Recognition network-based gas turbine state monitoring method and system, apparatus and medium

Abstract

A recognition network-based gas turbine state monitoring method, implemented by means of a pre-constructed monitoring model, and comprising: according to gas turbine data and by means of a first residual network module, acquiring first feature data (S101); according to the first feature data and by means of a second residual network module, acquiring second feature data (S102); and acquiring output data on the basis of the second feature data and by means of an output module, and using the output data as a state detection result of the gas turbine (S103). A recognition network-based gas turbine state monitoring model comprises: a first residual network module (710), a second residual network module (720) and an output module (730), the first residual network module (710) and the second residual network module (720) both comprising LSTM layers, wherein first feature data is acquired by means of the first residual network module (710) and according to gas turbine data, second feature data is acquired by means of the second residual network module (720) and according to the first feature data, output data is acquired by means of the output module (730) and on the basis of the second feature data, and the output data is used as a state detection result of the gas turbine. The present application improves the gas turbine state monitoring efficiency and state monitoring accuracy, thus being applicable to state monitoring for gas turbines of different models and specifications.

Description

Gas turbine condition monitoring method, system, device and medium based on identification network Technical Field

[0001] The present application relates to the technical field of gas turbine condition monitoring, and in particular to a gas turbine condition monitoring method, system, device and medium based on an identification network. Background Art

[0002] Heavy-duty gas turbines operate in complex environments and under highly variable operating conditions. As their operating time increases, the risk of failure increases significantly. A gas turbine failure can degrade control system performance, resulting in significant economic losses to industrial production processes. Therefore, monitoring the operating status of heavy-duty gas turbines is essential.

[0003] Existing methods for gas turbine condition monitoring are mainly based on statistical analysis or model building. Statistical analysis or model building requires feature extraction of the gas turbine. However, feature extraction for complex gas turbines is difficult and often requires manual feature extraction by professionals, which is time-consuming and labor-intensive, resulting in low efficiency of gas turbine condition monitoring.

[0004] Currently, no effective solution has been proposed for the problem of low accuracy in gas turbine condition monitoring in related technologies.

[0005] Summary of the Invention

[0006] The embodiments of the present application provide a method, system, device and medium for gas turbine state monitoring based on an identification network, so as to at least solve the problem of low efficiency of gas turbine state monitoring in related technologies.

[0007] In a first aspect, an embodiment of the present application provides a gas turbine state monitoring method based on a recognition network, characterized in that the method is implemented by a pre-built monitoring model, the monitoring model comprising: a first residual network module, a second residual network module and an output module; the first residual network module and the second residual network module both include an LSTM layer, and the method comprises:

[0008] Obtaining first feature data through the first residual network module according to the gas turbine data;

[0009] Obtaining second feature data through the second residual network module according to the first feature data;

[0010] Output data is acquired through the output module based on the second characteristic data, and the output data is used as a state detection result of the gas turbine.

[0011] In one embodiment, the LSTM layer includes a first LSTM layer; the first residual network module further includes: a first activation layer and a first focusing layer connected to the first LSTM layer;

[0012] The first activation layer is used to activate first activation data acquired based on the gas turbine data;

[0013] The first LSTM layer is used to obtain first sequence data according to the first activation data;

[0014] The first focusing layer acquires first focusing data based on the first sequence data; and / or

[0015] The LSTM layer further includes a second LSTM layer, and the first residual network module further includes: a second activation layer and a first maximum pooling processing layer connected to the second LSTM layer;

[0016] The second activation layer is used to activate second activation data based on the output of the second batch normalization layer;

[0017] The second LSTM layer is used to obtain second sequence data according to the second activation data output by the second batch normalization layer;

[0018] The first maximum pooling processing layer is used to obtain first pooled data according to the second sequence data.

[0019] In one embodiment, an attention layer is included between the first LSTM layer and the first focusing layer, and the attention layer is used to obtain attention data according to the first sequence data;

[0020] When acquiring the first focus data based on the first sequence data, the first focus layer is specifically used to: acquire the first focus data according to the attention data.

[0021] In one embodiment, the first residual network module further includes a first convolutional jump layer, the input of the first convolutional jump layer is used to receive the gas turbine data, and the output of the first convolutional jump layer is used to connect the output of the first residual network module.

[0022] In one embodiment, the second residual network module includes at least two feature extraction units;

[0023] The LSTM layer of the second residual network module includes: a third LSTM layer arranged in the feature extraction unit, and the second residual network module further includes: a third activation layer and a second focusing layer connected to the third LSTM layer;

[0024] The third activation layer is used to activate third activation data obtained based on the gas turbine data;

[0025] The third LSTM layer is used to obtain third sequence data according to the third activation data;

[0026] The second focusing layer acquires second focusing data based on the third sequence data; and / or

[0027] The LSTM layer of the second residual network module includes: a fourth LSTM layer arranged in the feature extraction unit, and the second residual network module further includes: a fourth activation layer and a second maximum pooling processing layer connected to the fourth LSTM layer;

[0028] The fourth activation layer is used to activate fourth activation data based on the output of the fourth batch normalization layer;

[0029] The fourth LSTM layer is used to obtain fourth sequence data according to the fourth activation data output by the fourth batch normalization layer;

[0030] The second maximum pooling processing layer is used to obtain second pooled data based on the fourth sequence data.

[0031] In one embodiment, the second residual network module further includes a second convolutional jump layer, the input of the second convolutional jump layer is used to receive data output by the first feature extraction unit, and the output of the second convolutional jump layer is used to connect the output of the second feature extraction unit.

[0032] In one embodiment, obtaining second feature data through the second residual network module according to the first feature data includes:

[0033] Acquire a plurality of feature extraction unit data through a plurality of the feature extraction units according to the first feature data;

[0034] Second feature data is obtained based on the data of a plurality of feature extraction units.

[0035] In one embodiment, the output module includes a global pooling processing layer, and the global pooling processing layer is used to obtain global data based on the second feature data.

[0036] In a second aspect, an embodiment of the present application provides a gas turbine condition monitoring system based on a recognition network, characterized in that the system is implemented by a pre-built monitoring model, the monitoring model comprising: a first residual network module, a second residual network module and an output module; the first residual network module and the second residual network module both include an LSTM layer, and the device comprises:

[0037] A first residual network module is used to obtain first feature data through the first residual network module according to the gas turbine data;

[0038] A second residual network module is used to merge a plurality of the second feature data and input the combined data into an output block to output a gas turbine state monitoring result;

[0039] The output block module obtains output data through the output module based on the second characteristic data, and uses the output data as the state detection result of the gas turbine.

[0040] In a third aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, a gas turbine state monitoring method based on an identification network as described in the first aspect above is implemented.

[0041] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a gas turbine state monitoring method based on an identification network as described in the first aspect above.

[0042] The gas turbine state monitoring method, system, device and medium based on the identification network provided by the embodiments of the present application have at least the following technical effects.

[0043] This is achieved through a pre-built monitoring model comprising a first residual network module, a second residual network module, and an output module. Both the first and second residual network modules include LSTM layers. The method includes: obtaining first feature data from the gas turbine data through the first residual network module; obtaining second feature data from the first feature data through the second residual network module; and obtaining output data based on the second feature data through the output module, which serves as the gas turbine status detection result. This method solves the problem of low gas turbine status monitoring efficiency, improves both efficiency and accuracy, and is applicable to gas turbine status monitoring of various models and specifications.

[0044] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0046] FIG1 is a flow chart of a gas turbine state monitoring method based on an identification network according to an exemplary embodiment of the present application;

[0047] FIG2 is a flow chart of step S101 according to an exemplary embodiment;

[0048] FIG3 is a structural block diagram of a first residual network module according to an exemplary embodiment;

[0049] FIG4 is a flow chart of step S102 according to an exemplary embodiment;

[0050] FIG5 is a block diagram showing a structure of a second residual network module according to an exemplary embodiment;

[0051] FIG6 is a flow chart of step S103 according to an exemplary embodiment;

[0052] FIG7 is a block diagram of a gas turbine condition monitoring system based on an identification network according to an embodiment of the present application;

[0053] Fig. 8 is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0055] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0056] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0057] Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by a person of ordinary skill in the technical field to which this application belongs. The words "one", "a", "the" and the like used in this application do not indicate a limit on quantity and may indicate the singular or plural. The terms "include", "comprise", "have" and any variations thereof used in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units that are inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The word "multiple" used in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0058] Based on the above situation, embodiments of the present application provide a gas turbine state monitoring method, system, device and medium based on an identification network.

[0059] In a first aspect, an embodiment of the present application provides a gas turbine state monitoring method based on an identification network. FIG1 is a flow chart of a gas turbine state monitoring method based on an identification network. As shown in FIG1 , the gas turbine state monitoring method based on an identification network is implemented by a pre-built monitoring model. The monitoring model includes: a first residual network module, a second residual network module, and an output module; the first residual network module and the second residual network module both include an LSTM layer. The method includes:

[0060] Step S101: Obtain first feature data through a first residual network module according to gas turbine data;

[0061] Step S102: Obtain second feature data through a second residual network module according to the first feature data;

[0062] Step S103: obtaining output data through the output module based on the second characteristic data, and using the output data as a state detection result of the gas turbine.

[0063] In summary, the gas turbine state monitoring method based on the identification network provided in the embodiment of the present application improves the gas turbine state monitoring efficiency and state monitoring accuracy through steps S101 to S103, is suitable for state monitoring of gas turbines of different models and specifications, and solves the problem of low efficiency of gas turbine state monitoring existing in related technologies.

[0064] FIG2 is a flow chart of step S101 according to an exemplary embodiment. As shown in FIG2 , step S101 includes:

[0065] Step S1011: Acquire gas turbine data;

[0066] Step S1012: construct a first residual network module;

[0067] Step S1013: input the gas turbine data into the first residual network module to obtain first feature data.

[0068] Step S1011 specifically includes obtaining time series operating condition data of the gas turbine and performing data augmentation on the time series data using a sliding average window. By applying the sliding average window to the original data, a series of different smoothed sequences can be generated, each representing a different time period in the original data. These new smoothed sequences can be used as additional training samples. By using the sliding average window to augment the time series data, the generalization and robustness of the model are improved.

[0069] Figure 3 is a block diagram of the first residual network module structure according to an exemplary embodiment. As shown in Figure 3, the first batch of normalization layers, the first activation layer, the first LSTM layer, the first attention layer, the first focus layer, the second batch of normalization layers, the second activation layer, the second LSTM layer and the first maximum pooling processing layer are constructed in sequence, wherein the output of the previous layer structure is used as the input of the next layer structure, and is connected through the first convolutional jump layer operation before the first batch of normalization layers and after the first maximum pooling processing layer.

[0070] In one embodiment, the LSTM layer includes: the first residual network module further includes: a first activation layer and a first focusing layer connected to the first LSTM layer;

[0071] The first activation layer and the first focusing layer connected to the first LSTM layer;

[0072] The first activation layer is used to activate first activation data acquired based on gas turbine data;

[0073] The first LSTM layer is used to obtain first sequence data according to the first activation data;

[0074] The first focusing layer acquires first focusing data based on the first sequence data.

[0075] Specifically, the first activation layer activates the first activation data obtained based on the gas turbine data. In a neural network, each neuron corresponds to an activation function, which nonlinearly maps the first batch of normalized input data and outputs a new signal as the input to the next layer (the first LSTM layer). Useful activation functions include, but are not limited to, Sigmoid, ReLU, Tanh, or Softmax.

[0076] The first LSTM layer obtains the first sequence data based on the first activation data. By introducing a gating mechanism, the first LSTM layer can selectively forget or update some past information, thereby better processing sequence data. First, the input xt and the hidden state ht-1 at the previous moment are combined to calculate a candidate value. Then, the activation layer function is used to calculate an input gate to control which information is remembered. The activation function is used to calculate a forget gate to control which past information is forgotten. A new memory unit is calculated based on the input gate and the candidate value. The activation function is used to calculate an output gate to control which information is output. The new hidden state is calculated based on the output gate and the memory unit, and the processed first sequence data is used as the input of the first focusing layer. The first LSTM layer can improve gradient vanishing and prevent overfitting.

[0077] The first focusing layer obtains the first focused data based on the processed first sequence data. This function typically partitions the input tensor into distinct regions and operates on each region to produce a smaller output tensor. The size and stride of the pooling window affect the size and shape of the pooled output tensor. All first sequence data is combined into a new tensor, which serves as the input to the next layer. Using the focusing layer reduces the number of features and improves computational efficiency.

[0078] In one embodiment, the LSTM layer includes a first LSTM layer; the first residual network module further includes: a first activation layer and a first focusing layer connected to the first LSTM layer; and or

[0079] The LSTM layer further includes a second LSTM layer, and the first residual network module further includes: a second activation layer and a first maximum pooling processing layer connected to the second LSTM layer;

[0080] The second activation layer is used to activate the second activation data based on the output of the second batch normalization layer;

[0081] The second LSTM layer is used to obtain second sequence data according to the second activation data output by the second batch normalization layer;

[0082] The first maximum pooling processing layer is used to obtain first pooled data according to the second sequence data.

[0083] Specifically, the second activation layer activates the first activation data obtained based on the gas turbine data. In a neural network, each neuron corresponds to an activation function, which nonlinearly maps the first batch of normalized input data and outputs a new signal as the input to the next layer (the second LSTM layer). Useful activation functions include, but are not limited to, Sigmoid, ReLU, Tanh, or Softmax.

[0084] The second LSTM layer obtains the second sequence data based on the first activation data. By introducing a gating mechanism, the second LSTM layer can selectively forget or update some past information, thereby better processing sequence data. First, the input xt and the hidden state ht-1 at the previous moment are combined to calculate a candidate value. Then, the activation layer function is used to calculate an input gate to control which information is remembered. The activation function is used to calculate a forget gate to control which past information is forgotten. A new memory unit is calculated based on the input gate and the candidate value. The activation function is used to calculate an output gate to control which information is output. The new hidden state is calculated based on the output gate and the memory unit, and the processed second sequence data is used as the input of the first focusing layer. The second LSTM layer can improve gradient vanishing and prevent overfitting.

[0085] The first maximum pooling processing layer is used to obtain the first pooled data based on the second sequence data. The purpose of maximum pooling is to select the most important features in the window, and the maximum value usually represents the most important feature in the window. First, define a pooling window size and stride. Its size can be selected according to the size of the input data. The stride represents the distance the pooling window slides each time. Move the pooling window from the upper left corner of the input data and find the maximum value in the window at each position. All the maximum values ​​are combined into a new tensor (i.e., the first pooled data), which will be used as the input of the next layer.

[0086] In one embodiment, an attention layer is included between the first LSTM layer and the first focusing layer, and the attention layer is used to obtain attention data based on the first sequence data; when the first focusing layer obtains the first focusing data based on the first sequence data, it is specifically used to: obtain the first focusing data based on the attention data.

[0087] Specifically, an attention layer is included between the first LSTM layer and the first focusing layer. The attention layer is used to obtain attention data based on the first sequence data and to obtain second focused data based on the attention data. The attention layer has three input tensors: query, key, and value. The input dimensions are determined based on the data passed down from the first LSTM layer. The attention mechanism function Multi-Head Attention is used to calculate the similarity matrix between the query and key. The similarity matrix is ​​calculated using the dot product or other functions, followed by a softmax operation on the matrix to determine the importance of each element. The weight matrix is ​​multiplied by the value tensor to calculate a new tensor. This new tensor is then fed into the next layer. The attention layer allows the network to adaptively focus on valid areas when learning features. In addition, the attention layer is located before the output layer of each block, allowing it to allocate different attention shares to different features based on the relevance of the target variable while maintaining information integrity.

[0088] In one embodiment, the first residual network module further includes a first convolutional jump layer, the input of the first convolutional jump layer is used to receive gas turbine data, and the output of the first convolutional jump layer is used to connect the output of the first residual network module.

[0089] Specifically, the first convolutional jump layer uses a standard convolution operation to pass the input data into the convolutional layer for feature extraction. During the process, different convolution kernels are used to convolve the input data to obtain more features. A direct connection is established between the output and input of the convolutional layer. The first convolutional jump layer can add the output of the convolutional layer directly to the input of the subsequent layer. The tensor after the jump layer is added to the tensor output by the convolutional layer to ensure that the two tensors have the same dimension. The added tensor is used as the input of the next layer to continue subsequent calculations and predictions. Convolutional jump layers can help the model better process deep-level features and improve the performance and accuracy of the model. When implementing it, it is necessary to establish a direct connection between the output and input of the convolutional layer, and add the tensor after the jump layer to the tensor output by the convolutional layer to obtain a better feature representation.

[0090] It's worth noting that the first normalization layer involves calculating the mean and variance of each feature dimension of the input data X in a mini-batch, where m represents the mini-batch size and xi represents the input data for the i-th sample. For each feature dimension, the input data in that dimension is normalized using the calculated mean and variance. The normalized result is linearly transformed and translated to obtain the final output. During training, since the mean and variance of each mini-batch vary, a moving average is applied to these mini-batches to obtain the global mean and variance. Batch normalization accelerates network convergence, improves model performance, and reduces the risk of overfitting.

[0091] The above step S1013, inputting the gas turbine data into the first residual network module to obtain the first feature data, specifically includes: inputting the gas turbine data into the input block to output the residual features or skip-layer features, wherein the residual features or skip-layer features are the first feature data.

[0092] FIG4 is a flow chart of step S102 according to an exemplary embodiment. As shown in FIG4 , step S102 obtains second feature data through a second residual network module according to the first feature data, including:

[0093] Step S1021: construct a second residual network module;

[0094] Step S1022: Input the first feature data obtained in step S101 into the second residual network module to obtain second feature data.

[0095] FIG5 is a block diagram of a second residual network module structure according to an exemplary embodiment. As shown in FIG5 , the second residual network module includes several feature extraction units. The feature extraction units specifically include: a third batch normalization layer, a third activation layer, a third LSTM layer, a second attention layer, a second focus layer, a fourth batch normalization layer, a fourth activation layer, a fourth LSTM layer, and a second maximum pooling layer constructed in sequence. The output of the previous layer serves as the input of the next layer. The layers before the third batch normalization layer and after the second maximum pooling layer are connected by a convolutional jump layer.

[0096] In one embodiment, the second residual network module includes at least two feature extraction units; the LSTM layer of the second residual network module includes: a third LSTM layer arranged in the feature extraction unit, and the second residual network module further includes: a third activation layer and a second focusing layer connected to the third LSTM layer;

[0097] The third activation layer is used to activate third activation data obtained based on gas turbine data;

[0098] The third LSTM layer is used to obtain third sequence data according to the third activation data;

[0099] The second focusing layer acquires second focusing data based on the third sequence data; and / or

[0100] The LSTM layer of the second residual network module includes: a fourth LSTM layer arranged in the feature extraction unit, and the second residual network module further includes: a fourth activation layer and a second maximum pooling processing layer connected to the fourth LSTM layer;

[0101] The fourth activation layer is used to activate the fourth activation data based on the output of the second batch normalization layer;

[0102] The fourth LSTM layer is used to obtain fourth sequence data according to the fourth activation data output by the second batch normalization layer;

[0103] The second maximum pooling processing layer is used to obtain second pooled data based on the fourth sequence data.

[0104] Specifically, it should be noted that, with the exception of the first convolutional skip layer in the first residual network module and the second convolutional skip layer in the second residual network module, the layer structure of each feature extraction unit is identical to that of the first residual network module. The second residual network module also includes a second convolutional skip layer, whose input receives the data output by the first feature extraction unit, and whose output is connected to the output of the second feature extraction unit. The second convolutional skip layer does not use standard convolution operations, but instead passes the input data to a non-convolutional layer for feature extraction. This process can involve a fully connected layer, a pooling layer, or a deconvolution layer. A direct connection is established between the output and input of the non-convolutional layer. The output of the non-convolutional layer can be directly added to the input of the subsequent layer. The tensor after the skip layer and the tensor output by the non-convolutional layer are added together. This operation ensures that the two tensors have the same shape. The added tensor serves as the input to the next layer for subsequent calculations and predictions.

[0105] In one embodiment, step S1022, inputting the first feature data obtained in step S101 into a second residual network module, obtaining the second feature data includes: obtaining a plurality of feature extraction unit data using a plurality of feature extraction units based on the first feature data; and obtaining second feature data based on the plurality of feature extraction unit data. The number of feature extraction units is determined based on the actual fault condition of the gas turbine. A greater number of feature extraction units results in clearer fault feature extraction and a slower speed. A smaller number of feature extraction units results in less clear fault features and a faster calculation speed. The first feature data is input into the plurality of feature extraction unit data, outputting the plurality of feature extraction unit data, and merging the plurality of feature extraction unit data to obtain the second feature data.

[0106] FIG6 is a flow chart of step S103 according to an exemplary embodiment. As shown in FIG6 , step S106 includes:

[0107] Step S1031: construct an output module;

[0108] Step S1032: input the second characteristic data into the output module, and use the output data as the state detection result of the gas turbine.

[0109] The above-mentioned step S1031, constructing the output module specifically includes: constructing the fifth batch normalization layer, the global pooling processing layer, the third focusing layer, the sixth batch normalization layer and the fourth focusing layer in sequence. Among them, the global pooling processing layer selects the maximum pooling or average pooling function, and the maximum pooling or average pooling function will operate on the entire input data and output a scalar. The pooling function is applied to the input data to obtain a scalar. The scalar represents the statistical information of all features in the input data. The scalar is used as the input of the next layer to continue subsequent calculations and predictions. The global pooling processing layer helps the model extract global features from the input data, thereby improving performance and accuracy. Among them, the fifth batch normalization layer and the sixth batch normalization layer have the same structure as the first batch normalization layer in the first residual network module, and the third focusing layer, the fourth focusing layer and the first focusing layer have the same structure.

[0110] Step S1032: input the second characteristic data into the output module, and use the output data as the state detection result of the gas turbine.

[0111] In summary, the gas turbine state monitoring method based on a recognition network provided in the embodiments of the present application is implemented using a pre-built monitoring model. The monitoring model includes: a first residual network module, a second residual network module, and an output module. Both the first residual network module and the second residual network module include LSTM layers. The method includes: obtaining first feature data based on gas turbine data through the first residual network module; obtaining second feature data based on the first feature data through the second residual network module; and obtaining output data based on the second feature data through the output module, and using the output data as the gas turbine state detection result. This method can quickly and accurately identify and monitor the state of a gas turbine with high reliability and accuracy, and is suitable for fields such as gas turbine state monitoring.

[0112] In a second aspect, an embodiment of the present application provides a gas turbine state monitoring system based on an identification network. FIG7 is a block diagram of a gas turbine state monitoring system based on an identification network according to an embodiment of the present application. As shown in FIG7 , the gas turbine state monitoring system based on the identification network is implemented by a pre-built monitoring model, and the monitoring model includes: a first residual network module, a second residual network module and an output module; the first residual network module and the second residual network module both include LSTM layers, and the device includes: a first residual network module 710, a second residual network module 720 and an output block module 730. Wherein:

[0113] A first residual network module 710 is configured to obtain first feature data through the first residual network module according to the gas turbine data;

[0114] The second residual network module 720 is used to combine the plurality of second feature data and input them into the output block to output the gas turbine state monitoring result;

[0115] The output block module 730 obtains output data through the output module based on the second characteristic data, and uses the output data as a state detection result of the gas turbine.

[0116] In summary, the present application provides a gas turbine state monitoring system based on an identification network, which improves the efficiency and accuracy of gas turbine state monitoring.

[0117] In one embodiment, the first residual network module 710 includes a first activation layer unit, a first LSTM layer unit, a first focus layer unit, a second LSTM layer unit, a second activation layer unit, and a first maximum pooling processing layer unit.

[0118] The first activation layer unit is used to activate first activation data acquired based on gas turbine data;

[0119] The first LSTM layer unit is used to obtain first sequence data according to the first activation data;

[0120] The first focusing layer unit acquires first focusing data based on the first sequence data; and / or

[0121] In one embodiment, the LSTM layer unit further includes a second LSTM layer unit, and the first residual network module further includes: a second activation layer unit and a first maximum pooling processing layer unit connected to the second LSTM layer unit;

[0122] The second activation layer unit is used to activate the second activation data based on the output of the first batch of normalization layers;

[0123] The second LSTM layer unit is used to obtain the second sequence data according to the second activation data output by the first batch of normalization layers;

[0124] A first attention layer unit is included between the first LSTM layer unit and the first focusing layer unit, and the first attention layer unit is used to obtain first attention data according to the first sequence data;

[0125] When obtaining the first focusing data based on the first sequence data, the first focusing layer is specifically used to: obtain the first focusing data according to the first attention data.

[0126] In one embodiment, the first residual network module further includes a first convolutional skip layer unit;

[0127] The input of the first convolutional skip layer unit is used to receive gas turbine data;

[0128] The output of the first convolutional skip layer unit is used to connect the output of the first residual network module.

[0129] In one embodiment, the second residual network module includes at least two feature extraction units;

[0130] The LSTM layer unit of the second residual network module includes: a third LSTM layer unit set in the feature extraction unit, and the second residual network module further includes: a third activation layer unit and a second focusing layer unit connected to the third LSTM layer unit;

[0131] The third activation layer is used to activate third activation data obtained based on gas turbine data;

[0132] The third LSTM layer is used to obtain third sequence data according to the third activation data;

[0133] The second focusing layer acquires second focusing data based on the third sequence data; and / or

[0134] The LSTM layer of the second residual network module includes: a fourth LSTM layer arranged in the feature extraction unit, and the second residual network module further includes: a fourth activation layer and a second maximum pooling processing layer connected to the fourth LSTM layer;

[0135] The fourth activation layer is used to activate the fourth activation data based on the output of the fourth batch normalization layer;

[0136] The fourth LSTM layer is used to obtain fourth sequence data according to the fourth activation data output by the fourth batch normalization layer;

[0137] The second maximum pooling processing layer is used to obtain second pooled data based on the fourth sequence data.

[0138] In one embodiment, the second residual network module further includes a second convolutional skip layer unit; the input of the second convolutional skip layer unit is used to receive data output by the first feature extraction unit;

[0139] The output of the second convolutional skip layer unit is used to connect the output of the second feature extraction unit.

[0140] In one embodiment, the second feature data is obtained by a second residual network module according to the first feature data, for:

[0141] Acquire a plurality of feature extraction unit data through a plurality of feature extraction units according to the first feature data;

[0142] Second feature data is obtained based on a plurality of feature extraction unit data.

[0143] In one embodiment, the output module includes a global pooling processing layer, which is used to obtain global data based on the second feature data.

[0144] In summary, the gas turbine state monitoring system based on the identification network provided in this application solves the problem of low efficiency of gas turbine state monitoring, improves the efficiency and accuracy of gas turbine state monitoring, and is suitable for state monitoring of gas turbines of different models and specifications.

[0145] It should be noted that the gas turbine condition monitoring system based on an identification network provided in this embodiment is used to implement the aforementioned embodiments, and details already described will not be repeated. As used above, the terms "module," "unit," "subunit," etc. may refer to a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the above embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0146] In a third aspect, an embodiment of the present application provides an electronic device. Figure 8 is a block diagram of an electronic device according to an exemplary embodiment. As shown in Figure 8 , the electronic device may include a processor 81 and a memory 82 storing computer program instructions.

[0147] Specifically, the processor 81 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0148] Among them, the memory 82 may include a large-capacity memory for data or instructions. By way of example and not limitation, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include removable or non-removable (or fixed) media. Where appropriate, the memory 82 may be inside or outside the data processing device. In a specific embodiment, the memory 82 is a non-volatile memory. In a specific embodiment, the memory 82 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0149] The memory 82 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 81 .

[0150] The processor 81 reads and executes computer program instructions stored in the memory 82 to implement any one of the gas turbine state monitoring methods based on the identification network in the above embodiments.

[0151] In one embodiment, the gas turbine condition monitoring device based on the identification network may further include a communication interface 83 and a bus 80. As shown in FIG8, the processor 81, the memory 82, and the communication interface 83 are connected via the bus 80 and communicate with each other.

[0152] The communication interface 83 is used to enable communication between the various modules, devices, units, and / or devices in the embodiments of the present application. The communication port 83 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0153] Bus 80 comprises hardware, software, or both, and couples components of the identification network-based gas turbine condition monitoring system. Bus 80 includes, but is not limited to, at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 80 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of the above. Bus 80 may include one or more buses, where appropriate. Although embodiments herein describe and illustrate a particular bus, this application contemplates any suitable bus or interconnect.

[0154] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the gas turbine state monitoring method based on the identification network provided in the first aspect.

[0155] The readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0156] In a possible embodiment, the present invention can also be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps of the gas turbine status monitoring method based on the identification network provided in the first aspect.

[0157] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or entirely on the remote device.

[0158] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0159] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A gas turbine condition monitoring method based on recognition network, characterized in that: The method is implemented by a pre-built monitoring model, which includes: a first residual network module, a second residual network module and an output module; the first residual network module and the second residual network module both include an LSTM layer, and the method includes: Acquire first feature data through the first residual network module according to the gas turbine data; Obtaining second feature data through the second residual network module according to the first feature data; Based on the second characteristic data, output data is acquired through the output module, and the output data is used as a state detection result of the gas turbine.

2. The method according to claim 1, characterized in that The LSTM layer includes a first LSTM layer; the first residual network module also includes: a first activation layer and a first focusing layer connected to the first LSTM layer; The first activation layer is used to activate first activation data acquired based on the gas turbine data; The first LSTM layer is used to obtain first sequence data according to the first activation data; The first focusing layer acquires first focusing data based on the first sequence data; and / or The LSTM layer also includes a second LSTM layer, and the first residual network module also includes: a second activation layer and a first maximum pooling processing layer connected to the second LSTM layer; The second activation layer is used to activate second activation data based on the output of the second batch normalization layer; The second LSTM layer is used to obtain second sequence data according to the second activation data output by the second batch normalization layer; The first maximum pooling processing layer is used to obtain first pooling data according to the second sequence data.

3. The method according to claim 2, characterized in that A first attention layer is included between the first LSTM layer and the first focusing layer, and the first attention layer is used to obtain first attention data according to the first sequence data; When acquiring the first focusing data based on the first sequence data, the first focusing layer is specifically used to: acquire the first focusing data according to the first attention data.

4. The method according to claim 2, characterized in that: The first residual network module also includes a first convolutional jump layer, the input of the first convolutional jump layer is used to receive the gas turbine data, and the output of the first convolutional jump layer is used to connect the output of the first residual network module.

5. The method according to claim 1, characterized in that The second residual network module includes at least two feature extraction units; The LSTM layer of the second residual network module includes: a third LSTM layer arranged in the feature extraction unit, and the second residual network module also includes: a third an activation layer and a second focusing layer; The third activation layer is used to activate third activation data acquired based on the gas turbine data; The third LSTM layer is used to obtain third sequence data according to the third activation data; The second focusing layer acquires second focusing data based on the third sequence data; and / or The LSTM layer of the second residual network module includes: a fourth LSTM layer arranged in the feature extraction unit, and the second residual network module also includes: a fourth activation layer and a second maximum pooling processing layer connected to the fourth LSTM layer; The fourth activation layer is used to activate fourth activation data based on the output of the fourth batch normalization layer; The fourth LSTM layer is used to obtain fourth sequence data according to the fourth activation data output by the fourth batch normalization layer; The second maximum pooling processing layer is used to obtain second pooling data according to the fourth sequence data.

6. The method according to claim 5, characterized in that The second residual network module also includes a second convolutional jump layer, the input of the second convolutional jump layer is used to receive data output by the first feature extraction unit, and the output of the second convolutional jump layer is used to connect the output of the second feature extraction unit.

7. The method according to claim 6, characterized in that Acquiring second feature data through the second residual network module according to the first feature data includes: Acquire a plurality of feature extraction unit data through a plurality of the feature extraction units according to the first feature data; The second feature data is obtained according to the data of a plurality of the feature extraction units.

8. The method according to claim 1, characterized in that The output module includes a global pooling processing layer, and the global pooling processing layer is used to obtain global data according to the second feature data.

9. A gas turbine condition monitoring system based on recognition network, characterized in that: The system is implemented by a pre-built monitoring model, which includes: a first residual network module, a second residual network module and an output module; the first residual network module and the second residual network module both include an LSTM layer, and the system includes: A first residual network module, used for acquiring first feature data through the first residual network module according to the gas turbine data; A second residual network module is used to merge a plurality of the second feature data and input them into an output block to output a gas turbine state monitoring result; The output block module obtains output data through the output module based on the second characteristic data, and uses the output data as the state detection result of the gas turbine.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the gas turbine condition monitoring method based on the identification network according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a gas turbine condition monitoring method based on an identification network as claimed in any one of claims 1 to 8 is implemented.

Images