Generator fault-based repetitive defect prediction method and system

Abstract

The invention discloses a repetitive defect prediction method and system based on a generator fault, and belongs to the field of fault prediction. The method comprises the steps that when a generator fault prediction model is trained and a preset fitting degree standard is reached, it is determined that training of the generator fault prediction model is completed; determining a repetitive defect occurrence rate of the generator, if the repetitive defect occurrence rate is higher than a preset repetitive defect occurrence rate threshold, determining that the generator has a risk of occurrence of repetitive defects, and performing clustering analysis on the real-time generator operation information, the real-time operation information and the real-time environment information to obtain a real-time clustering result; and determining a target fault mode according to the real-time clustering result and a preset fault mode list. According to the scheme, potential problems can be found in advance before the generator actually breaks down, and the reliability and operation safety of the generator are improved. By predicting and identifying repetitive defects, unnecessary comprehensive inspection or excessive maintenance can be avoided, the maintenance cost is reduced, and the maintenance efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault prediction, and in particular to a method and system for predicting repetitive defects based on generator faults. Background Art

[0002] With the rapid development of industry, generators, as the core equipment of power supply, play a vital role in the power system. However, various faults often occur in the operation of generators, among which repetitive defects are a particularly prominent problem, that is, the generator has the same fault many times under the same or similar working conditions and environment. This defect will affect the normal operation of the generator and have a serious impact on industrial production.

[0003] Traditional prediction of repetitive defects in generators mainly relies on human analysis and experience-based judgment. During the operation of the generator, staff will conduct real-time monitoring and manual inspections to observe the operating status of the generator. If any abnormality is found, they will combine previous experience and rules to determine whether there are potential repetitive defects.

[0004] However, due to the increasing complexity of the structure and function of the generator, relying on the experience and judgment of the staff may reduce the accuracy and reliability of the prediction results. In addition, the efficiency of manual prediction of repetitive defects of the generator is also low, and the cost of manual prediction is also high. Summary of the invention

[0005] The present application provides a method and system for predicting repetitive defects based on generator faults, which solves the problem that the structure and function of today's generators are becoming increasingly complex, and relying on the experience and judgment of the staff may reduce the accuracy and reliability of the prediction results. In addition, the efficiency of manual prediction of repetitive defects of generators is also low, and the cost of manual prediction is also high.

[0006] According to a first aspect of the present application, a method for predicting repetitive defects based on generator faults is provided, the method comprising:

[0007] Obtain historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and train the generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0008]

[0009] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability;

[0010] Calculating the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, determining that the generator fault prediction model training is completed;

[0011] Acquire real-time generator operation information, real-time operation information and real-time environmental information, input the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determine the occurrence rate of repetitive defects of the generator;

[0012] If the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, it is determined that the generator is at risk of repetitive defects, and a cluster analysis is performed on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain a real-time clustering result, and a target fault mode is determined based on the real-time clustering result and a preset fault mode list.

[0013] Further, after training the generator fault prediction model according to the historical generator operation information, the corresponding historical operation information, the corresponding historical environment information and a preset generator fault prediction model construction formula, the method further includes:

[0014] If the model fit of the generator fault prediction model does not meet the preset model fit standard, sample balancing is performed on the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information according to the preset fault mode list to obtain a sample balance result;

[0015] The generator fault prediction model is trained according to the sample balancing result and a preset generator fault prediction model construction formula until the model fit of the generator fault prediction model reaches a preset model fit standard.

[0016] Further, the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information are sample balanced according to the preset fault mode list to obtain a sample balance result, including:

[0017] Determine sample imbalance information according to the preset fault mode list, the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information; determine the target sample quantity and target sample information of the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information corresponding to each fault mode according to the sample imbalance information and the preset sample balancing method;

[0018] A sample balance result is determined according to the target sample quantity and the target sample information.

[0019] Furthermore, the process of determining the failure mode list includes:

[0020] Obtain historical generator operation information, corresponding historical operation information, corresponding historical environment information and corresponding failure mode when the generator fails, perform cluster analysis on the historical generator operation information, corresponding historical operation information and corresponding historical environment information under each failure mode, and obtain historical clustering results corresponding to each failure mode;

[0021] A failure mode list is determined according to the failure mode and the historical clustering result.

[0022] Further, determining a target failure mode according to the real-time clustering result and a preset failure mode list includes:

[0023] Calculating the Mahalanobis distance between the real-time clustering result and each historical clustering result in the preset fault mode list, and determining the historical clustering result corresponding to the smallest Mahalanobis distance as the target historical clustering result;

[0024] The failure mode corresponding to the target historical clustering result is determined as the target failure mode.

[0025] Furthermore, after determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes:

[0026] Determining whether the target failure mode meets a preset remote processing standard;

[0027] If the preset remote processing standard is met, a corresponding remote processing method is matched in the preset remote processing standard according to the target fault mode, and remote fault processing is performed on the generator according to the remote processing method.

[0028] Further, after determining whether the target failure mode meets a preset remote processing standard, the method further includes:

[0029] If the preset remote processing standard is not met, the target fault mode is sent to the handheld terminal of the staff, so that the staff can perform on-site fault processing on the generator according to the target fault mode.

[0030] Furthermore, after determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes:

[0031] Determine in real time whether the performance data of the generator fault prediction model meets the preset performance standard; wherein the performance standard includes the accuracy standard and the recall rate standard;

[0032] If the performance data of the generator fault prediction model does not meet the preset performance standard, all first generator operation information, corresponding first operation information and corresponding first environment information when the generator fails are obtained;

[0033] The generator fault prediction model is retrained according to the first generator operation information, the corresponding first operation information, the corresponding first environmental information and a preset generator fault prediction model construction formula until the model fit of the generator fault prediction model reaches a preset model fit standard.

[0034] Furthermore, after determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes:

[0035] regularly collecting the second generator operation information, the corresponding second operation information, and the corresponding second environment information when the engine fails according to a preset model update time interval;

[0036] The generator fault prediction model is retrained according to the second generator operation information, the corresponding second operation information, the corresponding second environment information and a preset generator fault prediction model construction formula until the model fit of the updated generator fault prediction model reaches a preset model fit standard.

[0037] According to a second aspect of the present application, a repetitive defect prediction system based on generator faults is provided, characterized in that the system comprises:

[0038] The model training module is used to obtain historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and train the generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0039]

[0040] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability;

[0041] A standard comparison module is used to calculate the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, it is determined that the training of the generator fault prediction model is completed;

[0042] A repetitive defect prediction module is used to obtain real-time generator operation information, real-time operation information and real-time environmental information, input the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determine the repetitive defect occurrence rate of the generator;

[0043] A fault mode determination module is used to determine that the generator is at risk of repetitive defects if the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, perform cluster analysis on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain real-time clustering results, and determine a target fault mode based on the real-time clustering results and a preset fault mode list.

[0044] In the embodiment of the present application, historical generator operation information, corresponding historical operation information and corresponding historical environment information are obtained when the generator fails, and the generator fault prediction model is trained according to the historical generator operation information, the corresponding historical operation information, the corresponding historical environment information and the preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0045]

[0046] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X condition; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability; the model fit of the generator fault prediction model is calculated, and if the model fit reaches the preset model fit standard, the generator is determined to be The fault prediction model training is completed; the real-time generator operation information, real-time operation information and real-time environmental information are obtained, and the real-time generator operation information, the real-time operation information and the real-time environmental information are input into the generator fault prediction model to determine the repetitive defect incidence rate of the generator; if the repetitive defect incidence rate is higher than the preset repetitive defect incidence rate threshold, it is determined that the generator has the risk of repetitive defects, and the real-time generator operation information, real-time operation information and real-time environmental information are clustered and analyzed to obtain real-time clustering results, and the target fault mode is determined according to the real-time clustering results and the preset fault mode list. By setting the fault prediction model based on the above-mentioned repetitive defect prediction method of generator fault, potential problems can be discovered in advance before the generator actually fails, thereby improving the reliability and operation safety of the generator. And by predicting and identifying repetitive defects, unnecessary comprehensive inspections or excessive repairs can be avoided, thereby reducing maintenance costs and improving maintenance efficiency.

[0047] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present application, nor are they intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0049] Figure 1 It is a flow chart of a method for predicting repetitive defects based on generator faults provided in Example 1 of the present application;

[0050] Figure 2 It is a flow chart of a method for predicting repetitive defects based on generator faults provided in the second embodiment of the present application;

[0051] Figure 3It is a flow chart of the method for predicting repetitive defects based on generator faults provided in the third embodiment of the present application.

[0052] Figure 4 It is a structural diagram of a repetitive defect prediction system based on generator faults provided in Example 4 of the present application. DETAILED DESCRIPTION

[0053] 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.

[0054] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0055] The following is a detailed description of the repetitive defect prediction method based on generator faults provided by the embodiment of the present application through specific embodiments and application scenarios in conjunction with the accompanying drawings.

[0056] Embodiment 1

[0057] Figure 1 is a flow chart of a method for predicting repetitive defects based on generator faults provided in the first embodiment of the present application. Figure 1 As shown, the specific steps include:

[0058] S101, obtaining historical generator operation information, corresponding historical operation information and corresponding historical environment information when a generator fails, and training a generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0059]

[0060] Among them, P(Y=1|X) is the probability of generator failure under the condition of given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability.

[0061] First of all, the usage scenario of this solution can be to train a generator fault prediction model based on the historical fault data of the generator, and use this model to predict the repetitive defects of the generator, and determine the failure mode of the generator when the repetitive defect incidence rate is higher than the preset repetitive defect incidence rate threshold.

[0062] Based on the above usage scenarios, it can be understood that the executor of the present application can be a repetitive defect prediction system based on generator faults that integrates the functions of training generator fault prediction models, determining the repetitive defect incidence rate of generators, and determining fault modes. No excessive limitations are made here.

[0063] In this solution, the historical generator operation information can be the operating status and performance indicators when the generator failure occurs. Specifically, it can include voltage, current, power, speed, temperature, generator operation time and load data, etc. This information records the operating conditions and working status of the generator when the failure occurs.

[0064] The corresponding historical operation information may include the operator's operation records, control instructions, and maintenance activities, etc., which can reflect the impact of human factors on the generator status.

[0065] The corresponding historical environmental information can describe the environmental conditions when the generator is running. Specifically, it can include ambient temperature, ambient humidity, light, air pressure, noise, and air quality. Environmental information can affect the performance and health of the generator.

[0066] The preset generator fault prediction model building formula can be a mathematical formula or algorithm used to train the generator fault prediction model. Usually, these formulas are based on historical data and use machine learning or statistical techniques to capture the relationship between generator faults and operation, operation and environment. In this solution, the generator fault prediction model building formula is:

[0067]

[0068] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability;

[0069] The influence coefficient of each parameter represents the degree of influence of each type of historical data on the probability of generator failure. These coefficients can be estimated using methods such as maximum likelihood estimation. These coefficients represent the degree of influence of each feature on the probability of generator failure. The process of training the model can include using known historical data, i.e., historical generator operation information, historical operation information, and historical environmental information, to fit the model so that it can predict whether the generator will fail under given input features. The process of model training is to adjust these coefficients so that the model can best match the known historical data and perform well on unknown data.

[0070] The generator fault prediction model may be a model constructed through training data, and is used to predict possible faults of the generator.

[0071] The historical generator operation information can be obtained from the sensors of the generator. Different operation information can be obtained using corresponding sensors, such as voltage sensors, current sensors, and temperature sensors. The corresponding historical operation information is obtained from the log records, and the historical environment information is obtained according to the environment sensor. Specifically, different environment types can be obtained using corresponding sensors, such as using a temperature sensor to obtain the environment temperature in the generator operation environment, and using a humidity sensor to obtain the environment humidity. Then the historical generator operation information, historical operation information, and historical environment information are integrated into a data set to ensure that each sample contains complete feature information. The generator fault prediction model is trained using the integrated data set and the preset generator fault prediction model construction formula. Specifically, each influence coefficient can be set to an initial value first, which can be initialized randomly or using some heuristic methods. Using the initialized parameters, the predicted probability value of each sample is calculated according to the preset generator fault prediction model construction formula. These predicted values ​​represent the probability of the generator failure under given input features. According to the predicted probability value and the actual label, the loss function of the model is calculated. Specifically, the logarithmic loss function can be used as the loss function of the generator fault prediction model. Then, the influencing parameters of the model are adjusted by maximizing the likelihood estimation or minimizing the loss function. Specifically, optimization algorithms such as gradient descent can be used to achieve this. For example, first set an initial value for each influencing parameter β0, β1, β2, and β3, use the initialized parameters, and build a formula to calculate the probability of generator failure for each sample according to the preset generator failure prediction model. For a given sample i, the loss function can be calculated using the following formula:

[0072]

[0073] Among them, y i is the actual label of the sample (0 or 1), is the corresponding predicted probability value. For the entire training set, the loss function of all samples can be calculated and averaged as the overall loss. Then, an optimization algorithm such as gradient descent can be used to update the influencing parameters of the model according to the gradient of the loss function.

[0074] On the basis of the above technical solution, optionally, after training the generator fault prediction model according to the historical generator operation information, the corresponding historical operation information, the corresponding historical environment information and the preset generator fault prediction model construction formula, the method further includes:

[0075] If the model fit of the generator fault prediction model does not meet the preset model fit standard, sample balancing is performed on the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information according to the preset fault mode list to obtain a sample balance result;

[0076] The generator fault prediction model is trained according to the sample balancing result and a preset generator fault prediction model construction formula until the model fit of the generator fault prediction model reaches a preset model fit standard.

[0077] In this solution, the sample balancing result can be an adjusted data set, so that the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information corresponding to each fault mode are more balanced in the data set. Through sample balancing, the model bias caused by uneven data distribution can be reduced or eliminated, so that the model can learn the information of each category, each time period and each feature more comprehensively. Specifically, it can include sample data balance, data diversity retention and feature distribution balance. For example, if there are four modes A, B, C and D in the preset fault mode list, in the original data set, the number of samples of fault mode A is large, while the number of samples of fault modes B, C and D is relatively small. The goal is to balance the number of samples of these four fault modes. Then the sample balancing result can include sample quantity balance, that is, after the balancing process, the number of samples of the four fault modes A, B, C and D is now roughly equal, then each fault mode has the same number of representatives in the data set, thereby avoiding the model bias towards the fault mode with a larger number. Data diversity preservation, that is, in the balancing process, the data diversity of each fault mode is preserved, so the balanced data set still contains a variety of different operation information, environmental information and generator operation information to reflect various situations in actual operation. Feature distribution balance, that is, in the balanced data set, the feature distribution of various fault modes should be relatively balanced, and no feature is too concentrated or missing in a certain fault mode.

[0078] The number of samples of each fault mode can be analyzed according to the preset fault mode list. Specifically, it can include query operations on the database or data set to find the records corresponding to each fault mode. It can be determined which fault modes have too many samples and which have too few samples, so as to identify the situations that need to be balanced. According to the analysis results of the number of samples, a specific sample balancing strategy can be formulated. For fault modes with too many samples, undersampling methods can be used, such as randomly reducing samples or reducing representative repeated samples based on clustering and other methods. For fault modes with too few samples, oversampling methods can be used, such as copying existing samples or synthesizing new samples using methods such as SMOTE. According to the formulated balancing strategy, actual sample balancing operations are performed on the historical generator operation information, the corresponding historical operation information, and the corresponding historical environment information. Specifically, it can include deleting, copying or synthesizing samples to ensure that the number of samples for each fault mode is balanced.

[0079] After the sample balance result is determined, the balanced historical generator operation information, the corresponding historical operation information, and the corresponding historical environment information can be determined from the sample balance result, and then these parameters are input into the preset generator fault prediction model construction formula. The predicted probability value of each sample is calculated according to the preset generator fault prediction model construction formula using the parameters initialized for the model. These predicted values ​​represent the probability of the generator failing under given input features. According to the predicted probability value and the actual label, the loss function of the model is calculated. Specifically, the logarithmic loss function can be used as the loss function of the generator fault prediction model. Then, the parameters of the model are adjusted by maximizing the likelihood estimate or minimizing the loss function. Specifically, this can be achieved using optimization algorithms such as gradient descent, and the above steps are repeated until the model fit of the generator fault prediction model reaches the preset model fit standard.

[0080] In this scheme, the problem of category imbalance in the data set can be solved through sample balancing, so that the model can more comprehensively learn the fault characteristics of each category and under each condition, thereby improving the prediction accuracy of the model and enabling it to more accurately predict the fault condition of the generator.

[0081] On the basis of the above technical solution, optionally, sample balancing is performed on the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information according to the preset fault mode list to obtain a sample balancing result, including:

[0082] Determine sample imbalance information according to the preset fault mode list, the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information; determine the target sample quantity and target sample information of the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information corresponding to each fault mode according to the sample imbalance information and the preset sample balancing method;

[0083] A sample balance result is determined according to the target sample quantity and the target sample information.

[0084] In this solution, sample imbalance information can be information about the difference in the number of samples of different fault modes in the data set, the difference in feature distribution between fault modes, and the inconsistency of data quality. For example, the number of samples of some fault modes may be much larger than that of other modes, causing the model to be too biased towards those categories with a large number during training, thereby ignoring the categories with a small number, resulting in poor performance of the model in predicting minority class faults. The data set consists of historical generator operation information, corresponding historical operation information, and corresponding historical environment information.

[0085] The preset sample balancing method can be a series of predetermined strategies to solve the sample imbalance problem, which aims to adjust the number of samples of each category in the data set so that the model can consider all categories more evenly during training. Specifically, it can include oversampling, undersampling, data synthesis, and using different weights or penalty functions to balance the losses of different categories.

[0086] A failure mode may be a specific symptom or characteristic that a generator exhibits when it fails.

[0087] The target sample number may refer to the number of samples of each failure mode that is expected to be achieved after the sample balancing operation. Specifically, this number may be determined based on the overall size of the data set, the performance requirements of the model, and the relative importance between categories.

[0088] The target sample information may refer to the key information and features that the sample expected to be generated or adjusted should contain during the sample balancing process. In this solution, the target sample information may be composed of the historical generator operation information, the corresponding historical operation information, and the corresponding historical environment information screened out after balancing.

[0089] According to the preset list of fault modes, historical generator operation information, operation information and environmental information related to these fault modes can be extracted from the historical database. For each fault mode, the corresponding number of samples can be counted. By comparing the number of samples of different fault modes, it is possible to identify which fault modes have too many samples and which have too few samples. Specifically, the imbalance of the samples can be calculated, for example, by measuring the ratio of the number of majority class samples to the number of minority class samples. The higher the imbalance, the more serious the sample imbalance problem. Then determine the total sample size of the data set, and use the preset sample balancing method to allocate a relatively balanced number of samples for each fault mode and determine the sample information according to business needs and model performance requirements. Specifically, it can include oversampling using the SMOTE algorithm, undersampling using cluster undersampling, or combining multiple methods for balancing. Finally, count the number of samples and sample information for each fault mode and combine them into the target sample size and target sample information.

[0090] After the target sample quantity and target sample information are determined, the target sample quantity and target sample information are combined to form a sample balance result.

[0091] In this scheme, by balancing the number of samples and sample information of different failure modes, the model can learn the characteristics of various failure modes more evenly, thereby improving the overall prediction performance.

[0092] S102, calculating the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, determining that the training of the generator fault prediction model is completed.

[0093] Model fit can refer to the degree to which the model fits the data. Specifically, it can be measured using a loss function. The smaller the value of the loss function, the better the model fits the data.

[0094] The preset model fit standard can be the optimization target in the training process, that is, by adjusting the model parameters, the prediction results of the model on the training data are made to match the actual labels as much as possible. In this solution, when using the loss function for model training, the value of the loss function can be minimized. Specifically, a threshold of the loss function value can be set. When the loss function value of the model drops below the threshold, it is considered that the model has reached the preset fit standard.

[0095] The model fit can be calculated using the following formula:

[0096]

[0097] Where N represents the input sample size, that is, the total number of samples involved in calculating the loss function; y irepresents the category label of the i-th sample; p i It represents the probability that the model predicts the i-th sample as a positive class, which is between 0 and 1, and represents the model's confidence that the sample belongs to the positive class.

[0098] In y i log(p i ), when y i When 1, this term calculates the logarithm of the probability that the model predicts the positive class and adds it to the loss. i )log(1-p i ), when y i When it is 0, this item calculates the logarithm of the probability that the model predicts a negative class and adds it to the loss. The value of the loss function can be calculated after each iteration. When the loss function value drops below the threshold of the loss function value, the model is considered to have reached the preset fitness standard.

[0099] When the model fit of the generator fault prediction model reaches the preset model fit standard, it can be determined that the generator fault prediction model training is completed. Specifically, the model fit can be evaluated by observing whether the loss function value of the model on the training data has reached the expected minimum value. When the model reaches the preset fit standard, it can be determined that the model training is completed.

[0100] S103, acquiring real-time generator operation information, real-time operation information and real-time environmental information, inputting the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determining the occurrence rate of repetitive defects of the generator.

[0101] The real-time generator operation information may be the current operating status and current performance indicators of the generator, and specifically may include voltage, current, power, speed, temperature, generator operation time, and load data.

[0102] Real-time operation information may include the current operator's operation records, control instructions, and maintenance activities.

[0103] Real-time environmental information can describe the real-time status of the environment around the generator, specifically, it can include ambient temperature, ambient humidity, light, air pressure, noise, and air quality.

[0104] The repetitive defect occurrence rate can be the frequency of repetitive defects occurring in the generator within a certain period of time. Specifically, it can be expressed as a percentage or frequency.

[0105] The real-time generator operation information can be collected using the generator's sensors, and the real-time operation information can be obtained through real-time updated log records. The real-time environment information around the generator can be monitored using environmental monitoring equipment, and the real-time generator operation information, real-time operation information, and real-time environment information can be input into the generator fault prediction model. According to the prediction results output by the model, the repetitive defect incidence rate of the generator is determined. Specifically, this can be a probability value, which indicates the possibility of repetitive defects in the generator under given conditions, or a frequency value, which indicates the number of repetitive defects in the generator within a certain period of time. For example, for a certain bearing of the engine, the vibration amplitude and frequency of the bearing can be collected through a vibration sensor. The number of times the operator starts and stops the generator and the load adjustment operations are obtained. Parameters such as ambient temperature and humidity are monitored, and the above data are transmitted to the fault prediction model. The prediction model analyzes the real-time data and outputs the probability of bearing failure. If the historical records show that the incidence of bearing failure is high under the same operating conditions, the prediction model will improve the probability prediction of bearing failure and output the corresponding repetitive defect incidence rate.

[0106] S104: If the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, it is determined that the generator is at risk of repetitive defects, and a cluster analysis is performed on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain a real-time clustering result, and a target fault mode is determined based on the real-time clustering result and a preset fault mode list.

[0107] The preset repetitive defect occurrence rate threshold may be a preset value used to determine whether the generator has a risk of repetitive defects. When the actually measured repetitive defect occurrence rate exceeds the threshold, it is considered that the generator has a risk of repetitive defects.

[0108] The real-time clustering results can be obtained by clustering the real-time generator operation information, real-time operation information, and real-time environment information. Cluster analysis is an unsupervised learning method that classifies similar objects into one category to discover structures and patterns in the data. In this solution, the purpose of cluster analysis can be to extract data points with similar characteristics from a large amount of real-time information to form different categories or groups. These clustering results can reflect the current operating status of the generator and may include possible abnormalities or failure modes.

[0109] The preset fault mode list may be a predefined set of fault modes, specifically, it may include various fault modes that may occur in the generator, specifically, it may include electrical faults, including winding short circuit or open circuit, which may be caused by insulation aging and mechanical damage or overheating. Overvoltage or overcurrent may be caused by power fluctuations, load changes or control system failures. Generator demagnetization may be caused by excitation system failure or field winding problems. Mechanical faults, including bearing damage or wear, that is, bearings may wear after long-term operation, resulting in increased vibration and reduced efficiency of the generator. Rotor imbalance, that is, caused by manufacturing defects, uneven materials or unbalanced mass accumulated during operation. Coupling failure, that is, the coupling connects the generator and other equipment. If the connection is poor or worn, it will cause vibration and power transmission problems. Cooling system failure, including insufficient cooling water flow or excessive temperature, will affect the heat dissipation effect of the generator and cause excessive temperature rise. Cooler blockage or leakage will reduce cooling efficiency, thereby affecting the operation of the generator. Control system failure, that is, malfunction of protection devices, such as malfunction of overcurrent protection and undervoltage protection, will cause the generator to shut down. Control system failure, such as PLC and DCS control system failure, will cause the generator to not operate normally. Lubrication system failure, including insufficient or contaminated lubricating oil, which will affect the lubrication effect of the moving parts inside the generator and increase wear. Lubricating oil pump failure, if the oil pump does not work properly, will cause poor lubrication.

[0110] The target fault mode can be determined based on the real-time clustering result and a preset fault mode list. When one or some categories in the real-time clustering result match a fault mode in the preset fault mode list, the fault mode is considered to be the target fault mode.

[0111] The determined repetitive defect rate of the generator can be compared with the preset repetitive defect rate threshold. If it is higher than the threshold, the real-time generator operation information, real-time operation information and real-time environmental information are clustered and analyzed. Specifically, similar data points can be classified into one category to find the structure and pattern in the data. Appropriate clustering algorithms can be selected, such as K-means and hierarchical clustering, and the number or conditions of clusters can be determined. Through cluster analysis of real-time information, real-time clustering results can be obtained, which reflect the current operating status of the generator and possible abnormalities or failure modes. Then, the real-time clustering results are matched with the preset failure mode list to find the most matching failure mode. If the real-time clustering results are highly matched with one or some preset failure modes, then it can be considered to be the target failure mode that the generator is currently experiencing.

[0112] In the embodiment of the present application, historical generator operation information, corresponding historical operation information and corresponding historical environment information are obtained when the generator fails, and the generator fault prediction model is trained according to the historical generator operation information, the corresponding historical operation information, the corresponding historical environment information and the preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0113]

[0114] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X condition; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability; the model fit of the generator fault prediction model is calculated, and if the model fit reaches the preset model fit standard, the generator is determined to be The fault prediction model training is completed; the real-time generator operation information, real-time operation information and real-time environmental information are obtained, and the real-time generator operation information, the real-time operation information and the real-time environmental information are input into the generator fault prediction model to determine the repetitive defect incidence rate of the generator; if the repetitive defect incidence rate is higher than the preset repetitive defect incidence rate threshold, it is determined that the generator has the risk of repetitive defects, and the real-time generator operation information, real-time operation information and real-time environmental information are clustered and analyzed to obtain real-time clustering results, and the target fault mode is determined according to the real-time clustering results and the preset fault mode list. By setting the fault prediction model based on the above-mentioned repetitive defect prediction method of generator fault, potential problems can be discovered in advance before the generator actually fails, thereby improving the reliability and operation safety of the generator. And by predicting and identifying repetitive defects, unnecessary comprehensive inspections or excessive repairs can be avoided, thereby reducing maintenance costs and improving maintenance efficiency.

[0115] Based on the above technical solution, optionally, the process of determining the failure mode list includes:

[0116] Obtain historical generator operation information, corresponding historical operation information, corresponding historical environment information and corresponding failure mode when the generator fails, perform cluster analysis on the historical generator operation information, corresponding historical operation information and corresponding historical environment information under each failure mode, and obtain historical clustering results corresponding to each failure mode;

[0117] A failure mode list is determined according to the failure mode and the historical clustering result.

[0118] In this solution, the historical clustering result can be a group or category obtained by clustering the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information when the generator fails. Each category represents a collection of similar failure modes and related information.

[0119] Historical data of generator failures can be collected, including historical generator operation information, historical operation information, and historical environmental information. Key features related to the failure mode can be extracted from the collected data. These features can reflect the changes in the generator's operating status, operating conditions, and environmental conditions. Then, a suitable clustering algorithm is selected, such as K-means, hierarchical clustering, and DBSCAN. The selected clustering algorithm is used to perform cluster analysis on the extracted features. Specifically, the algorithm can divide the samples into different categories based on the similarity between the data, and combine the categories into historical clustering results.

[0120] Each category in the historical clustering results can be matched with a known fault mode. Specifically, this can be achieved by comparing the feature distribution, sample description, and related performance of the fault mode in the clustering results. If the characteristics of a clustering result are highly consistent with the description of a known fault mode, then it can be considered that this clustering result represents this fault mode. Finally, the historical clustering results are combined with the corresponding fault modes to form a fault mode list. In the fault mode list, each fault mode and historical clustering result can be expressed as a fault mode-corresponding historical clustering result. For example, a sample of a clustering result mainly shows abnormal voltage drop and current instability, accompanied by a trend of temperature increase. These features are consistent with the stator winding fault in the fault mode, so it is determined that the fault mode corresponding to this clustering result is a stator winding fault.

[0121] In this solution, the failure modes in the list are obtained based on historical data and cluster analysis, thus improving the accuracy and reliability of practical applications.

[0122] On the basis of the above technical solution, optionally, determining the target failure mode according to the real-time clustering result and a preset failure mode list includes:

[0123] Calculating the Mahalanobis distance between the real-time clustering result and each historical clustering result in the preset fault mode list, and determining the historical clustering result corresponding to the smallest Mahalanobis distance as the target historical clustering result;

[0124] The failure mode corresponding to the target historical clustering result is determined as the target failure mode.

[0125] In this scheme, the Mahalanobis distance can represent the distance between a point and a distribution, or the covariance distance of the data.

[0126] The target historical clustering result may refer to a historical clustering result having a minimum Mahalanobis distance with the real-time clustering result.

[0127] The target failure mode may be a failure mode corresponding to a target historical clustering result.

[0128] For the real-time clustering result, the Mahalanobis distance between it and each historical clustering result can be calculated. Specifically, it can be calculated by the following formula:

[0129] D_M(x)=sqrt[(x-μ)′∑^(-1)(x-μ)];

[0130] Where x represents the eigenvector of the real-time clustering result; D_M(x) represents the Mahalanobis distance of the data point x; μ is the mean of the historical clustering result; Σ is the covariance matrix; Σ^(-1) is the inverse matrix of the covariance matrix; (x-μ)' represents the transpose of (x-μ).

[0131] After calculating all Mahalanobis distances, the minimum value can be found, and the corresponding historical clustering result is the target historical clustering result. Finally, the fault mode corresponding to the target historical clustering result can be found from the fault mode list as the target fault mode.

[0132] In this scheme, by using the Mahalanobis distance, the similarity between the real-time clustering results and the historical clustering results can be measured more accurately, thereby improving the accuracy and reliability of fault identification.

[0133] Based on the above technical solution, optionally, after determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes:

[0134] regularly collecting the second generator operation information, the corresponding second operation information, and the corresponding second environment information when the engine fails according to a preset model update time interval;

[0135] The generator fault prediction model is retrained according to the second generator operation information, the corresponding second operation information, the corresponding second environment information and a preset generator fault prediction model construction formula until the model fit of the updated generator fault prediction model reaches a preset model fit standard.

[0136] In this solution, the preset model update time interval can be a specified time period, which is used to regularly collect all generator operation information, operation information and environmental information of engine failures after the model is deployed, and use this information to update the generator failure prediction model. Specifically, this time interval can be set according to actual needs, such as updating once a week or a month.

[0137] The second generator operation information, the corresponding second operation information and the corresponding second environment information may be all generator operation information, operation information and environment information of the engine failure collected according to a preset model update time interval after the model is deployed.

[0138] The second generator operation information, the corresponding second operation information, and the corresponding second environmental information can be re-input into the preset generator fault prediction model construction formula. Specifically, each influencing coefficient can be first set to an initial value, and the initialized parameters can be used to calculate the predicted probability value of each sample according to the preset generator fault prediction model construction formula. The loss function of the model is calculated based on the predicted probability value and the actual label, and then the influencing parameters of the model are adjusted by maximizing the likelihood estimate or minimizing the loss function until the model fit of the generator fault prediction model reaches the preset model fit standard.

[0139] In this solution, regular model updates can ensure that the model can capture data changes in a timely manner, thereby maintaining the accuracy of model predictions and the reliability of the model.

[0140] Embodiment 2

[0141] Figure 2 is a flow chart of a method for predicting repetitive defects based on generator faults provided in the second embodiment of the present application, such as Figure 2 As shown, the specific method includes the following steps:

[0142] S201, obtaining historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and training a generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0143]

[0144] Among them, P(Y=1|X) is the probability of generator failure under the condition of given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability.

[0145] S202, calculating the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, determining that the training of the generator fault prediction model is completed.

[0146] S203, acquiring real-time generator operation information, real-time operation information and real-time environmental information, inputting the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determining the occurrence rate of repetitive defects of the generator.

[0147] S204: If the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, it is determined that the generator is at risk of repetitive defects, and a cluster analysis is performed on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain a real-time clustering result, and a target fault mode is determined based on the real-time clustering result and a preset fault mode list.

[0148] S205: Determine whether the target failure mode meets a preset remote processing standard.

[0149] The preset remote processing standard can be a standard for determining whether a specific fault mode can be processed remotely. Specifically, it can include different fault types: software faults, i.e., software configuration errors or conflicts, software updates or patch installations, software restarts or resets. Parameter adjustment faults, i.e., remote adjustment of generator operating parameters, calibration and optimization of E-Air control system parameters. Monitoring and diagnosis faults, i.e., analysis and diagnosis of remote monitoring data, and remote confirmation and processing of warning information. Each type of fault that can be processed remotely has a corresponding fault mode and a corresponding remote processing method.

[0150] After the target failure mode is determined, it can be matched in the preset remote processing standard. If the corresponding failure type can be found, it is considered to meet the preset remote processing standard.

[0151] Based on the above technical solution, optionally, after determining whether the target failure mode meets a preset remote processing standard, the method further includes:

[0152] If the preset remote processing standard is not met, the target fault mode is sent to the handheld terminal of the staff, so that the staff can perform on-site fault processing on the generator according to the target fault mode.

[0153] In this solution, the handheld terminal may be a portable data processing device with specific functions and characteristics. Specifically, it may perform data transmission through a wireless network to achieve communication with a remote server or other devices.

[0154] When the preset remote processing standards are not met, the target fault mode can be sent to the staff's handheld terminal through wireless communication technology. After receiving the target fault mode, the staff can quickly understand the fault-related information and go to the site for corresponding processing.

[0155] In this solution, target fault modes that do not meet the preset remote processing standards are sent to the staff, so that these faults can be handled in a timely manner, reducing the losses caused by equipment failures and improving the efficiency of fault repair.

[0156] S206: If the preset remote processing standard is met, a corresponding remote processing method is matched in the preset remote processing standard according to the target fault mode, and remote fault processing is performed on the generator according to the remote processing method.

[0157] Remote processing can refer to a method of troubleshooting a specific fault mode through remote technical means. Specifically, it can include remote monitoring and diagnosis, that is, obtaining the operating status and data of the generator in real time through a remote monitoring system, and using diagnostic software or algorithms to identify and analyze faults. Parameter adjustment and optimization, that is, the operating parameters of the generator, such as voltage, frequency, and load, can be adjusted remotely to optimize its performance or solve certain faults. Software update and configuration, that is, for software failures or situations where configuration updates are required, remotely upload new software versions or configuration files to achieve remote updates and configurations of the generator control system. Remote restart and reset, that is, for certain faults that can be solved by restarting or resetting, remotely restart the generator or reset its control system by sending control instructions.

[0158] If the conditions are met, the corresponding remote processing method can be found in the preset remote processing standard according to the target failure mode, and then the matched remote processing method can be executed through the remote control system. Specifically, it can include sending control instructions, uploading configuration files, and adjusting parameter values.

[0159] In this embodiment, by setting the remote processing standard, it is possible to quickly determine whether the fault can be processed remotely when a fault occurs, and match the appropriate remote processing method, thereby improving the efficiency of fault processing. In addition, remote processing is performed for faults that can be processed remotely, reducing dependence on staff and reducing labor costs.

[0160] Embodiment 3

[0161] Figure 3 is a flow chart of a method for predicting repetitive defects based on generator faults provided in the third embodiment of the present application, such as Figure 3 As shown, the specific method includes the following steps:

[0162] S301, obtaining historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and training a generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0163]

[0164] Among them, P(Y=1|X) is the probability of generator failure under the condition of given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability.

[0165] S302, calculating the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, determining that the training of the generator fault prediction model is completed.

[0166] S303, obtaining real-time generator operation information, real-time operation information and real-time environmental information, inputting the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determining the repetitive defect occurrence rate of the generator.

[0167] S304: If the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, it is determined that the generator is at risk of repetitive defects, and a cluster analysis is performed on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain a real-time clustering result, and a target fault mode is determined based on the real-time clustering result and a preset fault mode list.

[0168] S305, determining in real time whether the performance data of the generator fault prediction model meets a preset performance standard; wherein the performance standard includes an accuracy standard and a recall rate standard.

[0169] Performance data can be quantitative indicators used to measure the prediction effect of the model. Specifically, it can include accuracy, recall rate, precision, and F1 value.

[0170] The preset performance standard can refer to the threshold or requirement of the model performance indicator that is preset when evaluating the model performance. For example, for the accuracy standard, the model can be required to achieve at least 90% accuracy in predicting faults; for the recall standard, the model can be required to be able to identify at least 90% of the real faults.

[0171] When the model is deployed to actual applications, it will receive real-time operating data of the generator and make fault predictions. These real-time prediction results can be collected and compared with the actual fault conditions to calculate performance data such as the model's accuracy and recall rate. If the performance data meets the preset standards, the model's performance in the current application meets the requirements; otherwise, the model needs to be retrained.

[0172] S306: If the performance data of the generator fault prediction model does not meet the preset performance standard, all first generator operation information, corresponding first operation information and corresponding first environment information when the generator fails are obtained.

[0173] The first generator operation information, the corresponding first operation information and the corresponding first environmental information may be the generator operation information, the corresponding operation information and the corresponding environmental information when a current generator fails, that is, a collection of historical fault data used to train the model and fault data collected after the model is applied.

[0174] When the performance data of the generator fault prediction model does not meet the preset performance standards, the historical generator operation information, the corresponding historical operation information and the corresponding historical environmental information can be read from the database. At the same time, the generator operation information, the corresponding operation information and the corresponding environmental information output after the model is applied when the generator fails can be sorted out, and these data can be integrated to obtain the first generator operation information, the corresponding first operation information and the corresponding first environmental information.

[0175] S307, retraining the generator fault prediction model according to the first generator operation information, the corresponding first operation information, the corresponding first environmental information and the preset generator fault prediction model construction formula until the model fit of the generator fault prediction model reaches the preset model fit standard.

[0176] The first generator operation information, the corresponding first operation information, and the corresponding first environmental information can be re-input into the preset generator fault prediction model construction formula. Specifically, each influencing coefficient can be first set to an initial value, and the initialized parameters can be used to calculate the predicted probability value of each sample according to the preset generator fault prediction model construction formula. The loss function of the model is calculated based on the predicted probability value and the actual label, and then the influencing parameters of the model are adjusted by maximizing the likelihood estimate or minimizing the loss function until the model fit of the generator fault prediction model reaches the preset model fit standard.

[0177] In this embodiment, through real-time monitoring, the decline in model performance can be discovered in time, so that the model can be trained again to improve the accuracy of subsequent predictions of the model, and the stability and reliability of the model can be maintained.

[0178] Embodiment 4

[0179] Figure 4 is a schematic diagram of the structure of a repetitive defect prediction system based on generator faults provided in the fourth embodiment of the present application, such as Figure 4 As shown, specifically including the following:

[0180] The model training module 401 is used to obtain historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and train the generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0181]

[0182] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability;

[0183] A standard comparison module 402 is used to calculate the model fit of the generator fault prediction model. If the model fit reaches a preset model fit standard, it is determined that the training of the generator fault prediction model is completed.

[0184] A repetitive defect prediction module 403 is used to obtain real-time generator operation information, real-time operation information and real-time environmental information, input the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determine the repetitive defect occurrence rate of the generator;

[0185] The fault mode determination module 404 is used to determine that the generator is at risk of repetitive defects if the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, perform cluster analysis on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain real-time clustering results, and determine the target fault mode based on the real-time clustering results and a preset fault mode list.

[0186] In the embodiment of the present application, the model training module is used to obtain historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and train the generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is:

[0187]

[0188] Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability; the standard comparison module is used to calculate the model fit of the generator fault prediction model. If the model fit reaches the preset model fit standard, the generator fault prediction model training is determined. The training is completed; a repetitive defect prediction module is used to obtain real-time generator operation information, real-time operation information and real-time environmental information, input the real-time generator operation information, the real-time operation information and the real-time environmental information into the generator fault prediction model, and determine the repetitive defect incidence rate of the generator; a fault mode determination module is used to determine that the generator has a risk of repetitive defects if the repetitive defect incidence rate is higher than the preset repetitive defect incidence rate threshold, cluster analysis is performed on the real-time generator operation information, real-time operation information and real-time environmental information, and a real-time clustering result is obtained, and the target fault mode is determined according to the real-time clustering result and the preset fault mode list. Through the above-mentioned repetitive defect prediction system based on generator faults, a fault prediction model is set, and potential problems can be discovered in advance before the generator actually fails, thereby improving the reliability and operation safety of the generator. And by predicting and identifying repetitive defects, unnecessary comprehensive inspections or excessive repairs can be avoided, thereby reducing repair costs and improving repair efficiency.

[0189] The above are only preferred embodiments of the present application and the technical principles used. The present application is not limited to the specific embodiments described herein, and various obvious changes, readjustments and substitutions that can be made by those skilled in the art will not deviate from the scope of protection of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

[0190] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

Claims

1. A method for predicting repetitive defects based on generator faults, characterized in that: The method comprises: Obtain historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and train the generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is: Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability; Calculating the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, determining that the generator fault prediction model training is completed; Acquire real-time generator operation information, real-time operation information and real-time environmental information, input the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determine the occurrence rate of repetitive defects of the generator; If the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, it is determined that the generator is at risk of repetitive defects, and a cluster analysis is performed on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain a real-time clustering result, and a target fault mode is determined based on the real-time clustering result and a preset fault mode list.

2. The method for predicting repetitive defects based on generator faults according to claim 1, characterized in that: After training the generator fault prediction model according to the historical generator operation information, the corresponding historical operation information, the corresponding historical environment information and a preset generator fault prediction model construction formula, the method further includes: If the model fit of the generator fault prediction model does not meet the preset model fit standard, sample balancing is performed on the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information according to the preset fault mode list to obtain a sample balance result; The generator fault prediction model is trained according to the sample balancing result and a preset generator fault prediction model construction formula until the model fit of the generator fault prediction model reaches a preset model fit standard.

3. The method for predicting repetitive defects based on generator faults according to claim 2, characterized in that: According to the preset fault mode list, sample balancing is performed on the historical generator operation information, the corresponding historical operation information, and the corresponding historical environment information to obtain a sample balancing result, including: Determine sample imbalance information according to the preset fault mode list, the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information; determine the target sample quantity and target sample information of the historical generator operation information, the corresponding historical operation information and the corresponding historical environment information corresponding to each fault mode according to the sample imbalance information and the preset sample balancing method; A sample balance result is determined according to the target sample quantity and the target sample information.

4. The method for predicting repetitive defects based on generator faults according to claim 1, characterized in that: The process of determining the failure mode list includes: Obtain historical generator operation information, corresponding historical operation information, corresponding historical environment information and corresponding failure mode when the generator fails, perform cluster analysis on the historical generator operation information, corresponding historical operation information and corresponding historical environment information under each failure mode, and obtain historical clustering results corresponding to each failure mode; A failure mode list is determined according to the failure mode and the historical clustering result.

5. The method for predicting repetitive defects based on generator faults according to claim 4, characterized in that: Determining a target failure mode according to the real-time clustering result and a preset failure mode list includes: Calculating the Mahalanobis distance between the real-time clustering result and each historical clustering result in the preset fault mode list, and determining the historical clustering result corresponding to the smallest Mahalanobis distance as the target historical clustering result; The failure mode corresponding to the target historical clustering result is determined as the target failure mode.

6. The method for predicting repetitive defects based on generator faults according to claim 1, characterized in that: After determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes: Determining whether the target failure mode meets a preset remote processing standard; If the preset remote processing standard is met, a corresponding remote processing method is matched in the preset remote processing standard according to the target fault mode, and remote fault processing is performed on the generator according to the remote processing method.

7. The method for predicting repetitive defects based on generator faults according to claim 6, characterized in that: After determining whether the target failure mode meets a preset remote processing standard, the method further includes: If the preset remote processing standard is not met, the target fault mode is sent to the handheld terminal of the staff, so that the staff can perform on-site fault processing on the generator according to the target fault mode.

8. The method for predicting repetitive defects based on generator faults according to claim 1, characterized in that: After determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes: Determine in real time whether the performance data of the generator fault prediction model meets the preset performance standard; wherein the performance standard includes the accuracy standard and the recall rate standard; If the performance data of the generator fault prediction model does not meet the preset performance standard, all first generator operation information, corresponding first operation information and corresponding first environment information when the generator fails are obtained; The generator fault prediction model is retrained according to the first generator operation information, the corresponding first operation information, the corresponding first environmental information and a preset generator fault prediction model construction formula until the model fit of the generator fault prediction model reaches a preset model fit standard.

9. The method for predicting repetitive defects based on generator faults according to claim 1, characterized in that: After determining the target failure mode according to the real-time clustering result and the preset failure mode list, the method further includes: regularly collecting second generator operation information, corresponding second operation information, and corresponding second environment information when the engine fails according to a preset model update time interval; The generator fault prediction model is retrained according to the second generator operation information, the corresponding second operation information, the corresponding second environment information and a preset generator fault prediction model construction formula until the model fit of the updated generator fault prediction model reaches a preset model fit standard.

10. A repetitive defect prediction system based on generator faults, characterized in that: The system comprises: The model training module is used to obtain historical generator operation information, corresponding historical operation information and corresponding historical environment information when the generator fails, and train the generator fault prediction model according to the historical generator operation information, corresponding historical operation information, corresponding historical environment information and a preset generator fault prediction model construction formula; wherein the generator fault prediction model construction formula is: Wherein, P(Y=1|X) is the probability of generator failure under the given parameter X; X1 is the historical generator operation information; X2 is the historical operation information; X3 is the historical environmental information; β0 is the intercept term, that is, the basic probability of generator failure when all features are 0; β1 is the influence coefficient of historical generator operation information on the generator failure probability; β2 is the influence coefficient of historical operation information on the generator failure probability; β3 is the influence coefficient of historical environmental information on the generator failure probability; A standard comparison module is used to calculate the model fit of the generator fault prediction model, and if the model fit reaches a preset model fit standard, it is determined that the training of the generator fault prediction model is completed; A repetitive defect prediction module is used to obtain real-time generator operation information, real-time operation information and real-time environmental information, input the real-time generator operation information, the real-time operation information and the real-time environmental information into a generator fault prediction model, and determine the repetitive defect occurrence rate of the generator; A fault mode determination module is used to determine that the generator is at risk of repetitive defects if the repetitive defect incidence rate is higher than a preset repetitive defect incidence rate threshold, perform cluster analysis on the real-time generator operation information, real-time operation information, and real-time environmental information to obtain real-time clustering results, and determine a target fault mode based on the real-time clustering results and a preset fault mode list.

Images