Power plant equipment fault diagnosis method, device and equipment

Abstract

The embodiment of the present invention provides a method, device and equipment for diagnosing faults of power station equipment, which relates to the field of fault diagnosis technology. The method is applied to electronic equipment, wherein a fault diagnosis model is integrated inside the electronic equipment, and the network architecture of the fault diagnosis model is a Transformer model. The fault diagnosis model includes a primary layer and a high-level layer. The method includes: using the fault diagnosis model as a source model and freezing the primary layer of the source model; acquiring equipment data of power station equipment in real time, and based on the equipment data acquired in real time, looping the first loop step until a preset number of executions is reached to obtain a fault diagnosis result. Through the present invention, the accuracy of fault diagnosis of the fault diagnosis model can be effectively improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of fault diagnosis, and in particular, to a method, device and equipment for diagnosing faults of power station equipment. Background Art

[0002] Power plant equipment fault diagnosis is of great significance to ensure the safe, reliable, economical and efficient operation of power plants. At present, fault diagnosis models are usually used to diagnose power plant equipment. However, as power plant equipment ages, the performance parameters of power plant equipment will change. Fault diagnosis models are usually trained based on a specific set of data sets. If the data set does not match the performance parameters of the aged power plant equipment, it may cause a shift in data distribution, making it impossible for the fault diagnosis model to accurately identify the fault.

[0003] There is currently no better solution to the above problems. Summary of the invention

[0004] The embodiments of the present invention provide a method, apparatus and device for diagnosing power plant equipment faults, so as to effectively improve the accuracy of fault diagnosis of a fault diagnosis model.

[0005] According to an embodiment of the present invention, a method for diagnosing a fault of a power plant device is provided, which is applied to an electronic device, wherein a fault diagnosis model is integrated inside the electronic device, wherein the network architecture of the fault diagnosis model is a Transformer model, and the fault diagnosis model includes a primary layer and a high-level layer, wherein the method includes:

[0006] Using the fault diagnosis model as a source model and freezing the primary layer of the source model;

[0007] Acquire equipment data of power station equipment in real time, and based on the equipment data acquired in real time, cyclically execute the first cycle step until a preset number of executions is reached to obtain a fault diagnosis result;

[0008] Wherein, the first cycle step comprises:

[0009] Performing data analysis on the device data acquired in real time to obtain analysis results;

[0010] In the case where the analysis result is inconsistent with the previous analysis result, determining the current target task according to the analysis result;

[0011] Adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer is adapted to the current target task;

[0012] Constructing a transfer learning network model, wherein the network architecture of the transfer learning network model is a hybrid architecture of an LSTM model and a Transformer model;

[0013] According to the frozen primary layer and the adjusted high-level layer, transfer learning is performed on the source model through the transfer learning network model to obtain a transfer learning network model after transfer learning;

[0014] The transfer learning network model after the transfer learning is used to perform fault diagnosis on the power station equipment to obtain a fault diagnosis result.

[0015] In an exemplary embodiment, the primary layer of the fault diagnosis model includes at least one Encoder layer and a self-attention sublayer of each Encoder layer;

[0016] The freezing the primary layer of the source model comprises:

[0017] During the process of constructing the source model, traversing the layers of the source model, and determining at least one of the encoder layers or the self-attention sublayer of each encoder layer as a frozen layer;

[0018] The weight of the frozen layer is frozen so that the weight of the frozen layer is not updated during the training process of the source model.

[0019] In an exemplary embodiment, the method of performing transfer learning on the source model through the transfer learning network model according to the frozen primary layer and the adjusted high-level layer to obtain the transfer learning network model after transfer learning includes:

[0020] Based on the primary layer of the previous source model, add an LSTM module, and connect the output of the previous source model to the LSTM module to construct an initial transfer learning network model;

[0021] Training the initial transfer learning network model, wherein during the training of the initial transfer learning network model, the LSTM module and the high-level layer are updated in real time, and the primary layer is not updated;

[0022] The trained initial transfer learning network model is used as the transfer learning network model after transfer learning.

[0023] In an exemplary embodiment, adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer adapts to the current target task includes:

[0024] Defining a loss function according to the current target task;

[0025] Dividing the device data acquired in real time into a plurality of task data sets;

[0026] Using each of the task data sets to train the previous source model in sequence, wherein in each process of training the previous source model, the gradient of the loss function is calculated;

[0027] Execute the second loop step repeatedly until a preset number of executions is reached, so that the high-level layer adapts to the current target task, wherein the number of executions is greater than the number of trainings of the previous source model;

[0028] The second cycle step comprises:

[0029] Update the parameters of the previous high-level layer of the source model according to the gradient of the loss function.

[0030] In an exemplary embodiment, updating the parameters of the previous high-level layer of the source model according to the gradient of the loss function includes:

[0031] Adopting a gradient descent formula, updating the parameters of the previous high-level layer of the source model according to the gradient of the loss function;

[0032] Wherein, the gradient descent formula includes:

[0033]

[0034] In the formula, θ new is the updated parameters of the previous high-level layer of the source model, θ old is the parameter of the previous high-level layer of the source model before the update, η is the learning rate, which is used to control the step size of the parameter update each time the second cycle step is executed, J(θ) is the loss function, is the gradient of the loss function with respect to the parameter θ.

[0035] In an exemplary embodiment, performing data analysis on the device data acquired in real time to obtain analysis results includes:

[0036] Extracting key features of the device data acquired in real time;

[0037] Acquiring historical data of the power station equipment;

[0038] Constructing a normal performance baseline of the power plant equipment according to the historical data, and determining current equipment performance indicators according to key features of the equipment data acquired in real time;

[0039] Comparing the current device performance indicator with the normal performance baseline to obtain b comparison results, where b is a positive integer;

[0040] According to the b comparison results, determining m normal indicators and n abnormal indicators of the power station equipment, wherein m+n=b, and m and n are both integers;

[0041] The m normal indicators and the n abnormal indicators are used as the analysis results.

[0042] In an exemplary embodiment, when the analysis result is inconsistent with a previous analysis result, determining the current target task according to the analysis result includes:

[0043] If at least one of the abnormal indicators in the analysis result is different from the abnormal indicator in the previous analysis result; and / or

[0044] If there is at least one abnormal indicator in the analysis result that does not exist in the previous analysis result, a target indicator is obtained, wherein the target indicator is at least one abnormal indicator in the analysis result that is different from the previous analysis result and / or at least one abnormal indicator in the analysis result that does not exist in the previous analysis result;

[0045] According to the target indicators, determine the current target tasks.

[0046] In an exemplary embodiment, determining the current target task according to the target indicator includes:

[0047] Classify the target indicator according to corresponding properties to obtain the type of the target indicator, and obtain the target device corresponding to the type of the target indicator;

[0048] In combination with the historical maintenance database, the current target task is determined according to the type of the target indicator and the target equipment.

[0049] According to another embodiment of the present invention, there is provided a power plant equipment fault diagnosis device, comprising:

[0050] A freezing module, used for taking the fault diagnosis model as a source model and freezing the primary layer of the source model;

[0051] A loop module, used for acquiring equipment data of power station equipment in real time, and based on the equipment data acquired in real time, cyclically executing the first loop step until a preset number of executions is reached to obtain a fault diagnosis result;

[0052] Wherein, the first cycle step comprises:

[0053] Performing data analysis on the device data acquired in real time to obtain analysis results;

[0054] In the case where the analysis result is inconsistent with the previous analysis result, determining the current target task according to the analysis result;

[0055] Adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer is adapted to the current target task;

[0056] Constructing a transfer learning network model, wherein the network architecture of the transfer learning network model is a hybrid architecture of an LSTM model and a Transformer model;

[0057] According to the frozen primary layer and the adjusted high-level layer, transfer learning is performed on the source model through the transfer learning network model to obtain a transfer learning network model after transfer learning;

[0058] The transfer learning network model after the transfer learning is used to perform fault diagnosis on the power station equipment to obtain a fault diagnosis result.

[0059] According to yet another embodiment of the present invention, there is further provided an electronic device, including:

[0060] The above-mentioned power station equipment fault diagnosis device

[0061] a memory configured to store instructions; and

[0062] The processor is configured to call the instructions from the memory and implement the above-mentioned power plant equipment fault diagnosis method when executing the instructions.

[0063] Through the present invention, the method of transfer learning is adopted, and the source model is transferred and learned by adjusting the high-level layer of the source model and combining the transfer learning network model of the hybrid architecture of the LSTM and Transformer models. The features that the source model has learned can be used to reduce the dependence on the new data set, thereby solving the problem of data distribution offset to a certain extent. In addition, by freezing the primary layer of the source model and adjusting the high-level layer, the source model is transferred and learned using the transfer learning network model, so that the model can adapt to the performance parameter changes of the aging equipment. Through the transfer learning network model after transfer learning, the power station equipment can be diagnosed to obtain more accurate fault diagnosis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a hardware structure block diagram of a mobile terminal according to an embodiment of the present invention;

[0065] Figure 2 is a flow chart of a method for diagnosing a power plant equipment fault according to an embodiment of the present invention;

[0066] Figure 3is a flow chart of a first loop step according to an embodiment of the present invention;

[0067] Figure 4 4 is a structural block diagram of a power plant equipment fault diagnosis device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0069] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0070] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a mobile terminal of a power plant equipment fault diagnosis method according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0071] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as a computer program corresponding to a method for diagnosing faults of power plant equipment in an embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the mobile terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0072] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0073] The power plant equipment fault diagnosis method provided by the embodiment of the present application is described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0074] The embodiment of the present application discloses a method for diagnosing power station equipment faults.

[0075] Reference Figure 2 A power station equipment fault diagnosis method is applied to an electronic device, wherein a fault diagnosis model is integrated inside the electronic device, and the network architecture of the fault diagnosis model is a Transformer model. The fault diagnosis model includes a primary layer and a high-level layer. The method includes the following steps:

[0076] S110: Use the fault diagnosis model as a source model and freeze the primary layer of the source model.

[0077] The primary layer is used to extract basic features of the data. In this embodiment, the primary layer is a feature extraction layer. Freezing the primary layer of the source model means that the parameters of the primary layer are kept unchanged in the subsequent transfer learning process.

[0078] Since the primary layer is used to capture common features, the primary layer can be frozen to effectively improve the training speed of the subsequent transfer learning network model. In addition, freezing the primary layer can effectively prevent the source model from overfitting on the new task. Overfitting refers to the phenomenon that the source model performs well on the training data but performs poorly on the test data. By freezing the primary layer, the basic features of the source model will not be adjusted, which can reduce the risk of overfitting.

[0079] S120, acquiring equipment data of the power station equipment in real time, and based on the equipment data acquired in real time, cyclically executing the first loop step until a preset number of executions is reached to obtain a fault diagnosis result.

[0080] By cyclically executing the first cycle step, multiple data analyses and model adjustments can be performed to obtain more accurate fault diagnosis results. The power station equipment includes transformers, generators, and the like.

[0081] Reference Figure 3 , the first cycle steps include:

[0082] S130: Analyze the device data acquired in real time to obtain analysis results.

[0083] The analysis results are information about the device status and performance obtained based on the device data.

[0084] S140: When the analysis result is inconsistent with the previous analysis result, determine the current target task according to the analysis result.

[0085] In the case where the analysis result is inconsistent with the previous analysis result, the current target task is determined based on the analysis result, and it can be determined whether the current equipment has a fault and determine the further diagnostic task that needs to be performed, that is, the current target task. If the analysis result is inconsistent with the previous result, it means that the power plant equipment may have a fault or other abnormal conditions have occurred.

[0086] S150. Adjust the high-level layer of the previous source model according to the current target task to make the high-level layer adapt to the current target task.

[0087] The high-level layer is responsible for higher-level feature extraction and pattern recognition than the primary layer. This embodiment can adjust the high-level layer of the source model according to the current target task to better adapt to the current fault diagnosis requirements, that is, the current target task.

[0088] S160. Construct a transfer learning network model, wherein the network architecture of the transfer learning network model is a hybrid architecture of an LSTM model and a Transformer model.

[0089] The transfer learning network model of this embodiment combines the network architecture of the LSTM (Long Short-Term Memory) model and the Transformer model. Among them, the Transformer model is based on the self-attention mechanism, which can better capture the correlation in the sequence data, and each layer of the Transformer model can access the information of all positions of the entire input sequence. The LSTM model is a recurrent neural network suitable for sequence data processing, which is used to process sequence data.

[0090] S170. Perform transfer learning on the source model through a transfer learning network model according to the frozen primary layer and the adjusted advanced layer, to obtain a transfer learning network model after transfer learning.

[0091] The frozen primary layer can be the encoder layer of the Transformer model or the self-attention sublayer in the encoder layer, where each encoder layer contains a self-attention mechanism and a feed-forward neural network to extract basic contextual information and features of the input data.

[0092] This embodiment can use the transfer learning method to transfer the knowledge and features of the source model to the transfer learning network model. By freezing the primary layer and adjusting the advanced layer, the transfer learning network model can adapt to the performance parameter changes of the aging power station equipment.

[0093] S180, using the transfer learning network model after transfer learning to perform fault diagnosis on the power station equipment to obtain a fault diagnosis result.

[0094] The power station equipment is diagnosed for faults based on the transfer learning network model after transfer learning. The final fault diagnosis result can be obtained by analyzing the equipment data and predicting the model.

[0095] It should be noted that the transfer learning network model after transfer learning can be updated immediately when new data is received in the future. In order to facilitate the management of the transfer learning network model by the management personnel, the performance data of the transfer learning network model can be collected in real time, so as to facilitate the subsequent optimization of the transfer learning network model according to the performance data.

[0096] This embodiment adopts the method of transfer learning. By adjusting the high-level layer of the source model and combining the transfer learning network model with the hybrid architecture of the LSTM and Transformer models, the source model can be transferred and learned. The features that the source model has learned can be used to reduce the dependence on new data sets, thereby solving the problem of data distribution offset to a certain extent. In addition, by freezing the primary layer of the source model and adjusting the high-level layer, the source model is transferred and learned using the transfer learning network model, so that the model can adapt to the changes in the performance parameters of aging equipment. Through the transfer learning network model after transfer learning, fault diagnosis of power station equipment can be performed to obtain more accurate fault diagnosis results.

[0097] In one implementation of this embodiment, the primary layer of the fault diagnosis model includes at least one Encoder layer and a self-attention sublayer of each Encoder layer;

[0098] Freeze the primary layer of the source model, including the following steps:

[0099] S210. During the process of constructing the source model, traverse the layers of the source model and determine at least one Encoder layer or the self-attention sublayer of each Encoder layer as a frozen layer.

[0100] In this embodiment, during the process of constructing the source model, the primary layer can be determined to be in a non-trainable state, and the weight of the primary layer can remain unchanged during subsequent model training. The Encoder layer is the encoder layer of the Transformer model.

[0101] S220, freezing the weight of the frozen layer so that the weight of the frozen layer is not updated during the training process of the source model.

[0102] In this embodiment, the trainable property of the primary layer is set to False, so that the weights of the frozen layer are not updated during the training process of the source model.

[0103] This implementation method can retain the basic feature extraction capability learned by the source model on the original data by freezing the layer, which can effectively reduce the amount of calculation and improve the efficiency of the transfer learning process.

[0104] In one implementation of this embodiment, transfer learning is performed on the source model through a transfer learning network model according to the frozen primary layer and the adjusted high-level layer to obtain a transfer learning network model after transfer learning, including the following steps:

[0105] S310. Based on the primary layer of the previous source model, add an LSTM module, and connect the output of the previous source model to the LSTM module to construct an initial transfer learning network model.

[0106] S320, training an initial transfer learning network model, wherein, during the training of the initial transfer learning network model, the LSTM module and the advanced layer are updated in real time, and the primary layer is not updated.

[0107] S330, using the trained initial transfer learning network model as the transfer learning network model after transfer learning.

[0108] Since the first cycle step is executed cyclically, the transfer learning network model of the previous first cycle step is used as the previous source model.

[0109] On the basis of the previous source model, a new LSTM module is added to enhance the model's sequence processing capability. The output of the previous source model is connected to the newly added LSTM module to form a new network structure model, namely the initial transfer learning network model.

[0110] During the initial transfer learning network model training process, the LSTM modules and high-level layers are updated in real time, while the primary layers are not updated.

[0111] This implementation method realizes the diagnosis of power station equipment faults by adding LSTM modules, training the initial transfer learning network model and performing transfer learning, which can effectively improve the accuracy of fault diagnosis.

[0112] In one implementation of this embodiment, adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer adapts to the current target task includes the following steps:

[0113] S410, defining a loss function according to the current target task.

[0114] The loss function is used to measure the gap between the model's predicted results and the actual results, thereby guiding the model to optimize.

[0115] This embodiment may involve a loss function for measuring the performance of the transfer learning network model on the current target task according to the current target task.

[0116] S420: Divide the device data acquired in real time into a plurality of task data sets.

[0117] Specifically, the device data acquired in real time may be aggregated, and the aggregated device data may be divided into a plurality of task data sets.

[0118] S430, using each task data set to train the previous source model in sequence, wherein in each process of training the previous source model, the gradient of the loss function is calculated.

[0119] For each task data set, the transfer learning network model can be trained once according to the data set and the gradient of the loss function can be calculated, which is conducive to the continuous optimization and learning of the transfer learning network model on various tasks.

[0120] Specifically, the gradient of the loss function refers to the rate of change of the loss function relative to the model parameters. In machine learning, the gradient indicates the sensitivity of the loss function to the model parameters.

[0121] S440, looping the second loop step until a preset number of executions is reached to adapt the high-level layer to the current target task, wherein the number of executions is greater than the number of trainings of the previous source model.

[0122] By repeatedly executing the second loop step, the performance of the transfer learning network model on the current target task can be further improved, ensuring that the transfer learning network model is fully learned and optimized on multiple tasks.

[0123] The second cycle steps include:

[0124] S450. Update the parameters of the high-level layer of the previous source model according to the gradient of the loss function.

[0125] According to the gradient of the loss function, the high-level layer parameters of the source model are updated. By continuously using the gradient to optimize the model parameters, the transfer learning network model can gradually improve its performance to adapt to the current target task.

[0126] This implementation method adjusts and optimizes the high-level layers of the model to adapt it to the current target task, so as to maximize the performance and adaptability of the transfer learning network model on the current target task.

[0127] In one implementation of this embodiment, updating the parameters of the high-level layer of the previous source model according to the gradient of the loss function includes the following steps:

[0128] S510, using the gradient descent formula, updating the parameters of the high-level layer of the previous source model according to the gradient of the loss function.

[0129] Among them, the gradient descent formula includes:

[0130]

[0131] In the formula, θ new is the updated parameters of the high-level layer of the previous source model, θ old is the parameter of the high-level layer of the previous source model before updating, η is the learning rate, which is used to control the step size of parameter update each time the second cycle step is executed, J(θ) is the loss function, ▽ θ J(θ) is the gradient of the loss function with respect to the parameter θ.

[0132] Specifically, calculating the gradient of the loss function with respect to the parameter θ can be achieved by solving the partial derivative of the loss function with respect to each parameter.

[0133] Gradient descent is used to update model parameters to minimize the loss function. Its basic principle is to update the parameters in the direction where the loss function decreases fastest. Through continuous iterative updates, the gradient descent algorithm can gradually make the model approach the minimum value of the loss function.

[0134] This implementation adopts the gradient descent formula to update the high-level layer parameters of the previous source model, which can move towards the minimum value of the loss function. In this way, after each parameter update, the value of the loss function will decrease, thereby gradually improving the prediction performance of the model and improving the accuracy of fault diagnosis.

[0135] In one implementation of this embodiment, data analysis is performed on the device data acquired in real time to obtain analysis results, including the following steps:

[0136] S610: Extract key features of the device data acquired in real time.

[0137] S620: Obtain historical data of power station equipment.

[0138] S630: Build a normal performance baseline of the power station equipment based on historical data, and determine the current equipment performance indicators based on key features of the equipment data acquired in real time.

[0139] S640. Compare the current device performance indicator with the normal performance baseline to obtain b comparison results, where b is a positive integer.

[0140] S650. Determine m normal indicators and n abnormal indicators of the power station equipment according to b comparison results, where m+n=b, and m and n are both integers.

[0141] S660. Take m normal indicators and n abnormal indicators as analysis results.

[0142] In this embodiment, the key features of the equipment data acquired in real time can be extracted through a pre-trained model. The historical data includes the data collected historically from the power station equipment, which may include equipment temperature, equipment pressure, equipment voltage, current, etc. It may also include the operation data of the equipment in different time periods, fault records, maintenance logs, etc.

[0143] By analyzing and modeling historical data or using statistical principles, a baseline for normal performance of power plant equipment can be established. The baseline can be the average indicator or range of the equipment under normal operating conditions. By comparing the equipment data obtained in real time with the baseline, the performance indicators of the current equipment, such as the degree of deviation and abnormality, can be determined.

[0144] Taking the use of statistical principles to establish a normal performance baseline for power station equipment as an example, the normal performance baseline can be:

[0145] B i =μ i ±k·σ i ;

[0146] Wherein, μi and σi are the historical average and standard deviation of the performance index of the ith device, respectively, and k is a constant of the control limit. In this embodiment, k is 3, corresponding to a confidence level of 99.7%.

[0147] For each device performance indicator Pi extracted in real time, its deviation from the normal performance baseline can be calculated = |P i -μ i |.

[0148] By comparing the current equipment performance index with the normal performance baseline, a comparison result can be obtained. The comparison result can reflect the performance status of the current power station equipment, such as whether it is normal or abnormal.

[0149] Based on the comparison results, the equipment indicators can be divided into normal indicators and abnormal indicators. Normal indicators are indicators that are consistent with or close to the normal performance baseline, while abnormal indicators are indicators that deviate far from the normal performance baseline or are obviously abnormal. Based on the number of comparison results, the number of normal indicators and abnormal indicators can be determined.

[0150] This implementation method deeply analyzes the real-time acquired equipment data, extracts key features, establishes a normal performance baseline, determines the current equipment performance indicators, and determines normal indicators and abnormal indicators based on the comparison results, thereby achieving evaluation and analysis of equipment performance and operating status.

[0151] In one implementation of this embodiment, when the analysis result is inconsistent with the previous analysis result, determining the current target task according to the analysis result includes the following steps:

[0152] S710: If at least one abnormal indicator in the analysis result is different from the abnormal indicator in the previous analysis result; and / or

[0153] S720. If there is at least one abnormal indicator in the analysis result that does not exist in the previous analysis result, obtain a target indicator, wherein the target indicator is at least one abnormal indicator in the analysis result that is different from the previous analysis result and / or at least one abnormal indicator in the analysis result that does not exist in the previous analysis result.

[0154] S730. Determine the current target task according to the target indicator.

[0155] If at least one abnormal indicator in the analysis result is different from the abnormal indicator in the previous analysis result, it may indicate that the performance or operating status of the power plant equipment has changed; if at least one abnormal indicator in the analysis result is not present in the previous analysis result, it may indicate that a new abnormal situation has occurred in the power plant equipment. If any of the above situations occurs, the current target task is determined based on the target indicator.

[0156] For example, if an abnormal indicator is different from the previous analysis result, equipment fault diagnosis and maintenance work may be required to resolve the new abnormal situation. At this time, the current target task is fault diagnosis.

[0157] If there is an abnormal indicator that does not exist in the previous analysis result, it is necessary to understand the cause of the abnormal situation. In this case, the current target task is fault query.

[0158] This implementation method determines the current target task according to the difference in the analysis results, which helps to deal with equipment abnormalities in a timely manner.

[0159] In one implementation of this embodiment, determining the current target task according to the target indicator includes the following steps:

[0160] S810. Classify the target indicator according to the corresponding properties to obtain the type of the target indicator, and obtain the target device corresponding to the type of the target indicator. The type of the target indicator includes temperature abnormality type, pressure abnormality type, vibration abnormality type, current abnormality type, etc.

[0161] S820. Determine the current target task based on the type of target indicator and target equipment in combination with the historical maintenance database.

[0162] The historical maintenance database stores data detected during historical maintenance, including the type of target indicator and the corresponding target equipment.

[0163] The type of target indicator and the corresponding target equipment are matched with the historical maintenance database to determine the current target task. For example, if the temperature of power plant equipment A is abnormal, the current target task can be determined through the historical maintenance database.

[0164] This implementation method can determine the current target task based on the historical maintenance database, and can effectively ensure the accuracy of the transfer learning network model in fault diagnosis.

[0165] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present invention.

[0166] In this embodiment, a power plant equipment fault diagnosis device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made are not repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0167] Figure 4 is a structural block diagram of a power plant equipment fault diagnosis device according to an embodiment of the present invention. Figure 4 As shown, the device comprises:

[0168] A freezing module, used for taking the fault diagnosis model as a source model and freezing the primary layer of the source model;

[0169] A loop module is used to obtain equipment data of power station equipment in real time, and based on the equipment data obtained in real time, cyclically execute the first loop step until a preset number of executions is reached to obtain a fault diagnosis result;

[0170] The first cycle step includes:

[0171] Perform data analysis on the equipment data acquired in real time to obtain analysis results;

[0172] When the analysis result is inconsistent with the previous analysis result, the current target task is determined according to the analysis result;

[0173] Adjust the high-level layers of the previous source model according to the current target task so that the high-level layers are adapted to the current target task;

[0174] Construct a transfer learning network model, where the network architecture of the transfer learning network model is a hybrid architecture of LSTM layer and Transformer;

[0175] According to the frozen primary layer and the adjusted advanced layer, the source model is transferred through the transfer learning network model to obtain the transfer learning network model after transfer learning;

[0176] The transfer learning network model after transfer learning is used to diagnose the faults of power station equipment and obtain the fault diagnosis results.

[0177] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0178] An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0179] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0180] An embodiment of the present invention further provides an electronic device, including:

[0181] The above-mentioned power station equipment fault diagnosis device;

[0182] a memory configured to store instructions; and

[0183] The processor is configured to call instructions from the memory and implement the above-mentioned power plant equipment fault diagnosis method when executing the instructions.

[0184] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0185] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0186] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0187] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for diagnosing power plant equipment faults, characterized in that: Applied to an electronic device, the electronic device has a fault diagnosis model integrated therein, the network architecture of the fault diagnosis model is a Transformer model, the fault diagnosis model includes a primary layer and a high-level layer, and the method includes: Using the fault diagnosis model as a source model and freezing the primary layer of the source model; Acquire equipment data of power station equipment in real time, and based on the equipment data acquired in real time, cyclically execute the first cycle step until a preset number of executions is reached to obtain a fault diagnosis result; Wherein, the first cycle step comprises: Performing data analysis on the device data acquired in real time to obtain analysis results; In the case where the analysis result is inconsistent with the previous analysis result, determining the current target task according to the analysis result; Adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer is adapted to the current target task; Constructing a transfer learning network model, wherein the network architecture of the transfer learning network model is a hybrid architecture of an LSTM model and a Transformer model; According to the frozen primary layer and the adjusted high-level layer, transfer learning is performed on the source model through the transfer learning network model to obtain a transfer learning network model after transfer learning; Using the transfer learning network model after the transfer learning to perform fault diagnosis on the power station equipment to obtain a fault diagnosis result; The primary layer of the fault diagnosis model includes at least one Encoder layer and a self-attention sublayer of each Encoder layer; The freezing the primary layer of the source model comprises: During the process of constructing the source model, traversing the layers of the source model, and determining at least one of the encoder layers or the self-attention sublayer of each encoder layer as a frozen layer; Freezing the weight of the frozen layer so that the weight of the frozen layer is not updated during the training process of the source model; The method of performing transfer learning on the source model through the transfer learning network model according to the frozen primary layer and the adjusted high-level layer to obtain the transfer learning network model after transfer learning includes: Based on the primary layer of the previous source model, add an LSTM module, and connect the output of the previous source model to the LSTM module to construct an initial transfer learning network model; Training the initial transfer learning network model, wherein during the training of the initial transfer learning network model, the LSTM module and the high-level layer are updated in real time, and the primary layer is not updated; Using the trained initial transfer learning network model as the transfer learning network model after transfer learning; The step of adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer adapts to the current target task includes: Defining a loss function according to the current target task; Dividing the device data acquired in real time into a plurality of task data sets; Using each of the task data sets to train the previous source model in sequence, wherein in each process of training the previous source model, the gradient of the loss function is calculated; Execute the second loop step repeatedly until a preset number of executions is reached, so that the high-level layer adapts to the current target task, wherein the number of executions is greater than the number of trainings of the previous source model; The second cycle step comprises: Update the parameters of the previous high-level layer of the source model according to the gradient of the loss function.

2. The method according to claim 1, characterized in that: The updating of the parameters of the previous high-level layer of the source model according to the gradient of the loss function comprises: Adopting a gradient descent formula, updating the parameters of the previous high-level layer of the source model according to the gradient of the loss function; Wherein, the gradient descent formula includes: ； In the formula, θ new is the updated parameters of the high-level layer of the previous source model, θ old is the parameters of the high-level layer of the previous source model before updating, η is the learning rate, which is used to control the step size of parameter update each time the second cycle step is executed. J(θ) is the loss function, ▽ θ J(θ) The loss function is about the parameter θ gradient.

3. The method according to claim 1, characterized in that The performing of data analysis on the device data acquired in real time to obtain analysis results includes: Extracting key features of the device data acquired in real time; Acquiring historical data of the power station equipment; Constructing a normal performance baseline of the power plant equipment according to the historical data, and determining current equipment performance indicators according to key features of the equipment data acquired in real time; Comparing the current device performance indicator with the normal performance baseline to obtain b comparison results, where b is a positive integer; According to the b comparison results, determine m normal indicators and n abnormal indicators of the power station equipment, where m+n=b, and m and n are both integers; The m normal indicators and the n abnormal indicators are used as the analysis results.

4. The method according to claim 3, characterized in that When the analysis result is inconsistent with the previous analysis result, determining the current target task according to the analysis result includes: If at least one of the abnormal indicators in the analysis result is different from the abnormal indicator in the previous analysis result; and / or If there is at least one abnormal indicator in the analysis result that does not exist in the previous analysis result, a target indicator is obtained, wherein the target indicator is at least one abnormal indicator in the analysis result that is different from the previous analysis result and / or at least one abnormal indicator in the analysis result that does not exist in the previous analysis result; According to the target indicators, determine the current target tasks.

5. The method according to claim 4, characterized in that Determining the current target task according to the target indicator includes: Classify the target indicator according to corresponding properties to obtain the type of the target indicator, and obtain the target device corresponding to the type of the target indicator; In combination with the historical maintenance database, the current target task is determined according to the type of the target indicator and the target equipment.

6. A power plant equipment fault diagnosis device, applied to the power plant equipment fault diagnosis method according to any one of claims 1 to 5, characterized in that: include: A freezing module, used for taking the fault diagnosis model as a source model and freezing the primary layer of the source model; A loop module, used for acquiring equipment data of power station equipment in real time, and based on the equipment data acquired in real time, cyclically executing the first loop step until a preset number of executions is reached to obtain a fault diagnosis result; Wherein, the first cycle step comprises: Performing data analysis on the device data acquired in real time to obtain analysis results; In the case where the analysis result is inconsistent with the previous analysis result, determining the current target task according to the analysis result; Adjusting the high-level layer of the previous source model according to the current target task so that the high-level layer is adapted to the current target task; Constructing a transfer learning network model, wherein the network architecture of the transfer learning network model is a hybrid architecture of an LSTM model and a Transformer model; According to the frozen primary layer and the adjusted high-level layer, transfer learning is performed on the source model through the transfer learning network model to obtain a transfer learning network model after transfer learning; The transfer learning network model after the transfer learning is used to perform fault diagnosis on the power station equipment to obtain a fault diagnosis result.

7. An electronic device, characterized in that: include: A power plant equipment fault diagnosis device according to claim 6; a memory configured to store instructions; as well as The processor is configured to call the instruction from the memory and implement the power plant equipment fault diagnosis method according to any one of claims 1 to 5 when executing the instruction.

Images