A power generation equipment maintenance strategy dynamic generation method based on a vectorized knowledge graph

Abstract

The present application relates to the technical field of equipment maintenance, and more particularly to a power generation equipment maintenance strategy dynamic generation method based on a vectorized knowledge graph, comprising: obtaining a plurality of maintenance fault data; determining the length of the monitoring period of a target power generation equipment according to the historical fault frequency and load anomaly degree of the target power generation equipment, and reducing the length of the monitoring period based on the fault influence degree under a preset condition; determining whether to perform secondary determination according to the maximum abnormal signal characteristic value or to generate a maintenance strategy according to the vibration anomaly degree and temperature anomaly degree corresponding to the current period monitoring data. The present application can improve the maintenance efficiency and quality of power generation equipment.

Description

Technical Field

[0001] This invention relates to the field of equipment maintenance technology, and in particular to a method for dynamically generating maintenance strategies for power generation equipment based on vectorized knowledge graphs. Background Technology

[0002] With the continuous development of the power industry and the expanding scale of power generation equipment, regular maintenance of power equipment is necessary to ensure the safe and stable operation of the power system. However, the methods for formulating maintenance plans are relatively traditional and simplistic. In most cases, maintenance is carried out at fixed intervals without considering the actual operating conditions of the equipment, which can easily lead to over-maintenance. This not only results in a huge waste of resources, but frequent maintenance may also cause additional damage to the equipment due to improper operation, affecting its normal operation and service life. Therefore, how to improve the efficiency and quality of power generation equipment maintenance is an urgent problem to be solved by those skilled in the art.

[0003] Chinese Patent Publication No. CN116881658A discloses an intelligent condition assessment method and system for hydro-generator units. The method includes: collecting measurement point distribution data of the target hydro-generator unit to obtain the monitoring sensor deployment results; generating a monitoring sensor array; obtaining the operating data of the target hydro-generator unit within a preset time window to generate a real-time condition dataset; obtaining a first real-time condition dataset and a second real-time condition dataset; inputting the datasets into a unit measurement point condition assessment model for analysis to obtain real-time measurement point condition assessment results; obtaining unit equipment assessment results; and combining the real-time measurement point condition assessment results with the data into a maintenance strategy decision model to obtain a maintenance strategy decision scheme. However, the above scheme has the following problems: it cannot optimize the monitoring frequency according to the actual operating conditions of the power generation equipment, and it cannot adaptively adjust the maintenance strategy generation method according to the real-time equipment conditions, resulting in poor maintenance efficiency and quality of the power generation equipment. Summary of the Invention

[0004] To address this, the present invention provides a dynamic generation method for power generation equipment maintenance strategies based on vectorized knowledge graphs, which overcomes the problems in the prior art where the monitoring frequency cannot be optimized according to the actual operating conditions of the power generation equipment, and the maintenance strategy generation method cannot be adaptively adjusted according to the real-time status of the equipment, resulting in poor maintenance efficiency and quality of the power generation equipment.

[0005] To achieve the above objectives, this invention provides a method for dynamically generating maintenance strategies for power generation equipment based on vectorized knowledge graphs, comprising:

[0006] Obtain several fault inspection data;

[0007] The duration of the monitoring cycle for the target power generation equipment is determined based on the historical failure frequency and load anomaly of the target power generation equipment, and the duration of the monitoring cycle is adjusted to be reduced based on the failure impact under preset conditions.

[0008] Based on the vibration anomaly and temperature anomaly corresponding to the current cycle monitoring data, a secondary judgment or maintenance strategy generation is performed based on the maximum anomaly signal characterization value.

[0009] In the maintenance strategy generation process, the analysis and generation are determined based on the maximum data correlation, or the matching and generation are based on the data correlation.

[0010] In the analysis and generation process, the abnormal matching degree is determined based on the number matching degree and the location deviation degree of the abnormal monitoring points. The matching maintenance fault data corresponding to the current period monitoring data is determined based on the abnormal matching degree. The maintenance strategy is generated directly or in combination based on the strategy similarity corresponding to each matching maintenance fault data, or the maintenance strategy is generated based on the maximum abnormal matching threshold or the target vectorized knowledge graph.

[0011] Furthermore, the duration of the monitoring cycle for the target power generation equipment is determined based on the historical failure frequency and load anomaly of the target power generation equipment, and the duration of the monitoring cycle is adjusted to be reduced based on the failure impact under preset conditions.

[0012] The preset condition is that the fault impact is greater than the preset fault impact. The duration of the monitoring cycle of the target power generation equipment is negatively correlated with the historical fault frequency and the load anomaly. The decrease in the duration of the monitoring cycle is positively correlated with the fault impact.

[0013] Furthermore, the method for confirming the impact of the fault includes:

[0014] If the vibration difference at each monitoring point is greater than or equal to the preset vibration difference, the fault impact is determined based on the anomaly comparison coefficient.

[0015] If the vibration difference at each monitoring point is less than the preset vibration difference, the impact of the fault is determined based on the correlation of historical faults.

[0016] Furthermore, if the vibration anomaly degree corresponding to the current period monitoring data is greater than or equal to the preset vibration anomaly degree or the temperature anomaly degree is greater than or equal to the preset temperature anomaly degree, then a maintenance strategy is generated.

[0017] Furthermore, if the vibration anomaly degree corresponding to the current period monitoring data is less than the preset vibration anomaly degree and the temperature anomaly degree is less than the preset temperature anomaly degree, then a secondary judgment is made based on the anomaly signal characterization value.

[0018] In the secondary judgment, the abnormal signal characterization value corresponding to each monitoring point is determined based on the harmonic anomaly degree, and the maximum value among the abnormal signal characterization values ​​corresponding to each monitoring point is recorded as the maximum abnormal signal characterization value. Based on the maximum abnormal signal characterization value, it is determined whether to generate a maintenance strategy.

[0019] If the maximum abnormal signal characterization value is greater than or equal to the preset maximum abnormal signal characterization value, then a maintenance strategy will be generated.

[0020] If the maximum abnormal signal representation value is less than the preset maximum abnormal signal representation value, then there is no need to generate a maintenance strategy.

[0021] Furthermore, the abnormal signal characterization value is determined based on the kurtosis characterization value, including:

[0022] If the kurtosis characterization value is greater than or equal to the preset kurtosis characterization value, the abnormal signal characterization value is determined based on the spectrum comparison deviation value.

[0023] If the kurtosis characterization value is less than the preset kurtosis characterization value, the abnormal signal characterization value is determined by comparing the harmonic law.

[0024] Furthermore, if the maximum data correlation degree is greater than or equal to the preset maximum data correlation degree, then matching and generation are performed based on the data correlation degree;

[0025] During the matching process, the maintenance strategy corresponding to the maintenance fault data with the highest correlation to the current period monitoring data is used as the generated maintenance strategy.

[0026] Furthermore, if the maximum data correlation is less than the preset maximum data correlation, then analysis and generation will proceed.

[0027] Furthermore, if the strategy similarity corresponding to each matched maintenance fault data is greater than or equal to the preset strategy similarity, then maintenance strategies are generated directly or in combination based on the correlation degree of strategy differences.

[0028] If the correlation degree of the strategy difference is less than the preset correlation degree of the strategy difference, then a maintenance strategy is generated by combination.

[0029] If the correlation degree of strategy difference is greater than or equal to the preset correlation degree of strategy difference, then the maintenance strategy is generated directly.

[0030] The matched maintenance fault data refers to maintenance fault data whose abnormal matching degree with the current periodic monitoring data is greater than the preset abnormal matching degree.

[0031] Furthermore, if the strategy similarity corresponding to each matched maintenance fault data is less than the preset strategy similarity, then the maintenance strategy is determined based on the maximum abnormal matching threshold or the vectorized knowledge graph.

[0032] If the maximum anomaly matching threshold is greater than or equal to the preset maximum anomaly matching threshold, then a maintenance strategy is generated based on the anomaly matching threshold.

[0033] If the maximum anomaly matching threshold is less than the preset maximum anomaly matching threshold, then a maintenance strategy is generated based on the target vectorized knowledge graph.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows: the technical solution of the present invention reflects the reliability level and real-time operating pressure of the target power generation equipment by using historical fault frequency and load anomaly, and dynamically adapts the monitoring cycle length accordingly, avoiding the drawbacks of "over-monitoring causing resource waste" or "insufficient monitoring causing missed hidden dangers" in the fixed cycle mode; under preset conditions, the cycle is further shortened based on the fault impact, which can not only ensure the timeliness of early warning in high-risk scenarios, but also minimize the potential losses of fault spread, thereby improving the maintenance efficiency and quality of power generation equipment.

[0035] Furthermore, this invention effectively reflects the abnormal state of the target power generation equipment through vibration anomaly and temperature anomaly. When the vibration anomaly and temperature anomaly are large, maintenance strategies are directly generated, enabling rapid fault response. Secondary judgment is performed through the maximum abnormal signal characterization value to ensure that potential faults are locked in time and targeted strategies are generated, thereby minimizing the safety risk of fault expansion.

[0036] Furthermore, this invention effectively reflects the correlation between current periodic monitoring data and maintenance fault data by using the maximum data correlation degree. Then, different maintenance strategy generation methods are adaptively selected based on the maximum data correlation degree, making the generation of maintenance strategies more in line with actual application scenarios. When the maximum data correlation degree is large, matching and generating based on the data correlation degree can quickly call historical schemes and shorten the decision cycle. When the maximum data correlation degree is small, analysis and generation can improve the accuracy of maintenance strategy generation, thereby improving the maintenance efficiency and quality of power generation equipment.

[0037] Furthermore, this invention effectively reflects the similarity of operation and maintenance strategies by matching the strategy similarity corresponding to the fault data. When the strategy similarity is high, the correlation strength of the detailed differences between strategies is effectively reflected by the strategy difference correlation. Then, maintenance strategies are adaptively generated directly or in combination based on the strategy difference correlation, improving the accuracy of maintenance strategy generation. When the strategy similarity is low, the anomaly matching threshold effectively reflects the matching depth between the abnormal features of the current monitoring data and the abnormal features in the historical fault data. Then, maintenance strategies are generated based on the anomaly matching threshold or vectorized knowledge graph, improving the system's adaptability to complex or new faults and the reliability of maintenance strategy generation. Attached Figure Description

[0038] Figure 1This is a schematic diagram of the dynamic generation method for power generation equipment maintenance strategies based on vectorized knowledge graphs according to the present invention.

[0039] Figure 2 This is a flowchart illustrating the generation of a secondary judgment or maintenance strategy based on the abnormal signal characterization value according to the vibration anomaly degree and temperature anomaly degree corresponding to the current periodic monitoring data in this invention.

[0040] Figure 3 This is a flowchart illustrating the process of generating data based on the maximum data correlation degree or matching data based on the data correlation degree, as described in this invention.

[0041] Figure 4 This is a flowchart illustrating how the present invention generates maintenance strategies directly or in combination based on the correlation degree of strategy differences. Detailed Implementation

[0042] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0043] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0044] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0045] Please see Figures 1 to 4 As shown, this invention provides a method for dynamically generating maintenance strategies for power generation equipment based on vectorized knowledge graphs, including:

[0046] Obtain several fault inspection data;

[0047] The duration of the monitoring cycle for the target power generation equipment is determined based on the historical failure frequency and load anomaly of the target power generation equipment, and the duration of the monitoring cycle is adjusted to be reduced based on the failure impact under preset conditions.

[0048] Based on the vibration anomaly and temperature anomaly corresponding to the current cycle monitoring data, a secondary judgment or maintenance strategy generation is performed based on the maximum anomaly signal characterization value.

[0049] In the maintenance strategy generation process, the analysis and generation are determined based on the maximum data correlation, or the matching and generation are based on the data correlation.

[0050] In the analysis and generation process, the abnormal matching degree is determined based on the number matching degree and the location deviation degree of the abnormal monitoring points. The matching maintenance fault data corresponding to the current period monitoring data is determined based on the abnormal matching degree. The maintenance strategy is generated directly or in combination based on the strategy similarity corresponding to each matching maintenance fault data, or the maintenance strategy is generated based on the maximum abnormal matching threshold or the target vectorized knowledge graph.

[0051] The application scenario of this invention is the generation of maintenance strategies for power generation equipment. The target power generation equipment is a generator that needs to generate maintenance strategies. This invention sets up several monitoring points, each of which corresponds to a temperature sensor and a vibration sensor. The location and number of each monitoring point are not limited, and users can set them according to their actual needs. It should be noted that the location and number of fault points corresponding to each maintenance fault data in this invention are the same as the location and number of fault points corresponding to the target power generation equipment.

[0052] The current period monitoring data consists of the monitoring data corresponding to each monitoring point of the target power generation equipment in the current monitoring period. The monitoring data corresponding to a single monitoring point is the vibration velocity and temperature monitored in real time in the current monitoring period.

[0053] Individual maintenance fault data includes historical monitoring data and maintenance strategies for each monitoring point of the generator of the same model as the target power generation equipment. The historical monitoring data for each monitoring point is the vibration velocity and temperature monitored in real time during use. The maintenance strategy is a text composed of several keywords, such as "bearing, disassembly and inspection, wear measurement, grease replacement, 3-hour downtime, 2 maintenance workers, vibration ≤4.5mm / s acceptance, temperature ≤90℃ acceptance". This is content that is easy for those skilled in the art to understand, and will not be elaborated on in detail.

[0054] This invention includes several historical records. Each historical record contains at least one record of the historical fault frequency, load anomaly degree, fault impact degree, vibration difference degree, and anomaly comparison coefficient during the generation of a power equipment maintenance strategy. Each historical record also has a corresponding pass / fail marker. The pass / fail marker indicates whether the generation process of the power equipment maintenance strategy meets the user's requirements. The pass / fail marker can be manually recorded. It is understood that the user can determine whether the generation process of the power equipment maintenance strategy meets the requirements based on self-defined indicators. Self-defined indicators can be, but are not limited to, the number of errors, which will not be elaborated here. The number of errors refers to the number of times no problems were found when the target power equipment was maintained according to the generated maintenance strategy.

[0055] Specifically, the duration of the monitoring cycle for the target power generation equipment is determined based on the historical failure frequency and load anomaly of the target power generation equipment, and the duration of the monitoring cycle is adjusted to be reduced based on the failure impact under preset conditions.

[0056] The preset condition is that the fault impact is greater than the preset fault impact. The duration of the monitoring cycle of the target power generation equipment is negatively correlated with the historical fault frequency and the load anomaly. The decrease in the duration of the monitoring cycle is positively correlated with the fault impact.

[0057] Among them, the historical failure frequency is the number of times the target power generation equipment has failed;

[0058] Load anomaly degree = duration of abnormal load state of the target power generation equipment within a preset time period / duration of the preset time period; the greater the user's demand for improving the accuracy of complex anomaly degree determination, the longer the preset time period. A method for setting the preset time period is provided, which records the time interval of 30 days to the current time as the reference time, and records the time from the reference time to the current time as the preset time period; the abnormal load range is (0, a1)∪(a2, +∞), a1 = rated active power × 80%, a2 = rated active power × 110%. In this invention, the rated active power of the target power generation equipment is 600MW, and the power of the target power generation equipment is measured by a high-precision power analyzer;

[0059] The present invention sets a continuous cyclical monitoring cycle. The duration of a single monitoring cycle is 10 days - (historical fault frequency / average historical fault frequency corresponding to each reference historical record + load anomaly / average load anomaly corresponding to each reference historical record) × 5 days; the reference historical records are historical records that can meet the user's needs.

[0060] The reduction in the monitoring cycle duration = [(fault impact - preset fault impact) / fault impact] × 5d;

[0061] The preset fault impact value can be determined by the user based on the actual application scenario. The greater the user's need to improve the accuracy of the monitoring cycle setting, the smaller the preset fault impact value should be. A method for determining the preset fault impact value is provided, which detects historical records that do not require adjustment of the monitoring cycle duration based on the fault impact value, and records the average fault impact value corresponding to the historical records that meet the user's needs as the preset fault impact value.

[0062] Specifically, the methods for confirming the impact of the fault include:

[0063] If the vibration difference at each monitoring point is greater than or equal to the preset vibration difference, the fault impact is determined based on the anomaly comparison coefficient.

[0064] If the vibration difference at each monitoring point is less than the preset vibration difference, the impact of the fault is determined based on the correlation of historical faults.

[0065] In this invention, the time point is set by the user. A method for setting the time point is provided, which takes the start time of a single monitoring cycle or the start time of a single maintenance fault data as the starting point and sets a time point every 1 minute in chronological order.

[0066] Vibration variability is the average of the coefficients of variation of the target power generation equipment at each time point in the current monitoring cycle; the coefficient of variation at a single time point = the standard deviation of the vibration velocity at each monitoring point at that time point / the average of the vibration velocity at each monitoring point at that time point;

[0067] The user can determine the preset vibration difference value according to the actual application scenario. The smaller the preset vibration difference value, the greater the user's need to determine the fault impact based on the anomaly comparison coefficient. A method for determining the preset vibration difference value is provided, which detects the historical records of the user's determination of the fault impact based on the anomaly comparison coefficient, and records the average vibration difference value corresponding to the historical records that meet the user's needs as the preset vibration difference value.

[0068] Anomaly comparison coefficient = |Vibration difference degree of the target power generation equipment in the current monitoring cycle - average vibration difference degree of each first reference historical record|, where the first reference historical record is a historical record that does not require maintenance strategy generation and can meet user needs.

[0069] Historical records with vibration differences less than or equal to the vibration differences of the target power generation equipment in the current monitoring cycle and that meet user requirements are recorded as second reference historical records. Historical fault correlation = number of second reference historical records that need to generate maintenance strategies / total number of second reference historical records.

[0070] When determining the fault impact based on the anomaly comparison coefficient, the fault impact = anomaly comparison coefficient / average value of the anomaly comparison coefficients corresponding to each first reference historical record;

[0071] When determining the impact of a fault based on the correlation of historical faults, the impact of the fault is calculated as: historical fault correlation / average historical fault correlation corresponding to each first reference historical record.

[0072] Understandably, when the vibration difference between monitoring points is greater than or equal to the preset vibration difference, it means that the vibration difference has reached a significant level. At this time, the vibration data itself is sufficient to clearly show the degree of abnormality. Through the abnormality comparison coefficient, the deviation between the current abnormal state and the normal state can be directly quantified, thereby accurately determining the degree of fault impact. However, when the vibration difference between monitoring points is less than the preset vibration difference, the degree of abnormality reflected by the vibration signal is weak. It is difficult to accurately judge the fault impact based solely on the current vibration data. In this case, the historical fault correlation can indirectly reflect the actual probability of fault impact in the scenario by statistically analyzing the proportion of similar scenarios that require maintenance. This effectively avoids the limitation of insufficient data information in low-difference scenarios and ultimately achieves accurate quantification of fault impact under different equipment states.

[0073] Specifically, if the vibration anomaly degree corresponding to the current cycle monitoring data is greater than or equal to the preset vibration anomaly degree or the temperature anomaly degree is greater than or equal to the preset temperature anomaly degree, then a maintenance strategy is generated.

[0074] Among them, the vibration anomaly is the maximum value of the sub-vibration anomalies corresponding to each monitoring point. The minimum value of the vibration velocity monitored by each monitoring point in the historical record that does not require maintenance strategy generation and can meet user needs is recorded as v1, and the maximum value is recorded as v2. The sub-vibration anomaly corresponding to a single monitoring point = the duration of the vibration velocity corresponding to the monitoring point in the current monitoring cycle that is not in [v1, v2] / the duration of a single monitoring cycle.

[0075] Temperature anomaly is the maximum temperature at each monitoring point at each time point during the current monitoring period.

[0076] Users can determine the preset vibration anomaly degree and preset temperature anomaly degree based on the actual application scenario. The larger the preset vibration anomaly degree and preset temperature anomaly degree, the greater the user's need for secondary judgment based on the anomaly signal characterization value. The system provides a preset vibration anomaly degree and preset temperature anomaly degree, detects the historical records of the user's secondary judgment based on the anomaly signal characterization value, and records the average vibration anomaly degree and the average temperature anomaly degree corresponding to the historical records that meet the user's needs as the preset vibration anomaly degree and preset temperature anomaly degree, respectively.

[0077] Specifically, if the vibration anomaly degree corresponding to the current period monitoring data is less than the preset vibration anomaly degree and the temperature anomaly degree is less than the preset temperature anomaly degree, then a secondary judgment is made based on the anomaly signal characterization value.

[0078] In the secondary judgment, the abnormal signal characterization value corresponding to each monitoring point is determined based on the harmonic anomaly degree, and the maximum value among the abnormal signal characterization values ​​corresponding to each monitoring point is recorded as the maximum abnormal signal characterization value. Based on the maximum abnormal signal characterization value, it is determined whether to generate a maintenance strategy.

[0079] If the maximum abnormal signal characterization value is greater than or equal to the preset maximum abnormal signal characterization value, then a maintenance strategy will be generated.

[0080] If the maximum abnormal signal representation value is less than the preset maximum abnormal signal representation value, then there is no need to generate a maintenance strategy.

[0081] The user can determine the value of the preset maximum abnormal signal representation value according to the actual application scenario. The smaller the preset maximum abnormal signal representation value, the greater the user's need for maintenance strategy generation. A preset maximum abnormal signal representation value is provided. The user's historical records of maintenance strategy generation in the secondary judgment are detected, and the average value of the maximum abnormal signal representation value corresponding to the historical records that can meet the user's needs is recorded as the preset maximum abnormal signal representation value.

[0082] Specifically, determining the characterization values ​​of abnormal signals based on kurtosis characterization values ​​includes:

[0083] If the kurtosis characterization value is greater than or equal to the preset kurtosis characterization value, the abnormal signal characterization value is determined based on the spectrum comparison deviation value.

[0084] If the kurtosis characterization value is less than the preset kurtosis characterization value, the abnormal signal characterization value is determined by comparing the harmonic law.

[0085] Each monitoring point corresponds to a waveform graph with time as the horizontal axis and vibration velocity as the vertical axis for each monitoring cycle. The waveform graph with time as the horizontal axis and vibration velocity as the vertical axis is transformed into a two-dimensional image with time as the horizontal axis and vibration acceleration as the vertical axis. The acceleration corresponding to a single time point is the slope of the waveform graph at that time point. Then, the acceleration in the time domain is transformed into the frequency spectrum through Fourier transform to obtain the "frequency-acceleration" two-dimensional image. This is a common technique used by those skilled in the art, and will not be elaborated on in detail.

[0086] Kurtosis value = standard deviation of acceleration at each frequency / average difference of acceleration at each frequency; frequency is the frequency value corresponding to the horizontal axis of each peak in the "frequency-acceleration" two-dimensional image;

[0087] The user can determine the value of the preset kurtosis characterization value according to the actual application scenario. The larger the value of the preset kurtosis characterization value, the greater the user's need to determine the abnormal signal characterization value based on the harmonic law comparison degree. The preset kurtosis characterization value is defined as the average value of the kurtosis characterization values ​​corresponding to the historical records that can determine the abnormal signal characterization value based on the harmonic law comparison degree and meet the user's needs.

[0088] The spectrum comparison deviation value is the maximum value among the sub-frequency deviation values ​​corresponding to each frequency. The sub-frequency deviation value corresponding to a single frequency = the acceleration corresponding to that frequency - the average value of the acceleration corresponding to that frequency at that monitoring point in the historical records that do not require maintenance strategy generation and can meet user needs.

[0089] The formula for calculating the harmonic law comparison degree k is:

[0090]

[0091] Where f is the average frequency corresponding to the maximum acceleration at this monitoring point in each historical record that does not require maintenance strategy generation and can meet user needs, and A 1f Let A be the acceleration corresponding to 1f, ..., A nf Let nf be the acceleration.

[0092] When determining the characterization value of an abnormal signal based on the spectrum comparison deviation value, the characterization value of an abnormal signal = spectrum comparison deviation value / average value of spectrum comparison deviation values ​​corresponding to each historical record that can meet the user's needs;

[0093] When determining the abnormal signal characterization value based on the harmonic law comparison degree, the abnormal signal characterization value = harmonic law comparison degree / average harmonic law comparison degree corresponding to each historical record that can meet the user's needs;

[0094] Understandably, kurtosis can effectively reflect the relative intensity of transient impacts. When the kurtosis value is greater than or equal to the preset kurtosis value, the fault is mainly manifested as a significant increase in the amplitude of a specific frequency. By comparing the maximum deviation between the current spectrum and the historical normal spectrum, this amplitude anomaly can be captured directly and accurately. When the kurtosis value is less than the preset kurtosis value, it may be difficult to find obvious differences by directly comparing the amplitudes of each frequency. However, under normal conditions, the amplitude of higher-order harmonics will decrease rapidly with increasing order. When a fault occurs, this law will be broken, resulting in a significant change in the harmonic law comparison degree. Therefore, by using the harmonic law comparison degree, the qualitative change in the harmonic ratio relationship can be keenly captured.

[0095] Specifically, if the maximum data correlation degree is greater than or equal to the preset maximum data correlation degree, then matching and generation will be performed based on the data correlation degree;

[0096] During the matching process, the maintenance strategy corresponding to the maintenance fault data with the highest correlation to the current period monitoring data is used as the generated maintenance strategy.

[0097] Among them, the maximum data correlation degree is the maximum value of the data correlation degree between the current period monitoring data and each maintenance fault data;

[0098] The correlation between the current period monitoring data and the data of a single maintenance fault is calculated as follows: r1 / (average of the first correlation corresponding to the historical records that can meet user needs) + r2 / (average of the second correlation corresponding to the historical records that can meet user needs).

[0099] The formula for calculating the first correlation r1 between the current periodic monitoring data and the data from a single maintenance fault is as follows:

[0100]

[0101] Where i is 1, 2, 3, α11, α12 and α13 are the vibration anomaly degree, temperature anomaly degree and maximum anomaly signal characterization value corresponding to the current cycle monitoring data, respectively, α21, α22 and α23 are the vibration anomaly degree, temperature anomaly degree and maximum anomaly signal characterization value corresponding to a single maintenance fault data, β1 is the preset vibration anomaly degree, β2 is the preset temperature anomaly degree and β3 is the preset maximum anomaly signal characterization value;

[0102] The second correlation r2 between the current period monitoring data and the single maintenance fault data is the minimum value among the sub-correlation degrees corresponding to each monitoring point;

[0103] For the current cycle monitoring data and any maintenance fault data, the time points of the maintenance fault data are selected and analyzed in order from earliest to latest. For a single time point, the time point and the time points after the time point are selected in sequence until the number of selected time points is the same as the number of time points in the current cycle monitoring data. Then, the selected time points are recorded as an analysis time interval. The selection and analysis are continued for the time points that have not been selected and analyzed until the number of selected time points cannot reach the number of time points in the current cycle monitoring data, at which point the selection and analysis stops.

[0104] The sub-correlation degree corresponding to a single monitoring point is the maximum value of the interaction coefficients among the analysis time intervals of the current cycle monitoring data and the single maintenance fault data.

[0105] The formula for calculating the interaction coefficient r0 between the current periodic monitoring data and the individual maintenance fault data for a single analysis time interval is as follows:

[0106]

[0107] Where m is the number of time points in the current period's monitoring data; x k and y k These are the vibration velocity values ​​corresponding to the k-th time point within a single analysis time interval in the current cycle monitoring data and the individual maintenance fault data, respectively. This is the average value of the vibration velocity at each time point in the current cycle monitoring data. The average value of vibration velocity at each time point within a single analysis time interval in a single maintenance fault data set, k = 1, 2, 3, ..., m;

[0108] The user can determine the preset maximum data correlation value based on the actual application scenario. The smaller the preset maximum data correlation value, the greater the user's demand for matching and generating data based on data correlation. The system provides a preset maximum data correlation value, detects the historical records generated by the user based on data correlation, and records the average of the maximum data correlation values ​​corresponding to the historical records that meet the user's needs as the preset maximum data correlation value.

[0109] Specifically, if the maximum data correlation is less than the preset maximum data correlation, then analysis and generation will proceed.

[0110] Understandably, when the maximum data correlation is greater than or equal to the preset maximum data correlation, it indicates that the current periodic monitoring data and a certain maintenance fault data have a very high feature matching degree, and there are mature maintenance strategies that can effectively adapt to the current scenario. Therefore, matching and generation are performed based on the data correlation. When the maximum data correlation is less than the preset maximum data correlation, it indicates that the current data and all historical reference data have significant feature differences, and existing experience cannot fully cover the current scenario. Therefore, it is necessary to develop maintenance strategies through analysis and generation.

[0111] Specifically, if the strategy similarity corresponding to each matched maintenance fault data is greater than or equal to the preset strategy similarity, then maintenance strategies are generated directly or in combination based on the correlation degree of strategy differences.

[0112] If the correlation degree of the strategy difference is less than the preset correlation degree of the strategy difference, then a maintenance strategy is generated by combination.

[0113] If the correlation degree of strategy difference is greater than or equal to the preset correlation degree of strategy difference, then the maintenance strategy is generated directly.

[0114] The matched maintenance fault data refers to maintenance fault data whose abnormal matching degree with the current periodic monitoring data is greater than the preset abnormal matching degree.

[0115] Wherein, strategy similarity = number of identical keywords / average number of keywords in the maintenance strategies corresponding to each matched maintenance fault data; identical keywords are keywords that appear in all the maintenance strategies corresponding to each matched maintenance fault data;

[0116] The value of the preset strategy similarity can be determined by the user according to the actual application scenario. The greater the user's need to improve the accuracy of maintenance strategy generation, the greater the value of the preset strategy similarity. One preset strategy similarity value is provided, with a preset strategy similarity of 70%.

[0117] Other related words in the maintenance strategies corresponding to each matched maintenance fault data, excluding the same keywords, are recorded as differential keywords. The strategy difference correlation degree is the average of the evaluation values ​​corresponding to each differential keyword.

[0118] For a single differential keyword, the differential keyword is recorded as the target differential keyword, and other differential keywords other than the target keyword are recorded as reference differential keywords. The matching maintenance fault data that appears with the target differential keyword is recorded as the analysis data. The evaluation value corresponding to the target differential keyword = the average number of reference differential keywords corresponding to each analysis data / the total number of different differential keywords appearing in the maintenance strategy corresponding to each matching maintenance fault data.

[0119] The value of the preset strategy difference correlation degree can be determined by the user according to the actual application scenario. The larger the value of the preset strategy difference correlation degree, the greater the user's demand for the combined generation of maintenance strategies. A method for determining the value of the preset strategy difference correlation degree is provided, which detects the historical records of user combined generation of maintenance strategies and records the average value of the strategy difference correlation degree corresponding to the historical records that can meet the user's needs as the preset strategy difference correlation degree.

[0120] When generating maintenance strategies, select the matching maintenance fault data with the highest strategy representativeness and record it as the first data. Record other matching maintenance fault data outside the first data as the second data. Record the keywords appearing in the first data as the first keywords. Record the keywords in the second data that are different from the first data as the second keywords. Record the second keywords with a combination coefficient greater than the preset combination coefficient as the selected keywords. Use each first keyword and each selected keyword as the generated maintenance strategy.

[0121] For a single second keyword, the second data in which the second keyword appears is recorded as the second data of analysis, and the collocation coefficient corresponding to the second keyword is the average number of first keywords corresponding to each second data of analysis;

[0122] The user can determine the value of the preset combination usage coefficient according to the actual application scenario. The greater the user's need for precision to improve the effectiveness of the maintenance strategy, the greater the value of the preset combination usage coefficient. A method for determining the value of the preset combination usage coefficient is provided, which is the average value of the combination usage coefficients corresponding to each selected keyword in the historical records that can meet the user's needs.

[0123] When generating maintenance strategies directly, the maintenance strategy with the highest representativeness corresponding to the matching maintenance fault data is selected as the generated maintenance strategy.

[0124] The representativeness of the maintenance strategy corresponding to a single matched maintenance fault data is: the number of identical keywords in the maintenance strategy / the average number of identical keywords in the maintenance strategies corresponding to all matched maintenance fault data + the average evaluation value of each different keyword in the maintenance strategy / the average evaluation value of each different keyword in the maintenance strategy corresponding to all matched maintenance fault data.

[0125] An abnormal monitoring point is a monitoring point whose sub-vibration anomaly degree is greater than or equal to the preset vibration anomaly degree, or whose sub-temperature anomaly degree is greater than or equal to the preset temperature anomaly degree, or whose abnormal signal characterization value is greater than or equal to the preset abnormal signal characterization value.

[0126] For the current cycle monitoring data and a single maintenance fault data, the abnormal monitoring points that are the same as the abnormal monitoring points corresponding to the current cycle monitoring data and the abnormal monitoring points corresponding to the maintenance fault data are recorded as the same abnormal monitoring points. Other abnormal monitoring points that are not the same abnormal monitoring points corresponding to the current cycle monitoring data are recorded as the first abnormal monitoring points. Other abnormal monitoring points that are not the same abnormal monitoring points corresponding to the maintenance fault data are recorded as the second abnormal monitoring points.

[0127] Anomaly matching degree = Quantity matching degree / Average of the quantity matching degree of each maintenance fault data and the current cycle monitoring data + (1 - Position deviation degree / Average of the position deviation degree of each maintenance fault data and the current cycle monitoring data);

[0128] Quantity matching degree = the larger of the number of identical abnormal monitoring points and the number of abnormal monitoring points corresponding to the current period monitoring data and the single maintenance fault data;

[0129] Location deviation = |Average value of distance reference value corresponding to each first abnormal monitoring point -Average value of distance reference value corresponding to each second abnormal monitoring point|; The distance reference value corresponding to a single first abnormal monitoring point is the minimum of the shortest distances from that first abnormal monitoring point to each of the same abnormal monitoring points, and the distance reference value corresponding to a single second abnormal monitoring point is the minimum of the shortest distances from that second abnormal monitoring point to each of the same abnormal monitoring points;

[0130] The value of the preset anomaly matching degree can be determined according to the actual application scenario. The greater the user's demand for improving the accuracy of maintenance strategy generation, the greater the value of the preset anomaly matching degree. A method for determining the value of the preset anomaly matching degree is provided, which records the anomaly matching degree corresponding to each matched maintenance fault data in the historical records that can meet the user's needs as the preset anomaly matching degree.

[0131] Understandably, when the strategy similarity is greater than or equal to the preset strategy similarity, it indicates that the maintenance strategies corresponding to the matched maintenance fault data are similar. Therefore, maintenance strategies are generated directly or in combination based on the strategy difference correlation, and the strategy difference correlation effectively reflects the strength of the correlation between detailed differences. When the strategy difference correlation is less than the preset strategy difference correlation, it indicates that the correlation between detailed differences is small. Combining and generating maintenance strategies can integrate all effective details to ensure comprehensiveness. When the strategy difference correlation is greater than or equal to the preset strategy difference correlation, it indicates that the correlation between detailed differences is large. Directly generating maintenance strategies can focus on core measures and avoid strategy redundancy and operational complexity caused by over-integration of closely related details, thereby improving the efficiency and accuracy of maintenance execution.

[0132] Specifically, if the strategy similarity corresponding to each matched maintenance fault data is less than the preset strategy similarity, then the maintenance strategy is determined based on the maximum abnormal matching threshold or the vectorized knowledge graph.

[0133] If the maximum anomaly matching threshold is greater than or equal to the preset maximum anomaly matching threshold, then a maintenance strategy is generated based on the anomaly matching threshold.

[0134] If the maximum anomaly matching threshold is less than the preset maximum anomaly matching threshold, then a maintenance strategy is generated based on the target vectorized knowledge graph.

[0135] Among them, the maximum anomaly matching threshold is the maximum value among the anomaly matching thresholds corresponding to the current period monitoring data and each maintenance fault data;

[0136] The anomaly matching threshold between the current period monitoring data and a single maintenance fault data is calculated as: anomaly matching degree / preset anomaly matching degree + anomaly interaction degree / preset anomaly interaction degree; the anomaly interaction degree is the minimum value among the sub-correlation degrees corresponding to each anomaly monitoring point appearing in the current period monitoring data and the single maintenance fault data.

[0137] The values ​​of preset abnormal interaction degree and preset abnormal matching degree can be determined according to the actual application scenario. The greater the user's demand for generating maintenance strategies based on the abnormal matching threshold, the smaller the values ​​of preset abnormal interaction degree and preset abnormal matching degree. A method for determining the values ​​of preset abnormal interaction degree and preset abnormal matching degree is provided, which detects the historical records of users generating maintenance strategies based on the abnormal matching threshold, and records the average value of abnormal interaction degree and the average value of abnormal matching degree corresponding to the historical records that can meet the user's needs as preset abnormal interaction degree and preset abnormal matching degree, respectively.

[0138] When generating maintenance strategies based on anomaly matching thresholds, the maintenance strategy corresponding to the maintenance fault data with the largest anomaly matching threshold is used as the generated maintenance strategy.

[0139] When generating maintenance strategies based on target vectorized knowledge graphs, the current period monitoring data can be input into the trained target vectorized knowledge graph to generate maintenance strategies.

[0140] The construction process of the target vectorized knowledge graph includes: using several maintenance fault data as training data and training the target vectorized knowledge graph. The training process includes, but is not limited to, data preprocessing, dataset partitioning, and setting hyperparameters. During data preprocessing, the maintenance fault data is transformed into an "entity-relationship" structure that the knowledge graph can recognize, and the entities and relations are transformed into initial vectors. The vectors are divided into a training set (70%), a validation set (10%), and a test set (20%) in a 7:1:2 ratio. Users can set hyperparameters according to their actual needs. Hyperparameters include, but are not limited to, vector dimension, learning rate, and number of iterations. These are all common techniques used by those skilled in the art and will not be elaborated on in detail. The trained target vectorized knowledge graph takes the current period monitoring data as input and the maintenance strategy as output.

[0141] Understandably, when the strategy similarity is less than the preset strategy similarity, it indicates that the maintenance strategies corresponding to each matched maintenance fault data are significantly different. Therefore, the maintenance strategy is determined based on the maximum anomaly matching threshold or the vectorized knowledge graph. The maximum anomaly matching threshold effectively reflects the matching depth between the current anomaly features and historical faults. When the maximum anomaly matching threshold is greater than or equal to the preset maximum anomaly matching threshold, it indicates that the current anomaly features are highly similar to the anomaly of a certain matched maintenance fault data. In this case, generating a maintenance strategy based on the anomaly matching threshold can rely on the most similar historical anomaly cases to quickly reuse validated effective strategies, ensuring the accuracy of maintenance direction and response efficiency. When the maximum anomaly matching threshold is less than the preset maximum anomaly matching threshold, it indicates that the current anomaly features are significantly different from all matched maintenance fault data. In this case, generating a maintenance strategy through the target vectorized knowledge graph can infer and generate targeted strategies adapted to new features, breaking through the limitations of historical experience and ensuring effective response to new or complex faults.

[0142] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1.A method for dynamically generating a power plant maintenance strategy based on a vectorized knowledge graph, characterized in that, include: Obtain several fault inspection data; The duration of the monitoring cycle for the target power generation equipment is determined based on the historical failure frequency and load anomaly of the target power generation equipment, and the duration of the monitoring cycle is adjusted to be reduced based on the failure impact under preset conditions. If the vibration anomaly degree corresponding to the current cycle monitoring data is greater than or equal to the preset vibration anomaly degree or the temperature anomaly degree is greater than or equal to the preset temperature anomaly degree, then a maintenance strategy will be generated. If the vibration anomaly degree corresponding to the current period monitoring data is less than the preset vibration anomaly degree and the temperature anomaly degree is less than the preset temperature anomaly degree, then a secondary judgment is made based on the anomaly signal characterization value. In the secondary judgment, the maximum value of the abnormal signal characterization value corresponding to each monitoring point is recorded as the maximum abnormal signal characterization value, and the maintenance strategy generation is determined based on the maximum abnormal signal characterization value. If the maximum abnormal signal characterization value is greater than or equal to the preset maximum abnormal signal characterization value, then a maintenance strategy will be generated. In the maintenance strategy generation process, if the maximum data correlation degree is greater than or equal to the preset maximum data correlation degree, the strategy will be generated based on the data correlation degree; if the maximum data correlation degree is less than the preset maximum data correlation degree, the strategy will be generated based on analysis. During the analysis and generation process, the anomaly matching degree is determined based on the number matching degree and the location deviation degree of the anomaly monitoring points, and the matching maintenance fault data corresponding to the current cycle monitoring data is determined based on the anomaly matching degree. If the strategy similarity corresponding to each matched maintenance fault data is greater than or equal to the preset strategy similarity, then maintenance strategies are generated directly or in combination based on the correlation degree of strategy differences. If the correlation degree of the strategy difference is less than the preset correlation degree of the strategy difference, then a maintenance strategy is generated by combination. If the correlation degree of strategy difference is greater than or equal to the preset correlation degree of strategy difference, then the maintenance strategy is generated directly. If the strategy similarity corresponding to each matched maintenance fault data is less than the preset strategy similarity, then the maintenance strategy is determined based on the maximum abnormal matching threshold or the vectorized knowledge graph. If the maximum anomaly matching threshold is greater than or equal to the preset maximum anomaly matching threshold, then a maintenance strategy is generated based on the anomaly matching threshold. If the maximum anomaly matching threshold is less than the preset maximum anomaly matching threshold, then a maintenance strategy is generated based on the target vectorized knowledge graph. The preset condition is that the fault impact is greater than the preset fault impact, the duration of the monitoring cycle of the target power generation equipment is negatively correlated with the historical fault frequency and the load anomaly, and the decrease in the duration of the monitoring cycle is positively correlated with the fault impact. The method for confirming the degree of fault impact includes: if the vibration difference of each monitoring point is greater than or equal to the preset vibration difference, the degree of fault impact is determined according to the anomaly comparison coefficient; if the vibration difference of each monitoring point is less than the preset vibration difference, the degree of fault impact is determined according to the historical fault correlation. The method for confirming the abnormal signal characterization value includes: if the kurtosis characterization value is greater than or equal to the preset kurtosis characterization value, then the abnormal signal characterization value is determined based on the spectrum comparison deviation value; if the kurtosis characterization value is less than the preset kurtosis characterization value, then the abnormal signal characterization value is determined based on the harmonic law comparison degree; kurtosis characterization value = standard deviation of acceleration corresponding to each frequency / average difference of acceleration corresponding to each frequency; spectrum comparison deviation value is the maximum value among the sub-frequency deviation values ​​corresponding to each frequency, and the sub-frequency deviation value corresponding to a single frequency = acceleration corresponding to that frequency - average value of acceleration corresponding to that frequency at that monitoring point in the historical record that does not require maintenance strategy generation and can meet user needs; The formula for calculating the harmonic law comparison degree k is: ； in, This is the average frequency corresponding to the maximum acceleration at each monitoring point in historical records, which does not require maintenance strategy generation and can meet user needs. Let f be the acceleration corresponding to 1f, ... Let nf be the acceleration. Historical failure frequency refers to the number of times the target power generation equipment has failed. Load anomaly rate = Duration of the target power generation equipment being in an abnormal load state within a preset time period / Duration of the preset time period; Vibration variability is the average of the coefficients of variation of the target power generation equipment at each time point in the current monitoring cycle; the coefficient of variation for a single time point = the standard deviation of the vibration velocity at each monitoring point at that time point / the average of the vibration velocity at each monitoring point at that time point. 2.The vectorized knowledge graph based power generation equipment maintenance strategy dynamic generation method according to claim 1, characterized in that, During the matching process, the maintenance strategy corresponding to the maintenance fault data with the highest correlation to the current period monitoring data is used as the generated maintenance strategy.

Images