Power generation terminal equipment operation inspection system and method based on large model

Abstract

The invention relates to the technical field of power generation terminal equipment operation inspection, in particular to a power generation terminal equipment operation inspection system and method based on a large model. According to the invention, the real-time data of the power generation terminal equipment is synchronously obtained through the sliding window mechanism, so that the obtained real-time data is more effective; a decision tree is constructed through historical operation parameters and historical fault types, obtained through training, of the power generation terminal equipment in the historical operation process, the decision tree can be effectively constructed, and meanwhile the decision tree can more accurately output the fault types based on real-time data; and determining an operation inspection state based on the matching degree of the output fault type and the actual fault type, and determining an adjustment parameter based on the operation inspection state. According to the method, the parameters are adjusted based on the operation and inspection state, so that collaborative analysis of multi-source data is enhanced, the cross-modal association capability is improved, and the accuracy of operation and inspection of the power generation equipment is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power generation terminal equipment operation and inspection, and particularly relates to a power generation terminal equipment operation and inspection system and method based on a large model. BACKGROUND

[0002] The operation and inspection background technology of power generation terminal equipment has experienced a transformation from manual inspection to intelligent operation and maintenance. In the early stage, it mainly relied on regular manual inspection, paper records and simple instrument detection, which had the problems of low efficiency and high risk of missing detection. With the application of sensor technology and SCADA system, real-time monitoring and data collection of equipment state were realized. In recent years, combined with the Internet of Things platform, edge computing and digital twin technology, a closed-loop system of state perception-data analysis-fault warning was constructed, and non-contact detection means such as unmanned aerial vehicle inspection and infrared thermal imaging were introduced. The current technical trend is centered on AI-driven predictive maintenance, which realizes fault prediction through machine learning analysis of historical data, combines AR remote collaboration and robot automatic inspection, and forms an intelligent operation and inspection mode of "cloud-edge-end" cooperation, which significantly improves equipment reliability and operation and maintenance economy.

[0003] Chinese Patent Publication No. CN110112771 B discloses a method and device for detecting the operation state of a generator of a nuclear power plant. The state of the generator set is determined according to the on-off state of the generator outlet switch and the high-voltage side switch of the main transformer. When the generator is in normal operation, the generator set is in a split state, and the line current of the generator is greater than the minimum load current, it is determined that the generator is in an island operation state. Compared with the prior art, the present application accurately determines whether the generator is in an island operation state by the line current state of the generator outlet switch, the high-voltage side switch of the main transformer and the generator.

[0004] As can be seen, the prior art has the following problems: in the prior art, multi-source data is analyzed in isolation, and lacks cross-modal correlation capability, thereby reducing the accuracy of power equipment operation and inspection. SUMMARY

[0005] Therefore, the present application provides a power generation terminal equipment operation and inspection system and method based on a large model to overcome the problem that in the prior art, multi-source data is analyzed in isolation, and lacks cross-modal correlation capability, thereby reducing the accuracy of power equipment operation and inspection.

[0006] To achieve the above purpose, the present application provides a power generation terminal equipment operation and inspection system based on a large model, comprising:

[0007] a real-time data acquisition unit, configured to synchronously acquire infrared thermal images and vibration phase signals generated by a plurality of sensor monitoring points arranged at the power generation terminal device and collect electrical operation parameters through a sliding window mechanism, wherein all the data are time-aligned according to a set sliding window length;

[0008] a historical data acquisition unit, configured to acquire historical operation parameters and historical fault types of the power generation terminal device in a historical operation process, wherein the historical operation parameters include the infrared thermal images, the vibration phase signals and the electrical operation parameters;

[0009] a model training unit, connected to the historical data acquisition unit, configured to train and build a decision tree based on the data acquired by the historical data acquisition unit, wherein the training parameters required for building the decision tree include a maximum depth when training a constraint tree structure, a minimum sample number of a leaf node, a training iteration number of the decision tree and an initial learning rate;

[0010] a model output unit, connected to the real-time data acquisition unit and the historical data acquisition unit respectively, configured to input the data acquired by the real-time data acquisition unit into the decision tree and output a fault type of the power generation terminal device in an abnormal operation state;

[0011] an analysis unit, connected to the model output unit, configured to determine an operation inspection state based on a matching degree of the fault types output by the model output unit and actual fault types in a preset time period, adjust the maximum depth when training the constraint tree structure in the training decision tree based on the operation inspection state, and adjust the training iteration number of the decision tree when the adjustment number is equal to a preset number and the operation inspection state is still unqualified.

[0012] Further, the analysis unit is further configured to draw a matching degree-pre-set time period curve based on the matching degrees in a plurality of preset time periods, wherein the above steps are performed when the operation inspection state is determined to be a first state, the first state is that the matching degree is less than a first preset matching degree and greater than a second preset matching degree; the analysis unit is further configured to determine whether to adjust the maximum depth when training the constraint tree structure based on an average value of absolute values of slopes corresponding to a plurality of nodes of the curve; and the analysis unit is further configured to increase the maximum depth when training the constraint tree structure based on a ratio of a sample amount acquired in the preset time period to a preset sample amount, and the increase amplitude of the maximum depth when training the constraint tree structure is proportional to the ratio.

[0013] Further, the analysis unit is further configured to increase the maximum depth based on the minimum sample number of the leaf node, and the increase amplitude of the maximum depth is proportional to the minimum sample number of the leaf node.

[0014] Further, the analysis unit is further configured to repeatedly adjust the maximum depth of the training constraint tree structure at least once until the adjustment number reaches a preset number or the adjustment number is less than the preset number, wherein the average value of the absolute value of the slope of the adjusted matching degree-preset time period curve is obtained, and the repeated adjustment operation is performed when the average value is less than or equal to a preset average value, and the operation and inspection state is qualified when the matching degree is greater than or equal to the first preset matching degree; the analysis unit is further configured to obtain an optimized matching degree when the adjustment number is equal to the preset number, and determine the adjusted operation and inspection state based on the optimized matching degree; and the analysis unit is further configured to adjust the training iteration number of the decision tree based on the optimized matching degree, wherein the adjustment is performed when the adjusted operation and inspection state is the first state.

[0015] Further, the analysis unit is further configured to increase the training iteration number of the decision tree based on the ratio of the optimized matching degree to the first preset matching degree, and the increase amplitude of the training iteration number of the decision tree is inversely proportional to the ratio.

[0016] Further, the analysis unit is further configured to decrease the initial learning rate based on the training iteration number of the decision tree, and the decrease amplitude of the initial learning rate is proportional to the training iteration number.

[0017] Further, the analysis unit is further configured to calculate an optimized average value by calculating the average value of the absolute value of the slope of the matching degree-preset time period curve drawn based on the matching degree re-detected after the above adjustment step; the analysis unit is further configured to draw a rotor current-time curve based on the rotor current of the generator in each preset time period obtained by the real-time data obtaining unit, wherein the above step is performed when the operation and inspection state is determined to be a second state or the optimized average value is less than or equal to the preset average value, and the second state is that the matching degree is less than or equal to the second preset matching degree; and the analysis unit is further configured to reduce the sliding window length based on the variance of the absolute value of the slope of the rotor current-time curve, and the reduction amplitude of the sliding window length is proportional to the variance.

[0018] Further, the analysis unit is further configured to reduce the sliding step length based on the sliding window length, and the reduction amplitude of the sliding step length is inversely proportional to the sliding window length.

[0019] Further, the analysis unit is further configured to increase the range reference of the infrared sensor based on the sliding window length, and the increase amplitude of the range reference of the infrared sensor is inversely proportional to the sliding window length.

[0020] To achieve the above-mentioned purpose, the present application provides a power terminal equipment operation and inspection method based on a large model, comprising:

[0021] Real-time synchronous acquisition of infrared thermal images and vibration phase signals generated by a plurality of sensor monitoring points arranged at the power generation terminal equipment and collection of electrical operation parameters are performed through a sliding window mechanism, wherein all real-time data are time-aligned according to a set sliding window length;

[0022] Historical operation parameters and historical fault types of the power generation terminal equipment in a historical operation process are acquired, wherein the historical operation parameters include infrared thermal images, vibration phase signals and electrical operation parameters;

[0023] Data acquired based on the historical operation parameters and the historical fault types are used for training to construct a decision tree, wherein training parameters required for construction of the decision tree include a maximum depth during training of a constraint tree structure, a minimum number of leaf nodes, a training iteration number of the decision tree and an initial learning rate;

[0024] The acquired real-time data are input into the decision tree, and a fault type of the power generation terminal equipment during abnormal operation is output;

[0025] A matching degree between the output fault type and an actual fault type within a preset time period is used to determine an operation inspection state, and the maximum depth during training of the constraint tree structure in the training decision tree is adjusted based on the operation inspection state, and the training iteration number of the decision tree is adjusted when the adjustment number is equal to a preset number and the operation inspection state is still unqualified.

[0026] Compared with the prior art, the present application has the beneficial effects that the present application synchronously acquires real-time data of the power generation terminal equipment through a sliding window mechanism, so that the acquired real-time data are more effective, and a decision tree is constructed based on historical operation parameters and historical fault types of the power generation terminal equipment in a historical operation process acquired through training, so that the decision tree is effectively constructed, and the decision tree can more accurately output a fault type based on real-time data; and an operation inspection state is determined based on a matching degree between the output fault type and an actual fault type, and an adjustment parameter is determined based on the operation inspection state. The present application cooperatively analyzes multi-source data through a model, strengthens cross-modal correlation ability, and thus improves the accuracy of operation inspection of the power generation equipment.

[0027] Further, the present application determines a matching degree change fluctuation condition based on an average value of an absolute value of a slope of a matching degree-pre-set time period curve, and determines whether to adjust the maximum depth during training of the constraint tree structure based on the reason for unqualification of the operation inspection state, so that the maximum depth during training of the constraint tree structure is more accurately adjusted, so that data are more accurately predicted, and thus the accuracy of operation inspection of the power generation equipment is further improved.

[0028] Further, the present application can prevent the tree from over-refining noise and retain the resolution of rare faults when there are few key fault samples, thereby further improving the accuracy of the device operation inspection.

[0029] Further, the present application can solve the problem of unqualified operation inspection state caused by underfitting of the gradient boosting tree in the decision tree after training according to the original training iteration number, thereby further improving the efficiency of the operation inspection of the power terminal device.

[0030] Further, the present application can make the prediction result of the model more accurate, thereby further improving the accuracy of monitoring the operation inspection state of the power terminal device and further improving the efficiency of the operation inspection of the power terminal device.

[0031] Further, the present application can avoid ignoring the detailed mode due to the increase in the training iteration number, make the prediction result of the model more accurate, and thereby further improve the efficiency of the operation inspection of the power terminal device.

[0032] Further, the present application can make the decision tree have no time delay when detecting the fault type by adjusting the length of the sliding window, thereby matching the detected fault type with the actual time period, and further improving the accuracy of the operation inspection of the power terminal device.

[0033] Further, the present application can make the obtained data more accurate by adjusting the sliding step after adjusting the length of the sliding window, thereby further improving the prediction result of the decision tree and further improving the accuracy of the operation inspection of the power terminal device.

[0034] Further, the present application can reduce the sampling database and increase more effective data by adjusting the range reference of the infrared sensor of the obtained infrared thermal image after adjusting the sliding window and the sliding step, thereby making the prediction result of the decision tree more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The structure diagram of the power terminal device operation inspection system based on the large model according to the embodiment of the present application;

[0036] Figure 2 The step flow chart of the power terminal device operation inspection method based on the large model according to the embodiment of the present application;

[0037] Figure 3 FIG. 1 is a flowchart of a processing procedure when the operation inspection state is determined as the first state based on the matching degree and the preset matching degree according to an embodiment of the present application;

[0038] Figure 4 FIG. 2 is a flowchart of a processing procedure when the operation inspection state is determined as the second state based on the matching degree and the preset matching degree according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the objects and advantages of the present application clearer, the present application will be further described in conjunction with embodiments. It should be understood that the specific embodiments described herein merely serve the purpose of explaining the present application and are not intended to limit the present application.

[0040] The preferred embodiments of the present application will be described below with reference to the drawings. It should be understood by those skilled in the art that the embodiments merely serve the purpose of explaining the technical principles of the present application and are not intended to limit the protection scope of the present application.

[0041] It should be noted that in the description of the present application, unless explicitly defined and limited, the terms "mounting", "connection" and "linking" should be understood in a broad sense, for example, can be fixed connection, can also be detachable connection, or integral connection; can be mechanical connection, can also be electrical connection; can be direct connection, can also be indirect connection through an intermediate medium, or internal communication of two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0042] Please refer to Figure 1 FIG. 1 is a structure schematic diagram of a power generation terminal equipment operation inspection system based on a large model according to an embodiment of the present application.

[0043] The system comprises a real-time data acquisition unit, a historical data acquisition unit, a model training unit, a model output unit and an analysis unit.

[0044] The real-time data acquisition unit is used to synchronously acquire infrared thermographs and vibration phase signals generated by a plurality of sensor monitoring points arranged in the power generation terminal equipment through a sliding window mechanism, and to collect electrical operation parameters, wherein all the data are time-aligned according to a set sliding window length;

[0045] The historical data acquisition unit is used to acquire historical operation parameters and historical fault types of the power generation terminal equipment in the historical operation process, wherein the historical operation parameters comprise infrared thermographs, vibration phase signals and electrical operation parameters;

[0046] The model training unit is connected with the historical data acquisition unit, and is used to train based on the data acquired by the historical data acquisition unit to construct a decision tree, wherein the training parameters required for constructing the decision tree include the maximum depth when training the tree structure, the minimum sample number of leaf nodes, the training iteration number of the decision tree and the initial learning rate;

[0047] The model output unit is connected with the real-time data acquisition unit and the historical data acquisition unit respectively, and is used to input the data acquired by the real-time data acquisition unit into the decision tree, and output the fault type of the power terminal equipment in abnormal operation;

[0048] The analysis unit is connected with the model output unit, and is used to determine the operation inspection state based on the matching degree of the fault type output by the model output unit and the actual fault type in a preset time period, and adjust the maximum depth when training the tree structure in training the decision tree based on the operation inspection state, and adjust the training iteration number of the decision tree when the adjustment number is equal to a preset number and the operation inspection state is still unqualified.

[0049] Specifically, the core mechanism of the sliding window includes window division and dynamic adjustment and multi-source data alignment, wherein the window division and dynamic adjustment is that the sliding window divides the data stream through fixed length or variable length intervals, and the window can slide forward with the arrival of new data, so as to ensure continuous coverage of data while avoiding repeated calculation. The adjustment of window size and sliding step directly affects the data freshness and processing delay; the multi-source data alignment is that the sliding window can align the data from different sources through timestamps or serial numbers, and ensure the time consistency of the data in the window. This will not be repeated here.

[0050] Please refer to Figure 2 The figure is a step flow chart of the power terminal equipment operation inspection method based on a large model according to the embodiment of the application.

[0051] The process of the power terminal equipment operation inspection based on a large model is as follows:

[0052] S1, through the real-time data acquisition unit, the infrared thermograph and vibration phase signal generated by the plurality of sensor monitoring points arranged in the power terminal equipment are synchronously acquired by using the sliding window mechanism, and the electrical operation parameters are collected, wherein all the data are time-sequentially aligned according to the set sliding window length;

[0053] S2, the historical operation parameters and historical fault types of the power terminal equipment in the historical operation process are acquired by the historical data acquisition unit, wherein the historical operation parameters include the infrared thermograph, the vibration phase signal and the electrical operation parameters;

[0054] S3, a model training unit connected to the historical data acquisition unit trains based on the data acquired by the historical data acquisition unit to construct a decision tree, wherein the training parameters required for constructing the decision tree include the maximum depth when training the tree structure, the minimum number of leaf nodes, the number of training iterations of the decision tree, and the initial learning rate;

[0055] S4, a model output unit connected to the real-time data acquisition unit and the historical data acquisition unit respectively inputs the data acquired by the real-time data acquisition unit into the decision tree, and outputs the fault type of the power terminal device when it is running abnormally;

[0056] S5, an analysis unit connected to the model output unit determines the operation inspection state based on the matching degree of the fault type output by the model output unit and the actual fault type within a preset time period, and adjusts the maximum depth when training the tree structure in the training decision tree based on the operation inspection state, and adjusts the number of training iterations of the decision tree when the number of adjustments is equal to the preset number and the operation inspection state is still unqualified.

[0057] Please refer to Figure 3 The processing flow chart when the operation inspection state is determined to be the first state based on the matching degree and the pre-stored preset matching degree according to the embodiment of the application is shown in the figure. The analysis unit according to the embodiment of the application is also used to draw a matching degree-preset time period curve based on the matching degrees within a plurality of preset time periods, wherein the above steps are performed when the operation inspection state is determined to be the first state, and the first state is that the matching degree is less than the first preset matching degree and greater than the second preset matching degree; the analysis unit is also used to determine whether to adjust the maximum depth when training the tree structure based on the average value of the absolute values of the slopes corresponding to a plurality of nodes of the curve; the analysis unit is also used to increase the maximum depth when training the tree structure based on the ratio of the sample amount acquired within the preset time period to the preset sample amount, and the increase amplitude of the maximum depth when training the tree structure is proportional to the ratio.

[0058] Specifically, the matching degree L0 can be divided into the first preset matching degree L1 and the second preset matching degree L2, and the first preset matching degree L1 is set to 0.9 (i.e. 90%) and the second preset matching degree L2 is set to 0.75 (i.e. 75%) in the matching degree standard, and it should be noted that in other embodiments, the values of L1 and L2 can also be determined according to the operation inspection requirements of the power terminal device; the comparison process based on the matching degree L and L1 and L2 is as follows:

[0059] If the matching degree L is greater than or equal to the first preset matching degree L1, it indicates that the operation inspection state of the power terminal device is qualified;

[0060] If the matching degree L is less than the first preset matching degree L1 and greater than the second preset matching degree L2, the current operation inspection state is determined to be the first state;

[0061] If the matching degree L is less than the second preset matching degree L2, the current operation detection state is determined as the second state.

[0062] Specifically, if the operation detection state is determined as the first state, the matching degrees in a plurality of preset time periods are obtained, and a matching degree-pre-set time period curve is drawn, an average value of absolute values of slopes corresponding to a plurality of nodes of the curve is calculated, and if the average value is less than or equal to a preset average value, it indicates that the matching degree changes less, and the rare fault is not monitored due to a small amount of samples, which causes underfitting of the single decision tree, thereby leading to deviation of data prediction, and therefore, the maximum depth in training the constrained tree structure in training the decision tree is adjusted based on a ratio of the obtained sample amount to a preset sample amount, and the preset average value is 1.1%.

[0063] Specifically, the preset ratio Q0 of the obtained sample amount to the preset sample amount is 0.85, and the comparison process based on the ratio Q and the preset ratio Q0 is specifically as follows:

[0064] If the ratio Q is less than or equal to the preset ratio Q0, the maximum depth in training the constrained tree structure is adjusted to 1.07 times of the original maximum depth.

[0065] If the ratio Q is greater than the preset ratio Q0, the maximum depth in training the constrained tree structure is adjusted to 1.19 times of the original maximum depth.

[0066] Specifically, in order to prevent the tree from being excessively refined to noise and to retain the resolution ability of rare faults when the key fault samples are less, the maximum depth is adjusted based on a minimum sample number of a leaf node, wherein the preset minimum sample number R0 of the leaf node is 7, and the comparison process based on the minimum sample number R of the leaf node and the preset minimum sample number R0 of the leaf node is specifically as follows:

[0067] If the minimum sample number R of the leaf node is less than or equal to the preset minimum sample number R0 of the leaf node, the maximum depth in training the constrained tree structure is adjusted to 1.04 times of the maximum depth adjusted based on the sample amount.

[0068] If the minimum sample number R of the leaf node is less than or equal to the preset minimum sample number R0 of the leaf node, the maximum depth in training the constrained tree structure is adjusted to 1.21 times of the maximum depth adjusted based on the sample amount.

[0069] Specifically, the analysis unit is further configured to repeatedly adjust the maximum depth of the training constraint tree structure at least once until the adjustment number reaches a preset number or the adjustment number is less than the preset number, and the operation detection state is qualified, wherein an average value of an absolute value of a slope of the adjusted matching degree-pre-set time period curve is obtained, and the repeated adjustment operation is performed when the average value is less than or equal to a preset average value, and the standard for the operation detection state being qualified is that the matching degree is greater than or equal to the first preset matching degree; the analysis unit is further configured to obtain an optimized matching degree when the adjustment number is equal to the preset number, and determine the adjusted operation detection state based on the optimized matching degree; and the analysis unit is further configured to adjust the training iteration number of the decision tree based on the optimized matching degree, wherein the adjustment is performed when the adjusted operation detection state is the first state.

[0070] Specifically, the matching degree is adjusted based on the adjusted matching degree, the curve is redrawn, the average value of the absolute value of the slope of the curve is calculated, the maximum depth is repeatedly adjusted when the average value is less than or equal to the preset average value, the adjustment is stopped when the operation detection state is qualified when the adjustment number reaches the preset number or the adjustment number is less than the preset number, the optimized matching degree when the adjustment number is equal to the preset number is re-detected, and the operation detection state is determined based on the optimized matching degree. If the operation detection state is still the first state, it indicates that the gradient boosting tree in the decision tree appears under-fitting after training based on the original training iteration number, so that the result output by the decision tree is inaccurate, and the training iteration number of the decision tree is adjusted based on the ratio of the optimized matching degree to the first preset matching degree.

[0071] Specifically, the preset ratio T0 of the optimized matching degree to the first preset matching degree is 0.89, and the comparison process based on the ratio T and the preset ratio T0 is as follows:

[0072] If the ratio T is less than or equal to the preset ratio T0, the training iteration coefficient of the decision tree is adjusted to 2.3 times the original training iteration number.

[0073] If the ratio T is greater than the preset ratio T0, the training iteration coefficient of the decision tree is adjusted to 1.2 times the original training iteration number.

[0074] Specifically, in the gradient boosting framework, the decision tree is added as a weak learner, and the initial learning rate is required to control the contribution of each tree, which will not be described again.

[0075] Specifically, the preset training iteration number U0 of the decision tree is 500, that is, the preset total number of weak learners, and the comparison process based on the training iteration number U and the preset training iteration number U0 is as follows:

[0076] If the training iteration number U is less than or equal to the preset training iteration number U0, the initial learning rate is adjusted to 0.94 times the original initial learning rate.

[0077] If the number of training iterations U is greater than the preset number of training iterations U0, the initial learning rate is adjusted to 0.87 times the original initial learning rate.

[0078] Referring to Figure 4 The analysis unit is further configured to calculate an average value of absolute values of slopes of a matching degree-predefined time period curve drawn based on the matching degree re-detected after the adjusting, to obtain an optimized average value; the analysis unit is further configured to draw a rotor current-time curve based on rotor currents of the generator in each predefined time period obtained by the real-time data obtaining unit, wherein the above steps are performed when the operation detection state is determined to be the second state or the optimized average value is less than or equal to the preset average value, and the second state is that the matching degree is less than or equal to the second preset matching degree; and the analysis unit is further configured to reduce the length of the sliding window based on a variance of absolute values of slopes of the rotor current-time curve, and the reduction range of the length of the sliding window is proportional to the variance.

[0079] Specifically, the adjusted matching degree is re-acquired, the matching degree-predefined time period curve is redrawn based on the matching degree, and an average value of absolute values of slopes of the curve is calculated; if the average value of absolute values of slopes of a plurality of nodes of the curve is still less than or equal to the preset average value, it indicates that the decision tree detects the delay of the type of fault, that is, the type of fault should be detected in the A time period, but is detected in the A+1 time period, which causes a large number of detected types of faults to be mismatched with the actual time period; therefore, rotor currents of the generator in each predefined time period are obtained, and a rotor current-time curve is drawn, the length of the sliding window is reduced based on a variance of absolute values of slopes of the curve, thereby solving the problem of detection delay.

[0080] Specifically, the preset variance V0=0.5, and the comparison process based on the variance V and the preset variance V0 is specifically as follows:

[0081] If the variance V is less than or equal to the preset variance V0, the length of the sliding window is adjusted to 0.91 times the original length of the sliding window;

[0082] If the variance V is greater than the preset variance V0, the length of the sliding window is adjusted to 0.79 times the original length of the sliding window.

[0083] Specifically, after adjusting the length of the sliding window, the sliding step length also needs to be reduced synchronously, so that the key changes of the current can be captured more accurately. Therefore, the sliding step length is adjusted by the length of the sliding window, wherein the sliding step length is less than the length of the sliding window. The preset length of the sliding window W0 is 400 ms. The comparison process based on the length of the sliding window W and the preset length of the sliding window W0 is as follows:

[0084] If the length of the sliding window W is less than or equal to the preset length of the sliding window W0, the sliding step length is adjusted to 0.72 times the original sliding step length.

[0085] If the length of the sliding window W is greater than the preset length of the sliding window W0, the sliding step length is adjusted to 0.87 times the original sliding step length.

[0086] Specifically, after adjusting the length of the sliding window and the sliding step length, the sampling data will be saturated. Therefore, the range reference of the infrared sensor of the infrared thermal image obtained based on the length of the sliding window is adjusted during high-frequency monitoring, so that the sampling data is more effective.

[0087] Specifically, the preset length of the sliding window M0 is 350 ms. The comparison process based on the length of the sliding window W and the preset length of the sliding window M0 is as follows:

[0088] If the length of the sliding window W is less than or equal to the preset length of the sliding window M0, the range reference of the infrared sensor is adjusted to 1.98 times the original reference.

[0089] If the length of the sliding window W is greater than the preset length of the sliding window M0, the range reference of the infrared sensor is adjusted to 1.57 times the original reference.

[0090] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the drawings. However, those skilled in the art will readily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application. The technical solutions after these changes or replacements will all fall within the protection scope of the present application.

[0091] The above description is only the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various changes and modifications to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A power generation terminal equipment operation and maintenance system based on a large model, characterized in that, include: The real-time data acquisition unit is used to synchronously acquire infrared thermal images and vibration phase signals generated by several sensor monitoring points set on the power generation terminal equipment through a sliding window mechanism, and to collect electrical operating parameters. All data are time-aligned according to the set sliding window length. The historical data acquisition unit is used to acquire historical operating parameters and historical fault types of the power generation terminal equipment during historical operation. The historical operating parameters include infrared thermal images, vibration phase signals, and electrical operating parameters. The model training unit is connected to the historical data acquisition unit and is used to train a decision tree based on the data acquired by the historical data acquisition unit. The training parameters that need to be set when building the decision tree include the maximum depth when training the constraint tree structure, the minimum number of samples in the leaf nodes, the number of training iterations of the decision tree, and the initial learning rate. The model output unit is connected to the real-time data acquisition unit and the historical data acquisition unit respectively. It inputs the data acquired by the real-time data acquisition unit into the decision tree and outputs the fault type of the power generation terminal equipment during abnormal operation. An analysis unit, connected to the model output unit, is used to determine the operation and maintenance status based on the matching degree between the fault type output by the model output unit and the actual fault type within a preset time period. It also adjusts the maximum depth of the training constraint tree structure in the training decision tree based on the operation and maintenance status, and adjusts the number of training iterations of the decision tree when the number of adjustments equals the preset number and the operation and maintenance status is still unqualified.

2. The power generation terminal equipment operation and maintenance system based on a large model according to claim 1, characterized in that, The analysis unit is also used to plot a matching degree-preset time period curve based on the matching degree within multiple preset time periods. The above steps are performed when the operation and maintenance status is determined to be a first status. The first status is when the matching degree is less than a first preset matching degree and greater than a second preset matching degree. The analysis unit is also used to determine whether to adjust the maximum depth when training the constraint tree structure based on the average of the absolute values ​​of the slopes corresponding to several nodes of the curve. The analysis unit is also used to increase the maximum depth of the training constraint tree structure based on the ratio of the sample size obtained within a preset time period to the preset sample size, and the increase in the maximum depth of the training constraint tree structure is proportional to the ratio.

3. The power generation terminal equipment operation and maintenance system based on a large model according to claim 2, characterized in that, The analysis unit is also used to increase the maximum depth of the training constraint tree structure based on the minimum number of leaf node samples, and the increase in the maximum depth of the training constraint tree structure is proportional to the minimum number of leaf node samples.

4. The power generation terminal equipment operation and maintenance system based on a large model according to claim 3, characterized in that, The analysis unit is also used to repeatedly adjust the maximum depth of the training constraint tree structure at least once until the number of adjustments reaches a preset number or the number of adjustments is less than a preset number, at which point the operation and inspection status is qualified, and the adjustment is stopped. In this case, the average value of the absolute value of the slope of the curve of the adjusted matching degree - preset time period is obtained, and the adjustment operation is repeated when the average value is less than or equal to the preset average value. The standard for the operation and inspection status to be qualified is that the matching degree is greater than or equal to the first preset matching degree. The analysis unit is also used to obtain the optimal matching degree when the number of adjustments is equal to the preset number of adjustments, and to determine the adjusted operation and maintenance status based on the optimal matching degree; The analysis unit is also used to adjust the number of training iterations of the decision output based on the optimized matching degree, wherein the adjustment is performed when the adjusted operation and maintenance state is the first state.

5. The power generation terminal equipment operation and maintenance system based on a large model according to claim 4, characterized in that, The analysis unit is also used to increase the number of training iterations of the decision tree based on the ratio of the optimized matching degree to the first preset matching degree, and the increase in the number of training iterations of the decision tree is inversely proportional to the ratio.

6. The power generation terminal equipment operation and maintenance system based on a large model according to claim 5, characterized in that, The analysis unit is also used to reduce the initial learning rate based on the number of training iterations of the decision tree, and the reduction in the initial learning rate is proportional to the number of training iterations.

7. The power generation terminal equipment operation and maintenance system based on a large model according to claim 6, characterized in that, The analysis unit is also used to calculate the average value of the absolute value of the slope of the matching degree-preset time period curve plotted based on the matching degree re-detected after the above adjustment steps, so as to obtain the optimized average value. The analysis unit is also used to plot a rotor current-time curve based on the rotor current of the generator in each preset time period acquired by the real-time data acquisition unit. The above steps are performed when the operation and maintenance state is determined to be the second state or the optimized average value is less than or equal to the preset average value. The second state is when the matching degree is less than or equal to the second preset matching degree. The analysis unit is also used to reduce the sliding window length based on the variance of the absolute value of the slope of the rotor current-time curve, and the reduction in the sliding window length is proportional to the variance.

8. The power generation terminal equipment operation and maintenance system based on a large model according to claim 7, characterized in that, The analysis unit is also used to reduce the sliding step size based on the sliding window length, and the reduction in the sliding step size is inversely proportional to the sliding window length.

9. The power generation terminal equipment operation and maintenance system based on a large model according to claim 8, characterized in that, The analysis unit is also used to increase the range reference of the infrared sensor based on the length of the sliding window, and the increase in the range reference of the infrared sensor is inversely proportional to the length of the sliding window.

10. A method for operation and maintenance of power generation terminal equipment based on a large model, characterized in that, include: Through a sliding window mechanism, infrared thermal images and vibration phase signals generated by several sensor monitoring points set on the power generation terminal equipment are acquired in real time and synchronously, and electrical operating parameters are collected. All real-time data are time-aligned according to the set sliding window length. The historical operating parameters and fault types of the power generation terminal equipment during the historical operation process are obtained. The historical operating parameters include infrared thermal images, vibration phase signals and electrical operating parameters. The decision tree is constructed by training based on the data obtained from the historical operating parameters and historical fault types. The training parameters to be set when constructing the decision tree include the maximum depth when training the constraint tree structure, the minimum number of samples in the leaf nodes, the number of training iterations of the decision tree, and the initial learning rate. The acquired real-time data is input into the decision tree, and the fault type of the power generation terminal equipment during abnormal operation is output. The operation and maintenance status is determined based on the matching degree between the fault type output within a preset time period and the actual fault type. The maximum depth of the training constraint tree structure in the training decision tree is adjusted based on the operation and maintenance status. The number of training iterations of the decision tree is adjusted when the number of adjustments is equal to the preset number and the operation and maintenance status is still unqualified.

Images