Data Prediction Method and System for Thermal Power Plant Equipment

Abstract

The present invention discloses a data prediction method and system for thermal power plant equipment, specifically relating to the technical field of power equipment condition monitoring, and is used to solve the problem that the existing prediction of thermal power plant equipment leads to misjudgment of the health state due to the confusion between the working condition noise and the fault signal characteristics under dynamic working conditions; by collecting multi-source operation data and classifying them into load working condition intervals, extracting time-frequency domain characteristics to construct a working condition feature set; dynamically detecting the load switching frequency and analyzing the cross-condition consistency to mark the abnormal working condition intervals; combining the historical degradation stage matching results to decompose the non-abnormal working condition characteristics to generate a degradation feature subset; correcting the degradation features through the coupling of mechanical dynamics and thermodynamic parameters; finally, inversely positioning the fault source equipment along the fault propagation map and generating maintenance instructions; realizing the separation of dynamic working condition characteristics and the fusion of physical mechanisms, effectively improving the accuracy of thermal power plant equipment degradation prediction and the reliability of maintenance decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment status monitoring, and more specifically, to a data prediction method and system for thermal power plant equipment. Background Art

[0002] As the core energy supply unit of the power system, the predictive health management of key rotating equipment (such as steam turbines, boiler feed pumps, fans, etc.) of thermal power plants is a key direction of the industry's intelligent transformation; the increasingly frequent peak-shaving demand of the power grid requires the equipment to dynamically switch operation within the load range, resulting in complex non-steady-state characteristics of the monitoring parameters (vibration, temperature, pressure, etc.); the existing technology mainly constructs a health assessment model through multi-source sensor data fusion, and relies on threshold alarms, statistical filtering and other methods to identify equipment degradation trends. However, on the one hand, the operating noise caused by load fluctuations is highly coupled with the actual degradation signal of the equipment in the time and frequency domain, which is difficult to separate through conventional feature extraction methods; on the other hand, the data-driven model lacks embedded constraints on the physical mechanism of the equipment, resulting in the prediction results may deviate from the actual operating rules.

[0003] In the prior art, the prediction method for thermal power plant equipment has the defect of feature confusion under dynamic conditions. Specifically, the drastic fluctuations in the operating parameters of thermal power plant equipment during the peak-shaving process will mask the true degradation characteristics and cannot effectively distinguish between operating noise and fault signals. This feature confusion phenomenon will lead to misjudgment of the health status of thermal power plant equipment, causing the risk of false alarms or missed detections, thereby affecting the accuracy and reliability of predictive maintenance of thermal power plant equipment. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a data prediction method and system for thermal power plant equipment to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The data prediction method of thermal power plant equipment includes the following steps:

[0007] S1. Collect and pre-process the operating status data of thermal power plant equipment;

[0008] S2. Classify the operating status data according to the preset load condition interval, extract the time-frequency domain features within each load condition interval, and generate a condition feature set;

[0009] S3, detecting the switching frequency of the load condition interval, and if the switching frequency exceeds the switching frequency threshold, starting the dynamic analysis strategy switching to mark the abnormal condition interval;

[0010] S4. Decompose the set of operating condition features in the non-abnormal operating condition interval according to the similarity matching result between the historical degradation stage of the equipment and the current time-frequency domain features to generate subsets of equipment degradation features;

[0011] S5. Based on the coupling relationship between the mechanical dynamics parameters and the thermodynamic parameters of the equipment, perform parameter coupling correction on the subsets of equipment degradation features;

[0012] S6. Based on the corrected subsets of equipment degradation features, trace back the potential fault source equipment reversely along the equipment fault propagation map, determine the maintenance priority and generate maintenance instructions.

[0013] In a preferred embodiment, collect and preprocess the operating state data of the thermal power plant equipment, including:

[0014] Synchronously collect the vibration data, temperature data and pressure data of the thermal power plant equipment, and align the time stamps of the vibration data, temperature data and pressure data;

[0015] Perform data cleaning on the vibration data, temperature data and pressure data after time stamp alignment. The data cleaning includes using a sliding window filtering method to remove the high-frequency noise components in the vibration data, and removing abnormal outliers based on the correlation between the temperature data and the pressure data;

[0016] Resample the cleaned vibration data, temperature data and pressure data at a preset sampling frequency to generate a standardized operating state data set.

[0017] In a preferred embodiment, classify the operating state data according to the preset load operating condition intervals, extract the time-frequency domain features in each load operating condition interval, and generate a set of operating condition features, including:

[0018] Divide the standardized operating state data set into multiple load operating condition intervals according to the load value. The load operating condition intervals include low load intervals, medium load intervals and high load intervals;

[0019] Extract the time-frequency domain features of the vibration data, temperature data and pressure data in each load operating condition interval respectively;

[0020] Merge the time-frequency domain features in the same load operating condition interval according to the time window to generate a multi-dimensional set of operating condition features.

[0021] In a preferred embodiment, the time-frequency domain features of the vibration data include the vibration spectrum energy distribution, the root mean square value in the time domain and the centroid frequency in the frequency domain. The time-frequency domain features of the temperature data include the temperature change rate and the temperature fluctuation amplitude. The time-frequency domain features of the pressure data include the pressure fluctuation frequency and the pressure gradient.

[0022] In a preferred embodiment, the switching frequency of the detected load operating condition interval is detected. If the switching frequency exceeds the switching frequency threshold, a dynamic analysis strategy switch is initiated to mark the abnormal operating condition interval, including:

[0023] Based on a preset time window, count the number of switches in the load operating condition interval, calculate the switching frequency per unit time, and determine whether the switching frequency exceeds the switching frequency threshold;

[0024] If the switching frequency exceeds the switching frequency threshold, align the time-frequency domain features of adjacent load operating condition intervals by the dynamic time warping method, and extract the cross-condition consistency index;

[0025] If the cross-condition consistency index is lower than the consistency threshold, mark the load operating condition interval corresponding to the current time window as an abnormal operating condition interval, and trigger the re-extraction of the time-frequency domain features within the abnormal operating condition interval; otherwise, mark it as a non-abnormal operating condition interval.

[0026] In a preferred embodiment, extracting the cross-condition consistency index includes: performing alignment processing on the time-frequency domain features of adjacent load operating condition intervals, calculating the Pearson correlation coefficients for the aligned vibration spectrum energy distribution, temperature change rate, and pressure gradient respectively, and taking the average value of the Pearson correlation coefficients as the cross-condition consistency index.

[0027] In a preferred embodiment, decomposing the condition feature set of the non-abnormal operating condition interval according to the similarity matching result between the historical degradation stage of the device and the current time-frequency domain features to generate a device degradation feature subset, including:

[0028] For the current time-frequency domain features in the condition feature set of the non-abnormal operating condition interval, judge the matching degree between the current time-frequency domain features and the historical degradation stage of the device according to the similarity calculation result between the historical degradation stage label of the device and the current time-frequency domain features;

[0029] Among them, the historical degradation stage label of the device is divided by the vibration amplitude trend and temperature gradient trend in the historical operation data of the device;

[0030] If the matching degree is greater than or equal to the preset matching threshold, extract the device degradation feature subset from the condition feature set of the non-abnormal operating condition interval using the low-frequency narrowband decomposition mode;

[0031] If the matching degree is less than the preset matching threshold, based on the correlation between the harmonic component distribution of the vibration spectrum in the condition feature set of the non-abnormal operating condition interval and the critical speed of the rotor dynamic parameters, select the frequency band with concentrated harmonic energy for enhanced decomposition to generate the device degradation feature subset.

[0032] In a preferred embodiment, based on the coupling relationship between the mechanical dynamics parameters and the thermodynamic parameters of the device, parameter coupling correction is performed on the device degradation feature subset, including:

[0033] According to the correlation between the rotor speed and the vibration amplitude in the mechanical dynamics parameters of the device, amplitude-frequency characteristic correction is performed on the vibration amplitude in the device degradation feature subset;

[0034] Based on the matching relationship between the thermal expansion coefficient and the temperature gradient in the thermodynamic parameters, thermal stress compensation correction is performed on the temperature gradient in the device degradation feature subset;

[0035] The corrected vibration amplitude and temperature gradient are combined with the pressure gradient feature to generate a device degradation feature subset after parameter coupling correction.

[0036] In a preferred embodiment, based on the corrected device degradation feature subset, the potential fault source devices are traced back along the device fault propagation map, and the maintenance priority is determined and a maintenance instruction is generated, including:

[0037] Obtain the device fault propagation map, which is constructed based on the mechanical connection relationship and energy transfer path of the device, and includes the fault propagation direction and weight between device nodes;

[0038] Extract abnormal parameters from the device degradation feature subset after parameter coupling correction, and match the corresponding nodes in the device fault propagation map according to the type of abnormal parameters;

[0039] Traverse backward along the fault propagation direction of the device fault propagation map to locate the potential fault source device node of the abnormal parameter;

[0040] Determine the maintenance priority according to the number and position depth of the abnormal parameters of the potential fault source device node;

[0041] Based on the maintenance priority and the remaining service life of the device, a maintenance instruction including the fault source device identifier and the maintenance time is generated.

[0042] On the other hand, the present invention provides a data prediction system for thermal power plant equipment, including:

[0043] Data acquisition and processing module: Collect and preprocess the operation status data of thermal power plant equipment;

[0044] Working condition classification and extraction module: Classify the operation status data according to the preset load working condition interval, extract the time-frequency domain features in each load working condition interval, and generate a working condition feature set;

[0045] Dynamic strategy anomaly marking module: Detect the switching frequency of the load working condition interval. If the switching frequency exceeds the switching frequency threshold, start the dynamic analysis strategy switching to mark the abnormal working condition interval;

[0046] Degradation feature decomposition module: Decompose the set of operating condition features in the non-abnormal operating condition interval according to the similarity matching result between the historical degradation stage of the device and the current time-frequency domain features, and generate a device degradation feature subset.

[0047] Parameter coupling correction module: Based on the coupling relationship between the mechanical dynamics parameters and thermodynamic parameters of the device, perform parameter coupling correction on the device degradation feature subset.

[0048] Fault backtracking decision module: Based on the corrected device degradation feature subset, backtrack the potential fault source device along the device fault propagation map in reverse, determine the maintenance priority, and generate maintenance instructions.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. Through the separation of dynamic operating condition features and the integration of physical mechanisms, the accuracy of the health state assessment of thermal power plant equipment and the reliability of maintenance decisions are significantly improved; based on the dynamic classification of load operating condition intervals and time-frequency domain feature decomposition technology, the operating condition noise and the true degradation signal of the equipment during the peak shaving process are effectively distinguished, avoiding misjudgment problems caused by feature confusion under dynamic operating conditions; by detecting the load switching frequency and triggering the abnormal marking mechanism, data interference caused by sudden changes in operating conditions can be actively excluded, and at the same time, combined with the similarity matching of historical degradation stages, the feature decomposition mode is adaptively selected to ensure the feature extraction accuracy in different degradation stages, not only suppressing noise interference, but also achieving stable capture of degradation features, providing a high-quality data basis for subsequent analysis.

[0051] 2. Through the coupling correction of mechanical dynamics and thermodynamic parameters, the physical mechanism of the device is embedded in the data-driven model, making the prediction results strictly conform to the actual operation law; based on the reverse backtracking mechanism of the fault propagation map, the potential fault source device can be located from the system level, and combined with the priority evaluation of multi-dimensional degradation features, accurate maintenance instructions are generated, which not only avoids the prediction deviation caused by the traditional model deviating from the device mechanism, but also optimizes the allocation efficiency of maintenance resources through system-level fault location, and finally realizes the improvement of the full-link reliability from local anomaly detection to global maintenance decision-making. Description of the Drawings

[0052] Figure 1 It is a flowchart of the data prediction method for the thermal power plant equipment of the present invention;

[0053] Figure 2 It is a schematic structural diagram of the data prediction system for the thermal power plant equipment of the present invention. Detailed Embodiments

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1: Figure 1 A data prediction method for the equipment of a thermal power plant according to the present invention is provided, including the following steps:

[0056] S1. Collect the operation status data of the thermal power plant equipment and preprocess it;

[0057] S2. Classify the operation status data according to the preset load condition intervals, extract the time-frequency domain features within each load condition interval, and generate a set of condition features;

[0058] S3. Detect the switching frequency of the load condition intervals. If the switching frequency exceeds the switching frequency threshold, start the dynamic analysis strategy switch to mark the abnormal condition intervals;

[0059] S4. Decompose the set of condition features of the non-abnormal condition intervals according to the similarity matching result between the historical degradation stage of the equipment and the current time-frequency domain features, and generate a subset of equipment degradation features;

[0060] S5. Based on the coupling relationship between the mechanical dynamics parameters and the thermodynamic parameters of the equipment, perform parameter coupling correction on the subset of equipment degradation features;

[0061] S6. Based on the corrected subset of equipment degradation features, trace back the potential fault source equipment along the equipment fault propagation map in reverse, determine the maintenance priority, and generate a maintenance instruction.

[0062] The specific implementation of S1. Collecting the operation status data of the thermal power plant equipment and preprocessing it is as follows:

[0063] Vibration data, temperature data, and pressure data of thermal power plant equipment are synchronously collected through vibration sensors, temperature sensors, and pressure sensors, and the vibration data, temperature data, and pressure data are aligned in terms of time stamps. The vibration sensor uses a piezoelectric acceleration sensor, which is installed in the vertical direction of the steam turbine bearing pedestal to collect mechanical vibration signals of the equipment; the temperature sensor uses a platinum resistance temperature sensor, which is installed in the outlet pipeline of the boiler feed pump to monitor the temperature change of the medium; the pressure sensor uses a capacitive pressure transmitter, which is installed in the steam inlet pipeline of the steam turbine to measure the steam pressure. The specific method for time stamp alignment is to take the acquisition time point of the vibration data as the reference, and align the acquisition time points of the temperature data and pressure data to the sampling moment of the vibration data through an interpolation algorithm. For example, if the vibration data is continuously collected at time points t1, t2, t3, and the temperature data is collected at adjacent time points t0, t2, t4, then linear interpolation is used to calculate the interpolation of the temperature data at time points t1 and t3 to ensure that all sensor data is aligned at the same time point.

[0064] Data cleaning is performed on the vibration data, temperature data, and pressure data after time stamp alignment. The data cleaning includes the following processes: The sliding window filtering method is used to remove the high-frequency noise components in the vibration data. The window length of the sliding window is determined according to the main frequency period of the equipment vibration signal. For example, when the main frequency is 1 kHz, the window length is set to 10 sampling points, the sliding step is set to 1 sampling point, and the vibration data within the window is processed by mean filtering, calculating the average value of the data within the window to filter out high-frequency noise components with a frequency higher than 5 kHz. The specific method for removing abnormal outliers based on the correlation between temperature data and pressure data is to calculate the Pearson correlation coefficient between the temperature data and pressure data within the same time window. If the absolute value of the correlation coefficient is lower than the preset threshold, it is determined that there are abnormal data points within this time window and they are removed. The preset threshold is determined based on historical data statistics. For example, by analyzing the correlation coefficient distribution of temperature and pressure under normal operating conditions, the lower limit value covering 95% of the normal data is selected as the threshold. For example, the threshold is 0.6. If the absolute value of the correlation coefficient of a certain time window is lower than 0.6, all the data within this time window is removed.

[0065] The cleaned vibration data, temperature data, and pressure data are resampled at a preset sampling frequency to generate a standardized operating state dataset. The preset sampling frequency is set according to twice the highest vibration frequency of the equipment. For example, if the highest vibration frequency of the equipment is 5 kHz, the sampling frequency is set to 10 kHz.

[0066] The resampling method performs interpolation on the cleaned data, uniformly samples the vibration data, temperature data, and pressure data to a preset frequency, and normalizes the data to a numerical range of 0-1 through linear scaling. For example, the original range of the vibration amplitude is -5V to +5V, and after normalization, it is (original value + 5) / 10; the original range of the temperature data is 0-200°C, and after normalization, it is original value / 200; the original range of the pressure data is 0-10MPa, and after normalization, it is original value / 10.

[0067] S2. Classify the operation status data according to the preset load condition intervals, extract the time-frequency domain features within each load condition interval, and generate a set of condition features. The specific implementation is as follows:

[0068] Divide the standardized operation status data set into multiple load condition intervals according to the load value. The load range of each load condition interval is determined according to the load distribution statistics of the equipment's historical operation data. For example, by analyzing the load value distribution histogram of the thermal power plant equipment in the past year, determine that the low load interval corresponds to the lowest load range with a higher distribution frequency of the load value in the historical operation of the equipment, the medium load interval corresponds to the intermediate load range with a higher distribution frequency of the load value, and the high load interval corresponds to the highest load range with a higher distribution frequency of the load value. The number of load condition intervals is set according to the equipment's peak shaving requirements. For example, it is divided into three intervals, namely the low load interval, the medium load interval, and the high load interval.

[0069] Extract the time-frequency domain features of the vibration data, temperature data, and pressure data within each load condition interval respectively. The time-frequency domain features of the vibration data include the vibration spectrum energy distribution, the root mean square value in the time domain, and the centroid frequency in the frequency domain. Among them, the vibration spectrum energy distribution calculates the energy proportion of each frequency band through the fast Fourier transform. For example, divide the vibration data into multiple frequency bands according to the preset frequency resolution, and calculate the proportion of the energy of each frequency band in the total energy; the root mean square value in the time domain is the square root of the mean square value of the vibration amplitude within the time window, reflecting the overall energy level of the vibration signal; the centroid frequency in the frequency domain is the weighted average frequency of the vibration spectrum energy, reflecting the concentrated frequency band of the vibration energy.

[0070] The time-frequency domain features of the temperature data include the temperature change rate and the temperature fluctuation amplitude. The temperature change rate is the slope of the temperature change per unit time. For example, it is obtained by calculating the difference between the temperature values at adjacent time points divided by the time interval; the temperature fluctuation amplitude is the difference between the maximum and minimum temperature values within the time window, reflecting the temperature fluctuation range.

[0071] The time-frequency domain features of the pressure data include the pressure fluctuation frequency and the pressure gradient. The pressure fluctuation frequency is the number of times the pressure data exceeds the preset threshold per unit time. For example, the number of times the pressure value exceeds the average pressure value; the pressure gradient is the difference between the pressure values at adjacent time points, reflecting the severity of the pressure change.

[0072] Merge the vibration spectrum energy distribution, time-domain root mean square value, frequency-domain centroid frequency, temperature change rate, temperature fluctuation amplitude, pressure fluctuation frequency, and pressure gradient within the same load condition interval according to a time window to generate a multi-dimensional condition feature set. The length of the time window is set according to the equipment operation cycle or maintenance requirements. For example, it is set to 10 minutes, indicating that a set of multi-dimensional condition features is generated every 10 minutes. The merging method is to arrange all the time-frequency domain features within the same time window in a preset order as a feature vector. For example, the vibration spectrum energy distribution, time-domain root mean square value, frequency-domain centroid frequency, temperature change rate, temperature fluctuation amplitude, pressure fluctuation frequency, and pressure gradient are arranged in sequence to form a feature vector containing multiple dimensions, and each feature vector corresponds to the condition state within a time window.

[0073] S3. Detect the switching frequency of the load condition interval. If the switching frequency exceeds the switching frequency threshold, start the dynamic analysis strategy switching to mark the abnormal condition interval. The specific implementation is as follows:

[0074] Based on a preset time window, count the number of switches of the load condition interval, calculate the switching frequency per unit time, and determine whether the switching frequency exceeds the switching frequency threshold. The length of the preset time window is set according to the equipment peak shaving requirements. For example, the time window is set to 1 hour, and the number of switches of the load condition interval within this window is counted. The switching frequency is the number of switches divided by the length of the time window.

[0075] The switching frequency threshold is determined according to the statistical distribution of the load switching frequency in the equipment historical operation data. The specific method is as follows: Analyze the load switching records of the equipment in different peak shaving scenarios in the past year, count the distribution of the number of switches per hour, and select the upper limit of the number of switches covering 95% of the normal conditions as the switching frequency threshold. For example, if the statistical result shows that the number of switches per hour under normal conditions does not exceed 5 times, the switching frequency threshold is set to 5 times / hour. If the switching frequency of the current time window exceeds this threshold, it is determined as a high-frequency switching condition and subsequent analysis is triggered.

[0076] If the switching frequency exceeds the switching frequency threshold, align the time-frequency domain features of adjacent load condition intervals through the dynamic time warping method, and extract the cross-condition consistency index. The specific implementation of the dynamic time warping method is as follows: Take the vibration spectrum energy distribution, temperature change rate, and pressure gradient sequence corresponding to adjacent load condition intervals as inputs, and perform non-linear alignment of the time axis through the dynamic time warping algorithm to eliminate the time series offset caused by the difference in load switching speed. For example, if the duration of the previous load condition interval is 10 minutes and the corresponding vibration spectrum energy distribution sequence contains 10 data points, and the duration of the subsequent load condition interval is 5 minutes and the corresponding vibration spectrum energy distribution sequence contains 5 data points, then align the two sequences to the same time length through the dynamic time warping algorithm, for example, align to 10 minutes, and adjust the scaling ratio of the time axis during the alignment process so that key feature points (such as vibration energy peaks, temperature change inflection points) match in time. The temperature change rate and pressure gradient sequences after alignment adopt the same time axis adjustment rule to ensure the time synchronization of multi-source signals.

[0077] The calculation method of the cross-condition consistency index is as follows: Calculate the Pearson correlation coefficients for the aligned vibration spectrum energy distribution, temperature change rate, and pressure gradient sequences respectively, and take the average of the three Pearson correlation coefficients as the cross-condition consistency index. For example, if the correlation coefficient of the aligned vibration spectrum energy distribution sequence is 0.8, the correlation coefficient of the temperature change rate sequence is 0.7, and the correlation coefficient of the pressure gradient sequence is 0.6, then the cross-condition consistency index is (0.8 + 0.7 + 0.6) / 3 = 0.7.

[0078] If the cross-condition consistency index is lower than the consistency threshold, mark the load condition interval corresponding to the current time window as an abnormal condition interval, and trigger the re-extraction of the time-frequency domain features within the abnormal condition interval.

[0079] The consistency threshold is determined according to the statistical distribution of the cross-condition consistency index in the historical normal condition data. The specific method is as follows: Collect the cross-condition consistency index data of the equipment under normal conditions, calculate its mean and standard deviation, and select the mean minus twice the standard deviation as the threshold lower limit. For example, if the mean of the cross-condition consistency index under normal conditions is 0.8 and the standard deviation is 0.1, then the consistency threshold is set to 0.6. If the current index is lower than 0.6, it is determined as an abnormal condition interval.

[0080] The specific operation to trigger the re-extraction is as follows: Clear the time-frequency domain feature data within the abnormal condition interval, and re-extract the vibration spectrum energy distribution, temperature change rate, and pressure gradient features within this interval based on the method in step S2. For example, if a certain time window is marked as abnormal, extract the original data of this time window from the standardized operation state dataset, and re-execute the load condition interval classification and time-frequency domain feature extraction to generate an updated set of condition features.

[0081] If the cross-condition consistency index is not lower than the consistency threshold, mark the load condition interval corresponding to the current time window as a non-abnormal condition interval, retain the original time-frequency domain feature data, and enter the subsequent processing flow.

[0082] S4. Decompose the condition feature set of the non-abnormal condition interval according to the similarity matching result between the historical degradation stage of the device and the current time-frequency domain features to generate a device degradation feature subset. The specific implementation is as follows:

[0083] For the current time-frequency domain features in the condition feature set of the non-abnormal condition interval, judge the matching degree between the current time-frequency domain features and the historical degradation stage of the device according to the similarity calculation result between the historical degradation stage label of the device and the current time-frequency domain features. The non-abnormal condition interval is the load condition interval marked as non-abnormal in step S3, and the condition feature set is the multi-dimensional feature set including vibration spectrum energy distribution, time-domain root mean square value, frequency-domain centroid frequency, temperature change rate, temperature fluctuation amplitude, pressure fluctuation frequency, and pressure gradient generated in step S2.

[0084] The historical degradation stage label of the device is divided by the vibration amplitude trend and temperature gradient trend in the device historical operation data. The specific method is as follows: Collect the vibration amplitude and temperature gradient data in the device historical operation data, calculate the average growth rate of the vibration amplitude and the fluctuation amplitude of the temperature gradient according to the time window, and divide the device degradation process into the initial, middle, and final stages. For example, when the vibration amplitude growth rate is lower than 5% in three consecutive months, it is marked as the initial degradation stage; when the growth rate is between 5% and 10%, it is marked as the middle degradation stage; when the growth rate exceeds 10%, it is marked as the final degradation stage; when the temperature gradient fluctuation amplitude is lower than 2 °C / h in three consecutive months, it is marked as the initial degradation stage; when the fluctuation amplitude is between 2 °C / h and 5 °C / h, it is marked as the middle degradation stage; when the fluctuation amplitude exceeds 5 °C / h, it is marked as the final degradation stage.

[0085] If the matching degree is greater than or equal to the preset matching threshold, extract the equipment degradation feature subset from the set of working condition features in the non-abnormal working condition interval using the low-frequency narrowband decomposition mode. The specific implementation method of the low-frequency narrowband decomposition mode is: intercept the energy of the frequency band with the vibration spectrum energy distribution lower than the preset cut-off frequency in the set of working condition features in the non-abnormal working condition interval as the equipment degradation feature subset. The preset cut-off frequency is set according to twice the fundamental frequency of the equipment. For example, when the fundamental frequency of the equipment is 50 Hz, the cut-off frequency is set to 100 Hz, and the vibration spectrum energy distribution in the frequency band from 0 Hz to 100 Hz is intercepted. The preset matching threshold is statistically determined based on the calculation results of the similarity between the historical degradation stage labels of the equipment and the corresponding time-frequency domain features. For example, by analyzing the similarity data distribution in the historical normal degradation stage, select the lower limit value covering 95% of the normal data as the threshold. The specific operation is: collect 100 groups of similarity data of the equipment in the initial and middle degradation stages, calculate the mean value of 0.8 and the standard deviation of 0.1, then the preset matching threshold is set to 0.7 (mean - 1 times the standard deviation). If the current matching degree ≥ 0.7, it is determined as the stable degradation stage and the low-frequency narrowband decomposition is triggered.

[0086] If the matching degree is less than the preset matching threshold, based on the correlation between the harmonic component distribution of the vibration spectrum in the set of working condition features in the non-abnormal working condition interval and the critical speed of the rotor dynamic parameters, select the frequency band with concentrated harmonic energy for enhanced decomposition to generate the equipment degradation feature subset. The method for selecting the frequency band with concentrated harmonic energy is: extract the amplitude of the harmonic components of the vibration spectrum in the set of working condition features in the non-abnormal working condition interval. If the amplitude of a certain harmonic component exceeds the preset ratio of the fundamental frequency amplitude, it is determined as the frequency band with concentrated energy. The preset ratio is statistically determined based on the historical harmonic distribution of the equipment. For example, statistically analyze the ratio of the amplitude of the second harmonic to the amplitude of the fundamental frequency in the historical data of the equipment in the final degradation stage, select the lower limit value covering 90% of the normal data as the ratio threshold. If the threshold is 20%, when the amplitude of the second harmonic exceeds 20% of the amplitude of the fundamental frequency, select this frequency band for enhanced decomposition.

[0087] The correlation between the critical speed of the rotor dynamic parameters is specifically reflected as: determine the speed corresponding to the fundamental frequency according to the rotor critical speed calculation formula. For example, when the critical speed is 3000 rpm, the corresponding fundamental frequency is 50 Hz, the second harmonic corresponds to 100 Hz, and the third harmonic corresponds to 150 Hz. Based on this correlation, locate the frequency band with concentrated harmonic energy. The specific operation of the enhanced decomposition is: perform band-pass filtering on the selected frequency band with concentrated harmonic energy, extract the vibration energy distribution in this frequency band, and merge it with the temperature change rate and pressure gradient characteristics to generate the equipment degradation feature subset.

[0088] S5. Based on the coupling relationship between the mechanical dynamic parameters and thermodynamic parameters of the equipment, the parameter coupling correction of the equipment degradation feature subset is specifically implemented as:

[0089] According to the correlation between the rotor speed and the vibration amplitude in the mechanical dynamics parameters of the equipment, the amplitude-frequency characteristics of the vibration amplitude in the equipment degradation feature subset are corrected. The mechanical dynamics parameters of the equipment include the designed rotor speed and the critical speed. The critical speed is the critical rotor resonance speed disclosed in the equipment design parameters. For example, the critical speed of a certain type of steam turbine rotor is 3000 rpm.

[0090] The correlation between the rotor speed and the vibration amplitude is determined by the resonance response law in the rotor dynamics equation. Specifically: when the actual operating speed approaches the critical speed, the vibration amplitude increases non-linearly with the square of the ratio of the speed to the critical speed. The correction method is: calculate the vibration amplitude correction factor according to the ratio of the actual operating speed to the critical speed. For example, when the actual speed is 80% of the critical speed, the correction factor is 1 / (0.8 2 ) = 1.5625. Multiply the original vibration amplitude in the equipment degradation feature subset by the correction factor to obtain the corrected vibration amplitude. For example, if the original vibration amplitude is 0.4 mm / s², the corrected value is 0.4×1.5625 = 0.625 mm / s², to reflect the true vibration level when the rotor approaches the critical speed.

[0091] Based on the matching relationship between the thermal expansion coefficient and the temperature gradient in the thermodynamic parameters, the temperature gradient in the equipment degradation feature subset is corrected for thermal stress compensation. The thermodynamic parameters include the thermal expansion coefficient of the equipment material and the equipment structure size. The thermal expansion coefficient is the publicly available physical parameter of the equipment material (such as alloy steel). For example, the thermal expansion coefficient of alloy steel is 12×10 -6 / °C. The matching relationship between the temperature gradient and the thermal stress is determined by the thermal stress equation. Specifically: the thermal stress caused by the temperature gradient is proportional to the product of the thermal expansion coefficient, the temperature change rate, and the equipment structure size. The correction method is: calculate the theoretical thermal strain according to the thermal expansion coefficient and the equipment structure size. For example, if the length of the equipment shaft is 5 meters and the temperature gradient is 10 °C / h, the theoretical thermal strain is 12×10 -6 / °C × 5 × 10 3 mm × 10 °C = 0.6 mm. If the actual measured thermal strain is 0.5 mm, the compensation coefficient is 0.5 / 0.6 ≈ 0.833. Multiply the original temperature gradient in the equipment degradation feature subset by the compensation coefficient to obtain the corrected temperature gradient. For example, if the original temperature gradient is 12 °C / h, the corrected value is 12×0.833 ≈ 10 °C / h, to eliminate the measurement deviation caused by material thermal expansion.

[0092] Merge the corrected vibration amplitude, temperature gradient, and pressure gradient features to generate a subset of device degradation features after parameter coupling correction. The merging method is as follows: Align the corrected vibration amplitude, temperature gradient, and pressure gradient data according to the time window to form a multi-dimensional feature vector containing the corrected vibration amplitude, corrected temperature gradient, and original pressure gradient. The length of the time window is the same as the window length used when generating the operating condition feature set in step S2, for example, a 10-minute window. The alignment method is as follows: Based on the sampling time points of the corrected vibration amplitude and temperature gradient, align the pressure gradient data to the same time points through linear interpolation to ensure that all feature data are merged at the same time stamp. For example, within a certain time window, the corrected vibration amplitude is 0.625 mm / s², the corrected temperature gradient is 10 °C / h, and the pressure gradient is 0.2 MPa / s. Then the merged feature vector is [0.625, 10, 0.2], which is output as a subset of device degradation features after parameter coupling correction to the subsequent steps.

[0093] S6. Based on the subset of device degradation features after correction, trace back the potential fault source devices along the device fault propagation map in reverse, determine the maintenance priority, and generate maintenance instructions. The specific implementation is as follows:

[0094] Obtain the device fault propagation map, which is constructed based on the mechanical connection relationship and energy transfer path of the device, and includes the fault propagation direction and weight between device nodes. The mechanical connection relationship of the device is determined according to the physical connection structure in the process piping and instrumentation diagram of the device. For example, the steam turbine and the generator are connected by a coupling, the boiler feed pump and the pipeline are connected by a flange, and the pipeline and the valve are connected by a thread.

[0095] The energy transfer path is determined according to the medium flow direction. For example, steam flows from the boiler to the steam turbine, cooling water flows from the condenser to the cooling tower, and electric current flows from the generator to the transformer. The fault propagation weight is determined by statistically analyzing the fault chain reaction frequency between device nodes in the historical fault records. For example, if the probability of a pump fault causing a valve fault in the historical data is 70%, the propagation weight from the pump to the valve is set to 0.7; if the probability of a pipeline leak causing a pump fault is 50%, the propagation weight from the pipeline to the pump is set to 0.5.

[0096] Extract the abnormal parameters from the subset of device degradation features after parameter coupling correction, and match the corresponding nodes in the device fault propagation map according to the type of abnormal parameters.

[0097] The method for extracting abnormal parameters is as follows: Compare the vibration amplitude, temperature gradient, and pressure gradient in the device degradation feature subset after parameter coupling correction with the preset normal range respectively. If a certain parameter exceeds the normal range, it is determined as an abnormal parameter. The preset normal range is determined according to the statistical distribution of the device historical operation data. For example, the normal range of vibration amplitude is the mean ± 3 times the standard deviation, the normal range of temperature gradient is 5°C / h to 15°C / h, and the normal range of pressure gradient is 0 MPa / s to 0.3 MPa / s.

[0098] The matching rule for matching the corresponding nodes in the device fault propagation map according to the abnormal parameter type is as follows: Abnormal vibration amplitude is associated with the nodes of rotating machinery (such as steam turbine bearings, generator rotors), abnormal temperature gradient is associated with the nodes of heat exchange equipment (such as boiler pipes, condensers), and abnormal pressure gradient is associated with the nodes of fluid control (such as pumps, valves). For example, if the vibration amplitude in a certain time window in the device degradation feature subset after parameter coupling correction is 1.0 mm / s² (exceeding the normal range of 0.8 mm / s²), it is matched to the steam turbine bearing node; if the temperature gradient is 18°C / h (exceeding the normal range of 15°C / h), it is matched to the boiler pipe node.

[0099] Traverse reversely along the fault propagation direction of the device fault propagation map to locate the potential fault source device node of the abnormal parameter. The specific method of reverse traversal is: Starting from the device node where the abnormal parameter is detected, trace back layer by layer along the reverse path of the fault propagation direction until the root device without an upstream node is found. For example, if a pressure gradient abnormality is detected at the valve node, the reverse traversal path is valve → pump → pipe → boiler. If the historical fault weight of the pump node is 0.7 and there is an unrepaired record, it is determined that the pump node is the potential fault source; if the pipe node is normal and the boiler node is normal, the pump node is the final fault source. During the traversal process, preferentially select the path with a high propagation weight for backtracking. For example, the weight of the pump → valve path is 0.7, which is higher than the weight of the pipe → pump path of 0.5, so the pump → valve path is preferentially checked.

[0100] Determine the maintenance priority according to the number and position depth of the abnormal parameters of the potential fault source device node. The number of abnormal parameters is the total number of abnormal parameters of vibration amplitude, temperature gradient, and pressure gradient that appear simultaneously on the same device node. For example, if there are abnormal vibration amplitude and temperature gradient at a certain pump node, the number of abnormal parameters is 2. The position depth is the number of levels of the fault source node from the system entrance. For example, if the system entrance is the boiler, in the path of boiler → pipe → pump → valve, the depth of the boiler is 1, the depth of the pipe is 2, the depth of the pump is 3, and the depth of the valve is 4.

[0101] The maintenance priority calculation formula is: Priority score = Number of abnormal parameters × Location depth coefficient. The location depth coefficient is set according to the number of levels. For example, the coefficient for depth 1 is 1.0, for depth 2 is 1.2, for depth 3 is 1.5, and for depth 4 is 2.0. If the number of abnormal parameters of a pump node is 2 and the depth is 3, then the priority score is 2 × 1.5 = 3.0.

[0102] Based on the maintenance priority and the remaining service life of the equipment, a maintenance instruction including the identification of the faulty equipment and the maintenance time is generated. The remaining service life of the equipment is estimated according to the ratio of the cumulative operating time of the equipment to the designed life. For example, the designed life of a certain pump is 10 years, it has been operating for 8 years, and the remaining life is 2 years.

[0103] The rule for determining the maintenance time is: When the priority score is higher than the maintenance priority threshold and the remaining life is lower than the life threshold, emergency maintenance is triggered; when the priority score is higher than the threshold but the remaining life is higher than the threshold, scheduled maintenance is triggered. The maintenance priority threshold and the life threshold are set according to the equipment operation and maintenance strategy. For example, the priority threshold is set to 2.0 and the life threshold is set to 1 year. If the priority score of a pump node is 3.0 and the remaining life is 0.8 years, an emergency maintenance instruction is generated and the maintenance time is set within 24 hours; if the remaining life is 1.5 years, a scheduled maintenance instruction is generated and the maintenance time is set within 7 days. The content of the maintenance instruction includes the identification of the faulty equipment (such as equipment number, installation location), details of abnormal parameters (such as vibration amplitude 1.0 mm / s²), and recommended maintenance measures (such as bearing replacement, seal inspection).

[0104] Embodiment 2: Figure 2 The structural schematic diagram of the data prediction system for thermal power plant equipment of the present invention is given. The data prediction system for thermal power plant equipment includes:

[0105] Data acquisition and processing module: Collect the operation status data of thermal power plant equipment and preprocess it;

[0106] Operating condition classification and extraction module: Classify the operation status data according to the preset load operating condition intervals, extract the time-frequency domain features within each load operating condition interval, and generate an operating condition feature set;

[0107] Dynamic strategy anomaly marking module: Detect the switching frequency of the load operating condition intervals. If the switching frequency exceeds the switching frequency threshold, start the dynamic analysis strategy switching to mark the abnormal operating condition intervals;

[0108] Degradation feature decomposition module: Decompose the operating condition feature set of the non-abnormal operating condition intervals according to the similarity matching result between the historical degradation stage of the equipment and the current time-frequency domain features, and generate a subset of equipment degradation features;

[0109] Parameter coupling correction module: Based on the coupling relationship between the mechanical dynamics parameters and thermodynamic parameters of the device, perform parameter coupling correction on the subset of device degradation characteristics;

[0110] Fault backtracking decision module: Based on the corrected subset of device degradation characteristics, trace back the potential fault source devices along the device fault propagation map in reverse, determine the maintenance priority and generate maintenance instructions.

[0111] It should be noted that the present invention can be deployed on the device itself to achieve embedded applications, or can also run on a PC or other terminals with a user interface, so as to meet various hardware environments and usage requirements.

[0112] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more sets of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0113] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, and modules can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0114] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or modules can be in electrical, mechanical, or other forms.

[0115] The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0116] In addition, in each embodiment of the present application, the functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0117] If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0118] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0119] Finally, the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data prediction method for thermal power plant equipment, characterized in that: The steps include: S1. Collect and pre-process the operating status data of thermal power plant equipment; S2. Classify the operating status data according to the preset load condition interval, extract the time-frequency domain features within each load condition interval, and generate a condition feature set; S3, detecting the switching frequency of the load condition interval, and if the switching frequency exceeds the switching frequency threshold, starting the dynamic analysis strategy switching to mark the abnormal condition interval; S4. Decomposing the operating condition feature set of the non-abnormal operating condition interval according to the similarity matching results between the historical degradation stage of the equipment and the current time-frequency domain features to generate an equipment degradation feature subset; S5. Based on the coupling relationship between the mechanical dynamic parameters and thermodynamic parameters of the equipment, parameter coupling correction is performed on the equipment degradation feature subset; S6. Based on the corrected equipment degradation feature subset, trace back the potential fault source equipment along the equipment fault propagation map, determine the maintenance priority and generate maintenance instructions.

2. The data prediction method for thermal power plant equipment according to claim 1, characterized in that: Collect and pre-process the operating status data of thermal power plant equipment, including: Synchronously collect vibration data, temperature data and pressure data of thermal power plant equipment, and align the timestamps of vibration data, temperature data and pressure data; Data cleaning is performed on the vibration data, temperature data and pressure data after time stamp alignment. Data cleaning includes using a sliding window filtering method to remove high-frequency noise components in the vibration data and removing abnormal outliers based on the correlation between temperature data and pressure data. The vibration data, temperature data and pressure data after cleaning are resampled according to the preset sampling frequency to generate a standardized operating status data set.

3. The data prediction method for thermal power plant equipment according to claim 1, characterized in that: The operating status data is classified according to the preset load condition interval, and the time-frequency domain features in each load condition interval are extracted to generate a condition feature set, including: The standardized operating status data set is divided into multiple load condition intervals according to the load value, and the load condition intervals include a low load interval, a medium load interval and a high load interval; Extract time-frequency domain features from vibration data, temperature data and pressure data in each load condition interval; The time-frequency domain features within the same load condition interval are merged according to the time window to generate a multi-dimensional condition feature set.

4. The data prediction method for thermal power plant equipment according to claim 3, characterized in that: The time-frequency domain characteristics of vibration data include vibration spectrum energy distribution, time-domain root mean square value and frequency-domain centroid frequency; the time-frequency domain characteristics of temperature data include temperature change rate and temperature fluctuation amplitude; the time-frequency domain characteristics of pressure data include pressure fluctuation frequency and pressure gradient.

5. The data prediction method for thermal power plant equipment according to claim 1, characterized in that: Detect the switching frequency of the load condition interval. If the switching frequency exceeds the switching frequency threshold, start the dynamic analysis strategy switching to mark the abnormal condition interval, including: The number of switching times in the load condition interval is counted based on the preset time window, the switching frequency per unit time is calculated, and it is determined whether the switching frequency exceeds the switching frequency threshold; If the switching frequency exceeds the switching frequency threshold, the time-frequency domain features of adjacent load condition intervals are aligned through the dynamic time warping method, and the cross-condition consistency index is extracted; If the cross-operating condition consistency index is lower than the consistency threshold, the load operating condition interval corresponding to the current time window is marked as an abnormal operating condition interval, and the re-extraction of the time-frequency domain features in the abnormal operating condition interval is triggered; otherwise, it is marked as a non-abnormal operating condition interval.

6. The data prediction method for thermal power plant equipment according to claim 5, characterized in that: Extracting the cross-operating condition consistency index includes: aligning the time-frequency domain features of adjacent load condition intervals, calculating the Pearson correlation coefficient for the aligned vibration spectrum energy distribution, temperature change rate and pressure gradient, and taking the average value of the Pearson correlation coefficient as the cross-operating condition consistency index.

7. The data prediction method for thermal power plant equipment according to claim 1, characterized in that: According to the similarity matching results between the historical degradation stage of the equipment and the current time-frequency domain features, the operating condition feature set of the non-abnormal operating condition interval is decomposed to generate a subset of equipment degradation features, including: For the current time-frequency domain features in the working condition feature set of the non-abnormal working condition interval, the matching degree between the current time-frequency domain features and the historical degradation stage of the equipment is determined according to the similarity calculation result between the historical degradation stage label of the equipment and the current time-frequency domain features; Among them, the equipment historical degradation stage label is divided by the vibration amplitude trend and temperature gradient trend in the equipment historical operation data; If the matching degree is greater than or equal to a preset matching threshold, a subset of equipment degradation features is extracted from the operating condition feature set of the non-abnormal operating condition interval using a low-frequency narrowband decomposition mode; If the matching degree is less than the preset matching threshold, based on the correlation between the harmonic component distribution of the vibration spectrum in the operating condition feature set of the non-abnormal operating condition interval and the critical speed of the rotor dynamic parameters, the frequency band where the harmonic energy is concentrated is selected for enhanced decomposition to generate a subset of equipment degradation features.

8. The data prediction method for thermal power plant equipment according to claim 1, characterized in that: Based on the coupling relationship between the mechanical dynamic parameters and thermodynamic parameters of the equipment, parameter coupling correction is performed on the equipment degradation feature subset, including: According to the correlation between the rotor speed and the vibration amplitude in the mechanical dynamics parameters of the equipment, the amplitude-frequency characteristics of the vibration amplitude in the equipment degradation feature subset are corrected; Based on the matching relationship between the thermal expansion coefficient and the temperature gradient in the thermodynamic parameters, thermal stress compensation correction is performed on the temperature gradient in the equipment degradation feature subset; The corrected vibration amplitude and temperature gradient are combined with the pressure gradient features to generate a parameter-coupled corrected device degradation feature subset.

9. The data prediction method for thermal power plant equipment according to claim 1, characterized in that: Based on the corrected equipment degradation feature subset, the potential fault source equipment is traced back along the equipment fault propagation map to determine the maintenance priority and generate maintenance instructions, including: Obtain equipment fault propagation maps, which are constructed based on the mechanical connection relationship and energy transfer path of the equipment, and include the fault propagation direction and weight between equipment nodes; Extract abnormal parameters from the equipment degradation feature subset after parameter coupling correction, and match the corresponding nodes in the equipment fault propagation graph according to the abnormal parameter type; Traverse in reverse along the fault propagation direction of the equipment fault propagation map to locate the potential fault source equipment nodes with abnormal parameters; Determine maintenance priority based on the number and location depth of abnormal parameters of potential fault source device nodes; Based on the maintenance priority and the remaining service life of the equipment, a maintenance instruction is generated, including the equipment identification of the fault source and the maintenance time.

10. A data prediction system for thermal power plant equipment, used to implement the data prediction method for thermal power plant equipment according to any one of claims 1 to 9, characterized in that: include: Data acquisition and processing module: collects and pre-processes the operating status data of thermal power plant equipment; Working condition classification and extraction module: classifies the operating status data according to the preset load condition interval, extracts the time-frequency domain features within each load condition interval, and generates a working condition feature set; Dynamic strategy abnormality marking module: detects the switching frequency of the load condition interval. If the switching frequency exceeds the switching frequency threshold, the dynamic analysis strategy switching is initiated to mark the abnormal condition interval; Degradation feature decomposition module: Decomposes the operating condition feature set of the non-abnormal operating condition interval according to the similarity matching results between the historical degradation stage of the equipment and the current time-frequency domain features, and generates a subset of equipment degradation features; Parameter coupling correction module: Based on the coupling relationship between the mechanical dynamic parameters and thermodynamic parameters of the equipment, parameter coupling correction is performed on the equipment degradation feature subset; Fault backtracking decision module: Based on the corrected equipment degradation feature subset, it reversely traces back to the potential fault source equipment along the equipment fault propagation map, determines the maintenance priority and generates maintenance instructions.

Images