Intelligent object construction method of hydroelectric equipment body

Abstract

The invention relates to the field of hydroelectric equipment, and provides an intelligent object construction method of a hydroelectric equipment body, which comprises the following steps: establishing a hydroelectric equipment state intelligent sensing system, and obtaining state data of hydroelectric equipment; establishing a multi-source information fusion, feature extraction and self-learning analysis system based on the hydroelectric equipment; the state of the hydroelectric equipment is intelligently evaluated; carrying out early warning on faults of the hydroelectric equipment; through the above steps, the construction and evaluation of the intelligent object of the hydroelectric equipment are realized. By constructing a full-service process of multi-dimensional situation intelligent perception, multi-source information fusion, feature self-learning, fault self-diagnosis and maintenance self-decision making of the hydroelectric equipment, comprehensive monitoring evaluation and intelligent management of a hydroelectric equipment body are realized, the evaluation precision of the operating state of the hydroelectric equipment is greatly improved, early warning can be realized, and the service life of the hydroelectric equipment is prolonged. And the operation efficiency and the safety of the equipment are improved.

Description

Technical Field

[0001] The invention relates to the field of hydropower equipment, and in particular to a method for constructing an intelligent object of a hydropower equipment body. Background Art

[0002] With the development of science and technology, the automation and intelligence level of hydropower equipment is constantly improving. However, there are still some problems in the existing hydropower equipment monitoring, early warning and diagnosis systems, such as limited state perception means and inaccurate fault diagnosis. Specifically:

[0003] 1. The existing technology has blind spots in the perception of the status of hydropower equipment. The monitoring items and measurement points are not well arranged, resulting in problems such as inability to measure, inaccurate measurement, and incomplete measurement, which restricts the accurate grasp of the status of hydropower equipment.

[0004] 2. The level of mining of the massive data generated by the hydropower equipment itself is insufficient, and there are information islands, which makes it impossible to fully utilize the massive data to assist in power production and equipment maintenance.

[0005] 3. The means of early warning and diagnosis of hydropower equipment faults are single, and the historical health sample data are not fully utilized to establish an entity health sample library. The exploration of the occurrence and growth principles of the fault mechanism is unclear and unclear.

[0006] 4. The maintenance strategy still uses the model of regular maintenance + planned maintenance, which fails to fully consider the actual status of the hydropower equipment itself, especially fails to implement preventive maintenance of faults to reduce risks. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method for constructing an intelligent object of a hydropower equipment body, which can construct an intelligent object based on multi-source data collection and multi-source information fusion, and can intelligently evaluate the status of the hydropower equipment and issue early warnings for faults.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for constructing an intelligent object of a hydropower equipment body, comprising the following steps:

[0009] S01. Establish an intelligent sensing system for the status of hydropower equipment to obtain status data of hydropower equipment;

[0010] S02. Establish a multi-source information fusion, feature extraction and self-learning analysis system based on hydropower equipment;

[0011] S03. Intelligently evaluate the status of hydropower equipment;

[0012] S04. Early warning of failures of hydropower equipment;

[0013] The above steps are used to realize the construction and evaluation of intelligent objects of hydropower equipment.

[0014] In a preferred solution, as in step S01, the intelligent sensing system includes sensors arranged on the hydropower equipment, including voltage and current sensors, vibration sensors, temperature sensors, pressure sensors and air gap sensors;

[0015] Voltage and current sensors are used to monitor the voltage and current parameters of hydropower equipment in real time, as well as the changing parameters of voltage and current;

[0016] Vibration sensors are used to collect vibration and vibration change parameters of hydropower equipment;

[0017] The temperature sensor is used to collect the temperature and temperature change parameters of the hot spot position of the hydropower equipment;

[0018] Pressure sensors are used to monitor the pressure and pressure change parameters of pipelines, containers or pressure parts of hydropower equipment;

[0019] The air gap sensor is used to monitor the air gap parameters and air gap change parameters between relatively rotating devices.

[0020] In the preferred solution, the voltage and current sensors use current transformers and voltage transformers, and combine clock parameters to obtain the change parameters of voltage and current;

[0021] In a preferred solution, the vibration sensor uses an accelerometer or a displacement sensor to collect high-frequency or low-frequency vibration parameters, and obtains vibration change parameters in combination with clock parameters;

[0022] The temperature sensor uses a thermocouple, a thermistor or an infrared sensor, and combines the clock parameters to obtain the temperature change parameters;

[0023] The pressure sensor uses a membrane pressure sensor or a piezoelectric element sensor, and combines the clock parameters to obtain the pressure change parameters;

[0024] The air gap sensor adopts a Hall sensor and combines the clock parameters to obtain the air gap variation parameters.

[0025] In a preferred solution, in step S01, the intelligent sensing system includes sensors installed on the hydropower equipment and also includes a three-dimensional space sensor. The three-dimensional space sensor uses a gyroscope, a sonar sensor or a laser point cloud collector to detect the spatial position, shape or distance of the hydropower equipment components, and obtains the spatial position change parameters of the hydropower equipment components in combination with the clock parameters.

[0026] The preferred solution is as in step S01, further comprising an acoustic sensor, an ultrasonic detection sensor, a thermal imaging detection sensor, a motor current waveform sensor and a video monitoring sensor;

[0027] Acoustic sensors are used to collect audio signals from hydropower equipment;

[0028] Ultrasonic detection sensors are used to analyze internal cracks or corrosion in hydropower equipment parts after the equipment is shut down;

[0029] Thermal imaging detection sensors are used to detect the temperature distribution of hydropower equipment;

[0030] The motor current waveform sensor is used to detect the harmonic components of the motor current;

[0031] Video surveillance sensors are used to monitor visual abnormal conditions related to hydropower equipment.

[0032] The preferred solution is as in step S02, comprising the following steps:

[0033] S021. Preprocess the collected data, including one or more combinations of data cleaning, data screening, data integration, data transformation and data specification, integrate the static data, operation data, real-time dynamic data and unstructured data of the hydropower equipment body, and build an ontology service database;

[0034] S022. Extract feature information from the preprocessed data to reduce the amount of data and improve processing efficiency;

[0035] S023. Obtain a weight coefficient of each feature information, and use the weight coefficient of each feature information to perform weighted fusion on each feature information.

[0036] The preferred solution is as in step S021, comprising the following steps:

[0037] S0211. Ensure that all data sources are correctly identified and integrated, and that timestamps are synchronized;

[0038] S2212, converting data in different formats into a unified format;

[0039] S2213, data splitting, if the data field contains multiple information, split it into separate columns to facilitate analysis;

[0040] S2214, Data check: Confirm that there are no missing values ​​in the data set, including completeness check, consistency check and outlier detection;

[0041] S2215, Data correction: including error correction, filling missing values ​​and duplicate data processing;

[0042] S2216, Data transformation: Converting data into a format suitable for analysis;

[0043] Data reduction: scaling the data to ensure that indicators of different magnitudes can be compared within the same analytical framework;

[0044] Data integration: integrating data from multiple systems or sensors into a consistent analytical framework or view;

[0045] S2217. Evaluate the quality of the preprocessed data.

[0046] The preferred solution is as in step S021, building the ontology service database includes the following steps:

[0047] The state access controller is used to collect various data, and the monitoring data is transmitted in real time and stored in the Hbase database through the data upload service.

[0048] In the preferred solution, as in step S021, data collection also comes from a Web service, including collecting equipment ledger information, defect records, test data, operating status, maintenance history, and manufacturer-related information from a production management system;

[0049] The Web service is based on two Web service interfaces: one is a real-time data transmission interface, which is used to send and receive data immediately; when any equipment ledger information is added, updated, or deleted in the production management system, the data is transmitted in real time through the Web service and updated to the Hbase database;

[0050] The other is a query receiving interface based on a time period, which sends data query requests to the production management system at set time intervals and updates the returned data to the Hbase database.

[0051] The preferred solution is as in step S022, which adopts an unsupervised feature learning method to construct an equipment body health status degradation index, uses data in the Hbase data service library to mine key features of the normal state of the equipment body, establishes a time domain degradation index based on parameter regression fitting and a frequency domain degradation index based on feature reconstruction singular values, and quantitatively evaluates the equipment health status or mines fault information in real time.

[0052] In a preferred solution, as in step S022, the time domain degradation index construction includes the following steps:

[0053] S022a1: Taking the historical operating conditions of the unit in a healthy state as input samples and the time domain characteristics of vibration parameters as output samples, a time domain health model of the hydropower unit based on the time series prediction model is established;

[0054] S022a2: Input the operating condition data in the real-time monitoring sample of the hydropower unit into the time domain health model, that is, obtain the time domain characteristic health value of the vibration signal at the current moment, calculate the relative error between the actual value and the theoretical value of the time domain characteristic, and use it as the time domain fault symptom indicator.

[0055] The preferred solution is as follows: in step S022a1, constructing a time series prediction model and a time domain health model specifically includes:

[0056] Construct a random configuration network, whose structure includes: input layer, hidden layer and output layer, and initialize the input weight ω and bias b:

[0057]

[0058] Among them, ω l and b l Represent the input weight and bias of the lth hidden node respectively; λ represents ω l and b l The scaling factor, D max is the maximum number of hidden nodes in the hidden layer;

[0059] Sigmoid is selected as the activation function, and the output of the lth hidden node is calculated. When the hidden layer node is L-1, the output value of SCN is:

[0060]

[0061] where μ l is the output weight of the lth hidden node; the network error is calculated using formula (3):

[0062] y L-1 =YY L-1 =[e L-1,1 ,e L-1 ,2,…,e L-1,R ] (3) Use the least squares method to calculate the hidden layer output weight:

[0063]

[0064] Continuously increase the number of hidden layer nodes, use formula (3) to minimize the error value, and finally output the optimal model.

[0065] In the preferred solution, the historical operating condition sample data X(t) of the unit under normal operating conditions is used as input, and the time domain feature sample Y(t) is used as output. The unit state health model is constructed through the time series prediction model. The mathematical expression is:

[0066] Y2(t)=f SCN [X(t)] (5)

[0067] Where, X(t) = {x1(t), x2(t)}, Y(t) = {y1(t), y2(t)}, x1(t) is the water head, x2(t) is the active power, y1(t) is the peak-to-peak value of the vibration signal in the time domain, and y2(t) is the standard deviation of the vibration signal in the time domain.

[0068] In the preferred solution, in the real-time monitoring stage, the real-time operating condition data of the unit is input into the established health model to obtain the theoretical health value of the time domain characteristics under the current corresponding working conditions, and the relative difference between the actual value and the theoretical value of the time domain characteristics of the vibration signal is calculated, and after weighted calculation, it is used as the fault symptom indicator of the unit vibration time domain direction; the expression is:

[0069]

[0070] Where V1(t) is the fault symptom index of the vibration signal in the time domain, V′1(t) and V′2(t) are the symptom indicators corresponding to the time domain characteristic components, d1(t) and d2(t) are the actual values ​​of the time domain characteristics, and ω1 and ω2 are the weight coefficients.

[0071] In order to enhance the sensitivity of the fault symptom index to abnormal data, the weight value in the above formula is:

[0072]

[0073] Through the above steps, the theoretical value of time domain characteristic health under the current corresponding working conditions is obtained.

[0074] The preferred solution is as in step S022, where the frequency domain degradation index based on feature-reconstructed singular values ​​includes the following steps:

[0075] Extract the wavelet singular value eigenvectors of vibration signals under similar working conditions, construct a wavelet singular value eigenvector set under the same working condition range, and use the K-means clustering algorithm to obtain the center vector of the vector set as the standard feature of the vibration signal under the corresponding working condition;

[0076] The wavelet singular value eigenvector of the corresponding vibration signal in the real-time monitoring sample of the unit is extracted, and its Euclidean distance with the center vector is calculated as the fault symptom indicator in the frequency domain.

[0077] In the preferred solution, as in step S022, the specific steps of reconstructing the frequency domain degradation index of the singular value based on the feature are:

[0078] S022b1: Decompose the signal using a wavelet decomposition algorithm to obtain a series of wavelet reconstruction coefficients. Based on the vibration signal of the historical operating condition samples of the unit under normal operating conditions, the wavelet decomposition coefficients obtained by signal decomposition are used as the input matrix A of the singular value decomposition;

[0079] S022b2: Perform singular value decomposition on the signal matrix A. There exists an orthogonal matrix, namely the unitary matrix U(U∈C m×m ) and the orthogonal matrix V(V∈C n×n ), so that the following conditions are met:

[0080]

[0081] The above formula is the decomposition formula of singular values, which shows that matrix A can be decomposed into the product of three matrices. Among them, S is a diagonal matrix, that is, S = diag (σ1, σ2, …, σ r ),σ1,σ2,…,σ r are the singular values ​​of the matrix A, and σ1≥σ2≥…≥σ r >0,r=rank(A);

[0082] Transforming the above formula, we can get:

[0083]

[0084] The obtained singular value vector S is the signal eigenvector;

[0085] S022b3: The Euclidean distance between the real-time vibration signal wavelet singular value vector and the cluster center C is used as the frequency domain fault symptom indicator, and the calculation formula is:

[0086]

[0087] Where V2(t) is the fault symptom index of the vibration signal in the frequency domain, and q(t) is the singular value eigenvector of the measured vibration signal at time t.

[0088] The preferred solution is as in step S022, introducing the calculation method of stream mining into the fault diagnosis of hydropower equipment, using the time domain rolling calculation method to perform real-time incremental update of data features;

[0089] Based on the acquired monitoring data, characteristic values ​​reflecting the operating status of the equipment are extracted, including the peak-to-peak value and the frequency-multiplied amplitude of the time domain characteristics.

[0090] According to the operating characteristics of the hydropower units, the operating water head of the hydropower units basically follows the fluctuation changes with an annual cycle. The corresponding unit states will have certain differences under different water heads, and will also be different under different operating conditions. Therefore, a real-time rolling calculation of the characteristic values ​​of equipment status indicators with one year as the basic unit is established.

[0091] In a preferred solution, as in step S022, extracting a characteristic value reflecting the operating status of the device includes the following steps:

[0092] S022c1: Use the head and active power of the hydropower unit as stability constraints to screen the equipment status monitoring data to ensure that the real-time data used to calculate the characteristic value is derived from samples of normal equipment operation;

[0093] S022c2: Taking the current time as the reference point, select the historical data of the past year as the basic data set for calculating the characteristic value of the device, mark the first data point used for calculation as the starting point, and count the total number of qualified health data samples participating in the calculation, recorded as N;

[0094] S022c3: Use a time-decayed weight distribution mechanism, where the sum of the weights of all data points is 1, set the weight of the starting data point to 1 / N^2, and then increase the weight ratio of subsequent data points in sequence according to the rule of arithmetic progression; the weight distribution is given by the following formula:

[0095]

[0096] Where N represents the number of healthy data samples entering the calculation;

[0097] S022c4: Assume that the data sequence currently entering the eigenvalue calculation is X1, X2, X3, ...X N , the device characteristic value is Y, then the characteristic value Y is calculated by the following formula:

[0098]

[0099] This allows the latest data to have more weight, reflecting the impact of real-time data changes on the eigenvalues;

[0100] Repeat steps S022c1 to S022c4 with the latest time as the calculation node to generate a new eigenvalue. If there is no latest data, no calculation is required this time, and the current eigenvalue is equal to the last calculated value. This process is repeated to generate new eigenvalues.

[0101] The preferred solution is as in step S022c4:

[0102] Taking the current time point as a marker, a statistical calculation is performed every L hours. Suppose each feature value calculated in the past 24 hours is Y1, Y2, Y3…Y k , the current time point yesterday is marked, and each feature value calculated 24 hours ago is Where k = 24 / L, and the average of the characteristic values ​​for the past 24 hours is calculated:

[0103]

[0104] in, It indicates the average value of the features calculated 24 hours before the current time point. It is the feature average value calculated 24 hours after the current time point.

[0105] Similarly, the calculation is recalculated at a time interval of L hours, and the average eigenvalue of the 24-hour eigenvalue is calculated repeatedly based on the continuous generation of new eigenvalues.

[0106] In the preferred solution, as in step S023, the weighted fusion of feature information includes the following steps:

[0107] The entropy weight method, the statistical signal processing method based on Bayesian theory or the DS evidence theory are used for weighted fusion of feature information.

[0108] In the preferred solution, as in step S023, the process of determining the information weight using the entropy weight method includes the following steps:

[0109] Characteristic sequences with different meanings for the same hydropower equipment or system a 11 ,a 21 ,…,a m1 ,Through the amount of information contained in the feature, the weight of the feature is determined using the ,entropy value of the feature sequence;

[0110] S0231. Construct an evaluation matrix based on the relevant information of the object to be evaluated;

[0111] When the number of features is m and the sequence length is n, the initial evaluation matrix A is obtained:

[0112]

[0113] Where a ij is the evaluation value of the jth indicator sequence of the i-th feature;

[0114] S0232. Make the evaluation matrix A dimensionless and transform it into a standard matrix (b ij ) m×n ;

[0115] The evaluation matrix A is dimensionless through the mean method, so that the data interval of each indicator is between 0 and 1;

[0116]

[0117] The processed normalized matrix is ​​B:

[0118]

[0119] S0233. Calculate the entropy value e of the feature i ;

[0120]

[0121] Where: j=1,2,…,n, when b ij = 0, b ij lnbij =0;

[0122] S0234. Determine the feature entropy weight γ i ;

[0123]

[0124] Finally, the weight sequence of each feature is obtained: (γ1,γ2,…,γ m );

[0125] Perform weighted fusion on each feature to obtain a new fusion feature.

[0126] The preferred solution is as in step S03, whereby the status of the hydropower equipment is intelligently evaluated, risk factors are dynamically determined, and support is provided for equipment operation decision-making;

[0127] Develop intelligent application tools based on the data warehouse model to deeply mine the associations and implicit information between data to accurately identify the fault type, fault severity and fault change trend;

[0128] Develop equipment maintenance plans based on data analysis results and provide corresponding decision-making recommendations for sudden equipment failures.

[0129] In a preferred solution, as in step S04, the fused features are compared with the features of the healthy sample library to obtain a judgment result. If the result does not meet the preset conditions, an alarm prompt is issued;

[0130] Based on the structure of hydropower equipment and the fault rule mechanism, the fault reasoning and identification are realized according to the extracted equipment status characteristics and fault symptoms, and the cause of the fault is traced.

[0131] The present invention provides a method for constructing an intelligent object of a hydropower equipment body. By constructing a full business process of multi-dimensional situation intelligent perception of hydropower equipment - multi-source information fusion - feature self-learning - fault self-diagnosis - maintenance self-decision-making, the comprehensive monitoring, evaluation and intelligent management of the hydropower equipment body are realized, the evaluation accuracy of the operating status of the hydropower equipment is greatly improved, early warning can be achieved, and the operating efficiency and safety of the equipment are improved. The multi-source information fusion and feature self-learning method of the hydropower equipment of the present invention ensures the consistency and applicability of the data, and provides reliable data support for diagnosis and operation decision-making. The self-warning and self-diagnosis method of hydropower equipment faults driven by multi-dimensional self-learning features promotes multi-time, multi-dimensional data sharing, experience iteration and reuse. The self-evaluation and intelligent maintenance decision-making method of the hydropower equipment status realizes the automatic iteration of knowledge and intelligent maintenance of hydropower equipment operation decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0132] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0133] Figure 1 A flow chart for constructing the hydropower equipment body intelligent object of the present invention.

[0134] Figure 2 It is a schematic diagram of the multi-source information fusion and feature self-learning process of hydropower equipment of the present invention.

[0135] Figure 3 It is a schematic flow chart of the hydropower equipment fault self-warning and self-diagnosis method of the present invention. DETAILED DESCRIPTION

[0136] Embodiment 1:

[0137] like Figures 1 to 3 In the present invention, a method for constructing an intelligent object of a hydropower equipment body comprises the following steps:

[0138] S01. Establish an intelligent sensing system for the status of hydropower equipment to obtain status data of hydropower equipment;

[0139] In a preferred solution, as in step S01, the intelligent sensing system includes sensors arranged on the hydropower equipment, including voltage and current sensors, vibration sensors, temperature sensors, pressure sensors and air gap sensors;

[0140] Voltage and current sensors are used to monitor the voltage and current parameters of hydropower equipment in real time, as well as the changing parameters of voltage and current;

[0141] Vibration sensors are used to collect vibration and vibration change parameters of hydropower equipment;

[0142] The temperature sensor is used to collect the temperature and temperature change parameters of the hot spot position of the hydropower equipment;

[0143] Pressure sensors are used to monitor the pressure and pressure change parameters of pipelines, containers or pressure parts of hydropower equipment;

[0144] The air gap sensor is used to monitor the air gap parameters and air gap change parameters between relatively rotating devices.

[0145] In the preferred solution, the voltage and current sensors use current transformers and voltage transformers, and combine clock parameters to obtain the change parameters of voltage and current;

[0146] In a preferred solution, the vibration sensor uses an accelerometer or a displacement sensor to collect high-frequency or low-frequency vibration parameters, and obtains vibration change parameters in combination with clock parameters;

[0147] The temperature sensor uses a thermocouple, a thermistor or an infrared sensor, and combines the clock parameters to obtain the temperature change parameters;

[0148] The pressure sensor uses a membrane pressure sensor or a piezoelectric element sensor, and combines the clock parameters to obtain the pressure change parameters;

[0149] The air gap sensor adopts a Hall sensor and combines the clock parameters to obtain the air gap variation parameters.

[0150] In a preferred solution, in step S01, the intelligent sensing system includes sensors arranged on the hydropower equipment, and also includes a three-dimensional space sensor, which uses a gyroscope, a sonar sensor or a laser point cloud collector to detect the spatial position, shape or distance of the hydropower equipment components. The three-dimensional information of objects in the environment is captured by transmitting and receiving different types of signals, such as light, sound waves or radar waves, and converted into digital data. These data are used to create three-dimensional models, measure distances, and achieve motion tracking, etc. The spatial position change parameters of the hydropower equipment components are obtained in combination with clock parameters.

[0151] The preferred solution is as in step S01, further comprising an acoustic sensor, an ultrasonic detection sensor, a thermal imaging detection sensor, a motor current waveform sensor and a video monitoring sensor;

[0152] Acoustic sensors are used to collect audio signals from hydropower equipment. They use microphones to capture the sound of equipment in operation and detect abnormal noise or knocking sounds through sound analysis, thereby providing early warning of equipment failures. They are also used to measure water levels, flow rates, and underwater object detection.

[0153] Ultrasonic detection sensors are used to analyze internal cracks or corrosion in hydropower equipment parts after the equipment is shut down;

[0154] Thermal imaging detection sensors are used to detect the temperature distribution of hydropower equipment; for example, by using a thermal imaging camera, the temperature distribution on the surface of the equipment can be detected to identify potential hot spots or cooling problems.

[0155] The motor current waveform sensor is used to detect the harmonic components of the motor current and diagnose the motor's operating status and potential faults.

[0156] Video surveillance sensors are used to monitor visual abnormal conditions related to hydropower equipment, and observe the visual status of equipment in real time, such as oil leakage, abnormal smoke, etc. They can also be used to monitor the operating status of managers.

[0157] It is further preferred to deploy sensor energy harvesters, such as solar panels, to achieve self-powering of the sensors, ensuring that the sensors can continue to operate in remote, wireless or continuous monitoring situations. Use networked wireless sensors, which consist of many wireless sensor nodes distributed in space and cooperating with each other to form a wireless intercommunication network to monitor, sense and collect information of various environmental monitoring objects in real time. Use IoT sensors to support remote monitoring and management. Use drones to load a variety of sensors, such as hyperspectral sensors, infrared thermal imaging sensors, cameras, etc. Optical sensors. Use optical principles and optoelectronic technology to convert optical signals into electrical signals to detect and measure specific parameters or environmental changes.

[0158] S02. Establish a multi-source information fusion, feature extraction and self-learning analysis system based on hydropower equipment;

[0159] S03. Intelligently evaluate the status of hydropower equipment;

[0160] S04. Early warning of failures of hydropower equipment;

[0161] The above steps are used to realize the construction and evaluation of intelligent objects of hydropower equipment.

[0162] Embodiment 2:

[0163] like Figure 2 In the preferred solution, as in step S02, the following steps are included:

[0164] S021. Preprocess the collected data, including one or more combinations of data cleaning, data screening, data integration, data transformation and data specification, integrate the static data, operation data, real-time dynamic data and unstructured data of the hydropower equipment body, and build an ontology service database;

[0165] S022. Extract feature information from the preprocessed data to reduce the amount of data and improve processing efficiency;

[0166] S023. Obtain a weight coefficient of each feature information, and use the weight coefficient of each feature information to perform weighted fusion on each feature information.

[0167] The preferred solution is as in step S021, comprising the following steps:

[0168] S0211. Ensure that all data sources are correctly identified and integrated, and that timestamps are synchronized;

[0169] S2212, converting data in different formats into a unified format;

[0170] S2213, data splitting, if the data field contains multiple information, split it into separate columns to facilitate analysis;

[0171] S2214, Data check: Confirm that there are no missing values ​​in the data set, including completeness check, consistency check and outlier detection;

[0172] S2215, Data correction: including error correction, filling missing values ​​and duplicate data processing;

[0173] S2216, Data transformation: Converting data into a format suitable for analysis;

[0174] Data reduction: scaling the data to ensure that indicators of different magnitudes can be compared within the same analytical framework;

[0175] Data integration: integrating data from multiple systems or sensors into a consistent analytical framework or view;

[0176] S2217. Evaluate the quality of the preprocessed data.

[0177] like Figure 2 In the preferred solution, as in step S021, constructing the ontology service database includes the following steps:

[0178] The state access controller is used to collect various data, and the monitoring data is transmitted in real time and stored in the Hbase database through the data upload service. The Hbase database is a distributed, unstructured, column-based storage database, which is very suitable for storing and managing large-scale monitoring data. HBase is suitable for scenarios that process large amounts of unstructured or semi-structured data. HBase adopts a column family storage model. Each table consists of multiple column families. Columns can be added dynamically within a column family without pre-defining the specific names of the columns. This design is conducive to efficient processing of sparse data and supports flexible data models. Key-value storage: In HBase, data is stored in the form of key-value pairs, where the key is defined by the row key (Row Key) and the column is determined by the column family and the column qualifier (Column Qualifier). The row key is used to uniquely identify a row of data and supports efficient range scanning. In HBase, data is stored in the form of key-value pairs, where the key is defined by the row key (Row Key) and the column is determined by the column family and the column qualifier (Column Qualifier). The row key is used to uniquely identify a row of data and supports efficient range scanning. Although HBase does not support complex SQL operations: such as JOIN, complex transaction processing and other operations. The query mainly relies on row keys and column families. This structure is very suitable for analyzing complex hydropower equipment data. The query is implemented using Apache Hive or Phoenix tools. All data in the HBase of the present invention are stored in the form of byte arrays (byte[]). Operations such as numerical calculations are required. The terminal program converts the data and then calculates and processes it by itself. This solution realizes the sharing and collaboration between big data and terminal operable data. Since HBase supports real-time read and write operations, it is suitable for application scenarios where the hydropower equipment of the present invention requires low-latency access, such as real-time data analysis, message systems, and user behavior tracking.

[0179] In the preferred solution, as in step S021, data collection also comes from a Web service, including collecting equipment ledger information, defect records, test data, operating status, maintenance history, and manufacturer-related information from a production management system;

[0180] The Web service is based on two Web service interfaces: one is a real-time data transmission interface, which is used to send and receive data immediately; when any equipment ledger information is added, updated, or deleted in the production management system, the data is transmitted in real time through the Web service and updated to the Hbase database;

[0181] The other is a query receiving interface based on a time period, which sends data query requests to the production management system at set time intervals and updates the returned data to the Hbase database.

[0182] Embodiment 3:

[0183] The preferred solution is as in step S022, which adopts an unsupervised feature learning method to construct an equipment body health status degradation index, uses data in the Hbase data service library to mine key features of the normal state of the equipment body, establishes a time domain degradation index based on parameter regression fitting and a frequency domain degradation index based on feature reconstruction singular values, and quantitatively evaluates the equipment health status or mines fault information in real time.

[0184] The core value of unsupervised feature learning lies in its ability to transform the raw data of hydropower equipment into higher-level and more expressive feature representations. These features often capture the essential characteristics of hydropower equipment data, thereby facilitating subsequent machine learning tasks such as classification, clustering, anomaly detection or recommendation systems. Unsupervised feature learning specifically includes the following aspects:

[0185] Dimensionality reduction: Reducing the dimension of data while maintaining important structure or variation information of the data, such as principal component analysis (PCA) and independent component analysis (ICA).

[0186] Encoding and decoding: Using neural network architectures such as autoencoders to extract features by forcing the model to learn efficient compression and reconstruction of data.

[0187] Cluster analysis: Although clustering itself is an unsupervised learning task, it can also be viewed as a feature learning process because it attempts to cluster similar data points together, implicitly defining the substructure of the data. Its application is to cluster the feature data that causes the corresponding faults, thereby reflecting the causes of the faults.

[0188] Dictionary learning: Representing data by learning a set of basis vectors (dictionary elements) so that the data can be represented by a linear combination of these basis vectors, such as sparse coding methods.

[0189] Manifold learning: It helps reveal the intrinsic geometric structure of hydropower equipment data by finding low-dimensional manifold structures in high-dimensional data, such as local linear embedding (LLE) and isometric mapping (Isomap).

[0190] Deep learning methods, such as deep belief networks (DBNs) and variational autoencoders (VAEs), play an important role in unsupervised feature learning. They can learn multi-layer abstract feature representations.

[0191] In a preferred solution, as in step S022, the time domain degradation index construction includes the following steps:

[0192] S022a1: Taking the historical operating conditions of the unit in a healthy state as input samples and the time domain characteristics of the vibration parameters as output samples, a time domain health model of the hydropower unit based on the time series prediction SCN model is established;

[0193] S022a2: Input the operating condition data in the real-time monitoring sample of the hydropower unit into the time domain health model, that is, obtain the time domain characteristic health value of the vibration signal at the current moment, calculate the relative error between the actual value and the theoretical value of the time domain characteristic, and use it as the time domain fault symptom indicator.

[0194] The preferred solution is as follows: in step S022a1, constructing a time series prediction model and a time domain health model specifically includes:

[0195] Construct a random configuration network, whose structure includes: input layer, hidden layer and output layer, and initialize the input weight ω and bias b:

[0196]

[0197] Among them, ω l and b l Represent the input weight and bias of the lth hidden node respectively; λ represents ω l and b l The scaling factor, D max is the maximum number of hidden nodes in the hidden layer;

[0198] Sigmoid is selected as the activation function, and the output of the lth hidden node is calculated. When the number of hidden layer nodes is L-1, the output value of SCN is:

[0199]

[0200] where μ l is the output weight of the lth hidden node; the network error is calculated using formula (3):

[0201] y L-1 =YY L-1 =[e L-1,1 ,e L-1,2 ,…,e L-1,R ] (3)

[0202] The output weight of the hidden layer is calculated using the least squares method:

[0203]

[0204] Continuously increase the number of hidden layer nodes, use formula (3) to minimize the error value, and finally output the optimal model.

[0205] In the preferred solution, the historical operating condition sample data X(t) of the unit under normal operating conditions is used as input, and the time domain feature sample Y(t) is used as output. The unit state health model is constructed through the time series prediction model. The mathematical expression is:

[0206] Y2(t)=fSCN [X(t)] (5)

[0207] Where, X(t) = {x1(t), x2(t)}, Y(t) = {y1(t), y2(t)}, x1(t) is the water head, x2(t) is the active power, y1(t) is the peak-to-peak value of the vibration signal in the time domain, and y2(t) is the standard deviation of the vibration signal in the time domain.

[0208] In the preferred solution, in the real-time monitoring stage, the real-time operating condition data of the unit is input into the established health model to obtain the theoretical health value of the time domain characteristics under the current corresponding working conditions, and the relative difference between the actual value and the theoretical value of the time domain characteristics of the vibration signal is calculated, and after weighted calculation, it is used as the fault symptom indicator of the unit vibration time domain direction; the expression is:

[0209]

[0210] Where V1(t) is the fault symptom index of the vibration signal in the time domain, V′1(t) and V′2(t) are the symptom indicators corresponding to the time domain characteristic components, d1(t) and d2(t) are the actual values ​​of the time domain characteristics, and ω1 and ω2 are the weight coefficients.

[0211] In order to enhance the sensitivity of the fault symptom index to abnormal data, the weight value in the above formula is:

[0212]

[0213] Through the above steps, the theoretical value of time domain characteristic health under the current corresponding working conditions is obtained.

[0214] The preferred solution is as in step S022, where the frequency domain degradation index based on feature-reconstructed singular values ​​includes the following steps:

[0215] Extract the wavelet singular value eigenvectors of vibration signals under similar working conditions, construct a wavelet singular value eigenvector set under the same working condition range, and use the K-means clustering algorithm to obtain the center vector of the vector set as the standard feature of the vibration signal under the corresponding working condition;

[0216] The wavelet singular value eigenvector of the corresponding vibration signal in the real-time monitoring sample of the unit is extracted, and its Euclidean distance with the center vector is calculated as the fault symptom indicator in the frequency domain.

[0217] In the preferred solution, as in step S022, the specific steps of reconstructing the frequency domain degradation index of the singular value based on the feature are:

[0218] S022b1: Decompose the signal using a wavelet decomposition algorithm to obtain a series of wavelet reconstruction coefficients. Based on the vibration signal of the historical operating condition samples of the unit under normal operating conditions, the wavelet decomposition coefficients obtained by signal decomposition are used as the input matrix A of the singular value decomposition;

[0219] S022b2: Perform singular value decomposition on the signal matrix A. There exists an orthogonal matrix, namely the unitary matrix U(U∈C m×m ) and the orthogonal matrix V(V∈C n×n ), so that the following conditions are met:

[0220]

[0221] The above formula is the decomposition formula of singular values, which shows that matrix A can be decomposed into the product of three matrices. Among them, S is a diagonal matrix, that is, S = diag (σ1, σ2, …, σ r ),σ1,σ2,…,σ r are the singular values ​​of the matrix A, and σ1≥σ2≥…≥σ r >0,r=rank(A);

[0222] Transforming the above formula, we can get:

[0223]

[0224] The obtained singular value vector S is the signal eigenvector;

[0225] S022b3: The Euclidean distance between the real-time vibration signal wavelet singular value vector and the cluster center C is used as the frequency domain fault symptom indicator, and the calculation formula is:

[0226]

[0227] Where V2(t) is the fault symptom index of the vibration signal in the frequency domain, and q(t) is the singular value eigenvector of the measured vibration signal at time t.

[0228] Embodiment 5:

[0229] like Figure 3 In the preferred solution, as in step S022, the calculation method of stream mining is introduced into the fault diagnosis of hydropower equipment, and the time domain rolling calculation method is used to perform real-time incremental updates on data features;

[0230] Stream mining is also known as manifold learning: it aims to find low-dimensional manifold structures in high-dimensional data, such as local linear embedding (LLE) and isometric mapping (Isomap), to help reveal the intrinsic geometric structure of hydropower equipment data.

[0231] Based on the acquired monitoring data, characteristic values ​​reflecting the operating status of the equipment are extracted, including the peak-to-peak value and the frequency-multiplied amplitude of the time domain characteristics.

[0232] Embodiment 6:

[0233] In a preferred solution, as in step S022, extracting a characteristic value reflecting the operating status of the device includes the following steps:

[0234] S022c1: Use the head and active power of the hydropower unit as stability constraints to screen the equipment status monitoring data to ensure that the real-time data used to calculate the characteristic value is derived from samples of normal equipment operation;

[0235] S022c2: Taking the current time as the reference point, select the historical data of the past year as the basic data set for calculating the characteristic value of the device, mark the first data point used for calculation as the starting point, and count the total number of qualified health data samples participating in the calculation, recorded as N;

[0236] S022c3: Use a time-decayed weight distribution mechanism, where the sum of the weights of all data points is 1, set the weight of the starting data point to 1 / N^2, and then increase the weight ratio of subsequent data points in sequence according to the rule of arithmetic progression; the weight distribution is given by the following formula:

[0237]

[0238] Where N represents the number of healthy data samples entering the calculation;

[0239] S022c4: Assume that the data sequence currently entering the eigenvalue calculation is X1, X2, X3, ...X N , the device characteristic value is Y, then the characteristic value Y is calculated by the following formula:

[0240]

[0241] This allows the latest data to have more weight, reflecting the impact of real-time data changes on the eigenvalues;

[0242] Because the change of equipment status is a slow and gradual process, in order to save computing resources, statistical calculations are performed at intervals of L hours. When a new calculation starts, the data of the first m hours of the previous calculation of the year is discarded, and the latest L hours of newly collected data are added. That is, steps S022c1 to S022c4 are repeated with the latest time as the calculation node to generate new eigenvalues. If there is no latest data, no calculation is required this time, and the current eigenvalue is equal to the last calculated value, and new eigenvalues ​​are generated repeatedly in this way.

[0243] Embodiment 7:

[0244] The preferred solution is as in step S022c4:

[0245] Taking the current time point as a marker, a statistical calculation is performed every L hours. Suppose each feature value calculated in the past 24 hours is Y1, Y2, Y3…Y k , the current time point yesterday is marked, and each feature value calculated 24 hours ago is Where k = 24 / L, and the average of the characteristic values ​​for the past 24 hours is calculated:

[0246]

[0247] in, It indicates the average value of the features calculated 24 hours before the current time point. It is the feature average value calculated 24 hours after the current time point.

[0248] Similarly, the calculation is recalculated at a time interval of L hours, and the average eigenvalue of the 24-hour eigenvalue is calculated repeatedly based on the continuous generation of new eigenvalues.

[0249] Embodiment 8:

[0250] Obtain the weight coefficients of each feature information, and use these weight coefficients to perform weighted fusion on each feature information. In the fusion process, preferably, a variety of data fusion technologies are used to further improve the effect of data processing, such as statistical signal processing methods based on Bayesian theory, DS evidence theory and entropy weight method. In addition, deep learning methods can also be used to extract spatial and temporal features of various types of data and fuse them.

[0251] In the preferred solution, as in step S023, the weighted fusion of feature information includes the following steps:

[0252] The entropy weight method, the statistical signal processing method based on Bayesian theory or the DS evidence theory are used for weighted fusion of feature information.

[0253] In the preferred solution, as in step S023, the process of determining the information weight using the entropy weight method includes the following steps:

[0254] Characteristic sequences with different meanings for the same hydropower equipment or system a 11 ,a 21 ,…,a m1 ,Through the amount of information contained in the feature, the weight of the feature is determined using the ,entropy value of the feature sequence;

[0255] S0231. Construct an evaluation matrix based on the relevant information of the object to be evaluated;

[0256] When the number of features is m and the sequence length is n, the initial evaluation matrix A is obtained:

[0257]

[0258] Where a ij is the evaluation value of the jth indicator sequence of the i-th feature;

[0259] S0232. Make the evaluation matrix A dimensionless and transform it into a standard matrix (b ij ) m×n ;

[0260] The evaluation matrix A is dimensionless through the mean method, so that the data interval of each indicator is between 0 and 1;

[0261]

[0262] The processed normalized matrix is ​​B:

[0263]

[0264] S0233. Calculate the entropy value e of the feature i ;

[0265]

[0266] Where: j=1,2,…,n, when b ij = 0, b ij lnb ij =0;

[0267] S0234. Determine the feature entropy weight γ i ;

[0268]

[0269] Finally, the weight sequence of each feature is obtained: (γ1,γ2,…,γ m );

[0270] Perform weighted fusion on each feature to obtain a new fusion feature.

[0271] The preferred solution is as in step S03, whereby the status of the hydropower equipment is intelligently evaluated, risk factors are dynamically determined, and support is provided for equipment operation decision-making;

[0272] Develop intelligent application tools based on the data warehouse model to deeply mine the associations and implicit information between data in order to accurately identify the type of fault, the severity of the fault and the changing trend of the fault; Intelligent application tools include data aggregation and intelligent push tools suitable for mobile phones and other tools, and clustering tools built based on deep machine learning. This tool can cluster the relevant data involved in the fault to facilitate the analysis of the cause of the fault, and continuously perform weighted iteration on the clustering results based on the analysis result data, that is, the data with a greater correlation with the hydropower equipment failure is increased in weight, while the data with a greater correlation with the hydropower equipment failure is reduced in weight, ultimately revealing the generation and development mechanism of the fault and forming a fault prediction mechanism.

[0273] According to the data analysis results, an equipment maintenance plan is formulated, and corresponding decision-making suggestions are provided for sudden equipment failures. Preferably, intelligent selection is made from historical maintenance plans, and suggestions are made. After a new maintenance plan is formed based on the hydropower equipment itself, it is iteratively optimized after the maintenance is completed, and then the newly formed maintenance plan is incorporated into the historical maintenance plan.

[0274] Embodiment 9:

[0275] like Figure 3 In the preferred solution, as in step S04, the fused features are compared with the features of the healthy sample library to obtain a judgment result. If the result does not meet the preset conditions, an alarm prompt is issued;

[0276] Based on the structure of hydropower equipment and the fault rule mechanism, according to the extracted equipment status characteristics and fault symptoms, the fault reasoning and identification are realized, and the cause of the fault is traced. Through the aforementioned data clustering processing and combined with the iteration of fault feedback, the hydropower equipment data is associated with the fault, and the weight between the corresponding data and the fault is analyzed, thereby revealing the fault generation mechanism.

[0277] The above-mentioned embodiments are only preferred technical solutions of the present invention and should not be regarded as limitations of the present invention. The embodiments and features in the embodiments of the present application can be arbitrarily combined with each other without conflict. The protection scope of the present invention shall be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for constructing an intelligent object of a hydropower equipment body, characterized in that The following steps are involved: S01. Establish an intelligent sensing system for the status of hydropower equipment to obtain status data of hydropower equipment; S02. Establish a multi-source information fusion, feature extraction and self-learning analysis system based on hydropower equipment; S03. Intelligently evaluate the status of hydropower equipment; S04. Early warning of failures of hydropower equipment; The above steps are used to realize the construction and evaluation of intelligent objects of hydropower equipment.

2. The method for constructing an intelligent object of a hydropower equipment body according to claim 1, characterized in that: In step S01, the intelligent sensing system includes sensors arranged on the hydropower equipment, including voltage and current sensors, vibration sensors, temperature sensors, pressure sensors and air gap sensors; Voltage and current sensors are used to monitor the voltage and current parameters of hydropower equipment in real time, as well as the changing parameters of voltage and current; Vibration sensors are used to collect vibration and vibration change parameters of hydropower equipment; The temperature sensor is used to collect the temperature and temperature change parameters of the hot spot position of the hydropower equipment; Pressure sensors are used to monitor the pressure and pressure change parameters of pipelines, containers or pressure parts of hydropower equipment; The air gap sensor is used to monitor the air gap parameters and air gap change parameters between relatively rotating devices.

3. The method for constructing an intelligent object of a hydropower equipment body according to claim 2, characterized in that: The voltage and current sensors use current transformers and voltage transformers and combine clock parameters to obtain the change parameters of voltage and current.

4. The method for constructing an intelligent object of a hydropower equipment body according to claim 2, characterized in that: The vibration sensor uses an accelerometer or a displacement sensor to collect high-frequency or low-frequency vibration parameters, and combines the clock parameters to obtain vibration change parameters; The temperature sensor uses a thermocouple, a thermistor or an infrared sensor, and combines the clock parameters to obtain the temperature change parameters; The pressure sensor uses a membrane pressure sensor or a piezoelectric element sensor, and combines the clock parameters to obtain the pressure change parameters; The air gap sensor adopts a Hall sensor and combines the clock parameters to obtain the air gap variation parameters.

5. The method for constructing an intelligent object of a hydropower equipment body according to claim 2, characterized in that: In step S01, the intelligent sensing system includes sensors installed on the hydropower equipment and also includes a three-dimensional space sensor. The three-dimensional space sensor uses a gyroscope, a sonar sensor or a laser point cloud collector to detect the spatial position, shape or distance of the hydropower equipment components, and obtains the spatial position change parameters of the hydropower equipment components in combination with clock parameters.

6. The method for constructing an intelligent object of a hydropower equipment body according to claim 2, characterized in that: In step S01, it also includes an acoustic sensor, an ultrasonic detection sensor, a thermal imaging detection sensor, a motor current waveform sensor and a video monitoring sensor; Acoustic sensors are used to collect audio signals from hydropower equipment; Ultrasonic detection sensors are used to analyze internal cracks or corrosion in hydropower equipment parts after the equipment is shut down; Thermal imaging detection sensors are used to detect the temperature distribution of hydropower equipment; The motor current waveform sensor is used to detect the harmonic components of the motor current; Video surveillance sensors are used to monitor visual abnormal conditions related to hydropower equipment.

7. The method for constructing an intelligent object of a hydropower equipment body according to claim 1, characterized in that the steps In S02, the following steps are included: S021. Preprocess the collected data, including one or more combinations of data cleaning, data screening, data integration, data transformation and data specification, integrate the static data, operation data, real-time dynamic data and unstructured data of the hydropower equipment body, and build an ontology service database; S022. Extract feature information from the preprocessed data to reduce the amount of data and improve processing efficiency; S023. Obtain a weight coefficient of each feature information, and use the weight coefficient of each feature information to perform weighted fusion on each feature information.

8. The method for constructing an intelligent object of a hydropower equipment body according to claim 7, characterized in that Step S021 includes the following steps: S0211. Ensure that all data sources are correctly identified and integrated, and that timestamps are synchronized; S2212, converting data in different formats into a unified format; S2213, data splitting, if the data field contains multiple information, split it into separate columns to facilitate analysis; S2214, Data check: Confirm that there are no missing values ​​in the data set, including completeness check, consistency check and outlier detection; S2215, Data correction: including error correction, filling missing values ​​and duplicate data processing; S2216, Data transformation: Converting data into a format suitable for analysis; Data reduction: scaling the data to ensure that indicators of different magnitudes can be compared within the same analytical framework; Data integration: integrating data from multiple systems or sensors into a consistent analytical framework or view; S2217. Evaluate the quality of the preprocessed data.

9. The method for constructing an intelligent object of a hydropower equipment body according to claim 7, characterized in that In step S021, constructing the ontology service database includes the following steps: The state access controller is used to collect various data, and the monitoring data is transmitted in real time and stored in the Hbase database through the data upload service.

10. The method for constructing an intelligent object of a hydropower equipment body according to claim 7, characterized in that In step S021, data collection also comes from Web services, including collecting equipment ledger information, defect records, test data, operating status, maintenance history and manufacturer-related information from the production management system; The Web service is based on two Web service interfaces: one is a real-time data transmission interface, which is used to send and receive data immediately; when any equipment ledger information is added, updated, or deleted in the production management system, the data is transmitted in real time through the Web service and updated to the Hbase database; The other is a query receiving interface based on a time period, which sends data query requests to the production management system at set time intervals and updates the returned data to the Hbase database.

11. The method for constructing an intelligent object of a hydropower equipment body according to claim 7, characterized in that In step S022, an unsupervised feature learning method is used to construct an equipment health status degradation index. The data in the Hbase data service library is used to mine the key features of the normal state of the equipment. A time domain degradation index based on parameter regression fitting and a frequency domain degradation index based on feature reconstruction singular values ​​are established to quantitatively evaluate the equipment health status or mine fault information in real time.

12. The method for constructing an intelligent object of a hydropower equipment body according to claim 11, characterized in that In step S022, the time domain degradation index construction includes the following steps: S022a1: Taking the historical operating conditions of the unit in a healthy state as input samples and the time domain characteristics of vibration parameters as output samples, a time domain health model of the hydropower unit based on the time series prediction model is established; S022a2: Input the operating condition data in the real-time monitoring sample of the hydropower unit into the time domain health model, that is, obtain the time domain characteristic health value of the vibration signal at the current moment, calculate the relative error between the actual value and the theoretical value of the time domain characteristic, and use it as the time domain fault symptom indicator.

13. The method for constructing an intelligent object of a hydropower equipment body according to claim 12, characterized in that In step S022a1: constructing a time series prediction model, the time domain health model specifically includes: Construct a random configuration network, whose structure includes: input layer, hidden layer and output layer, and initialize the input weight ω and bias b: Among them, ω l and b l Represent the input weight and bias of the lth hidden node respectively; λ represents ω l and b l The scaling factor, D max is the maximum number of hidden nodes in the hidden layer; Sigmoid is selected as the activation function, and the output of the lth hidden node is calculated. When the hidden layer node is L-1, the output value of SCN is: where μ l is the output weight of the lth hidden node; the network error is calculated using formula (3): and L-1 =YY L-1 =[e L-1,1 ,and L-1,2 ,…,and L-1,R ] (3) The output weight of the hidden layer is calculated using the least squares method: Continuously increase the number of hidden layer nodes, use formula (3) to minimize the error value, and finally output the optimal model.

14. The method for constructing an intelligent object of a hydropower equipment body according to claim 13, characterized in that : Taking the historical operating condition sample data X(t) of the unit under normal operation as input and the time domain feature sample Y(t) as output, the unit status health model is constructed through the time series prediction model. The mathematical expression is: Y2(t)=f SCN [X(t)] (5) Where, X(t) = {x1(t), x2(t)}, Y(t) = {y1(t), y2(t)}, x1(t) is the water head, x2(t) is the active power, y1(t) is the peak-to-peak value of the vibration signal in the time domain, and y2(t) is the standard deviation of the vibration signal in the time domain.

15. The method for constructing an intelligent object of a hydropower equipment body according to claim 14, characterized in that: In the real-time monitoring stage, the real-time operating data of the unit is input into the established health model to obtain the theoretical value of the time domain characteristic health under the current corresponding working condition, and the relative difference between the actual value and the theoretical value of the time domain characteristic of the vibration signal is calculated. After weighted calculation, it is used as the fault symptom indicator of the unit vibration time domain direction; the expression is: Where V1(t) is the fault symptom index of the vibration signal in the time domain, V′1(t) and V′2(t) are the symptom indicators corresponding to the time domain characteristic components, d1(t) and d2(t) are the actual values ​​of the time domain characteristics, and ω1 and ω2 are the weight coefficients. In order to enhance the sensitivity of the fault symptom index to abnormal data, the weight value in the above formula is: Through the above steps, the theoretical value of time domain characteristic health under the current corresponding working conditions is obtained.

16. The method for constructing an intelligent object of a hydropower equipment body according to claim 11, characterized in that In step S022, reconstructing the frequency domain degradation index of the singular value based on the feature includes the following steps: Extract the wavelet singular value eigenvectors of vibration signals under similar working conditions, construct a wavelet singular value eigenvector set under the same working condition range, and use the K-means clustering algorithm to obtain the center vector of the vector set as the standard feature of the vibration signal under the corresponding working condition; The wavelet singular value eigenvector of the corresponding vibration signal in the real-time monitoring sample of the unit is extracted, and its Euclidean distance with the center vector is calculated as the fault symptom indicator in the frequency domain.

17. The method for constructing an intelligent object of a hydropower equipment body according to claim 16, characterized in that In step S022, the specific steps of reconstructing the frequency domain degradation index of the singular value based on the feature are: S022b1: Decompose the signal using a wavelet decomposition algorithm to obtain a series of wavelet reconstruction coefficients. Based on the vibration signal of the historical operating condition samples of the unit under normal operating conditions, the wavelet decomposition coefficients obtained by signal decomposition are used as the input matrix A of the singular value decomposition; S022b2: Perform singular value decomposition on the signal matrix A. There exists an orthogonal matrix, namely the unitary matrix U(U∈C m×m ) and the orthogonal matrix V(V∈C n×n ), so that the following conditions are met: The above formula is the decomposition formula of singular values, which shows that matrix A can be decomposed into the product of three matrices. Among them, s is a diagonal matrix, that is, S = diag (σ1, σ2,…, σ r ),σ1,σ2,…,σ r are the singular values ​​of the matrix A, and σ1≥σ2≥…≥σ r >0,r=rank(A); Transforming the above formula, we can get: The obtained singular value vector S is the signal eigenvector; S022b3: The Euclidean distance between the real-time vibration signal wavelet singular value vector and the cluster center C is used as the frequency domain fault symptom indicator, and the calculation formula is: Where V2(t) is the fault symptom index of the vibration signal in the frequency domain, and q(t) is the singular value eigenvector of the measured vibration signal at time t.

18. The method for constructing an intelligent object of a hydropower equipment body according to claim 7, characterized in that In step S022, the calculation method of stream mining is introduced into the fault diagnosis of hydropower equipment, and the time domain rolling calculation method is used to perform real-time incremental update of data features; Based on the acquired monitoring data, characteristic values ​​reflecting the operating status of the equipment are extracted, including the peak-to-peak value and the frequency-multiplied amplitude of the time domain characteristics.

19. The method for constructing an intelligent object of a hydropower equipment body according to claim 7, characterized in that In step S022, extracting the characteristic value reflecting the operation status of the device includes the following steps: S022c1: Use the head and active power of the hydropower unit as stability constraints to screen the equipment status monitoring data to ensure that the real-time data used to calculate the characteristic value is derived from samples of normal equipment operation; S022c2: Taking the current time as the reference point, select the historical data of the past year as the basic data set for calculating the characteristic value of the device, mark the first data point used for calculation as the starting point, and count the total number of qualified health data samples participating in the calculation, recorded as N; S022c3: Use a time-decayed weight distribution mechanism, where the sum of the weights of all data points is 1, set the weight of the starting data point to 1 / N^2, and then increase the weight ratio of subsequent data points in sequence according to the rule of arithmetic progression; the weight distribution is given by the following formula: Where N represents the number of healthy data samples entering the calculation; S022c4: Assume that the data sequence currently entering the eigenvalue calculation is X1, X2, X3, ...X N , the device characteristic value is Y, then the characteristic value Y is calculated by the following formula: This allows the latest data to have more weight, reflecting the impact of real-time data changes on the eigenvalues; Repeat steps S022c1 to S022c4 with the latest time as the calculation node to generate a new eigenvalue. If there is no latest data, no calculation is required this time, and the current eigenvalue is equal to the last calculated value. This process is repeated to generate new eigenvalues.

20. The method for constructing an intelligent object of a hydropower equipment body according to claim 19, characterized in that In step S022c4: Taking the current time point as a marker, a statistical calculation is performed every L hours. Suppose each feature value calculated in the past 24 hours is Y1, Y2, Y3…Y k , the current time point yesterday is marked, and each feature value calculated 24 hours ago is Where k = 24 / L, and the average of the characteristic values ​​for the past 24 hours is calculated: in, It indicates the average value of the features calculated 24 hours before the current time point. It is the feature average value calculated 24 hours after the current time point. Similarly, the calculation is recalculated at a time interval of L hours, and the average eigenvalue of the 24-hour eigenvalue is calculated repeatedly based on the continuous generation of new eigenvalues.

21. A method for constructing an intelligent object of a hydropower equipment body according to any one of claims 7, 19 and 20, characterized in that In step S023, the weighted fusion of feature information includes the following steps: The entropy weight method, the statistical signal processing method based on Bayesian theory or the DS evidence theory are used for weighted fusion of feature information.

22. A method for constructing an intelligent object of a hydropower equipment body according to claim 21, characterized in that In step S023, the process of determining the information weight using the entropy weight method includes the following steps: Characteristic sequences with different meanings for the same hydropower equipment or system a 11 ,a 21 ,…,a m1 ,Through the amount of information contained in the feature, the weight of the feature is determined using the ,entropy value of the feature sequence; S0231. Construct an evaluation matrix based on the relevant information of the object to be evaluated; When the number of features is m and the sequence length is n, the initial evaluation matrix A is obtained: Where a ij is the evaluation value of the jth indicator sequence of the i-th feature; S0232. Make the evaluation matrix A dimensionless and transform it into a standard matrix (b ij ) m×n ; The evaluation matrix A is dimensionless through the mean method, so that the data interval of each indicator is between 0 and 1; The processed normalized matrix is ​​B: S0233. Calculate the entropy value e of the feature i ; Where: j=1,2,…,n, when b ij = 0, b ij lnb ij =0; S0234. Determine the feature entropy weight γ i ; Finally, the weight sequence of each feature is obtained: (γ1,γ2,…,γ m ); Perform weighted fusion on each feature to obtain a new fusion feature.

23. The method for constructing an intelligent object of a hydropower equipment body according to claim 1, characterized in that the steps In S03, the status of hydropower equipment is intelligently evaluated, risk factors are dynamically determined, and support is provided for equipment operation decision-making; Develop intelligent application tools based on the data warehouse model to deeply mine the associations and implicit information between data to accurately identify the fault type, fault severity and fault change trend; Develop equipment maintenance plans based on data analysis results and provide corresponding decision-making recommendations for sudden equipment failures.

24. A method for constructing an intelligent object of a hydropower equipment body according to any one of claims 1 to 22, characterized in that the steps In S04, the fused features are compared with the features of the healthy sample library to obtain a judgment result. If the result does not meet the preset conditions, an alarm prompt is issued; Based on the structure of hydropower equipment and the fault rule mechanism, the fault reasoning and identification are realized according to the extracted equipment status characteristics and fault symptoms, and the cause of the fault is traced.

Images