, resulting in 168 complete prediction points, corresponding to the hourly pressure trend over the next 7 days. Any negative predictions generated in any step are corrected to zero.

[0081] The multi-level early warning analysis module is responsible for comparing gas pressure prediction data with thresholds and outputting tiered early warnings. The threshold system includes three levels: Exceeding Limit Value (default initial pressure -0.05MPa), Warning 2 Value (default initial pressure -0.04MPa), and Warning 1 Value (default initial pressure -0.02MPa). These three values ​​must maintain a strictly increasing relationship. If a threshold level is not explicitly specified in the configuration table (value -99999), the default offset value will be automatically used.

[0082] The early warning logic is executed sequentially according to priority: First, it checks for limit violations, iterating through 168 prediction points. If a predicted value is found to be less than or equal to the limit violation value, an alarm of level "Limit Violation Warning" is immediately generated, recording the expected date and time of the first limit violation point. If no limit violations occur throughout the process, it continues to check attention values ​​2 and 1, triggering only the highest priority hit item. Simultaneously, the same three-level comparison is performed on the latest actual value, generating three real-time status alarms: "Current Limit Violation," "Current Attention Value 2," and "Current Attention Value 1," which can be used in combination with prediction warnings.

[0083] Alarm Service Interface Module: Alarm information is sent to the unified alarm service process via the alarm client interface. Each alarm carries the alarm type (using device status analysis type 260), device ID, substation ID, alarm status, and descriptive text. The descriptive text includes the specific date and pressure value predicted for exceeding the limit, such as "Predicted SF6 pressure will reach the warning value of 0.45MPa on May 28, 2026". When the alarm disappears (pressure recovers or no alarms for two consecutive cycles), a recovery notification with the content "normal" is automatically sent. The module internally maintains a suppression table based on device ID and alarm level; alarms of the same device and level are not sent repeatedly within the validity period.

[0084] Predictive data write-back module: The 168 prediction points generated by GM(1,1) are written back to the SCADA sampling library via a batch write interface. When writing, the start time, time interval (1 hour), and complete data array are specified. The SCADA side can then display the prediction data in the historical database as a dashed line or a trend curve overlaid with different colors. In addition, the value of the first prediction point is synchronously updated to the prediction value field of the equipment early warning monitoring table, allowing the latest gas pressure prediction data to be directly presented on the main monitoring page.

[0085] The system runs as a background daemon process. The main loop executes once every 300 seconds, and each loop processes all configured SF6 monitoring devices serially. Data querying, preprocessing, prediction, and early warning judgment for the same device are completed within a single loop. The process registers via the message bus upon startup and releases alarm client resources upon exit.

Claims

1. A method for early warning of SF6 gas pressure trends based on a grey prediction algorithm, characterized in that, include: Historical gas pressure data of SF6 electrical equipment were collected according to the collection cycle and duration. Historical gas pressure data was preprocessed. The grey prediction model is used to predict the trend of preprocessed historical gas pressure data and generate gas pressure prediction data for a preset time period in the future. The gas pressure prediction data is compared with the preset multi-level early warning thresholds to generate graded early warning information; The tiered early warning information is sent to the alarm service for display and circulation; the gas pressure prediction data is written into the sampling database, and the latest prediction value is updated in the equipment monitoring table.

2. The SF6 gas pressure trend early warning method according to claim 1, characterized in that, The preprocessing of historical gas pressure data includes: marking samples with pressure values ​​less than or equal to a preset value or greater than a preset value as invalid values; filling the data with the average of valid values ​​when the proportion of invalid values ​​does not exceed a preset ratio; and checking the most recent preset number of samples when the proportion exceeds a preset ratio. If there are invalid values, the prediction is terminated and an error code is returned. Otherwise, all data is filled with the average of the most recent preset number of valid samples.

3. The SF6 gas pressure trend early warning method according to claim 1, characterized in that, The gray prediction model for trend prediction includes: performing a cumulative generation, background value construction, least squares parameter identification, and cumulative subtraction and restoration operation on the preprocessed historical gas pressure data, and iteratively producing full gas pressure prediction data in a rolling prediction manner.

4. The SF6 gas pressure trend early warning method according to claim 1, characterized in that, The rolling prediction adds data points to the end of the historical input gas pressure data after each generation and removes an equal number of old points. Iteratively, it generates prediction points with preset values, and negative gas pressure prediction data is reset to zero.

5. The SF6 gas pressure trend early warning method according to claim 1, characterized in that, The multi-level warning thresholds include the over-limit value, the attention 2 value, and the attention 1 value.

6. The SF6 gas pressure trend early warning method according to claim 1, characterized in that, The method for generating graded early warning information is as follows: traverse the gas pressure prediction data, check the position where the first value is below the threshold according to the priority of exceeding the limit value, attention 2 value, and attention 1 value, generate the corresponding level of early warning and attach the expected time of exceeding the limit; at the same time, supplement the current status alarm according to the current actual value.

7. The SF6 gas pressure trend early warning method according to claim 1, characterized in that, The historical gas pressure data collection period is a preset time interval, and the collection length is a preset duration, totaling a preset number of sample points.

8. An SF6 gas pressure trend early warning system based on a grey prediction algorithm, characterized in that, include: The data acquisition module obtains historical gas pressure data from the SCADA system; The data processing module cleans and replaces anomalies in the raw data; The grey prediction engine module generates future pressure prediction data for a preset duration based on the grey prediction model. The multi-level early warning analysis module compares the predicted value with multi-level thresholds such as the limit-breaking value and the attention value and outputs graded early warnings; the alarm service interface module pushes the early warning to the alarm service and updates the database status. The prediction data write-back module writes the gas pressure prediction data into the SCADA sampling library to support display and traceability.

9. The SF6 gas pressure trend early warning system according to claim 7, characterized in that, The gray prediction engine module includes a data preprocessing submodule to determine the stationarity of gas pressure prediction data: when the number of samples deviating from the mean by more than a set multiple exceeds a preset proportion of the total, the mean is used to replace the entire data; when all data in the gas pressure prediction are identical, the last value is used to fill all prediction points.

10. The SF6 gas pressure trend early warning system according to claim 7, characterized in that, The multi-level early warning analysis module automatically sends a recovery notification when an early warning is restored; the alarm service interface module suppresses early warnings of the same level for the same device to avoid duplicate alarms.

Images