Full-condition dynamic intelligent monitoring method, device and system for pumped storage power station

Abstract

The application provides a full-condition dynamic intelligent monitoring method, device and system for a pumped storage power station, the method comprising: acquiring multi-source data of full-life cycle of all devices of the pumped storage power station to construct a multi-dimensional data set and perform multi-dimensional quantitative evaluation; dividing operating conditions for the devices whose monitoring priority reaches a preset level, calculating fixed threshold adaptation degrees of each condition under different operating directions, identifying high-risk conditions with threshold failure and performing risk grading; splitting a two-way operating data set based on the multi-dimensional data set, calculating a two-way causal correlation relationship between devices, constructing a time-space coupling correlation feature set, constructing a full-condition dynamic adaptation model cluster, and intelligently monitoring and warning the conditions of the pumped storage power station devices whose risk level reaches a preset threshold, which can greatly reduce the false alarm and missed alarm rate of device monitoring, realize early fault warning and rapid root cause positioning, and improve the operation safety and operation and maintenance efficiency of the pumped storage power station.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to a dynamic intelligent monitoring method, device and system for all operating conditions of a pumped storage power station. Background Technology

[0002] With the accelerated construction of new power systems, pumped storage power stations, as the most technologically mature and largest-capacity energy storage power source, undertake the core tasks of peak shaving, frequency regulation, phase regulation, and emergency backup in the power system. Compared with conventional hydropower stations, pumped storage power stations have significant characteristics such as bidirectional reversible operation, frequent switching of operating conditions, a high proportion of transition processes, and strong global coupling. The core units can switch operating conditions more than 10 times per day on average, with the transition process accounting for more than 30% of the operating time, and equipment failures frequently occur during the switching of operating conditions.

[0003] Current alarm analysis methods based on fixed thresholds ignore the significant differences in normal operating conditions under various working conditions and fail to consider the correlation between equipment faults, making it difficult to identify and locate fault sources. Pumped storage power stations have multiple operating conditions and a high proportion of transient processes. Existing technologies cannot meet the intelligent monitoring needs of pumped storage power stations, and there is an urgent need to develop a dynamic intelligent monitoring technology for all operating conditions specifically designed for the characteristics of pumped storage power stations. Summary of the Invention

[0004] The purpose of this application is to provide a dynamic intelligent monitoring method, device and system for all operating conditions of pumped storage power stations, which can significantly reduce the false alarm and missed alarm rate of equipment monitoring, realize early warning of faults and rapid root cause location, and improve the operation safety and maintenance efficiency of pumped storage power stations.

[0005] Firstly, this application provides a dynamic intelligent monitoring method for all operating conditions of a pumped storage power station. The method includes: acquiring multi-source data of all equipment in the pumped storage power station throughout its entire lifecycle; constructing a multidimensional dataset based on the multi-source data, and performing multi-dimensional quantitative evaluation of all equipment based on the multidimensional dataset to determine the equipment monitoring priority classification results; dividing the operating conditions for equipment whose monitoring priority reaches a preset level, calculating the fixed threshold adaptability of each operating condition under different operating directions, identifying high-risk operating conditions with threshold failure and classifying the risks; splitting the bidirectional operating dataset based on the multidimensional dataset, calculating the bidirectional causal relationship between equipment, quantifying the temporal influence intensity and delay characteristics between parameters under different operating conditions, and constructing a spatiotemporal coupling correlation feature set; constructing a dynamic adaptation model cluster for all operating conditions based on the spatiotemporal coupling correlation feature set, and intelligently monitoring and warning of high-risk operating conditions where the risk level of the pumped storage power station equipment reaches a preset threshold; the intelligent monitoring and warning includes dynamic identification of operating conditions, steady-state adaptive monitoring, transient monitoring of transition processes, and dynamic model scheduling.

[0006] Furthermore, the steps outlined above for constructing a multidimensional dataset based on multi-source data and performing multi-dimensional quantitative evaluation of all equipment at the station based on the multidimensional dataset to determine the equipment monitoring priority classification results include: preprocessing the multi-source data; preprocessing includes: outlier removal, missing value completion, timestamp synchronization, and data normalization; labeling the preprocessed multi-source data to construct a multidimensional dataset; labels include: operating condition labels, fault labels, and environmental labels; performing multi-dimensional quantitative evaluation of each device based on the multidimensional dataset to determine the comprehensive risk score for each device; and classifying the monitoring priorities of the devices according to the comprehensive risk scores corresponding to multiple devices to obtain the equipment monitoring priority classification results.

[0007] Furthermore, the steps described above, which involve multi-dimensional quantitative evaluation of each device based on a multidimensional dataset to determine the comprehensive risk score for each device, include: for each device, performing the following steps: statistically analyzing the number of operating condition switching times and the maximum load change rate of a single switching within a preset period to calculate the operating condition impact intensity of the device; extracting historical fault records of the device and calculating the historical fault contribution based on fault occurrence frequency, fault downtime, and fault impact range; calculating the dynamic change amplitude of each monitoring parameter of the device under different operating conditions to obtain the dynamic response characteristics of the parameters; and weightedly integrating the operating condition impact intensity, historical fault contribution, and parameter dynamic response characteristics to obtain the comprehensive risk score of the device.

[0008] Furthermore, the steps outlined above for classifying operating conditions for equipment with monitoring priority reaching a preset level, calculating the fixed threshold fit for each operating condition under different operating directions, identifying high-risk operating conditions with threshold failure, and classifying risks include: dividing the operating conditions of the pumped storage power station into steady-state operating conditions and transient transition operating conditions; dividing the same steady-state operating condition into multiple sub-operating condition segments according to upstream and downstream water level ranges and ambient temperature ranges; dividing the transient transition operating condition into four sub-operating condition segments: start-up transition, shutdown transition, bidirectional switching transition, and load mutation transition; for each sub-operating condition segment, based on the fixed threshold range set by the monitoring system for each parameter of equipment with monitoring priority reaching a preset level, calculating the deviation between the parameter distribution under the sub-operating condition and the fixed threshold range, and obtaining the fixed threshold fit for the sub-operating condition; when the fixed threshold fit for any sub-operating condition segment is lower than the preset threshold, it is determined to be a high-risk operating condition, and risk classification is performed in conjunction with the failure rate and severity of the sub-operating condition.

[0009] Furthermore, the steps described above, which involve splitting the multidimensional dataset into bidirectional operational datasets, calculating bidirectional causal relationships between devices, quantifying the temporal influence intensity and delay characteristics of parameters under different operating conditions, and constructing a spatiotemporal coupling correlation feature set, include: splitting the multidimensional dataset into two independent datasets based on the pumping operation direction and the power generation operation direction; calculating the global influence coefficient of global environmental parameters on device monitoring parameters for each dataset; using environmental parameters with global influence coefficients greater than a preset threshold as basic influence factors, and employing Granger causality test combined with temporal mutual information analysis to calculate the bidirectional causal correlation matrix between device parameters under different operating conditions; and based on the bidirectional causal correlation matrix between device parameters, selecting parameter pairs with stable influence intensity greater than a preset threshold, and constructing spatiotemporal coupling correlation feature sets by operating direction and operating condition, combined with their temporal delay characteristics.

[0010] Furthermore, the steps for calculating the global impact coefficient of global environmental parameters on equipment monitoring parameters include: obtaining synchronous time series data within a preset historical period corresponding to the global environmental parameters and equipment monitoring parameters; establishing a regression model between the global environmental parameters and equipment monitoring parameters using a linear regression method based on the synchronous time series data; and calculating the determination coefficient of the regression model as the global impact coefficient of the global environmental parameters on equipment monitoring parameters.

[0011] Furthermore, the steps described above for constructing a full-condition dynamic adaptation model cluster based on a spatiotemporal coupled correlation feature set to intelligently monitor and warn of high-risk operating conditions where the risk level of pumped storage power station equipment reaches a preset threshold include: the full-condition dynamic adaptation model cluster includes: a condition dynamic identification unit, a steady-state adaptive monitoring unit, a transient monitoring unit for the transition process, and a model dynamic scheduling unit; the condition dynamic identification unit identifies the current operating condition type and stage in real time; the steady-state adaptive monitoring unit outputs the probability of equipment anomalies within a preset time period based on core key equipment parameters, related equipment parameters, and global environmental parameters; the transient monitoring unit outputs the probability of anomalies in the transition process and suspected fault locations based on high-resolution time-series data of the entire transition process; the model dynamic scheduling unit automatically switches the corresponding monitoring model according to the output of the condition dynamic identification unit and preloads the model parameters of the target operating condition when a condition switching command is received; the output results of the full-condition dynamic adaptation model cluster and the fixed threshold judgment results are integrated to generate multi-level alarm signals, and the root cause of the fault is automatically deduced based on the spatiotemporal coupled correlation feature set.

[0012] Furthermore, the above method also includes: recording the entire process of each working condition switching at high resolution and constructing a standard template library for working condition switching; calculating the dynamic deviation between the current switching process and the standard template in real time, and triggering an alarm when the deviation exceeds a preset threshold; periodically updating the equipment monitoring priority classification results, high-risk working condition set, and spatiotemporal coupling correlation feature set based on the latest operating data; and dynamically adjusting model parameters and alarm thresholds according to the equipment performance degradation trend to achieve system self-learning and self-optimization.

[0013] Secondly, this application also provides a dynamic intelligent monitoring device for all operating conditions of a pumped storage power station. The device includes: a data acquisition module for acquiring multi-source data throughout the entire lifecycle of all equipment in the pumped storage power station; a multi-dimensional data fusion and evaluation module for constructing a multi-dimensional dataset based on the multi-source data and performing multi-dimensional quantitative evaluation of all equipment based on the multi-dimensional dataset to determine the equipment monitoring priority classification results; and a threshold adaptation analysis module for dividing equipment with a preset monitoring priority into bidirectional operating conditions, calculating the fixed threshold adaptation degree of each operating condition under different operating directions, and identifying threshold failures. The system includes a high-risk operating condition detection module and a risk classification module; a correlation analysis module for bidirectional splitting of the operating dataset based on a multidimensional dataset, calculating the bidirectional causal relationship between devices, quantifying the temporal influence intensity and delay characteristics of parameters under different operating conditions, and constructing a spatiotemporal coupling correlation feature set; and a model cluster early warning module for constructing a dynamic adaptation model cluster for all operating conditions based on the spatiotemporal coupling correlation feature set, and intelligently monitoring and warning of high-risk operating conditions where the risk level of pumped storage power station equipment reaches a preset threshold. The intelligent monitoring and early warning includes dynamic identification of operating conditions, steady-state adaptive monitoring, transient monitoring of transition processes, and dynamic model scheduling.

[0014] Thirdly, this application also provides a dynamic intelligent monitoring system for all operating conditions of a pumped storage power station, the system being used to perform the method described in the first aspect.

[0015] The method, device, and system for dynamic intelligent monitoring of pumped storage power stations under all operating conditions provided in this application generate a priority classification system for equipment monitoring through multi-dimensional quantitative evaluation, identify high-risk operating conditions by combining global environmental parameter correction, quantify the bidirectional causal relationship between equipment using a spatiotemporal decoupling algorithm, and construct a dynamic model cluster consisting of operating condition identification, steady-state monitoring, and transient monitoring of transition processes to achieve dual-track intelligent monitoring under all operating conditions. This invention can significantly reduce the false alarm and missed alarm rates of equipment monitoring, achieve early warning of faults and rapid root cause location, and improve the operational safety and maintenance efficiency of pumped storage power stations. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a dynamic intelligent monitoring method for all operating conditions of a pumped storage power station, provided as an embodiment of this application; Figure 2 A flowchart illustrating the overall process of a dynamic intelligent monitoring method for a pumped storage power station under all operating conditions, as provided in this application embodiment. Figure 3 A schematic diagram illustrating the principle of a spatiotemporal decoupling correlation analysis algorithm provided in this application embodiment; Figure 4 A schematic diagram illustrating a transition process timing segmentation modeling provided in an embodiment of this application; Figure 5 An overall architecture diagram of a dynamic intelligent monitoring system for all operating conditions of a pumped storage power station provided in an embodiment of this application; Figure 6 A logical diagram illustrating the entire process of working condition switching control provided in this application embodiment; Figure 7 This is a structural block diagram of a dynamic intelligent monitoring device for a pumped storage power station under all operating conditions, provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To address the problems of equipment screening failure, incomplete operating condition coverage, correlation analysis errors, and poor model adaptability in existing technologies for pumped storage power station applications, this application provides a method, device, and system for full-condition dynamic intelligent monitoring of pumped storage power stations. This system covers the entire process of data governance, equipment classification, operating condition identification, correlation analysis, model building, monitoring and early warning, and closed-loop optimization. It is adapted to the operational characteristics of pumped storage power stations, including bidirectional operation, high-frequency operating condition switching, a large proportion of transition processes, and significant global environmental impacts, thereby solving the aforementioned problems.

[0020] To facilitate understanding of this embodiment, a detailed description of a dynamic intelligent monitoring method for a pumped storage power station under all operating conditions, as disclosed in this application, will be provided first. Figure 1 A flowchart of a dynamic intelligent monitoring method for a pumped storage power station under all operating conditions, provided in this application embodiment, is included in the following steps: Step S102: Obtain multi-source data for the entire lifecycle of all equipment in the pumped storage power station; The system collects multi-source heterogeneous data from all equipment throughout their entire lifecycle via standardized industrial communication interfaces. This includes, but is not limited to: second-level operating data from the computer monitoring system in Safety Zone I (including upstream and downstream water level data and environmental parameter data), millisecond-level vibration and temperature data from the unit status monitoring system in Safety Zone II, and ledger data, maintenance records, and fault reports from the production management system in the management information zone.

[0021] Step S104: Construct a multidimensional dataset based on multi-source data, and conduct a multidimensional quantitative evaluation of all equipment in the station based on the multidimensional dataset to determine the equipment monitoring priority classification results; In this step, multi-source data can be preprocessed first, and then a multidimensional dataset can be constructed based on the preprocessed data. Preprocessing may include outlier removal, missing value completion, timestamp synchronization, and data normalization. Three core labels are automatically assigned to the preprocessed data: operating condition labels, fault labels, and environmental labels, with manual correction supported.

[0022] Furthermore, each piece of equipment is quantitatively assessed from three dimensions to comprehensively reflect its operational risks. All calculations are based on operational data within a preset historical period. Based on the distribution range of the overall risk score for all equipment at the station, the equipment is divided into three levels: core critical equipment, important auxiliary equipment, and general equipment. The monitoring priority decreases progressively for each level.

[0023] Step S106: For equipment whose monitoring priority reaches the preset level, divide the operating conditions, calculate the fixed threshold adaptability of each operating condition under different operating directions, identify high-risk operating conditions of threshold failure and classify the risks. Equipment whose monitoring priority reaches the preset level refers to core critical equipment and important auxiliary equipment. This step, through full-condition threshold adaptation analysis and high-risk condition identification, can identify high-risk conditions that are prone to false alarms or missed alarms due to fixed threshold failure, achieving full-condition coverage without blind spots. The specific identification process will be detailed later.

[0024] In this step, based on fixed threshold adaptability, historical failure rate, and failure severity, all operating conditions are divided into three levels: Level 1 high-risk operating condition, Level 2 high-risk operating condition, and normal operating condition. High-resolution monitoring for the transition process is activated for Level 1 high-risk operating conditions, large-scale steady-state monitoring is activated for Level 2 high-risk operating conditions, and conventional threshold monitoring is used for normal operating conditions.

[0025] Step S108: Based on the multidimensional dataset, split the bidirectional running dataset and calculate the bidirectional causal relationship between devices, quantify the temporal influence intensity and delay characteristics between parameters under different operating conditions, and construct a spatiotemporal coupling correlation feature set; To address the strong bidirectional coupling characteristic of pumped storage power station equipment, this step involves splitting the bidirectional operational dataset to calculate the bidirectional causal relationships between devices. This quantifies the temporal influence strength and delay characteristics of parameters under different operating conditions, ultimately constructing a spatiotemporal coupling correlation feature set. This feature set is used to quantify the causal relationships between devices, parameters, and between the environment and devices, providing high-quality input features for the model. Each record in the spatiotemporal coupling correlation feature set contains the following fields: operating direction, operating condition type, device to which the influencing parameter belongs, name of the influencing parameter, device to which the affected parameter belongs, name of the affected parameter, strength of stable influence, and time delay.

[0026] Step S110: Construct a dynamic adaptation model cluster for all operating conditions based on the spatiotemporal coupling correlation feature set, and intelligently monitor and warn of high-risk operating conditions where the risk level of pumped storage power station equipment reaches a preset threshold.

[0027] This step constructs a model cluster adapted to the high-frequency switching characteristics of pumped storage systems, enabling differentiated and accurate monitoring of steady-state and transient processes. The aforementioned intelligent monitoring and early warning system includes dynamic condition identification, steady-state adaptive monitoring, transient process monitoring, and model dynamic scheduling. The full-condition dynamically adapted model cluster comprises: a dynamic condition identification unit, a steady-state adaptive monitoring unit, a transient process monitoring unit, and a model dynamic scheduling unit. Multiple identification results are output through parallel processing by different units.

[0028] By integrating the output of large model clusters with traditional fixed threshold judgment results, multi-level early warning information is generated: Multi-source anomaly fusion rules are established, and any anomaly triggered by any judgment path will trigger an alarm through the platform. Based on the anomaly probability and fault severity, alarms are divided into three levels: Level 1 alarm (urgent), Level 2 alarm (important), and Level 3 alarm (general). Furthermore, based on spatiotemporal coupled correlation feature sets, the fault propagation path and root cause can be automatically deduced in descending order of correlation strength, providing targeted handling suggestions for operations and maintenance personnel.

[0029] The dynamic intelligent monitoring method for pumped storage power stations under all operating conditions provided in this application generates an equipment priority classification system through multi-dimensional quantitative evaluation, identifies high-risk operating conditions by combining global environmental parameters, and quantifies the bidirectional causal relationship between equipment using a spatiotemporal decoupling algorithm. It constructs a dynamic model cluster consisting of operating condition identification, steady-state monitoring, and transient monitoring of transition processes, achieving dual-track intelligent monitoring under all operating conditions. This significantly reduces the false alarm and missed alarm rates of equipment monitoring, enables early fault warning and rapid root cause location, and improves the operational safety and maintenance efficiency of pumped storage power stations.

[0030] This application also provides another method for dynamic intelligent monitoring of pumped storage power stations under all operating conditions. This method is implemented based on the previous embodiment. See [link to previous embodiment]. Figure 2 As shown, the specific process includes the following: S1, Multi-source data fusion and device priority classification.

[0031] This step is used to accurately distinguish monitoring priorities from the massive number of devices across the entire site, avoiding redundant computing power, while ensuring that high-risk devices receive focused monitoring.

[0032] S11. Multi-source data acquisition and standardized preprocessing.

[0033] Multi-source heterogeneous data from all equipment throughout its entire lifecycle is collected via standardized industrial communication interfaces. This includes, but is not limited to: second-level operational data from the Safety Zone I computer monitoring system (including upstream and downstream water level data and environmental parameter data); millisecond-level vibration and temperature data from the Safety Zone II unit status monitoring system; ledger data, maintenance records, and fault reports from the Management Information Zone production management system. All collected data undergoes a unified, standardized preprocessing procedure. Outlier removal: A statistical distribution-based method is used to identify and remove abrupt changes in data that exceed the reasonable physical range, as well as continuous constant data caused by sensor malfunctions; the span value of all parameters in a single monitoring record is calculated, and when the span value exceeds a preset threshold, the record is determined to be an outlier and removed.

[0034] Missing value completion: For short-term missing data with a duration less than the preset threshold, linear interpolation is used for completion; for long-term missing data with a duration greater than or equal to the preset threshold, historical averages under the same working conditions and environmental conditions are used for completion, and the completed data is marked with a low confidence level.

[0035] Timestamp synchronization: The timestamps of all data are synchronized to the power station's GPS clock reference to ensure time consistency of data from different systems and sampling frequencies, with the time synchronization error not exceeding a preset threshold.

[0036] Data normalization: Normalize monitoring parameters of different dimensions to eliminate the impact of dimensional differences on subsequent analysis.

[0037] S12. Construction of a multidimensional labeling system: This involves labeling the preprocessed multi-source data to construct a multidimensional dataset. Three core labels are automatically assigned to the preprocessed data, with manual correction supported. Operating condition labels cover all operating states of pumped storage power stations, divided into steady-state operation and transient operation categories. The steady-state operation category includes pumping steady-state, power generation steady-state, phase-shifting steady-state, and standby shutdown; the transient operation category includes pumping start-up, power generation start-up, pumping shutdown, power generation shutdown, pumping to power generation, power generation to pumping, pumping to phase shifting, phase shifting to pumping, power generation to phase shifting, phase shifting to power generation, load shedding, and emergency shutdown.

[0038] Fault Labels: These include the fault occurrence time, fault end time, fault location, fault type, fault severity, and fault handling records. Fault severity is categorized into three levels: general fault, important fault, and major fault.

[0039] Environmental label: includes upstream and downstream water level range, ambient temperature range, ambient humidity range, and seasonal information.

[0040] S13. Multi-dimensional quantitative assessment of equipment: Based on a multidimensional dataset, a multi-dimensional quantitative assessment of each device is conducted to determine the comprehensive risk score of each device.

[0041] Each device is quantitatively assessed across three dimensions to comprehensively reflect its operational risk. All calculations are based on operational data within a preset historical period. Specifically, the following steps are performed for each device: (1) Calculate the impact intensity of the equipment's operating conditions by statistically analyzing the number of times the equipment switches operating conditions within a preset period and the maximum load change rate of a single switch. Working condition impact strength The formula used to quantify the degree of mechanical and electrical impact experienced by equipment during operation is as follows: ; in, This represents the total number of times the equipment's operating conditions have switched within a preset historical period. The total duration of the preset historical cycle; Let be the load change rate during the i-th operating condition switch; , The preset weighting coefficients, and .

[0042] (2) Extract the historical fault records of the equipment, and calculate the historical fault contribution of the equipment based on the frequency of fault occurrence, the duration of fault downtime and the scope of fault impact; Historical fault contribution The formula used to quantify the impact of equipment failures on the overall safe operation of a power plant is as follows: ; in, This is the total number of equipment failures within a preset historical period; The total duration of the preset historical cycle; Let be the downtime of the j-th fault; This is the influence range coefficient for the j-th fault, with a value ranging from 0 to 1. The larger the influence range, the larger the value. , , The preset weighting coefficients, and .

[0043] (3) Calculate the dynamic change range of each monitoring parameter of the equipment under different operating conditions to obtain the dynamic response characteristics of the parameters; Parameter dynamic response characteristics The formula used to reflect the sensitivity of equipment parameters to changes in operating conditions is as follows: ; in, This represents the maximum value of the average of the k-th monitored parameter under all different operating conditions. It represents the minimum mean value of the k-th monitoring parameter under all different operating conditions.

[0044] (4) The impact intensity of the working condition, the contribution of historical faults and the dynamic response characteristics of parameters are weighted and integrated to obtain the comprehensive risk score of the equipment.

[0045] The comprehensive risk score of the equipment is obtained by weighting and integrating the indicators from the above three dimensions: ; in, Assess the overall risk level of the equipment; , , The preset comprehensive weighting coefficient, and .

[0046] The aforementioned eight weighting coefficients were jointly calibrated using analytic hierarchy process (AHP) and the operational characteristics of pumped storage power stations. Specifically, the weighting coefficient for operational condition impact intensity was determined based on the contribution of operational condition switching frequency and load mutations to equipment lifespan loss; the weighting coefficient for historical fault contribution was normalized based on the actual impact of fault frequency, downtime, and affected area on the safe operation of the power station; and the weighting coefficient for comprehensive risk score was determined based on the actual proportion of operational condition impact intensity, historical fault contribution, and parameter dynamic response characteristics in the risk composition of pumped storage equipment. The weighting determination process included: constructing a judgment matrix based on equipment importance; using AHP to complete initial weighting; verifying and correcting consistency through historical fault data and operational characteristics; and periodically updating the weights based on new data during the system's self-learning cycle to ensure the weight distribution matches the actual operational risks of the units.

[0047] S14, Equipment monitoring priority classification, that is, based on the comprehensive risk scores corresponding to multiple devices, the monitoring priority of the devices is classified, and the equipment monitoring priority classification result is obtained.

[0048] In this embodiment, based on the distribution range of the overall risk score of all equipment in the station, the equipment is divided into three levels: (1) Core and critical equipment: These are the core equipment for the operation of the power plant, and their failures can lead to unit shutdown or even a complete plant shutdown, with a huge impact. Typical equipment includes water pumps and turbines, generator motors, governors, and inlet ball valves / butterfly valves.

[0049] (2) Important auxiliary equipment: The comprehensive risk score is in the middle range. It provides direct support for the core and critical equipment. Failure will affect the normal operation of the core equipment and may lead to load reduction or unplanned shutdown of the unit. Typical equipment includes main transformer, excitation system, oil pressurization device and cooling water system.

[0050] (3) General equipment: The overall risk score is in the lowest range, the operating conditions change slowly, the scope of the fault impact is small, and it will not directly cause the core equipment to shut down. Typical equipment includes gates, drainage systems, and plant power auxiliary equipment.

[0051] Large-scale intelligent monitoring is only enabled for core and critical equipment and important auxiliary equipment; general equipment will continue to use the existing conventional threshold monitoring method.

[0052] S2, threshold adaptation for all working conditions and identification of high-risk working conditions.

[0053] The operating conditions of the pumped storage power station will be divided into steady-state operating conditions and transient transition operating conditions. The same steady-state operating condition will be further divided into multiple sub-operating condition segments according to the upstream and downstream water level range and the ambient temperature range. The transient transition operating condition will be divided into four sub-operating condition segments: start-up transition, shutdown transition, bidirectional switching transition, and load change transition. For each sub-operating condition segment, based on the fixed threshold range set by the monitoring system for each parameter of equipment with a preset monitoring priority level, the deviation of the parameter distribution under the sub-operating condition from the fixed threshold range will be calculated to obtain the fixed threshold fit degree of the sub-operating condition. When the fixed threshold fit degree of any sub-operating condition segment is lower than the preset threshold, it will be judged as a high-risk operating condition, and the risk will be classified in combination with the failure rate and failure severity of the sub-operating condition.

[0054] This step is used to identify high-risk operating conditions that are prone to false alarms or missed alarms due to fixed threshold failure, so as to achieve full coverage without blind spots.

[0055] S21, pumping and accumulating pumps are divided into all operating conditions.

[0056] A comprehensive operating condition classification system specific to pumped storage power stations is established, dividing all operating conditions of the power station into two main categories: steady-state operating conditions and transient operating conditions, which are further subdivided into all the specific operating conditions mentioned above. For each specific operating condition, independent datasets are established for both pumping operation and power generation operation.

[0057] S22, Fixed threshold fit calculation.

[0058] Extract the fixed threshold range set for each parameter of each device with a preset monitoring priority level in the existing monitoring system, and calculate the adaptability of the fixed threshold under each operating condition according to the following steps: (1) Calculation of characteristic distribution trend: For the k-th monitoring parameter of the equipment, extract all historical data of it under the m-th working condition, and calculate the characteristic distribution trend of the parameter under the working condition: ; in, The characteristic distribution of the k-th parameter under the m-th working condition; This represents the total number of historical records where the value of the k-th parameter is greater than the midpoint of its fixed threshold interval under the m-th working condition. This represents the total number of historical records for the k-th parameter under the m-th operating condition.

[0059] (2) Calculation of overall distribution difference: Calculate the average value of the characteristic distribution of the k-th parameter under all working conditions. : ; in This represents the total number of operating conditions.

[0060] (3) Then calculate the overall distribution difference of the k-th parameter under the m-th working condition. : ; (4) Calculation of fixed threshold fit of equipment under working condition: The deviation of the maximum value of the overall distribution difference of all parameters of the equipment under the m-th working condition from 1 is taken as the fixed threshold fit of the equipment under that working condition. : ; Overall distribution differences The smaller the value, the lower the fit of the fixed threshold under this working condition, and the more severe the failure.

[0061] S23, Global environment parameter segmentation correction.

[0062] Considering the significant global impact of upstream and downstream water levels and ambient temperature on the operating parameters of pumped storage power station equipment, environmental segmentation corrections are made for the steady-state operating conditions: The same steady-state condition is divided into multiple continuous two-dimensional sub-condition segments according to the preset range of upstream and downstream water levels and the preset range of ambient temperature.

[0063] Using the same method described above, calculate the fixed threshold fit for each two-dimensional sub-working condition segment.

[0064] If the fixed threshold fit of any two-dimensional sub-condition segment is lower than the preset threshold, the steady-state condition is determined to be a threshold failure condition.

[0065] S24. High-risk operating condition identification and classification. Based on fixed threshold adaptability, historical failure rate, and failure severity, all operating conditions are divided into three levels: (1) Level 1 high-risk operating conditions: Operating conditions with extremely low fixed threshold adaptability, high historical failure rate, and serious failure consequences. Typical operating conditions include pumping water to generate electricity, power generation to pumping water, emergency shutdown, and load shedding.

[0066] (2) Level II high-risk operating conditions: Operating conditions with low fixed threshold adaptability, or moderate historical failure rate and mild failure consequences. Typical operating conditions include low-load pumping steady state, low-load power generation steady state, pumping start-up, and power generation start-up.

[0067] (3) Normal operating conditions: Operating conditions with high fixed threshold adaptability and low historical failure rate. Typical operating conditions include medium-to-high load pumping steady state, medium-to-high load power generation steady state, and standby.

[0068] For Level 1 high-risk operating conditions, high-resolution monitoring of the transition process is activated; for Level 2 high-risk operating conditions, large-scale steady-state monitoring is activated; and for routine operating conditions, conventional threshold monitoring is used.

[0069] S3, Spatiotemporal decoupling bidirectional correlation analysis.

[0070] like Figure 3 As shown, this step is used to quantify the causal relationships between devices, parameters, and between the environment and devices, providing high-quality input features for the model.

[0071] S31. Separate the operation direction data, that is, split the multidimensional dataset into two independent dedicated datasets according to the pumping operation direction and the power generation operation direction.

[0072] All historical data were completely separated according to the pumping operation direction and the power generation operation direction, forming two independent and non-interfering direction-specific datasets. All subsequent correlation analysis and model training were conducted independently within their respective direction-specific datasets, completely avoiding erroneous correlations caused by the reversal of parameter characteristics due to bidirectional operation.

[0073] S32. Global Environmental Impact Quantification: This involves calculating the global impact coefficient of global environmental parameters on equipment monitoring parameters for each specific dataset. This includes calculating the global impact coefficients of upstream and downstream water levels, ambient temperature, and other global environmental parameters on all monitoring parameters of all core and critical equipment. (1) Obtain the synchronous time series data within the preset historical period corresponding to the global environmental parameters and equipment monitoring parameters respectively; for the global environmental parameters E and equipment monitoring parameters X, extract the synchronous time series data E(t) and X(t) within the preset historical period.

[0074] (2) Based on synchronous time series data, a regression model between global environmental parameters and equipment monitoring parameters is established using linear regression; a regression model between environmental parameters and equipment parameters is established using linear regression. ; in For regression coefficients, For the intercept term, This is the residual term.

[0075] (3) Calculate the determination coefficient of the regression model as the global influence coefficient of the global environmental parameters on the equipment monitoring parameters.

[0076] Calculate the coefficient of determination R of the regression model 2 Use it as a global environment parameter For equipment parameters Global impact coefficient: ; Global Influence Coefficient The value ranges from 0 to 1, with a larger value indicating a more significant impact of environmental parameters on equipment parameters. Environmental parameters with a global impact coefficient greater than a preset threshold are included as basic impact factors in subsequent correlation analysis to preemptively eliminate the impact of environmental changes on equipment parameters and reduce the probability of spurious correlations from the source.

[0077] S33. Two-way causal relationship calculation: taking environmental parameters with a global influence coefficient greater than a preset threshold as basic influence factors, and using Granger causality test combined with time series mutual information analysis, the two-way causal relationship matrix between equipment parameters under different operating conditions is calculated.

[0078] The Granger causality test combined with time-series mutual information analysis was used to calculate the bidirectional causal relationship between equipment parameters under different operating conditions, and to quantify the influence intensity and time delay characteristics.

[0079] Definition and Calculation of Time Delay Characteristics: Time delay characteristics refer to the time interval within which a change in one parameter causes a change in another parameter. For parameters... and Calculate different time delays Mutual information below: ; in , For marginal probability density, The joint probability density. The time delay that maximizes mutual information. That is, the time delay of parameter X with respect to parameter Y: ; Calculation of causal influence intensity: at the optimal time delay The Granger causality test is then used to calculate the causal influence of parameter X on parameter Y. : ; in, For prediction using only historical values ​​of Y The residual variance; To predict using historical values ​​of both X and Y The residual variance. Similarly, calculate the causal influence strength of parameter Y on parameter X. This method generates a bidirectional causal correlation matrix that includes parameter pairs, influence direction, causal influence strength, and time delay. Compared to existing methods that only calculate unidirectional correlation coefficients, this embodiment can accurately capture the bidirectional coupling relationships commonly found between pumped storage power station equipment.

[0080] S34. Correlation stability screening, which is based on the bidirectional causal correlation matrix between equipment parameters, screening parameter pairs whose stable influence intensity is greater than a preset threshold, and combining their time delay characteristics to construct a spatiotemporal coupling correlation feature set for different operating directions and working conditions.

[0081] Stability screening is performed on each association in the bidirectional causal association matrix: Historical data is divided into multiple consecutive time windows of equal length; within each time window, the corresponding causal influence strength is recalculated using the method described above; the variance of the causal influence strength across all time windows is then calculated. .when Greater than the preset variance threshold If the association is deemed unstable, it is removed from the list; for stable associations, the average causal influence strength over all time windows is calculated. This represents the strength of the stable influence of the association.

[0082] The correlations with stable influence intensity greater than the preset influence threshold are selected, and combined with their time delay characteristics, a spatiotemporal coupling correlation feature set is formed according to the running direction and the working condition.

[0083] Each record in the spatiotemporal coupling correlation feature set contains the following fields: running direction, operating condition type, equipment to which the influencing parameter belongs, name of the influencing parameter, equipment to which the affected parameter belongs, name of the affected parameter, strength of stable influence, and time delay.

[0084] S4, Full-condition dynamic model cluster monitoring. This step constructs a model cluster adapted to the high-frequency switching characteristics of pumped storage systems to achieve differentiated and accurate monitoring of steady-state and transient processes.

[0085] The full-condition dynamic adaptation model cluster includes: a condition dynamic identification unit, a steady-state adaptive monitoring unit, a transient monitoring unit for transition processes, and a model dynamic scheduling unit. The condition dynamic identification unit identifies the current condition type and stage in real time. The steady-state adaptive monitoring unit outputs the probability of equipment anomalies within a preset time period based on core key equipment parameters, related equipment parameters, and global environmental parameters. The transient monitoring unit outputs the probability of anomalies in the transition process and suspected fault locations based on high-resolution time-series data of the entire transition process. The model dynamic scheduling unit automatically switches the corresponding monitoring model according to the output of the condition dynamic identification unit and preloads the model parameters of the target condition when a condition switching command is received. The output results of the full-condition dynamic adaptation model cluster are integrated with fixed threshold judgment results to generate multi-level alarm signals, and the root cause of the fault is automatically deduced based on a spatiotemporal coupled correlation feature set. The details of each unit are described below: S41. Construction of the Dynamic Operating Condition Recognition Unit: A lightweight temporal neural network is used to construct the dynamic operating condition recognition unit. Input characteristics: active power, reactive power, guide vane opening, circuit breaker status, variable frequency starter operating status, and excitation system status.

[0086] Output results: current operating condition type, current operating condition stage (preparation stage, execution stage, stabilization stage), and operating condition identification confidence level.

[0087] Training and Optimization: The model is trained using historical operating data from the entire station, and an early stopping strategy is employed to prevent overfitting. The model features a lightweight design to ensure inference latency does not exceed a preset threshold, enabling instantaneous recognition during operating condition transitions.

[0088] S42. Construction of steady-state adaptive monitoring unit: Independent steady-state adaptive monitoring models are constructed for the pumping operation direction and the power generation operation direction respectively. Input features: key monitoring parameters of core critical equipment, related equipment impact parameters corresponding to the spatiotemporal coupling correlation feature set, and global environmental parameters.

[0089] Output label: Whether the device will experience an anomaly within a preset time period. If the device experiences an anomaly within the preset time period after the target historical period, the label is 1; otherwise, it is 0.

[0090] Model Training: Historical data is divided into training, validation, and test sets according to a preset ratio, ensuring that all fault samples are included in the test set. A large model architecture suitable for time series data is adopted, and training is performed through pre-training and fine-tuning. During training, the false negative rate is optimized first, followed by the false positive rate.

[0091] Output result: The probability of device malfunction within a preset time period in the future.

[0092] S43. Construction of the transient monitoring unit for the transition process, such as Figure 4 As shown, a transient monitoring unit for the transition process is constructed using a multi-scale temporal convolutional network, specifically designed for processing short-time, high-resolution temporal data of the transition process. Input characteristics: High-resolution time-series data of the entire transition process, with a sampling frequency not lower than a preset threshold.

[0093] Data preprocessing: The transition process is divided into multiple consecutive time segments according to the time axis, and data normalization is performed on each segment.

[0094] Feature extraction: Features at different time scales are extracted for each time segment, and then multi-scale feature fusion is performed to effectively capture transient change features during the transition process.

[0095] Output results: probability of abnormality in the transient process, suspected fault location, and fault severity.

[0096] S44. Model dynamic scheduling mechanism: Establish a model dynamic scheduling mechanism based on working condition identification. It receives the output results of the dynamic identification unit in real time and automatically switches to the monitoring model corresponding to the working condition.

[0097] When the system receives a condition switching command from the central control room, it loads the parameters of the monitoring model corresponding to the target condition in advance, prepares for monitoring, and completely eliminates the monitoring blind spot at the moment of condition switching.

[0098] When the confidence level of the working condition identification is lower than the preset threshold, multiple monitoring models for possible working conditions are activated simultaneously, and the weights of each model are dynamically adjusted according to the changes in the working condition identification results.

[0099] S45, Multi-source anomaly fusion and early warning.

[0100] By integrating the output of a large model cluster with the results of traditional fixed threshold judgments, multi-level early warning information is generated. Establish multi-source anomaly fusion rules, and trigger an alarm through the platform if any judgment path causes an anomaly. Based on the anomaly probability and fault severity, alarms are divided into three levels: Level 1 alarm (urgent), Level 2 alarm (important), and Level 3 alarm (general).

[0101] Based on the spatiotemporal coupling correlation feature set, the fault propagation path and root cause are automatically deduced in descending order of correlation strength, providing targeted handling suggestions for operation and maintenance personnel.

[0102] S5 enables full-process control of operating condition switching.

[0103] (1) Record the entire process of each working condition switching process in high resolution and build a standard template library for working condition switching; When a working condition switching command is received, the high-resolution data acquisition mode is automatically activated to record the data from the issuance of the command to the establishment of a steady state.

[0104] After accumulating sufficient normal operating condition switching data, a standard template library for operating condition switching is constructed by using statistical learning methods, categorized by type and environmental condition.

[0105] (2) Calculate the dynamic deviation between the current switching process and the standard template in real time, and trigger an alarm when the deviation exceeds the preset threshold.

[0106] The system calculates the dynamic deviation between the current switching curve and the corresponding standard template in real time. When the deviation exceeds a preset threshold, an early alarm is triggered to provide early warning of potential problems such as jamming, asynchrony, and excessive impact during the switching process. Key indicators for each operating condition switch are recorded, including total switching time, duration of each stage, maximum vibration value, maximum pressure value, and maximum current value. An operating condition switch quality assessment report is generated for equipment status tracking and maintenance plan development.

[0107] S6, System Self-Learning and Iterative Optimization. This means periodically updating the equipment monitoring priority classification results, high-risk operating condition set, and spatiotemporal coupling correlation feature set based on the latest operational data; and dynamically adjusting model parameters and alarm thresholds according to the equipment performance degradation trend to achieve system self-learning and self-optimization.

[0108] Specifically, a closed-loop self-learning and iterative optimization mechanism should be established to enable the system to continuously adapt to equipment performance degradation and changes in power plant operation modes. The system regularly uses newly added operational data to recalculate the comprehensive risk score of equipment and update the equipment priority classification system; it also regularly recalculates the fixed threshold adaptability of each operating condition and updates the high-risk operating condition set; it regularly re-performs bidirectional causal correlation analysis and updates the spatiotemporal coupling correlation feature set; and it regularly fine-tunes all models using newly added normal and fault data, dynamically adjusting model parameters and alarm thresholds. The self-learning process is conducted offline, without affecting the system's real-time monitoring function.

[0109] This application's embodiments utilize the impact intensity of operating conditions and the contribution of historical faults to achieve precise classification of pumped storage power station equipment, reducing computing power consumption while ensuring monitoring effectiveness; it incorporates transient transition conditions into the scope of intelligent monitoring and performs segmented corrections based on global hydrological and meteorological parameters, achieving comprehensive coverage of high-risk operating conditions of pumped storage power stations; it addresses the shortcomings of existing unidirectional impact analysis through spatiotemporal decoupling bidirectional correlation analysis, and improves the accuracy of correlation analysis and eliminates false correlations by quantifying the temporal delay characteristics between parameters; it designs differentiated model architectures for the different characteristics of steady-state and transient processes, realizing automatic model scheduling and preloading, which can improve the accuracy of fault identification during transient processes.

[0110] like Figure 5 As shown in the figure, this embodiment discloses a full-condition dynamic intelligent monitoring system for implementing the above method. It adopts a layered distributed architecture and connects to the production control area through a one-way safety isolation device. It only reads data and does not issue any control commands, thus ensuring the absolute safety of the production system.

[0111] The overall system architecture is divided into a data layer, an analysis layer, and an application layer. Each layer interacts with the other through standardized interfaces and works together to complete functions such as data access, analysis and calculation, monitoring and early warning, and closed-loop optimization.

[0112] The module components and detailed functions are as follows: The multi-dimensional data fusion and evaluation module is deployed at the data layer and serves as the system's data entry point. It includes the following units: a multi-source data acquisition and cleaning unit, responsible for acquiring heterogeneous data from multiple sources through standardized interfaces and performing preprocessing operations such as outlier removal, missing value completion, timestamp synchronization, and data normalization; a tagging system construction unit, responsible for automatically labeling data with operating condition tags, fault tags, and environmental tags, and providing a manual correction interface; and a multi-dimensional equipment evaluation unit, responsible for calculating the equipment's operating condition impact intensity, historical fault contribution, and parameter dynamic response characteristics according to the above formulas, generating an equipment priority classification system.

[0113] The full-condition threshold adaptation analysis module is deployed in the analysis layer and includes the following units: Full-condition classification unit: responsible for establishing a dedicated full-condition classification system for pumped storage power stations and classifying data into different conditions. Threshold adaptation calculation unit: responsible for calculating the characteristic distribution pattern, overall distribution differences, and fixed threshold adaptation of equipment conditions according to the above formula. Environmental impact correction unit: responsible for dividing steady-state conditions into multiple two-dimensional sub-condition segments and calculating the threshold adaptation of each sub-condition segment. High-risk condition classification unit: responsible for identifying high-risk conditions and classifying their risks.

[0114] The spatiotemporal decoupling correlation analysis module, deployed at the analysis layer, is one of the core innovative modules of this system. It includes the following units: Operation Direction Separation Unit: Responsible for completely separating historical data according to the pumping operation direction and the power generation operation direction. Global Impact Analysis Unit: Responsible for calculating the impact coefficients of global environmental parameters on equipment parameters according to the above formula. Bidirectional Causal Correlation Calculation Unit: Responsible for calculating the strength and time delay characteristics of bidirectional causal influence between equipment parameters according to the above formula. Correlation Stability Screening Unit: Responsible for screening stable and significant correlations and generating a spatiotemporal coupling correlation feature set.

[0115] The full-condition dynamic model cluster module is deployed at the analysis layer and is the core computing module of the system. It includes the following units: Condition Dynamic Identification Unit: Responsible for real-time identification of the current condition type and stage. Steady-State Adaptive Monitoring Unit: Responsible for building and running steady-state monitoring models in different directions, outputting the probability of equipment anomalies. Transient Process Monitoring Unit: Responsible for building and running specialized monitoring models for transient processes, outputting the probability of transient process anomalies and the location of faults. Model Dynamic Scheduling Unit: Responsible for automatically scheduling the corresponding monitoring models based on the condition identification results and implementing model preloading.

[0116] The multi-source anomaly fusion and early warning module is deployed at the application layer and includes the following units: Anomaly Fusion Unit: Responsible for fusing the output results of the large model and the results of traditional threshold judgment. Alarm Grading Unit: Responsible for generating three-level alarms based on anomaly probability and fault severity. Fault Root Cause Inference Unit: Responsible for automatically inferring the fault propagation path and root cause based on spatiotemporal coupled correlation feature sets. Alarm Push Unit: Responsible for pushing alarm information through various methods such as SMS, mobile applications, and in-site broadcasts.

[0117] The full-process control module for operating condition switching, such as Figure 6 As shown, deployed at the application layer, it includes the following units: A handover process recording unit: responsible for recording high-resolution data throughout the entire handover process. A standard template library unit: responsible for building and maintaining a standard template library for handover processes categorized by type and environmental condition. A deviation calculation and early warning unit: responsible for calculating the dynamic deviation between the current handover curve and the standard template in real time, and triggering an early warning when the deviation exceeds the limit. A handover quality assessment unit: responsible for extracting key handover indicators and generating a handover quality assessment report.

[0118] The system's self-learning and optimization module is deployed in the analysis layer and includes the following units: Equipment Priority Update Unit: Responsible for periodically updating the equipment priority classification system. Operating Condition Set Update Unit: Responsible for periodically updating the high-risk operating condition set. Related Feature Update Unit: Responsible for periodically updating the spatiotemporal coupling related feature set. Model Fine-tuning Unit: Responsible for periodically fine-tuning all models using newly added data.

[0119] The system is deployed in the power plant's information management area and connects to the production control area via a one-way network security isolation device. Only production data is allowed to flow unidirectionally into the intelligent monitoring system; any control commands written back to the production control area are strictly prohibited. The system uses an independent server cluster and does not share hardware resources with other business systems, ensuring system stability and security. The system provides a comprehensive access control mechanism, with different user roles having different operating permissions to guarantee data security.

[0120] This embodiment uses the guide vane jamming failure during the power transition process of pumping as an example to explain the working process of the present invention in detail: After receiving the pumping and power relay command from the central control room, the system preloads the parameters of the transient monitoring model for the pumping and power relay transition process and starts the high-resolution data acquisition mode.

[0121] The dynamic operating condition identification unit identifies the current operating condition as "pumping and power generation" in real time, and the model dynamic scheduling unit automatically switches to the transient monitoring unit for the transition process.

[0122] During the transition, the system collects high-resolution time-series data in real time of parameters such as guide vane opening, governor oil pressure, volute pressure, and unit vibration. The transient monitoring unit extracts multi-scale features from the input data during the transition and finds that the rate of change of guide vane opening deviates significantly from the standard curve, while the governor oil pressure exhibits abnormal fluctuations.

[0123] The model output transition process has an abnormal probability greater than the preset threshold. The suspected fault location is the guide vane mechanism, triggering a level 2 alarm.

[0124] Based on the spatiotemporal coupling correlation feature set, the fault root cause inference unit automatically infers that the fault root cause is abnormal increase in governor oil pressure due to guide vane jamming, and pushes the handling suggestion: immediately stop the machine to check the guide vane mechanism and clean foreign objects in the guide vane gap.

[0125] Maintenance personnel promptly shut down the unit for inspection based on the system's early warning. They discovered that foreign objects were causing jamming in the guide vane gap, thus preventing a major accident that could have resulted in guide vane breakage and unit damage.

[0126] After the fault is handled, the system adds the fault data to the fault sample library and fine-tunes the transient monitoring model of the transition process in the next self-learning cycle to further improve the model's accuracy in identifying similar faults.

[0127] Based on the above method embodiments, this application also provides a dynamic intelligent monitoring device for all operating conditions of a pumped storage power station, see [link to relevant documentation]. Figure 7As shown, the device includes: a data acquisition module 702, used to acquire multi-source data of all equipment in the pumped storage power station throughout its entire lifecycle; a multi-dimensional data fusion and evaluation module 704, used to construct a multi-dimensional dataset based on the multi-source data, and to perform multi-dimensional quantitative evaluation of all equipment in the station based on the multi-dimensional dataset to determine the equipment monitoring priority classification results; and a threshold adaptation analysis module 706, used to divide the equipment with monitoring priority reaching a preset level into bidirectional operating conditions, calculate the fixed threshold adaptation degree of each operating condition under different operating directions, identify high-risk operating conditions of threshold failure, and perform risk classification. The correlation analysis module 708 is used to split the bidirectional operating dataset based on the multidimensional dataset, calculate the bidirectional causal relationship between devices, quantify the temporal influence intensity and delay characteristics between parameters under different operating conditions, and construct a spatiotemporal coupling correlation feature set. The model cluster early warning module 710 is used to construct a dynamic adaptation model cluster for all operating conditions based on the spatiotemporal coupling correlation feature set, and to intelligently monitor and warn of high-risk operating conditions where the risk level of pumped storage power station equipment reaches a preset threshold. The intelligent monitoring and early warning includes dynamic identification of operating conditions, steady-state adaptive monitoring, transient monitoring of transition processes, and dynamic model scheduling.

[0128] Furthermore, the aforementioned multidimensional data fusion and evaluation module 704 is used to preprocess multi-source data. Preprocessing includes outlier removal, missing value completion, timestamp synchronization, and data normalization. Based on the preprocessed multi-source data, labels are applied to construct a multidimensional dataset. Labels include operating condition labels, fault labels, and environmental labels. Based on the multidimensional dataset, a multidimensional quantitative evaluation is performed on each device to determine the comprehensive risk score for each device. According to the comprehensive risk scores corresponding to multiple devices, the monitoring priority is graded for each device to obtain the device monitoring priority grading results.

[0129] Furthermore, the aforementioned multi-dimensional data fusion and evaluation module 704 is used to perform the following steps for each device: statistically analyze the number of operating condition switching times and the maximum load change rate of a single switching time within a preset period, and calculate the operating condition impact intensity of the device; extract historical fault records of the device, and calculate the historical fault contribution of the device based on the fault occurrence frequency, fault downtime, and fault impact range; calculate the dynamic change amplitude of each monitoring parameter of the device under different operating conditions, and obtain the dynamic response characteristics of the parameters; and perform weighted fusion of the operating condition impact intensity, historical fault contribution, and parameter dynamic response characteristics to obtain the comprehensive risk score of the device.

[0130] Furthermore, the aforementioned threshold adaptation analysis module 706 is used to divide the operating conditions of the pumped storage power station into steady-state operating conditions and transient transition operating conditions. The same steady-state operating condition is further divided into multiple sub-operating condition segments according to the upstream and downstream water level range and the ambient temperature range. The transient transition operating condition is divided into four sub-operating condition segments: start-up transition, shutdown transition, bidirectional switching transition, and load change transition. For each sub-operating condition segment, based on the fixed threshold range set by the monitoring system for each parameter of equipment with a monitoring priority reaching a preset level, the deviation of the parameter distribution under the sub-operating condition from the fixed threshold range is calculated to obtain the fixed threshold adaptation degree of the sub-operating condition. When the fixed threshold adaptation degree of any sub-operating condition segment is lower than the preset threshold, it is determined to be a high-risk operating condition, and risk classification is performed in combination with the failure rate and failure severity of the sub-operating condition.

[0131] Furthermore, the aforementioned correlation analysis module 708 is used to split the multidimensional dataset into two independent datasets according to the pumping operation direction and the power generation operation direction; for each dataset, the global influence coefficient of global environmental parameters on equipment monitoring parameters is calculated; environmental parameters with a global influence coefficient greater than a preset threshold are used as basic influence factors, and a method combining Granger causality test and time-series mutual information analysis is used to calculate the bidirectional causal correlation matrix between equipment parameters under different operating conditions; based on the bidirectional causal correlation matrix between equipment parameters, parameter pairs with stable influence intensity greater than a preset threshold are selected, and combined with their time-series delay characteristics, a spatiotemporal coupling correlation feature set is constructed for each operating direction and operating condition.

[0132] Furthermore, the aforementioned correlation analysis module 708 is used to acquire synchronous time series data within a preset historical period corresponding to the global environmental parameters and equipment monitoring parameters, respectively; based on the synchronous time series data, a regression model between the global environmental parameters and equipment monitoring parameters is established using a linear regression method; and the determination coefficient of the regression model is calculated as the global influence coefficient of the global environmental parameters on the equipment monitoring parameters.

[0133] Furthermore, the aforementioned model cluster early warning module 710, used for dynamic adaptation of the model cluster under all operating conditions, includes: a dynamic operating condition identification unit, a steady-state adaptive monitoring unit, a transient monitoring unit for the transition process, and a model dynamic scheduling unit. Through the dynamic operating condition identification unit, the current operating condition type and stage are identified in real time. Through the steady-state adaptive monitoring unit, based on core key equipment parameters, related equipment parameters, and global environmental parameters, the probability of equipment anomalies within a preset time period is output. Through the transient monitoring unit, based on high-resolution time-series data of the entire transition process, the probability of anomalies in the transition process and suspected fault locations are output. Through the model dynamic scheduling unit, the corresponding monitoring model is automatically switched according to the output of the dynamic operating condition identification unit, and the model parameters of the target operating condition are preloaded when a operating condition switching command is received. The output results of the dynamic adaptation model cluster under all operating conditions and the fixed threshold judgment results are integrated to generate multi-level alarm signals, and the root cause of the fault is automatically deduced based on the spatiotemporal coupling correlation feature set.

[0134] Furthermore, the aforementioned device also includes: a self-learning module, used to record the entire process of each working condition switching at high resolution and build a standard template library for working condition switching; to calculate the dynamic deviation between the current switching process and the standard template in real time, and to trigger an alarm when the deviation exceeds a preset threshold; to periodically update the equipment monitoring priority classification results, high-risk working condition set and spatiotemporal coupling correlation feature set according to the latest operating data; and to dynamically adjust the model parameters and alarm thresholds according to the equipment performance degradation trend, so as to realize the system's self-learning and self-optimization.

[0135] The device provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts of the device embodiment not mentioned can be referred to the corresponding content in the aforementioned method embodiment.

[0136] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions cause the processor to implement the above-described method. For specific implementation details, please refer to the foregoing method embodiments, which will not be repeated here.

[0137] The computer program products of the methods, apparatus, and electronic devices provided in the embodiments of this application include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementations, please refer to the method embodiments, which will not be repeated here.

[0138] Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0139] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0141] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A dynamic intelligent monitoring method for all operating conditions of a pumped storage power station, characterized in that, The method includes: Acquire multi-source data throughout the entire lifecycle of all equipment in a pumped storage power station; A multidimensional dataset is constructed based on the multi-source data, and a multidimensional quantitative evaluation of all equipment in the station is performed based on the multidimensional dataset to determine the equipment monitoring priority classification results. For equipment whose monitoring priority reaches a preset level, the operating conditions are divided, and the fixed threshold adaptability of each operating condition under different operating directions is calculated. High-risk operating conditions with threshold failure are identified and risk classification is performed. Based on the multidimensional dataset, the bidirectional running dataset is split, and the bidirectional causal relationship between devices is calculated. The temporal influence intensity and delay characteristics between parameters under different operating conditions are quantified, and a spatiotemporal coupling correlation feature set is constructed. Based on the spatiotemporal coupling correlation feature set, a full-condition dynamic adaptation model cluster is constructed to intelligently monitor and warn of high-risk conditions where the risk level of pumped storage power station equipment reaches a preset threshold. The intelligent monitoring and early warning includes dynamic identification of operating conditions, steady-state adaptive monitoring, transient monitoring of transition processes, and dynamic scheduling of models.

2. The method according to claim 1, characterized in that, The steps of constructing a multidimensional dataset based on the multi-source data, and performing a multidimensional quantitative evaluation of all equipment in the station based on the multidimensional dataset to determine the equipment monitoring priority classification results include: The multi-source data is preprocessed; the preprocessing includes: outlier removal, missing value completion, timestamp synchronization, and data normalization; A multidimensional dataset is constructed by labeling preprocessed multi-source data; the labels include: operating condition labels, fault labels, and environmental labels. Based on the multidimensional dataset, a multidimensional quantitative assessment is performed on each device to determine the comprehensive risk score for each device. Based on the comprehensive risk scores corresponding to multiple devices, the monitoring priorities of the devices are classified, and the monitoring priority classification results are obtained.

3. The method according to claim 2, characterized in that, Based on the aforementioned multidimensional dataset, the steps for performing multidimensional quantitative assessments on each device and determining the comprehensive risk score for each device include: For each device, perform the following steps: The number of operating condition switching times of the equipment within a preset period and the maximum load change rate of a single switching are statistically analyzed to calculate the operating condition impact intensity of the equipment. Extract the historical fault records of the equipment, and calculate the historical fault contribution of the equipment based on the fault occurrence frequency, fault downtime, and fault impact range; Calculate the dynamic change range of each monitored parameter of the device under different operating conditions to obtain the dynamic response characteristics of the parameters; The comprehensive risk score of the equipment is obtained by weighting and fusing the impact intensity of the working condition, the contribution of historical faults, and the dynamic response characteristics of the parameters.

4. The method according to claim 1, characterized in that, For equipment whose monitoring priority reaches a preset level, the following steps are taken: dividing the operating conditions, calculating the fixed threshold adaptability of each operating condition under different operating directions, identifying high-risk operating conditions for threshold failure, and classifying the risks. The operating conditions of pumped storage power stations are divided into steady-state operating conditions and transient transition operating conditions. The same steady-state operating condition is further divided into multiple sub-operating condition segments according to the upstream and downstream water level range and the ambient temperature range. The transient transition operating conditions are divided into four sub-operating condition segments: start-up transition, shutdown transition, bidirectional switching transition, and load change transition. For each sub-operating condition, based on the fixed threshold range set by the monitoring system for each parameter of the device whose monitoring priority reaches the preset level, the deviation between the parameter distribution under the sub-operating condition and the fixed threshold range is calculated to obtain the fixed threshold fit of the sub-operating condition. When the fixed threshold adaptability of any sub-operating condition is lower than the preset threshold, the sub-operating condition is determined to be a high-risk condition, and risk classification is performed based on the failure rate and failure severity of the sub-operating condition.

5. The method according to claim 4, characterized in that, The steps of splitting the bidirectional running dataset based on the multidimensional dataset, calculating the bidirectional causal relationship between devices, quantifying the temporal influence strength and delay characteristics between parameters under different operating conditions, and constructing a spatiotemporally coupled correlation feature set include: The multidimensional dataset is split into two independent datasets according to the pumping operation direction and the power generation operation direction; For each dedicated dataset, calculate the global impact coefficient of global environmental parameters on device monitoring parameters; Environmental parameters with a global impact coefficient greater than a preset threshold are used as basic impact factors. The Granger causality test combined with time-series mutual information analysis is used to calculate the bidirectional causal relationship matrix between equipment parameters under different operating conditions. Based on the bidirectional causal correlation matrix among the device parameters, parameter pairs with stable influence strength greater than a preset threshold are selected, and spatiotemporal coupling correlation feature sets are constructed according to their time delay characteristics, based on their operating direction and operating condition.

6. The method according to claim 5, characterized in that, The steps for calculating the global influence coefficient of global environmental parameters on equipment monitoring parameters include: Acquire synchronized time series data within preset historical periods corresponding to global environmental parameters and device monitoring parameters; Based on the synchronous time series data, a regression model between global environmental parameters and equipment monitoring parameters is established using the linear regression method; The determination coefficient of the regression model is calculated and used as the global influence coefficient of the global environmental parameters on the equipment monitoring parameters.

7. The method according to claim 1, characterized in that, The steps for constructing a full-condition dynamic adaptation model cluster based on the aforementioned spatiotemporal coupled correlation feature set, and for intelligent monitoring and early warning of high-risk operating conditions where the risk level of pumped storage power station equipment reaches a preset threshold, include: The full-condition dynamic adaptation model cluster includes: a condition dynamic identification unit, a steady-state adaptive monitoring unit, a transient monitoring unit for transition processes, and a model dynamic scheduling unit; The dynamic working condition identification unit identifies the current working condition type and stage in real time. The steady-state adaptive monitoring unit outputs the probability of equipment anomalies within a preset time period based on core key equipment parameters, related equipment parameters, and global environmental parameters. The transient monitoring unit for the transient process outputs the probability of anomalies and suspected fault locations based on high-resolution time-series data of the entire transient process. The model dynamic scheduling unit automatically switches the corresponding monitoring model according to the output of the working condition dynamic identification unit, and preloads the model parameters of the target working condition when a working condition switching instruction is received. By integrating the output results of the full-condition dynamic adaptation model cluster and the fixed threshold judgment results, multi-level alarm signals are generated, and the root cause of the fault is automatically deduced based on the spatiotemporal coupling correlation feature set.

8. The method according to claim 1, characterized in that, The method further includes: Record the entire process of each working condition switch in high resolution and build a standard template library for working condition switching. The system calculates the dynamic deviation between the current switching process and the standard template in real time, and triggers an alarm when the deviation exceeds a preset threshold. The equipment monitoring priority classification results, the high-risk operating condition set, and the spatiotemporal coupling correlation feature set are updated periodically based on the latest operational data. Based on the trend of equipment performance degradation, the model parameters and alarm thresholds are dynamically adjusted to achieve self-learning and self-optimization of the system.

9. A dynamic intelligent monitoring device for all operating conditions of a pumped storage power station, characterized in that, The device includes: The data acquisition module is used to acquire multi-source data throughout the entire lifecycle of all equipment in a pumped storage power station. The multidimensional data fusion and evaluation module is used to construct a multidimensional dataset based on the multi-source data, and to perform multidimensional quantitative evaluation of all equipment in the station based on the multidimensional dataset to determine the equipment monitoring priority classification results. The threshold adaptation analysis module is used to divide the bidirectional operating conditions for devices whose monitoring priority reaches a preset level, calculate the fixed threshold adaptation degree of each operating condition under different operating directions, identify high-risk operating conditions of threshold failure and classify the risks. The correlation analysis module is used to split the bidirectional running dataset based on the multidimensional dataset, calculate the bidirectional causal correlation between devices, quantify the temporal influence intensity and delay characteristics between parameters under different operating conditions, and construct a spatiotemporal coupling correlation feature set. The model cluster early warning module is used to construct a dynamic adaptation model cluster for all operating conditions based on the spatiotemporal coupling correlation feature set, and to intelligently monitor and warn of high-risk operating conditions where the risk level of pumped storage power station equipment reaches a preset threshold; the intelligent monitoring and early warning includes dynamic identification of operating conditions, steady-state adaptive monitoring, transient monitoring of transition processes, and dynamic model scheduling.

10. A dynamic intelligent monitoring system for all operating conditions of a pumped storage power station, characterized in that, The system is used to perform the method as described in any one of claims 1-8.

Images