Hydropower unit fault diagnosis method and system based on digital twin

Abstract

The present invention relates to the field of unit fault diagnosis technology, specifically including a hydropower unit fault diagnosis method and system based on digital twins, comprising: dividing the startup phase of the target hydropower unit, analyzing the startup transition interference items, establishing a filter channel to evaluate the interference degree, combining real-time data anomaly analysis, and using a sensor algorithm to establish an early warning model; establishing a digital twin model based on the equipment topology, combining the early warning model to predict potential faults, solving the technical problem that it is difficult to distinguish temporary normal fluctuations from hydropower unit failures through empirical thresholds during the startup process of the hydropower unit, and the limitations of fault diagnosis analysis, and realizing effective separation and quantitative evaluation of various interference factors under different working conditions in the startup phase. By constructing a real-time data anomaly analysis module, it accurately captures anomalies in the startup process, thereby improving the sensitivity of fault symptom detection, establishing a startup jam early warning model, and comprehensively improving the technical effect of the accuracy of hydropower unit fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field related to unit fault diagnosis, and in particular to a hydropower unit fault diagnosis method and system based on digital twins. Background Art

[0002] In the field of modern hydropower generation, the reliability of hydropower units is a key factor in the stable operation of power systems. Common fault diagnosis methods for hydropower units often rely on fixed threshold judgments or algorithm models driven by fault experience data, such as fault tree and other related fault analysis and diagnosis mechanism models, to provide support for the overall status diagnosis and fault analysis of the unit. However, during the startup process, especially in the startup phase of hydropower units, relying solely on fixed threshold judgments or historical fault case analysis has limitations, and more refined and intelligent fault identification methods are needed.

[0003] In summary, the prior art has technical problems in that it is difficult to distinguish temporary normal fluctuations from hydropower unit failures through empirical thresholds during the startup process of the hydropower unit, and the fault diagnosis and analysis has limitations. Summary of the Invention

[0004] This application provides a hydropower unit fault diagnosis method and system based on digital twins, aiming to solve the technical problems in the prior art where it is difficult to distinguish temporary normal fluctuations from hydropower unit failures through empirical thresholds during the startup of the hydropower unit, and the fault diagnosis analysis has limitations.

[0005] In view of the above problems, the technical solution to implement this application is:

[0006] On the one hand, the present application provides a hydropower unit fault diagnosis method based on digital twins, wherein the method includes: dividing the target hydropower unit into stages, determining the startup stage of the target hydropower unit and reading real-time data, wherein the real-time data includes current, voltage, and active power adjustment values, and the startup stage includes a hydropower unit standby stage, a hydropower unit kinetic energy climbing stage, and a hydropower unit grid-connected working stage;

[0007] Determine a first interference item set based on interference item analysis of the hydroelectric generator standby phase and the hydroelectric generator kinetic energy climbing phase when starting up;

[0008] Determine a second interference item set based on the interference item analysis during the hydroelectric generator kinetic energy climbing stage and the hydroelectric generator grid-connected working stage;

[0009] establishing an interference item filtering channel based on the first interference item set and the second interference item set, outputting the interference degree of the startup phase and extracting time segments during the startup phase when the interference degree exceeds an interference threshold, wherein the interference threshold includes an absolute threshold and a difference threshold;

[0010] Calculating the mean unit power of the target hydropower unit and performing abnormal data analysis based on the data distribution of the real-time data; establishing a startup jam warning model using a perceptron algorithm based on the mean unit power of the target hydropower unit and the abnormal data analysis results, combined with active power regulation feedback, in time segments where the interference level exceeds the interference threshold during the startup phase;

[0011] Based on the topological structure relationship of the target hydropower unit equipment, a digital twin model is established. Combined with the startup jam warning model, potential faults of the target hydropower unit during the startup process are predicted, and fault diagnosis and alarm are performed based on the potential fault prediction results.

[0012] On the other hand, the present application provides a hydropower unit fault diagnosis system based on digital twins, wherein the system includes: a stage division module for performing stage division based on a target hydropower unit, determining the startup stage of the target hydropower unit and reading real-time data, wherein the real-time data includes current, voltage, and active power adjustment values, and the startup stage includes a hydropower unit standby stage, a hydropower unit kinetic energy climbing stage, and a hydropower unit grid-connected working stage;

[0013] A first interference item analysis module is configured to analyze interference items during startup based on the standby phase and the kinetic energy ramp-up phase of the hydroelectric generator, and determine a first interference item set;

[0014] A second interference item analysis module is configured to analyze the interference items during the startup and transition based on the hydroelectric generator kinetic energy climbing phase and the hydroelectric generator grid-connected working phase, and determine a second interference item set;

[0015] an interference item filtering module, configured to establish an interference item filtering channel based on the first interference item set and the second interference item set, output the interference degree of the startup phase, and extract time segments during the startup phase where the interference degree exceeds an interference threshold, where the interference threshold includes an absolute threshold and a difference threshold;

[0016] an abnormal data analysis module, configured to calculate the unit power mean of the target hydropower unit and perform abnormal data analysis based on the data distribution of the real-time data, and establish a startup jam warning model using a perceptron algorithm based on the unit power mean of the target hydropower unit and the abnormal data analysis results in combination with active power regulation feedback, in time segments where the interference level exceeds the interference threshold during the startup phase;

[0017] The fault diagnosis and alarm module is used to establish a digital twin model based on the topological structure relationship of the target hydropower unit equipment, combine the startup jam warning model, predict the potential faults of the target hydropower unit during the startup process, and perform fault diagnosis and alarm based on the potential fault prediction results.

[0018] In summary, the one or more technical solutions provided in this application realize the effective separation and quantitative evaluation of various interference factors under different operating conditions during the startup phase. By constructing a real-time data anomaly analysis module, it accurately captures anomalies in the startup process, thereby improving the sensitivity of fault sign detection, establishing a startup jam warning model, and comprehensively improving the technical effect of the fault diagnosis accuracy of hydropower units. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart of a hydropower unit fault diagnosis method based on digital twin is provided for this application;

[0020] Figure 2 A flow chart of adjusting the parameters of the startup jam warning model in the hydropower unit fault diagnosis method based on digital twin is provided for this application;

[0021] Figure 3 A structural diagram of a hydropower unit fault diagnosis system based on digital twin is provided for this application.

[0022] Explanation of reference numerals: stage division module M100, first interference item analysis module M200, second interference item analysis module M300, interference item filtering module M400, abnormal data analysis module M500, fault diagnosis and alarm module M600. DETAILED DESCRIPTION

[0023] Example 1

[0024] The present application is described in detail below with reference to the accompanying drawings. Figure 1 As shown, the present application provides a hydropower unit fault diagnosis method based on digital twins, wherein the method includes:

[0025] S1: Divide the target hydropower unit into stages, determine the startup stage of the target hydropower unit and read real-time data, wherein the real-time data includes current, voltage, and active power adjustment values. The startup stage includes a hydropower unit standby stage, a hydropower unit kinetic energy ramp-up stage, and a hydropower unit grid-connected operation stage.

[0026] Relying solely on fixed thresholds or historical failure case analysis has limitations. For example, the stress state of each component changes significantly during startup, causing large fluctuations in real-time monitoring data. Threshold comparisons cannot accurately distinguish between temporary fluctuations during normal startup and true fault signals. The startup phase involves transitions between multiple operating modes, and the types and causes of potential faults in each mode vary.

[0027] By dividing the process into stages, we carefully analyzed the entire startup process, from the standby phase to the kinetic energy ramp-up phase and finally to the grid-connected phase. We paid special attention to the interference analysis during the startup sequence, established an interference filtering channel, and accurately quantified the interference level during the startup phase. Combining unit power mean calculation and real-time data anomaly analysis, we used a sensor algorithm to build a startup jam warning model. This model provides fault diagnosis and warning throughout the startup process, ensuring the safe operation of the target hydropower unit during the startup phase.

[0028] According to the workflow and operating procedures of the target hydropower unit, the operating status of the target hydropower unit is clearly divided into three main startup stages: the hydropower unit standby stage, the hydropower unit kinetic energy climbing stage, and the hydropower unit grid-connected operation stage; synchronous data acquisition equipment for the target hydropower unit is configured, and the synchronous data acquisition equipment includes relevant electrical parameter acquisition equipment such as voltmeters and ammeters;

[0029] It should be noted that in the subdivided startup phases corresponding to the hydropower generator standby phase, the hydropower generator kinetic energy climbing phase, and the hydropower generator grid-connected working phase, the analysis of real-time data and fault diagnosis have different focuses: in the hydropower generator standby phase, detect and confirm whether the basic electrical parameters of the unit are within the normal range, such as checking whether the current and voltage remain constant and fluctuate around the rated value, and monitor the initial state and response speed of the active power regulation device in the standby state to ensure that corresponding adjustments can be made quickly after the start-up command is issued. In addition, in the hydropower generator standby phase, attention should also be paid to non-electrical parameters such as upstream water flow and water head, which will affect the selection of the unit start-up timing and the safety of the initial operation;

[0030] During the stage when the hydroelectric generator's kinetic energy increases, as the turbine blades begin to capture the energy of the water flow, the focus of monitoring shifts to the dynamic trends of current and voltage as the speed increases. This analysis aims to determine whether these trends follow the expected performance curves, as well as the followability and accuracy of the active power regulation system. Specifically, the system should be able to effectively convert mechanical energy into electrical energy and adjust it in a timely manner according to load demand. At the same time, changes in physical quantities such as unit vibration and noise, as well as thermodynamic parameters such as temperature and pressure, should be observed to detect any early signs of failure that may lead to reduced efficiency or component wear.

[0031] During the hydropower generator grid-connected phase, after the unit is successfully connected, real-time data monitoring focuses on the current and voltage quality under grid synchronization conditions to ensure compliance with the harmonic, frequency, and voltage fluctuation limits specified by the grid access regulations. Furthermore, the active power regulation system is monitored for accurate tracking of dispatch instructions and stable output power under actual load conditions. The reactive power regulation capability is also examined to ensure good dynamic response and stability across the entire operating range. Furthermore, the impact of factors such as the water level difference between the upper and lower basins and the coordinated operation of upstream and downstream equipment on the overall efficiency of the unit is comprehensively considered. Simulation analysis is conducted through a digital twin model to proactively prevent system-level failure risks.

[0032] Therefore, when reading real-time data, the emphasis is slightly different in each stage. Specifically, when the target hydropower unit is in the standby stage, real-time data is read to monitor the basic levels of current and voltage and the preparatory state of active power regulation during the standby period; after the target hydropower unit enters the kinetic energy climbing stage, the changes in real-time data are continuously collected and recorded, paying attention to the dynamic changes of current and voltage as the water flow energy increases, and how the active power regulation value responds to the speed increase process of the unit; after the target hydropower unit enters the grid-connected working stage, real-time data is continued to be monitored and recorded, focusing on the stable operating status after grid connection, including whether the current and voltage comply with the grid specifications, whether the active power regulation reaches the predetermined value, and whether there are fluctuations. The startup stage of the hydropower unit is refined into the hydropower standby stage, the hydropower kinetic energy climbing stage, and the hydropower grid-connected working stage, to provide support for subsequent analysis.

[0033] S2: analyzing interference items during startup based on the hydroelectric generator standby phase and the hydroelectric generator kinetic energy ramp-up phase, and determining a first interference item set;

[0034] S3: Analyzing interference items during startup based on the hydroelectric generator kinetic energy ramp-up phase and the hydroelectric generator grid-connected operation phase to determine a second set of interference items;

[0035] S4: establishing an interference item filtering channel based on the first interference item set and the second interference item set, outputting the interference degree of the startup phase and extracting time segments during the startup phase when the interference degree exceeds an interference threshold, wherein the interference threshold includes an absolute threshold and a difference threshold;

[0036] Integrate and analyze the data of each detailed startup phase to identify the transition points between each detailed startup phase, paying particular attention to possible interference items during the startup transition, such as changes during the transition from standby to ramp-up, and from ramp-up to grid connection. Furthermore, determine the start and end time points of the hydropower unit's standby phase, hydropower unit kinetic energy ramp-up phase, and grid connection phase. Through real-time monitoring, continuously collect key parameter data such as current, voltage, and active power regulation values ​​during this process.

[0037] Analyze the transition from the standby phase to the kinetic energy ramp-up phase. Specifically, analyze the real-time data changes from the end of the standby phase to the beginning of the kinetic energy ramp-up phase to identify factors that may affect smooth startup during this period, such as sudden current fluctuations and voltage instability. Based on pre-defined interference factor categories, screen out interference variables with significant impact during this phase and group them into the first interference item set.

[0038] Analyze the transition from the kinetic energy ramp-up phase to the grid-connected phase. Specifically, when the kinetic energy ramp-up phase is about to end and the unit is about to be connected to the grid, analyze the changing trends of the real-time data again to find interference sources that may cause grid connection difficulties or have a negative impact on the stability of the unit. Use the same method as the transition analysis from the standby phase to the kinetic energy ramp-up phase to incorporate the identified key interference factors into the second interference item set.

[0039] An algorithm model for processing interference items is configured, wherein the interference item filtering channel is capable of performing real-time calculation and analysis based on variables included in the first interference item set and the second interference item set; the real-time data is continuously processed, and each interference factor is quantitatively scored using the interference item filtering channel, and a continuous interference degree signal curve is established based on the scoring results;

[0040] Integrate the interference signal during the entire startup phase to obtain the interference level value within each time period; set absolute threshold values ​​(e.g., the maximum allowable instantaneous interference amount) and differential threshold values ​​(e.g., the relative change amplitude of the interference amount at adjacent time points) based on engineering practice experience and historical failure cases; examine the interference level curve to find multiple time segments within the startup phase time limit when the interference level exceeds the absolute threshold and differential threshold in the interference threshold, and record the multiple time segments within the startup phase time limit when the interference level exceeds the absolute threshold and differential threshold in the interference threshold;

[0041] In summary, during the startup phase of the target hydropower unit, the real-time data of each detailed startup phase is dynamically monitored, and an effective method is used to quantify the degree of interference, and then determine when the preset interference threshold is exceeded, thereby accurately extracting the problem periods that may be caused by various interferences during the startup process of the hydropower unit, providing support for targeted fault diagnosis and alarm.

[0042] S5: Calculating the average unit power of the target hydropower unit and performing abnormal data analysis based on the data distribution of the real-time data. During the time segment when the interference level exceeds the interference threshold during the startup phase, based on the average unit power of the target hydropower unit and the abnormal data analysis results, combined with active power regulation feedback, a perceptron algorithm is used to establish a startup jam warning model.

[0043] S6: Based on the topological structure relationship of the target hydropower unit equipment, a digital twin model is established, and combined with the startup jam warning model, potential faults of the target hydropower unit during the startup process are predicted, and fault diagnosis alarms are performed based on the potential fault prediction results.

[0044] Based on the real-time data of the target hydropower unit during the startup phase, the average active power at each time interval (e.g., at the second level) is calculated to obtain the power mean curve per unit time; the overall distribution characteristics of the real-time data during the startup phase are analyzed, including the data trend, peak values, valley values, and fluctuation range; reasonable statistical thresholds are set to detect whether there are abnormal data points that deviate from the normal range based on the distribution of real-time data; and special attention is paid to the occurrence and characteristics of abnormal data during the time period when the identified interference level exceeds the preset threshold;

[0045] Using a perceptron algorithm or other machine learning method suitable for fault warning, the model uses unit power mean data, abnormal data analysis results, and feedback from the active power regulation system as input features. The model is trained using the training data set to identify patterns that may cause startup stuck or delayed. Furthermore, the model parameters are calibrated to ensure that the model can accurately predict the risk of startup stuck when new real-time data arrives.

[0046] Create a virtual digital twin model based on the actual physical structure, electrical system, and control logic of the target hydropower unit. This digital twin model should be able to reflect the interactions and dynamic responses between the unit's internal components and have simulation capabilities. Integrate the startup jam warning model into the digital twin model so that the virtual model can reflect the actual unit status in real time.

[0047] A digital twin model with an integrated startup jam warning model is used to simulate various scenarios during the startup process of the target hydropower unit. Furthermore, during the simulation process, potential fault scenarios that may occur are predicted, such as rotor imbalance, excessive electromagnetic vibration, and increased bearing wear. The simulation results are analyzed to sort and prioritize the causes and degrees of faults that may cause startup jams. When the digital twin model predicts that the probability of a potential fault exceeds a certain threshold, a fault diagnosis alarm mechanism is triggered, and a startup jam warning model is established to comprehensively improve the fault diagnosis accuracy of the hydropower unit during the startup phase.

[0048] Furthermore, the present application method includes:

[0049] Obtain upstream and downstream water level information of the target hydropower unit, and establish a comprehensive diagnostic system based on the topological structure of the target hydropower unit equipment;

[0050] Based on the integrated diagnostic system and the operation data of upstream and downstream equipment, the upstream water level information is compared with the downstream water level information to obtain the availability of the alarm result;

[0051] The alarm result availability is used as identification information and added to the potential fault prediction result.

[0052] Install or connect data acquisition equipment, set up water level monitoring sensors in the upstream and downstream areas of the target hydropower unit, and continuously collect real-time water level information; synchronously obtain water level data from the upstream and downstream basins, and integrate and store them; further, analyze the network topology of the target hydropower unit and its upstream and downstream related equipment, clarify the correlation and impact path between equipment; understand the logical connection and operating rules between the hydropower unit and the upstream reservoir, downstream river, gate and other control facilities;

[0053] Based on the equipment topology relationship, combined with the physical characteristics and operating principles of the hydropower unit, a comprehensive diagnostic system covering various environmental factors such as water level, flow rate, flow, equipment status, etc. is established; the comprehensive diagnostic system should be able to consider the impact of water level changes on unit performance, load distribution and potential failures, and be able to integrate the operating data of equipment at different levels.

[0054] Linked analysis of upstream and downstream water level information with real-time operating data of hydropower units and other related equipment identifies the relationship between water level fluctuations and equipment operating status. A series of judgment conditions and thresholds are set to compare the actual water level with a preset safety range or ideal operating range, and the reliability and effectiveness of the current alarm result, i.e., the alarm result availability, is calculated. Furthermore, methods such as statistics, machine learning, or expert systems are used to study and verify large amounts of historical data to improve the accuracy of alarm availability.

[0055] The availability of alarm results is used as an important reference indicator, quantified and attached to the existing potential fault prediction results; if the availability of alarm results is high, the credibility of the corresponding potential fault prediction is enhanced; if the availability is low, the weight of the prediction result is reduced or other possible causes are further investigated; the final fault prediction report will not only include the fault type and possibility, but also display the availability of alarm results related to water level as an important basis for auxiliary decision-making; when the potential fault prediction results show a high risk, combined with the availability of alarm results, early warning signals are issued in a timely manner, so as to check related risks in advance and improve the stability of the target hydropower unit.

[0056] Furthermore, if Figure 2 As shown, the upstream water level information and downstream water level information of the target hydropower unit are obtained, and the method of the present application further includes:

[0057] Connecting to a data acquisition module to collect watershed monitoring data, wherein the watershed monitoring data includes turbidity, flow rate, and flow velocity of the upstream and downstream watersheds of the target hydropower unit;

[0058] Merging the watershed monitoring data with the upstream water level information and the downstream water level information, and combining them with the real-time data to obtain a comprehensive situation data combination;

[0059] By combining the comprehensive situation data, analyzing the potential failure influencing factors of the changes in the watershed monitoring data on the startup phase of the target hydropower unit;

[0060] Based on the potential fault influencing factors, the startup stuck warning model parameters are adjusted.

[0061] Install appropriate water quality monitors (turbidity meters), flow meters, velocity meters, and other data acquisition equipment at key locations in the upstream and downstream basins of the target hydropower units. Connect these data acquisition modules and start an automated data acquisition process to acquire and record parameter data such as turbidity, flow, and velocity in the upstream and downstream basins in real time.

[0062] Integrate the real-time watershed monitoring data, including turbidity, flow rate, and velocity, with the existing upstream and downstream water level information; perform data alignment to ensure the consistency of data in each dimension in the same time series; after the watershed monitoring data is merged with the upstream and downstream water level information, it is combined with the real-time data to form a comprehensive situation data combination that fully reflects the overall hydrological status of the watershed where the target hydropower unit is located;

[0063] Analyze the temporal trends and interactions of various parameters in the comprehensive situation data combination using statistical analysis methods or machine learning algorithms. Study how specific hydrological conditions (such as abnormally high / low water levels, sudden changes in water flow, and deteriorating water quality) affect the normal operation of hydropower units during the startup phase. Identify potential fault influencing factors that may induce startup stuck or other faults. Identify which changes in hydrological parameters have an impact on the startup stuck warning model. Incorporate potential fault influencing factors into the startup stuck warning model as a basis for adjusting model parameters.

[0064] Based on the correlation between potential fault influencing factors and the probability of fault occurrence, the specific parameters of the startup jam warning model are optimized; through backtesting, cross-validation and other means, it is verified whether the adjusted model parameters can more accurately predict the startup failure risk of the target hydropower unit under different hydrological conditions; the model parameters are continuously iterated until the model accuracy meets the expected performance on the historical data set, thereby enhancing the early warning capability of potential faults during the startup phase of the hydropower unit.

[0065] Furthermore, based on the data distribution of the real-time data, abnormal data analysis is performed. The method of the present application includes:

[0066] Extracting data distribution characteristics of the real-time data, selecting kernel functions and penalty factors, and establishing a single-classification OCSVN framework;

[0067] Based on the single-classification OCSVN framework, the model parameters are optimized through training sample data to determine the abnormal data analysis channel;

[0068] The interference item filtering channel is connected to the abnormal data analysis channel to obtain an abnormal data processing model.

[0069] Calculate statistical data (such as mean, variance, kurtosis, skewness, etc.), frequency distribution histograms, correlation matrices, etc. to obtain the data distribution characteristics of real-time data, which are used to characterize the normal behavior and potential patterns of the data; construct a single-class OCSVN (One-Class Support Vector Network) framework: based on the distribution characteristics of the real-time data, select a suitable kernel function, such as a linear kernel, Gaussian kernel (RBF), polynomial kernel, etc., to map the original data in a high-dimensional space; determine a penalty factor, which is used to balance model complexity and generalization ability. Too large or too small a penalty factor will affect the performance of the model; use the selected kernel function and penalty factor to initialize a single-class support vector machine framework. OCSVN is mainly used to identify the boundaries of normal samples in the data set to detect abnormal data that deviates from the normal range.

[0070] Prepare training sample data, which should contain real-time data points representing normal operating conditions. Use the training sample data to train the model under the OCSVN framework, adjusting model parameters through optimization algorithms (such as the Lagrange multiplier method and the SMO algorithm) to find the optimal hyperplane that maximizes the distance between the data points and the decision boundary. Furthermore, solve the support vector problem, aiming to keep the training data as tightly as possible within the region defined by the hyperplane while keeping unlabeled abnormal data away from this region.

[0071] After model training, based on optimized parameters, the model will be able to distinguish between normal and abnormal data. The OCSVN model now constitutes an abnormal data analysis channel. By predicting new real-time data points, the model will classify them as normal or abnormal. Abnormal data points are those that the model deems to deviate significantly from the normal distribution.

[0072] The interference item filtering channel based on the startup phase that has been constructed is connected to the abnormal data analysis channel. Among them, the interference item filtering channel first filters the real-time data to eliminate or weaken the fluctuations caused by conventional interference factors in the startup phase; the abnormal data analysis channel then processes the filtered data to further identify possible fault signals in the abnormal startup phase; the interference item filtering channel and the abnormal data analysis channel are connected to form an abnormal data processing model, so that the abnormal data processing model can effectively reduce noise interference while accurately detecting abnormal phenomena that may cause hydropower unit startup jams or other faults.

[0073] Furthermore, a comprehensive diagnostic system is established based on the topological structure of the target hydropower unit equipment. The present application method also includes:

[0074] Analyzing the real-time data and extracting key operating parameters based on the startup phase of the target hydropower unit;

[0075] Based on the startup transition period corresponding to the first interference item set and in combination with the key operating parameters, a first time window segment is set, wherein the first time window segment includes a non-masked activation instruction of the abnormal data processing model;

[0076] The first time window in the start-up transition period of the first interference item set is segmented and inserted into the abnormal data analysis channel to capture abnormal data in the start-up transition period corresponding to the first interference item set.

[0077] Furthermore, the present application method also includes:

[0078] Based on the start-up transition period corresponding to the second interference item set and in combination with the key operating parameters, setting a second time window segment, wherein the second time window segment includes a non-masked activation instruction of the abnormal data processing model;

[0079] The second time window in the start-up transition period of the second interference item set is segmented and inserted into the abnormal data analysis channel to capture abnormal data in the start-up transition period corresponding to the second interference item set.

[0080] During the transition from the standby phase to the kinetic energy ramp-up phase, key operating parameters such as rotor speed, torque, generator efficiency, and excitation current are extracted from real-time data during the startup phase of the target hydropower unit, based on the unit's actual operating status and the relationships between devices. In-depth analysis of these key operating parameters is conducted to understand their dynamic changes during different time periods during the startup phase, identifying characteristic values ​​that are closely related to startup performance.

[0081] Determine a startup transition period corresponding to the first set of interference items, i.e., the startup transition period is the period most susceptible to interference from the external environment or internal conditions; use historical data and field measured data to analyze the impact characteristics and specific manifestations of the first set of interference items during the startup transition phase (e.g., from the standby phase to the kinetic energy ramp-up phase); combine theoretical analysis and actual observation results to set specific time window segments for the startup transition period corresponding to the first set of interference items; within the corresponding time window segments, special attention should be paid to abnormal data, and therefore a non-shielded activation instruction is set, which means that the abnormal data processing model is in a state of enhanced sensitivity or priority processing during this time period;

[0082] Insert an abnormal data analysis channel. Specifically, integrate the information of the set first time window segment into the workflow of the abnormal data analysis channel. When the real-time data falls into the time window segment, the abnormal data processing model will automatically activate and increase the response level to ensure that the abnormal data in the startup transition period is captured and analyzed in a timely and accurate manner. Within the first time window segment, the abnormal data processing model will conduct an in-depth scan and judgment of the real-time data to identify data points that are beyond the normal range or do not conform to the expected rules. Furthermore, for the captured suspected abnormal data, a secondary diagnosis will be performed through the abnormal data processing model. In short, in the specific startup phase and critical time window, the detection and diagnosis of abnormal behavior of the target hydropower unit will be strengthened, thereby improving the system's fault prediction accuracy and early warning capabilities.

[0083] Using the same approach, during the transition from the kinetic energy ramp-up phase to the grid-connected operation phase, the startup transition period corresponding to the second interference item set is another key transition phase, namely, from the hydropower generator kinetic energy ramp-up phase to the grid-connected operation phase. During this phase, we determine which interference factors may significantly affect unit performance or cause potential failures, thus forming the second interference item set.

[0084] During the startup transition period corresponding to the second interference item set, a detailed study is conducted on key operating parameters in the real-time data, including but not limited to speed, voltage fluctuations, frequency changes, mechanical vibrations, and other parameters. By analyzing the characteristics and trends of these parameters during this period, key operating parameters closely associated with the second interference item are identified. A second time window segment is defined based on the timing and duration of the possible anomaly caused by the second interference item set. Within the second time window segment, a non-shielded activation instruction of the abnormal data processing model is set, which means that the model will give the highest level of response and processing priority to abnormal data within the second time window segment.

[0085] Insert an abnormal data analysis channel. Specifically, embed the relevant parameter configuration of the second time window segment into the existing abnormal data analysis channel; design system logic so that when real-time data enters the second time window segment, the abnormal data processing model can automatically switch to the non-shielded activation mode; monitor the data flow of the target hydropower unit in the startup transition period corresponding to the second interference item set in real time; when the real-time data deviates from the normal range or shows an atypical pattern within the second time window segment, the abnormal data processing model responds quickly and accurately captures the abnormal data; in short, effectively monitor and warn the potential fault risks of the target hydropower unit in the transition process from the kinetic energy climbing stage to the grid-connected working stage during the startup process, thereby improving the ability of fault diagnosis and prevention.

[0086] Furthermore, a comprehensive diagnostic system is established based on the topological structure of the target hydropower unit equipment. The present application method also includes:

[0087] Based on the grid-side controller and each machine-side rectifier of the target hydropower unit, monitoring the startup interaction signal and obtaining multi-address control link information;

[0088] During the hydroelectric generator standby stage, the hydroelectric generator kinetic energy climbing stage, and the hydroelectric generator grid-connected working stage, the rotor speed range is analyzed to obtain the speed change law and torque characteristic curve during the startup process;

[0089] Based on the speed variation rule and torque characteristic curve during the startup process, fault location is performed in combination with the multi-access control link information.

[0090] Determine the internal equipment composition of the target hydropower unit, its physical connections, and its logical control relationships, and build a detailed equipment topology model, including the grid-side controller, individual generator-side rectifiers, and the interfaces and communication paths between them. Deploy a sensor network and data acquisition system to monitor and record the startup interaction signals between the grid-side controller and individual generator-side rectifiers in real time. The startup interaction signals contain important information reflecting system status and control instructions.

[0091] By parsing and analyzing the startup interaction signals, we can extract multi-access control link information, that is, how different devices work together, transmit commands, and respond to each other, so as to understand the control flow and potential problems in the entire startup process.

[0092] During the different startup phases of the hydropower unit (standby phase, kinetic energy ramp-up phase, and grid-connected phase), rotor speed data is continuously collected while monitoring changes in torque output. A detailed analysis of the speed range in each phase is conducted to identify key indicators such as the speed increase rate, stabilization point, and mutation point. Based on this analysis, a speed change model for the startup process is constructed.

[0093] The torque-speed characteristic curve is simultaneously recorded and analyzed, including the maximum torque point, torque fluctuations, and the presence of abnormal peaks or valleys, to characterize equipment performance and health. The acquired multi-access control link information is combined with speed variation patterns and torque characteristic curves. By comparing the expected behavior under normal operating conditions with the actual observed behavior, a preliminary determination of potential faulty links or components is made.

[0094] Based on the comprehensive diagnostic system, data analysis technologies (such as pattern recognition and machine learning) are used to further refine fault location. For example, if an abnormality is found in the torque curve and a certain control link responds laggingly or incorrectly, it may indicate that a specific component (such as the excitation system or speed regulation device) has a fault or performance degradation. Fault reports and diagnostic suggestions are generated to guide on-site maintenance personnel to conduct targeted inspections and repairs to ensure the safe and efficient operation of the hydropower unit.

[0095] Furthermore, based on the potential fault prediction results, fault diagnosis and warning are performed. The present application method also includes:

[0096] Establish fault cascading computing power constraints by presetting standard information;

[0097] Extract key fault indicators based on the fault maintenance records of the target hydropower units;

[0098] Based on the digital twin model and the startup jam warning model, an in-depth simulation is performed with the key fault indicators as the center to obtain fault chain data, including gearbox fault chain data and transmission chain fault chain data;

[0099] Based on the fault cascading data, using the fault cascading computing power constraints, multi-dimensional fault propagation path tolerance deduction is performed to evaluate the operation fault risk index;

[0100] The potential fault prediction result is marked with a risk level according to the operational fault risk index.

[0101] Referencing GB / T 15468-2020 "Basic Technical Requirements for Hydraulic Turbines" and GB / T 38334-2019 "Technical Specifications for Black Start of Hydropower Stations," we refined technical and safety requirements related to fault diagnosis, defined the conditions and restrictions for triggering fault cascades, and set fault cascade computing power constraints that comply with industry standards. These constraints can be used to limit the scope of fault impact and the likelihood of fault spread, ensuring the accuracy and compliance of subsequent analysis.

[0102] Review the historical fault maintenance records of the target hydropower unit and extract representative key fault indicators, such as fault frequency, severity, associated components, and fault stage. Use the established digital twin model to accurately simulate the actual operating status of the target hydropower unit. Combined with the startup jam warning model, conduct in-depth dynamic simulation of the extracted key fault indicators and analyze the potential chain reactions caused by different fault scenarios.

[0103] Through simulations, we obtain data on the chain of failures of key components, such as gearbox and transmission chain failures, including but not limited to the likelihood of failure, the probability of chain failures, and expected losses. Based on this data, and subject to pre-set constraints on the chain of failure computing power, we use advanced computing methods to conduct a multi-dimensional deduction of the possible propagation paths of the faults, assess the propagation range and consequences of each fault state, and then quantify the operational failure risk index that may be caused by various fault conditions. The operational failure risk index comprehensively reflects risk factors in multiple aspects, such as the probability of failure, potential losses, and operational stability.

[0104] Based on the calculated operational failure risk index, the potential failure prediction results are divided into different risk levels, such as low risk, medium risk, high risk or emergency risk levels; for each predicted potential failure, a corresponding risk level label is assigned to form an intuitive risk warning report, and this information is transmitted to the monitoring system and decision-making platform in a timely manner, triggering the corresponding level of alarm notification, providing a scientific basis for operation and maintenance personnel to take preventive measures or emergency response.

[0105] In summary, the beneficial effects of the embodiments of the present application are:

[0106] 1. Real-time monitoring and precise analysis of various state parameters and environmental variables during the startup phase of hydropower units, effectively filtering out irrelevant interference, collecting real-time data and conducting comprehensive situation data analysis, effectively identifying potential fault influencing factors that cause startup jams, and by building a startup jam warning model and combining multi-source data fusion with machine learning algorithms, achieving synchronous warning of faults occurring during the startup process of hydropower units.

[0107] 2. A highly simulated hydropower unit operating environment was established using the digital twin model. Starting from the equipment topology, the correlation between the components was deeply analyzed, and the interaction between the equipment and the linkage effect of the entire system were comprehensively considered to improve the accuracy and coverage of fault diagnosis.

[0108] 3. By studying the fault modes under different operating conditions during the startup phase of the hydropower unit, combining multi-access control link information and speed and torque characteristic curves, we can refine the processing of interference factors and abnormal data to improve the accuracy of fault diagnosis. Through deep simulation and fault chain computing power constraints, we can quantitatively evaluate and grade the fault risks.

[0109] 4. Based on the startup phase of the target hydropower unit, real-time data is analyzed and key operating parameters are extracted. Based on the startup transition period corresponding to the first interference item set and in combination with key operating parameters, a first time window segment is set. This first time window segment includes a non-masked activation instruction for the abnormal data processing model. The first time window segment within the startup transition period of the first interference item set is inserted into the abnormal data analysis channel to capture abnormal data within the startup transition period corresponding to the first interference item set. During the specific startup phase and critical time window, the detection and diagnosis of abnormal behavior of the target hydropower unit is strengthened, thereby improving the system's fault prediction accuracy and early warning capabilities.

[0110] Example 2

[0111] Based on the same inventive concept as the hydropower unit fault diagnosis method based on digital twin in the above embodiment, Figure 3 As shown, an embodiment of the present application provides a hydropower unit fault diagnosis system based on digital twins, wherein the system includes:

[0112] The stage division module M100 is used to divide the target hydropower unit into stages, determine the startup stage of the target hydropower unit and read real-time data, wherein the real-time data includes current, voltage, and active power adjustment values. The startup stage includes the hydropower unit standby stage, the hydropower unit kinetic energy climbing stage, and the hydropower unit grid-connected operation stage;

[0113] A first interference item analysis module M200 is configured to analyze interference items during startup based on the standby phase and the kinetic energy ramp-up phase of the hydroelectric generator, and determine a first interference item set;

[0114] The second interference item analysis module M300 is configured to analyze the interference items during the startup and transition based on the hydroelectric generator kinetic energy ramp-up phase and the hydroelectric generator grid-connected operation phase, and determine a second interference item set;

[0115] an interference item filtering module M400, configured to establish an interference item filtering channel based on the first interference item set and the second interference item set, output the interference level during the startup phase, and extract time segments during the startup phase where the interference level exceeds an interference threshold, where the interference threshold includes an absolute threshold and a difference threshold;

[0116] The abnormal data analysis module M500 is used to calculate the unit power mean of the target hydropower unit and perform abnormal data analysis based on the data distribution of the real-time data. In the time segment when the interference level exceeds the interference threshold during the startup phase, based on the unit power mean of the target hydropower unit and the abnormal data analysis results, combined with active power regulation feedback, a perceptron algorithm is used to establish a startup jam warning model;

[0117] The fault diagnosis and alarm module M600 is used to establish a digital twin model based on the topological structure relationship of the target hydropower unit equipment, combine the startup jam warning model, predict the potential faults of the target hydropower unit during the startup process, and perform fault diagnosis and alarm based on the potential fault prediction results.

[0118] Furthermore, the digital twin-based hydropower unit fault diagnosis system is also used to perform the following method:

[0119] Obtain upstream and downstream water level information of the target hydropower unit, and establish a comprehensive diagnostic system based on the topological structure of the target hydropower unit equipment;

[0120] Based on the integrated diagnostic system and the operation data of upstream and downstream equipment, the upstream water level information is compared with the downstream water level information to obtain the availability of the alarm result;

[0121] The alarm result availability is used as identification information and added to the potential fault prediction result.

[0122] Furthermore, the digital twin-based hydropower unit fault diagnosis system is also used to perform the following method:

[0123] Connecting to a data acquisition module to collect watershed monitoring data, wherein the watershed monitoring data includes turbidity, flow rate, and flow velocity of the upstream and downstream watersheds of the target hydropower unit;

[0124] Merging the watershed monitoring data with the upstream water level information and the downstream water level information, and combining them with the real-time data to obtain a comprehensive situation data combination;

[0125] By combining the comprehensive situation data, analyzing the potential failure influencing factors of the changes in the watershed monitoring data on the startup phase of the target hydropower unit;

[0126] Based on the potential fault influencing factors, the startup stuck warning model parameters are adjusted.

[0127] Furthermore, the abnormal data analysis module M500 is used to perform the following method:

[0128] Extracting data distribution characteristics of the real-time data, selecting kernel functions and penalty factors, and establishing a single-classification OCSVN framework;

[0129] Based on the single-classification OCSVN framework, the model parameters are optimized through training sample data to determine the abnormal data analysis channel;

[0130] The interference item filtering channel is connected to the abnormal data analysis channel to obtain an abnormal data processing model.

[0131] Furthermore, the digital twin-based hydropower unit fault diagnosis system is also used to perform the following method:

[0132] Analyzing the real-time data and extracting key operating parameters based on the startup phase of the target hydropower unit;

[0133] Based on the startup transition period corresponding to the first interference item set and in combination with the key operating parameters, a first time window segment is set, wherein the first time window segment includes a non-masked activation instruction of the abnormal data processing model;

[0134] The first time window in the start-up transition period of the first interference item set is segmented and inserted into the abnormal data analysis channel to capture abnormal data in the start-up transition period corresponding to the first interference item set.

[0135] Furthermore, the digital twin-based hydropower unit fault diagnosis system is also used to perform the following method:

[0136] Based on the start-up transition period corresponding to the second interference item set and in combination with the key operating parameters, setting a second time window segment, wherein the second time window segment includes a non-masked activation instruction of the abnormal data processing model;

[0137] The second time window in the start-up transition period of the second interference item set is segmented and inserted into the abnormal data analysis channel to capture abnormal data in the start-up transition period corresponding to the second interference item set.

[0138] Furthermore, the digital twin-based hydropower unit fault diagnosis system is also used to perform the following method:

[0139] Based on the grid-side controller and each machine-side rectifier of the target hydropower unit, monitoring the startup interaction signal and obtaining multi-address control link information;

[0140] During the hydroelectric generator standby stage, the hydroelectric generator kinetic energy climbing stage, and the hydroelectric generator grid-connected working stage, the rotor speed range is analyzed to obtain the speed change law and torque characteristic curve during the startup process;

[0141] Based on the speed variation rule and torque characteristic curve during the startup process, fault location is performed in combination with the multi-access control link information.

[0142] Furthermore, the fault diagnosis and alarm module M600 is used to perform the following method:

[0143] Establish fault cascading computing power constraints by presetting standard information;

[0144] Extract key fault indicators based on the fault maintenance records of the target hydropower units;

[0145] Based on the digital twin model and the startup jam warning model, an in-depth simulation is performed with the key fault indicators as the center to obtain fault chain data, including gearbox fault chain data and transmission chain fault chain data;

[0146] Based on the fault cascading data, using the fault cascading computing power constraints, multi-dimensional fault propagation path tolerance deduction is performed to evaluate the operation fault risk index;

[0147] The potential fault prediction result is marked with a risk level according to the operational fault risk index.

[0148] In summary, any step can be stored as a computer instruction or program in an unlimited computer memory and can be called and recognized by an unlimited computer processor, without any unnecessary restrictions.

[0149] Furthermore, the above technical solution only reflects the preferred technical solution of the technical solution of the embodiment of the present application. Some changes that may be made to certain parts thereof by technical personnel in this technical field all reflect the novel principles of the embodiment of the present application. Obviously, technical personnel in this field can make various changes and modifications to the present application without departing from the scope of the present application.

Claims

1. A hydropower unit fault diagnosis method based on digital twins, characterized in that: The method comprises: Based on the target hydropower unit, the startup phase of the target hydropower unit is determined and real-time data is read, wherein the real-time data includes current, voltage, and active power adjustment values. The startup phase includes a hydropower unit standby phase, a hydropower unit kinetic energy ramp-up phase, and a hydropower unit grid-connected operation phase. Determine a first interference item set based on interference item analysis of the hydroelectric generator standby phase and the hydroelectric generator kinetic energy climbing phase when starting up; Determine a second interference item set based on the interference item analysis during the hydroelectric generator kinetic energy climbing stage and the hydroelectric generator grid-connected working stage; establishing an interference item filtering channel based on the first interference item set and the second interference item set, outputting the interference degree of the startup phase and extracting time segments during the startup phase when the interference degree exceeds an interference threshold, wherein the interference threshold includes an absolute threshold and a difference threshold; Calculating the mean unit power of the target hydropower unit and performing abnormal data analysis based on the data distribution of the real-time data; establishing a startup jam warning model using a perceptron algorithm based on the mean unit power of the target hydropower unit and the abnormal data analysis results, combined with active power regulation feedback, in time segments where the interference level exceeds the interference threshold during the startup phase; Based on the topological structure relationship of the target hydropower unit equipment, a digital twin model is established. Combined with the startup jam warning model, potential faults of the target hydropower unit during the startup process are predicted, and fault diagnosis and alarm are performed based on the potential fault prediction results.

2. The hydropower unit fault diagnosis method based on digital twin according to claim 1, characterized in that: The method comprises: Obtain upstream and downstream water level information of the target hydropower unit, and establish a comprehensive diagnostic system based on the topological structure of the target hydropower unit equipment; Based on the integrated diagnostic system and the operation data of upstream and downstream equipment, the upstream water level information is compared with the downstream water level information to obtain the availability of the alarm result; The alarm result availability is used as identification information and added to the potential fault prediction result.

3. The hydropower unit fault diagnosis method based on digital twin according to claim 2 is characterized in that: Acquiring upstream water level information and downstream water level information of a target hydropower unit, the method further includes: Connecting to a data acquisition module to collect watershed monitoring data, wherein the watershed monitoring data includes turbidity, flow rate, and flow velocity of the upstream and downstream watersheds of the target hydropower unit; Merging the watershed monitoring data with the upstream water level information and the downstream water level information, and combining them with the real-time data to obtain a comprehensive situation data combination; By combining the comprehensive situation data, analyzing the potential failure influencing factors of the changes in the watershed monitoring data on the startup phase of the target hydropower unit; Based on the potential fault influencing factors, the startup stuck warning model parameters are adjusted.

4. The hydropower unit fault diagnosis method based on digital twin according to claim 2, characterized in that: Performing abnormal data analysis based on the data distribution of the real-time data, the method comprising: Extracting data distribution characteristics of the real-time data, selecting kernel functions and penalty factors, and establishing a single-classification OCSVN framework; Based on the single-classification OCSVN framework, the model parameters are optimized through training sample data to determine the abnormal data analysis channel; The interference item filtering channel is connected to the abnormal data analysis channel to obtain an abnormal data processing model.

5. The hydropower unit fault diagnosis method based on digital twin according to claim 4 is characterized in that: In combination with the topological structure relationship of the target hydropower unit equipment, a comprehensive diagnostic system is established, and the method further includes: Analyzing the real-time data and extracting key operating parameters based on the startup phase of the target hydropower unit; Based on the startup transition period corresponding to the first interference item set and in combination with the key operating parameters, a first time window segment is set, wherein the first time window segment includes a non-masked activation instruction of the abnormal data processing model; The first time window in the start-up transition period of the first interference item set is segmented and inserted into the abnormal data analysis channel to capture abnormal data in the start-up transition period corresponding to the first interference item set.

6. The hydropower unit fault diagnosis method based on digital twin according to claim 5 is characterized in that: The method further comprises: Based on the start-up transition period corresponding to the second interference item set and in combination with the key operating parameters, setting a second time window segment, wherein the second time window segment includes a non-masked activation instruction of the abnormal data processing model; The second time window in the start-up transition period of the second interference item set is segmented and inserted into the abnormal data analysis channel to capture abnormal data in the start-up transition period corresponding to the second interference item set.

7. The hydropower unit fault diagnosis method based on digital twin according to claim 2, characterized in that: In combination with the topological structure relationship of the target hydropower unit equipment, a comprehensive diagnostic system is established, and the method further includes: Based on the grid-side controller and each machine-side rectifier of the target hydropower unit, monitoring the startup interaction signal and obtaining multi-address control link information; During the hydroelectric generator standby stage, the hydroelectric generator kinetic energy climbing stage, and the hydroelectric generator grid-connected working stage, the rotor speed range is analyzed to obtain the speed change law and torque characteristic curve during the startup process; Based on the speed variation rule and torque characteristic curve during the startup process, fault location is performed in combination with the multi-access control link information.

8. The hydropower unit fault diagnosis method based on digital twin according to claim 7 is characterized in that: Producing a fault diagnosis alarm based on the potential fault prediction result, the method further includes: Establish fault cascading computing power constraints by presetting standard information; Extract key fault indicators based on the fault maintenance records of the target hydropower units; Based on the digital twin model and the startup jam warning model, an in-depth simulation is performed with the key fault indicators as the center to obtain fault chain data, including gearbox fault chain data and transmission chain fault chain data; Based on the fault cascading data, using the fault cascading computing power constraints, multi-dimensional fault propagation path tolerance deduction is performed to evaluate the operation fault risk index; The potential fault prediction result is marked with a risk level according to the operational fault risk index.

9. The hydropower unit fault diagnosis system based on digital twin is characterized by: A system for implementing the digital twin-based hydropower unit fault diagnosis method according to any one of claims 1 to 8, comprising: a stage division module, configured to divide the target hydropower unit into stages, determine the startup stage of the target hydropower unit, and read real-time data, wherein the real-time data includes current, voltage, and active power adjustment values, and the startup stage includes a hydropower unit standby stage, a hydropower unit kinetic energy ramp-up stage, and a hydropower unit grid-connected operation stage; A first interference item analysis module is configured to analyze interference items during startup based on the standby phase and the kinetic energy ramp-up phase of the hydroelectric generator, and determine a first interference item set; A second interference item analysis module is configured to analyze the interference items during the startup and transition based on the hydroelectric generator kinetic energy climbing phase and the hydroelectric generator grid-connected working phase, and determine a second interference item set; an interference item filtering module, configured to establish an interference item filtering channel based on the first interference item set and the second interference item set, output the interference degree of the startup phase, and extract time segments during the startup phase where the interference degree exceeds an interference threshold, where the interference threshold includes an absolute threshold and a difference threshold; an abnormal data analysis module, configured to calculate the unit power mean of the target hydropower unit and perform abnormal data analysis based on the data distribution of the real-time data, and establish a startup jam warning model using a perceptron algorithm based on the unit power mean of the target hydropower unit and the abnormal data analysis results in combination with active power regulation feedback, in time segments where the interference level exceeds the interference threshold during the startup phase; The fault diagnosis and alarm module is used to establish a digital twin model based on the topological structure relationship of the target hydropower unit equipment, combine the startup jam warning model, predict the potential faults of the target hydropower unit during the startup process, and perform fault diagnosis and alarm based on the potential fault prediction results.

Images