Fault prediction method for computer monitoring system of hydropower station based on big data analysis

Abstract

The present application relates to the field of intelligent monitoring and fault prediction, and discloses a fault prediction method for a hydropower station computer monitoring system based on big data analysis, including the following steps: collecting multidimensional sensor data of key equipment in the hydropower station and preprocessing the data; constructing a Morse function and calculating its gradient and Hessian matrix, extracting the topological features of the system, and using the Morse‑Smale complex to divide the system state and distinguish between stable, transitional and fault states; based on topological features and time series data, using the Transformer model to predict faults and calculate the probability of equipment failure; based on the calculated fault discrimination score value and combined with a dynamic threshold adjustment strategy, if the score value exceeds the preset threshold, the alarm mechanism of the SCADA monitoring system is triggered to achieve real-time fault warning. The present invention improves the accuracy and real-time performance of equipment fault prediction through comprehensive data analysis and deep learning models, ensuring the stable operation and safe maintenance of hydropower station equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and fault prediction, and in particular to a fault prediction method for a hydropower station computer monitoring system based on big data analysis. Background Art

[0002] At present, the equipment monitoring systems of hydropower stations mostly rely on traditional condition monitoring and analysis methods, usually using alarm mechanisms based on preset rules, or simple statistical analysis methods to evaluate the health status of equipment. These methods rely on single-dimensional data for analysis and often ignore the complex interactions and potential nonlinear relationships in equipment operation. With the changes in the equipment operating environment and the influence of various factors, traditional methods have certain limitations in the accuracy and real-time nature of fault prediction.

[0003] Fault prediction in existing technologies usually uses a simple statistical analysis method based on thresholds. For abnormal situations that occur during equipment operation, judgments are often made based on pre-set alarm thresholds. However, this static threshold setting may lead to false alarms or missed alarms, and cannot fully consider the differences between different equipment or operating environments. In addition, as the complexity of equipment and monitoring data increases, traditional methods cannot effectively process large amounts of multidimensional sensor data and lack sufficient capabilities for in-depth analysis and prediction.

[0004] In terms of data processing, existing technologies often use simple data denoising and anomaly detection technologies, but their accuracy and adaptability are insufficient and cannot achieve ideal results in large-scale data processing and real-time monitoring. At the same time, existing methods usually lack comprehensive modeling and multi-level analysis of the system's operating status, and often only focus on local information while ignoring changes in the overall operating status.

[0005] Therefore, the existing technology has significant deficiencies in processing complex multidimensional data and improving fault prediction accuracy and real-time performance. A new method is needed to overcome these shortcomings and achieve more accurate prediction and real-time monitoring of equipment operating status. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a fault prediction method for a hydropower station computer monitoring system based on big data analysis. Through big data analysis and topological data analysis, the accuracy and real-time performance of fault prediction of hydropower station equipment are improved, overcoming the shortcomings of traditional monitoring methods in complex multi-dimensional data processing and real-time early warning.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A fault prediction method for a hydropower station computer monitoring system based on big data analysis comprises the following steps:

[0008] Collect multi-dimensional sensor data from the hydropower station computer monitoring system and perform pre-processing;

[0009] Constructing a Morse function based on the sensor data, calculating the gradient and the Hessian matrix, extracting the topological features of the system operation state, and constructing a Morse-Smale complex based on the topological features to divide the system operation state area;

[0010] The prediction model is used to model the topological characteristics of the Morse-Smale complex and time series data to predict the possibility of failure;

[0011] The fault discrimination score value is calculated based on the prediction results. When the score value exceeds the preset threshold, the fault alarm of the monitoring system is triggered.

[0012] Preferably, the sensor data includes vibration signals, temperature signals, load power and system entropy values.

[0013] Preferably, the preprocessing step includes using wavelet transform to remove noise, using local outlier factor algorithm to detect outliers, and performing Min-Max normalization on all data.

[0014] Preferably, the step of constructing a Morse function based on the sensor data comprises:

[0015] Selecting multiple sensor data variables representing the operating state of the system, including at least a vibration signal, a temperature signal, a load power and a system entropy value;

[0016] Set weight parameters and weight each variable according to its influence on system stability;

[0017] A Morse function is constructed to reflect the stability of the system's operating state and meet the differentiability condition. The construction form of the Morse function is a weighted combination of multiple sensor variables to ensure accurate representation of the system's state topological structure.

[0018] Preferably, the steps of constructing the Morse-Smale complex include:

[0019] Calculate the gradient vector of the Morse function and solve the gradient vanishing point to determine the critical point distribution of the Morse complex;

[0020] Calculate the Hessian matrix and analyze the Morse index based on its eigenvalue to determine the system stability under different states;

[0021] The system state regions are divided based on the Morse index, where the stable region corresponds to the critical point where the Morse index is zero, the transition region corresponds to the critical point where the Morse index is one, and the unstable region corresponds to the critical point where the Morse index is greater than one;

[0022] A Morse-Smale complex is constructed, and the topological structure of the system operation state is formed through the intersection of the stable manifold and the unstable manifold. The partition boundary of the Morse-Smale complex is adjusted according to the change of the system state.

[0023] Preferably, the step of constructing the input features of the prediction model includes:

[0024] Extract the topological features of the Morse-Smale complex, including at least the distribution of critical points, the boundary of the stable region and the Betti number;

[0025] Calculate time series characteristics of sensor data, including trend rate, mean, variance, and autoregressive eigenvalues;

[0026] The topological features of the Morse-Smale complex are fused with the time series features to form an input feature matrix, and features of different dimensions are uniformly processed through a standardization method.

[0027] Preferably, the prediction model uses a Transformer structure to perform prediction calculations, including the following steps:

[0028] Construct the input sequence matrix and perform nonlinear mapping on the input features through the embedding layer;

[0029] Calculate the query matrix, key matrix and value matrix of the input feature matrix, and calculate the weighted attention distribution based on the multi-head self-attention mechanism to obtain long-term dependencies;

[0030] A feedforward neural network is used to perform nonlinear transformation on the output of the attention mechanism, and residual connections and layer normalization are used to improve training stability.

[0031] Calculate the prediction output and adjust the prediction parameters based on historical data to improve the estimation accuracy of the probability of failure.

[0032] Preferably, the step of calculating the fault discrimination score value comprises:

[0033] Set a multi-dimensional fault feature vector, including the topological features of the Morse-Smale complex, the time series features, and the output probability value of the prediction model;

[0034] The fault feature vector is converted into a fault score value by using a weighted nonlinear mapping method, wherein the fault discrimination score value is calculated as follows:

[0035]

[0036] in, is the weighting coefficient of the fault feature, is the corresponding eigenvalue, is a nonlinear mapping function;

[0037] The weighting coefficient is adjusted based on historical operating data to improve the calculation accuracy of the fault score value.

[0038] Preferably, the fault alarm triggering threshold of the monitoring system adopts a dynamic adjustment strategy, including the following steps:

[0039] Calculate the statistical distribution of the current fault discrimination score value and determine the mean and standard deviation within the set time window;

[0040] Adopt an adaptive adjustment strategy to calculate the dynamic alarm threshold based on statistical characteristics, where the dynamic threshold is calculated as follows:

[0041]

[0042] in, is the mean of the rating values, is the standard deviation of the ratings, is the preset adjustment factor;

[0043] When the fault discrimination score value exceeds the dynamic alarm threshold, the monitoring system alarm is triggered and an alarm log is recorded to optimize subsequent threshold settings.

[0044] The present invention also provides a fault prediction device for a hydropower station computer monitoring system based on big data analysis, comprising:

[0045] A data acquisition module, used to collect multi-dimensional sensor data of the hydropower station computer monitoring system and pre-process the data;

[0046] A topological analysis module, used to construct a Morse function based on the sensor data, calculate the gradient and the Hessian matrix, extract the topological features of the system operation state, and construct a Morse-Smale complex based on the topological features to divide the system operation state area;

[0047] A fault prediction module is used to model the topological features of the Morse-Smale complex and time series data using a prediction model to predict the possibility of a fault;

[0048] The discrimination and alarm module is used to calculate the fault discrimination score value according to the prediction result, and trigger the fault alarm of the monitoring system when the score value exceeds a preset threshold.

[0049] The present invention provides a fault prediction method for a hydropower station computer monitoring system based on big data analysis. It has the following beneficial effects:

[0050] 1. The present invention combines multi-dimensional sensor data and topological analysis technology, performs topological division of the system based on the Morse-Smale complex, and performs fault prediction in combination with the Transformer model of deep learning, which can accurately identify the operating status and potential faults of hydropower station equipment. Through multi-level and multi-dimensional information processing, the present invention greatly improves the accuracy of fault prediction and ensures the timely discovery of potential faults.

[0051] 2. The present invention provides an effective real-time monitoring method by calculating the fault discrimination score and adopting a dynamic threshold adjustment strategy to issue an early warning before the equipment failure occurs. This mechanism not only improves the safety of the equipment, but also helps the operation and maintenance personnel take preventive measures before the failure occurs, thereby reducing equipment downtime and maintenance costs.

[0052] 3. The fault prediction system of the present invention can adapt to the working status and operating environment of different devices. The adjustment of dynamic thresholds is based on the statistical characteristics of historical operating data, so that the system can flexibly respond to the usage scenarios of different devices and avoid the problem of false positives or false negatives caused by static thresholds.

[0053] 4. The present invention combines multiple data sources and comprehensively analyzes the operating status of the equipment by integrating Morse function with time series data. This method fully utilizes the potential of sensor data through multi-dimensional feature extraction and improves the ability to identify equipment failure modes.

[0054] 5. The present invention can identify potential faults in advance and provide alarms, so that equipment maintenance personnel can perform maintenance or replace parts before the fault occurs. This predictive maintenance not only improves the operating efficiency of the equipment, but also effectively reduces the cost of emergency repairs and equipment downtime, and improves the overall operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0056] Figure 2 It is a schematic diagram of the device structure of the present invention.

[0057] Among them, 10, data acquisition module; 20, topology analysis module; 30, fault prediction module; 40, identification and alarm module. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The present invention provides a fault prediction method for a hydropower station computer monitoring system based on big data analysis, uses Morse-Smale complexes for topological analysis, and models the system state in combination with a prediction model to achieve early warning of faults.

[0060] like Figure 1 As shown, the fault prediction method of the hydropower station computer monitoring system based on big data analysis may include the following steps:

[0061] S1. Collect sensor data such as vibration, temperature, power, entropy value, etc. of key equipment in hydropower stations, and improve data quality through denoising, anomaly detection and normalization processing;

[0062] S2, construct the Morse function, calculate the gradient and Hessian matrix, extract the system topology characteristics, and use the Morse-Smale complex to divide the system operation state area to distinguish between stable, transitional and fault states;

[0063] S3, based on the topological features of the Morse-Smale complex and time series data, use the Transformer prediction model to build a model, calculate the future trend of the equipment operation status, and predict the probability of failure;

[0064] S4. Calculate the fault discrimination score value and adopt a dynamic threshold adjustment strategy. If the score exceeds the preset threshold, the alarm mechanism of the SCADA monitoring system is triggered to achieve real-time fault warning.

[0065] Next, each step of the method and its specific implementation process will be described in detail.

[0066] For step S1, in this embodiment, data collection is the primary step of the fault prediction method, which directly affects the accuracy of subsequent data processing and fault prediction models. Specifically, the sensor data in this step includes but is not limited to the vibration signal, temperature signal, power signal, and entropy value of the device. These sensor data represent the operating status and health status of the device, and can provide the system with early warning signals about the occurrence of faults.

[0067] First, the vibration signal It is an important parameter to measure the mechanical status of the equipment and can reflect whether the equipment has mechanical wear or imbalance problems. It is mainly used to monitor the temperature changes of motors, bearings, generators and other equipment. Excessive temperature usually means that there is a potential failure risk in the equipment. Power signal It can reflect the load condition of the unit. Overload or unbalanced load may cause equipment failure. Through information entropy calculation, the system stability is evaluated and the system's working efficiency and abnormal status are indirectly reflected.

[0068] After the sensor data is collected, in order to ensure the quality of the data, the collected data must be preprocessed. Data preprocessing mainly includes denoising, anomaly detection and normalization, the purpose of which is to remove interference signals in the sensor data and ensure the accuracy of the data, thereby improving the reliability of subsequent analysis and fault prediction.

[0069] De-noising is the first step of pre-processing, and its purpose is to remove high-frequency noise in sensor data. To this end, the wavelet transform method is used to smooth the signal in this embodiment. Wavelet transform achieves the effect of denoising by decomposing the signal into different frequency components and applying appropriate filtering operations on each frequency component. The formula is:

[0070]

[0071] in, is the wavelet coefficient, is the wavelet basis function, It is the denoised signal. Wavelet transform can effectively retain the useful components in the signal while removing irrelevant noise.

[0072] Anomaly detection is the second step of preprocessing, which aims to identify and remove data points that do not conform to the normal range. To this end, this embodiment uses the local outlier factor (LOF) algorithm for outlier detection. The LOF algorithm determines whether a data point is abnormal based on the neighborhood density. If the neighborhood density of a point is significantly lower than that of its neighbors, the point is considered abnormal. The LOF algorithm can effectively identify abnormal data points and remove them. The formula is as follows:

[0073]

[0074] in, is the local reachable density, For point of Nearest neighbor. In this way, the LOF algorithm can clearly distinguish normal data from abnormal data, thus ensuring the quality and reliability of the data.

[0075] Normalization is the last step of preprocessing, and its purpose is to map all sensor data to a uniform numerical range for subsequent analysis and modeling. Usually, the range of data normalization is set to [0,1]. In this way, the imbalance problem caused by different dimensions of different features can be avoided. In this embodiment, the Min-Max normalization method is used, which maps the data to the [0,1] range through the following formula:

[0076]

[0077] in, is the original data, and are the minimum and maximum values ​​of the data, respectively. This is the normalized data. Through this method, the numerical range of all data features is unified, providing a data basis for subsequent model training and fault prediction.

[0078] Through the above preprocessing steps, the collected hydropower station monitoring data will be cleaned and standardized, thus providing data input for the subsequent Morse function construction and fault prediction.

[0079] For step S2, in this embodiment, by constructing the Morse function, calculating its gradient and Hessian matrix, the topological features of the system are extracted, and the Morse-Smale complex is constructed based on these features. By dividing the operating state area of ​​the system, this step can distinguish the stable, transitional, and fault states of the system, thereby providing a reliable basis for subsequent fault prediction and early warning.

[0080] In this embodiment, a Morse function representing the operating state of the system is first constructed. The purpose of the Morse function is to use sensor data to reflect the overall state of the system by weighted combination of different parameters. Specifically, the constructed Morse function can be expressed as:

[0081]

[0082] in:

[0083] Represents the entropy value of the system, which is used to measure the stability or disorder of the system;

[0084] The vibration signal representing the equipment is an important indicator of mechanical failure;

[0085] Represents the load power of the equipment and reflects the working load status of the equipment;

[0086] Represents the temperature signal of the device, used to monitor the thermal status of the device;

[0087] , , is the weight coefficient, which is used to weight the influence of each parameter on the system stability.

[0088] The construction of the Morse function enables it to reflect changes in system status and its early warning effect on equipment failure. By weighting these parameters, the present invention ensures that the constructed Morse function can accurately capture key changes in equipment operation and provide necessary stability analysis.

[0089] Next, we use the constructed Morse function to calculate its gradient and Hessian matrix. The gradient is a tool to describe the rate of change of a function, which provides information about the change of the system state in different directions. , the changing trend and abnormal state of the system can be determined. Specifically, the gradient vector is:

[0090]

[0091] The calculation of gradient vectors helps to identify the key change points of the system and further understand the dynamic behavior of the system under different states.

[0092] The Hessian matrix is ​​a second-order derivative matrix used to describe the quadratic change of the system state. The eigenvalues ​​of the Hessian matrix can reflect the stability of the system and the fault warning signal. The calculation of the Hessian matrix is:

[0093]

[0094] The Hessian matrix can be used to determine the stability region of the system and provide important information for subsequent fault prediction. Specifically, by analyzing the eigenvalues ​​of the Hessian matrix, it can be determined whether the system is in a normal state, a transition state, or a fault state.

[0095] Furthermore, based on the Morse function and its calculated gradient and Hessian matrix, this embodiment divides the operating state region of the system by constructing a Morse-Smale complex. The Morse-Smale complex is a topological tool used to analyze the topological structure of the system state space, especially in the presence of critical points and saddle points. By constructing the complex, the stable state, transition state, and fault state of the system can be distinguished. The specific steps are as follows:

[0096] Calculate the critical point of the Morse function: based on the gradient information Calculating critical points, which correspond to key changes in system state, can help identify precursors to failure.

[0097] Calculate the Morse index: Based on the eigenvalues ​​of the Hessian matrix , the Morse index can be calculated to determine whether the system is in different states:

[0098] When k = 0, it means that the system is at a stable point, that is, the equipment is operating normally;

[0099] When k = 1, it means that the system is in a saddle point, i.e., a transition state, which may be a precursor to failure;

[0100] When k=2, it means that the system is at an unstable point, that is, the system is in a fault state.

[0101] Divide system status areas: Based on the analysis of Morse index and critical points, the system's operating status is divided into stable area, transition area and fault area. These areas can help accurately identify the current operating status of the equipment and provide strong support for subsequent fault prediction.

[0102] Through the construction of the Morse-Smale complex, this embodiment can effectively divide the system state and implement fault detection and early warning on this basis. Each divided area (stable, transition, fault) corresponds to different equipment operation characteristics, providing a basis for subsequent fault prediction and model training.

[0103] Step S2 in this embodiment combines the construction of Morse function, the calculation of gradient and Hessian matrix, and the partition of Morse-Smale complex to form a complete system state analysis framework. Through this framework, in-depth topological analysis of the equipment operation status of the hydropower station can be performed, and the operation anomalies of the equipment can be effectively identified, providing reliable data support for subsequent fault prediction.

[0104] For step S3, in this embodiment, based on the topological features extracted by the Morse-Smale complex and time series data, the Transformer prediction model is used to model the equipment operation status and further predict the probability of failure. This step aims to combine topological analysis with time series data, use deep learning methods to accurately predict the future state of the system, and provide effective support for fault warning.

[0105] First, the data input to the prediction model consists of two parts: topological features and time series data. Among them, the topological features include the system state information extracted by the Morse-Smale complex constructed in step S2. Specifically, the topological features include key features such as critical point distribution, stable region boundary, Betti number, etc. These features reflect the topological structure and state changes of the system at different time points. The time series data comes from the equipment operation data collected by the sensor, such as vibration signals, temperature signals, load power, etc., which reflect the actual operation status and trend of the equipment.

[0106] In this embodiment, the time series data includes the historical operation data of the equipment, which is mainly used to capture the evolution trend of the system status. By modeling the time series, the dynamic characteristics and change rules of the system can be captured, thereby providing valuable reference for subsequent fault prediction. Common time series features such as trend change rate, mean, variance, etc. are all processed as input features in this step.

[0107] Next, based on the above two types of data, this embodiment uses the Transformer model to model, predict the future operation trend of the equipment, and calculate its failure probability. The Transformer model is a deep learning structure based on the self-attention mechanism, which is widely used in processing time series data and long-term dependency tasks, and has the advantage of processing sequence data.

[0108] The working principle of the Transformer model includes several key steps. In this embodiment, the core part of the Transformer model is the multi-head self-attention mechanism, which enables the model to focus on different parts of the input data from different angles. Specifically, the model first maps the input topological features and time series data to a higher-dimensional space through the embedding layer and generates query, key, and value matrices. Each matrix is ​​calculated by the following formula:

[0109]

[0110] in, For input data, , , are the weight matrices for query, key, and value, respectively. By calculating the dot product of the query matrix and the key matrix, we can get the weighted attention distribution:

[0111]

[0112] here, is the dimension scaling factor of the key matrix, Used to generate probability distribution. Finally, the model uses the attention distribution to perform weighted averaging on the value matrix to obtain the context information for each time step.

[0113] Through the above process, the Transformer model can capture the long-term dependencies between various time points in the input data, thereby learning the evolution of the equipment's operating status. Based on this contextual information, the model performs nonlinear transformations on the data through a feedforward neural network and calculates the prediction results for each time step.

[0114] In this step, the prediction output is the probability distribution of the future state of the equipment, which is specifically expressed as the probability of failure. This probability value can be used as the basis for fault warning, indicating the risk of equipment failure at a certain point in the future. Through the training process, the model continuously adjusts internal parameters to optimize the prediction accuracy.

[0115] By using the Transformer model, this embodiment can effectively predict the future trend of the equipment, especially when facing time series data with long-term dependencies, the Transformer model can effectively capture these complex dependencies and provide accurate fault prediction results. In addition, since the Transformer model has strong parallel computing capabilities, the fault prediction process of this embodiment has high computing efficiency and is suitable for large industrial systems such as hydropower stations.

[0116] In this embodiment, the Transformer model can not only effectively process multidimensional time series data, but also integrate the topological features of the Morse-Smale complex with the time series data, thereby achieving comprehensive modeling of the equipment operation status. This modeling method can provide efficient fault prediction and provide a scientific basis for the health management and maintenance decision-making of the hydropower station.

[0117] For step S4, in this embodiment, based on the equipment operation status information obtained in the previous steps, the fault discrimination score is calculated, and the fault alarm is triggered through the dynamic threshold adjustment strategy. The purpose of this step is to evaluate the risk of equipment failure based on the output results of the prediction model, and to issue an alarm when necessary to ensure the safe operation of the hydropower station.

[0118] First, in this embodiment, the calculation of the fault discrimination score is based on multiple input features, mainly including topological features extracted by Morse-Smale complex, historical time series data and fault probability values ​​predicted by Transformer model. The calculation formula of the fault score is as follows:

[0119]

[0120] in, is the weight coefficient of the feature, For each feature value, is a nonlinear mapping function. Through this formula, the system integrates multi-dimensional feature information into a scoring value, reflecting the possibility of equipment failure.

[0121] Specifically, the calculation of the score value not only depends on the input of each sensor data, but also combines the topological features of the Morse-Smale complex (such as critical point distribution, stable area, etc.) to improve the accuracy of the prediction through a weighted fusion method. The importance in fault prediction is different, the weight It reflects the influence of the feature on the system fault identification. Therefore, by adjusting these weights, the contribution of different features to the final score can be flexibly controlled.

[0122] Next, the dynamic threshold adjustment strategy is applied to the evaluation of the fault discrimination score. The goal of the dynamic threshold is to adaptively adjust the alarm threshold according to the current operating status of the system and the statistical characteristics of historical data in order to accurately determine the risk level of the equipment. Specifically, the threshold can be calculated using the following formula:

[0123]

[0124] in, is the mean of the rating values, is the standard deviation of the ratings, In this way, the dynamic adjustment strategy automatically sets a suitable alarm threshold according to the distribution of the score value, thereby ensuring the flexibility and accuracy of the alarm mechanism.

[0125] The core idea of ​​the dynamic threshold adjustment strategy is: if the score value Exceeding the set threshold , the device is considered to be at a high risk of failure and the alarm mechanism needs to be triggered. The advantage of dynamic adjustment is that it can make real-time adjustments based on the device's historical operating data, current status, and system changes, thereby avoiding false alarms or missed alarms caused by static threshold settings.

[0126] In addition, the calculation of the fault discrimination score and the dynamic threshold adjustment strategy in this embodiment also involve continuous learning and optimization of historical data. In the process of fault prediction, the system gradually accumulates data according to the actual operation of the equipment and adjusts the weights. and threshold , ensuring that fault identification and alarm mechanisms become more accurate over time. Each alarm event will be logged as a basis for subsequent adjustment of thresholds and optimization of models.

[0127] Finally, when the fault discrimination score value Exceeding the dynamically adjusted alarm threshold When the system triggers the SCADA monitoring system alarm mechanism of the hydropower station, the alarm mechanism can send a warning signal to the operation and maintenance personnel and provide early warning information of equipment failure. In this way, this embodiment can not only detect the abnormal state of the equipment in real time, but also reduce the probability of failure through a timely alarm mechanism, thereby ensuring the safe and reliable operation of the hydropower station.

[0128] In general, the present invention provides a method of collecting and preprocessing sensor data such as vibration, temperature, power and entropy of key equipment, constructing a Morse function and calculating its gradient and Hessian matrix, extracting the topological characteristics of the system, and then using the Morse-Smale complex to divide the system into stable, transitional and fault state regions. Based on these topological characteristics and time series data, the Transformer prediction model is used to predict the probability of fault occurrence, and the fault discrimination score is calculated in combination with the dynamic threshold adjustment strategy, and finally, by triggering the alarm mechanism of the SCADA monitoring system, real-time fault warning is achieved. This method can effectively improve the health management and fault prediction capabilities of hydropower station equipment and ensure the stable operation of the equipment.

[0129] The device for predicting faults of a hydropower station computer monitoring system based on big data analysis described below and the method for predicting faults of a hydropower station computer monitoring system based on big data analysis described above can be referred to each other.

[0130] Please see attached Figure 2 The present invention also provides a fault prediction device for a hydropower station computer monitoring system based on big data analysis, comprising:

[0131] The data acquisition module 10 is used to collect multi-dimensional sensor data of the hydropower station computer monitoring system and pre-process the data;

[0132] A topological analysis module 20, configured to construct a Morse function based on the sensor data, calculate a gradient and a Hessian matrix, extract topological features of the system operation state, and construct a Morse-Smale complex based on the topological features to divide the system operation state area;

[0133] A fault prediction module 30 is used to model the topological characteristics of the Morse-Smale complex and the time series data using a prediction model to predict the possibility of a fault;

[0134] The discrimination and alarm module 40 is used to calculate a fault discrimination score value according to the prediction result, and trigger a fault alarm of the monitoring system when the score value exceeds a preset threshold.

[0135] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0136] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fault prediction method for a hydropower station computer monitoring system based on big data analysis, characterized in that: The following steps are involved: Collect multi-dimensional sensor data from the hydropower station computer monitoring system and perform pre-processing; Constructing a Morse function based on the sensor data, calculating the gradient and the Hessian matrix, extracting the topological features of the system operation state, and constructing a Morse-Smale complex based on the topological features to divide the system operation state area; The prediction model is used to model the topological characteristics of the Morse-Smale complex and time series data to predict the possibility of failure; Calculate the fault discrimination score value based on the prediction results, and trigger the fault alarm of the monitoring system when the score value exceeds the preset threshold; The steps of constructing the input features of the prediction model include: Extract the topological features of the Morse-Smale complex, including at least the distribution of critical points, the boundary of the stable region and the Betti number; Calculate time series characteristics of sensor data, including trend rate, mean, variance, and autoregressive eigenvalues; The topological features of the Morse-Smale complex are fused with the time series features to form an input feature matrix, and the features of different dimensions are uniformly processed by a standardization method; The prediction model uses a Transformer structure to perform prediction calculations, including the following steps: Construct the input sequence matrix and perform nonlinear mapping on the input features through the embedding layer; Calculate the query matrix, key matrix and value matrix of the input feature matrix, and calculate the weighted attention distribution based on the multi-head self-attention mechanism to obtain long-term dependencies; A feedforward neural network is used to perform nonlinear transformation on the output of the attention mechanism, and residual connections and layer normalization are used to improve training stability. Calculate the prediction output and adjust the prediction parameters based on historical data to improve the estimation accuracy of the probability of failure.

2. The method for predicting faults of a hydropower station computer monitoring system based on big data analysis according to claim 1 is characterized in that: The sensor data includes vibration signals, temperature signals, load power and system entropy values.

3. The method for predicting faults of a hydropower station computer monitoring system based on big data analysis according to claim 1 is characterized in that: The preprocessing step includes using wavelet transform to remove noise, using local outlier factor algorithm to detect outliers, and performing Min-Max normalization on all data.

4. The method for predicting faults of a hydropower station computer monitoring system based on big data analysis according to claim 1 is characterized in that: The step of constructing a Morse function based on the sensor data comprises: Selecting multiple sensor data variables representing the operating state of the system, including at least a vibration signal, a temperature signal, a load power and a system entropy value; Set weight parameters and weight each variable according to its influence on system stability; A Morse function is constructed to reflect the stability of the system's operating state and meet the differentiability condition. The construction form of the Morse function is a weighted combination of multiple sensor variables to ensure accurate representation of the system's state topological structure.

5. The method for predicting faults of a hydropower station computer monitoring system based on big data analysis according to claim 4 is characterized in that: The steps to construct the Morse-Smale complex include: Calculate the gradient vector of the Morse function and solve the gradient vanishing point to determine the critical point distribution of the Morse complex; Calculate the Hessian matrix and analyze the Morse index based on its eigenvalues ​​to determine the system stability under different states; The system state regions are divided based on the Morse index, where the stable region corresponds to the critical point where the Morse index is zero, the transition region corresponds to the critical point where the Morse index is one, and the unstable region corresponds to the critical point where the Morse index is greater than one; A Morse-Smale complex is constructed, and the topological structure of the system operation state is formed through the intersection of the stable manifold and the unstable manifold. The partition boundary of the Morse-Smale complex is adjusted according to the change of the system state.

6. The method for predicting faults of a hydropower station computer monitoring system based on big data analysis according to claim 1 is characterized in that: The step of calculating the fault discrimination score value comprises: Set a multi-dimensional fault feature vector, including the topological features of the Morse-Smale complex, the time series features, and the output probability value of the prediction model; The fault feature vector is converted into a fault score value by using a weighted nonlinear mapping method, wherein the fault discrimination score value is calculated as follows: ； in, is the weighting coefficient of the fault feature, is the corresponding eigenvalue, is a nonlinear mapping function; The weighting coefficient is adjusted based on historical operating data to improve the calculation accuracy of the fault score value.

7. The method for predicting faults in a hydropower station computer monitoring system based on big data analysis according to claim 6 is characterized in that: The fault alarm triggering threshold of the monitoring system adopts a dynamic adjustment strategy, including the following steps: Calculate the statistical distribution of the current fault discrimination score value and determine the mean and standard deviation within the set time window; Adopt an adaptive adjustment strategy to calculate the dynamic alarm threshold based on statistical characteristics, where the dynamic threshold is calculated as follows: ； in, is the mean of the rating values, is the standard deviation of the ratings, is the preset adjustment factor; When the fault discrimination score value exceeds the dynamic alarm threshold, the monitoring system alarm is triggered and an alarm log is recorded to optimize subsequent threshold settings.

8. A fault prediction device for a hydropower station computer monitoring system based on big data analysis, used to execute the method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module, used to collect multi-dimensional sensor data of the hydropower station computer monitoring system and pre-process the data; A topological analysis module, used to construct a Morse function based on the sensor data, calculate the gradient and the Hessian matrix, extract the topological features of the system operation state, and construct a Morse-Smale complex based on the topological features to divide the system operation state area; A fault prediction module is used to model the topological features of the Morse-Smale complex and time series data using a prediction model to predict the possibility of a fault; The discrimination and alarm module is used to calculate the fault discrimination score value according to the prediction result, and trigger the fault alarm of the monitoring system when the score value exceeds a preset threshold.

Images