Power plant equipment abnormal rapid identification method based on cloud edge collaboration

Abstract

The application discloses a power station equipment abnormal rapid identification method based on cloud edge cooperation and relates to the technical field of power station equipment data analysis. The method first pre-processes and initially screens edge time series data reflecting the real-time running state of power station equipment on the edge side to improve the timeliness of data uploading to the cloud, then performs trend comparison analysis based on the initial screening result to determine the running state fluctuation law, simultaneously performs multi-dimensional correlation analysis and data cooperation management, generates a visual fault risk assessment report, and finally identifies the equipment fault level according to the report and the real-time running data on the edge side and synchronously dynamically optimizes the identification capability through cloud edge cooperation to form a complete abnormal identification process. The method realizes the improvement of data transmission timeliness, the clearness of equipment running law and the dynamic optimization of identification capability under cloud edge cooperation, and effectively solves the problem of composite fault omission caused by non-fault fluctuation misjudgment.

Description

Technical Field

[0001] This invention relates to the field of power plant equipment data analysis technology, and in particular to a method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration. Background Technology

[0002] With the acceleration of the global energy transition, traditional centralized power plants are deeply integrated with distributed photovoltaic, wind power, and energy storage power plants. The scale of power plant operation is expanding, the operating conditions of equipment are becoming more complex, and the popularization of industrial internet technology is driving power plants into the era of digital monitoring. The amount of high-frequency data such as vibration, temperature, current, and acoustic data generated by equipment in real time is increasing exponentially, which places higher demands on the real-time and accuracy of equipment anomaly identification. The power plant equipment here includes core power generation equipment (such as steam turbine generators, photovoltaic inverters, wind turbine generators, etc.), auxiliary system equipment (such as cooling pumps, oil pumps, transformers), and monitoring and sensing equipment (such as vibration sensors, temperature sensors, pressure sensors). Their stable operation directly determines the power generation efficiency and safety of the power plant.

[0003] Existing technologies mainly achieve anomaly identification in two ways: one is to rely on manual inspection combined with a local programmable logic controller (PLC) system, where maintenance personnel regularly check the equipment's operating status on-site, and use the PLC system to monitor key parameters such as voltage and temperature at thresholds, triggering an alarm once the data exceeds the limits; the other is to use centralized cloud analysis, which transmits data such as vibration and current generated by all equipment in the power station to the cloud in real time, and uses machine learning models such as support vector machines and neural networks to analyze the data, thereby identifying anomalies.

[0004] For example, Chinese invention patent CN112417363B discloses a substation load analysis method and system, which includes: randomly selecting a target substation within the analysis range, acquiring the main transformer voltage data and high-voltage side line data of the target substation at a certain moment; calculating the total capacity of the main transformer, the total active load of the main transformer at high voltage, the total active load of the main transformer at medium voltage, and the total active load of the main transformer at low voltage of the target substation; calculating the active load value of the target substation and filtering out the portion less than a preset first threshold; determining whether there is an abnormal loss in the active load value of the target substation, correcting the active load value with abnormal loss, or calculating the load rate of the target substation; obtaining the load rate of all target substations within the selected analysis range, and outputting the load analysis results for all time periods and all target substations.

[0005] However, centralized cloud analysis is prone to network bandwidth congestion due to high-frequency data transmission and large data volume, resulting in data transmission delays (which can reach minutes in remote power plants). This makes it difficult to capture instantaneous equipment anomalies (such as short-term current fluctuations) in a timely manner. Furthermore, PLC systems do not fully consider multi-parameter correlation logic and environmental interference factors, making it difficult to balance the real-time performance and accuracy of anomaly identification. This can lead to missed fault detection, such as misjudging environmentally induced temperature fluctuations as equipment failures. There is also the problem of untimely identification of power plant equipment anomalies, leading to misjudgment of non-fault-related instantaneous fluctuations, and consequently, missed detection of complex faults. Summary of the Invention

[0006] To address the technical problems in existing technologies, this invention provides a method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration. The technical solution is as follows: Step 1: At the edge side of power plant equipment operation status monitoring, preprocessing and initial anomaly screening of edge time-series data under cloud-edge collaboration are performed to ensure the integrity and reliability of the verification data segments and key features uploaded to the cloud. The edge time-series data reflects the real-time operating status and core parameter change trends of the power plant equipment. Step 2: Based on the anomaly screening results, trend comparison analysis is performed to determine the fluctuation pattern of the power plant equipment's operating status. Simultaneously, multi-dimensional correlation analysis and data collaborative management are conducted to generate a fault risk assessment report for visualizing the fault types and impact range of the power plant equipment. Step 3: Based on the equipment fault risk assessment report and real-time operating data from the edge side, equipment anomaly identification is performed to determine the fault level of the power plant equipment. Simultaneously, dynamic identification optimization based on cloud-edge collaboration is performed to improve the efficiency and reliability of power plant equipment anomaly identification. The real-time operating data from the edge side reflects the real-time operating conditions and parameter dynamic changes of the power plant equipment after verification.

[0007] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0008] 1. This invention preprocesses and initially screens anomalies in edge-time data reflecting the real-time operating status of power plant equipment at the edge. This step significantly reduces the amount of high-frequency data transmission, avoids network bandwidth congestion common in centralized analysis, and solves the minute-level latency problem in data transmission for remote power plants, laying the foundation for timely detection of instantaneous anomalies such as short-term current fluctuations. Based on the initial screening results, trend comparison analysis is conducted to clarify the fluctuation patterns of equipment operation. Simultaneously, a visualized fault risk assessment report is generated through multi-dimensional correlation analysis and data collaborative management. This step compensates for the shortcomings of PLC systems in not fully considering multi-parameter correlation logic and environmental interference factors, accurately distinguishing between environmentally induced temperature fluctuations and equipment faults, and reducing misjudgments of non-fault fluctuations. Combining the fault risk assessment report with real-time edge operating data, the fault level of equipment is identified, and the identification capability is dynamically optimized through cloud-edge collaboration. This step relies on precise analysis to ensure identification accuracy while improving efficiency through collaborative optimization, effectively balancing the real-time nature and accuracy of anomaly identification, avoiding missed detection of equipment faults and complex faults, and comprehensively improving the reliability of power plant equipment anomaly identification.

[0009] 2. First, acquire the time-series data of electrical and mechanical parameters of the power station equipment. For the electrical parameter time-series data, use wavelet filtering to remove electromagnetic interference, obtaining the current peak deviation rate and voltage fluctuation amplitude. For the mechanical parameter time-series data, use mean filtering to eliminate ambient temperature interference, obtaining the temperature change gradient and vibration frequency spectrum peaks. Filtering eliminates electromagnetic and ambient temperature interference, resolving the problem of misjudgment caused by the PLC system's failure to consider environmental interference, and improving data accuracy. Then, process the data into binary form, and generate an initial anomaly screening table (1 for abnormal, 0 for normal) based on a reference range. Only upload verification data segments with at least one parameter of 1; all values ​​of 0 are stored at the edge. This process significantly reduces the amount of uploaded data, avoids bandwidth congestion and minute-level delays, helps to promptly capture instantaneous anomalies such as short-term current fluctuations, and reduces misjudgments of non-fault fluctuations through accurate initial screening, laying the foundation for avoiding subsequent missed fault detections.

[0010] 3. A timestamp consistency score is obtained by quantifying the deviation between the timestamp of the verified data segment and the preset sampling interval, marking qualified / abnormal data. Only data with qualified timestamps and normal initial screening parameters are allowed to be uploaded; otherwise, supplementary collection and re-verification are performed. This step ensures the accuracy of uploaded data time, avoiding analysis errors caused by time deviations and laying the foundation for accurate troubleshooting. After successful verification, the anomaly root cause investigation compares the duration of suspected data anomalies with historical data under the same operating conditions to obtain anomaly matching degree, determining whether the anomaly is due to the equipment itself. If the root cause is not determined or the investigation limit is reached, it is marked for re-inspection, and the results are synchronously archived in the cloud. This process compares historical data with external interference, fully considering multi-parameter correlations and environmental factors, avoiding misjudging environmentally induced fluctuations as equipment failures, solving the problem of missed fault detection, and simultaneously reducing transmission volume by uploading only qualified data, alleviating bandwidth congestion and latency, and helping to capture instantaneous anomalies in a timely manner.

[0011] 4. First, the fluctuation amplitude of each parameter in the reviewed data segment is quantified by comparing its proportion with historical data from the same period under the same operating conditions to obtain the degree of consistency of the fluctuation trend, thereby determining the fluctuation pattern of the operating status. This step clarifies the state pattern by comparing historical data, providing a benchmark for accurate anomaly identification and reducing the possibility of misjudgment. The multi-dimensional correlation analysis process first statistically analyzes the change amplitude of the edge time-series data under abnormal fluctuations and preset environmental factors within the same time window, then standardizes and calculates the synchronization coefficient: when the change direction is the same, it is obtained by reverse conversion of the percentage of overlapping fluctuation amplitudes and the peak time difference; when the direction is opposite, it is obtained by reverse superposition of the remaining percentage and the valley time difference, and the average value is taken to obtain the change synchronization score. If the change synchronization score exceeds the preset value, it is marked as an environmental factor anomaly and the environmental cause is traced; otherwise, it is marked as an equipment anomaly and the historical fault cause is traced, and then an anomaly cause-state change relationship graph is constructed. This process accurately distinguishes between environmental interference and equipment problems, solves the problem of misjudging environmentally induced fluctuations, and assists in quickly locating the scope of fault impact. Simultaneously, the first abnormal timestamp of electrical and mechanical parameters is extracted from the graph, and the interval is calculated to determine whether it is a composite anomaly. This mechanism captures the characteristics of abnormal cross-dimensional diffusion, effectively identifies complex faults caused by the same root cause, and improves the targeting and efficiency of troubleshooting.

[0012] 5. Cross-validation is conducted based on historical fault level characteristics from the equipment fault risk assessment report and real-time edge operation data to obtain cross-validation scores. The grading standards cover the entire score range and are mutually exclusive, ensuring clear and accurate fault determination. Cross-validation scores are obtained as follows: fault feature data corresponding to the corresponding validation score range is extracted, and the abnormal percentage duration of each data point is calculated; the difference between the sum of the abnormal percentage duration and the historical average duration of the same level is calculated, and the difference between the number of fault feature data points and the historical number is also calculated; the absolute values ​​of the two differences are added together and then converted in reverse to quantify the matching degree between historical features and real-time data. This process, based on dual verification of historical fault features and real-time data, fully considers parameter correlation logic, avoids misjudging faults due to single data points, and solves the problem of missed or incorrect judgments caused by PLC systems not considering multiple parameter correlations; furthermore, relying on edge real-time data reduces dependence on high-frequency data transmission from the cloud, alleviates bandwidth congestion and latency, helps to identify equipment anomalies in a timely manner, and improves the real-time performance and accuracy of anomaly identification.

[0013] 6. By first calculating the feature drift of real-time edge data, if it exceeds the corresponding reference value, the edge side automatically updates local anomaly parameters and compresses the drift feature data before uploading; otherwise, it maintains the original parameters and only synchronizes data without anomalies. This process transmits data on demand, reducing invalid transmission and alleviating bandwidth congestion and latency issues. When updating local parameters, the deviation between the drift feature data and the historical fault feature database in the cloud is used to map the impact of local parameter updates and the impact of cloud data processing priority. The former is then used to compensate for the original parameters to generate optimized parameters and replace them, improving the edge-side recognition adaptability. In cloud processing, the priority impact is used to improve parsing speed: if the target is met, the upload frequency is shortened based on the parsing capacity margin to optimize interaction efficiency; if the target is not met after a timeout, an alert is issued, prompting the investigation of node load and data format compatibility issues to quickly restore parsing efficiency. The entire process achieves dynamic optimization through cloud-edge collaboration, balancing the real-time nature and accuracy of anomaly recognition, and avoiding missed detection of instantaneous anomalies due to parameter lag. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating a method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration, provided in an embodiment of the present invention.

[0016] Figure 2 This is a flowchart of edge time series data preprocessing and anomaly screening provided in an embodiment of the present invention;

[0017] Figure 3 This is one of the flowcharts for multi-dimensional correlation analysis and data collaborative management provided in the embodiments of the present invention;

[0018] Figure 4 The second flowchart for multi-dimensional correlation analysis and data collaborative management provided in this embodiment of the invention;

[0019] Figure 5 This is a schematic diagram of the relationship between abnormal causes and state changes provided in an embodiment of the present invention;

[0020] Figure 6 A flowchart illustrating the cloud-edge collaborative dynamic identification optimization provided in this embodiment of the invention;

[0021] Figure 7 This is the homepage of the power plant operation and maintenance management platform provided in this embodiment of the invention;

[0022] Figure 8 This is a schematic diagram of the fault management module in the power plant operation and maintenance management platform provided in an embodiment of the present invention;

[0023] Figure 9 This is a schematic diagram of the cloud-edge collaboration module in the power plant operation and maintenance management platform provided in this embodiment of the invention. Detailed Implementation

[0024] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0025] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0026] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0027] In the operation of power plant equipment, timely and accurate identification of anomalies is crucial to ensuring stable power supply. Traditional methods suffer from problems such as identification lag and insufficient accuracy, making them unsuitable for complex operating conditions. Therefore, this invention provides a method for rapid anomaly identification in power plant equipment based on cloud-edge collaboration, such as... Figure 1 The flowchart shown is for a method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration. The processing flow of this method may include:

[0028] Step 1: At the edge of the power plant equipment operation status monitoring, preprocess and initially screen the edge time series data under cloud-edge collaboration to ensure the integrity and reliability of the verification data segments and key features uploaded to the cloud. The edge time series data is used to reflect the real-time operation status and core parameter change trends of the power plant equipment. The edge time series data includes the peak current deviation rate, which reflects the load stability of the electrical circuit of the power plant equipment; the voltage fluctuation amplitude, which reflects the power grid power supply fluctuation; the temperature change gradient, which reflects the heat dissipation status of the mechanical components of the power plant equipment; and the vibration frequency spectrum peak, which reflects the operational stability of the mechanical components of the power plant equipment.

[0029] Step two involves trend comparison analysis based on the initial anomaly screening results to determine the fluctuation patterns of the power plant equipment's operating status. Simultaneously, multi-dimensional correlation analysis and data collaborative management are conducted to generate a fault risk assessment report that visualizes the types and impact range of power plant equipment faults. Power plant equipment fault types include composite anomalies and single-dimensional anomalies. Composite anomalies indicate a correlation between electrical and mechanical parameter anomalies, possibly caused by the same root cause (such as aging of the stator winding insulation of the power plant equipment). Single-dimensional anomalies indicate an anomaly in only one dimension (electrical or mechanical) of the electrical and mechanical parameters, while the other dimension is normal, and the two anomalies are unrelated. They are usually caused by specific problems specific to the corresponding dimension. For example, electrical parameter anomalies may originate from instantaneous fluctuations in the power grid or poor contact in local circuits, while mechanical parameter anomalies may originate from insufficient lubrication of a single component or local vibration interference.

[0030] Step 3: Based on the equipment failure risk assessment report and real-time operating data from the edge side, equipment anomaly identification is performed to determine the failure level of the power plant equipment. At the same time, dynamic identification optimization through cloud-edge collaboration is carried out to improve the efficiency and reliability of power plant equipment anomaly identification. The real-time operating data from the edge side is used to reflect the real-time operating conditions and dynamic changes of parameters of the power plant equipment after verification. The failure levels of the power plant equipment include Level 1, Level 2, and Level 3 failures, with the corresponding failure severity increasing from low to high, and the urgency of matching processing gradually increasing. The real-time operating data from the edge side represents the edge time-series data subset after edge-side preprocessing (such as consistency verification and mean filtering) and anomaly screening, after removing invalid interference.

[0031] Among them, the peak current deviation rate represents the percentage of the difference between the effective value of the three-phase current and the peak value of the rated current of the power station equipment, accounting for the peak value of the rated current; the voltage fluctuation amplitude represents the difference between the maximum and minimum instantaneous values ​​of the bus voltage within a consecutive preset sampling period (usually set to 10 sampling periods); the temperature change gradient represents the ratio of the difference in the operating temperature of the power station equipment to the sampling time interval within two adjacent sampling periods (sampling interval of 1 minute); and the vibration frequency spectrum peak represents the frequency value corresponding to the frequency point with the highest energy value in the frequency domain graph after performing a Fourier transform on the vibration acceleration signal of the power station equipment. The preset sampling period and the setting of two adjacent sampling periods are determined based on the operating characteristics of the power station equipment (such as response speed and stabilization period) and industry monitoring standards. In practical applications, fine-tuning can be performed according to the equipment model and operating conditions to balance data timeliness and identification accuracy, and adapt to the monitoring needs in different scenarios.

[0032] It is important to understand that the calculations of parameters such as peak current deviation rate and voltage fluctuation amplitude, as well as the setting of the sampling period, are all monitoring data corresponding to the overall operating status of the power plant equipment, rather than focusing on a single component. These parameters comprehensively reflect the overall operational stability of the equipment from electrical (current, voltage) and mechanical (temperature, vibration) dimensions. For example, current and voltage data reflect the operating conditions of the entire electrical system of the equipment, while operating temperature and vibration data relate to the overall coordinated state of the equipment's mechanical transmission system. Ultimately, through multi-dimensional data fusion, a judgment is made on whether the power plant equipment is operating normally as a whole, rather than monitoring a single component.

[0033] In a specific implementation, such as in the operation of thermal power plant units, preprocessing and initial screening of time-series data such as current peak deviation rate and vibration frequency spectrum peaks at the edge can quickly eliminate environmental interference data, avoid invalid information occupying bandwidth, and allow the cloud to focus on core data verification. Addressing the issue of photovoltaic power plants easily misjudging single-dimensional anomalies as complex anomalies due to light fluctuations, multi-dimensional correlation analysis can accurately distinguish the correlation between electrical parameters (such as voltage fluctuation amplitude) and mechanical parameters (such as temperature change gradient) anomalies, reducing ineffective troubleshooting by maintenance personnel. Simultaneously, cloud-edge collaborative dynamic optimization can update edge-side parameters in real time based on feature drift. When wind turbines in wind power plants are operating under high load, it can shorten the response time for fault level determination. Level 3 severe faults (such as main component wear) can quickly trigger early warnings, buying time for emergency repairs, reducing daily power generation losses due to equipment downtime, and ensuring the safe and stable operation of the power plant.

[0034] like Figure 2The flowchart shown illustrates the edge time-series data preprocessing and anomaly screening. Its design logic is as follows: First, initial edge time-series data of the power plant equipment is acquired, and key parameters are extracted from both electrical and non-electrical dimensions (i.e., mechanical dimensions). An anomaly screening table is generated through binary processing. By determining whether a parameter is 1, the data nature is quickly distinguished: if at least one parameter is 1, it is determined to be a data segment requiring review and uploaded to the cloud for traceability; otherwise, it is determined to be time-series data without anomalies and stored on the edge side. This achieves efficient preliminary data screening, reduces the uploading of invalid data, and improves the efficiency of anomaly detection.

[0035] Further understanding is needed regarding the preprocessing and anomaly screening of edge time-series data under cloud-edge collaboration. The specific process includes: acquiring initial edge time-series data of power plant equipment under different monitoring dimensions, including electrical parameter time-series data and mechanical parameter time-series data. The electrical parameter time-series data includes the effective values ​​of three-phase current and the instantaneous values ​​of bus voltage; the mechanical parameter time-series data includes operating temperature and vibration acceleration. For the electrical parameter time-series data, a wavelet filtering algorithm is used to remove noise signals caused by electromagnetic interference to obtain the current peak deviation rate and voltage fluctuation amplitude. For the mechanical parameter time-series data, a mean filtering algorithm is used to remove temperature measurement errors caused by ambient temperature interference to obtain the temperature change gradient and vibration frequency spectrum. The acquired edge time-series data are processed into binary form and combined with the reference range in the database to generate an anomaly screening data table to visually distinguish whether there are time-series anomalies in the edge time-series data. Here, 1 indicates that the corresponding edge time-series data has time-series anomalies, and 0 indicates that the corresponding edge time-series data does not have time-series anomalies. If at least one of the four parameters in the anomaly screening data table is 1, the corresponding edge time-series data is determined to be a verification data segment and uploaded to the cloud for anomaly root cause investigation. If the binary results of all four parameters in the anomaly screening data table are 0, the corresponding edge time-series data is determined to be without anomalies and is not uploaded to the cloud. Instead, the corresponding edge time-series data is stored only on the edge side.

[0036] The specific process of binary processing is as follows: the four parameters after filtering—current peak deviation rate, voltage fluctuation amplitude, temperature change gradient, and vibration frequency spectrum peak—are compared with the reference ranges of the corresponding parameters in the database. If the current peak deviation rate is within the preset deviation threshold (e.g., ±5% of the rated value), the corresponding binary processing result is recorded as 0; otherwise, it is recorded as 1. Similarly, voltage fluctuation amplitude within the allowable fluctuation range (e.g., ±2% of the rated voltage) is recorded as 0, and exceeding it is recorded as 1. Temperature change gradient within the normal gradient range (e.g., ≤2℃ / min) is recorded as 0, and exceeding the range is recorded as 1. Vibration frequency spectrum peak within the safe peak range is recorded as 0, and exceeding the range is recorded as 1. The 0 and 1 values ​​visually indicate whether each parameter is abnormal.

[0037] The specific values ​​mentioned above (such as ±5% of the rated value, ±2% of the rated voltage, etc.) are based on two criteria: first, the design operating thresholds of various parameters in the power plant equipment's factory technical manual are referenced, which are the basic standards for the safe operation of power plant equipment; second, industry operation and maintenance standards (such as power equipment condition monitoring standards) are combined, taking into account both the timeliness of fault warnings and the false judgment rate. The values ​​were finally determined after multiple rounds of actual testing and verification to ensure that the values ​​not only conform to the characteristics of the equipment but also adapt to the needs of on-site operation and maintenance.

[0038] It's important to understand that the selection of the three-phase current RMS value and bus voltage instantaneous value from the electrical parameters is crucial because they directly reflect the load stability and power supply quality of the power station's electrical circuits. Electromagnetic interference can easily generate noise, affecting the judgment of electrical faults such as short circuits and overloads. Among the mechanical parameters, operating temperature and vibration acceleration are core indicators of the operating status of mechanical components. Ambient temperature interference can mask abnormal heat dissipation, while vibration signals can directly reflect mechanical problems such as component wear and imbalance. These two types of parameters respectively cover the core electrical and mechanical dimensions of the equipment; their anomalies can provide early warnings for most equipment faults, and targeted interference handling can improve data reliability.

[0039] Different filtering algorithms are employed because the interference characteristics and data features of electrical and mechanical parameters differ significantly. Electrical parameters (current, voltage) are susceptible to electromagnetic interference, and clutter signals are complex in frequency and highly transient. Wavelet filtering excels at decomposing multi-frequency signals and can accurately separate interference from valid signals. Mechanical parameters (temperature, vibration) exhibit slow-drifting errors due to ambient temperature interference. Mean filtering is more effective at smoothing stationary interference and can preserve key trend features such as temperature gradients and vibration spectral peaks. Algorithm selection based on suitability maximizes interference removal and ensures data validity.

[0040] In this embodiment, by selectively choosing core electrical and mechanical parameters and adapting filtering algorithms, interference from electromagnetic fields and ambient temperature is precisely eliminated, significantly improving the reliability of data such as current peak deviation rate and vibration frequency spectrum peak. The binary visualization initial screening and hierarchical transmission mechanism only uploads verification data containing anomalies, significantly reducing the amount of data transmitted between the cloud and the edge, saving bandwidth resources, while storing data without anomalies on the edge side. This can provide early warning of most equipment failures, improving the efficiency of operation and maintenance response while ensuring the stable operation of power plant equipment.

[0041] Further, the specific steps for investigating the root cause of anomalies are as follows: After the consistency verification is passed, the duration of the anomaly in the edge time series data corresponding to the verified data segment within the preset monitoring period is statistically analyzed and compared with the historical duration of the anomaly in the historical edge time series data of the corresponding power station equipment under the same operating conditions in the cloud. Based on the relative difference between the anomaly duration and the historical anomaly duration, the anomaly matching degree is obtained, which is the absolute value of the difference between the anomaly duration and the historical anomaly duration. If the anomaly matching degree is not less than the reference anomaly matching degree (usually set to 80%), it is determined that the equipment itself is abnormal, the investigation is stopped, and an equipment anomaly warning is issued. If the anomaly matching degree is less than the reference anomaly matching degree, further comparison is made with the database. If the duration of concurrent external interference factors (such as heavy rain, strong winds, and grid voltage fluctuations) exceeds the preset duration, it is determined to be non-equipment-related abnormality, and the investigation is stopped. Otherwise, the corresponding edge time series data is marked as unconfirmed abnormal data and synchronized to the cloud to prompt the preset personnel for auxiliary analysis, ensuring that hidden equipment failures are not overlooked and the impact of unrecorded external interference is not misjudged. If the root cause of the abnormality is still not determined within the preset investigation period or the number of investigations reaches the limit, the corresponding power station equipment is marked as equipment to be re-inspected, and the preset personnel are prompted to conduct manual investigation. The results of the investigation of the root cause of the abnormality (equipment-related abnormality / non-equipment-related abnormality / requiring re-inspection) are synchronized to the cloud for classified archiving and storage.

[0042] Specifically, consistency verification is used to ensure the accuracy and integrity of the verification data segments uploaded to the cloud in the time dimension. The specific process is as follows: the deviation between the sampling timestamp of the verification data segment and the preset sampling interval (usually 10 seconds) is quantified, and the result is expressed as a timestamp consistency score; verification data segments with timestamp consistency scores within the corresponding consistency score range are marked as timestamp qualified data, otherwise, they are marked as timestamp abnormal data; if the binary results of the four parameters in the abnormal initial screening data table corresponding to the timestamp qualified data are all 0, then the consistency verification is deemed qualified and can be uploaded to the cloud; otherwise, the edge side is prompted to supplement the edge time series data of the corresponding time period and re-verify.

[0043] Specifically, the duration of historical anomalies under the same operating conditions for corresponding power plant equipment is extracted from the cloud-based historical fault feature database; this duration is defined as the historical anomaly duration. The consistency score range represents the range between the maximum and minimum values ​​of the historical timestamp consistency scores for each historical power plant equipment under the same operating conditions in the cloud-based historical fault feature database. The aforementioned reference anomaly matching degree and preset sampling interval are set by default based on equipment type, operating conditions, and historical fault data. In practical applications, these settings can be dynamically adjusted based on long-term fault misjudgment rate and data transmission stability, combined with maintenance feedback, to adapt to different scenario requirements.

[0044] It's important to understand that prompting the edge side to supplement and re-verify edge time-series data for the corresponding time period is crucial to ensuring that the verification data uploaded to the cloud is both time-accurate and data-valid, preventing invalid or erroneous data from interfering with subsequent anomaly root cause investigation. The specific reasons are as follows: Firstly, if the timestamp is invalid, it indicates a time deviation in data collection or transmission, which will prevent the cloud from accurately associating the device's concurrent operating status and environmental factors, affecting the timeliness and accuracy of anomaly tracing. Secondly, even if the timestamp is valid, if the initial screening four parameters are abnormal (binary results containing 1), it indicates that the data itself may contain unresolved interference or potential anomalies. Directly uploading this data would increase the cost of ineffective cloud-based investigation and could even lead to misjudgments. Therefore, re-verifying and supplementing the data provides a reliable foundation for subsequent investigations.

[0045] In this embodiment, prolonged external interference (such as heavy rain or power grid fluctuations) can easily lead to abnormal equipment parameters. Since such abnormalities are not due to equipment malfunctions, determining whether the duration of the interference exceeds a preset value allows for a rapid distinction between temporary abnormalities caused by interference and inherent equipment problems, preventing misdiagnosis and reducing ineffective maintenance. By comparing data with historical data, equipment malfunctions are quickly identified. Combined with external interference investigation, misdiagnosis is avoided. The pending data and manual review mechanism address both latent faults and unrecorded interference. Categorized archiving facilitates traceability, significantly improving maintenance efficiency and reducing ineffective investigations and missed fault diagnoses.

[0046] Furthermore, trend comparison analysis is performed based on the initial anomaly screening results. The specific process is as follows: the fluctuation amplitude of each parameter in the verification data segment obtained from the initial anomaly screening is quantified by comparing it with the fluctuation amplitude of historical data under the same conditions in the same period, to obtain the fluctuation trend consistency degree, which reflects the changing trend of the power plant equipment's operating status. It is then determined whether the obtained fluctuation trend consistency degree meets the preset trend stability condition. If the fluctuation trend consistency degree meets the trend stability condition, it is determined that the fluctuation of the power plant equipment's operating status is within the allowable range set by preset personnel. Otherwise, it is determined that the operating status of the power plant equipment has abnormal fluctuations, triggering multi-dimensional correlation analysis and data collaborative management. The trend stability condition indicates that the fluctuation trend consistency degree is greater than the preset fluctuation trend consistency degree (usually set to 80%).

[0047] In this embodiment, the fluctuation amplitude of each parameter (such as current peak deviation rate and temperature change gradient) in the verification data segment is first extracted during the monitoring period. Then, the fluctuation amplitude of the corresponding parameter under the same historical operating conditions in the cloud is retrieved. The ratio of the two fluctuation amplitudes for each set of parameters is calculated, i.e., the division operation, to obtain the single parameter matching ratio. By performing harmonic averaging on all single parameter matching ratios, the fluctuation trend matching degree reflecting the trend of operating status change is finally obtained. Among them, the harmonic averaging process avoids extreme value interference and improves the rationality and reliability of the matching degree result. Under normal circumstances, the preset fluctuation trend matching degree is mainly set based on the historical normal operation data of the power plant equipment: the fluctuation amplitude of each parameter (such as temperature and current) under the long-term stable operating conditions of the equipment is extracted, the fluctuation trend matching degree of the data under the same operating conditions in the same period is calculated, the average value of multiple statistical results is taken, and at the same time, combined with the common standards of stable operation of equipment in the industry, the fault false judgment rate and the missed judgment rate are balanced, and finally 80% is determined as the judgment threshold of stable operating trend of power plant equipment.

[0048] like Figure 3 One of the flowcharts shown is for multi-dimensional correlation analysis and data collaborative management, such as... Figure 4 The second flowchart of the multi-dimensional correlation analysis and data collaborative management shown has the following design logic: First, statistical analysis of edge time-series data under abnormal fluctuations is performed to obtain the data fluctuation amplitude and environmental factor change amplitude within the same time window. After standardization, the synchronization coefficient of each dimension is calculated, and the average is taken to obtain the change synchronization score. By comparing with preset values, environmental factor anomalies and equipment anomalies are distinguished. Then, the real-time operating status is correlated to construct an anomaly cause-state change relationship map. Next, based on the anomaly cause-state change relationship map, the timestamp of the anomaly occurrence is obtained, and the anomaly occurrence interval is calculated. Then, it is determined whether the interval is less than the preset time window: if so, it is marked as a composite anomaly, indicating that multiple anomaly causes interact within a short period of time; if not, it is marked as a single-dimensional anomaly, i.e., caused by a single cause, providing a basis for subsequent anomaly type judgment and realizing more accurate classification and tracing of power plant equipment anomalies.

[0049] Further understanding is needed regarding multi-dimensional correlation analysis and collaborative data management, which includes a first step and a second step. The first step specifically includes: statistically analyzing the edge time series data corresponding to abnormal fluctuations, obtaining the fluctuation amplitude of different edge time series data within the same time window, and synchronously obtaining the change amplitude of the preset environmental factor change state within the corresponding time window; converting all fluctuation amplitudes and change amplitudes into dimensionless values ​​in the 0-1 interval to eliminate dimensional differences, calculating the synchronization coefficient of each edge time series data and change amplitude, and taking the average to obtain the change synchronization score, which measures the synchronization between the edge time series data fluctuation and the preset environmental factor change state.

[0050] Within the same time window, if the edge time series data and the preset environmental factor change state change in the same direction, the corresponding synchronization coefficient is obtained by calculating the overlap ratio of the fluctuation amplitudes of the two and combining it with the reverse conversion value of the time difference of the corresponding peak. If the edge time series data and the preset environmental factor change state change in different directions, the corresponding synchronization coefficient is obtained by calculating the remaining ratio after the fluctuation amplitudes of the two are superimposed in reverse and combining it with the reverse conversion value of the time difference of the corresponding valley.

[0051] If the change synchronization score is greater than the preset change synchronization score, the corresponding anomaly is marked as an environmental factor anomaly, and the cause is traced based on the environmental monitoring data of the same period (such as temperature and humidity change curves, air pressure fluctuation records); otherwise, the corresponding anomaly is marked as an anomaly of the power station equipment itself, and the cause is traced based on the equipment's historical fault data (such as past fault parameters and maintenance records of similar components). The preset change synchronization score is represented by the sum and average of the historical synchronization coefficients of the power station equipment in the multi-dimensional analysis process. The traceability results of the two types of anomalies are collaboratively correlated with the real-time operating status of the power station equipment to construct a relationship map of anomaly causes and status changes, which is used to reflect the correspondence between different causes and the status changes of the power station equipment, and to help quickly locate the scope of the fault impact.

[0052] The second step specifically includes: obtaining the nodes where electrical and mechanical parameters first show abnormalities from the relationship graph of abnormal causes and state changes, and obtaining the corresponding timestamps. The time length between the two timestamps is calculated to obtain the abnormal occurrence interval of electrical and mechanical parameters, which is used to reflect the diffusion speed and sequence of abnormalities between the electrical and mechanical dimensions, and to reflect the cross-dimensional diffusion characteristics of abnormalities. If the obtained abnormal occurrence interval is not greater than the preset time window, the corresponding abnormality is marked as a composite abnormality, and the preset personnel are prompted to conduct cross-dimensional parameter linkage verification. The preset time window is usually set based on the correlation characteristics of the electrical and mechanical systems of the power plant equipment, the historical cross-dimensional fault diffusion time, and industry operation and maintenance standards, and is usually 5-15 minutes. Otherwise, the corresponding abnormality is marked as a single-dimensional abnormality, and the preset personnel are prompted to conduct a special investigation only for the corresponding dimension.

[0053] The 5-15 minute time window is derived from statistical analysis of power plant equipment operation data: extracting historical cases of cross-faults between electrical and mechanical systems, recording the time it takes for an anomaly to spread from one dimension to another, mostly concentrated in 5-12 minutes; combining the physical relationship between electrical and mechanical components of the equipment (such as the response delay of motors and transmission mechanisms), and referring to the industry's standard for the golden period for handling fault propagation, the final 5-15 minute time window is determined to balance timely early warning and prevention of misjudgment.

[0054] Specifically, such as Figure 5The diagram illustrating the relationship between anomaly causes and state changes uses a multi-layered color-coded ladder structure, with different colors corresponding to different anomaly levels, cause types, and dimensional attributes. The top dark red represents high-level anomalies caused by strong environmental factors such as sudden changes in temperature and humidity, and fluctuations in grid voltage. These anomalies are often the first to trigger and have a wide impact. Below, the red and orange ladders correspond to medium-to-high-level anomalies caused by environmental factors such as strong winds and heavy rain, and serious equipment faults such as aging stator winding insulation. Red leans more towards environmental causes, while orange leans more towards problems with the equipment's mechanical system itself. Further down, the green and cyan ladders correspond to moderate anomalies in the equipment's electrical or mechanical systems, such as poor contact in local circuits or insufficient lubrication of components. The blue and purple ladders correspond to lower-level equipment anomalies, such as loose foundation bolts or deviations in component installation accuracy. The bottom light-colored section corresponds to low-level anomalies such as slight sensor drift or oversights in routine inspections, mostly caused by minor problems with the equipment itself or oversights in operation and maintenance management. By using color differentiation, the distribution of anomalies from cause type, dimensional attributes to severity level can be presented intuitively, helping to quickly grasp the core characteristics of anomalies and providing clear guidance for subsequent cross-dimensional or single-dimensional investigations.

[0055] The results of two types of anomaly tracing were categorized and analyzed to clarify the specific causes of environmental anomalies (such as sudden changes in temperature and humidity, and power grid fluctuations) and equipment anomalies (such as circuit aging), as well as the key parameters associated with each cause (such as current, temperature, and vibration). Real-time operating status data of the power plant equipment corresponding to each cause was extracted, and the causes were precisely aligned with the instantaneous values ​​and time-series change curves of parameters during the same period using timestamps, ensuring that the causes and status changes matched in the time dimension. Using the causes as the horizontal dimension and equipment status change parameters as the vertical dimension, a relationship graph of anomaly causes and status changes was constructed. Labels (such as colors and symbols) were used to represent the correspondence between different causes and parameter fluctuations, supplementing information such as the duration of the anomaly and the affected equipment modules, clearly showing how the causes triggered changes in equipment status.

[0056] In this embodiment, by quantifying the synchronicity between equipment data and environmental changes, and combining historical thresholds, environmental interference (such as sudden changes in temperature and humidity) and equipment malfunctions (such as component wear) are accurately distinguished, avoiding ineffective troubleshooting. The constructed relationship graph clearly presents the correspondence between the cause and changes in equipment status, significantly shortening the source tracing time. Based on the abnormal intervals of electrical and mechanical parameters, complex and single-dimensional anomalies are efficiently identified, making cross-dimensional collaborative troubleshooting more targeted and reducing excessive maintenance. Overall, the system achieves accurate determination of the root cause and type of anomalies, optimizes the allocation of maintenance resources, and ensures stable equipment operation.

[0057] Furthermore, equipment anomaly identification is performed, specifically: based on historical fault level characteristics in the equipment fault risk assessment report and real-time edge operation data, cross-validation scores are obtained, and cross-validation is performed simultaneously to quantify the degree of matching between real-time data and historical fault characteristics, thereby determining the current fault level of the equipment; the specific process for cross-validation is as follows:

[0058] If the cross-validation score is within the first validation score range (usually set to 80-95 points), the power plant equipment corresponding to the current edge real-time operation data is determined to meet the characteristics of a minor fault and is initially marked as a Level 1 fault; if the cross-validation score is within the second validation score range (usually set to 50-79 points), the power plant equipment corresponding to the current edge real-time operation data is determined to meet the characteristics of a moderate fault and is initially marked as a Level 2 fault; if the cross-validation score is within the third validation score range (usually set to 0-49 points), the power plant equipment corresponding to the current edge real-time operation data is determined to meet the characteristics of a severe fault and is initially marked as a Level 3 fault; the first validation score range, the second validation score range, and the third validation score range represent mutually exclusive intervals that cover the entire score range, divided from low to high based on the cross-validation score.

[0059] In this score range, 95-100 points indicates a fault-free state, meaning the equipment is not classified as fault-free. Within this range, real-time edge operation data closely matches the characteristics of historical fault-free equipment, the duration of anomalies is close to zero, and the number of fault-related data points is almost identical to historical fault-free data, indicating stable equipment operation and no need to label fault levels. However, it's important to note that in scenarios with extremely high requirements for operational safety and stability, such as critical units in nuclear power plants or main transformer systems in large hydropower stations, a score of 95-100 points alone is insufficient to meet the stringent standards; a score of 100 is required to indicate a fault-free state. In these scenarios, equipment failure could lead to major safety accidents or huge losses. Therefore, real-time edge data must perfectly match the characteristics of historical fault-free equipment, with both the duration of anomalies and the number of fault-related data points being zero, ensuring the equipment is in an absolutely stable operating state.

[0060] The method for obtaining cross-validation scores is as follows: Real-time edge operation data within the corresponding validation score interval during the cross-validation process is obtained and recorded as fault feature data. The abnormal percentage duration of each fault feature data is calculated, representing the proportion of the total abnormal duration of each fault feature data to the total duration of the preset monitoring period. The abnormal percentage durations are summed and the difference is calculated with the average abnormal percentage duration under the same historical fault level to obtain the percentage duration difference. Simultaneously, the difference in the number of fault feature data is obtained, i.e., the difference between the number of fault feature data and the corresponding historical data. The absolute values ​​of the two differences are added together and then converted in reverse to obtain the cross-validation score used to quantify the degree of matching between historical fault level features and edge real-time operation data.

[0061] In this embodiment, the verification score range (80-95 points, 50-79 points, 0-49 points) is determined based on the distribution of characteristic parameters of different levels of faults in the historical fault data of power plant equipment (such as the duration of abnormality and the amount of fault characteristic data), combined with the actual severity of the fault and the priority of operation and maintenance response. In different scenarios (such as high-load operation of new energy power plants and maintenance of old power plant equipment), the range boundaries can be fine-tuned according to the equipment failure rate, the effect of historical fault handling and industry standards to improve the adaptability of the level determination.

[0062] By inversely converting the sum of the absolute values ​​of the differences in duration and data quantity, the relationship between a larger difference and a lower match is transformed into an intuitive scoring logic where a higher score indicates a higher match. This makes the cross-validation score more aligned with the practical needs of fault severity determination and facilitates rapid quantification of the degree of matching. This anomaly identification step quantifies the match between real-time data and historical fault characteristics through cross-validation, accurately classifying fault levels and avoiding subjective judgment errors. The clear scoring range and inverse conversion logic make severity determination efficient and traceable, helping maintenance personnel quickly pinpoint the severity of faults, develop targeted solutions, and improve equipment fault response efficiency.

[0063] like Figure 6 The flowchart shown illustrates the dynamic identification optimization process in cloud-edge collaboration. Its design logic is as follows: First, it determines whether the feature drift is greater than the reference feature drift. If it is, the local anomaly parameters are updated; otherwise, the original identification parameters are maintained. After updating, the impact of the local parameter update and the impact of the cloud priority are mapped. New parameters are generated through compensation calculations. Next, the cloud parsing speed is increased, and it is determined whether the corresponding reference value is reached. If not, monitoring continues. If the parsing speed meets the standard in the next moment, the data upload frequency is dynamically shortened; if not, a cloud interaction anomaly warning is triggered. This achieves dynamic optimization of anomaly identification parameters under cloud-edge collaboration, balancing edge-side parameter adaptability and cloud parsing efficiency. Through hierarchical judgment and feedback adjustment, the system ensures efficient and stable data interaction and anomaly identification when device features drift.

[0064] Further understanding is needed regarding the dynamic identification and optimization of cloud-edge collaboration, which includes: if the feature drift of the real-time edge data within the corresponding verification score interval is greater than the reference feature drift, the edge side automatically updates the local anomaly parameters and compresses and uploads the corresponding drift feature data to the cloud; otherwise, the edge side maintains the original identification parameters and only synchronizes the real-time edge data without anomalies to the cloud; the feature drift represents the result obtained by inputting the real-time load rate change slope and temperature gradient change rate into the feature drift mapping set.

[0065] The automatic update of local anomaly parameters on the edge side follows a specific optimization process: The drift feature data uploaded from the edge side is mapped to the deviation value of the historical fault feature database in the cloud, resulting in the corresponding impact of local parameter updates and the impact of cloud data processing priority. The drift feature data reflects the characteristic offset state of the real-time operating parameters of the edge-side power station equipment as the operating conditions change. The impact of local parameter updates is then compensated with the original identification parameters on the edge side to generate optimized anomaly identification parameters, which replace the original local parameters, completing the local update.

[0066] By prioritizing cloud data processing, the cloud parsing speed of drift feature data is improved. If the cloud parsing speed reaches the reference cloud parsing speed within the preset monitoring period, the upload frequency of edge data is dynamically shortened based on the cloud parsing capacity margin, thereby optimizing the efficiency of cloud-edge data interaction. The upload frequency of edge data represents the time interval for synchronizing data from the edge to the cloud. If the cloud parsing speed does not reach the reference cloud parsing speed within the preset monitoring period, monitoring continues. If the cloud parsing speed still does not reach the reference cloud parsing speed at the end of the next preset monitoring period, a cloud interaction anomaly warning is issued to prompt the preset personnel to check the load of the cloud parsing node and simultaneously check whether the compression format of the drift feature data is compatible with the parsing requirements, so as to quickly restore parsing efficiency.

[0067] Specifically, the reference feature drift is represented by the sum and average of feature drift data recorded by power plant equipment during changes in operating conditions (such as drift values ​​under different loads and environments). The reference cloud parsing speed is represented by the sum and average of the parsing time data of historical power plant equipment drift feature data in the cloud (time taken under different data volumes and compression formats), and then converted into the result representing the corresponding parsing speed. Data such as the slope of real-time load rate change and the rate of change of temperature gradient under different operating conditions of the equipment are collected to extract feature drift samples; the degree of drift influence corresponding to the samples is labeled, and a mapping rule between feature data and drift results is established; combined with the historical optimization effect, iterative adjustments are made to form a mapping set, namely the feature drift mapping set.

[0068] The impact of local parameter updates is the correction magnitude of the original identification parameters on the edge side, determining the direction of parameter optimization. The impact of cloud data processing priority is used to allocate priority weights for cloud parsing resources. Based on the impact of cloud data processing priority, dedicated parsing threads are allocated to high-priority drift feature data, prioritizing computing resources, reducing queuing time, and accelerating parsing speed. The remaining cloud parsing capacity (total processing capacity - current occupied capacity) is monitored. If the remaining capacity is sufficient (e.g., exceeding 30%), it indicates that the cloud can handle more data. The edge side automatically shortens the data upload interval (e.g., originally 10 minutes) to a duration that adapts to the remaining capacity (e.g., 5 minutes), optimizing the efficiency of cloud-edge data interaction.

[0069] In this embodiment, data transmission is tiered by determining feature drift, uploading critical data only when drift exceeds limits. This significantly reduces cloud-edge interaction, saving bandwidth resources and reducing cloud processing pressure. A dynamic update mechanism for local parameters on the edge side adapts to changes in equipment operating conditions in real time, preventing identification failures due to feature shifts and improving anomaly identification accuracy. Cloud priority scheduling and parsing capabilities are linked for rapid parsing of important drift data, while dynamic adjustment of upload frequency ensures precise resource allocation, effectively improving cloud-edge interaction efficiency. The overall process achieves intelligent collaboration and dynamic optimization of cloud-edge resources, significantly improving the real-time performance, accuracy, and robustness of power plant equipment anomaly identification.

[0070] It should be added that, such as Figure 7 The homepage of the power plant operation and maintenance management platform, as shown, utilizes a cloud-edge collaborative method for rapid identification of power plant equipment anomalies. The page displays data such as the total number of devices, online / offline / alarm status, and load rate and operating status of devices in different zones. All of this data relies on the cloud-edge collaborative architecture, with real-time device parameters collected at the edge and rapidly analyzed by cloud algorithms, providing fundamental data support for anomaly identification. The inverter overload warning in the initial anomaly screening warning list is a preliminary result of this identification method. Viewing the details allows for linking to subsequent fault management and cloud-edge collaborative modules, making the homepage an overview window for rapid anomaly identification and helping maintenance personnel perceive equipment anomaly trends immediately.

[0071] like Figure 8 The diagram shows the fault management module in the power plant operation and maintenance management platform. The cross-validation and calculation process combines multi-dimensional equipment characteristics (such as efficiency and temperature) at the edge with historical fault models in the cloud to further verify the initial screening results of anomalies. In the equipment fault level determination results, the determination of photovoltaic panel feature exceeding the limit relies on the linkage analysis of real-time edge monitoring and cloud algorithm under cloud-edge collaboration to ensure the accuracy of fault level (such as level 2 fault) determination. At the same time, triggering operations such as edge parameter updates can feed back the fault analysis results to the cloud-edge collaboration system, optimize the parameter model of subsequent anomaly identification, and realize the closed-loop linkage of fault management and anomaly identification methods.

[0072] like Figure 9The diagram illustrates the cloud-edge collaboration module in the power plant operation and maintenance management platform. Data such as load rate and temperature gradient change rate in the feature drift calculation are collected in real time from the edge and transmitted to the cloud for computation, providing dynamic feature basis for anomaly identification. Operations such as edge parameter updates and cloud intelligent compression of uploaded parameters in the optimization status ensure efficient and accurate data interaction between the edge and the cloud, allowing the smooth flow of real-time, massive data required for anomaly identification. Records such as upload frequency adjustment and data compression in the dynamic optimization operations and records demonstrate that data transmission and processing are optimized through cloud-edge collaboration, thereby laying a solid technical foundation for rapid anomaly identification and ensuring that the identification method runs accurately and quickly under efficient edge-cloud collaboration.

[0073] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0074] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0075] In various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0076] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0077] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration, characterized in that, Includes the following steps: Step 1: On the edge side of power plant equipment operation status monitoring, preprocess and initially screen the edge time series data under cloud-edge collaboration to ensure the integrity and reliability of the verification data segments and key features uploaded to the cloud. The edge time series data is used to reflect the real-time operation status and core parameter change trends of power plant equipment. Step 2: Based on the initial screening results of anomalies, conduct trend comparison analysis to determine the fluctuation pattern of the operating status of power plant equipment. At the same time, conduct multi-dimensional correlation analysis and data collaborative management to generate a fault risk assessment report for visualizing the fault types and impact range of power plant equipment. Step 3: Based on the equipment failure risk assessment report and real-time operating data from the edge side, equipment anomaly identification is performed to determine the failure level of the power plant equipment. At the same time, dynamic identification optimization of cloud-edge collaboration is carried out to improve the efficiency and reliability of power plant equipment anomaly identification. The real-time operating data from the edge side is used to reflect the real-time operating conditions and dynamic changes of parameters of the power plant equipment after verification. The aforementioned multi-dimensional correlation analysis and collaborative data management include: Statistical analysis of edge time series data corresponding to abnormal fluctuations, obtaining the fluctuation amplitude of different edge time series data within the same time window, and simultaneously obtaining the change amplitude of preset environmental factors within the corresponding time window; All fluctuation and change amplitudes are standardized, and the synchronization coefficients of each edge time series data and change amplitude are calculated. The average value is then used to obtain the change synchronization score, which measures the synchronization between the fluctuation of edge time series data and the change state of preset environmental factors. Within the same time window, if the edge time series data and the preset environmental factor change state change in the same direction, the corresponding synchronization coefficient is obtained by calculating the overlap ratio of the fluctuation amplitude of the two and combining it with the reverse conversion value of the time difference of the corresponding peak. If the direction of change of the edge time series data is different from that of the preset environmental factor change state, the corresponding synchronization coefficient is obtained by calculating the remaining proportion after the fluctuation amplitudes of the two are superimposed in reverse, and combined with the reverse conversion value of the time difference of the corresponding valley value. If the change synchronization score is greater than the preset change synchronization score, the corresponding anomaly will be marked as an environmental factor anomaly, and the cause will be traced based on the environmental monitoring data of the same period. Otherwise, the corresponding anomaly will be marked as an anomaly of the power station equipment itself, and the cause will be traced based on the equipment's historical fault data. The tracing results of the two types of anomalies are correlated with the real-time operating status of the power plant equipment to construct a relationship map of anomaly causes and status changes, which reflects the correspondence between different causes and changes in the status of power plant equipment.

2. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 1, characterized in that, The specific process of preprocessing and anomaly screening of edge time-series data under cloud-edge collaboration includes: Acquire initial edge time-series data of power plant equipment under different monitoring dimensions, including electrical parameter time-series data and mechanical parameter time-series data. The electrical parameter time-series data includes the effective value of three-phase current and the instantaneous value of bus voltage. The mechanical parameter time-series data includes operating temperature and vibration acceleration. For electrical parameter time series data, wavelet filtering algorithm is used to remove noise signals caused by electromagnetic interference in order to obtain the peak current deviation rate, which reflects the load stability of the electrical circuit of the power station equipment, and the voltage fluctuation amplitude, which reflects the power grid power supply fluctuation. For the time series data of mechanical parameters, the mean filtering algorithm is used to remove the temperature measurement error caused by the interference of ambient temperature, so as to obtain the temperature change gradient reflecting the heat dissipation state of the mechanical components of the power station equipment, and the vibration frequency spectrum peak reflecting the operating stability of the mechanical components of the power station equipment. The acquired edge time series data are processed into binary data and combined with the reference range in the database to generate an anomaly screening data table to visually distinguish whether there are time series anomalies in the edge time series data. The edge time series data includes current peak deviation rate, voltage fluctuation amplitude, temperature change gradient and vibration frequency spectrum peak. If at least one of the four parameters in the initial screening data table is 1, the corresponding edge time series data will be identified as a verification data segment and uploaded to the cloud for anomaly root cause investigation. If the binary results of all four parameters in the anomaly screening data table are 0, the corresponding edge time series data will be determined as having no anomalies and will not be uploaded to the cloud. Instead, the corresponding edge time series data will be stored only on the edge side.

3. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 2, characterized in that, The specific steps for investigating the root cause of anomalies are as follows: After the consistency verification is passed, the abnormal duration of the edge time series data corresponding to the verification data segment within the preset monitoring period is statistically analyzed and compared with the historical abnormal duration of the historical edge time series data of the corresponding power station equipment under the same operating conditions in the cloud. Based on the relative difference between the abnormal duration and the historical abnormal duration, the abnormal matching degree is obtained. The consistency verification is used to ensure the accuracy and data integrity of the verification data segment uploaded to the cloud in the time dimension. If the abnormal matching degree is not less than the reference abnormal matching degree, it is determined that the equipment itself is abnormal, the investigation is stopped, and an equipment abnormality warning is issued. If the abnormal matching degree is less than the reference abnormal matching degree, the occurrence duration of the same external interference factors in the database is further compared. If the occurrence duration is greater than the preset occurrence duration, it is determined that it is not an abnormality of the device itself and the investigation is stopped. Otherwise, the corresponding edge time series data is marked as abnormal data to be confirmed and synchronized to the cloud to prompt the preset personnel for auxiliary analysis. If the root cause of the anomaly is not determined within the preset inspection period or the maximum number of inspections is reached, the corresponding power plant equipment will be marked as equipment to be re-inspected, and the preset personnel will be prompted to conduct manual inspection. The results of the anomaly root cause investigation will be synchronized to the cloud for categorized archiving and storage.

4. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 3, characterized in that, The consistency verification process involves the following steps: The deviation between the sampling timestamp of the reviewed data segment and the preset sampling interval is quantified, and the result is expressed as a timestamp consistency score. Data segments whose timestamp consistency scores are within the corresponding consistency score range are marked as qualified timestamp data; otherwise, they are marked as abnormal timestamp data. If the binary results of all four parameters in the abnormal initial screening data table corresponding to the timestamp qualified data are all 0, then the consistency check is deemed qualified and the data is uploaded to the cloud. Otherwise, the edge side is prompted to collect additional edge time series data for the corresponding time period and re-verify.

5. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 1, characterized in that, The specific process for determining the trend comparison based on the initial anomaly screening results is as follows: The fluctuation range of each parameter in the verification data segment obtained by the initial screening of anomalies is quantified by the proportion of the fluctuation range of the same period and the same operating conditions in the historical data, so as to obtain the fluctuation trend matching degree to reflect the changing trend of the operating status of the power plant equipment. Determine whether the obtained fluctuation trend matches the preset trend stability conditions; If the fluctuation trend matches the trend stability condition, it is determined that the fluctuation of the operating status of the power station equipment is within the allowable range; otherwise, it is determined that the operating status of the power station equipment has abnormal fluctuations, triggering multi-dimensional correlation analysis and data collaborative management. The trend stability condition indicates that the degree of agreement between the fluctuation trends is greater than the preset degree of agreement between the fluctuation trends.

6. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 1, characterized in that, The multi-dimensional correlation analysis and data collaborative management also include: From the relationship graph of abnormal cause-state change, the nodes where the electrical parameters and mechanical parameters first appear abnormal are obtained respectively, and the corresponding timestamps are obtained. The time length between the two timestamps is calculated to obtain the abnormal occurrence interval of electrical parameters and mechanical parameters, which is used to reflect the diffusion speed and sequence of abnormalities between electrical and mechanical dimensions. If the interval between the occurrence of an anomaly is no greater than the preset time window, the corresponding anomaly will be marked as a composite anomaly, and the preset personnel will be prompted to perform cross-dimensional parameter linkage verification. The composite anomaly indicates that there is a correlation between electrical parameter and mechanical parameter anomalies. Conversely, the corresponding anomaly will be marked as a single-dimensional anomaly, and the designated personnel will be prompted to conduct a specific investigation only on the corresponding dimension.

7. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 1, characterized in that, The process of identifying device malfunctions specifically involves: Based on the historical fault level characteristics in the equipment failure risk assessment report and the real-time edge operation data, cross-validation scores are obtained, and cross-validation is performed to quantify the degree of matching between real-time data and historical fault characteristics. The specific process for performing cross-validation is as follows: If the cross-validation score falls within the first validation score range, it is initially marked as a level one fault. If the cross-validation score falls within the second validation score range, it is initially labeled as a level 2 fault. If the cross-validation score is within the third validation score range, it is initially marked as a level three fault; The first verification score interval, the second verification score interval, and the third verification score interval represent mutually exclusive intervals that cover the entire score range, divided from low to high based on cross-validation scores. The fault severity corresponding to the first-level fault, the second-level fault, and the third-level fault increases from low to high, and the urgency of matching and processing gradually increases.

8. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 7, characterized in that, The method for obtaining the cross-validation score is as follows: Obtain real-time running data at the edge of the corresponding validation score interval during the cross-validation process and record it as fault feature data. Calculate the abnormal proportion and duration of each fault feature data. The sum of the duration of each abnormality percentage is calculated, and the difference is calculated with the average duration of the abnormality percentage under the same fault level in history to obtain the duration difference of the percentage. At the same time, the difference in the number of fault feature data is obtained. The absolute values ​​of the two differences are added together and then converted in reverse to obtain the cross-validation score, which is used to quantify the degree of matching between historical fault level features and real-time edge operation data.

9. The method for rapid identification of power plant equipment anomalies based on cloud-edge collaboration as described in claim 8, characterized in that, The dynamic identification and optimization for cloud-edge collaboration includes: If the feature drift of the real-time running data at the edge of the corresponding verification score interval is greater than the reference feature drift, the edge side will automatically update the local anomaly parameters and compress and upload the corresponding drift feature data to the cloud. Otherwise, maintain the original identification parameters and only synchronize real-time data from the edge side without anomalies to the cloud; The feature drift degree represents the result obtained by inputting the real-time load rate change slope and temperature gradient change rate into the feature drift mapping set; The edge side automatically updates local anomaly parameters, and the specific optimization process is as follows: The drift feature data uploaded from the edge side is mapped to the deviation value of the historical fault feature database in the cloud to obtain the corresponding local parameter update impact and cloud data processing priority impact. The local parameter update impact is compensated by the original edge-side identification parameters to generate optimized edge-side anomaly identification parameters, which are then used to replace the original local parameters, thus completing the local update. By prioritizing cloud data processing, the cloud parsing speed of drift feature data can be improved. If the cloud parsing speed corresponding to the preset monitoring period is increased to the reference cloud parsing speed, the upload frequency of edge data will be dynamically shortened based on the cloud parsing capacity margin, thereby optimizing the efficiency of cloud-edge data interaction. If the cloud parsing speed does not reach the reference cloud parsing speed within the preset monitoring period, monitoring will continue. If the cloud parsing speed still does not reach the reference cloud parsing speed at the end of the next preset monitoring period, a cloud interaction anomaly warning will be issued. The drift feature data is used to reflect the characteristic offset state of the real-time operating parameters of the edge-side power station equipment as the operating conditions change, and the upload frequency of the edge-side data represents the time interval for synchronizing data from the edge side to the cloud.

Images