Gas turbine early warning analysis platform based on diagnosis knowledge base

Abstract

The invention discloses a gas turbine early-warning analysis platform based on a diagnosis knowledge base, and the platform comprises a data end which carries out the preliminary processing of the real-time monitoring data of a gas turbine, transmits the monitoring data meeting the requirements to the data end for storage, marks the monitoring data not meeting the requirements as fluctuation data, and carries out the early-warning analysis of the gas turbine. The fluctuation data is transmitted to an analysis end for prediction and analysis; the analysis end is used for obtaining a next fault maintenance time value and a trend time value corresponding to the fluctuation data by fitting historical data of gas turbines of the same model, and comparing a combination time node formed by combining the trend time value and a starting time node of the fluctuation data with the next fault maintenance time value; relates to the technical field of gas turbine fault early warning, and solves the problems that an existing gas turbine fault early warning system lacks statistical representativeness and is poor in generalization ability during use, the reliability of a model cannot be quoted, and ineffective early warning is generated due to the fact that deviation data generated during working are prone to misjudgment.

Description

Technical Field

[0001] The present invention relates to the technical field of gas turbine fault early warning technology, and in particular to a gas turbine early warning analysis platform based on a diagnostic knowledge base. Background Art

[0002] While gas turbines are booming, they frequently fail due to their harsh internal working environment and complex structure. What is more serious is that the traditional ARIMA detection algorithm can only detect serious deformation of components, and the sensitivity of this detection algorithm is low. In the invention patent with application number CN201910853890.3, a gas turbine fault warning system based on the SARIMA model is proposed, which includes a gas turbine data support module for data collection, data encryption, data processing, data decryption, data transmission and data storage of real-time data of the gas turbine of the SIS system. It has the advantages of accurate prediction results and multi-level professional analysis of gas turbine exhaust temperature.

[0003] This fault warning system only makes predictions based on data from a single gas turbine. This system lacks statistical representativeness, and the model will over-adapt to the noise or special characteristics of a single gas turbine, resulting in extremely poor generalization ability and an inability to verify the reliability of the model. Secondly, when fault judgment is made by measuring the degree of similarity between actual values and predicted values, since the gas turbine is not completely stable during operation, misjudgment is likely to occur when deviated data or data collection errors occur, resulting in invalid warnings. Therefore, it is necessary to address the shortcomings of existing technologies. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a gas turbine early warning analysis platform based on a diagnostic knowledge base, which solves the problems of the existing gas turbine fault early warning system lacking statistical representativeness and poor generalization ability when used, and being unable to verify the reliability of the model. At the same time, the deviation data generated during operation is prone to misjudgment, resulting in the generation of invalid early warnings.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a gas turbine early warning analysis platform based on a diagnostic knowledge base, comprising:

[0006] The data end processes the real-time monitoring data of the gas turbine and transmits the monitoring data that meets the requirements to the data end for storage. The monitoring data that does not meet the requirements is marked as fluctuation data, and then the fluctuation data is transmitted to the analysis end for prediction analysis.

[0007] On the analysis side, the next fault repair time value and trend time value corresponding to the set of fluctuation data are obtained by calculating the historical data of the same model gas turbine. The combined time node obtained by combining the trend time value and the starting time node of the fluctuation data is compared with the next fault repair time value, and a verification signal is generated if there is a prediction error;

[0008] The early warning end obtains the value of the fluctuation data corresponding to the time nodes of different fault degrees based on the verification signal, and compares the value of the fluctuation data with the values corresponding to different fault degrees. If they match, a fault warning is issued, otherwise a verification signal is generated.

[0009] Preferably, the specific method of preliminary processing is:

[0010] The real-time monitoring data collected by sensors are uniformly marked as category data;

[0011] A coordinate system is established, and the category data is set in the coordinate system with the occurrence time node and its actual value as the coordinates, and a fitting formula for all category data is generated. By substituting the occurrence time node into the fitting formula, the fitting value corresponding to the actual value of each category data is obtained, and then the absolute value of the difference between the actual value and the fitting value is compared with the fluctuation coefficient α, where α is a preset value. The category data corresponding to values less than α are transferred to the database for storage, and the category data corresponding to values greater than α are screened out, and then the screened category data are screened out for a second time.

[0012] Preferably, the specific method of secondary screening is:

[0013] Establish a number axis and statistical interval respectively, where the nodes of the number axis are set with equal time length, and the range of the statistical interval is several times the distance between two adjacent nodes of the number axis. Mark the selected type data on the number axis according to their occurrence time nodes;

[0014] By moving the statistical interval along the origin and nodes of the number axis in sequence, the number of time nodes occurring within it is counted until the sum of the numbers is greater than θ, where θ is a preset value. All types of data within the statistical interval and on its right are marked as fluctuation data, and the time node of the occurrence of the leftmost fluctuation data is used as the starting time node of the entire set of fluctuation data. The fluctuation data is then transmitted to the analysis end for early warning analysis.

[0015] Preferably, the specific method of analysis and prediction is:

[0016] The gas turbine to be monitored is the target gas turbine, and several gas turbines of the same model and with the same working environment are the reference gas turbines, and the historical operating data of the reference gas turbines are obtained;

[0017] Obtain the fault repair time value Si and trend time value Ei corresponding to the fluctuation data of the reference gas turbine, where i = 1, 2, 3, ..., n, S1 represents the time node when the gas turbine was first repaired, Sn represents the time node when the gas turbine was repaired for the nth time, and the trend time value Ei is the time between the starting time node of each set of fluctuation data and the corresponding next fault repair time node. Obtain the fault repair time value Td and its corresponding trend time value Qd corresponding to the target gas turbine, where d = 1, 2, 3, ..., n-1;

[0018] The maintenance time values Sn corresponding to several groups of reference gas turbines are calculated to obtain an average value Snp, and Snp is used as the predicted value of the next fault maintenance time value Tn of the target gas turbine. In the same way, the average value Enp of several groups of trend time values En is used as the predicted value of the trend time value Qn corresponding to the next fault maintenance of the target gas turbine;

[0019] Take the starting time node of the fluctuation data after the fault repair time value T(n-1) as the starting point, and the moment after Qn as the combined time node, and compare the combined time node with the repair time value Tn:

[0020] If they are equal, it means the forecast is accurate and normal monitoring and early warning are carried out;

[0021] Otherwise, it indicates that there is a prediction error, and a verification signal is generated and transmitted to the verification end for error correction.

[0022] Preferably, the specific method of fault warning is:

[0023] Obtain the fault node value Fij corresponding to each reference gas turbine trend time value Ei, where j = 1, 2, 3, and 4, representing a mild fault degree, a moderate fault degree, a severe fault degree, and a serious fault degree, respectively, wherein Fij represents the starting time of occurrence of different fault degrees, and use the average values Fn1p, Fn2p, Fn3p, and Fn4p corresponding to the fault node values Fn1, Fn2, Fn3, and Fn4, respectively, as the predicted values of the fault node values Kn1, Kn2, Kn3, and Kn4 corresponding to the target gas turbine trend time value Qn;

[0024] Based on the fluctuation data of the target gas turbine, the actual values corresponding to different fault degrees are marked as YZj, and the actual values of the fluctuation data corresponding to the occurrence time node and the fault node value Knj are marked as BDj;

[0025] The judgment is made through YZj-FD≤BDj≤YZj+FD, where FD is the preset data floating coefficient. If it is met, the corresponding early warning signal is generated to remind the maintenance personnel to perform maintenance, and the combined time node is used as the actual value of the target gas turbine fault repair time value Tn and transmitted to the data end for storage. Otherwise, it indicates that there is a prediction error of the fault node value Knj, and a verification signal is generated and transmitted to the verification end for error verification.

[0026] Preferably, the actual values of the mild fault degree, moderate fault degree, severe fault degree and serious fault degree are YZ1, YZ2, YZ3 and YZ4 respectively, and 25% YZ4 = YZ1, 50% YZ4 = YZ2, 75% YZ4 = YZ3, wherein the value of the time period between the fault node value Kn4 and the starting time node for generating the fluctuation data is the same as the trend time value Qn, and immediate maintenance is required when the serious fault degree is reached.

[0027] Preferably, the early warning analysis platform also includes a data terminal for entering reference data, maintenance data corresponding to the gas turbine and monitoring data, and storing the reference data, maintenance data and monitoring data, and also for transmitting parameter data and maintenance data to the collection terminal and the early warning terminal to provide a data basis for data analysis.

[0028] Preferably, the early warning analysis platform also includes a verification end for receiving a verification signal transmitted by the early warning end, obtaining the aging degree of the mechanical structure corresponding to the target gas turbine fluctuation data, and verifying whether the prediction error is caused by excessive aging of the target mechanical structure, wherein the target mechanical structure refers to the mechanical components in the gas turbine that need to be monitored.

[0029] Preferably, the specific verification method for excessive aging of the target mechanical structure is:

[0030] Obtain the mean temperature H2, mean humidity D2, mean stress Y2, and corresponding weight values Z1, Z2, and Z3 of the target mechanical structure when it is operating stably between the fault repair time value Sn-1 and the fault node value Kn1, where the specific values of the weight values Z1, Z2, and Z3 are preset by maintenance personnel based on the degree of influence of temperature, humidity, and stress on material aging, and obtain the specific values of the standard temperature H1, standard humidity D1, and standard stress Y1, where the standard temperature H1, standard humidity D1, and standard stress Y1 are experimental data obtained from aging tests;

[0031] according to Where V is the influence coefficient corresponding to the change in aging degree caused by actual temperature, humidity and stress changes;

[0032] according to Get the deviation coefficient β corresponding to the actual value;

[0033] The judgment is made by V-ε≤β≤V+ε, where ε is a preset value;

[0034] If it is met, it indicates prediction error caused by material aging;

[0035] If β < V - ε, it indicates that there is a prediction error caused by the failure of the associated mechanical structure;

[0036] If β>V+ε, it means that there is a data monitoring anomaly.

[0037] Preferably, the specific method of obtaining experimental data is:

[0038] Mechanical structures of the same model are obtained as samples, and accelerated aging tests are performed on the samples under the conditions of standard temperature H1, standard humidity D1 and standard stress Y1. Sampling is also performed regularly to test the mechanical and chemical properties of the mechanical structure materials. Accelerated aging tests are then performed on the samples with two of the three parameters, standard temperature H1, standard humidity D1 and standard stress Y1, as fixed values and the other parameter as a variable, so as to obtain the degree of influence of temperature, humidity and stress on sample aging, and obtain the corresponding weight values Z1, Z2 and Z3 respectively.

[0039] Beneficial effects

[0040] The present invention provides a gas turbine early warning analysis platform based on a diagnostic knowledge base. Compared with the existing technology, it has the following advantages:

[0041] (1) By fitting the historical data of the reference gas turbine, the next fault maintenance time value and its corresponding trend time value corresponding to the target gas turbine are obtained to improve the accuracy of the early warning analysis results. Based on the predicted values of different fault nodes, the set value and actual value of the fluctuation data are compared, which can not only verify the accuracy of the prediction analysis, but also correct the fault maintenance value.

[0042] (2) By obtaining the influence coefficient of the current aging degree of the target mechanical structure on the performance and comparing it with the deviation coefficient of the actual data, the specific cause of the prediction error can be analyzed to improve the accuracy of subsequent prediction analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a system framework diagram of the early warning analysis platform of the present invention;

[0044] Figure 2 This is a workflow diagram of the acquisition end of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] See also Figure 1-2 The present invention provides a gas turbine early warning analysis platform based on a diagnostic knowledge base:

[0047] As the first embodiment of the present application, it specifically includes a collection end and an analysis end;

[0048] The acquisition end is used to obtain real-time monitoring data of the gas turbine from the sensor, perform preliminary processing on the monitoring data, transmit the monitoring data that meets the requirements to the data end for storage, and mark the monitoring data that does not meet the requirements as fluctuation data. The fluctuation data is then transmitted to the analysis end for prediction and analysis of the next fault repair time value and trend time value;

[0049] The specific methods of initial processing are:

[0050] The real-time temperature data, pressure data, vibration data, and speed data collected by the sensor are uniformly marked as type data;

[0051] A coordinate system is established with time as the horizontal axis and value as the vertical axis, and the category data is set in the coordinate system with the occurrence time node and its actual value as the coordinates, wherein the occurrence time node is the moment value corresponding to the actual value of the single category data, and a fitting formula for all category data is generated. By substituting the occurrence time node into the fitting formula, the fitting value corresponding to the actual value of each category data is obtained, and the actual value and the value of the corresponding fitting data are subtracted, and then the absolute value of the difference is compared with the fluctuation coefficient α, wherein α is a preset value. If it is less than α, the specific corresponding category data is transferred to the database for storage. If it is greater than α, the corresponding category data is screened out, and the screened category data is screened out for secondary screening;

[0052] The specific method of secondary screening is:

[0053] Establish a number axis with equal time length of nodes, and mark the selected type data on the number axis according to their occurrence time nodes, establish a statistical interval, and the range of the statistical interval is several times the distance between two adjacent nodes of the number axis, and then overlap the left side of the statistical interval with the origin of the number axis and each node in order from left to right, and count the number of occurrence time nodes in the statistical interval at each movement, until the number and value are greater than θ, where θ is a preset value.

[0054] All types of data within and to the right of the statistical interval are marked as fluctuation data, and the occurrence time node of the fluctuation data on the leftmost side of the number axis in this group of fluctuation data is used as the starting time node of the entire group of fluctuation data, and then the fluctuation data is transmitted to the analysis end for early warning analysis.

[0055] The analysis end is used to receive the fluctuation data transmitted by the collection end and obtain the historical operation data of the gas turbine of the same model. By calculating the historical operation data, the fault repair time value and trend time value corresponding to the next fault of the group of fluctuation data are obtained, and the obtained fault repair time value and trend time value are transmitted to the early warning end for early warning analysis;

[0056] The specific methods of analysis and prediction are:

[0057] For ease of expression, the monitored gas turbine is marked as a target gas turbine, and several gas turbines of the same model and similar working environment as the target gas turbine are obtained and marked as reference gas turbines; and historical operating data of the reference gas turbines is obtained;

[0058] For the reference gas turbine, obtain each set of fluctuation data corresponding to it, as well as the fault repair time value Si and trend time value Ei corresponding to each set of fluctuation data, where i = 1, 2, 3, ..., n, where n represents the sequence number of repair times, S1 represents the time node when the gas turbine was first repaired, Sn represents the time node when the gas turbine was repaired for the nth time, and the trend time value Ei is the time length from the starting time node of each set of fluctuation data to the corresponding next fault repair time node;

[0059] For the target gas turbine, obtain a fault repair time value Td corresponding to each fault repair of the target gas turbine, and obtain a trend time value Qd corresponding to each fault repair of the target gas turbine, where d=1, 2, 3, ..., n-1;

[0060] According to the fault repair time value Sn of each reference gas turbine, the average value Snp corresponding to several groups of Sn is calculated, and Snp is used as the predicted value of the next fault repair time value Tn of the target gas turbine;

[0061] The average value Enp corresponding to several groups of En is calculated based on the fault trend time value En of each reference gas turbine, and Enp is used as the predicted value of the trend time value Qn corresponding to the next fault maintenance of the target gas turbine;

[0062] Take the starting time node of the fluctuation data after the fault repair time value T(n-1) as the starting point, and the moment after Qn as the combined time node, and compare the combined time node with the repair time value Tn:

[0063] If they are equal, it means the forecast is accurate and normal monitoring and early warning are carried out;

[0064] Otherwise, it indicates that there is a prediction error, and a verification signal is generated and transmitted to the verification end for error correction.

[0065] As the second embodiment of the present application, it specifically includes a data terminal and an early warning terminal;

[0066] The data end is used to enter reference data, maintenance data corresponding to the gas turbine, and monitoring data, and store the reference data, maintenance data, and monitoring data. The reference data includes expert experience, historical fault data, and equipment manuals. The maintenance data includes the type of fault and its corresponding maintenance time. It is also used to transmit parameter data and maintenance data to the collection end and the early warning end to provide a data basis for data analysis.

[0067] The early warning end is used to receive the execution signal transmitted by the analysis end, obtain the value of the fluctuation data corresponding to the time nodes of different fault degrees according to the verification signal, compare the value of the fluctuation data with the values corresponding to different fault degrees, and then issue a fault early warning for the fluctuation data;

[0068] The specific methods of fault warning are:

[0069] Obtain the fault node value Fij corresponding to the trend time value Ei of each reference gas turbine, where j = 1, 2, 3, 4, where Fij represents the starting time of occurrence of different fault degrees;

[0070] Then, the average values Fn1p, Fn2p, Fn3p, and Fn4p corresponding to the fault node values Fn1, Fn2, Fn3, and Fn4 are obtained respectively, and the average values Fn1p, Fn2p, Fn3p, and Fn4p are used as the predicted values of the fault node values Kn1, Kn2, Kn3, and Kn4 corresponding to the target gas turbine trend time value Qn;

[0071] Specifically, within the trend time value Qn, Kn1 represents the starting time corresponding to the occurrence of a minor fault, Kn2 represents the starting time corresponding to the occurrence of a moderate fault, Kn3 represents the starting time corresponding to the occurrence of a severe fault, and Kn4 represents the starting time corresponding to the occurrence of a serious fault. The value of the time period between the fault node value Kn4 and the starting time node of the fluctuation data is the same as the trend time value Qn, that is, the fault node value Qn4 corresponds to the maintenance time value Tn. When the fault reaches a serious level, immediate maintenance is required.

[0072] The classification method of mild fault, moderate fault, severe fault and serious fault is as follows:

[0073] The actual values corresponding to different fault degrees are marked as YZj, where j = 1, 2, 3, 4. The value corresponding to YZ1 is used as the judgment standard for a mild fault, the value corresponding to YZ2 is used as the judgment standard for a mild fault, and the value corresponding to YZ3 is used as the judgment standard for a mild fault. The value of the fluctuation data corresponding to a serious fault is marked as YZ4, and 25% YZ4 = YZ1, 50% YZ4 = YZ2, and 75% YZ4 = YZ3. The specific value of YZj is set by the maintenance personnel based on maintenance experience, and the specific value of YZj corresponding to different fault types is different.

[0074] Based on the fluctuation data of the target gas turbine, the actual values of the four fluctuation data corresponding to the occurrence time nodes and the fault node values Kn1, Kn2, Kn3, and Kn4 are marked as BDj, and judged by YZj-FD≤BDj≤YZj+FD, where FD is the preset data floating coefficient. If it meets the requirements, the corresponding mild warning signal, moderate warning signal, severe warning signal, and serious warning signal are generated to remind the maintenance personnel to perform maintenance, and the actual value of the target gas turbine fault maintenance time value Tn combined with the time node is transmitted to the data end for storage and used for correction of the average value. Otherwise, it indicates that there is a prediction error of the fault node value Knj, and a verification signal is generated and transmitted to the verification end for error verification.

[0075] As a third embodiment of the present application, it specifically includes a verification terminal, which is used to receive a verification signal transmitted by the early warning terminal, obtain the aging degree of the mechanical structure corresponding to the fluctuation data of the target gas turbine, and verify whether the prediction error is caused by excessive aging of the target mechanical structure, where the target mechanical structure refers to the mechanical component in the gas turbine that needs to be monitored:

[0076] The specific method of obtaining experimental data is as follows:

[0077] Obtain mechanical structures of the same model and batch as samples, place the samples in an aging tester, and conduct accelerated aging tests under the conditions of standard temperature H1, standard humidity D1, and standard stress Y1. Regularly sample and test the mechanical properties (such as strength, hardness, and toughness) and chemical properties (microstructure changes) of the mechanical structure materials. Then, conduct accelerated aging tests on the samples with two of the three parameters of standard temperature H1, standard humidity D1, and standard stress Y1 as fixed values and the other parameter as a variable, so as to obtain the degree of influence of temperature, humidity, and stress on sample aging, and obtain the corresponding weight values Z1, Z2, and Z3, respectively. The specific values of the weight values Z1, Z2, and Z3 are preset by maintenance personnel based on the degree of influence of temperature, humidity, and stress on material aging;

[0078] According to Kn1-S(n-1)=G, where G is the operation time value between the fault repair time value S(n-1) and the fault node value Kn1;

[0079] Obtain temperature change data of the target mechanical structure when it is in stable operation within the operation time G, obtain the mean temperature H2 of the mechanical structure in stable operation by averaging, and obtain the weight value Z1 corresponding to the mean temperature H2;

[0080] Using the same processing method as for the temperature value, obtain the mean humidity D2, mean stress Y2, and corresponding weight values Z2 and Z3 corresponding to the mechanical structure;

[0081] according to Where V is the influence coefficient corresponding to the change in aging degree caused by actual temperature, humidity and stress changes;

[0082] according to Get the deviation coefficient β corresponding to the actual value;

[0083] The judgment is made by V-ε≤β≤V+ε, where ε is the preset deviation floating value;

[0084] If it matches, it indicates a prediction error caused by material aging, and a replacement signal is generated to remind maintenance personnel to replace the mechanical structure according to the actual situation;

[0085] If β < V - ε, it indicates that there is a prediction error caused by a fault in the associated mechanical structure. A feedback signal is generated and transmitted to the acquisition end to obtain real-time monitoring data of the associated mechanical structure. The same processing method as the target mechanical structure is used to obtain whether the associated mechanical structure has a fault and the corresponding fault degree.

[0086] If β>V+ε, it indicates that there is a data monitoring anomaly, and an abnormal signal is generated to remind maintenance personnel to manually detect various parameters of the sensor and gas turbine respectively.

[0087] By comparing the influence coefficient of material aging on the data with the deviation coefficient of the actual data value, it is possible to determine whether the predicted error of the fault node value is caused by excessive material aging. The specific cause of the predicted error is obtained by analyzing the predicted error, and then a corresponding signal is generated to remind maintenance personnel to make corresponding maintenance treatments.

[0088] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A gas turbine early warning analysis platform based on a diagnostic knowledge base, characterized by: include: The data end processes the real-time monitoring data of the gas turbine and transmits the monitoring data that meets the requirements to the data end for storage. The monitoring data that does not meet the requirements is marked as fluctuation data, and then the fluctuation data is transmitted to the analysis end for prediction analysis. On the analysis side, the next fault repair time value and trend time value corresponding to the set of fluctuation data are obtained by calculating the historical data of the same model gas turbine. The combined time node obtained by combining the trend time value and the starting time node of the fluctuation data is compared with the next fault repair time value, and a verification signal is generated if there is a prediction error; The early warning end obtains the value of the fluctuation data corresponding to the time nodes of different fault degrees based on the verification signal, and compares the value of the fluctuation data with the values corresponding to different fault degrees. If they match, a fault warning is issued, otherwise a verification signal is generated.

2. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 1 is characterized in that: The specific methods of initial processing are: The real-time monitoring data collected by sensors are uniformly marked as category data; A coordinate system is established, and the category data is set in the coordinate system with the occurrence time node and its actual value as the coordinates, and a fitting formula for all category data is generated. By substituting the occurrence time node into the fitting formula, the fitting value corresponding to the actual value of each category data is obtained, and then the absolute value of the difference between the actual value and the fitting value is compared with the fluctuation coefficient α, where α is a preset value. The category data corresponding to values less than α are transferred to the database for storage, and the category data corresponding to values greater than α are screened out, and then the screened category data are screened out for a second time.

3. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 2 is characterized in that: The specific method of secondary screening is: Establish a number axis and statistical interval respectively, where the nodes of the number axis are set with equal time length, and the range of the statistical interval is several times the distance between two adjacent nodes of the number axis. Mark the selected type data on the number axis according to their occurrence time nodes; By moving the statistical interval along the origin and nodes of the number axis in sequence, the number of time nodes occurring within it is counted until the sum of the numbers is greater than θ, where θ is a preset value. All types of data within the statistical interval and on its right are marked as fluctuation data, and the time node of the occurrence of the leftmost fluctuation data is used as the starting time node of the entire set of fluctuation data. The fluctuation data is then transmitted to the analysis end for early warning analysis.

4. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 1 is characterized in that: The specific methods of analysis and prediction are: The gas turbine to be monitored is the target gas turbine, and several gas turbines of the same model and with the same working environment are the reference gas turbines, and the historical operating data of the reference gas turbines are obtained; Obtain the fault repair time value Si and trend time value Ei corresponding to the fluctuation data of the reference gas turbine, where i = 1, 2, 3, ..., n, S1 represents the time node when the gas turbine was first repaired, Sn represents the time node when the gas turbine was repaired for the nth time, and the trend time value Ei is the time between the starting time node of each set of fluctuation data and the corresponding next fault repair time node. Obtain the fault repair time value Td and its corresponding trend time value Qd corresponding to the target gas turbine, where d = 1, 2, 3, ..., n-1; The maintenance time values Sn corresponding to several groups of reference gas turbines are calculated to obtain an average value Snp, and Snp is used as the predicted value of the next fault maintenance time value Tn of the target gas turbine. In the same way, the average value Enp of several groups of trend time values En is used as the predicted value of the trend time value Qn corresponding to the next fault maintenance of the target gas turbine; Take the starting time node of the fluctuation data after the fault repair time value T(n-1) as the starting point, and the moment after Qn as the combined time node, and compare the combined time node with the repair time value Tn: If they are equal, it means the forecast is accurate and normal monitoring and early warning are carried out; Otherwise, it indicates that there is a prediction error, and a verification signal is generated and transmitted to the verification end for error correction.

5. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 1 is characterized in that: The specific methods of fault warning are: Obtain the fault node value Fij corresponding to each reference gas turbine trend time value Ei, where j = 1, 2, 3, and 4, representing a mild fault degree, a moderate fault degree, a severe fault degree, and a serious fault degree, respectively, wherein Fij represents the starting time of occurrence of different fault degrees, and use the average values Fn1p, Fn2p, Fn3p, and Fn4p corresponding to the fault node values Fn1, Fn2, Fn3, and Fn4, respectively, as the predicted values of the fault node values Kn1, Kn2, Kn3, and Kn4 corresponding to the target gas turbine trend time value Qn; Based on the fluctuation data of the target gas turbine, the actual values corresponding to different fault degrees are marked as YZj, and the actual values of the fluctuation data corresponding to the occurrence time node and the fault node value Knj are marked as BDj; The judgment is made through YZj-FD≤BDj≤YZj+FD, where FD is the preset data floating coefficient. If it is met, the corresponding early warning signal is generated to remind the maintenance personnel to perform maintenance, and the combined time node is used as the actual value of the target gas turbine fault repair time value Tn and transmitted to the data end for storage. Otherwise, it indicates that there is a prediction error of the fault node value Knj, and a verification signal is generated and transmitted to the verification end for error verification.

6. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 5 is characterized by: The actual values of the mild fault degree, moderate fault degree, severe fault degree and serious fault degree are YZ1, YZ2, YZ3 and YZ4 respectively, and 25% YZ4 = YZ1, 50% YZ4 = YZ2, 75% YZ4 = YZ3. The value of the time period between the fault node value Kn4 and the starting time node of the fluctuation data is the same as the trend time value Qn, and immediate maintenance is required when the fault degree is serious.

7. The gas turbine early warning analysis platform based on a diagnostic knowledge base according to claim 1 is characterized in that: The early warning analysis platform also includes a data terminal for entering reference data, maintenance data corresponding to the gas turbine, and monitoring data, and storing the reference data, maintenance data, and monitoring data. It is also used to transmit parameter data and maintenance data to the collection terminal and the early warning terminal to provide a data basis for data analysis.

8. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 1 is characterized in that: The early warning analysis platform also includes a verification terminal for receiving a verification signal transmitted by the early warning terminal, obtaining the aging degree of the mechanical structure corresponding to the target gas turbine fluctuation data, and verifying whether the prediction error is caused by excessive aging of the target mechanical structure, where the target mechanical structure refers to the mechanical components in the gas turbine that need to be monitored.

9. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 8 is characterized by: The specific verification method for excessive aging of the target mechanical structure is: Obtain the mean temperature H2, mean humidity D2, mean stress Y2, and corresponding weight values Z1, Z2, and Z3 of the target mechanical structure when it is operating stably between the fault repair time value Sn-1 and the fault node value Kn1, where the specific values of the weight values Z1, Z2, and Z3 are preset by maintenance personnel based on the degree of influence of temperature, humidity, and stress on material aging, and obtain the specific values of the standard temperature H1, standard humidity D1, and standard stress Y1, where the standard temperature H1, standard humidity D1, and standard stress Y1 are experimental data obtained from aging tests; according to Where V is the influence coefficient corresponding to the change in aging degree caused by actual temperature, humidity and stress changes; according to Get the deviation coefficient β corresponding to the actual value; The judgment is made by V-ε≤β≤V+ε, where ε is a preset value; If it is met, it indicates prediction error caused by material aging; If β < V - ε, it indicates that there is a prediction error caused by the failure of the associated mechanical structure; If β>V+ε, it means that there is a data monitoring anomaly.

10. The gas turbine early warning analysis platform based on the diagnostic knowledge base according to claim 9 is characterized in that: The specific method of obtaining experimental data is as follows: Mechanical structures of the same model are obtained as samples, and accelerated aging tests are performed on the samples under the conditions of standard temperature H1, standard humidity D1 and standard stress Y1. Sampling is also performed regularly to test the mechanical and chemical properties of the mechanical structure materials. Accelerated aging tests are then performed on the samples with two of the three parameters, standard temperature H1, standard humidity D1 and standard stress Y1, as fixed values and the other parameter as a variable, so as to obtain the degree of influence of temperature, humidity and stress on sample aging, and obtain the corresponding weight values Z1, Z2 and Z3 respectively.

Images