Intelligent alarm method for computer monitoring system of hydroelectric power plant

Abstract

The present invention relates to the technical field of hydroelectric power generation, and provides an intelligent alarm method for a computer monitoring system of a hydroelectric power plant.SOLUTION: The historical operation data of the hydroelectric power plant is preprocessed to obtain a first data set, and a data model is constructed based on the first data set. Acquiring first related information according to a physical structure of various equipment of the hydroelectric power plant; The second related information is acquired based on the operation data of various facilities of the hydroelectric power generation plant. Constructing a device association model according to the first association information, the second association information and the data model; The operation data of the target equipment is monitored in real time, and the operation state information of the target equipment is dynamically acquired based on the equipment-related model. Issuing an alarm according to a preset alarm strategy when the driving state information is abnormal; The present invention solves the problem of high false alarm rate in the prior art.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the field of hydropower technology, and in particular to an intelligent alarm method for computer monitoring systems in hydropower plants. [Background technology]

[0002] The monitoring system of a hydropower plant is an important means of ensuring the safe, stable, and economical operation of the plant. Traditional alarm methods are mainly based on threshold comparison, which has problems such as large amount of alarm information, low accuracy, and low discriminability, which increases the burden on personnel and may have a negative impact on the safe operation of the plant.

[0003] The alarm method in traditional hydropower plant monitoring systems mainly relies on threshold comparison method, which is simple and straightforward, but has some major problems in practical application: 1) Large volume of alarm information: The threshold comparison method generates a large volume of alarm information because every parameter change exceeding a preset threshold triggers an alarm, forcing personnel to deal with a lot of unimportant alarm information. 2) Low accuracy: Thresholds are usually static and cannot adapt to the complex changes that occur during actual driving, resulting in many false alarms and oversights. False alarms waste personnel's time and effort, and oversights can lead to overlooking actual problems. 3) Low discriminability: Traditional alarm information often lacks context and detailed explanations, making it difficult for personnel to quickly understand the severity and cause of the alarm. 4) Lack of intelligent analysis: Threshold comparison methods usually do not include intelligent analysis functions and cannot deeply analyze and understand data, so they cannot provide deeper insights. 5) Difficulty in adapting to complex situations: When faced with complex driving environments and changing operating conditions, fixed thresholds cannot accurately reflect the actual state of the system, resulting in insufficient responsiveness of the alarm system. 6) Dependence on human judgment: Due to the volume and complexity of alarm information, personnel must spend a lot of time analyzing and determining which alarms should be dealt with immediately and which can be temporarily ignored. 7) Difficult to update and maintain: As plant equipment and operating conditions change, thresholds need to be updated frequently, but manual threshold adjustments are time-consuming and prone to inaccuracies. 8) Lack of predictive capabilities: Traditional threshold alarm systems typically lack the ability to predict failures and provide alerts only after a problem has occurred, not before a problem occurs. Summary of the Invention

[0004] In order to overcome the drawbacks of the prior art, the present invention aims to provide an intelligent alarm method for a computer monitoring system of a hydroelectric power station, and to solve the problems such as the high false alarm rate in the past.

[0005] The technical means adopted by the present invention to solve the above problems are as follows. The intelligent alarm method for a computer monitoring system of a hydroelectric power plant includes the following steps: S100, preprocessing historical operation data of equipment of the hydroelectric power plant to obtain a first data set. S200, constructing a data model based on the first data set. S300, firstly, obtain relevant information based on the physical structure of various equipment in the hydropower plant. S400, obtaining second related information based on the operation data of various equipment of the hydroelectric power plant. S500: Building an equipment related model based on the first related information, the second related information, and the data model. S600 monitors the operation data of the target equipment in real time, dynamically obtains the operation status information of the target equipment based on the equipment-related model, and issues an alarm based on a pre-set alarm strategy if an abnormality exists in the operation status information.

[0006] As a preferred technical measure, step S200 includes the following steps: S211: extracting feature data from the first data set, where the feature data includes one or more of the following: mean value, variance, peak-to-peak value, and slope; S212: Perform feature selection on the feature data to obtain a second dataset. Based on the second dataset, typical normal samples, and typical fault samples, a sample set is constructed, where the sample set includes a training set and a test set. S213, construct a supervised learning model, train the supervised learning model through a training set, test the supervised learning model that meets the training conditions based on the preset indicators in a test set, dynamically tune the supervised learning model based on the test result, and record the supervised learning model after dynamic tuning as a data model.

[0007] As a preferred technical measure, step S300 includes the following steps: S310, extracting the physical connection relationships between equipment according to the system topology structure in the hydroelectric power plant. S320: Obtaining failure prediction propagation path information based on the physical connection relationship. S330, obtaining the first related information based on the fault prediction propagation path information, the past fault information, and the corresponding countermeasures.

[0008] As a preferred technical measure, step S400 includes the following steps: S410, calculating statistical correlations between different equipment operating parameters; S420: Obtain frequent co-occurrence pattern information of equipment status based on a data analysis algorithm, where the data analysis algorithm includes one or more of the following: Apriori algorithm, FP-growth algorithm. S430, obtaining the second association information based on the frequent co-occurrence pattern information and the expert library information, wherein the second association information includes strong association pattern information with a predetermined value selected from a large number of association rules for hydroelectric power plant equipment.

[0009] As a preferred technical measure, step S420 includes the following steps: S421, determining an input data set for a data analysis algorithm based on statistical correlation; S422, based on the input dataset, obtain a target frequent itemset. S423, obtaining frequent co-occurrence pattern information based on the target frequent item set.

[0010] As a preferred technical measure, step S421 includes the following steps: A100, the operating state parameters of the equipment are selected as a transaction set for related analysis. A200, discretizing the operating state parameters and converting the operating state parameters into Boolean or enumerated values. A300, all equipment states at each time sampling point are treated as one transaction record, and multiple transaction records constitute the input data set for the data analysis algorithm.

[0011] As a preferred technical measure, step S422 includes the following steps: B100: Set a minimum support threshold. Obtain the occurrence frequency of each state based on the input data set. Eliminate itemsets in the input data set whose support is lower than the minimum support threshold to obtain a first frequent itemset, where the first frequent itemset is a frequent set of a single item. B200 performs a self-join operation on the first frequent itemset to generate a first candidate itemset containing two items. The input dataset is scanned to obtain the actual support of the first candidate itemset. The first candidate itemsets whose actual support is lower than a minimum support threshold are filtered out to obtain a second frequent itemset, where the actual support is the actual number of occurrences of each first candidate itemset. B300 performs a self-join operation on the Pth frequent itemset to generate a Pth candidate itemset containing two items. The input dataset is scanned to obtain the actual support of the Pth candidate itemset. Each actual support is the actual number of times each Pth candidate itemset appears. The Pth candidate itemsets whose actual support is lower than the minimum support threshold are filtered out to obtain the P+1th frequent itemset. If the actual support of each Pth candidate itemset is lower than the minimum support threshold, the Pth frequent itemset is recorded as the target frequent itemset, where P≧2.

[0012] As a preferred technical measure, step S423 includes the following steps: C100 enumerates all non-empty subsets of the target frequent itemset. Each non-empty subset, together with its complement, constitutes one candidate association rule. C200, calculate the confidence of each candidate association rule. C300, filtering out candidate association rules whose confidence is less than a minimum confidence threshold, and obtaining frequent co-occurrence pattern information.

[0013] As a preferred technical measure, step S430 includes the following steps: S431: The rules in the frequent co-occurrence pattern information are ranked according to their support and reliability. S432, based on the expert library information, screen and remove superficial rules in the strong association rule set to obtain second association information.

[0014] As a preferred technical measure, in step S600, dynamically obtaining the operating status information of the target equipment based on the equipment-related model includes the following steps: D100, based on the equipment-related model, obtains the connectivity index of operation data. D200,Based on the equipment association model, we obtain the centrality index of each node, where each node corresponds to one equipment. D300 uses a community discovery algorithm to analyze connectivity indexes and centrality indexes to obtain equipment subgroups whose actual closeness exceeds the closeness threshold. D400, analyze the equipment subgroup to obtain operation status information, where the operation status information includes potential fault nodes and corresponding fault impacts and abnormal future expansion paths.

[0015] Compared with the prior art, the present invention has the following advantageous effects. The present invention provides an equipment relationship model based on primary relationship information, secondary relationship information, and a data model, providing a model containing comprehensive relationship information between equipment. This model enables rapid identification of the root cause of failures and improved fault diagnosis efficiency. This equipment relationship model enables real-time analysis of multiple operating data from monitored equipment and in-depth analysis of equipment data and relationships. This enables a comprehensive view of the equipment, discovering potential hidden relationships between equipment and better predicting and preventing potential failures and damage, thereby improving the safety and reliability of hydropower plants. Predictive maintenance and optimized maintenance planning can reduce maintenance costs and unplanned downtime. Triggering warning strategies can effectively implement preventive maintenance, reduce maintenance costs due to equipment failures, and maximize the long-term stable operation of hydropower plant equipment. Intelligent monitoring and warnings make operation management more efficient and accurate, helping to improve the overall operating efficiency and production capacity of hydropower plants. This comprehensive intelligent warning method not only ensures the normal operation of hydropower plants, but also provides solid data support and basis for future operation and maintenance decisions. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a flow diagram of an intelligent alarm method for a computer monitoring system of a hydroelectric power station provided by an embodiment of the present invention. [Figure 2] FIG. 2 is a schematic flow diagram of a method for constructing a data model provided by an embodiment of the present invention. [Figure 3] FIG. 3 is a flow diagram of a method for obtaining first related information provided by an embodiment of the present invention. [Figure 4] FIG. 4 is a flow diagram of a method for obtaining second related information provided by an embodiment of the present invention. [Figure 5] FIG. 5 is a flow diagram of a method for obtaining frequent co-occurrence pattern information provided by an embodiment of the present invention. [Figure 6] FIG. 6 is a schematic flow diagram of a method for determining an input data set provided by an embodiment of the present invention. [Figure 7] FIG. 7 is a flow diagram of a method for obtaining target frequent itemsets provided by an embodiment of the present invention. [Figure 8] FIG. 8 is a flow diagram of a method for obtaining frequent co-occurrence pattern information provided by an embodiment of the present invention. [Figure 9] FIG. 9 is a flow diagram of a method for obtaining second related information based on frequent co-occurrence pattern information and expert library information provided by an embodiment of the present invention. [Figure 10] FIG. 10 is a schematic diagram of the operation status information flow of the target equipment provided by an embodiment of the present invention. [Figure 11] FIG. 11 is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0017] The present invention will be described in more detail below with reference to embodiments and drawings, but the embodiments of the present invention are not limited thereto. [Embodiment 1] As shown in Figures 1 to 11, referring to Figure 1, the first aspect of the present application discloses an intelligent alarm method for a computer monitoring system of a hydroelectric power plant, which specifically includes the following steps: S100: Preprocessing past operation data of various equipment of the hydroelectric power plant to obtain a first data set, the operation data including at least one of a state quantity, a telemetry quantity, and a teleindication quantity. Specifically, preprocessing includes processes such as cleaning and normalization to remove outliers and noise and ensure data quality and usability. These operational data may include other data such as state quantities (temperature, pressure, etc.), telemetry quantities (flow rate, voltage, etc.), teleindication quantities (switch state, etc.), and the like.

[0018] S200, constructing a data model based on the first data set. In this step, the constructed data model is a data model that reflects the operating characteristics of the equipment. Specifically, the constructed data model can be a regression model, a time series model, etc.

[0019] S300 obtains first related information based on the physical structure of various equipment in the hydropower plant, and the first related information includes physical connection relationships between the equipment, a causal chain model, and a fault knowledge base. In this step, the first-level related information helps to understand the dependencies and influence links between equipment, contributing to a more accurate analysis of the root causes of equipment failures. Through the causal chain model, possible fault propagation paths can be quickly identified and identified, improving the efficiency of fault diagnosis.

[0020] S400, obtaining second related information based on the operation data of various equipment of the hydroelectric power plant, the second related information including statistical correlation and frequent co-occurrence pattern information. In this step, the secondary correlation information reveals statistical correlations and co-occurrence patterns between equipment, helping to discover equipment-related relationships that may exist but are not obvious. The statistical correlation and co-occurrence pattern information provides a basis for comprehensive analysis of the complex relationships between equipment, contributing to improving the reliability and stability of the entire system.

[0021] S500, constructing an equipment relation model based on the first relation information, the second relation information, and the data model. In this step, the equipment-related model comprehensively considers physical structures, statistical relationships, and predictive models to provide comprehensive analysis capabilities for the operating status of the entire system. The model helps the system make intelligent decisions and respond in complex environments, improving the system's real-time and adaptability.

[0022] S600 monitors multiple operation data of the target equipment in real time and dynamically obtains operation status information of the target equipment based on the equipment-related model. If an abnormality exists in the driving state information, an alarm is issued based on a preset alarm strategy. In this step, intelligent monitoring and alarm systems can detect changes in equipment status in real time and respond quickly, reducing potential system losses and risks. Timely alarms help implement preventive maintenance measures and reduce the impact of equipment failures on system operation.

[0023] The intelligent alarm method for a computerized monitoring system of a hydropower plant disclosed in this application provides a model containing comprehensive inter-equipment relationship information through an equipment relationship model constructed based on first-related information, second-related information, and a data model. This model helps to quickly identify the root cause of a failure and improve the efficiency of fault diagnosis. This equipment relationship model enables real-time analysis of multiple operating data of the monitored equipment and in-depth analysis of equipment data and relationships, thereby discovering potential hidden relationships between equipment from a comprehensive perspective and better predicting and preventing possible failures and damage. This improves the safety and reliability of the hydropower plant. Predictive maintenance and optimized maintenance planning can reduce maintenance costs and unplanned downtime. Triggering alarm strategies can effectively implement preventive maintenance, reduce maintenance costs due to equipment failures, and maximize the long-term stable operation of hydropower plant equipment. Intelligent monitoring and alarms make operation management more efficient and accurate, contributing to improving the overall operating efficiency and production capacity of the hydropower plant. This comprehensive intelligent alarm method not only ensures the normal operation of the hydropower plant, but also provides solid data support and basis for future operation and maintenance decisions.

[0024] The method disclosed in the present invention is intelligent and automated, easy for users to operate, and can adjust its monitoring and alarm strategies according to different equipment and environmental conditions, boasting strong adaptability and flexibility, improving the accuracy and efficiency of alarms, reducing the burden on personnel, and ensuring the safe and stable operation of power plants. It also contributes to improving the safety management level of hydropower plants, improving operational efficiency and economic benefits, and represents an important development direction for the modernization and intelligentization of hydropower plants.

[0025] In this embodiment, the method for obtaining the preset alarm strategy is as follows: Determine the alarm grade criteria based on the importance of the equipment and the fault result information, and the alarm grade criteria include different grades of accident levels, different grades of fault levels, and different grades of warning levels. Set the trigger conditions, confirmation mechanisms, processing flows, etc. of various alarms for typical operating conditions, and establish a standard alarm strategy library.

[0026] Specifically, the pre-configured alarm strategies may include a target-oriented alarm screening strategy, a grade alarm strategy, a condition alarm strategy, a delay alarm strategy, and a related quantity alarm strategy. By setting different levels of accident, failure and warning levels, alarm priorities can be clearly distinguished. This helps maintenance personnel quickly determine which alarm should be handled first when multiple alarms are received, and helps them allocate resources rationally to ensure the stable operation of important equipment.

[0027] By setting alarm trigger conditions for typical operating conditions, the system can automatically issue alarms under certain circumstances, reducing the burden of human monitoring while increasing the timeliness and accuracy of responses. Maintenance personnel can respond quickly based on pre-defined alarm strategies, effectively shortening fault handling time. Establishing a standard alarm strategy library means that there are clear processing flows and confirmation mechanisms for various alarms, which helps maintenance teams form unified operating practices, increase the standardization of fault handling, and reduce processing delays and errors caused by human factors.

[0028] Pre-configured alarm strategies can realize real-time monitoring and early warning of equipment status, allowing potential problems to be detected and dealt with in a timely manner, avoiding huge losses caused by equipment damage and system shutdowns. This not only reduces maintenance costs, but also reduces production losses caused by equipment failures. A reasonable alarm strategy and a timely handling mechanism can significantly improve the reliability of hydropower plant systems. Pre-configured alarm strategies can promptly alert maintenance personnel when equipment abnormalities occur, preventing equipment failures from having a serious impact on system operation.

[0029] Referring to FIG. 2, the method for establishing a data model specifically includes the following steps: S210, extracting feature data from the first dataset. A feature selection process is performed on the feature data to obtain a second data set. Here, the feature data includes one or more of the mean value, variance, peak-to-peak value, and slope. In this step, the feature data is a feature quantity that is meaningful for determining the equipment state, and is a statistical quantity such as a mean value, a variance, a peak-to-peak value, and a slope. Here, feature selection involves removing redundant and noisy features and reducing the complexity of the model. Common feature selection methods include filtering, wrapping, and embedding. By performing feature selection processing on feature data, the complexity of model training and the influence of noise can be reduced, while at the same time improving the generalization ability of the model. Transform the features as needed to improve their expressiveness, for example, by performing polynomial expansion or logarithmic transformation for nonlinear relationships.

[0030] S220, constructing a sample set based on the second data set, the typical normal samples and the typical fault samples, where the sample set includes a training set and a test set. Here, the training set is used for model learning and the test set is used for performance evaluation.

[0031] S230, training the supervised learning model constructed through the training set, testing the supervised learning model that meets the training conditions based on the preset indicators on the test set, dynamically tuning the supervised learning model based on the test results, and notating the supervised learning model after dynamic tuning as a data model. The supervised learning models that are constructed can be models such as logistic regression, decision trees, support vector machines, and neural networks. Here, the preset metrics may be accuracy, recall, F1-score, etc. Dynamic tuning methods may include model hyperparameter tuning, feature selection strategies, data preprocessing methods, etc. Dynamic tuning allows the model to better adapt to changes in data and fault conditions in real-world scenarios, improving the model's prediction accuracy and stability. Through repeated testing and adjustment, the model's performance and generalization ability can be optimized, ensuring its effective application to new data in the future.

[0032] Additionally, methods such as oversampling and undersampling can be used to balance the dataset and improve the generalization ability of the model. In this step, the model's hyperparameters can be rationally set to control the model's complexity and learning ability, and the optimal parameter combination can be prioritized through methods such as grid search and random search.

[0033] A model is trained on the training set, and the model parameters are iteratively updated until a pre-set stopping criterion is reached, such as convergence of the loss function or the number of training epochs. When evaluating the model performance on the test set, common metrics such as accuracy rate, recall rate, F1 score, and ROC curve can be used to comprehensively examine the classification effect of the model. Specifically, error analysis is performed to identify misclassified samples, analyze the causes, and add representative samples or modify features as necessary. During model tuning, performance can be further improved by adjusting the model structure, introducing regularization terms, and attempting ensemble learning.

[0034] The data model establishment method disclosed in this embodiment can achieve timely prediction and diagnosis of equipment failure through feature extraction and model training, thereby improving equipment reliability and production efficiency. Accurate failure prediction helps enterprises take preventive maintenance measures before equipment failure occurs, reducing maintenance costs and production downtime. The application of data models can optimize production planning and resource allocation, improving production efficiency and product quality.

[0035] Through the above steps, a data model highly related to the equipment's operating status has been established, which can automatically learn fault symptom features from equipment operating data, quickly determine the current situation, and achieve the intelligent monitoring goals of early detection, early warning, and early treatment of equipment faults.

[0036] Compared with traditional alarm methods based on on-site experience, this modeling method fully exploits the value of vast amounts of historical data, extracts more comprehensive and abundant feature indicators, and learns the multi-dimensional operating rules of equipment, significantly improving the coverage and accuracy of alarms, reducing the rate of false alarms and missed alerts, while also reducing the workload of personnel, freeing them from excessive monitoring tasks and allowing them to focus on analyzing and responding to abnormal situations. For example, for a hydroelectric turbine generator unit, more than a dozen monitored variables, such as generator temperature, current, voltage, hydroelectric turbine pressure, flow rate, and vibration, were selected as features to establish a multivariate monitored variable fusion support vector machine classification model. In actual application, this model achieved an alarm prediction time of more than 30 minutes for several typical faults, such as a generator winding inter-turn short circuit and a water-conducting bearing overheat, saving valuable time for the safe control and maintenance of the generator unit.

[0037] Referring to FIG. 3, the method for obtaining first related information specifically includes the following steps. S310,Based on the system topology structure in the hydroelectric power plant, the physical connection relationships between various equipment are extracted. Here, the physical connections between facilities are public buses, connecting switches, and the like. For example, the physical connection method between a hydroelectric turbine and a generator can be identified by checking the design drawings and the on-site layout. The input / output interfaces of each piece of equipment and the sensors and actuators to be connected are determined. In this step, extracting physical connection relationships helps understand the information flow paths between equipment, laying the foundation for subsequent fault prediction and propagation path analysis. Effective physical connection information reduces misunderstandings and erroneous inferences, improving the accuracy and reliability of predictions.

[0038] S320: Obtain predicted failure propagation path information based on the physical connection relationship. Specifically, the system combines physical connection relationships and equipment operation information to identify the possible paths along which a fault may propagate within a hydroelectric power plant. It takes into account data transmission, signal transmission, and energy transfer paths between pieces of equipment. By acquiring propagation path information, it is possible to identify potential fault propagation methods and issue advance warnings to potentially affected equipment and system components. After a fault occurs, the source of the fault can be quickly identified, accelerating the speed of fault rectification and shortening the duration of production downtime. For example, to predict faults in a hydroelectric turbine, the system analyzes the signal transmission paths between the turbine and the hydroelectric turbine control system, turbine governor, and power conditioning system, and identifies possible fault propagation mechanisms (such as abnormal signal transmission or data loss) that exist on each path.

[0039] S330: Obtain first related information based on the failure prediction propagation path information, the past failure information, and the corresponding countermeasures. Specifically, the system compares and analyzes the propagation path information of the predicted failure with past failure data to identify the equipment or system that is most likely to be involved in the failure prediction. It then comprehensively considers the experience and strategies for dealing with the failure to formulate targeted countermeasures. Obtaining the most relevant information enables a rapid response to potential failure events, and preventative or emergency countermeasures can be implemented to reduce the impact of the failure on production. This improves the accuracy and timeliness of failure prediction, optimizes maintenance strategies, and reduces operating and maintenance costs. For example, if a problem is found in an abnormal signal transmission path during fault prediction for a hydroelectric turbine control system, we analyze past data to see how that path behaved in past faults. Based on fault handling records and technical experience, we can make recommendations for optimizing or replacing related sensors and signal processing equipment. The method for obtaining first relevant information disclosed in this embodiment can accurately analyze physical connections and predict fault propagation paths, allowing for timely detection of potential fault risks and the adoption of preventive measures to ensure the stable operation of equipment and systems. Based on historical data and the first relevant information, maintenance strategies and budget allocations can be optimized to ensure the effective use of maintenance resources and cost control. By reducing unplanned downtime and maintenance time, the utilization rate and energy production efficiency of hydropower plants can be improved, enhancing market competitiveness.

[0040] Referring to FIG. 4, the method for obtaining second related information specifically includes the following steps. S410: Calculate statistical correlations between different equipment operating parameters. Here, the statistical correlation can be the Pearson correlation coefficient, mutual information, or the like. Specifically, a correlation coefficient such as the Pearson correlation coefficient or the Spearman correlation coefficient is calculated for each pair of equipment operation parameters (e.g., hydroelectric turbine rotation speed and generator output). Statistical correlation reflects the strength of the linear or nonlinear relationship between equipment operation parameters, and is useful for understanding the mutual influence and degree of interlocking between equipment. For example, in addition to calculating the cross-correlation coefficient using statistical methods for the operating data of hydraulic turbines and generators, the correlation between other equipment parameters such as pressure and temperature is also analyzed to comprehensively understand the equipment operating status. Determining the correlation between equipment operation parameters can help identify possible common failure modes and impact paths, improving the accuracy of failure prediction, optimizing equipment operation control strategies, and ensuring optimal equipment operation, extending equipment life, and improving reliability.

[0041] S420: Obtain frequent co-occurrence pattern information of equipment status based on a data analysis algorithm, where the data analysis algorithm includes one or more of an Apriori algorithm and an FP-growth algorithm. Specifically, the Apriori algorithm or FP-growth algorithm is applied to analyze frequently appearing patterns and rules in equipment status data. These patterns indicate common combinations of equipment operating states and may predict potential common faults or operating modes. For example, the algorithm analyzes the status data of equipment such as hydroelectric turbines, turbine governors, and generators to extract patterns that frequently co-occur between them (such as operating modes under specific operating conditions).The algorithm automatically discovers these patterns and generates specific frequent item sets or association rules. In this embodiment, the frequent co-occurrence patterns provide important clues to equipment status changes and help predict possible abnormal operating conditions or failure conditions of the equipment, thereby optimizing the equipment operation monitoring and warning system, taking timely measures to prevent failures, and improving the stability and availability of the equipment.

[0042] S430, obtaining the second related information based on the frequent co-occurrence pattern information and the expert library information. The second association information includes a plurality of strong association pattern information having a predetermined value selected from the mass association rules of the hydroelectric power plant facilities. Specifically, we combine frequent co-occurrence patterns with expert experience to select strongly associated patterns with a predetermined value. Strongly associated patterns may indicate important combinations of conditions in equipment operation or tendencies toward failure under specific conditions. By analyzing the frequently occurring patterns obtained through the analysis and combining them with expert knowledge to determine their importance and application value in real-world scenarios, relevant patterns that have practical significance for equipment operating status monitoring and fault prediction can be selected. In the technical means disclosed in this embodiment, obtaining the second related information contributes to optimizing fault prediction and maintenance management strategies, improving prediction accuracy and efficiency, and helping the operation team quickly identify potential equipment problems, take timely maintenance measures, and reduce the impact of faults on production. The method for obtaining second-related information disclosed in this embodiment uses statistical correlation analysis and frequent co-occurrence pattern analysis to enhance the understanding and prediction capabilities of equipment operating conditions, contributing to the early detection and intervention of potential problems. Based on the second-related information, more precise maintenance plans and resource allocation strategies can be formulated, improving maintenance efficiency and cost control. The results of statistical analysis and data analysis can be integrated to provide decision-making support for management and optimize operation and maintenance decision-making.

[0043] Referring to FIG. 5, the method for obtaining frequent co-occurrence pattern information specifically includes the following steps. S421: Determine an input data set for a data analysis algorithm based on statistical correlation. Specifically, based on the results of the statistical correlation analysis in the previous stage, equipment operation parameters with significant correlations are selected as input features for the data analysis algorithm. For example, parameters such as the hydroelectric turbine rotation speed, water level, and generator output are selected as features for the input data set. Assume that statistical analysis shows a significant linear correlation between the hydro-turbine rotational speed and the generator output. An input dataset is constructed using the hydro-turbine rotational speed and the generator output as input features for the data analysis algorithm. This step determines the appropriate input data set, improving the efficiency and accuracy of the data analysis algorithm and avoiding the interference of invalid features. The features selected through statistical correlation are likely to reflect key information about the equipment's operating status, which is useful for subsequent analysis of frequent co-occurrence patterns.

[0044] S422: Obtain a target frequent itemset based on the input dataset. Specifically, the Apriori algorithm or FP-growth algorithm is applied to analyze frequent itemsets in the dataset. These itemsets reflect patterns that frequently co-occur in the equipment condition data. Here, a frequent itemset refers to a set of features or attributes that frequently co-occur in the dataset. For a selected input data set (e.g., hydro turbine rotational speed, water level, generator power output), the Apriori algorithm is used to find frequent itemsets for these parameters. The algorithm identifies patterns of frequent occurrence of combinations of these parameters under various operating conditions. In the method disclosed in this embodiment, obtaining target frequent itemsets helps understand common patterns and conditions of equipment operating states, and provides clues for predicting possible equipment state changes and failure trends. These frequent itemsets serve as the basis for subsequent analysis to help identify key patterns and behaviors in equipment operation.

[0045] S423: Obtain frequent co-occurrence pattern information based on the target frequent item set. Specifically, based on the target frequent itemset, the equipment status data is further analyzed to identify frequently co-occurring patterns therein, which may include co-change patterns and event sequences of equipment status under specific conditions. Assuming that frequent item sets under various operating conditions of hydro-turbine rotation speed and water level are obtained through data analysis, these item sets are further analyzed to identify a frequent co-occurrence pattern that the water level decreases when the hydro-turbine rotation speed increases. In this step, obtaining frequent co-occurrence pattern information helps understand the patterns and rules of equipment operation status changes, providing a basis for fault prediction and preventive maintenance, allowing potential equipment problems to be detected in advance and measures to prevent or mitigate faults to be taken. The method for acquiring frequent co-occurrence pattern information disclosed in this embodiment can use the frequent co-occurrence pattern information obtained through statistical correlation analysis and data analysis to more comprehensively understand the changes and correlations in equipment operating conditions, thereby improving the depth and accuracy of the analysis. Based on the frequent co-occurrence pattern information, an intelligent failure prediction model can be built to provide advance warning of potential equipment failures and abnormalities, thereby reducing operational risks. Data-driven equipment condition analysis can optimize equipment operation and maintenance strategies and resource allocation, improving maintenance efficiency and equipment availability and reducing maintenance costs.

[0046] Referring to FIG. 6, the method for determining the input data set specifically includes the following steps: A100: Select operating state parameters of multiple pieces of equipment as a transaction set for related analysis. Specifically, a plurality of equipment operating status parameters, such as switch amounts, warning signals, and limit exceeding flags, are selected and made into a transaction set for related analysis. Specifically, based on the equipment type and importance of the hydropower plant, representative operating state parameters are selected for analysis, including the hydropower turbine rotation speed, water level, and generator output, to ensure that the selected parameters can effectively reflect the equipment's operating state and performance changes. For example, the rotation speed, output, and control valve opening of the hydraulic turbine, generator, and turbine governor are selected as the transaction set for the correlation analysis. These parameters cover the core indicators of equipment operation and are useful for understanding the mutual influence and interrelationship between equipment. In this step, selecting appropriate operating state parameters can improve the effectiveness and accuracy of the relevant analysis, avoid irrelevant or unimportant parameters from interfering with the data analysis results, and provide a comprehensive understanding of the equipment's operating status, which can provide a basis for subsequent data processing and model construction.

[0047] A200, rationally discretizing the operating state parameters and converting them into Boolean or enumerated values. For example, a temperature limit exceeding alarm is converted into two values: "yes" and "no." Specifically, based on the characteristics of the specific operating state parameters, an appropriate discretization method (such as threshold division or classification criteria) is determined to convert the continuous numerical parameters into discrete Boolean values ​​(normal / abnormal) or enumerated values ​​(low / medium / high). For example, the rotational speed of the hydraulic turbine can be set to discrete states such as low speed, medium speed, high speed, etc. The power generator output can be set to discrete states such as low load, medium load, high load, etc. In this step, discretization simplifies the processing complexity of data analysis algorithms, reduces the dimension of parameters, and makes related analysis more effective and efficient. After converting to discrete values, changes in the equipment's operating status can be more intuitively understood, facilitating subsequent pattern analysis and the discovery of related rules.

[0048] A300, all equipment states at each time sampling point constitute one transaction record, and multiple transaction records constitute the input data set of the data analysis algorithm. Specifically, at each time point, all selected equipment status parameters in the hydroelectric power plant are organized into one transaction record according to their discretized values, so that each transaction record contains the status information of each equipment at the same time point and serves as the basic unit for the related analysis. For example, assuming sampling every minute, the equipment status data (hydro turbine rotation speed, discretized state of generator output) for each minute constitutes one transaction record, and transaction records for multiple time points are combined to form a complete input data set, which is used for subsequent association rule analysis. In this step, each transaction record contains the status information of multiple pieces of equipment at the same time point, which helps to discover temporal correlations and co-occurrence patterns between pieces of equipment. Data analysis algorithms can use these transaction records to identify frequently co-occurring patterns, thereby helping to predict possible interlocking failures or abnormal conditions of the equipment. The input dataset determination method disclosed in this embodiment can comprehensively understand the operating status and changes of each piece of equipment in a hydropower plant through the selection of appropriate operating state parameters and effective discretization processing. The construction of a dataset based on transaction records helps data analysis algorithms accurately analyze frequent co-occurrence patterns between equipment states and improve prediction accuracy. The integration of discretization processing and the transaction record construction method can provide hydropower plants with optimized decision-making support for operation and maintenance management and fault prediction.

[0049] Referring to FIG. 7, the method for obtaining target frequent itemsets specifically includes the following steps: B100, setting a minimum support threshold. Based on the input data set, the frequency of occurrence of each state is obtained. Remove itemsets in the input dataset whose support is less than a minimum support threshold to obtain a first frequent itemset, which is a frequent set of a single item. Specifically, a minimum support threshold (e.g., 10%) is set. The input dataset is scanned, the frequency of occurrence of each state is counted, and a 1-frequent itemset (i.e., the first frequent itemset) is generated. The minimum support is a pre-set threshold that defines the minimum occurrence frequency or probability of a frequent itemset. Only itemsets that meet the minimum support are considered frequent; otherwise, they are excluded from the analysis results. Assume that the minimum support threshold is set to 0.1 (i.e., itemsets appear at least 10% frequently in the dataset). This means that only itemsets with an appearance frequency of more than 10% will be considered frequent and used in subsequent frequent itemset analysis. Setting a minimum support threshold can filter out itemsets that appear relatively infrequently in the dataset, improve the accuracy and validity of the analysis results, and focus on analyzing frequent itemsets that truly have certain importance and trends, reducing unnecessary calculations and analysis work.

[0050] B200 performs a self-join operation on the first frequent itemset to generate multiple first candidate itemsets containing two items. Scan the input dataset and obtain the actual support of the first candidate itemsets, which is the actual number of occurrences of each first candidate itemset. The first candidate itemsets whose actual support is less than the minimum support threshold are filtered out to obtain a second frequent itemset, and B300 is cyclically performed based on the second frequent itemset. Specifically, a self-join operation is performed on the 1-frequent itemsets to generate a 2-candidate itemset (i.e., the first candidate itemset), the input dataset is scanned, the support of the 2-candidate itemsets is calculated, and the 2-frequent itemset (i.e., the second frequent itemset) is filtered. In this step, self-join refers to the process of generating candidate itemsets from known frequent itemsets, and these candidate itemsets contain more items but still maintain their frequency. In the case of the Apriori algorithm, the self-join step generates a new candidate itemset by combining two frequent itemsets. For example, suppose the first frequent itemset contains a single item such as {hydro turbine rotation speed}, {generator output}, etc. Through the self-join operation, a candidate itemset containing two items such as {hydro turbine rotation speed, generator output} can be generated. In this step, the self-join operation expands the combinatorial possibilities of frequent itemsets and helps discover more complex frequent itemset patterns. In the continuous iterative process, new candidate itemsets that satisfy a preset minimum support can be generated, laying the foundation for frequent itemset analysis in the next step.

[0051] B300 performs a self-join operation on the Pth frequent itemsets to generate a plurality of Pth candidate itemsets containing two items, where P≧2. The input dataset is scanned to obtain multiple actual supports for the P-th candidate itemsets, where each actual support is the actual number of occurrences of each class of P-th candidate itemsets. The Pth candidate itemsets whose actual support is less than the minimum support threshold are filtered out to obtain the P+1th frequent itemsets. If the actual support of all of the Pth candidate itemsets is less than the minimum support threshold, the Pth frequent itemset is recorded as the target frequent itemset. Specifically, step B300 is repeated to generate k-candidate itemsets and k-frequent itemsets for each layer, and this continues until no new frequent itemsets are generated. Repeating step B300 means that if P is 2, a self-join operation is performed on the second frequent itemset to generate multiple second candidate itemsets containing two items. The input dataset is scanned to obtain multiple actual supports for the second candidate itemsets. Each actual support is the actual number of times each type of second candidate itemset appears. If the actual support of the second candidate item set is less than the minimum support threshold, the second frequent itemset is recorded as the candidate frequent itemset, i.e., the process stops when no new frequent itemsets are generated. For a given Pth frequent itemset, a new candidate itemset containing P+1 items is generated through a self-join operation. This step is a key step in the Apriori algorithm, where a higher-order candidate itemset is generated by combining frequent itemsets. For example, suppose the second frequent itemset contains itemsets such as {hydro turbine rotation speed, generator output}, {hydro turbine rotation speed, turbine governor status}, etc. Through self-joining, a new candidate itemset can be generated that contains the three items {hydro turbine rotation speed, generator output, turbine governor status}. In this step, the recursive nature of the self-join operation gradually expands the complexity of frequent itemsets, allowing for the discovery of longer frequent itemset patterns. At each step, newly generated candidate itemsets are guaranteed to meet the preset minimum support requirement, ensuring the accuracy and practicality of the analysis results. The method for obtaining target frequent itemsets disclosed in this embodiment can efficiently find and select frequent itemsets that meet the conditions through stepwise self-join operations. By setting a minimum support threshold and self-join operations, a certain degree of accuracy and interpretability of the analysis results can be ensured. The implementation of the entire technical solution can provide a basis for decision-making support and optimization of hydropower plant equipment status, thereby improving the efficiency and reliability of equipment operation.

[0052] Referring to FIG. 8, the method for obtaining frequent co-occurrence pattern information specifically includes the following steps: C100 enumerates all non-empty subsets of the target frequent itemset. Each non-empty subset, together with its complement, constitutes one candidate association rule. Specifically, for each frequent itemset, we enumerate all of its non-empty subsets, where a non-empty subset is a subset that selects at least one item from the frequent itemset. For example, if the target frequent itemset is {hydro turbine rotation speed, generator output}, the non-empty subsets that can be generated are {hydro turbine rotation speed}, {generator output}, {hydro turbine rotation speed, generator output}. In this step, multiple candidate association rules can be generated by enumerating non-empty subsets, where each rule represents a type of combinatorial relationship within the frequent itemset. The generation of these rules forms the basis for subsequent confidence calculations, which helps discover associations and patterns between different items.

[0053] C200, calculate the confidence of each candidate association rule. In this embodiment, the confidence is the quotient of the support of the candidate association rule divided by the support of the antecedent subset, and reflects the degree of reliability of the rule. Specifically, confidence refers to the conditional probability of the rule "If X appears, Y will appear," and is calculated as follows: confidence(X>Y) = support(X∪Y) / support(X), where support(X∪Y) represents the frequency with which X and Y appear together, and support(X) represents the frequency with which X appears alone. For example, for the rule {hydro turbine rotation speed}->{generator output}, its reliability is calculated as: Support({hydro turbine rotation speed, generator output}) / Support({hydro turbine rotation speed}). In this step, confidence assessment helps identify and quantify the degree of association between rules. Rules with high confidence usually have stronger associations, and filtering rules with confidence below a threshold can reduce unnecessary information and improve the effectiveness of association rule analysis.

[0054] C300, filtering out candidate association rules whose confidence is less than a minimum confidence threshold, and obtaining frequent co-occurrence pattern information. Here, the minimum confidence threshold can be set to 80%, and rules whose confidence does not meet the requirement can be filtered out to obtain frequent co-occurrence pattern information. In this embodiment, the frequent co-occurrence pattern information is strong association rule set information. Specifically, all calculated candidate association rules are compared with a set minimum reliability threshold, and rules whose reliability is equal to or greater than the threshold are retained as frequent co-occurrence pattern information. For example, if the minimum confidence threshold is set to 0.6, only association rules with a confidence level of 0.6 or higher are retained. The rule selected is {hydro turbine rotation speed} > {generator output}, which indicates a strong association between these two. In this step, selecting highly reliable association rules helps to understand the common association patterns between equipment status parameters, and these frequent co-occurrence pattern information can be used as the basis for equipment operation status analysis, fault prediction, and optimization decision-making. The method for obtaining frequent co-occurrence pattern information disclosed in this embodiment can deeply analyze and understand the relationships and patterns between equipment status parameters through association rule analysis. The discovered frequent co-occurrence pattern information can help predict possible interlocking failures or abnormal conditions of equipment and perform preventive maintenance in advance. This provides scientific basis and data support for operation and maintenance management and equipment optimization, and improves the operating efficiency and reliability of hydropower plant equipment. Specifically, for each frequent itemset, we enumerate all its non-empty subsets. Each subset and its complement constitutes a candidate association rule. For example, if a frequent itemset contains {A,B,C}, its non-empty subsets include {A}, {B}, {C}, {A,B}, {A,C}, {B,C}. For each candidate rule, we calculate its confidence, which is defined as the rule's support divided by the support of its antecedent subset. For example, the confidence of the rule {A,B}->{C} is calculated as Support({A,B,C}) / Support({A,B}). Set a minimum confidence threshold, e.g., 80%. Filter out rules with confidence below the threshold and retain only those with confidence above the threshold. These retained rules constitute a strong association rule set, and these rules are considered to have relatively reliable association relationships. By computing all possible association rules for a frequent itemset and filtering out rules that do not meet a set confidence threshold, we find strong association rules in a dataset. These association rules can help us understand related phenomena in the data and can be used in applications such as building recommendation systems and market analysis.

[0055] Referring to FIG. 9, the method for obtaining second related information based on frequent co-occurrence pattern information and expert library information specifically includes the following steps. S431: The rules in the frequent co-occurrence pattern information are sorted based on their support and reliability. It is also possible to use other evaluation indices such as lift or chi-square to comprehensively consider the importance of the rules. Here, Support measures how often a rule appears in a dataset, i.e., it is the probability that the union of the itemsets on the left and right sides of the rule appears in the dataset. Confidence evaluates the reliability of a rule, i.e., it is the probability that the conclusion of a rule will also appear when the condition part of the rule appears. For example, the support and reliability of the rules {hydroelectric turbine rotation speed} -> {generator output} and {water level} -> {turbine governor status} in the frequent co-occurrence pattern are calculated. The rules are sorted based on their support and reliability, and rules with high support and reliability are given priority. In this step, sorting helps experts and decision makers quickly understand and identify the most influential and relevant rules within frequent co-occurrence patterns, and prioritizes the processing of rules with high support and confidence, thereby improving the efficiency of analysis and decision-making.

[0056] S432, based on the expert library information, filter out superficial rules from the strong association rule set to obtain second association information. The purpose of this method is to retain valuable novel rules by removing superficial rules from the set of strong association rules. That is, by utilizing the experience of experts in the expert library information, valuable strong association patterns are selected from a large number of association rules, and potential relationships between facilities are revealed. Here, "superficial rules" refer to rules that appear to have relevance or high confidence on the surface, but in reality have no real practical meaning or causal relationship. Expert library information provides understanding of the actual relationships behind the rules and assists in identifying and eliminating these types of superficial rules. For example, suppose there is a high-confidence rule {water level}>{turbine governor status} in the frequent co-occurrence pattern, but if experts determine that this is merely a coincidence and has no real causal relationship, they can filter out this kind of superficial rule based on their knowledge to obtain more accurate related information. In this step, superficial rules are filtered out to improve the accuracy and practicality of the association rule analysis and ensure the scientific nature and explainability of the analysis results, thereby avoiding misunderstandings and unnecessary judgment bias caused by the interference of superficial rules. The method of obtaining second related information based on frequent co-occurrence pattern information and expert library information disclosed in this embodiment enables deeper and more accurate related information to be obtained through sorting and expert selection, which helps to comprehensively understand the relationship between equipment status and performance. This provides more accurate and reliable decision support for maintenance management and equipment optimization, contributing to improving the operational efficiency and reliability of equipment. The integrated use of data analysis technology and domain expert knowledge enables higher-level equipment status analysis and failure prediction. Through the implementation and integration of the above steps, a complete relevant regulation analysis and analysis process can be established, providing scientific basis and technical support for decision-making on equipment management and maintenance management of hydropower plants.

[0057] Below, a specific analysis process when the data analysis algorithm is the Apriori algorithm will be explained using an actual application example. A year's worth of historical data from six generator units at a hydroelectric power plant includes more than 300 telemetry and teleindication data, with a total data volume exceeding 10 million items. We select 20 representative parameters, including generator temperature, pressure, flow rate, and electrical quantity, as the analysis targets, and apply the Apriori algorithm to deeply analyze the associated rules hidden within them. Through frequent itemset analysis, typical combination patterns of several equipment conditions are discovered. For example, the combination {generator stator winding over-temperature, generator U-phase current over-current, and excitation current insufficiency} appears repeatedly in the past data, forming a frequent alarm pattern (i.e., information included in the frequent co-occurrence pattern information). Further association rules are generated to obtain some highly reliable decision rules, such as: Insufficient excitation current and excessive generator U-phase current → excessive generator stator winding temperature (support rate 12%, reliability 85%). This rule shows that when underexcitation and excessive single-phase current of the generator occur simultaneously, there is an 85% probability that the generator stator winding will become overheated. Analyzing the physical mechanism, it can be seen that the low excitation current weakens the magnetic field, worsening the stator current imbalance, and causing the local winding heat generation to cause the temperature rise. The equipment expert further explained that this type of frequent alarm pattern reflects the excitation imbalance problem of the generator unit. When the combined excitation abnormality and current imbalance alarm appears, it is a sign of a stator winding problem, and the relevant responsible person needs to be notified in advance to carry out an inspection. This provides an important basis for subsequent improvement of the related model and rational allocation of alarm items.

[0058] Similarly, a group of predictable related rules were discovered, such as {Trip of No. 2 main converter low-voltage side disconnector → Stop of No. 2 unit} and {Low cooling water pressure of No. 1 unit and increased vibration of No. 1 unit → High bearing temperature of No. 1 unit}, which greatly expanded the analytical scope of the personnel in charge and deepened their understanding of the inherent relationships in the equipment. The Apriori algorithm is known to be excellent at automatically discovering frequent co-occurring patterns from large-scale discrete events and describing the relationships and causal relationships between things using intuitive logical formulas. This type of data-driven association analysis complements the blind spots of artificial experience, reveals objectively existing laws that are often overlooked, and is a valuable complement to other analytical methods. Of course, frequent co-occurrence does not necessarily mean a definite causal relationship. The association rules discovered by Apriori require evaluation and interpretation based on physical mechanisms to determine the rationality and value of the association and to combine it with actual operations to create an operational equipment association model that can guide monitoring and warning work. In this embodiment, the constructed equipment-related model can be a graph model, for example, equipment topology and operation-related information, i.e., first related information, second related information, and data model, are expressed in a graph format such as a directed graph, an undirected graph, or a weighted graph. Specifically, the graph is structured such that each node represents a facility object such as a generator or a transformer. When two facilities are physically or logically related, an edge is connected between the corresponding nodes. The relationship may be based on multiple dimensions, such as facility topology connections, signal configurations, and operational data correlations. Each edge can then be assigned a weight value based on factors such as association strength, association type, etc., and the weights can be static, dynamic, or adjustable.

[0059] Referring to FIG. 10, the method for obtaining the operating status information of the target equipment specifically includes the following steps: D100, obtains connectivity indexes for multiple operation data based on equipment-related models. Through this step, the aim is to evaluate the relational complexity of the system. The connectivity index reflects the tightness of the connections between facilities, while the fragmented connections explain the looseness of the system. Specifically, graph theory techniques (shortest path, network flow, etc.) can be used to calculate the connection strength or path density between facilities, where the connectivity index can reflect the effectiveness and speed of information transmission between facilities and the cooperation that may exist between facilities. For example, in a facility-related model, assuming that there is a direct connection between facilities A, B, and C, the length of the shortest path from facility A to facility B or the number of paths can be calculated and used as part of the connectivity index. In this step, connectivity indicators can help identify close connections between equipment, contribute to understanding information transmission and interdependencies during equipment operation, and provide a data basis for subsequent equipment status analysis, which is particularly important in identifying equipment subgroups and failure potential nodes.

[0060] D200,Based on the equipment association model, we obtain the centrality index of each node, where each node corresponds to one equipment. Here, the centrality index includes degree centrality, betweenness centrality, closeness centrality, etc. Nodes with high centrality are more important in association analysis. Specifically, centrality indices (betweenness centrality, degree centrality, etc.) are used to measure the central position and influence of equipment nodes in the network. These indices can help identify key devices, i.e., nodes that occupy important positions in equipment-related networks. For example, in a facility association model, facility A may be assumed to have high degree centrality because it has direct connections to multiple other facilities, and its importance within the network can be quantified by calculating the degree or other centrality measure of facility A. In this step, centrality indexes contribute to identifying and focusing on important nodes in the equipment network, which have a significant impact on the stability and operating status of the entire network, and can provide a basis for subsequent equipment subgroup discovery and operating status analysis, which is particularly useful in identifying potential failure factor nodes.

[0061] D300 uses a community discovery algorithm to analyze connectivity and centrality indices to obtain equipment subgroups whose actual closeness exceeds the closeness threshold. Here, these equipment subgroups often have common physical or logical risk factors. Here, the community discovery algorithm can identify dense and interconnected subgroups in the network, which can be understood as sets of equipment with high interaction and dependency in the equipment network. For example, suppose a community discovery algorithm identifies a subgroup consisting of facilities A, B, and C. This subgroup has high internal connectivity and centrality, which may suggest that they exhibit high levels of cooperation in operation or are jointly subject to similar external influences. In this step, community discovery can help identify functional modules or integrated units in the equipment network, contribute to understanding the operating conditions and interaction patterns exhibited by the equipment at the group level, and provide more precise and detailed information support for subsequent equipment status analysis and identification of potential fault nodes.

[0062] D400 analyzes the equipment subgroup and obtains the operating status information, which includes potential fault nodes, corresponding fault impacts, and future expansion paths of the anomaly. Further fault diagnosis and condition analysis are performed on the determined equipment subgroup to identify potential fault nodes that may exist. Specifically, we analyze the interactions and influences between equipment subgroups and predict possible abnormal expansion paths and impact ranges. For example, if equipment A is identified as a potential failure node in an identified equipment subgroup, it may have a significant impact on the operating status of the entire subgroup. The failure impact of equipment A and the possible abnormality propagation path can be further analyzed to provide detailed information for maintenance and operation decision-making. This step helps to detect possible potential failures in advance through detailed analysis of equipment subgroups, and take preventive maintenance measures to prevent the adverse impact of equipment failures on system operation. It provides comprehensive assessment and prediction of equipment status, helping to optimize equipment operation strategies and maintenance plans. The method for obtaining operating status information for target equipment disclosed in this embodiment uses connectivity indicators, centrality indicators, and community discovery algorithms to perform in-depth analysis of the location and role of equipment within the network and achieve a comprehensive understanding of the equipment's operating status. By identifying potential failure nodes and abnormal expansion paths, measures can be taken in advance to prevent equipment failure and optimize equipment operation and maintenance strategies. Data analysis and expert knowledge are integrated to provide scientific evidence and feasible suggestions for equipment management and maintenance operation decision-making. Through the implementation and integration of the above steps, an effective equipment-related model analysis process can be established, providing support and guidance for comprehensive assessment and optimization of the operating status of hydropower plant equipment. Specifically, graph theory and algorithms (shortest path, connectivity, etc.) are used to analyze the importance and impact paths of equipment nodes, and nodes that may be the origin of a system failure are identified. The failure node is precisely located through an analysis of the abnormality propagation path. Inference and prediction are performed on the graph (locating the failure node, evaluating the impact of the failure, etc.) to simulate the propagation and spread of equipment abnormalities. In other words, the path and impact range of the failure propagation within the system are evaluated, and how abnormalities are propagated and spread between pieces of equipment is understood. Simulations are performed to mimic the propagation process of equipment abnormality events, and the possible system response and impact are predicted.

[0063] A detailed explanation will be given below using a vibration-induced fault scenario of a hydroelectric power plant unit as an example. The main components of the unit are referred to as nodes, which include the spiral casing, the upper cover, the water guide mechanism, the volute casing, the draft tube, the main shaft, the impeller, the generator stator, and the rotor. Based on the unit structure, physical connection relationships between component nodes are established, for example, the impeller is connected to the main shaft, the stator and rotor are clearance-fitted, and the water guide mechanism and draft tube are connected. Based on the vibration sensor arrangement of each component, the corresponding relationship between the signal node and the equipment node is established, and the logical relationship between the monitoring point and the measured object is characterized. Based on correlation analysis of past vibration data, weights are added to some physical connection edges and logical corresponding edges to dynamically reflect the strength of the correlation. The calculation results clearly show that the number of connected subgraphs in this graph is small and the average node degree is high, indicating a high degree of correlation between each part of the unit and that holistic consideration is required. Degree centrality and betweenness centrality indicate that the main axis node is located at the center of the graph and is an important component connecting the upstream and downstream flows. Failures on the main axis are likely to spread to upstream hydraulic components and downstream power generation components. The community discovery algorithm extracted three closely related subgroups of parts, corresponding to the hydro-turbine flowpath parts group, the main shaft parts group, and the generator parts group, respectively, which are relatively consistent with the physical division of the unit. If a rotor vibration signal frequently triggers an alarm, the calculated shortest path analysis shows that this rotor is directly connected to the main shaft and has a clearance fit with the stator. As a preliminary guess, the rotor vibration is likely caused by the main shaft vibration or the stator vibration.

[0064] Further analysis of the vibration trends of the main shaft and stator reveals that the amplitude of the main shaft vibration is large and the frequency is close to that of the rotor vibration, and it is determined that the main shaft is the source of the fault. After tracing to the main shaft node, backward analysis shows that the problem with the main shaft may affect the impeller fixing and sealing of the water guide mechanism, further worsening the vibration state of the hydro-turbine flow path parts. Vibration spectrum analysis showed that there was an imbalance in the main shaft, and that imbalance correction was necessary. At the same time, inspection of the affected parts, such as the impeller fixing bolts and the water guide mechanism sealing ring, was recommended to prevent the problem from expanding. Compared with the isolated monitoring of a single facility, this method is supported by graph theory algorithms to analyze the mutual influence between facilities from the perspective of combinatorial topology, making the correlation analysis more systematic and standardized. Especially in large-scale systems, graph models can quickly identify risk areas, reduce inspection scope, and reduce human blindness.

[0065] Furthermore, for core equipment, it is possible to build precise mechanism models, such as the dynamic model of the hydraulic turbine speed governing system. Specifically, we will establish mathematical models of components such as the hydraulic turbine, generator, governor, and PID controller (torque equation for the hydraulic turbine, actuator equation for the governor, etc.). Based on the physical connection relationships of the components (such as direct connection between the hydro turbine output shaft and the generator shaft, or using the governor output signal as the input to the PID controller), each component model is connected in series and parallel to form a complete system model. The input / output interface of the model is defined, and the input (water head, volute casing opening, etc.) and output (unit rotation speed, power, etc.) correspond one-to-one with the actual monitored quantities. Select typical speed control process test data (time series curves of rotation speed, opening, power, etc. under operating conditions such as unit load remote adjustment and load disconnection). The monitoring data is introduced into a mechanism model, and optimization methods such as least squares method and genetic algorithm are used to fit the model parameters so that the simulated curve matches the measured curve as closely as possible. The model accuracy is verified under different operating conditions, and the model structure and parameters are fine-tuned to ensure that the model can reflect the true dynamic characteristics of the speed control system. Based on the mechanism model, the relationship between component parameters and system response is analyzed, and sensitivity analysis is performed. Through simulation tests, the expression characteristics and propagation laws of equipment failures are studied to guide the selection of alarm quantities.

[0066] Below, we will show a case analysis of a governor system failure using the governor system of a hydroelectric power plant as an example. The torque balance equation and flow continuity equation of the hydraulic turbine are established to describe the nonlinear coupling relationship between the unit rotation speed, head, opening and power output. A PID controller model is established to depict the logical relationship between the power system frequency deviation signal and the governor operation. A dynamic model of the governor's actuator and servo system is established to reflect the response characteristics after the governor receives a control command. The inputs and outputs of the hydraulic turbine, controller, and actuator are connected to construct a closed-loop speed control system simulation model. Using the full load rejection test data of the unit, the hydro turbine dynamic characteristic parameters in the model, the proportional, integral and derivative coefficients of the governor, the actuator time constants, etc. are identified. When changing the PID parameters of the governor while keeping other parameters fixed, a larger proportional coefficient results in a faster system response but can cause overshoot, while a larger integral coefficient results in a smaller steady-state error but can cause low-frequency oscillation. When the time constant of the actuator is changed while other parameters are fixed, the larger the time constant, the larger the transmission delay from the governor to the hydro-turbine, and the worse the dynamic performance of the system. If the proportional coefficient of the speed governor is set too high, the simulation shows that when the unit cuts off the unit load, the rotation speed significantly overshoots, causing high-frequency vibrations and requiring multiple adjustments to restore stability, which coincides with the instability observed in actual operation. When the actuator malfunctions, changing the actuator time constant causes the governor output signal to lose linkage with the guide vane opening in the simulation, resulting in the actual guide vane opening not reaching the governor setting value and reducing the control accuracy of the system. From the above, we can see that improper governor parameter settings and actuator failures directly cause abnormalities in the unit's rotation speed and output response. By monitoring the distortion of the rotation speed and output curves in real time and establishing rational alarm logic, potential problems in the speed governor system can be detected early.

[0067] This case uses mechanism modeling to analyze the interactions between processes within the equipment based on physical mechanisms, and reveals the impact of local defects on overall performance. Unlike purely data-driven correlation analysis, mechanism models incorporate domain knowledge, allowing for a more precise explanation of the underlying logic of equipment failures and a more convincing relationship between failure manifestations and root causes. Mechanism modeling can simulate the dynamic behavior of equipment in different dimensions, such as the time domain and frequency domain, and provide rich characteristic information, which is advantageous for the deployment of intelligent diagnosis. Mechanism models can be implemented online in real time, and through means such as residual analysis, potential risks such as system parameter drift and performance degradation can be detected in a timely manner, providing a reliable basis for equipment health management.

[0068] Furthermore, the causal relationships of the mechanism model are clear, and the model parameters often have physical significance, making them highly interpretable and easy for on-site workers to understand and accept. Through mechanism simulation training, maintenance personnel can deepen their understanding of complex equipment and strengthen their ability to judge equipment status and troubleshoot based on mechanisms. By conducting multi-dimensional and multi-angle correlation analysis, we can comprehensively dig into the complex underlying relationships between equipment from data and knowledge. This correlation analysis works from shallow to deep, from the surface to the inside, taking into account prior information such as physical structure and measurement point placement, while fully utilizing statistical laws contained in massive operational data. At the same time, we utilize expert experience and mechanism models to evaluate and optimize, ultimately forming an equipment correlation model that combines breadth and depth. Compared with the isolated monitoring of a single piece of equipment, the equipment-related model portrays the underlying mechanism of the system from a holistic perspective, enhancing the systematicity and objectivity of monitoring. On the one hand, alarms for abnormal equipment can quickly identify the root cause and trace the upstream and downstream equipment of the fault propagation. On the other hand, through the combined alarms of related equipment, the development trend of the situation can be predicted, and feasible control suggestions can be put forward, realizing intelligent warning and intelligent decision-making. For example, if an internal short circuit occurs in a transformer, the concentration of dissolved gases in the oil will rise sharply, and parameters such as oil temperature and oil level will also show abnormalities. At the same time, the impact of the short circuit current will cause a series of chain reactions, such as a sudden drop in bus voltage and overcurrent tripping of connecting switches. While traditional alarms only individually list the limit exceedance situations for each monitored quantity, related analysis reveals the entire process of fault propagation through a graph model, quickly identifying the fault location and shortening the identification time, providing a reliable basis for emergency repairs. For example, if a generator excitation system is short-circuited, the related analysis can not only detect abnormal changes in the excitation current, but also predict subsequent effects such as frequency fluctuations. At the same time, the related model can suggest optimal excitation control measures, guiding personnel to adjust the excitation method in a timely manner or shut down the faulty exciter, thereby minimizing the impact of the accident. In this way, equipment-related modeling makes alarms more comprehensive and systematic, actively revealing potential problems that are difficult to observe directly but pose a high risk, truly achieving "prevention."

[0069] Furthermore, the present invention includes intelligent alarm generation and display. Specifically, it monitors equipment operation data in real time and dynamically evaluates the equipment's health status by combining it with alarm models and strategies. If an abnormality is detected, a corresponding alarm is triggered. Depending on the alarm level, a corresponding alarm message is generated, including attribute information such as the alarm target, alarm cause, and alarm time, and displayed on the monitoring screen. Alarm messages are intelligently sorted and displayed based on filter conditions and priority, and can be used to attract personnel's attention using various methods such as pop-up windows, audio, and light. If an abnormality is detected, a corresponding alarm is activated.

[0070] Furthermore, this application also includes the optimization of alarm strategies based on safety reinforcement learning. Specifically, a Markov decision process (MDP) model of the equipment health state and alarm state is established, where the state space S represents the equipment health state, the action space A represents the presence or absence of an alarm, and the reward function R represents the utility brought by the alarm. By adopting a low-dimensional MDP technique, the high-dimensional state space is mapped to a low-dimensional feature space, reducing the complexity of searching the state space. Common dimensionality reduction methods include principal component analysis and independent component analysis. By introducing safety constraints, we modify the reward function of the original MDP model by penalizing alarm states that endanger the safety of the system and rewarding alarm states that are beneficial to the safety of the system, thereby ensuring that the reinforcement learning process complies with the safety requirements. It employs reinforcement learning algorithms such as Q-learning and SARSA to learn optimal alert strategies through interaction with the environment, where state transition probabilities and reward functions can be set based on actual data statistics or expert experience. The learned alarm strategy is applied online to select the optimal alarm action based on the real-time health status of the asset, while the strategy is continuously optimized and updated to adapt to changing asset conditions. Depending on the alarm level, a corresponding alarm message including attribute information such as the alarm target, alarm cause, and alarm time is generated and displayed on the monitoring screen. Alarm messages are intelligently sorted and displayed based on filter criteria and priority, and are alerted to personnel in a variety of ways, including pop-ups, audio and visual alerts.

[0071] Below, application examples will be given and explained in detail. The safety alarms for six units at a certain power plant face two major challenges: first, there are thousands of unit state parameters and complex nonlinear relationships, making them difficult to model using traditional rule-based or simple threshold-based alarm methods; second, excessively frequent alarms can disrupt normal operation, while insufficient alarms can cause accidents. The key to intelligent systems is how to optimize alarm strategies while ensuring safety. To address this issue, we attempted to optimize the alarm logic using a safety reinforcement learning algorithm based on low-dimensional MDP. First, we selected 200 representative parameters, including unit vibration, temperature, and current, to construct a high-dimensional equipment state space. At the same time, we classified alarm actions into two categories: "alarm" and "no alarm." Based on historical data and expert experience, we initially set safety utility values ​​for different state-action combinations. Next, we used principal component analysis to reduce the 200-dimensional state to 20, significantly compressing the state space. After the dimension reduction, we further introduced safety constraints to guide the strategy search by imposing severe penalties on alarm actions that cause unit trips and providing additional rewards for alarm actions that successfully avoid failures. Next, we conducted strategy learning using the Q-learning algorithm. Q-learning continuously tries different state-action combinations and updates the action value function Q(s, a) based on the returns fed back from the environment, thereby gradually improving the long-term utility of the strategy. During the learning process, we used an ε-greedy strategy to balance exploration and exploitation and avoid prematurely falling into a local optimum. After 10,000 rounds of training, the algorithm converged to a safe and efficient alerting strategy that autonomously adjusts alert sensitivity based on the equipment's condition, monitors and waits to avoid excessive alerts when the equipment experiences minor anomalies, and issues accurate and immediate alerts with a severity level indication when the equipment experiences multiple consecutive anomalies. In practical application, this intelligent alarm strategy reduces the missed detection rate by 50% and the false alarm rate by 80% compared to traditional methods, significantly improving the reliability and safety of alarms. The strategy is continuously optimized according to dynamic changes in the equipment, allowing personnel to grasp the equipment status in a timely manner and respond to unexpected risks with ease. This case demonstrates the advantages of safety reinforcement learning in optimizing equipment alarms: by efficiently modeling equipment states using low-dimensional MDPs, extracting inherent laws of state-behavior mapping using reinforcement learning, and embedding domain knowledge through safety constraints, the alarm strategy becomes more intelligent, safe, and explainable. This algorithm offers a new perspective on improving the automation and intelligence of equipment alarms.

[0072] Furthermore, the present invention also includes alarm information inquiry and management. Specifically, it classifies and compiles statistics on current and past alarms, performs trend analysis, and generates alarm reports to provide data support for equipment maintenance. Personnel at different levels are given the authority to query and handle corresponding alarms. Critical alarms can only be cleared after an approval process. The alarm strategy and knowledge base support online editing and self-learning optimization, forming an expert system to continuously improve the system's intelligence level. Case 1 (object-oriented alarm filtering): In the interactive topology diagram of the power distribution automation main station system, each equipment object has its own independent alarm status light. Personnel can quickly screen for abnormal equipment based on voltage level, equipment type, etc. If the bus voltage exceeds the limit value, the bus object will flash red to warn the user. By selecting the corresponding object, important information such as the bus voltage and connected switches can be viewed, facilitating timely response. Case 2 (Conditional Alarm): For a main transformer oil temperature alarm, it is generally necessary to simultaneously verify the load status of the transformer. The conditional statement can be expressed as follows: An over-temperature alarm will be issued only if the transformer load exceeds 80% of the rated load for 30 consecutive minutes and the oil temperature exceeds 85 degrees for 5 minutes. This will significantly reduce unnecessary frequent alarms. Case 3 (Related Quantity Alarm): When the generator stator winding temperature exceeds the safety limit and an alarm occurs, the system automatically displays the related quantities, such as the generator stator current, cooling water temperature, and unit active power. The cause of the temperature rise can be quickly analyzed without the need for personnel to search for it, and effective measures, such as control or cooling water adjustment, can be taken. The alarm response time is reduced from minutes to seconds. This intelligent alarm method makes full use of big data and artificial intelligence technology to realize automated and intelligent alarms for hydropower plants, improving the accuracy, reliability and operability of alarms. This system can be widely applied to the centralized control operation of large and medium-sized hydropower plants, bringing a new management mode to equipment status monitoring, fault warning and decision-making support.

[0073] According to a second aspect, an embodiment of the present invention also provides an intelligent alarm system for a computer monitoring system of a hydroelectric power plant, including: The pre-processing module is used to pre-process historical operation data of various equipment of the hydroelectric power plant to obtain a first data set, The operation data includes at least one of a state quantity, a telemetry quantity, and a teleindication quantity. The establishment module is used to establish a data model based on the first data set. The first related information acquisition module is used to acquire first related information based on the physical structure of various equipment in the hydroelectric power plant, and the first related information includes physical connection relationships between the equipment, a causal chain model, and a fault knowledge base. The second related information acquisition module is used to acquire second related information based on the operation data of various equipment of the hydroelectric power plant, and the second related information includes statistical correlation and frequent co-occurrence pattern information. The construction module is used to construct an equipment relational model based on the first relational information, the second relational information, and the data model. The alarm module is used to monitor multiple operation data of the target equipment in real time and dynamically obtain operation status information of the target equipment based on the equipment-related model. If an abnormality exists in the driving state information, an alarm is issued based on a preset alarm strategy. It should be noted that the intelligent alarm method for the computer monitoring system of the hydroelectric power plant disclosed in the first aspect of the present invention is applicable to the intelligent alarm system for the computer monitoring system of the hydroelectric power plant disclosed in the second aspect of the present application, and therefore the description thereof will be omitted here.

[0074] A computer device according to an embodiment of the present invention includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory can include one or more computer program products, which can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory can include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory can include, for example, read-only memory (ROM), a hard disk, flash memory, etc. The processor is a central processing unit (CPU) or other form of processing unit having data processing capability and / or instruction execution capability, and is capable of controlling other components in the computer device to perform desired functions. In one embodiment of the present invention, the processor is used to execute the computer-readable instructions stored in the memory and cause the computer device to perform all or part of the steps of the intelligent alarm method for a computer monitoring system of a hydroelectric power plant in each of the above-mentioned embodiments of the present invention. Those skilled in the art will recognize that in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses, interfaces, etc., and these well-known structures are also included in the scope of protection of the present invention.

[0075] 11 shows a structure suitable for realizing a computer device in an embodiment of the present invention. The computer device shown in FIG. 11 is merely an example and does not pose any limitation on the functionality and scope of use of the embodiment of the present invention. As shown in Figure 11, the computer device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.), which performs various appropriate operations and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). The RAM also stores various programs and data necessary for the operation of the computer device. The processor, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus. Typically, input devices such as sensors or visual information gathering devices, output devices such as displays, storage devices such as magnetic tapes or hard disks, and communication devices are connected to the I / O interface. The communication devices enable the computing device to communicate and exchange data with other devices (such as edge computing devices) via wired or wireless communication. While FIG. 11 illustrates a computing device having various devices, not all of the devices illustrated in the figure need be implemented. Alternatively, more or fewer devices may be implemented. In particular, according to embodiments of the present invention, the processes described with reference to the flowcharts above can be implemented as a computer software program. For example, embodiments of the present invention include a computer program product including a computer program carried on a non-transitory computer-readable medium, the computer program including program code for performing the methods illustrated in the flowcharts. In such embodiments, the computer program is downloaded and installed from a network via a communications device, or installed from a storage device, or installed from ROM. When the computer program is executed by a processor, all or part of the steps of the intelligent alarm method for a computer monitoring system for a hydroelectric power plant according to embodiments of the present invention are performed. For detailed explanations of this embodiment, please refer to the corresponding explanations in the above-mentioned embodiments, and the explanation will be omitted here.

[0076] A computer-readable storage medium according to an embodiment of the present invention stores non-transitory computer-readable instructions, which, when executed by a processor, perform all or part of the steps of the intelligent alarm method for a computer monitoring system of a hydroelectric power plant according to each of the above-mentioned embodiments of the present invention. The computer-readable storage medium includes, but is not limited to, optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or removable hard disk), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge). For detailed explanations of this embodiment, please refer to the corresponding explanations in the above-mentioned embodiments, and the explanation will be omitted here. As described above, the present invention can be successfully implemented.

[0077] All features disclosed in any embodiment herein, or steps in any method or process implicitly disclosed, can be combined and / or extended or substituted in any manner, except for mutually exclusive features and / or steps. The above description is merely a preferred embodiment of the present invention and does not limit the present invention in any way. Based on the technical substance of the present invention, any simple modification, equivalent replacement or improvement made to the above embodiment within the spirit and principle of the present invention falls within the protection scope of the technical means of the present invention.

Claims

1. 1. An intelligent alarm method for a computer monitoring system of a hydroelectric power plant, comprising: S100, pre-processing historical operating data of the hydroelectric power plant equipment to obtain a first data set; S200. Constructing a data model based on the first data set; S300, obtaining first relevant information based on the physical structure of various equipment of the hydroelectric power plant; S400, obtaining second related information based on the operation data of various equipment of the hydroelectric power plant; S500, constructing an equipment relation model based on the first relation information, the second relation information, and the data model; S600: monitoring the operation data of the target equipment in real time, dynamically obtaining the operation status information of the target equipment based on the equipment-related model, and issuing an alarm based on a preset alarm strategy when an abnormality exists in the operation status information; An intelligent alarm method for a computer monitoring system of a hydroelectric power station, comprising the steps above.

2. Step S200 includes the following steps: S211. Extracting feature data from the first dataset, the feature data including one or more of mean, variance, peak-to-peak value, and slope; S212: performing feature selection processing on the feature data to obtain a second dataset; and constructing a sample set based on the second dataset, typical normal samples, and typical fault samples, the sample set including a training set and a test set; S213: constructing a supervised learning model, training the supervised learning model through the training set, testing the supervised learning model that meets the training conditions based on the preset indicators in the test set, dynamically tuning the supervised learning model based on the test result, and recording the supervised learning model after dynamic tuning as the data model. The intelligent alarm method for a computer monitoring system of a hydropower station as described in claim 1,

3. Step S300 includes the following steps: S310, extracting physical connection relationships between equipment according to the system topology structure in the hydroelectric power plant; S320: Obtaining failure prediction propagation path information based on the physical connection relationship; S330: obtaining the first relevant information based on the fault prediction propagation path information, past fault information and corresponding countermeasures.

4. Step S400 includes the following steps: S410, calculating statistical correlations between different equipment operating parameters; S420. Obtain frequent co-occurrence pattern information of equipment status according to a data analysis algorithm, wherein the data analysis algorithm includes one or more of an Apriori algorithm and an FP-growth algorithm; S430: Obtaining the second related information based on the frequent co-occurrence pattern information and the expert library information, wherein the second related information includes strong related pattern information with a predetermined value selected from the mass related rules of the hydroelectric power plant equipment.

5. Step S420 includes the following steps: S421. Determine an input data set for the data analysis algorithm based on statistical correlation; S422. Obtain a target frequent itemset based on the input data set; S423: Obtaining the frequent co-occurrence pattern information based on the target frequent item set. The intelligent alarm method for a computer monitoring system of a hydroelectric power station according to claim 4 .

6. Step S421 includes the following steps: A100, selecting the operating state parameters of the equipment as a transaction set for related analysis; A200, discretizing the operating state parameters and converting the operating state parameters into Boolean or enumerated values; A300, the intelligent alarm method for a computer monitoring system of a hydroelectric power plant according to claim 5, characterized in that all equipment statuses at each time sampling point are taken as one transaction record, and a plurality of transaction records constitute the input data set of the data analysis algorithm.

7. Step S422 includes the following steps: B100. Set a minimum support threshold, and obtain the occurrence frequency of each state according to the input data set. Remove itemsets in the input data set whose support is lower than the minimum support threshold to obtain a first frequent itemset, wherein the first frequent itemset is a frequent set of a single item; B200, performing a self-join operation on the first frequent itemsets to generate a first candidate itemset containing two items; scanning the input dataset to obtain actual support of the first candidate itemsets; filtering out first candidate itemsets whose actual support is lower than the minimum support threshold; and obtaining a second frequent itemset, wherein the actual support is the actual number of occurrences of each first candidate itemset; B300: performing a self-join operation on the Pth frequent itemsets to generate a Pth candidate itemset containing two items; scanning the input dataset to obtain actual supports of the Pth candidate itemsets, where each actual support is the actual number of times each Pth candidate itemset appears; filtering out Pth candidate itemsets whose actual supports are lower than the minimum support threshold to obtain a P+1th frequent itemset; and if the actual supports of all Pth candidate itemsets are lower than the minimum support threshold, recording the Pth frequent itemset as the target frequent itemset, where P≧2.

8. Step S423 includes the following steps: C100, enumerating all non-empty subsets of the target frequent itemset, each non-empty subset together with its complement constituting one candidate association rule; C200, calculate the confidence of each candidate association rule; C300, filtering out candidate association rules whose confidence is less than a minimum confidence threshold, to obtain the frequent co-occurrence pattern information. The intelligent alarm method for a computer monitoring system of a hydroelectric power plant according to claim 7 .

9. Step S430 includes the following steps: S431: ranking the rules in the frequent co-occurrence pattern information according to their support and reliability; S432: Based on the expert library information, screening and eliminating superficial rules in the strong association rule set to obtain the second association information. The intelligent alarm method for a computer monitoring system of a hydroelectric power station as described in claim 8.

10. In step S600, dynamically acquire the operating state information of the target equipment based on the equipment-related model; D100, obtaining a connectivity index of operation data based on the equipment-related model; D200, based on the equipment association model, obtain a centrality index for each node, where each node corresponds to one equipment; D300, using a community discovery algorithm to analyze the connectivity index and the centrality index, and obtain equipment subgroups whose actual closeness exceeds a closeness threshold; D400, analyze the equipment subgroup to obtain the operating state information, and the operating state information includes potential failure nodes and corresponding failure impacts, and abnormal future expansion paths; The intelligent alarm method for a computer monitoring system of a hydroelectric power station according to any one of claims 1 to 9, comprising the steps above.

Images