Hydroelectric generating set intelligent early warning method and system fusing working condition data

Abstract

The invention relates to the technical field of equipment monitoring, and provides a hydroelectric generating set intelligent early warning method and system fusing working condition data, and the method comprises the steps: accessing multi-source heterogeneous operation data, and carrying out the format unification and working condition label preprocessing of the data; based on a parameter probability density analysis and Gaussian fitting method, establishing a mapping relation between real-time data and unit operation conditions; a multi-monitoring-point behavior relation model is constructed, dynamic threshold analysis, trend analysis and analogy analysis are combined, system abnormity is recognized, and graded early warning is generated; on the basis of a neural network evaluation model, unit health state grades including normal, attention, abnormity and severity levels are divided, and an equipment health evaluation result is output; a working condition-behavior model is constructed, a current operation working condition is identified, future behaviors of the unit are predicted, and early detection and early warning of an abnormal state are realized; and integrating the working condition-behavior model into a unit monitoring system to realize real-time working condition identification, behavior prediction and model self-learning feedback.

Description

Technical Field

[0001] This invention relates to the field of equipment monitoring technology, and in particular to an intelligent early warning method and system for hydropower units that integrates operating condition data. Background Technology

[0002] The power industry is currently proposing to build a global energy internet enterprise. Hydropower stations, as an indispensable part of the power grid's energy supply chain, play a crucial role in the power system, including peak shaving, phase regulation, frequency regulation, and emergency backup, and are of paramount importance to the safety and stability of the power grid operation. Years of monitoring and operation of monitoring systems have provided massive amounts of data for the operation and maintenance management of hydropower stations. However, there is currently a lack of intelligent methods and support for the practical application of big data, lagging significantly behind the intelligent technology applications in the construction, manufacturing, medical, and transportation industries, such as autonomous driving, robotics, intelligent buildings, and intelligent assisted treatment. For example, the application of artificial intelligence technology for early warning in hydropower units is currently limited, and its integration with AI and big data analysis is shallow, still relying on human experience for data analysis, performance judgment, and decision-making. Furthermore, there is a lack of effective processing and analysis of monitoring data, making it impossible to predict and warn of abnormal situations in a timely manner, or to detect abnormal trends in advance, thus failing to predict equipment failure and accident risks.

[0003] To address the shortcomings of existing technologies, there is an urgent need to provide an intelligent early warning method for hydropower units that integrates operating condition data, in order to achieve one-stop operation and maintenance analysis and management of electromechanical equipment, thereby effectively improving the comprehensive operation and maintenance level of hydropower station electromechanical equipment. Summary of the Invention

[0004] This invention provides an intelligent early warning method and system for hydropower units that integrates operating condition data, in order to solve the problems existing in current early warning methods for hydropower units.

[0005] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides an intelligent early warning method for hydropower units that integrates operating condition data, comprising: Access heterogeneous operational data from multiple sources, including monitoring systems and online monitoring systems, and perform data format unification and operational condition label preprocessing. Based on parametric probability density analysis and Gaussian fitting methods, a mapping relationship between real-time data and unit operating conditions is established to achieve a strong correlation between data and operating conditions. Construct a multi-monitoring point behavior relationship model, and combine dynamic threshold analysis, trend analysis and analogy analysis to identify system anomalies and generate graded early warnings; Based on a neural network evaluation model, the health status of the unit is classified into four levels: normal, attention, abnormal and severe, and the equipment health assessment results are output. Construct a condition-behavior model to identify the current operating conditions and predict the future behavior of the unit, so as to achieve early detection and warning of abnormal conditions; The operating condition-behavior model is integrated into the unit monitoring system to achieve real-time operating condition identification, behavior prediction, and model self-learning feedback.

[0006] Optionally, the data access step may further include: The PISDK interface is used to connect to the PI database to obtain historical data of measurement points, and high-frequency data acquisition is achieved by using scheduled tasks and multi-threading technology. The collected data is converted into a format and stored in the time-series database InfluxDB, dynamically partitioned by date. Configure automatic reconnection and breakpoint resume mechanisms to ensure the continuity and integrity of data access.

[0007] Optionally, establishing the mapping relationship between real-time data and operating conditions specifically includes: Probability density analysis was performed on the unit operating parameters, and the A / D test was used to determine whether the data conformed to a normal distribution. A parameter distribution model is established based on Gaussian fitting method, and the mean and standard deviation are extracted as working condition identification features. Based on the characteristics of parameter distribution, the operating conditions of the generating units are classified, including steady-state, transient, pumping, and power generation conditions.

[0008] Optionally, the construction of the behavioral relationship model includes: Specific features are extracted for a single monitoring point, and correlation feature fusion analysis is performed for multiple monitoring points; An anomaly detection model based on correlation coefficient and similarity calculation is constructed by introducing state influence factors. Early warning output is provided using at least one of the following methods: vector target area map early warning, spectrum distribution target area early warning, trend early warning, and limit violation early warning.

[0009] Optionally, the classification of the health status level is based on: Multi-dimensional fusion analysis of equipment operation indicators, including vibration, sway, and pressure pulsation spectrum characteristics; A multi-task learning framework based on deep neural networks treats different working conditions as independent tasks for joint training. Overall health status level of output devices and corresponding optimization operation suggestions.

[0010] Optionally, the construction of the working condition-behavior model includes: Clustering algorithms are used to divide the operating data into working conditions, and to find working condition families that are compact within each cluster and separate between each cluster. A multi-task learning neural network model is constructed based on clustering results to achieve working condition recognition and behavior prediction. By analyzing the residuals between real-time data and predicted data, anomaly warnings are triggered and fed back to the model optimization stage.

[0011] In a second aspect, embodiments of this application provide an intelligent early warning system for hydropower units that integrates operating condition data, used to implement the method described in any one of the first aspects, comprising: The data access module is used to collect and preprocess operational data from the PI database and the online monitoring system; The working condition identification module is used to establish a data-working condition mapping relationship based on probability density analysis and Gaussian fitting method; The behavior analysis module is used to build a behavior relationship model of multiple monitoring points and to realize anomaly detection; The health assessment module is used to output the device's health status level based on a neural network model; The early warning output module is used to generate and publish tiered early warning information; The model management module is used for the training, deployment, and dynamic optimization of the work condition-behavior model.

[0012] Optionally, the data access module is further configured with: Multi-threaded data acquisition unit, format conversion unit, table-segmented storage unit, and disconnection reconnection unit; It supports access to at least one data source from InfluxDB, PI, or relational databases.

[0013] Optionally, the health assessment module includes: The feature extraction unit is used to extract spectral features from vibration, sway, and pressure pulsation data; Multi-task neural network units are used to jointly model the equipment state under different operating conditions; The evaluation output unit is used to output the health level and operational optimization suggestions.

[0014] Optionally, the warning output module supports: Multiple visualization and early warning methods are available, including vector target area maps, spectrum distribution target areas, trend charts, and over-limit alarms. Warning information will be pushed out via at least one of the following methods: email, SMS, or system message.

[0015] Beneficial effects: The intelligent early warning method for hydropower units, which integrates operating condition data, provides this invention. It establishes a global model of the normal behavior of the hydropower generator, considering the influence of operating conditions. The operating condition classification results are used as labels to establish a deep neural network-based normal behavior model for the hydropower generator. The stability and generalization boundary of this model are theoretically proven. This model treats each operating condition of the hydropower generator as a task, integrating multiple operating conditions into a deep neural network model within a multi-task learning framework. The normal behavior model of the hydropower generator based on multi-operating condition classification is compared with that based on a single operating condition and without operating condition classification. Attached Figure Description

[0016] Figure 1 A flowchart of a preferred embodiment of the intelligent early warning method for hydropower units that integrates operating condition data; Figure 2 This is a schematic diagram of data access according to a preferred embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0019] Please see Figure 1 This application provides an intelligent early warning method for hydropower units that integrates operating condition data, including: Access heterogeneous operational data from multiple sources, including monitoring systems and online monitoring systems, and perform data format unification and operational condition label preprocessing. Based on parametric probability density analysis and Gaussian fitting methods, a mapping relationship between real-time data and unit operating conditions is established to achieve a strong correlation between data and operating conditions. Construct a multi-monitoring point behavior relationship model, and combine dynamic threshold analysis, trend analysis and analogy analysis to identify system anomalies and generate graded early warnings; Based on a neural network evaluation model, the health status of the unit is classified into four levels: normal, attention, abnormal and severe, and the equipment health assessment results are output. Construct a condition-behavior model to identify the current operating conditions and predict the future behavior of the unit, so as to achieve early detection and warning of abnormal conditions; The operating condition-behavior model is integrated into the unit monitoring system to achieve real-time operating condition identification, behavior prediction, and model self-learning feedback.

[0020] The specific steps in the embodiment satisfy the following: 1. Data access to monitoring systems and online monitoring systems The steps to access the program are as follows: ① Install PISDK on the server. The program imports the necessary PISDK DLL files (PISDK.dll, PISDKCommom.dll, PITimeServer.dll), which are mainly used to connect to the PI database and call the APIs within them.

[0021] ② Query the prepared measurement points. Here, we query the historical values ​​of each measurement point for the five minutes before the current time. This query is performed using scheduled tasks and multi-threading technology.

[0022] ③ Data queried from the PI database is batch-converted to the required InfluxDB database format.

[0023] ④ The InfluxDB database uses a table partitioning method to store data. The database is dynamically partitioned according to the date of each day, and the data obtained from PI data each day is stored in the corresponding date table in the InfluxDB database.

[0024] ⑤ Reconnection mechanism: To prevent errors or sudden network outages during the process, the program automatically reconnects. See attached diagram for data access illustration. Figure 2 .

[0025] After the data is input, debugging and comparison are performed.

[0026] 2. Perform correlation analysis on monitoring data. Based on the specific feature analysis of a single monitoring point, and the correlation feature analysis of multiple monitoring points, a behavioral relationship model is constructed. By introducing the influencing factors of the state, and integrating dynamic threshold analysis, similar influencing factor trend analysis, and similar influencing factor analogy analysis, intelligent analysis of system anomalies is achieved. The system automatically calculates the performance indicators of the system's task execution based on the indicator analysis calculation model, and evaluates performance quality. Then, based on the state anomaly detection and performance quality evaluation results, alarm or early warning information is automatically generated.

[0027] The system intelligently identifies the operating conditions and historical data of equipment, and performs configurable automatic analysis and calculation of key equipment status characteristic indicators and equipment fault indicators, including events such as exceeding normal operation thresholds, exceeding gradient limits, and exceeding correlation coefficient limits. Early warning methods are as follows: vector target area map early warning and spectral distribution target area early warning are used to monitor changes in the distribution characteristics of various spectral components of unit vibration, sway, and pressure pulsation, enabling the detection of early fault signs, including vector target area map early warning, spectral distribution target area early warning, trend early warning, and vibration sway measurement value exceeding limit early warning.

[0028] 3. Operating condition analysis of intelligent equipment in hydropower stations Based on big data platforms and neural network evaluation models, we formulate equipment operating condition evaluation standards and health status prediction standards, and develop a smart equipment operating condition analysis module for hydropower stations. By studying equipment health status assessment, we specify the equipment's operating conditions and health evaluation standards, analyze and evaluate the data of various indicators reflecting the equipment's health status, and finally derive the overall health status level of the equipment, namely, four levels: normal, attention, abnormal, and severe.

[0029] First, real-time operating data of the unit is collected through sensors and online monitoring systems. Then, advanced fault diagnosis theories and data modeling methods are used to conduct in-depth analysis of the data. After obtaining the analysis results, a comprehensive assessment of the unit's operating conditions is conducted, and based on this, targeted optimization suggestions and improvement measures are proposed.

[0030] 1) Establish a mapping relationship between real-time data and unit operating conditions. First, the various operating conditions of the unit are clearly defined. The operating conditions can be defined by specific operating parameters and conditions. Based on the operating characteristics of the unit and the definition of the operating conditions, real-time data closely related to the operating conditions of the unit are determined. These data may include temperature, pressure, flow rate, speed, power, etc.

[0031] By using methods such as parametric probability density analysis and Gaussian fitting, historical real-time data and corresponding unit operating conditions are analyzed. Based on the data analysis results, a mapping model between real-time data and unit operating conditions is established. When a certain real-time data exceeds a certain threshold, it is determined that the unit has entered a certain specific operating condition. The established mapping model is applied to real-time data analysis, and the unit's operating condition is predicted based on the input of real-time data.

[0032] 2) Data is strongly correlated and bound to the unit's operating conditions. ① Parametric probability density analysis: For each parameter, its probability density function (PDF) or probability distribution can be calculated to view the distribution and range of parameter values. By comparing the parameter PDFs under different working conditions, parameter patterns associated with specific working conditions can be discovered.

[0033] The AD test assesses whether a sample comes from a specified distribution. When the data does indeed come from the specified distribution, it can be assumed that the cumulative distribution function of the data follows that distribution. This method tests the uniformity of the data distribution by calculating distance, for data from an assumed distribution function F. <...< The test statistic A can be expressed as: (1) (2) Before performing any statistical test, you first need to choose a significance level (usually denoted as α), which represents the threshold for rejecting the null hypothesis.

[0034] A distribution model was tested on the fault data. The P-value between the selected normal distribution and the theoretical distribution was calculated using Matlab software (the selection of the distribution was determined by comparing the P-value with the significance level). The results showed that the data conformed to the normal distribution. The distribution of the relevant parameters obtained from the calculation can be used to determine whether the unit's operating condition under the parameters is a transient state or a steady state.

[0035] ②Gauss fitting Gaussian fitting can provide statistical information such as the mean and standard deviation of the parameter distribution. It uses mathematical models to describe and predict patterns of data change, and can be used to view the range and trend of data variation.

[0036] The Gaussian function, also known as the normal distribution function, when used as a probability density function, resembles a bell-shaped curve. Its mathematical expression is as follows: f(x) = (3) Where x is the independent variable, f(x) is the value of the function at a certain point x, the parameter b is the mean, representing the center position of the bell curve, and c is the standard deviation, which determines the height of the curve.

[0037] Data fitting is performed using the Gaussian function. The fitting process is based on the least squares method. By adjusting the parameters of the Gaussian function, the residual between the function curve and the actual data is minimized, and the best-fit curve can be obtained. This helps to understand the central tendency and dispersion of the data and can be used to describe the overall trend of the data.

[0038] Establish a trend prediction model: The collected data are relevant historical data, which are continuous and predictable. The mean and standard deviation of the normal distribution are estimated based on the historical data. Using the estimated mean and standard deviation, a trend prediction model based on the normal distribution can be constructed. That is, the probability density function of the normal distribution can be used to predict the range or probability of values ​​at a future point in time.

[0039] Establish anomaly trend early warning: ① Set thresholds: Based on business needs and data characteristics, set one or more thresholds. These thresholds will be used to determine whether the data is abnormal; ② Calculate confidence intervals: Based on the mean and standard deviation of the normal distribution, confidence intervals at different confidence levels can be calculated. These confidence intervals will be used to determine whether the fluctuation range of the data is normal; ③ Anomaly detection: When a new data point appears, calculate its probability of falling within the confidence interval. If the probability is lower than a certain preset significance level (e.g., 5%), the data point is considered abnormal; ④ Early warning trigger: When an abnormal data point is detected, trigger the early warning mechanism. The early warning can be notified to relevant personnel via email, SMS, push notifications, etc.

[0040] 4. Establish a working condition-behavior model Operating condition identification and behavior prediction are core steps in realizing the operating condition-behavior model. By comparing and analyzing real-time monitoring data with the established operating condition-behavior model, the current operating condition type of the unit can be identified, and future behavior of the unit can be predicted based on this operating condition type. Anomaly detection and early warning are important guarantees for realizing the operating condition-behavior model. By comparing and analyzing real-time monitoring data with data predicted by the behavior model, anomalies or deviations from normal operating ranges in unit operation can be detected in a timely manner. Once these problems are detected, the system will promptly issue early warning signals to notify relevant personnel for handling and intervention to ensure the stable operation and safety of the unit.

[0041] 5. Application The established mapping model is applied to real-time data analysis. Based on the input of real-time data, the operating conditions of the unit are predicted. The prediction results are compared with the actual situation, and the model is continuously fed back and adjusted to improve the real-time performance and accuracy of the model.

[0042] By integrating the established mapping model into the unit's monitoring system, real-time data can be automatically correlated with and displayed in relation to the unit's operating conditions.

[0043] This application embodiment also provides an intelligent early warning system for hydropower units that integrates operating condition data, used to implement any one of the methods for intelligent early warning of hydropower units that integrates operating condition data, characterized in that it includes: The data access module is used to collect and preprocess operational data from the PI database and the online monitoring system; The working condition identification module is used to establish a data-working condition mapping relationship based on probability density analysis and Gaussian fitting method; The behavior analysis module is used to build a behavior relationship model of multiple monitoring points and to realize anomaly detection; The health assessment module is used to output the device's health status level based on a neural network model; The early warning output module is used to generate and publish tiered early warning information; The model management module is used for the training, deployment, and dynamic optimization of the work condition-behavior model.

[0044] Optionally, the data access module is further configured with: Multi-threaded data acquisition unit, format conversion unit, table-segmented storage unit, and disconnection reconnection unit; It supports access to at least one data source from InfluxDB, PI, or relational databases.

[0045] Optionally, the health assessment module includes: The feature extraction unit is used to extract spectral features from vibration, sway, and pressure pulsation data; Multi-task neural network units are used to jointly model the equipment state under different operating conditions; The evaluation output unit is used to output the health level and operational optimization suggestions.

[0046] Optionally, the warning output module supports: Multiple visualization and early warning methods are available, including vector target area maps, spectrum distribution target areas, trend charts, and over-limit alarms. Warning information will be pushed out via at least one of the following methods: email, SMS, or system message.

[0047] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for intelligent early warning of a hydroelectric generating unit fused with working condition data, characterized in that, The application comprises the following steps: Accessing multi-source heterogeneous operation data from monitoring systems and online monitoring systems, and performing format unification and working condition label preprocessing on the data; Based on parameter probability density analysis and Gaussian fitting method, the mapping relationship between real-time data and unit operation condition is established, and the strong correlation binding between data and condition is realized; A multi-monitoring point behavior relationship model is constructed, combined with dynamic threshold analysis, trend analysis and analogy analysis, to identify system abnormalities and generate graded warnings; Based on neural network evaluation model, the health state level of the unit is divided, including normal, attention, abnormal and serious four levels, and the equipment health evaluation result is output; A working condition-behavior model is constructed to identify the current operating condition and predict the future behavior of the unit, so as to realize early detection and warning of abnormal state; The working condition-behavior model is integrated into the unit monitoring system to realize real-time working condition identification, behavior prediction and model self-learning feedback.

2. The intelligent warning method of fusion working condition data of hydroelectric generating unit according to claim 1, characterized in that, In the data access step, it also includes: Using PISDK interface to connect PI database, obtaining historical data of measuring points, and using timing task and multi-thread technology to realize high-frequency data acquisition; The collected data is converted in format and stored in time series database InfluxDB, and dynamically stored in tables according to date; Automatic reconnection and breakpoint resume mechanism is configured to ensure the continuity and integrity of data access.

3. The intelligent warning method of fusion working condition data of hydroelectric generating unit according to claim 1, characterized in that, The mapping relationship between real-time data and operating condition is established, which specifically includes: Probability density analysis is performed on the operating parameters of the unit, and A-D test is used to determine whether the data conforms to normal distribution; Based on Gaussian fitting method, parameter distribution model is established, and mean and standard deviation are extracted as working condition identification features; According to the parameter distribution characteristics, the unit operating condition is divided, including steady state, transient state, pumping and power generation conditions.

4. The intelligent warning method of fusion working condition data of hydroelectric generating unit according to claim 1, characterized in that, The construction of the behavior relationship model includes: Special feature extraction is performed on a single monitoring point, and correlation feature fusion analysis is performed on multiple monitoring points; State influence factor is introduced to construct an abnormal detection model based on correlation coefficient and similarity calculation; At least one of the following warning output modes is adopted: vector target area graph warning, spectrum distribution target area warning, trend warning and out-of-limit warning.

5. The intelligent warning method of fusion working condition data of hydroelectric generating unit according to claim 1, characterized in that, The division of health state level is based on: Multi-dimensional fusion analysis of equipment operation indicators, including vibration, swing, pressure pulsation spectrum characteristics; Based on the multi-task learning framework of deep neural network, different working conditions are regarded as independent tasks for joint training; The overall health state level of the equipment and the corresponding optimization operation suggestion are output.

6. The intelligent warning method of fusion working condition data of hydroelectric generating unit according to claim 1, characterized in that, The construction of the working condition-behavior model includes: Using clustering algorithm to divide the working condition of the running data, and finding the working condition family with tightness within the class and separation between the classes; Based on the clustering results, a multi-task learning neural network model is constructed to realize working condition identification and behavior prediction; Through residual error analysis of real-time data and predicted data, abnormal warning is triggered and fed back to the model optimization link.

7. A fusion working condition data intelligent early warning system for hydroelectric generating units, for implementing the method according to any one of claims 1-6, characterized in that, It comprises: A data access module for collecting and preprocessing operation data from PI database and online monitoring system; A working condition identification module for establishing data-condition mapping relationship based on probability density analysis and Gaussian fitting method; A behavior analysis module for constructing a multi-monitoring point behavior relationship model and realizing abnormal detection; The health assessment module is configured to output a device health state level based on a neural network model; The early warning output module is configured to generate and issue a graded early warning information; The model management module is configured to train, deploy and dynamically optimize the working condition-behavior model.

8. The intelligent warning system for fusion working condition data of hydroelectric generating units according to claim 7, characterized in that, The data access module is further configured to have: A multi-thread data acquisition unit, a format conversion unit, a sub-table storage unit and a broken line reconnection unit; Support at least one of InfluxDB, PI, and a relational database data source access.

9. The intelligent warning system for fusion working condition data of hydroelectric generating units according to claim 7, characterized in that, The health assessment module includes: A feature extraction unit configured to extract frequency spectrum features from vibration, swing, and pressure pulsation data; A multi-task neural network unit configured to jointly model the device state under different working conditions; An evaluation output unit configured to output a health level and an operation optimization suggestion.

10. The system of claim 7, wherein, The early warning output module supports: Multiple visualization early warning methods such as vector target area map, frequency spectrum distribution target area, trend chart, and out-of-limit alarm; Early warning information is pushed through at least one of email, SMS, and system message.

Images