CN-121980443-A - On-line monitoring method and system for stress characteristics of hydroelectric generating set

CN121980443ACN 121980443 ACN121980443 ACN 121980443ACN-121980443-A

Abstract

The invention discloses a method and a system for on-line monitoring of the stress characteristics of a hydroelectric generating set, wherein the method comprises the following steps: a01: the stress data is collected in a multi-dimensional mode; a02: carrying out multi-source data fusion pretreatment; a03: stress characteristics are adaptively extracted, and A04: modeling and evaluating stress states; a05: the system comprises an intelligent sensing node of the Internet of things, a data preprocessing module, a feature extraction module, a state evaluation module, an early warning feedback module and a digital twin calibration module; according to the invention, a SVM-LSTM hybrid model is used, a normal state classifier is established by the SVM module, the LSTM module outputs a stress prediction value and calculates the deviation degree, the hybrid model combines the SVM classification advantage and the LSTM prediction capability, a normal state benchmark is accurately established, the real-time stress state deviation degree is quantized, a scientific basis is provided for state evaluation, the evaluation result is compared with a digital twin model simulation stress value, and model parameters are corrected.

Inventors

MENG PENG
LIU PENGFEI
ZHAO LEI
WANG HONGTENG
SUN BO
WANG LEI
BAI GUANGHUI
XU YANHE
FENG JIANJUN
WANG DAQUAN

Assignees

华电电力科学研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20251229

Claims (10)

1. The on-line monitoring method for the stress characteristics of the hydroelectric generating set is characterized by comprising the following steps of: A01, stress data are collected in a multi-dimensional mode, intelligent sensing nodes of the Internet of things are arranged at stress concentration positions of a rotating wheel, a main shaft and a frame of the hydroelectric generating set, stress values, vibration signals and temperature parameters are synchronously collected, and the stress values, the vibration signals and the temperature parameters are transmitted to a local data aggregation node through a wireless transmission module; a02, multisource data fusion preprocessing, namely sequentially performing wavelet threshold denoising, laida criterion outlier correction and space-time alignment processing on the original data acquired in the step A01 to acquire a fusion data set; a03, adaptively extracting stress features, extracting local spatial features and time sequence dependent features by adopting a CNN-RNN fusion model based on a fusion data set, and generating feature vectors by combining temperature stress correlation analysis; a04, modeling and evaluating the stress state, inputting the feature vector in the step A03 into an SVM-LSTM hybrid model, establishing a normal state reference and calculating the real-time stress state deviation degree; and A05, stress early warning and dynamic feedback, triggering multi-stage early warning according to the deviation degree in the step A04, and dynamically adjusting the sampling rate and updating the model parameters to form closed-loop monitoring.
2. The on-line monitoring method of the stress characteristics of the hydroelectric generating set according to claim 1, wherein the intelligent sensing node of the internet of things in the step A01 comprises a fiber bragg grating stress sensor, a piezoelectric acceleration sensor and a temperature sensor, and the wireless transmission module adopts a LoRaWAN protocol to realize data transmission.
3. The method for on-line monitoring of the stress characteristics of the hydroelectric generating set according to claim 1, wherein the wavelet threshold denoising in the step a02 adopts the following formula: ; Wherein, the For the wavelet transform coefficients, In order for the post-denoising coefficient to be a good value, As a result of the threshold value being set, Is the standard deviation of the noise, which is the standard deviation of the noise, For the data length, the noise component is suppressed by a soft threshold function.
4. The method for on-line monitoring of the stress characteristics of the hydroelectric generating set according to claim 1, wherein in the step A02, the time-space alignment is realized by a time stamp synchronization algorithm, the time errors of different sensor data are controlled within +/-1 ms, and the space coordinates are mapped and matched by a preset three-dimensional model coordinate of the hydroelectric generating set.
5. The method for on-line monitoring of stress characteristics of a hydroelectric generating set according to claim 1, wherein the working process of the SVM-LSTM hybrid model in step a04 comprises: the SVM module adopts RBF kernel functions to establish a normal state classifier, and the establishment formula is specifically as follows: ; Wherein, the As a function of the kernel, In order to be a lagrange multiplier, Is a label; LSTM module for outputting stress prediction value The degree of deviation is calculated by the following formula: wherein As a result of the actual measurement of the value, For normalized deviation.
6. The method for on-line monitoring of stress characteristics of a hydroelectric generating set according to claim 1, wherein the step a05 of multi-stage early warning comprises: Safety threshold Early warning threshold Alarm threshold The sampling rate is automatically increased to 2kHz in the early warning state, and the learning rate of the SVM-LSTM hybrid model is updated through an incremental learning algorithm, wherein the formula is expressed as follows: The formula indicates the learning rate Will follow Is decreased by an increase in (c).
7. The method for on-line monitoring of the stress characteristics of a hydroelectric generating set according to claim 1, wherein the step a04 further comprises digital twin dynamic calibration, specifically: Comparing the stress state deviation result with a simulation stress value of a digital twin model of the unit, and correcting model parameters by the following formula: ; Wherein, the As a parameter of the model, it is possible to provide, In order for the rate of learning to be high, As a loss function.
8. The on-line monitoring method of the stress characteristics of the hydroelectric generating set according to claim 1, wherein the CNN-RNN fusion model in the step a03 comprises a convolution layer and an LSTM layer, the convolution layer is used for extracting spatial features, and the LSTM layer is used for capturing time sequence features.
9. The system is used for executing the method of claim 1, and comprises an intelligent sensing node of the Internet of things, a data preprocessing module, a feature extraction module, a state evaluation module, an early warning feedback module and a digital twin calibration module.
10. The system of claim 9, wherein the intelligent sensing node of the internet of things is used for collecting stress values, vibration signals and temperature parameters, the data preprocessing module is used for preprocessing data obtained by the intelligent sensing node of the internet of things, the feature extraction module is used for adaptively extracting stress features, the state evaluation module is used for modeling and evaluating stress states, the early warning feedback module is used for stress early warning and dynamic feedback processing, and the digital twin calibration module is used for comparing results of the state evaluation module with simulated stress values of a digital twin model of the unit and realizing dynamic calibration according to corrected model parameters.

Description

On-line monitoring method and system for stress characteristics of hydroelectric generating set Technical Field The invention relates to the field of hydroelectric machine stress monitoring, in particular to a method and a system for on-line monitoring of the stress characteristics of a hydroelectric machine set. Background With the development of the hydropower industry to the large-scale and intelligent directions, the running working conditions of the hydroelectric generating set are increasingly complex, key components such as a rotating wheel, a main shaft and a frame are easy to cause fatigue damage and even structural failure due to stress concentration under the action of long-term dynamic load, the safety and stable running of the set and the reliability of an electric power system are seriously threatened, and the limitations of long detection period, poor data continuity, incapability of reflecting dynamic stress change in real time and the like exist in the traditional offline detection and single-point stress monitoring method, so that the requirements of the refined operation and maintenance of modern hydropower equipment are difficult to meet; In recent years, the rapid development of the Internet of things, artificial intelligence and digital twin technology provides a new approach for monitoring the stress of the hydroelectric generating set, the real-time acquisition of multi-source data is realized by deploying intelligent sensors at key positions, and stress characteristics can be effectively extracted and equipment states can be predicted by combining data fusion and a deep learning algorithm, however, the prior art still faces the challenges of multi-source heterogeneous data noise interference, complex stress characteristic space-time coupling relation, insufficient model self-adaption capability and the like. Disclosure of Invention The present invention is directed to a method and a system for on-line monitoring of the stress characteristics of a hydroelectric generating set in order to solve the above-mentioned problems. The aim of the invention can be achieved by the following technical scheme: In a first aspect, the invention provides a method for online monitoring stress characteristics of a hydroelectric generating set, comprising the following steps: A01, stress data are collected in a multi-dimensional mode, intelligent sensing nodes of the Internet of things are arranged at stress concentration positions of a rotating wheel, a main shaft and a frame of the hydroelectric generating set, stress values, vibration signals and temperature parameters are synchronously collected, and the stress values, the vibration signals and the temperature parameters are transmitted to a local data aggregation node through a wireless transmission module; a02, multisource data fusion preprocessing, namely sequentially performing wavelet threshold denoising, laida criterion outlier correction and space-time alignment processing on the original data acquired in the step A01 to acquire a fusion data set; a03, adaptively extracting stress features, extracting local spatial features and time sequence dependent features by adopting a CNN-RNN fusion model based on a fusion data set, and generating feature vectors by combining temperature stress correlation analysis; a04, modeling and evaluating the stress state, inputting the feature vector in the step A03 into an SVM-LSTM hybrid model, establishing a normal state reference and calculating the real-time stress state deviation degree; and A05, stress early warning and dynamic feedback, triggering multi-stage early warning according to the deviation degree in the step A04, and dynamically adjusting the sampling rate and updating the model parameters to form closed-loop monitoring. The intelligent sensing node of the Internet of things in the step A01 comprises a fiber bragg grating stress sensor, a piezoelectric acceleration sensor and a temperature sensor, and the wireless transmission module adopts the LoRaWAN protocol to realize data transmission. The various high-precision sensors can comprehensively and accurately collect the stress, vibration and temperature data of the hydroelectric generating set, key operation parameters are covered, stable data transmission in a long distance is ensured by the application of the LoRaWAN protocol, wiring cost and construction difficulty are reduced, and deployment flexibility and reliability of a monitoring system are improved. Further, the wavelet threshold denoising in the step a02 adopts the following formula: ; Wherein, the For the wavelet transform coefficients,In order for the post-denoising coefficient to be a good value,As a result of the threshold value being set,Is the standard deviation of the noise, which is the standard deviation of the noise,For the data length, the noise component is suppressed by a soft threshold function. Based on a wavelet threshold denoising algorithm, high-frequency noise in