Power plant auxiliary machine vibration fault diagnosis method and system based on deep learning

Abstract

The invention relates to the technical field of power plant auxiliary equipment, in particular to a power plant auxiliary equipment vibration fault diagnosis method and system based on deep learning. Comprises: establishing a fault database of power plant auxiliary equipment; setting a fault diagnosis model and a plurality of equipment points based on the fault database, and acquiring a monitoring data packet of each monitoring point according to a preset acquisition time node; judging whether an early warning instruction is generated or not according to all the monitoring data packets and the fault diagnosis model; the method comprises the following steps of: performing structural analysis on the power plant auxiliary equipment, setting a plurality of equipment points, constructing an abnormal signal set of each equipment point by utilizing a deep learning technology, setting a corresponding processing sub-model and a global diagnosis model according to the abnormal signal set, and analyzing a time sequence association relationship among characteristic sub-signals in a single abnormal signal set to obtain a diagnosis result of the abnormal signal set. And the corresponding processing sub-model is constructed, so that the processing efficiency of the monitoring data of each equipment point and the screening efficiency of the fault signal are improved, and the fault diagnosis efficiency of the power plant auxiliary equipment is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power plant auxiliary machinery, in particular to a power plant auxiliary machinery vibration fault diagnosis method and system based on deep learning. BACKGROUND

[0002] Power plant auxiliary machinery, such as fans, water pumps, coal mills, etc., is the key equipment for the safe and stable operation of thermal power generating units. It is prone to various mechanical vibration faults under long-term high-speed and heavy-load complex working conditions. At present, the mainstream fault diagnosis method mainly relies on traditional vibration signal analysis techniques, such as frequency spectrum analysis, envelope demodulation, etc., and combines expert experience for judgment.

[0003] However, such methods have obvious limitations: first, it highly depends on the personal knowledge and experience of the diagnosis expert, and the accuracy and consistency of the diagnosis results are difficult to guarantee. Second, the traditional method is not sensitive to early and weak fault characteristics in the signal, and it is difficult to achieve early warning of faults. Third, in the face of the characteristics of variable operation conditions of auxiliary machinery and strong non-stationary characteristics of vibration signals, the traditional diagnosis model based on fixed threshold has poor adaptability and high false alarm and missed alarm rates. Therefore, the existing technology cannot meet the urgent needs of modern smart power plants for real-time, accurate and intelligent diagnosis and early warning of auxiliary machinery status. SUMMARY

[0004] The purpose of the present application is to solve the above technical problems, and the present application provides a power plant auxiliary machinery vibration fault diagnosis method and system based on deep learning, aiming to improve the fault diagnosis efficiency of power plant auxiliary machinery and improve the operation and maintenance efficiency of power plant auxiliary machinery to ensure the safe operation of power plants.

[0005] In some embodiments of the present application, by analyzing the structure of the power plant auxiliary machinery, multiple equipment points are set, and deep learning technology is used to construct abnormal signal sets of each equipment point. According to the abnormal signal set, a corresponding processing sub-model and a global diagnosis model are set to improve the recognition efficiency of potential fault risks of the power plant auxiliary machinery and the overall operation and maintenance efficiency to ensure the safe operation of the power plant.

[0006] In some embodiments of the present application, by analyzing the time sequence correlation between each feature sub-signal in a single abnormal signal set, a corresponding processing sub-model is constructed to improve the processing efficiency of the monitoring data of each equipment point and the screening efficiency of the fault signal, thereby improving the fault diagnosis efficiency of the power plant auxiliary machinery.

[0007] In some embodiments of the present application, a power plant auxiliary machinery vibration fault diagnosis method based on deep learning is provided, which comprises:

[0008] establishing a fault database of the power plant auxiliary machinery;

[0009] Based on the fault database, a fault diagnosis model and multiple equipment points are set up, and monitoring data packets of each monitoring point are obtained according to the preset collection time nodes.

[0010] Determine whether to generate an early warning instruction based on all monitoring data packets and the fault diagnosis model;

[0011] The fault diagnosis model includes a diagnosis model and a data processing model.

[0012] In some embodiments of this application, a fault diagnosis model is defined, including:

[0013] Establish a sequence of device points A, A = (a1, a2, ..., a3) i …a n ), where a i Let i be the i-th device point; n is the number of device points;

[0014] Based on the equipment point sequence A, set a sequentially. i For the target device point;

[0015] Generate training data packets for target device points based on the fault database;

[0016] An abnormal signal set for the target device is generated based on the training data package, and the abnormal signal set includes multiple feature sub-signals.

[0017] The processing sub-model for the target device point is set based on the abnormal signal set;

[0018] Generate the processing sub-models for each device point in sequence;

[0019] Generate a data processing model based on all processing sub-models;

[0020] Obtain the abnormal signal set of each device point, and construct a diagnostic model based on the complete abnormal signal set;

[0021] The fault diagnosis model is set based on the diagnostic model and the processing sub-model.

[0022] In some embodiments of this application, a processing sub-model for the target device point is defined, including:

[0023] Generate a feature sub-signal sequence B based on the abnormal signal set of the target device point;

[0024] B = (b1, b2, ..., bb) i …b m ), where b i Let be the i-th feature sub-signal of the target device point; m is the number of feature sub-signals of the target device point;

[0025] b is set sequentially according to the characteristic sub-signal sequence B. i For target feature signals;

[0026] Generate the temporal correlation values ​​between the target feature signal and each feature sub-signal;

[0027] Construct an attention substructure for the target feature signal based on all temporal correlation values;

[0028] The attention substructures of each feature sub-signal are generated sequentially;

[0029] A processing sub-model for the target device point is generated based on the abnormal signal set and the full attention substructure.

[0030] In some embodiments of this application, the construction of the attention substructure of the target feature signal includes:

[0031] b is set sequentially according to the characteristic sub-signal sequence B. i Signals to be correlated;

[0032] Generate the temporal correlation value c of the signal to be correlated with the target feature signal;

[0033]

[0034] Where θ1 is the number of related indicators; β i Let j be the influence factor of the i-th correlation indicator; i It is a reference value for generating the i-th correlation index based on the target feature signal and the signal to be correlated;

[0035] Preset time-series correlation threshold C1;

[0036] If c > C1, the signal to be correlated is set as the time-series correlation signal of the target feature signal;

[0037] Obtain all temporal correlation signals of the target feature signal and generate the attention substructure of the target feature signal.

[0038] In some embodiments of this application, determining whether to generate an early warning instruction based on all monitoring data packets and the fault diagnosis model includes:

[0039] Based on the equipment point sequence A, let a be sequentially set. i For the device to be diagnosed;

[0040] Obtain the monitoring data packet of the device to be diagnosed at the current acquisition time node;

[0041] A primary processing model is generated based on the equipment to be diagnosed and the fault diagnosis model.

[0042] Multiple runtime periods are established based on the primary processing model and monitoring data packets;

[0043] Establish a sequence of time periods T, where T = (t1, t2, ..., tt) i…t r ), where t i Let r be the i-th running time period; r is the number of running time periods.

[0044] Set t1 as the first target time period;

[0045] Extract all first-level feature signals for the first target time period based on the first-level processing model;

[0046] Generate the first signal sub-packet for the first target time period;

[0047] Set t2 as the second target time period;

[0048] The extraction sub-strategy for the second target time period is set based on the primary processing model and the first signal sub-packet;

[0049] Generate a second signal sub-packet for the second target time period based on the sub-strategy;

[0050] Repeated iterations are performed to generate signal sub-packets for each runtime period;

[0051] The risk assessment value d of the device to be diagnosed is generated based on all signal sub-packets;

[0052] The risk assessment values ​​for each equipment point are generated sequentially.

[0053] Whether to generate an early warning instruction is determined based on all risk assessment values.

[0054] In some embodiments of this application, the extraction sub-strategy for the second target time period is set, including:

[0055] Establish the characteristic sub-signal sequence B1 of the device to be diagnosed;

[0056] B1=(b 11 b 12 …b 1i …b 1m1 ), where b 1i Let be the i-th feature sub-signal of the device to be diagnosed; m1 is the number of feature sub-signals of the device to be diagnosed.

[0057] Based on the feature sub-signal sequence B1, bi is sequentially set as the sub-signal to be evaluated;

[0058] Generate the matching value k between the sub-signal to be evaluated and the first signal sub-packet;

[0059] Preset matching threshold K1;

[0060] If k > K1, the sub-signal to be evaluated is set as the risk sub-signal for the second target time period;

[0061] The extraction sub-strategy is set based on all risk sub-signals in the second target time period.

[0062] In some embodiments of this application, generating the risk assessment value d of the device to be diagnosed includes:

[0063]

[0064] Where m1 is the number of feature sub-signals of the device to be diagnosed; μ i s is the influence factor of the i-th characteristic sub-signal in the device to be diagnosed; i It is an auxiliary risk value generated from the i-th characteristic sub-signal of the device point to be diagnosed based on all signal sub-packets.

[0065] In some embodiments of this application, determining whether to generate an early warning instruction based on all risk assessment values ​​includes:

[0066] Generate the early warning evaluation value f for the current data collection time point;

[0067]

[0068]

[0069] Where e is the risk compensation coefficient; n is the number of equipment points; η i d is the influence factor of the i-th equipment point; i Let be the risk assessment value of the i-th device at the current data collection time; U1 is the preset first conversion coefficient; d' is the preset risk assessment value threshold; Y(i) is the selection coefficient; if (d i -d′)>0, Y(i)=1; if (d i -d') < 0, Y(i) = 0;

[0070] Preset warning evaluation threshold F1;

[0071] If f > F1, a Level 1 early warning instruction is generated at the current data collection time point.

[0072] In some embodiments of this application, a deep learning-based power plant auxiliary equipment vibration fault diagnosis system is provided, comprising:

[0073] The processing unit is used to establish a fault database for power plant auxiliary equipment;

[0074] The processing unit is also used to set a fault diagnosis model and multiple equipment points based on the fault database;

[0075] The acquisition unit includes multiple acquisition sub-modules, which are located at various device points;

[0076] The acquisition unit is used to acquire monitoring data packets for each monitoring point according to a preset acquisition time node;

[0077] The processing unit includes:

[0078] The first processing module is used to set up the fault diagnosis model;

[0079] The fault diagnosis model includes: a diagnosis model and a data processing model;

[0080] The second processing module is used to determine whether to generate an early warning command based on all monitoring data packets and the fault diagnosis model.

[0081] In some embodiments of this application, the first processing module is further configured to:

[0082] Establish a sequence of device points A, A = (a1, a2, ..., a3) i …a n ), where a i Let i be the i-th device point; n is the number of device points;

[0083] Based on the equipment point sequence A, set a sequentially. i For the target device point;

[0084] Generate training data packets for target device points based on the fault database;

[0085] An abnormal signal set for the target device is generated based on the training data package, and the abnormal signal set includes multiple feature sub-signals.

[0086] The processing sub-model for the target device point is set based on the abnormal signal set;

[0087] Generate the processing sub-models for each device point in sequence;

[0088] Generate a data processing model based on all processing sub-models;

[0089] Obtain the abnormal signal set of each device point, and construct a diagnostic model based on the complete abnormal signal set;

[0090] The fault diagnosis model is set based on the diagnostic model and the processing sub-model.

[0091] Compared with existing technologies, the beneficial effects of the deep learning-based method and system for diagnosing vibration faults in power plant auxiliary equipment described in this application are as follows:

[0092] By performing structural analysis on the auxiliary equipment of the power plant, setting multiple equipment points, and using deep learning technology to construct abnormal signal sets for each equipment point, and setting corresponding processing sub-models and global diagnostic models based on the abnormal signal sets, the efficiency of identifying potential fault risks of the auxiliary equipment of the power plant and the overall operation and maintenance efficiency are improved, ensuring the safe operation of the power plant.

[0093] By analyzing the temporal correlation between various characteristic sub-signals in a single abnormal signal set, a corresponding processing sub-model is constructed to improve the processing efficiency of monitoring data at various equipment points and the screening efficiency of fault signals, thereby improving the fault diagnosis efficiency of power plant auxiliary equipment. Attached Figure Description

[0094] Figure 1 This is a flowchart illustrating a preferred embodiment of the present application of a deep learning-based method for diagnosing vibration faults in power plant auxiliary equipment. Detailed Implementation

[0095] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0096] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0097] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0098] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0099] like Figure 1 As shown in the preferred embodiment of this application, a method for diagnosing vibration faults in power plant auxiliary equipment based on deep learning includes:

[0100] S101: Establish a fault database for power plant auxiliary equipment;

[0101] S102: Based on the fault database, set up a fault diagnosis model and multiple equipment points, and obtain monitoring data packets for each monitoring point according to the preset acquisition time nodes;

[0102] S103: Determine whether to generate an early warning command based on all monitoring data packets and the fault diagnosis model;

[0103] The fault diagnosis model includes a diagnosis model and a data processing model.

[0104] Specifically, the equipment structure of power plant auxiliary equipment (such as fans, water pumps, coal mills, circulating water pumps, etc.) is analyzed, and several key parts (such as bearing housings, non-drive ends of motors, etc., which can generate vibration signals and reflect potential faults in the power plant) are screened out. Based on all key parts, multiple equipment points are set, where each equipment point represents a key part.

[0105] Specifically, the fault database includes historical fault parameters of power plant auxiliary equipment (location of occurrence, fault type, and historical vibration signals of each equipment point, etc.).

[0106] Specifically, each equipment point is equipped with a data acquisition submodule, which is preferably a vibration sensor. The data acquisition submodule is used to collect the operating vibration signals of each equipment point in real time.

[0107] Specifically, the fault diagnosis model is set up, including:

[0108] Establish a sequence of device points A, A = (a1, a2, ..., a3) i …a n ), where a i Let i be the i-th device point; n is the number of device points;

[0109] Based on the equipment point sequence A, set a sequentially. i For the target device point;

[0110] Generate training data packets for target device points based on the fault database;

[0111] An abnormal signal set for the target device is generated based on the training data package. The abnormal signal set includes multiple feature sub-signals.

[0112] The processing sub-model for the target device point is set based on the abnormal signal set;

[0113] Generate the processing sub-models for each device point in sequence;

[0114] Generate a data processing model based on all processing sub-models;

[0115] Obtain the abnormal signal set of each device point, and construct a diagnostic model based on the complete abnormal signal set;

[0116] The fault diagnosis model is set based on the diagnostic model and the processing sub-model.

[0117] Specifically, by analyzing relevant data in the fault database, all vibration signals collected at the target equipment point are selected to generate corresponding training data packages.

[0118] Specifically, deep learning technology is used to filter and analyze all data in the training data package, extracting all feature sub-signals related to the fault, and generating a corresponding abnormal signal set based on the extraction results. This also involves generating an analysis and extraction process for each feature sub-signal.

[0119] Specifically, a diagnostic model is constructed by analyzing all characteristic sub-signals in each abnormal signal set. The diagnostic model can analyze all collected vibration signals and provide timely warnings of potential fault risks.

[0120] It is understood that in the above embodiments, by performing structural analysis on the auxiliary equipment of the power plant, setting multiple equipment points, and using deep learning technology to construct abnormal signal sets for each equipment point, and setting corresponding processing sub-models and global diagnostic models based on the abnormal signal sets, the efficiency of identifying potential fault risks of the auxiliary equipment of the power plant and the overall operation and maintenance efficiency are improved, ensuring the safe operation of the power plant.

[0121] In a preferred embodiment of this application, the processing sub-model for setting the target device point includes:

[0122] Generate a feature sub-signal sequence B based on the abnormal signal set of the target device point;

[0123] B = (b1, b2, ..., bb) i …b m ), where b i Let be the i-th feature sub-signal of the target device point; m is the number of feature sub-signals of the target device point;

[0124] b is set sequentially according to the characteristic sub-signal sequence B. i For target feature signals;

[0125] Generate the temporal correlation values ​​between the target feature signal and each feature sub-signal;

[0126] Construct an attention substructure for the target feature signal based on all temporal correlation values;

[0127] The attention substructures of each feature sub-signal are generated sequentially;

[0128] A processing sub-model for the target device point is generated based on the abnormal signal set and the full attention substructure.

[0129] Specifically, constructing the attention substructure of the target feature signal includes:

[0130] b is set sequentially according to the characteristic sub-signal sequence B. i Signals to be correlated;

[0131] Generate the temporal correlation value c of the signal to be correlated with the target feature signal;

[0132]

[0133] Where θ1 is the number of related indicators; β i Let j be the influence factor of the i-th correlation indicator; i It is a reference value for generating the i-th correlation index based on the target feature signal and the signal to be correlated;

[0134] Preset time-series correlation threshold C1;

[0135] If c > C1, the signal to be correlated is set as the time-series correlation signal of the target feature signal;

[0136] Obtain all temporal correlation signals of the target feature signal and generate the attention substructure of the target feature signal.

[0137] Specifically, the correlation indicators include, but are not limited to, the probability of the target feature signal and the feature signal to be correlated occurring simultaneously, the probability of the target feature signal occurring after the target feature signal, and the similarity of the fault types corresponding to the two signals, etc., which are multiple parameters that represent the temporal correlation between the two signals. By quantifying each correlation indicator, the reference values ​​of each correlation indicator are made to be within the same range.

[0138] Specifically, the higher the reference value of each correlation indicator, the stronger the correlation between the two signals (i.e., the higher the probability that the signal to be correlated will exist after the target characteristic signal appears). The influence factor of each correlation indicator can be set according to its degree of reflection of the correlation between the two; the greater the degree of reflection, the larger the corresponding influence factor.

[0139] Specifically, the time-series correlation threshold can be set based on historical parameters. If the time-series correlation value between the target feature signal and the feature signal to be correlated is greater than the preset time-series correlation threshold, it means that if the target feature signal is extracted in the current time interval, the feature signal to be correlated needs to be prioritized in the analysis of the monitoring data collected in the next time interval.

[0140] Specifically, by constructing an attention substructure for target feature signals, the specific feature signal analysis range can be defined in a timely manner after the target feature signals are collected, thereby improving the efficiency of monitoring data processing and fault signal screening.

[0141] Specifically, the temporal correlation value is a one-way relationship. For example, the temporal correlation value of the first feature signal to the second feature signal is c1, and the temporal correlation value of the second feature signal to the first feature signal is c2. c1 and c2 are not completely equal; they may be the same or different.

[0142] In a preferred embodiment of this application, determining whether to generate an early warning instruction based on all monitoring data packets and the fault diagnosis model includes:

[0143] Based on the equipment point sequence A, let a be sequentially set. i For the device to be diagnosed;

[0144] Obtain the monitoring data packet of the device to be diagnosed at the current acquisition time node;

[0145] A primary processing model is generated based on the equipment to be diagnosed and the fault diagnosis model.

[0146] Multiple runtime periods are established based on the primary processing model and monitoring data packets;

[0147] Establish a sequence of time periods T, where T = (t1, t2, ..., tt) i …t r ), where t i Let r be the i-th running time period; r is the number of running time periods.

[0148] Set t1 as the first target time period;

[0149] Extract all first-level feature signals for the first target time period based on the first-level processing model;

[0150] Generate the first signal sub-packet for the first target time period;

[0151] Set t2 as the second target time period;

[0152] The extraction sub-strategy for the second target time period is set based on the primary processing model and the first signal sub-packet;

[0153] Generate a second signal sub-packet for the second target time period based on the sub-strategy;

[0154] Repeated iterations are performed to generate signal sub-packets for each runtime period;

[0155] The risk assessment value d of the device to be diagnosed is generated based on all signal sub-packets;

[0156] The risk assessment values ​​for each equipment point are generated sequentially.

[0157] Whether to generate an early warning instruction is determined based on all risk assessment values.

[0158] Specifically, based on all the characteristic sub-signals in the abnormal signal library of the device to be diagnosed, the monitoring data within the first target time period is comprehensively analyzed, the characteristic sub-signals are extracted, and each extracted characteristic sub-signal is set as a first-level characteristic signal.

[0159] Specifically, based on all time-series correlated signals included in the attention substructure of each primary feature signal, the analysis scope of the monitoring data in the second target time period is defined (i.e., up to all time-series correlated signals that need to be filtered and extracted). This process is repeated to complete the analysis of monitoring data within each operating time period.

[0160] Specifically, the extraction sub-strategy for the second target time period is set, including:

[0161] Establish the characteristic sub-signal sequence B1 of the device to be diagnosed;

[0162] B1=(b 11 b 12 …b 1i …b 1m1 ), where b 1i Let be the i-th feature sub-signal of the device to be diagnosed; m1 is the number of feature sub-signals of the device to be diagnosed.

[0163] Based on the feature sub-signal sequence B1, bi is sequentially set as the sub-signal to be evaluated;

[0164] Generate the matching value k between the sub-signal to be evaluated and the first signal sub-packet;

[0165] Preset matching threshold K1;

[0166] If k > K1, the sub-signal to be evaluated is set as the risk sub-signal for the second target time period;

[0167] The extraction sub-strategy is set based on all risk sub-signals in the second target time period.

[0168] Specifically, if the current feature sub-signal b1i is a time-series correlated signal of any first-level feature signal in the first signal sub-packet, then its matching value is set to 1; otherwise, the matching value is set to 0. The matching value threshold can be set according to historical parameters, and the value range of the matching value threshold is (0, 1). In this application, the matching value threshold is preferably 0.5.

[0169] Specifically, the sub-strategy extraction refers to the analysis and extraction process corresponding to each risk sub-signal in the monitoring data of the second target time period.

[0170] It is understood that in the above embodiments, by analyzing the temporal correlation between various feature sub-signals in a single abnormal signal set, a corresponding processing sub-model is constructed to improve the processing efficiency of monitoring data at various equipment points and the screening efficiency of fault signals, thereby improving the fault diagnosis efficiency of power plant auxiliary equipment.

[0171] In a preferred embodiment of this application, generating the risk assessment value d of the device to be diagnosed includes:

[0172]

[0173] Where m1 is the number of feature sub-signals of the device to be diagnosed; μ i s is the influence factor of the i-th characteristic sub-signal in the device to be diagnosed; i It is an auxiliary risk value generated from the i-th characteristic sub-signal of the device point to be diagnosed based on all signal sub-packets.

[0174] Specifically, the auxiliary risk value of each feature sub-signal is set according to its frequency of occurrence in each signal sub-packet and its degree of mapping to the operational fault. The higher the frequency of occurrence, the higher the degree of mapping to the operational fault, and the greater the corresponding auxiliary risk value. If the current feature sub-signal does not appear in any signal sub-packet, the auxiliary risk value of the feature sub-signal is 0.

[0175] Specifically, the influence factor of each characteristic sub-signal can be set according to its correlation with the operational fault. The greater the correlation, the larger the reference value of the corresponding influence factor.

[0176] Specifically, the decision to generate an early warning instruction is based on all risk assessment values, including:

[0177] Generate the early warning evaluation value f for the current data collection time point;

[0178]

[0179] Where e is the risk compensation coefficient; n is the number of equipment points; η i d is the influence factor of the i-th equipment point; i Let be the risk assessment value of the i-th device at the current data collection time; U1 is the preset first conversion coefficient; d' is the preset risk assessment value threshold; Y(i) is the selection coefficient; if (d i -d′)>0, Y(i)=1; if (d i -d') < 0, Y(i) = 0;

[0180] Preset warning evaluation threshold F1;

[0181] If f > F1, a Level 1 early warning instruction is generated at the current data collection time point.

[0182] Specifically, the warning evaluation threshold can be set based on historical parameters. When the real-time warning evaluation value is greater than the preset warning evaluation threshold, it indicates that there is a potential operational fault in the auxiliary equipment of the power plant, and timely maintenance is required to ensure the safe operation of the power plant as a whole.

[0183] Specifically, the risk assessment threshold can be set based on historical parameters. If the risk assessment value of the current equipment point is greater than the preset risk assessment threshold, it indicates that the current equipment point is in an abnormal operating state.

[0184] Specifically, by presetting a first conversion coefficient, the risk compensation coefficient e is kept within a preset value range, and The larger the value of , the larger the corresponding risk compensation coefficient e. The mapping relationship between the two can be set according to historical parameters.

[0185] Specifically, the impact factor for each equipment point can be set based on the historical failure frequency of its corresponding equipment structure. The higher the historical failure frequency, the larger the reference value of the corresponding impact factor. The mapping relationship between the two can be set based on historical parameters.

[0186] In another preferred embodiment of the deep learning-based vibration fault diagnosis method for power plant auxiliary equipment based on any of the above preferred embodiments, this preferred embodiment provides a deep learning-based vibration fault diagnosis system for power plant auxiliary equipment, comprising:

[0187] The processing unit is used to establish a fault database for power plant auxiliary equipment;

[0188] The processing unit is also used to set up a fault diagnosis model and multiple equipment points based on the fault database;

[0189] The data acquisition unit includes multiple data acquisition sub-modules, which are located at various device points.

[0190] The acquisition unit is used to acquire monitoring data packets for each monitoring point according to preset acquisition time nodes;

[0191] The processing unit includes:

[0192] The first processing module is used to set up the fault diagnosis model;

[0193] The fault diagnosis model includes: a diagnostic model and a data processing model;

[0194] The second processing module is used to determine whether to generate an early warning command based on all monitoring data packets and the fault diagnosis model.

[0195] In a preferred embodiment of this application, the first processing module is further configured to:

[0196] Establish a sequence of device points A, A = (a1, a2, ..., a3) i …a n ), where a i Let i be the i-th device point; n is the number of device points;

[0197] Based on the equipment point sequence A, set a sequentially. i For the target device point;

[0198] Generate training data packets for target device points based on the fault database;

[0199] An abnormal signal set for the target device is generated based on the training data package. The abnormal signal set includes multiple feature sub-signals.

[0200] The processing sub-model for the target device point is set based on the abnormal signal set;

[0201] Generate the processing sub-models for each device point in sequence;

[0202] Generate a data processing model based on all processing sub-models;

[0203] Obtain the abnormal signal set of each device point, and construct a diagnostic model based on the complete abnormal signal set;

[0204] The fault diagnosis model is set based on the diagnostic model and the processing sub-model.

[0205] Based on the first concept of this application, by performing structural analysis on the auxiliary equipment of the power plant, setting multiple equipment points, and using deep learning technology to construct abnormal signal sets for each equipment point, and setting corresponding processing sub-models and global diagnostic models based on the abnormal signal sets, the efficiency of identifying potential fault risks of the auxiliary equipment of the power plant and the overall operation and maintenance efficiency are improved, thereby ensuring the safe operation of the power plant.

[0206] According to the second concept of this application, by analyzing the temporal correlation between various characteristic sub-signals in a single abnormal signal set, a corresponding processing sub-model is constructed to improve the processing efficiency of monitoring data at various equipment points and the screening efficiency of fault signals, thereby improving the fault diagnosis efficiency of power plant auxiliary equipment.

[0207] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application.

Claims

1. A method for diagnosing vibration faults in power plant auxiliary equipment based on deep learning, characterized in that, include: Establish a fault database for power plant auxiliary equipment; Based on the fault database, a fault diagnosis model and multiple equipment points are set up, and monitoring data packets of each monitoring point are obtained according to the preset collection time nodes. Determine whether to generate an early warning instruction based on all monitoring data packets and the fault diagnosis model; The fault diagnosis model includes a diagnosis model and a data processing model.

2. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 1, characterized in that, Define the fault diagnosis model, including: Establish a sequence of device points A, A = (a1, a2, ..., a3) i …a n ), where a i Let i be the i-th device point; n is the number of device points; Based on the equipment point sequence A, set a sequentially. i For the target device point; Generate training data packets for target device points based on the fault database; An abnormal signal set for the target device is generated based on the training data package, and the abnormal signal set includes multiple feature sub-signals. The processing sub-model for the target device point is set based on the abnormal signal set; Generate the processing sub-models for each device point in sequence; Generate a data processing model based on all processing sub-models; Obtain the abnormal signal set of each device point, and construct a diagnostic model based on the complete abnormal signal set; The fault diagnosis model is set based on the diagnostic model and the processing sub-model.

3. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 2, characterized in that, Define the processing sub-model for the target device point, including: Generate a feature sub-signal sequence B based on the abnormal signal set of the target device point; B = (b1, b2, ..., bb) i …b m ), where b i Let be the i-th feature sub-signal of the target device point; m is the number of feature sub-signals of the target device point; b is set sequentially according to the characteristic sub-signal sequence B. i For target feature signals; Generate the temporal correlation values ​​between the target feature signal and each feature sub-signal; Construct an attention substructure for the target feature signal based on all temporal correlation values; The attention substructures of each feature sub-signal are generated sequentially; A processing sub-model for the target device point is generated based on the abnormal signal set and the full attention substructure.

4. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 3, characterized in that, Constructing the attention substructure of the target feature signal includes: b is set sequentially according to the characteristic sub-signal sequence B. i Signals to be correlated; Generate the temporal correlation value c of the signal to be correlated with the target feature signal; Where θ1 is the number of related indicators; β i Let j be the influence factor of the i-th correlation indicator; i It is a reference value for generating the i-th correlation index based on the target feature signal and the signal to be correlated; Preset time-series correlation threshold C1; If c > C1, the signal to be correlated is set as the time-series correlation signal of the target feature signal; Obtain all temporal correlation signals of the target feature signal and generate the attention substructure of the target feature signal.

5. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 4, characterized in that, Based on all monitoring data packets and the fault diagnosis model, determine whether to generate an early warning instruction, including: Based on the equipment point sequence A, let a be sequentially set. i For the device to be diagnosed; Obtain the monitoring data packet of the device to be diagnosed at the current acquisition time node; A primary processing model is generated based on the equipment to be diagnosed and the fault diagnosis model. Multiple runtime periods are established based on the primary processing model and monitoring data packets; Establish a sequence of time periods T, where T = (t1, t2, ..., tt) i …t r ), where t i Let r be the i-th running time period; r is the number of running time periods. Set t1 as the first target time period; Extract all first-level feature signals for the first target time period based on the first-level processing model; Generate the first signal sub-packet for the first target time period; Set t2 as the second target time period; The extraction sub-strategy for the second target time period is set based on the primary processing model and the first signal sub-packet; Generate a second signal sub-packet for the second target time period based on the sub-strategy; Repeated iterations are performed to generate signal sub-packets for each runtime period; The risk assessment value d of the device to be diagnosed is generated based on all signal sub-packets; The risk assessment values ​​for each equipment point are generated sequentially. Whether to generate an early warning instruction is determined based on all risk assessment values.

6. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 5, characterized in that, Define the extraction sub-strategy for the second target time period, including: Establish the characteristic sub-signal sequence B1 of the device to be diagnosed; B1=(b 11 b 12 …b 1i …b 1m1 ), where b 1i Let be the i-th feature sub-signal of the device to be diagnosed; m1 is the number of feature sub-signals of the device to be diagnosed. Based on the feature sub-signal sequence B1, bi is sequentially set as the sub-signal to be evaluated; Generate the matching value k between the sub-signal to be evaluated and the first signal sub-packet; Preset matching threshold K1; If k > K1, the sub-signal to be evaluated is set as the risk sub-signal for the second target time period; The extraction sub-strategy is set based on all risk sub-signals in the second target time period.

7. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 5, characterized in that, Generate the risk assessment value d for the device to be diagnosed, including: Where m1 is the number of feature sub-signals of the device point to be diagnosed; μ i s is the influence factor of the i-th characteristic sub-signal in the device to be diagnosed; i It is an auxiliary risk value generated from the i-th characteristic sub-signal of the device point to be diagnosed based on all signal sub-packets.

8. The deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in claim 7, characterized in that, Whether to generate an early warning instruction is determined based on all risk assessment values, including: Generate the early warning evaluation value f for the current data collection time point; Where e is the risk compensation coefficient; n is the number of equipment points; η i d is the influence factor of the i-th equipment point; i Let be the risk assessment value of the i-th device at the current data collection time; U1 is the preset first conversion coefficient; d' is the preset risk assessment value threshold; Y(i) is the selection coefficient; if (d i -d')>0, Y(i)=1; if (d i -d') < 0, Y(i) = 0; Preset warning evaluation threshold F1; If f > F1, a Level 1 early warning instruction is generated at the current data collection time point.

9. A deep learning-based vibration fault diagnosis system for power plant auxiliary equipment, employing the deep learning-based vibration fault diagnosis method for power plant auxiliary equipment as described in any one of claims 1-8, characterized in that, include: The processing unit is used to establish a fault database for power plant auxiliary equipment; The processing unit is also used to set a fault diagnosis model and multiple equipment points based on the fault database; The acquisition unit includes multiple acquisition sub-modules, which are located at various device points; The acquisition unit is used to acquire monitoring data packets for each monitoring point according to a preset acquisition time node; The processing unit includes: The first processing module is used to set up the fault diagnosis model; The fault diagnosis model includes: a diagnosis model and a data processing model; The second processing module is used to determine whether to generate an early warning command based on all monitoring data packets and the fault diagnosis model.

10. The deep learning-based power plant auxiliary equipment vibration fault diagnosis system as described in claim 9, characterized in that, The first processing module is also used for: Establish a sequence of device points A, A = (a1, a2, ..., a3) i …a n ), where a i Let i be the i-th device point; n is the number of device points; Based on the equipment point sequence A, set a sequentially. i For the target device point; Generate training data packets for target device points based on the fault database; An abnormal signal set for the target device is generated based on the training data package, and the abnormal signal set includes multiple feature sub-signals. The processing sub-model for the target device point is set based on the abnormal signal set; Generate the processing sub-models for each device point in sequence; Generate a data processing model based on all processing sub-models; Obtain the abnormal signal set of each device point, and construct a diagnostic model based on the complete abnormal signal set; The fault diagnosis model is set based on the diagnostic model and the processing sub-model.

Images