Optimization method and device for hydraulic power plant unit equipment defect data identification process

Abstract

The invention discloses an optimization method and device for a hydraulic power plant unit equipment defect data recognition process, and relates to the field of big data, and the optimization method comprises the steps: installing a plurality of types of sensors at key parts of a unit, so as to achieve the real-time monitoring of the operation state of the unit equipment; training a fault diagnosis model based on an integrated machine learning algorithm; based on the real-time monitoring data and the trained fault diagnosis model, constructing an equipment defect diagnosis automatic decision-making system; and optimizing a unit maintenance strategy and spare part management by utilizing historical data accumulated by the defect diagnosis system. Through the scheme of the invention, limited annotation data can be expanded, so that the generalization ability of the model is enhanced, and the robustness of the system on the anti-interference ability is improved.

Description

Technical Field

[0001] The present application relates to the field of big data, and in particular to a method and device for optimizing a defect data identification process of a hydropower plant unit equipment. Background Art

[0002] The unit equipment of a hydropower plant is an important infrastructure of the national economy, and its stable operation is crucial to the safety of national electricity consumption, economic development and people's livelihood. However, in the long-term operation process, the unit equipment has the risk of various defects, such as wear, corrosion, aging of parts, etc. If these defects are not discovered and handled in time, they may cause equipment failure or even major accidents. Therefore, timely identification of unit equipment defects is particularly important for preventive maintenance and ensuring the safe operation of equipment. At present, the main technologies used for defect data identification of hydropower plant unit equipment include manual inspection, vibration monitoring and infrared thermal imaging detection. Manual inspection relies on direct observation by the human eye, and there are problems such as high work intensity and susceptibility to subjective factors. Vibration monitoring and infrared thermal imaging detection use sensors to monitor the operating status of the equipment, which can obtain more objective data. However, these technologies also have certain defects, such as lack of sufficient sensitivity to some potential defects and inability to fully cover all equipment components.

[0003] In order to improve the accuracy and coverage of defect identification of unit equipment, people have begun to explore the use of artificial intelligence technology to perform intelligent analysis of various monitoring data. By building a machine learning model, data features such as vibration and temperature can be automatically learned to identify abnormal patterns. In recent years, with the rise of deep learning technology, researchers have also tried to use advanced algorithms such as convolutional neural networks to learn high-dimensional data such as videos and images in order to better capture the microscopic features of equipment defects. However, existing artificial intelligence methods also face some challenges when applied to hydropower plant scenarios. The first is the lack of labeled data. Due to the complexity of the actual operating environment of hydropower plants, it is not easy to obtain complete and detailed defect labeling data. Secondly, due to the wide variety of equipment and the variety of various defect modes, the model needs to have a strong generalization ability to work effectively. Furthermore, the operating environment of hydropower plants is complex and changeable, and is easily affected by many interference factors, which puts the robustness of the model to the test.

[0004] Therefore, there is an urgent need for a technical solution that can expand the limited labeled data to enhance the generalization ability of the model and improve the robustness of the system in anti-interference ability. Summary of the invention

[0005] In order to solve the deficiencies of the prior art, the embodiment of the present application provides a method and device for optimizing the defect data identification process of a hydropower plant unit equipment. The present application solves the technical problems of the prior art such as insufficient generalization ability and poor anti-interference robustness.

[0006] The embodiment of the present application provides an optimization method for the defect data identification process of hydropower plant unit equipment, including: installing multiple types of sensors at key parts of the unit to achieve real-time monitoring of the operating status of the unit equipment; training a fault diagnosis model based on an integrated machine learning algorithm; building an equipment defect diagnosis automatic decision-making system based on real-time monitoring data and the trained fault diagnosis model; wherein, when the monitoring data is abnormal, the system automatically inputs relevant features into the diagnosis model to identify the defect type and severity, and outputs disposal suggestions; and utilizing the historical data accumulated by the defect diagnosis system to optimize the unit maintenance strategy and spare parts management.

[0007] In a possible implementation, multiple types of sensors are installed at key parts of the unit to achieve real-time monitoring of the operating status of the unit equipment, including: setting a vibration sensor to monitor the vibration level of bearings and impeller components; setting a temperature sensor to monitor the temperature rise of generator windings and main shaft parts; setting a pressure sensor to monitor the water pressure changes at the turbine inlet and outlet and tailwater pipe parts; setting an electrical parameter sensor to monitor indicators including the voltage, current and power of the generator; the sensor collects data at a preset frequency and uploads it to the monitoring center through a bus network; when the data exceeds the threshold range, the system automatically triggers an early warning prompt.

[0008] In a possible implementation, a fault diagnosis model is trained based on an integrated machine learning algorithm: based on historical operating data, key parameters reflecting the state of the unit are selected as modeling features; wherein the modeling features include generator stator winding temperature, generator rotor temperature, thrust bearing temperature, unit vibration values ​​at multiple measuring points, generator active power, generator voltage and generator current; characteristic parameter samples of the unit under various operating conditions are collected, and corresponding state categories including normal, stator winding fault or rotor friction are annotated to form a labeled sample set; multiple classifiers are trained based on the sample set to learn the mapping relationship from features to state categories; the output results of multiple classifiers are linearly integrated to obtain a fault diagnosis model.

[0009] In a possible implementation, based on real-time monitoring data and a trained fault diagnosis model, an automatic decision-making system for equipment defect diagnosis is constructed, including: when abnormal vibration or temperature data of the unit is monitored, the system inputs relevant features into the fault diagnosis model to obtain the fault type; based on the corresponding vibration spectrum and temperature distribution monitoring data, combined with the motor structure diagram and the expert knowledge base, the specific location of the defect is inferred; based on the current operating parameters, the target electrical formula is substituted to calculate the severity and development trend of the fault; based on the calculation results, the system outputs one or more optional maintenance strategies including "reduced load operation, waiting for opportunity maintenance" and "shutdown for emergency repair".

[0010] In a possible implementation, the historical data accumulated by the defect diagnosis system is used to optimize the unit maintenance strategy and spare parts management, including: establishing a remaining life prediction model for unit components based on historical operating data to perform predictive maintenance; and / or, statistically modeling the consumption patterns of wearing parts to predict the demand for various spare parts, and combining factors including procurement cost, warehousing cost and out-of-stock risk to determine the optimal inventory level of each spare part.

[0011] In a possible implementation, multiple classifiers are trained based on a sample set to learn the mapping relationship from features to state categories, including: constructing an optimal classification hyperplane in a high-dimensional feature space based on a Gaussian kernel; and integrating multiple decision trees through bagging to vote on sample categories.

[0012] In a possible implementation, establishing a prediction model for the remaining life of a unit component based on historical operation data to perform predictive maintenance includes: fitting the remaining life of a target component; wherein the fitting method includes: Where L(t) represents the remaining life prediction at time t, η represents the life characteristic parameter, m represents the life index, t represents the time index value, and C represents the constant calibration factor.

[0013] In one possible implementation, statistical modeling is performed on the consumption patterns of consumable parts to predict the demand for various spare parts, and factors including procurement cost, storage cost and out-of-stock risk are combined to determine the optimal inventory level of each spare part, including: Where s represents the optimal inventory level, C o represents the cost of a single order, D represents the annual demand, C h Indicates the annual storage cost of spare parts per unit.

[0014] An embodiment of the present application also provides an optimization device for a hydropower plant unit equipment defect data identification process, comprising: a processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, and when the instructions are executed by the processor, the processor executes the method described in the above embodiment.

[0015] In the optimization method and device for the defect data identification process of a hydropower plant unit equipment provided above, the embodiment of the present application can expand the limited labeled data through the setting of multiple sensors and multiple integrated classifiers to enhance the generalization ability of the model and improve the robustness of the system in anti-interference ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic diagram of a flow chart of a method for optimizing a hydropower plant unit equipment defect data identification process provided in an embodiment of the present application;

[0018] Figure 2 A flowchart of a method for integrating a machine learning algorithm provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application.

[0020] Those skilled in the art will appreciate that the terms "first", "second" and the like in the embodiments of the present application are only used to distinguish different steps, devices or modules, etc., and do not represent any specific technical meaning, nor do they represent the necessary logical order between them. It should also be understood that in the embodiments of the present application, "multiple" may refer to two or more, and "at least one" may refer to one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present application, in the absence of explicit limitation or contrary revelation given in the context, it can generally be understood as one or more. In addition, the term "and / or" in the present application is only a kind of association relationship describing the associated objects, indicating that there can be three kinds of relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present application generally indicates that the associated objects before and after are an "or" relationship. It should also be understood that the description of each embodiment in the present application emphasizes the differences between the embodiments, and the same or similar parts can refer to each other. For the sake of brevity, they will not be repeated one by one.

[0021] At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present application and its application or use. The techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the specification. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0023] Figure 1 The present invention provides a schematic diagram of a flow chart of a method for optimizing a hydropower plant unit equipment defect data identification process. Figure 1 As shown, in step S101, multiple types of sensors are installed at key locations of the unit to achieve real-time monitoring of the operating status of the unit equipment.

[0024] Specifically, in one embodiment, it includes: setting a vibration sensor to monitor the vibration level of bearings and impeller components; setting a temperature sensor to monitor the temperature rise of generator windings and main shaft parts; setting a pressure sensor to monitor the water pressure changes of turbine inlet and outlet and tailwater pipe parts; setting an electrical parameter sensor to monitor indicators including voltage, current and power of the generator; the sensor collects data at a preset frequency and uploads it to the monitoring center through the bus network; when the data exceeds the threshold range, the system automatically triggers an early warning prompt.

[0025] Furthermore, vibration sensors are installed on bearings and impeller components, and they can monitor the vibration levels of these components. Vibration analysis is a very critical aspect of fault diagnosis, because abnormal vibration of the unit is often a precursor to internal damage. By analyzing the frequency, amplitude and duration of vibration data, it is possible to predict whether components such as bearings and impellers have problems such as wear, looseness or imbalance.

[0026] Temperature sensors are used to monitor temperature rise in generator windings and main shafts. Abnormal temperature rise is a direct indicator of overload, cooling system failure or insulation damage. Continuous temperature monitoring helps to detect these problems in time to avoid further damage to the equipment.

[0027] Pressure sensors are installed at the turbine inlet and outlet and tailwater pipe to monitor changes in water pressure. Abnormal fluctuations in water pressure may indicate pipe blockage, reduced pump efficiency or sealing problems. The data from pressure sensors is crucial to ensure the normal operation of turbines.

[0028] Electrical parameter sensors monitor the voltage, current, power and other indicators of the generator. Abnormal changes in these electrical parameters can reflect abnormalities in the motor windings, magnetic field or circuit failures, which is crucial to maintaining the stable operation and efficiency of the motor.

[0029] These sensors collect data periodically at a preset frequency and upload the data to the monitoring center in real time through the bus network. The monitoring center has threshold settings. When the data sent back from the sensor exceeds these thresholds, the system will automatically trigger an early warning prompt. This automated early warning mechanism is set based on historical data analysis and fault pattern recognition. Threshold events may include a sharp increase in vibration frequency, an abnormal increase in temperature, a significant decrease in water pressure, or instability in electrical parameters. The early warning system allows operators to respond quickly and conduct further inspections and maintenance to prevent equipment failures and potential production stagnation. The monitoring center will analyze and process this data. When any monitoring parameter exceeds the preset threshold range, the system will automatically trigger an early warning prompt to remind the operation and maintenance personnel to promptly inspect and troubleshoot the cause of the fault. For example, the temperature of a phase winding of the generator of unit #1 suddenly increased, exceeding the rated value of 80℃ to 100℃. The temperature sensor detects the abnormality, the monitoring interface pops up a red warning, and a text message is sent to the operator on duty. In this way, measures can be taken as early as possible to troubleshoot the cause before the equipment is damaged. This optimization method achieves comprehensive, real-time and efficient monitoring of the equipment by deploying various sensors at key locations of the unit, which can minimize the risk of equipment failure, ensure the safe operation of the unit, and improve the overall power generation efficiency and economic benefits.

[0030] At step S102, the fault diagnosis model is trained based on the integrated machine learning algorithm. Specifically, the raw data collected from the sensor needs to be preprocessed first, including removing outliers, interpolating missing data, standardization, etc., to ensure the integrity and consistency of the data. Next, feature engineering is performed on the preprocessed data to extract or construct features that are meaningful to fault diagnosis from the raw data, such as statistical features (mean, variance, etc.), time-frequency domain features, wavelet transform features, etc. Reasonable feature engineering can provide more valuable input information for the model. The fault diagnosis model is trained using an integrated learning method. Integrated learning integrates the prediction results of multiple basic models (such as decision trees, support vector machines, etc.) by weight, and can usually obtain better generalization performance than a single model. Common integrated methods include Bagging, Boosting, Stacking, etc. In the model training process, the K-fold cross-validation method can be used to divide the data into a training set and a validation set, and the model performance is evaluated on the validation set to avoid overfitting and improve the generalization ability of the model. For different basic models and integration methods, it is necessary to adjust their hyperparameters (such as the depth of the decision tree, the kernel function of the SVM, etc.) to obtain the best performance. Common optimization methods include grid search, random search, etc. Finally, multiple optimized basic models are integrated into the final fault diagnosis model (such as weighted average, majority voting, etc.) according to certain rules. Figure 2 The specific method will be shown in Figure 2 (see the description for details).

[0031] In step S103, an automatic decision-making system for equipment defect diagnosis is constructed based on real-time monitoring data and a trained fault diagnosis model; wherein, when an abnormality occurs in the monitoring data, the system automatically inputs relevant features into the diagnosis model to identify the defect type and severity, and outputs disposal suggestions.

[0032] In one embodiment, it includes: when abnormal vibration or temperature data of the unit is monitored, the system inputs relevant features into the fault diagnosis model to obtain the fault type; based on the corresponding vibration spectrum and temperature distribution monitoring data, combined with the motor structure diagram and the expert knowledge base, the specific location of the defect is inferred; based on the current operating parameters, the target electrical formula is substituted to calculate the severity and development trend of the fault; based on the calculation results, the system outputs one or more optional maintenance strategies including "reduced load operation, waiting for opportunity maintenance" and "shutdown for emergency repair".

[0033] Take a hydroelectric generator set as an example: when the vibration and temperature data of the unit are detected to be abnormal, the system inputs the relevant feature x = [T s , T r , T t , V1, V2, ...] to the SVM model to obtain the fault type For example, "short circuit between stator winding turns." Further analysis of monitoring data such as vibration spectrum and temperature distribution, combined with the motor structure diagram and expert knowledge base, infers the specific location of the defect, such as "#2 generator B phase U1 coil 3rd turn." According to the current load P, vibration level V, insulation resistance R and other parameters, substitute them into the formula to calculate the severity and development trend of the fault, such as the short circuit current between winding turns: Among them U N is the rated phase voltage, R f is the short-circuit fault resistance, R i is the winding wire resistance. f The larger the value, the more serious the fault is and the faster it should be handled. The system provides 2-3 optional maintenance strategies, such as "reducing load operation and waiting for maintenance", "stopping for emergency repair", etc., and analyzes the impact of each plan on power generation and maintenance costs for decision makers to weigh.

[0034] The advantage of this automatic decision-making system is that it can realize the automation of intelligent fault diagnosis through the support of machine learning algorithms and knowledge bases, and conduct quantitative analysis based on mathematical models such as electrical formulas to provide an objective basis for decision-making; at the same time, it integrates expert experience rules into the system to improve the interpretability and reliability of diagnosis. The system is expected to improve the timeliness and scientificity of equipment operation and maintenance, and avoid economic losses caused by human misjudgment.

[0035] Furthermore, in another embodiment, the system first collects the operating data of the equipment in real time through sensors installed at key locations, such as physical parameters such as vibration, temperature, and pressure. These data are the basis for system operation analysis and are used to monitor the immediate status of the equipment. When abnormal fluctuations occur in the monitoring data, the system will automatically start the fault detection program. During the fault detection process, the system uses the abnormal data as input and passes it to a well-trained fault diagnosis model. The model is based on deep learning or machine learning algorithms and has been trained to identify the characteristics and patterns of various equipment failures. Through the analysis of the model, specific defect types and fault severity can be identified. Among them, the algorithms and calculation formulas involved in this diagnostic process can be, for example:

[0036] The fault type identification formula is: F(x) = W x + b, where F(x) represents the fault classification function, x is the input feature vector (such as vibration spectrum and temperature data), and W and b are the weight and bias parameters of the model respectively. Fault severity calculation: S = ∫(T(t)-T norm )dt, where T(t) represents real-time temperature data, T norm It is the temperature baseline under normal operating conditions, and S represents the accumulated temperature anomaly, which is used to assess the severity of the fault.

[0037] The precise reasoning system of the fault location can not only identify the fault type and assess its severity, but also use the vibration spectrum analysis and temperature distribution data, combined with the motor structure diagram of the equipment and the expert knowledge base, to accurately infer the specific location where the fault occurred. This step is assisted by the following formula: Vibration spectrum analysis formula: V(f) = F(v(t)), where v(t) represents the vibration signal in the time domain, V(f) is the frequency domain signal obtained by Fourier transform, and f represents the frequency.

[0038] Decision support and maintenance strategy Based on the type, severity and specific location of the fault, the system will automatically propose corresponding maintenance suggestions and maintenance strategies. This includes but is not limited to strategies such as reducing equipment load, waiting for maintenance or emergency shutdown for repair. The selection of these strategies depends on the following evaluation formula: Maintenance decision model: D = αS + βC, where D represents the maintenance decision index, S is the severity of the fault, C represents the maintenance cost coefficient, and α and β are model parameters used to adjust the weights of fault severity and maintenance cost.

[0039] In step S104, the historical data accumulated by the defect diagnosis system is used to optimize the unit maintenance strategy and spare parts management.

[0040] In one embodiment, based on historical operation data, a prediction model for the remaining life of a unit component is established to perform predictive maintenance, which includes: fitting the remaining life of the target component; wherein the fitting method includes:

[0041]

[0042] Among them, L(t) represents the remaining life prediction at time t, and η represents the life characteristic parameter, which is a characteristic quantity for calibrating the overall service life of the component, and is closely related to the design life, material properties and working environment of the component. This parameter is obtained through historical data analysis and experimental data statistics, and reflects the average performance of the component. m represents the life index, which describes the relationship between the failure mode of the component and time, and is an index that characterizes the aging rate of the component. The size of the life index can reflect the speed of component aging. The larger the value, the faster the life of the component decays. t represents the time index value, which is usually the time unit calculated from the start of the component operation. It can be hours, days or other time units, determined according to the specific equipment operating environment and component characteristics.

[0043] This formula can accurately describe the aging failure process of components over time. By fitting the historical operating data, the optimal values ​​of characteristic parameters such as η and m can be obtained, so as to predict the remaining life L(t) for a given time t. The fitting process can use classical techniques such as least squares method and maximum likelihood estimation method, or more advanced machine learning algorithms such as neural networks. The quality of fitting depends on the adequacy of historical data, the applicability of the formula, and the effectiveness of the estimation method. Once a reliable remaining life prediction model is obtained, the remaining life of the component can be predicted based on the current operating time or number of cycles. When the predicted value is lower than a certain threshold, the maintenance plan can be triggered in time to avoid failures. At the same time, the output of the model can also provide a basis for formulating maintenance strategies and purchasing spare parts.

[0044] The advantage of this method is that it uses a large amount of historical data to establish a systematic mathematical model, which has the characteristics of reasonable prediction and strong interpretability. However, it should also be noted that for special components or working conditions, it may be necessary to improve the formula form and introduce more influencing factors to improve the fitting accuracy.

[0045] In another embodiment, the consumption pattern of wearing parts is statistically modeled to predict the demand for various spare parts, and factors including procurement cost, storage cost and out-of-stock risk are combined to determine the optimal inventory level of each spare part. First, the core of spare parts demand forecasting is to establish a mathematical model through historical data to accurately calculate the demand for spare parts. This usually involves statistical methods such as time series analysis and regression analysis, so that future demand trends can be predicted based on historical consumption data. For example, exponential smoothing, ARIMA model (autoregressive integrated moving average model) or seasonal forecasting model can be used to analyze and predict the cyclical and trend changes in spare parts demand, including:

[0046]

[0047] Where s represents the optimal inventory level, C o represents the cost of a single order, which covers the administrative processing costs, transportation costs and ordering costs of each order. D represents the annual demand, which is the annual consumption calculated based on historical data. C hrepresents the annual storage cost per spare part, which includes cost factors such as storage space fees, insurance premiums, and loss risks. This formula is derived from classical inventory management theory and is based on some assumptions, such as known and constant demand, no delays in out-of-stock and replenishment, etc. It attempts to find a balance between ordering costs and inventory holding costs to minimize total costs. Excessive ordering costs will lead to frequent ordering and increase management burden; while excessive inventory holding costs mean excessive capital occupation. This formula multiplies the two and takes the square root, so that when the two change, the optimal inventory level will also adjust accordingly. In practical applications, it is also necessary to combine other factors such as out-of-stock risk and replenishment cycle to comprehensively determine the optimal inventory strategy. In addition, for some key spare parts, safety stock can be appropriately increased as a buffer.

[0048] In addition, you can also consider introducing correction factors to cope with the volatility and uncertainty of demand, so as to adjust the inventory level more accurately. For example, introducing service level targets and out-of-stock costs to form a comprehensive inventory strategy model that includes safety stock:

[0049]

[0050] Among them, s ′ represents the adjusted inventory level; Z is the normal distribution quantile corresponding to the service level; σ D is the standard deviation of demand, which is used to measure the volatility of demand. Furthermore, sensitivity analysis can be used to evaluate the impact of changes in various parameters on the optimal inventory level, such as C o , D and C h How changes in inventory affect the final inventory strategy. This analysis helps companies make flexible inventory adjustments under different market environments. In addition, combined with actual supply chain management practices, it is also necessary to consider multiple factors such as procurement cycle, supplier reliability, and spare parts substitutability, and to achieve more refined inventory management by establishing a more complex multi-factor dynamic optimization model.

[0051] For example, Monte Carlo simulation or stochastic optimization techniques can be combined to simulate inventory performance under different scenarios, thereby providing data support for decision makers. Ultimately, this method based on mathematical models and statistical analysis can help companies effectively balance procurement costs, warehousing costs and out-of-stock risks, and achieve overall optimization and cost control of the supply chain.

[0052] Figure 2 A schematic diagram of a method flow of integrating a machine learning algorithm provided in an embodiment of the present application. Figure 2As shown, at step S201, based on historical operating data, key parameters reflecting the state of the unit are selected as modeling features; wherein the modeling features include generator stator winding temperature, generator rotor temperature, thrust bearing temperature, unit vibration values ​​at multiple measuring points, generator active power, generator voltage and generator current.

[0053] In step S202, characteristic parameter samples of the unit under various operating conditions are collected and labeled with corresponding status categories including normal, stator winding fault or rotor friction to form a labeled sample set.

[0054] In step S203, multiple classifiers are trained based on the sample set to learn the mapping relationship from features to state categories, including: constructing an optimal classification hyperplane in a high-dimensional feature space based on a Gaussian kernel; integrating multiple decision trees through bagging to vote on sample categories.

[0055] In one embodiment, in order to improve the performance and generalization ability of the classifier, the Gaussian kernel in the kernel method can be used to process nonlinearly separable data. The kernel method is a technology in the support vector machine (SVM) that implicitly maps the data to a high-dimensional space by introducing a kernel function. In this high-dimensional space, data that was originally linearly inseparable in the original feature space may become linearly separable. The Gaussian kernel, also known as the radial basis function (RBF), is a kernel function, and its expression is K(x, x′)=exp(-γ||xx′|| 2 ), where γ is a positive hyperparameter that controls the width of the kernel function.

[0056] The support vector machine using Gaussian kernel first needs to select appropriate parameters γ and penalty parameter C, and then solve an optimization problem to find the optimal classification hyperplane that can maximize the boundary between categories. This optimization problem can be expressed as: The constraints here are: Where w is the normal vector of the hyperplane, b is the offset term, and ξ i is a slack variable that allows a certain amount of classification error, and φ is the mapping function induced by the Gaussian kernel.

[0057] In order to further improve the classification performance, an integrated learning strategy, such as bagging, can be used to integrate the power of multiple classifiers. Bagging, or Bootstrap Aggregating, creates multiple training sets by performing multiple sampling with replacement on the original data set. Each training set is used to train a base classifier, usually a decision tree. Each decision tree is trained independently of its corresponding sample set, and ultimately decides the category of the new sample by voting. Assume that M decision trees are trained. For a new sample x, each decision tree T m A prediction result T will be output m (x). The final classification result is determined by majority vote: Where 1 is the indicator function, if T m If (x)=k is true, the function value is 1, otherwise it is 0. This method improves the robustness and accuracy of the model by increasing the diversity of the classifier.

[0058] At step S204, the output results of multiple classifiers are linearly integrated to obtain a fault diagnosis model. First, the role and characteristics of each classifier need to be clarified. In a fault diagnosis system, each classifier is usually trained for a specific fault type or equipment state in order to capture the characteristics of different fault modes.

[0059] These classifiers may be based on different algorithms, such as support vector machines (SVM), decision trees, neural networks, etc., each of which exhibits different advantages and limitations in specific situations. The basic idea of ​​linear integration is to integrate the judgment results of multiple classifiers through certain mathematical methods, such as weighted average or voting mechanism, to form a comprehensive output. The determination of weights is usually based on the performance of each classifier in the past, such as its accuracy, recall rate or F1 score. These weights can be automatically adjusted through optimization algorithms, such as gradient descent, genetic algorithm, etc., to achieve the best integration effect.

[0060] The advantage of the ensemble method is that it can integrate the wisdom of multiple classifiers to make up for the shortcomings of a single classifier. For example, a classifier may show higher sensitivity to certain types of faults, while other classifiers may perform better on other types of faults. Through reasonable weight distribution and ensemble strategies, the final diagnostic model can maintain high accuracy and robustness in many situations.

[0061] In addition, linear integration can also prevent overfitting and improve the generalization ability of the model by introducing regularization terms. For example, you can consider adding L1 or L2 regularization.

[0062] Furthermore, an embodiment of the present application also provides an optimization device for a hydropower plant unit equipment defect data identification process, comprising: a processor, a memory, and a system bus; the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute any of the above methods.

[0063] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any one of the above-mentioned optimization methods.

[0064] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the optimization method described in each embodiment of the present application or some parts of the embodiments.

[0065] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0066] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0067] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the defect data identification process of a hydropower plant unit equipment, characterized in that: include: By installing multiple types of sensors at key locations of the unit, real-time monitoring of the unit's equipment operating status can be achieved; Train fault diagnosis models based on integrated machine learning algorithms; Based on real-time monitoring data and trained fault diagnosis models, an automatic decision-making system for equipment defect diagnosis is built. When abnormalities occur in monitoring data, the system automatically inputs relevant features into the diagnosis model to identify the defect type and severity, and outputs disposal suggestions. Utilize historical data accumulated by the defect diagnosis system to optimize unit maintenance strategies and spare parts management.

2. The optimization method according to claim 1, characterized in that: The real-time monitoring of the equipment operation status of the unit is achieved by installing multiple types of sensors at key locations of the unit, including: Vibration sensors are provided to monitor vibration levels of components including bearings and impellers; Temperature sensors are installed to monitor the temperature rise of the generator windings and main shaft. Pressure sensors are provided to monitor water pressure changes at the turbine inlet and outlet and the tailwater pipe; An electrical parameter sensor is provided to monitor indicators including voltage, current and power of the generator; The sensor collects data at a preset frequency and uploads it to the monitoring center through the bus network; when the data exceeds the threshold range, the system automatically triggers an early warning prompt.

3. The optimization method according to claim 1, characterized in that: Among them, the fault diagnosis model is trained based on the integrated machine learning algorithm: Based on historical operation data, key parameters reflecting the status of the unit are selected as modeling features; wherein the modeling features include generator stator winding temperature, generator rotor temperature, thrust bearing temperature, unit vibration values ​​at multiple measuring points, generator active power, generator voltage and generator current; Collect characteristic parameter samples under various operating conditions of the unit and label them with corresponding status categories including normal, stator winding fault or rotor collision to form a labeled sample set; Train multiple classifiers based on the sample set to learn the mapping relationship from features to state categories; The output results of multiple classifiers are linearly integrated to obtain a fault diagnosis model.

4. The optimization method according to claim 1, characterized in that: The construction of an automatic decision-making system for equipment defect diagnosis based on real-time monitoring data and trained fault diagnosis models includes: When abnormal vibration or temperature data of the unit is detected, the system inputs relevant features into the fault diagnosis model to obtain the fault type; Based on the corresponding vibration spectrum and temperature distribution monitoring data, combined with the motor structure diagram and expert knowledge base, the specific location of the defect can be inferred; Substitute the current operating parameters into the target electrical formula to calculate the fault severity and development trend; Based on the calculation results, the system outputs one or more optional maintenance strategies including "reduced load operation, waiting for maintenance" and "shutdown for emergency repair".

5. The optimization method according to claim 1, characterized in that: The use of historical data accumulated by the defect diagnosis system to optimize unit maintenance strategies and spare parts management includes: Based on historical operation data, a prediction model for the remaining life of unit components is established to perform predictive maintenance; and / or, Statistical modeling of the consumption patterns of wearing parts is used to predict the demand for various spare parts, and factors including procurement cost, warehousing cost and out-of-stock risk are combined to determine the optimal inventory level for each spare part.

6. The optimization method according to claim 1, characterized in that: Among them, multiple classifiers are trained based on the sample set to learn the mapping relationship from features to state categories, including: Construct the optimal classification hyperplane in high-dimensional feature space based on Gaussian kernel; Multiple decision trees are integrated through Bagging to vote on sample categories.

7. The optimization method according to claim 5, characterized in that: The remaining life prediction model of unit components is established based on historical operation data to carry out predictive maintenance, including: The remaining life of the target component is fitted; wherein the fitting method comprises: Where L(t) represents the remaining life prediction at time t, η represents the life characteristic parameter, m represents the life index, t represents the time index value, and C represents the constant calibration factor.

8. The optimization method according to claim 5, characterized in that: The consumption patterns of wearing parts are statistically modeled to predict the demand for various spare parts, and the optimal inventory level of each spare part is determined by combining factors including procurement cost, storage cost and out-of-stock risk. Where s represents the optimal inventory level, C o represents the cost of a single order, D represents the annual demand, C h Indicates the annual storage cost of spare parts per unit.

9. An optimization device for the defect data identification process of a hydropower plant unit, characterized in that: include: A processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; The memory is used to store one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by the processor, the processor executes the method of claims 1-8.

Images