A health state identification system and method for a hydroelectric generating unit

Abstract

The application discloses a health state identification system method for a hydroelectric generating set, and the system comprises a collector and a processor. The collector collects operation data of the hydroelectric generating set and defined unit performance indexes. The processor processes the data collected by the collector. The processor comprises a learning and training module, a model identification module, an inspection iteration module and a pre-warning output module. The learning and training is based on the health state through the identification system method, and feature variables are extracted as health samples to construct an identification model. In combination with the mechanical structure and mechanism of the hydroelectric generator, identification is carried out from a basic model, a knowledge model, a physical model to a mathematical model, parameters in the model are determined, and a suitable learning algorithm or mathematical model is selected to perform model evaluation and prediction. The application can provide reliable data basis for a system platform of safe and optimized operation, fault diagnosis and state maintenance of the hydroelectric generating set.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of state identification, abnormal monitoring and diagnostic analysis of hydroelectric generating units, in particular to a hydroelectric generating unit health state identification system and method. BACKGROUND

[0002] At present, most domestic hydropower plants use a diagnostic platform based on a state monitoring system for fault diagnosis and operation and maintenance management. In actual use, it is found that such systems are mostly at the level of "state monitoring", and can only issue an alarm when a fault occurs in the unit, prompting the fault location. The ability to predict potential fault sources and analyze fault causes is weak, and operation, maintenance and repair strategies cannot be proposed.

[0003] The traditional analysis method does not comprehensively consider information fusion at each level in essence, resulting in frequent false alarms, false reports and missed reports in the analysis conclusion. The main factors from the perspective of data processing are: mutual contradiction between high-dimensional data information reflecting the operating state of the unit; mismatch between data processing method and information data; influence of environmental changes and interference information; coupling effect of two or more faults leading to fuzzy fault information connotation; inaccurate preset alarm and fault standards, etc.

[0004] The traditional system lacks artificial intelligence data fusion algorithms. When the dimension of the acquired unit data is more, the possibility of information contradiction and information entropy increase is also greater, and the reliability and adaptability of identification are lower.

[0005] At present, the data export analysis group of the hydroelectric generating unit often uses field tests and simple limit value comparison methods for state evaluation and monitoring and early warning. This method is affected by many factors such as working conditions (water regime) changes, unit structure, measurement equipment and personnel, lacks early warning ability for potential faults and new faults that have never occurred, and cannot fully represent the current actual health state and operation service capability of the unit. SUMMARY

[0006] To solve the above problems and defects, the present application proposes a hydroelectric generating unit health state identification system and method, which aims to extract the health sample characteristics in the unit data, discover the inherent law of the running equipment, understand the operation law of the unit, automatically perform state identification and prediction, output related early warning applications, guide the operation mode to avoid, make the unit run in a healthy area, and ensure the safe operation of the unit. The output results are continuously improved through iterative learning. It provides reliable data basis for the safe and optimal operation, fault diagnosis and state maintenance of the hydroelectric generating unit.

[0007] The application carries out health state-based learning training through a recognition system method, extracts feature variables as a recognition model constructed by health samples, and combines the mechanical structure and mechanism of the hydroelectric generator to expand recognition from a basic model, a knowledge model, a physical model to a mathematical model, determine parameters in the model, and select a suitable learning algorithm or mathematical model to carry out model evaluation and prediction.

[0008] The technical solution of the application is as follows:

[0009] A hydroelectric generator set health state recognition system comprises a collector and a processor, the collector collects hydroelectric generator set operation data and defined unit performance indexes, and the processor processes the data collected by the collector, and the processor comprises a learning training module, a model recognition module, an inspection iteration module and a warning output module.

[0010] The learning training module carries out learning training and feature extraction, extracts representative feature variables from the data collected by the collector as health samples to construct a recognition model through learning training;

[0011] The model recognition module combines the mechanical structure and mechanism of the hydroelectric generator to determine parameters in the model and select a suitable learning algorithm or mathematical model, and carries out model evaluation on a test set;

[0012] The inspection and iteration module adjusts and optimizes the model according to the model evaluation result, re-trains and evaluates by adjusting input variables, changing algorithm parameters and increasing training data until a model meeting the accuracy and precision requirements is obtained;

[0013] The warning output module carries out model prediction, carries out unit performance prediction and optimizes unit performance based on the recognized model, and outputs the prediction result to the staff and the monitoring system for unit state evaluation and abnormality judgment.

[0014] Further, in the learning training module, a multi-dimensional Gaussian model is used for classification, the health state features in the unit data are obtained through a learning algorithm, a safety boundary is automatically set, and a dangerous area of the unit is automatically found, wherein the multi-dimensional Gaussian model algorithm is as follows:

[0015]

[0016] wherein, represents a feature vector of dimensional unit state data, is the average value of these vectors, Σ represents the covariance matrix of all unit state vectors, and T is a threshold value for determining the sample category.

[0017] Further, the learning algorithm model based on big data in the learning training module is as follows:

[0018]

[0019] In the formula, X is original data, Y is an analysis target, S(E(X), Y) is a supervision function, is a learning parameter matrix, is an output result, is an evaluation function; according to different input health state data sets X, a result for a specific device state is trained

[0020] Further, in the model identification module, a health sample learning algorithm based on probability evaluation is adopted, wherein a numerical probability evaluation method is:

[0021]

[0022] wherein, is fault state data, is full state data.

[0023] The present application also relates to a water turbine generator set health state identification system method, comprising collecting water turbine generator set operation data and defined unit performance indexes, and processing the data collected by the collector, wherein:

[0024] learning training and feature extraction are performed, representative feature variables are extracted from the data collected by the collector as health samples through learning training, and an identification model is constructed;

[0025] combined with the mechanical structure and mechanism of the water turbine generator, identification is carried out from a basic model, a knowledge model, a physical model to a mathematical model, parameters in the model are determined, a suitable learning algorithm or mathematical model is selected, and model evaluation is performed on a test set;

[0026] according to the model evaluation result, the model is adjusted and optimized, training and evaluation are re-performed through adjusting input variables, changing algorithm parameters and increasing training data, until a model meeting the accuracy and accuracy requirements is achieved;

[0027] model prediction is performed, unit performance prediction and optimization of unit performance are performed based on the identified model, and the prediction result is output to the staff and the monitoring system, and is used for unit state evaluation and abnormality judgment.

[0028] The system method of the present application adopts different data fusion algorithms for different levels, solves the effective processing problem of high-dimensional collected data, the algorithm has strong adaptability for the complexity and particularity of the water turbine generator set state, improves the accuracy and reliability of automatic identification application, and each layer is combined to complete safety early warning, hidden danger discovery, optimization operation, energy efficiency management and other application targets.

[0029] The application can realize automatic completion of health state identification of the hydroelectric generating set, early identification of abnormal state, and automatic formation of a preliminary safety boundary. Then, data of one to two head periods of the unit operation are accessed for training iteration, and the model parameters are gradually improved, so that the accuracy of analysis and prediction of the system is continuously improved.

[0030] Through the health sample identification algorithm, the possible faults and defects of the unit can be timely evaluated and predicted, the degradation trend and fault warning of the hydroelectric generating set are early warned, the safety boundary warning protection is provided for different operating conditions of the unit, the identification, analysis and early warning tasks of the unit state are automatically completed, compared with the traditional method, the application condition range is wide, the early warning result is more accurate, and the problem that the traditional preset threshold warning can only protect the single operating condition of the unit is effectively solved.

[0031] The application can automatically identify the pre-fault or early fault of the unit under different operating conditions and push the early warning, and according to the identification result, maintenance suggestions are proposed, so that the stable operation and power generation efficiency of the unit are improved, the degradation trend is found in time, and the loss caused by unnecessary periodical routine maintenance work is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 The figure is a system block diagram of the application;

[0033] Figure 2 The figure is a system block diagram of the application; DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments will be described clearly and completely in combination with the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0035] Unless otherwise defined, technical terms or scientific terms used in the embodiments of the present application shall have the meanings that are commonly understood by a person of ordinary skill in the art. The words "first", "second", and similar words of distinction do not by themselves indicate any order, quantity, or importance, but are used to distinguish one element from another. The words "include" or "contain" and similar words mean that the elements or objects before the words encompass the elements or objects listed after the words and their equivalents, without excluding other elements or objects. The words "mount", "connect", and "join" should be interpreted broadly, for example, they can be fixed connections, detachable connections, or integral connections; they can be direct connections or indirect connections through an intermediate medium; and they can be internal connections of two elements. The words "up", "down", "left", "right", "horizontal", and "vertical" are only used to describe the relative positions of the components in the drawings, and these directional terms are relative concepts, which can be changed accordingly according to the positions of the components in the drawings.

[0036] As shown in Figure 1 The health state identification system of the hydroelectric generating set of the present embodiment includes a collector and a processor. The collector collects the operation data of the hydroelectric generating set and the defined performance indicators of the set, including but not limited to mechanical parameters, electrical parameters, hydrological parameters, water level, etc., and defines the performance indicators of the set (such as efficiency, etc.).

[0037] The processor processes the data collected by the collector. The processor includes a learning and training module, a model identification module, a verification and iteration module, and a pre-warning output module. Among them:

[0038] The learning and training module performs learning and training and feature extraction. The representative feature variables are extracted from the data collected by the collector as health samples through learning and training, and the identification model is constructed.

[0039] The model identification module identifies from the basic model, the knowledge model, the physical model to the mathematical model in combination with the mechanical structure and mechanism of the hydroelectric generator, determines the parameters in the model, selects the appropriate learning algorithm or mathematical model, and evaluates the model on the test set.

[0040] The verification and iteration module adjusts and optimizes the model according to the model evaluation results. The model is retrained and evaluated by adjusting the input variables, changing the algorithm parameters, and increasing the training data until the model meets the accuracy and accuracy requirements. The model is evaluated on the test set, and the evaluation index can be mean absolute error (MAE), mean square error (MSE), or R value, etc.

[0041] The early warning output module performs model predictions, using the identified models to predict and optimize unit performance. The prediction results are then output to staff and the monitoring system for unit status assessment and anomaly detection. Simultaneously, based on the prediction results, appropriate maintenance and repair plans can be developed promptly.

[0042] The first aspect of this invention is to use big data to conduct machine learning training on historical data of hydro-generator units and establish an identification model. The model identification module combines the learned and trained model with the structure and operation mechanism of the hydro-generator to predict the elements of the covariate matrix between variables represented by relevant parameters. This includes the identification process from basic model, knowledge model, physical model to mathematical model, to determine whether the parameters in the model are appropriate, thereby selecting a model with the best fit.

[0043] The verification and iteration module evaluates the model obtained from the identification module using a verification dataset. Based on the evaluation results, iterative training is performed again to gradually optimize the model from coarse to fine. The number of iterations is determined by the evaluation results and continues until convergence or a pre-defined stopping criterion is met.

[0044] By using model samples to examine and identify online data, abnormal conditions of the generating units can be detected, and diagnostic and decision support can be provided, thereby improving the diagnostic and analytical capabilities of the hydropower operation and maintenance monitoring platform.

[0045] This includes classifying the input data using a multidimensional Gaussian model, learning algorithms to acquire health status characteristics from the unit data, thereby discovering the inherent patterns of operating equipment, automatically setting safety boundaries, and automatically identifying hazardous areas of the unit.

[0046] The multidimensional Gaussian model algorithm of this invention is as follows:

[0047]

[0048] in, The dimension of the unit status data is indicated as follows: eigenvectors, It is the average of these vectors, where Σ represents the state vectors of all units. The covariance matrix. T is the threshold for determining the sample class.

[0049] In this embodiment, a multidimensional Gaussian function is used to classify the input data. Technical measures are taken in the selection of algorithms and data cleaning methods before modeling to reduce the impact of irrelevant data on quality. Different data processing methods are used for data cleaning based on the different operating states and structural mechanisms of the generator sets, and multiple datasets are matched and analyzed.

[0050] This embodiment utilizes a linear regression algorithm model, inputting multidimensional data of the generating unit in matrix form into the model. It then analyzes the relevant multidimensional data to understand the unit's operational patterns. The model uses mathematical probability methods to calculate the correlation between the unit's health status and various characteristics, i.e., the "weight parameters." After training, data is input into the model according to "feature values," and the model outputs a reasonable prediction result. This automatically predicts the unit's condition and outputs relevant early warnings to guide operational procedures and prevent adverse events, ensuring the unit operates within a healthy range and guaranteeing its safe operation. Through continuous iterative learning, the model continuously improves its output results.

[0051] This embodiment includes a health sample learning algorithm based on probability assessment. In the absence of fault identification data, it transforms quantitative assessment into numerical probability assessment, converting the incomputability of a "random problem" into a deterministic algorithm that calculates "empirical frequencies" based on historical data. The model is then validated through practical application and "trial and error learning." The numerical probability assessment method includes:

[0052]

[0053] In the formula For fault status data functions, This is a full-state data function. The fault state data over time time t. The full-state data of the relative monitoring equipment The probability assessment uses automatic learning to establish health samples of the equipment, thereby establishing equipment health assessment standards and solving practical problems in assessing complex conditions.

[0054] The influence of the characteristic values ​​of different variables on the result value is calculated, and the samples are identified by classifying the subjective understanding of equipment status by experts, thereby realizing the health assessment and trend prediction of the unit and completing the diagnostic and predictive analysis.

[0055] Therefore, this invention continuously upgrades and iterates through machine self-learning, improving the depth and breadth of the algorithm's functions in optimizing the operation, assessing the status, and diagnosing faults of hydro-generator units, thereby enhancing the operational stability of hydro-generator units and ensuring their safe and reliable operation.

[0056] One of the identification methods used in this embodiment is the high-dimensional data correlation analysis method. A data credibility function is established, and multiple factors such as objective physical laws and constraints between different data categories are considered. Combined with the water turbine structure and professional experience, the output of the neural network is used as the allocation value of the credibility function. The single credibility values ​​of multiple feature values ​​are integrated into a comprehensive credibility function value to complete the classification and identification of abnormal data states.

[0057] The present invention completes the learning and training of data through a learning and training module. The learning algorithm model based on big data is as follows:

[0058]

[0059] In the formula, X represents the original data, Y represents the analysis objective, and S(E(X), Y) represents the supervision function. To learn the parameter matrix, For the output results, This is the evaluation function. The algorithm model is trained based on different input health status datasets X to produce results specific to the state of each device. Its core is the extraction, transformation and loading of state data, followed by data learning, training and evaluation, thereby completing the discovery and condensation of value, and finally obtaining analytical results that have a continuous improvement effect on the entire system.

[0060] This embodiment of the system establishes a classification algorithm model based on big data learning algorithms to achieve the evaluation and prediction goals. The core of its operating mechanism is to train the model through learning algorithms to understand the inherent operational regularities of the unit, use these regularity models to warn of dangerous boundaries, optimize operating modes, and combine macroscopic energy input / output data with specific equipment data to generate specific application interpretations and conclusions. The system automatically completes data analysis, automatically outputs analysis results, automatically pushes results, and automatically detects, analyzes, and judges abnormal operating conditions through learning algorithms.

[0061] like Figure 2 As shown in the figure, this embodiment provides a method for identifying the health status of a hydro-generator unit:

[0062] In this embodiment, relevant data from the turbine unit monitoring system and status monitoring equipment during normal operation are collected as training data for the system. Mathematical tools are used to extract, transform, and load known data types and mechanism-based status types, and a training dataset containing various operating conditions and health statuses of the turbine unit during normal operation is constructed.

[0063] The health status training dataset in this embodiment includes: the turbine unit's start-up status, vibration, swing, pulsation, head, flow rate, active power, reactive power, stator current, rotor current, temperature, etc., as well as related equipment data such as excitation, speed regulation, and auxiliary equipment.

[0064] In this embodiment, relevant data on specific identification equipment of the turbine unit are initially extracted from the training dataset, and correlation analysis is performed to reduce the quality impact of irrelevant data. The input data is then classified using a multidimensional Gaussian function. Technical measures are taken regarding algorithm selection and data cleaning methods before modeling. Based on different operating states and structural mechanisms of the generator unit, multiple datasets are matched and analyzed. Appropriate data processing methods are used to continuously clean and filter the data, finally determining X as the input data space.

[0065] In this embodiment, the health status data space X of the hydro-generator unit is input, and based on operator and mechanism constraints, with Y as the objective... To learn the parameter matrix, The evaluation function is S(F(X), Y), which is the supervision function, and semi-supervised learning training is performed. All functions involved in this embodiment are existing function models, which the system calls according to specific needs.

[0066] In this embodiment, through learning and training based on artificial intelligence algorithms, health samples of hydro-generator units are calculated and identified, and the identification model results are classified and stored to form a health sample identification model library.

[0067] In this embodiment, the identification model is analyzed and evaluated from both macro and micro data perspectives, taking into account the application objectives. Compared with traditional data analysis, this system method attempts to evaluate the inherent laws of the identification model with a learning attitude. This exploratory evaluation of cognition can better uncover the value and laws behind the data, which is the core of the data analysis technology in this embodiment, thereby generating a wealth of applications.

[0068] In this embodiment, while the model is being evaluated, the identification model is being tested and iteratively trained. At the same time, the model's effectiveness is being continuously optimized and evaluated under the physical mechanisms and constraints of the hydro-generator unit.

[0069] In this embodiment, based on the health status identification model, the operating data of the hydro-generator unit is accessed to carry out predictive analysis applications for related equipment, including safety boundary mode optimization, operation mode optimization, scheduling mode optimization, operating cost optimization, hidden danger discovery, unit life prediction, maintenance cost prediction, etc.

[0070] In this embodiment, for the real-time operating data input from a specific hydro-generator unit, different feature samples will be obtained. These feature samples need to be identified and classified again. The samples that have been identified and formed patterns will become the knowledge reserves of the same type of unit and be stored in the knowledge base.

[0071] In this embodiment, guided by model prediction and knowledge base, data-driven results are output to complete the identification and application of the health status of the hydro-generator unit, including safety boundary mode optimization, operation mode optimization, scheduling mode optimization, operating cost optimization, hidden danger discovery, unit life prediction, maintenance cost prediction, etc.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A health status identification system for a hydro-generator unit, characterized in that: It includes a data acquisition unit and a processor. The data acquisition unit collects operating data and defined unit performance indicators of the hydro-generator unit. The processor processes the data collected by the data acquisition unit and includes a learning and training module, a model identification module, a verification and iteration module, and an early warning output module. The learning and training module performs learning and feature extraction. Through learning and training, representative feature variables are extracted from the data collected by the collector as health samples to build an identification model. The model identification module combines the mechanical structure and mechanism of the hydro-generator to determine the parameters in the model and select a suitable learning algorithm or mathematical model; the model is then evaluated on the test set. The verification and iteration module adjusts and optimizes the model based on the model evaluation results. By adjusting input variables, changing algorithm parameters, and increasing training data, the model is retrained and evaluated until it meets the accuracy and precision requirements. The early warning output module performs model prediction, and based on the identified model, it predicts and optimizes the unit performance, and outputs the prediction results to staff and the monitoring system for unit status assessment and anomaly detection.

2. The system according to claim 1, characterized in that: In the learning and training module, a multidimensional Gaussian model is used for classification. The learning algorithm acquires health status features from the unit data, automatically sets safety boundaries, and automatically identifies hazardous areas within the unit. The multidimensional Gaussian model algorithm is as follows: ； in, This represents a feature vector of dimension D representing the unit status data. This is the average of these vectors, Σ represents the state vectors of all units, X is the covariance matrix, and T is the threshold for determining the sample category.

3. The system according to claim 1, characterized in that: In the learning and training module, the learning algorithm model based on big data is as follows: ； In the formula, X The original data, Y To analyze the target, S ( E ( X ), Y ) is the supervision function. To learn the parameter matrix, For the output results, The evaluation function is based on different input health status datasets. X， Training to produce results specific to the condition of the equipment. .

4. The system according to claim 1, characterized in that: In the model identification module, a health sample learning algorithm based on probability assessment is adopted, wherein the numerical probability assessment method is: ； Where Nφe(t) represents the fault state data, and Nφ(t) represents the full state data.

5. A method for identifying the health status of a hydro-generator unit based on the system described in claim 1, characterized in that: This includes collecting operating data and defined performance indicators of the hydro-generator units, and processing the data collected by the data acquisition device, including: The learning and training process involves extracting representative feature variables from the data collected by the collector as health samples to construct an identification model. Based on the mechanical structure and mechanism of hydro-generators, we identify the basic model, knowledge model, physical model and mathematical model, determine the parameters in the model, select the appropriate learning algorithm or mathematical model, and evaluate the model on the test set. Based on the model evaluation results, the model is adjusted and optimized by adjusting input variables, changing algorithm parameters, and increasing training data, and then retraining and evaluating until a model that meets the accuracy and precision requirements is achieved. Model prediction is performed, and unit performance is predicted and optimized based on the identified model. The prediction results are output to staff and monitoring systems for unit status assessment and anomaly detection.

Images