Unmanned hydroelectric generating set dynamic monitoring fault early warning method based on edge computing

Abstract

The application provides an unmanned hydroelectric generating set dynamic monitoring fault early warning method based on edge computing, and relates to the technical field of hydroelectric power plant equipment fault early warning.The steps of the application are as follows: S1, collecting multi-source data of hydroelectric equipment and identifying the running state;S2, calculating the hydroelectric working condition correction factor;S3, determining the early warning threshold of the hydroelectric parameter;S4, performing parameter anomaly determination and calculating the comprehensive anomaly score;S5, performing hierarchical early warning according to the comprehensive anomaly score correction value and performing edge side linkage disposal.The beneficial effects are as follows: the early warning threshold uses dynamic characteristic benchmarks, which can adapt to different equipment states and working conditions, improving the applicability;the concept of abnormal group is introduced, fully considering the influence of the coupling relationship of the parameters on the comprehensive anomaly score, so that the comprehensive anomaly score can more accurately describe the abnormality degree;the data processing and operation and maintenance are mostly performed at the edge, and only when the early warning level is too high, the cloud is jointly processed, which can reduce the workload of the cloud.

Description

Technical Field

[0001] This invention relates to the field of fault early warning technology for hydropower plant equipment, and specifically to a fault early warning method for dynamic monitoring of unmanned hydropower units based on edge computing. Background Technology

[0002] There are currently four core technical problems in the monitoring and fault early warning of hydropower plant equipment: Early warning relies on fixed thresholds, which cannot adapt to changes in operating conditions. Traditional fixed threshold alarms cannot adapt to operating conditions such as unit start-up and shutdown, water level fluctuations, seasonal changes, and equipment aging, resulting in a high false alarm rate and an inability to predict in advance.

[0003] Centralized processing suffers from high latency and strong network dependence. Data needs to be uploaded to the station control layer or cloud computing. It becomes invalid when the network is interrupted, which cannot meet the millisecond-level safety early warning requirements of hydropower equipment.

[0004] Lacking the ability to learn historical characteristics at the device level, and without establishing a device-specific operating characteristic library, it only makes simple limit-crossing judgments and cannot identify early degradation trends.

[0005] The early warning system is limited to a single dimension and lacks a joint judgment mechanism. It relies solely on single measurement points and single parameters for judgment, making it susceptible to interference and resulting in insufficient accuracy and reliability of the early warning. Summary of the Invention

[0006] The main objective of this invention is to provide a dynamic monitoring and fault early warning method for unmanned hydropower units based on edge computing, thereby solving the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a dynamic monitoring and fault early warning method for unmanned hydropower units based on edge computing, comprising the following steps: S1. Collect multi-source data from hydropower equipment and identify its operating status; S2, Calculate the correction factor for hydropower operating conditions; S3. Determine the early warning thresholds for hydropower parameters; S4. Perform parameter anomaly detection and calculate the comprehensive anomaly score; S5. Based on the comprehensive anomaly score correction value, conduct graded early warning and carry out edge-side linkage handling.

[0008] Furthermore, the multi-source data includes: bearing temperature, winding temperature, cooling water temperature, system oil pressure, return oil pressure, operating oil pressure, pump running time, pump start frequency, vibration amplitude of key components, flow rate, voltage, current, insulation resistance, unit load, guide vane opening, flow rate, efficiency, speed, main shaft runout, volute pressure value, and unit guide vane operating rate. The importance level of the parameters is divided into three levels: high, medium, and low. The corresponding sampling frequency is set according to the importance level of the parameters.

[0009] Furthermore, the process of identifying the operating status is as follows: S101. Classify the types of operating states; the classified operating states include: shutdown and static state, startup process state, no-load idling state, grid-connected and loaded state, steady-state operation state, load reduction process state, and disconnection and shutdown state. S102. Determine the operating status and determine the operating status correction value based on the operating status; S103. Calculate the stability coefficient, and confirm the operating status a second time based on the stability coefficient to obtain the final operating status of the hydropower equipment. The expression for the stability coefficient is as follows: (1); in, The stability coefficient, To find the function with the maximum value, , These are the unit load change rate and the unit rated load, respectively. , These are the rate of change of the upstream and downstream water level difference and the rated upstream and downstream water level difference designed for the unit, respectively. , These are the rate of change of unit speed and the rated speed of the unit, respectively. The process of secondary confirmation of the running status is as follows: Set a drastic change threshold; when the stability coefficient is less than the drastic change threshold, the final operating state is confirmed as a drastic change state, and the warning shielding mode is automatically entered; in the automatic warning shielding mode, only data and trends are recorded, and no subsequent operations are performed; when the stability coefficient is not less than the drastic change threshold, the operating state determined in step S102 is taken as the final operating state, and subsequent operations are performed.

[0010] Furthermore, the expression for the hydropower operating condition correction factor is as follows: (2); in, For hydropower operating condition correction factors, , , These are the upstream and downstream water level difference factor, the load factor, and the flow rate factor, respectively. , , These are the weights of the upstream and downstream water level difference factor, the load factor, and the flow factor, respectively. The expression for the upstream and downstream water level difference factor is as follows: (3); in, This represents the actual value of the water level difference between the upstream and downstream areas. The expression for the load factor is as follows: (4); in, Real-time load of the unit; The expression for the flow factor is as follows: (5); in, For real-time water flow rate, This represents the flow rate corresponding to the unit's optimal efficiency point under the current upstream and downstream water level difference.

[0011] Furthermore, in step S3, the warning threshold includes: an upper warning threshold and a lower warning threshold; the expression for the upper warning threshold is as follows: (6); in, This represents the upper limit of the warning threshold for a certain parameter. This is the average of all normal data for this parameter over 30 consecutive days. This is the 99th score of all normal data for this parameter over 30 consecutive days. This is a correction value for the running status. This is the amount of compensation for aging drift. This is the sensitivity coefficient; The expression for the lower limit of the warning threshold is as follows: (7); in, This is the lower limit of the warning threshold. The score is 1 point for all normal data for this parameter over 30 consecutive days. The sensitivity coefficient is related to the importance level of the parameter.

[0012] Furthermore, the detailed process of step S4 is as follows: S401. Construct a parameter coupling association library that includes partial parameter coupling relationships; S402. Determine the warning threshold and anomaly judgment rules, perform anomaly judgment, and obtain anomaly points; if there is a coupling relationship between two anomaly points, then one of the anomaly points is called the anomaly coupling point of the other anomaly point; S403. Calculate the comprehensive anomaly score based on the outliers. The detailed process is as follows: Identify outliers without any abnormal coupling points, record these outliers as single outliers, and divide the remaining outliers into several outlier groups according to the outlier group classification criteria. The criteria for classifying anomaly groups are: any anomaly in an anomaly group must be coupled with at least one other anomaly in the same anomaly group, and any anomaly in an anomaly group must not be coupled with any anomaly in any other anomaly group. For any anomaly group, the anomaly point with the most anomaly coupling points is recorded as the main anomaly point of the group; if an anomaly group has more than one anomaly point with the most anomaly coupling points, then the multiple anomaly points with the most anomaly coupling points are selected as candidate main anomaly points. The deviation score for each outlier is calculated using the following expression: (8); in, The deviation score for a given outlier. This is the real-time value of the anomaly point. This refers to the warning threshold exceeded by this anomaly. Real-time deviation of outliers; The overall anomaly score is calculated based on the deviation value of the outliers, as shown in the following expression: (9); in, To calculate the overall abnormal score, For abnormal base scores, Extra points will be awarded for exceptions. For the first Deviation score for a single outlier This represents the total number of single outliers. For the first The deviation score of the main outlier of each outlier group. This represents the total number of outliers. For the first The first abnormal group The deviation score of each non-major outlier. for The total number of non-major anomalies in each anomaly cluster; S404. Predict the changing trend of anomalies based on the anomaly trend prediction model deployed on the edge side. The detailed process is as follows: Extract the basic feature parameters of the anomaly clusters; the basic feature parameters include: number of anomaly clusters, number of anomaly points in each anomaly cluster, real-time deviation of anomaly points, average deviation of anomaly clusters, duration of anomaly clusters, historical time-series monitoring data of each anomaly point in the anomaly cluster, and coupling and correlation topology matrix. The expression for the average deviation of outliers is: (10); in, The average deviation of a certain outlier group For this abnormal group of species Real-time deviation of each outlier point This represents the total number of outliers in the group. The trend of change is reflected by the rate of deviation of the outlier group from the total rate, which is expressed as follows: (11); in, For outliers deviating from the total rate, , The first Deviation rate and deviation weight of each outlier group; S405. Adjust the overall anomaly score based on the changing trend; The corrected expression for the overall anomaly score is as follows: (12); in, This is a correction value for the overall abnormal score. This represents the influence coefficient of the abnormal rate.

[0013] Furthermore, the specific rules for anomaly detection are as follows: For any parameter, obtain its current value. When the current value exceeds the warning threshold of the parameter, it is judged as abnormal. A parameter judged as abnormal is recorded as an abnormal point.

[0014] Furthermore, the deviation rate expression for the outlier is as follows: (13); in, , The first The average deviation of each anomaly group between the current period and the previous period. The duration of the cycle; The outlier deviation weight expression is as follows: (14); in, For the first The total number of outliers in each outlier group. For the first The total number of outliers in each outlier group.

[0015] Furthermore, the anomaly rate influence coefficient is determined by the deviation rate of the anomaly group from the total rate, and must follow the following rules: when the deviation rate of the anomaly group from the total rate is large, the comprehensive anomaly score is also large; when the deviation rate of the anomaly group from the total rate is small, the comprehensive anomaly score is also small.

[0016] Furthermore, the detailed process of step S5 is as follows: S501. Classify the warning level according to the warning level threshold; The warning levels are divided into four levels, from low to high, based on their severity: Level 1, Level 2, Level 3, and Level 4. The warning level thresholds are divided into Level 1 warning threshold, Level 2 warning threshold, and Level 3 warning threshold. When the overall anomaly score is less than the first-level warning threshold, the warning level is level 1; When the comprehensive anomaly score is greater than or equal to the Level 1 warning threshold but less than the Level 2 warning threshold, the warning level is Level 2. When the comprehensive anomaly score is greater than or equal to the Level 2 warning threshold but less than the Level 3 warning threshold, the warning level is Level 3. When the overall anomaly score exceeds the level 3 warning threshold, the warning level is 4. S502. Implement edge-side coordinated response based on the warning level; When the warning level is Level 1, the response plan is: no action will be taken for the time being, and monitoring will continue; When the warning level is 2, the handling plan is as follows: local pop-up notification, in-site message push, and logs are stored on the edge side without uploading to the cloud; When the warning level is level 3, the handling plan is as follows: the edge side issues control commands to automatically switch to the backup auxiliary machine, adjust the system operating conditions, and at the same time upload a short message warning to the cloud; When the warning level is 4, the response plan is as follows: the edge side shall immediately implement on-site emergency protection measures; At the same time, it proactively initiates edge and cloud collaboration, uploads complete fault messages, abnormal equipment information, and equipment operation snapshots, and coordinates with the plant-wide monitoring platform and operation and maintenance management system to take operation and maintenance measures.

[0017] Beneficial effects: (1) Determining the sampling frequency based on the importance of parameters can reduce resource consumption; (2) Introducing a stability coefficient to make a secondary judgment on the operating status can avoid issuing early warnings under the condition of drastic fluctuations in equipment, thus avoiding waste of resources; (3) The early warning threshold is a dynamic threshold that can realize dynamic monitoring of the unit, adapt to different equipment statuses and operating conditions, and improve applicability; (4) The importance of parameters is also introduced into the early warning threshold, which can reduce the false alarm rate of important parameters and the false alarm rate of less important parameters; (5) Introduce the concept of anomaly clusters, fully consider the impact of parameter coupling relationship on the comprehensive anomaly score, and modify the comprehensive anomaly score in combination with the development trend of anomaly clusters so that the comprehensive anomaly score can more accurately describe the degree of anomaly. (6) Most of the data processing and maintenance are done at the edge. Only when the warning level is too high will it be processed in conjunction with the cloud, which can reduce the workload of the cloud. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0019] Example 1 like Figure 1 As shown, the edge computing-based dynamic monitoring and fault early warning method for unmanned hydropower units includes the following steps: S1. Collect multi-source data from hydropower equipment and identify its operating status; Multi-source data includes: bearing temperature, winding temperature, cooling water temperature, system oil pressure, return oil pressure, operating oil pressure, pump running time, pump start frequency, vibration amplitude of key components, flow rate, voltage, current, insulation resistance, unit load, guide vane opening, flow rate, efficiency, speed, main shaft runout, volute pressure value, and unit guide vane operating rate. Since different data have different importance, if a uniform sampling frequency is used, too high a frequency will lead to waste of resources, while too low a frequency will lead to insufficient sampling of some important data. Therefore, different sampling frequencies are set for different parameters. This embodiment provides a simple solution: introduce parameter importance levels, which are divided into three levels: high, medium, and low. Set the sampling frequency of high-level parameters to 10ms / time, medium-level parameters to 100ms / time, and low-level parameters to 200ms / time. The detailed process for identifying the running status is as follows: S101. Classify the operating states; the classified operating states include: shutdown and static state, startup process state, no-load idling state, grid-connected and loaded state, steady-state operation state, load reduction process state, and disconnection and shutdown state; the classification criteria are as follows: The shutdown and static state must simultaneously meet the following conditions: the speed is 0, the circuit breaker is open, and there is no start command; The startup process must simultaneously meet the following conditions: the speed is within the range of 0~95% of the rated speed, there is a startup command, and the circuit breaker is tripped; The no-load running state must simultaneously meet the following conditions: the speed is within the range of 95% to 105% of the rated speed, the circuit breaker is open, and the load is 0; For grid-connected load conditions, the following conditions must be met simultaneously: the rotational speed is the rated speed, the circuit breaker is closed, and the load is within the range of 0~100% of the rated load. Steady-state operation must simultaneously meet the following conditions: the rotational speed is the rated speed, the circuit breaker is closed, the load fluctuation does not exceed ±2% of the rated load, and the duration is not less than 30 seconds. The load reduction process must simultaneously meet the following conditions: the load reduction rate is not less than 5% of the rated load / min and the circuit breaker is closed; The tripping and shutdown state must simultaneously meet the following conditions: the circuit breaker is open and the speed drops from 100% to 0; S102. Determine the operating status and determine the operating status correction value based on the operating status; the relationship between the operating status and the operating status correction value is as follows: The operating status correction value corresponding to the stopped and stationary state is 0.1; The running status correction value corresponding to the startup process status is 0.2; The operating status correction value corresponding to the no-load and idling state is 0.6; The operating status correction value corresponding to the grid-connected and load-bearing state is 0.7; The operating state correction value corresponding to the steady-state operating state is 1.0; The operating status correction value corresponding to the load reduction process is 0.4; The operating status correction value corresponding to the disconnection and shutdown state is 0.1; S103. Calculate the stability coefficient, and confirm the operating status a second time based on the stability coefficient to obtain the final operating status of the hydropower equipment. The expression for the stability coefficient is as follows: (1); in, The stability coefficient, To find the function with the maximum value, , These are the unit load change rate and the unit rated load, respectively. , These are the rate of change of the upstream and downstream water level difference and the rated upstream and downstream water level difference designed for the unit, respectively. , These are the rate of change of unit speed and the rated speed of the unit, respectively. The process of secondary confirmation of the running status is as follows: A threshold for drastic change is set; in this embodiment, the value is 0.3. When the stability coefficient is less than the threshold for drastic change, the final operating state is confirmed as a state of drastic change in operating conditions, and the warning shielding mode is automatically entered. In the warning shielding mode, only data and trends are recorded, and no subsequent operations are performed. When the stability coefficient is not less than the threshold for drastic change, the operating state determined in step S102 is taken as the final operating state, and subsequent operations are performed.

[0020] S2, Calculate the correction factor for hydropower operating conditions; The hydropower operating condition correction factor is related to the upstream and downstream water level difference, load, and flow rate. It can be calculated by weighting the upstream and downstream water level difference factor, load factor, and flow rate factor. The expression for the hydropower operating condition correction factor is as follows: (2); in, For hydropower operating condition correction factors, , , These are the upstream and downstream water level difference factor, the load factor, and the flow rate factor, respectively. , , The weights are the upstream and downstream water level difference factor, the load factor, and the flow factor, respectively, and the sum of the weights of the three is 1; The expression for the upstream and downstream water level difference factor is as follows: (3); in, This represents the actual value of the water level difference between the upstream and downstream areas. The expression for the load factor is as follows: (4); in, This represents the real-time load of the generating unit. Since the influence of hydropower load on other parameters is generally nonlinear, the load factor is set as a piecewise function here. The expression for the flow factor is as follows: (5); in, For real-time water flow rate, This is the flow rate corresponding to the unit's optimal efficiency point under the current upstream and downstream water level difference. This value can be obtained by fitting historical operating condition curves.

[0021] S3. Determine the early warning thresholds for hydropower parameters; The warning threshold includes an upper warning threshold and a lower warning threshold. The warning threshold is a dynamic threshold, and its value is dynamically adjusted to achieve dynamic monitoring. The expression for the upper warning threshold is as follows: (6); in, This represents the upper limit of the warning threshold for a certain parameter. This is the average of all normal data for this parameter over 30 consecutive days. This is the 99th percentile score of all normal data for this parameter over 30 consecutive days. A 99th percentile score means that this value is greater than 99% of the data values ​​for this parameter among all normal data values ​​over 30 consecutive days. This is a correction value for the running status. This is the amount of compensation for aging drift. This is the sensitivity coefficient; The expression for the lower limit of the warning threshold is as follows: (7); in, This is the lower limit of the warning threshold. The score is 1 point for all normal data of this parameter over 30 consecutive days. A score of 1 point means that among all normal data values ​​of this parameter over 30 consecutive days, this value is less than 1% of the data values. The sensitivity coefficient essentially reflects the degree of sensitivity of the corresponding parameter in early warning. Taking the upper limit of the early warning threshold as an example, with other parameters fixed, since this value corresponds to... and The difference is negative, so the larger the sensitivity coefficient, the lower the upper limit of the warning threshold, and the easier it is for this parameter to trigger a warning. Therefore, in the preferred embodiment, the sensitivity coefficient is determined by combining the importance level of the parameter in step S1. This ensures that important parameters are more likely to trigger warnings, while also reducing the false alarm probability of less important parameters. Specifically, when the parameter importance level is high, the sensitivity coefficient is 1.2; when the parameter importance level is medium, the sensitivity coefficient is 1.0; and when the parameter importance level is medium, the sensitivity coefficient is 0.8. Because equipment ages during use, the parameter operating range will shift. Therefore, an aging drift compensation is introduced to compensate for this shift. The aging drift compensation is generally obtained by subtracting the average value of the normal data of the parameter in the first month from the average value of the normal data of the parameter in the current month.

[0022] S4. Perform parameter anomaly detection and calculate the comprehensive anomaly score. The detailed process is as follows: S401. Construct a parameter coupling association library that includes some parameter coupling relationships; since there are coupling relationships between some parameters, the coupling parameters need to be considered when judging anomalies, so as to reduce the false alarm rate of early warnings. The coupling rules in the coupling association library are as follows: Bearing temperature is coupled with cooling water temperature; Winding temperature is coupled with unit load; The system oil pressure and return oil pressure are coupled; Operating hydraulic pressure is coupled with oil level; System oil pressure is coupled with unit load; The vibration amplitude of key components is coupled with their rotational speed; Spindle runout is coupled with unit load; Guide vane opening is coupled with unit load; Unit load is coupled with current; The water level in the sump is coupled with the pump's operating time. The water level in the sump is coupled with the pump start-up frequency; Upstream and downstream water level difference and flow rate; Cooling water temperature and flow rate; S402. Determine the anomaly detection rules and perform anomaly detection; The specific rules for anomaly detection are as follows: For any parameter, obtain its current value. When the current value exceeds the warning threshold of the parameter, it is judged as abnormal. A parameter judged as abnormal is recorded as an abnormal point. If there is a coupling relationship between two abnormal points, one of the abnormal points is called the abnormal coupling point of the other abnormal point. S403. Calculate the comprehensive anomaly score based on the outliers. The detailed process is as follows: Identify outliers without any abnormal coupling points, record these outliers as single outliers, and divide the remaining outliers into several outlier groups according to the outlier group classification criteria. The criteria for classifying anomaly groups are: any anomaly in an anomaly group must be coupled with at least one other anomaly in the same anomaly group, and any anomaly in an anomaly group must not be coupled with any anomaly in any other anomaly group. For any anomaly group, the anomaly point with the most anomalous coupling points is recorded as the principal anomaly point of the group; if an anomaly group has more than one anomaly point with the most anomalous coupling points, then the multiple anomaly points with the most anomalous coupling points are selected as candidate principal anomaly points; for example, in an anomaly group, A and B each have 2 anomalous coupling points, then A and B are selected as candidate principal anomaly points. The deviation score for each outlier is calculated using the following expression: (8); in, The deviation score for a given outlier. This is the real-time value of the anomaly point. This refers to the warning threshold exceeded by the anomaly point. For example, when anomaly point A exceeds the upper limit of the anomaly threshold, the upper limit of the anomaly threshold is [a certain percentage] of the warning threshold for anomaly point A. When point A is less than the lower limit of the anomaly threshold, the lower limit of the anomaly threshold is the value of point A. , Real-time deviation of outliers; The overall anomaly score is calculated based on the deviation value of the outliers, as shown in the following expression: (9); in, To calculate the overall abnormal score, For abnormal base scores, Extra points will be awarded for exceptions. For the first Deviation score for a single outlier This represents the total number of single outliers. For the first The deviation score of the main outlier of each outlier group. This represents the total number of outliers. For the first The first abnormal group The deviation score of each non-major outlier. for The total number of non-major anomalies in each anomaly cluster; S404. Predict the changing trend of anomalies based on the anomaly trend prediction model deployed on the edge side. The detailed process is as follows: Extract the basic feature parameters of the anomaly clusters; the basic feature parameters include: number of anomaly clusters, number of anomaly points in each anomaly cluster, real-time deviation of anomaly points, average deviation of anomaly clusters, duration of anomaly clusters, historical time-series monitoring data of each anomaly point in the anomaly cluster, and coupling and correlation topology matrix. The expression for the average deviation of outliers is: (10); in, The average deviation of a certain outlier group For this abnormal group of species Real-time deviation of each outlier point This represents the total number of outliers in the group. The trend of change is reflected by the rate of deviation of the outlier group from the total rate, which is expressed as follows: (11); in, For outliers deviating from the total rate, , The first Deviation rate and deviation weight of each outlier group; For any outlier group, its deviation rate is related to the average deviation of that group; the expression for the deviation rate of an outlier group is as follows: (12); in, , The first The average deviation of each anomaly group between the current period and the previous period. The duration of the cycle; Since the risks posed by different anomaly groups are not the same, anomaly group weights are set based on the number of anomalies within each group. This ensures that high-anomaly groups have a greater impact on risk trends and improve prediction accuracy. The anomaly group deviation weight expression is as follows: (13); in, For the first The total number of outliers in each outlier group. For the first The total number of outliers in each outlier group; S405. Adjust the overall anomaly score based on the changing trend; The corrected expression for the overall anomaly score is as follows: (14); in, This is a correction value for the overall abnormal score. The abnormal rate influence coefficient; The anomaly rate influence coefficient is determined by the deviation rate of the anomaly group from the total rate, and the following rules must be followed: when the deviation rate of the anomaly group from the total rate is large, the comprehensive anomaly score is also large; when the deviation rate of the anomaly group from the total rate is small, the comprehensive anomaly score is also small. This embodiment provides an adaptive anomaly rate influence coefficient, expressed as follows: (15); in, This is an adjustment coefficient, which is used to adjust the weight of the impact of the rate of deviation of the anomaly group from the total rate of deviation on the comprehensive anomaly score. It is determined manually by the staff.

[0023] S5. Based on the comprehensive anomaly score correction value, a graded early warning is issued and edge-side coordinated handling is carried out. The detailed process is as follows: S501. Classify the warning level according to the warning level threshold; The warning levels are divided into four levels, from low to high, based on their severity: Level 1, Level 2, Level 3, and Level 4. The warning level thresholds are divided into Level 1 warning threshold, Level 2 warning threshold, and Level 3 warning threshold. When the overall anomaly score is less than the first-level warning threshold, the warning level is level 1; When the comprehensive anomaly score is greater than or equal to the Level 1 warning threshold but less than the Level 2 warning threshold, the warning level is Level 2. When the comprehensive anomaly score is greater than or equal to the Level 2 warning threshold but less than the Level 3 warning threshold, the warning level is Level 3. When the overall anomaly score exceeds the level 3 warning threshold, the warning level is 4. S502. Implement edge-side coordinated response based on the warning level; When the warning level is Level 1, the response plan is: no action will be taken for the time being, and monitoring will continue; When the warning level is 2, the handling plan is as follows: local pop-up notification, in-site message push, and logs are stored on the edge side without uploading to the cloud; When the warning level is level 3, the handling plan is as follows: the edge side issues control commands to automatically switch to the backup auxiliary machine, adjust the system operating conditions, and at the same time upload a short message warning to the cloud; When the warning level is 4, the response plan is as follows: the edge side shall immediately implement local emergency protection actions, including audible and visual hard alarms, safety circuit interlocking, and emergency load limiting or shutdown of the unit. At the same time, it proactively initiates edge and cloud collaboration, uploads complete fault messages, abnormal equipment information, and equipment operation snapshots, and coordinates with the plant-wide monitoring platform and operation and maintenance management system to take operation and maintenance measures.

[0024] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing, characterized in that, Includes the following steps: S1. Collect multi-source data from hydropower equipment and identify its operating status; S2, Calculate the correction factor for hydropower operating conditions; S3. Determine the early warning thresholds for hydropower parameters; S4. Perform parameter anomaly detection and calculate the comprehensive anomaly score; S5. Based on the comprehensive anomaly score correction value, conduct graded early warning and carry out edge-side linkage handling.

2. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 1, characterized in that, Multi-source data includes: bearing temperature, winding temperature, cooling water temperature, system oil pressure, return oil pressure, operating oil pressure, pump running time, pump start frequency, vibration amplitude of key components, flow rate, voltage, current, insulation resistance, unit load, guide vane opening, flow rate, efficiency, speed, main shaft runout, volute pressure value, and unit guide vane operating rate. The importance level of the parameters is divided into three levels: high, medium, and low. The corresponding sampling frequency is set according to the importance level of the parameters.

3. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 2, characterized in that, The process of identifying the running status is as follows: S101. Classify the types of operating states; the classified operating states include: shutdown and static state, startup process state, no-load idling state, grid-connected and loaded state, steady-state operation state, load reduction process state, and disconnection and shutdown state. S102. Determine the operating status and determine the operating status correction value based on the operating status; S103. Calculate the stability coefficient, and confirm the operating status a second time based on the stability coefficient to obtain the final operating status of the hydropower equipment. The expression for the stability coefficient is as follows: （1）； in, The stability coefficient, To find the function with the maximum value, , These are the unit load change rate and the unit rated load, respectively. , These are the rate of change of the upstream and downstream water level difference and the rated upstream and downstream water level difference designed for the unit, respectively. , These are the rate of change of unit speed and the rated speed of the unit, respectively. The process of secondary confirmation of the running status is as follows: Set a drastic change threshold; when the stability coefficient is less than the drastic change threshold, the final operating state is confirmed as a drastic change state, and the warning shielding mode is automatically entered; in the automatic warning shielding mode, only data and trends are recorded, and no subsequent operations are performed; when the stability coefficient is not less than the drastic change threshold, the operating state determined in step S102 is taken as the final operating state, and subsequent operations are performed.

4. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 1, characterized in that, The expression for the hydropower operating condition correction factor is as follows: （2）； in, For hydropower operating condition correction factors, , , These are the upstream and downstream water level difference factor, the load factor, and the flow rate factor, respectively. , , These are the weights of the upstream and downstream water level difference factor, the load factor, and the flow factor, respectively. The expression for the upstream and downstream water level difference factor is as follows: （3）； in, This represents the actual value of the water level difference between the upstream and downstream areas. The expression for the load factor is as follows: （4）； in, Real-time load of the unit; The expression for the flow factor is as follows: （5）； in, For real-time water flow rate, This represents the flow rate corresponding to the unit's optimal efficiency point under the current upstream and downstream water level difference.

5. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 3 or 4, characterized in that, In step S3, the warning threshold includes: an upper warning threshold and a lower warning threshold; the expression for the upper warning threshold is as follows: （6）； in, This represents the upper limit of the warning threshold for a certain parameter. This is the average of all normal data for this parameter over 30 consecutive days. This is the 99th score of all normal data for this parameter over 30 consecutive days. This is a correction value for the running status. This is the amount of compensation for aging drift. This is the sensitivity coefficient; The expression for the lower limit of the warning threshold is as follows: （7）； in, This is the lower limit of the warning threshold. The score is 1 point for all normal data for this parameter over 30 consecutive days. The sensitivity coefficient is related to the importance level of the parameter.

6. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 5, characterized in that, The detailed process of step S4 is as follows: S401. Construct a parameter coupling association library that includes partial parameter coupling relationships; S402. Determine the warning threshold and anomaly judgment rules, perform anomaly judgment, and obtain anomaly points; if there is a coupling relationship between two anomaly points, then one of the anomaly points is called the anomaly coupling point of the other anomaly point; S403. Calculate the comprehensive anomaly score based on the outliers. The detailed process is as follows: Identify outliers without any abnormal coupling points, record these outliers as single outliers, and divide the remaining outliers into several outlier groups according to the outlier group classification criteria. The criteria for classifying anomaly groups are: any anomaly in an anomaly group must be coupled with at least one other anomaly in the same anomaly group, and any anomaly in an anomaly group must not be coupled with any anomaly in any other anomaly group. For any anomaly group, the anomaly point with the most anomaly coupling points is recorded as the main anomaly point of the group; if an anomaly group has more than one anomaly point with the most anomaly coupling points, then the multiple anomaly points with the most anomaly coupling points are selected as candidate main anomaly points. The deviation score for each outlier is calculated using the following expression: （8）； in, The deviation score for a given outlier. This is the real-time value of the anomaly point. This refers to the warning threshold exceeded by this anomaly. Real-time deviation of outliers; The overall anomaly score is calculated based on the deviation value of the outliers, as shown in the following expression: （9）； in, To calculate the overall abnormal score, For abnormal base scores, Extra points will be awarded for exceptions. For the first Deviation score for a single outlier This represents the total number of single outliers. For the first The deviation score of the main outlier of each outlier group. This represents the total number of outliers. For the first The first abnormal group The deviation score of each non-major outlier. for The total number of non-major anomalies in each anomaly cluster; S404. Predict the changing trend of anomalies based on the anomaly trend prediction model deployed on the edge side. The detailed process is as follows: Extract the basic feature parameters of the anomaly clusters; the basic feature parameters include: number of anomaly clusters, number of anomaly points in each anomaly cluster, real-time deviation of anomaly points, average deviation of anomaly clusters, duration of anomaly clusters, historical time-series monitoring data of each anomaly point in the anomaly cluster, and coupling and correlation topology matrix. The expression for the average deviation of outliers is: （10）； in, The average deviation of a certain outlier group For this abnormal group of species Real-time deviation of each outlier point This represents the total number of outliers in the group. The trend of change is reflected by the rate of deviation of the outlier group from the total rate, which is expressed as follows: （11）； in, For outliers deviating from the total rate, , The first Deviation rate and deviation weight of each outlier group; S405. Adjust the overall anomaly score based on the changing trend; The corrected expression for the overall anomaly score is as follows: （12）； in, This is a correction value for the overall abnormal score. This represents the influence coefficient of the abnormal rate.

7. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 6, characterized in that, The specific rules for anomaly detection are as follows: For any parameter, obtain its current value. When the current value exceeds the warning threshold of the parameter, it is judged as abnormal. A parameter judged as abnormal is recorded as an abnormal point.

8. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 6, characterized in that, The deviation rate expression for the outlier is as follows: （13）； in, , The first The average deviation of each anomaly group between the current period and the previous period. The duration of the cycle; The outlier deviation weight expression is as follows: （14）； in, For the first The total number of outliers in each outlier group. For the first The total number of outliers in each outlier group.

9. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 6, characterized in that, The anomaly rate influence coefficient is determined by the deviation rate of the anomaly group from the total rate, and must follow the following rules: when the deviation rate of the anomaly group from the total rate is large, the comprehensive anomaly score is also large; when the deviation rate of the anomaly group from the total rate is small, the comprehensive anomaly score is also small.

10. The method for dynamic monitoring and fault early warning of unmanned hydropower units based on edge computing according to claim 6, characterized in that, The detailed process of step S5 is as follows: S501. Classify the warning level according to the warning level threshold; The warning levels are divided into four levels, from low to high, based on their severity: Level 1, Level 2, Level 3, and Level 4. The warning level thresholds are divided into Level 1 warning threshold, Level 2 warning threshold, and Level 3 warning threshold. When the overall anomaly score is less than the first-level warning threshold, the warning level is level 1; When the comprehensive anomaly score is greater than or equal to the Level 1 warning threshold but less than the Level 2 warning threshold, the warning level is Level 2. When the comprehensive anomaly score is greater than or equal to the Level 2 warning threshold but less than the Level 3 warning threshold, the warning level is Level 3. When the overall anomaly score exceeds the level 3 warning threshold, the warning level is 4. S502. Implement edge-side coordinated response based on the warning level; When the warning level is Level 1, the response plan is: no action will be taken for the time being, and monitoring will continue; When the warning level is 2, the handling plan is as follows: local pop-up notification, in-site message push, and logs are stored on the edge side without uploading to the cloud; When the warning level is level 3, the handling plan is as follows: the edge side issues control commands to automatically switch to the backup auxiliary machine, adjust the system operating conditions, and at the same time upload a short message warning to the cloud; When the warning level is 4, the response plan is as follows: the edge side shall immediately implement on-site emergency protection measures; At the same time, it proactively initiates edge and cloud collaboration, uploads complete fault messages, abnormal equipment information, and equipment operation snapshots, and coordinates with the plant-wide monitoring platform and operation and maintenance management system to take operation and maintenance measures.

Images