Fault diagnosis method, device and electronic equipment for hydropower unit

The time sequence data of the hydropower unit is processed through the convolutional neural network model, and the problem of fault diagnosis relies on manual acquisition of vibration signal characteristics in the existing technology is solved, which improves the accuracy and accuracy of fault diagnosis, and achieves the accuracy of the operating status of the hydropower unit judge.

CN114528874BActive Publication Date: 2024-12-06WUQIANG XISHUI POWER PLANT OF WULING ELECTRIC POWER CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210082089.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-24
Publication Date
2024-12-06
Estimated Expiration
2042-01-24

AI Technical Summary

Technical Problem

The existing water-power unit fault diagnosis methods rely on manual acquisition of vibration signal characteristics, which are highly subjective, resulting in inaccurate diagnosis results. How to improve the accuracy of fault diagnosis has become an important challenge.

Method used

The time series data of the hydropower set is processed by using the convolutional neural network model. The input data is generated through overlapping sample segmentation of the timing data and the complementary set of empirical modal decomposition, and the type tag is obtained by using the trained convolutional neural network model to determine Health status.

Benefits of technology

It improves the accuracy and accuracy of fault diagnosis of hydropower units, reduces the influence of manual subjectivity, and achieves a relatively accurate judgment of operating status.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114528874B_ABST
    Figure CN114528874B_ABST
Patent Text Reader

Abstract

The present disclosure proposes a fault diagnosis method, device and electronic equipment for a hydropower unit, which relates to the technical field of fault diagnosis of hydropower units, wherein the method comprises: obtaining time series data of the hydropower unit; processing the time series data to obtain corresponding input data; processing the input data using a trained convolutional neural network model to obtain corresponding type labels; and determining the operating status of the hydropower unit according to the type labels. Thus, by using the trained convolutional neural network model and combining the time series data of the hydropower unit, the operating status of the hydropower unit can be determined more accurately, thereby effectively improving the diagnostic accuracy of the fault diagnosis of the hydropower unit.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of fault diagnosis of hydropower units, and in particular to a fault diagnosis method, device and electronic equipment for hydropower units. Background Art

[0002] Hydropower stations play an important role in the power grid, such as providing electricity and serving as emergency backup. With the grid connection of intermittent energy sources such as solar energy and wind energy, hydropower stations are taking on more peak load and frequency regulation functions in the power grid. The safe and stable operation of hydropower units is of great significance to the safety and stability of hydropower stations and even the entire power grid.

[0003] Generally, the vibration signal characteristics of the hydropower unit during operation can be analyzed to determine whether the unit has a fault. In the related art, it is usually necessary to manually obtain the vibration signal characteristics of the hydropower unit during operation, and the subjectivity of manually selecting characteristics is relatively strong, which may have a significant impact on the diagnosis results of the hydropower unit. Therefore, how to improve the accuracy of fault diagnosis of hydropower units is crucial. Summary of the invention

[0004] The present disclosure aims to solve one of the technical problems in the related art at least to some extent.

[0005] The first embodiment of the present disclosure provides a fault diagnosis method for a hydropower unit, comprising:

[0006] Obtain time series data of hydropower units;

[0007] Processing the time series data to obtain corresponding input data;

[0008] Processing the input data using the trained convolutional neural network model to obtain corresponding type labels;

[0009] The operating status of the hydroelectric generating unit is determined according to the type label.

[0010] Optionally, processing the time series data to obtain corresponding input data includes:

[0011] Perform overlapping sample segmentation on the time series data to determine multiple groups of sample sequences;

[0012] Perform complementary set empirical mode decomposition (CEEMD) on each group of sample sequences to obtain each intrinsic mode function (IMF) corresponding to each group of sample sequences;

[0013] Input data is determined according to the correlation coefficient between each of the IMFs and the corresponding sample sequence.

[0014] Optionally, the convolutional neural network model is trained and generated in the following manner:

[0015] Acquire R group training sample sequences of the hydropower unit, wherein the training sample sequences include operation data of the hydropower unit and corresponding operation labels, and R is an arbitrary positive number;

[0016] Processing each group of the training sample sequences to generate an input matrix corresponding to each group of the training sample sequences;

[0017] Processing the input matrix using the initial model to obtain a predicted label corresponding to the input matrix;

[0018] Determine a corresponding loss value based on the difference between the predicted label corresponding to each of the input matrices and the running label;

[0019] Based on the loss value, the initial model is reversely modified to generate a trained convolutional neural network model.

[0020] Optionally, the step of obtaining a training sample sequence of a group R of hydropower generating units includes:

[0021] Obtaining T groups of time series data of the hydropower unit, where T is any positive number;

[0022] The T groups of time series data are segmented into overlapping samples to generate R groups of sample sequences.

[0023] Optionally, processing each group of the training sample sequences to generate a corresponding input matrix includes:

[0024] Introducing paired positive and negative white noises with a mean of zero into each group of the training sample sequences to obtain a first sequence of the training sample sequences after the positive white noise is introduced and a second sequence of the training sample sequences after the negative white noise is introduced;

[0025] Performing empirical mode decomposition (EMD) on the first sequence and the second sequence corresponding to each group of the training sample sequences to obtain corresponding IMF components;

[0026] According to the correlation coefficient between each of the IMF components and the corresponding training sample sequence, the first M IMF components with the largest correlation coefficient in each group of the training sample sequences are determined as the corresponding input matrix.

[0027] Optionally, the processing the input matrix using the initial model to obtain a predicted label corresponding to the input matrix includes:

[0028] The input matrix is ​​input into the initial model so that the corresponding prediction label is obtained after being processed by the initial model, wherein the initial model includes an input layer, a convolution layer, a pooling layer, a flatten layer, a fully connected layer and a softmax layer.

[0029] The second embodiment of the present disclosure provides a fault diagnosis device for a hydropower unit, comprising:

[0030] An acquisition module, used to acquire the time series data of the hydropower unit;

[0031] A processing module, used for processing the time series data to obtain corresponding input data;

[0032] An input module, used to process the input data using the trained convolutional neural network model to obtain a corresponding type label;

[0033] A determination module is used to determine the operating status of the hydropower unit according to the type label.

[0034] Optionally, the processing module is specifically used to:

[0035] Perform overlapping sample segmentation on the time series data to determine multiple groups of sample sequences;

[0036] Perform complementary set empirical mode decomposition (CEEMD) on each group of sample sequences to obtain each intrinsic mode function (IMF) corresponding to each group of sample sequences;

[0037] Input data is determined according to the correlation coefficient between each of the IMFs and the corresponding sample sequence.

[0038] Optionally, the acquisition module is further used to acquire R group training sample sequences of the hydropower unit, wherein the training sample sequences include operation data of the hydropower unit and corresponding operation labels, and R is an arbitrary positive number;

[0039] The processing module is further used to process each group of the training sample sequences to generate an input matrix corresponding to each group of the training sample sequences;

[0040] The processing module is further used to process the input matrix using the initial model to obtain the prediction label corresponding to the input matrix;

[0041] The determination module is further used to determine a corresponding loss value based on a difference between a predicted label corresponding to each of the input matrices and the running label;

[0042] The device further includes a generating module, which is used for:

[0043] Based on the loss value, the initial model is reversely modified to generate a trained convolutional neural network model.

[0044] Optionally, the acquisition module is specifically used to:

[0045] Obtaining T groups of time series data of the hydropower unit, where T is any positive number;

[0046] The T groups of time series data are segmented into overlapping samples to generate R groups of sample sequences.

[0047] Optionally, the processing module is specifically used to:

[0048] Introducing paired positive and negative white noises with a mean of zero into each group of the training sample sequences to obtain a first sequence of the training sample sequences after the positive white noise is introduced and a second sequence of the training sample sequences after the negative white noise is introduced;

[0049] Performing empirical mode decomposition (EMD) on the first sequence and the second sequence corresponding to each group of the training sample sequences to obtain corresponding IMF components;

[0050] According to the correlation coefficient between each of the IMF components and the corresponding training sample sequence, the first M IMF components with the largest correlation coefficient in each group of the training sample sequences are determined as the corresponding input matrix.

[0051] Optionally, the processing module is specifically used to:

[0052] The input matrix is ​​input into the initial model so that the corresponding prediction label is obtained after being processed by the initial model, wherein the initial model includes an input layer, a convolution layer, a pooling layer, a flatten layer, a fully connected layer and a softmax layer.

[0053] The third aspect embodiment of the present disclosure proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the fault diagnosis method for the hydropower unit proposed in the first aspect embodiment of the present disclosure is implemented.

[0054] The fourth aspect of the present disclosure provides a non-temporary computer-readable storage medium storing a computer program, which, when executed by a processor, implements the fault diagnosis method for a hydropower unit as provided in the first aspect of the present disclosure.

[0055] The fifth aspect of the present disclosure provides a computer program product. When an instruction processor in the computer program product is executed, the fault diagnosis method for a hydropower unit provided in the first aspect of the present disclosure is executed.

[0056] The fault diagnosis method, device and electronic device of the hydropower unit provided by the present disclosure can first obtain the time series data of the hydropower unit, and then process the time series data to obtain the corresponding input data, and then use the trained convolutional neural network model to process the input data to obtain the corresponding type label, and then determine the operating status of the hydropower unit according to the type label. Therefore, by using the trained convolutional neural network model and combining the time series data of the hydropower unit, the operating status of the hydropower unit can be determined more accurately, and the diagnostic accuracy of the fault diagnosis of the hydropower unit is also effectively improved.

[0057] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description or learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic flow chart of a fault diagnosis method for a hydropower unit provided by another embodiment of the present disclosure;

[0059] Figure 2 A schematic flow chart of a fault diagnosis method for a hydropower unit provided by another embodiment of the present disclosure;

[0060] Figure 2A A schematic diagram of the structure of an initial model provided by an embodiment of the present disclosure;

[0061] Figure 2B A schematic diagram of prediction labels and operation labels corresponding to a training sample sequence provided by an embodiment of the present disclosure;

[0062] Figure 3 A schematic diagram of the structure of a fault diagnosis device for a hydropower unit provided by another embodiment of the present disclosure;

[0063] Figure 4 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0064] Embodiments of the present disclosure are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, and should not be construed as limiting the present disclosure.

[0065] The following describes the fault diagnosis method, device and electronic equipment for a hydropower unit according to an embodiment of the present disclosure with reference to the accompanying drawings.

[0066] The fault diagnosis control method for a hydropower unit according to an embodiment of the present disclosure may be executed by a fault diagnosis control device for a hydropower unit provided by an embodiment of the present disclosure, and the device may be configured in an electronic device.

[0067] Figure 1 The figure is a flow chart of the fault diagnosis method of the hydropower unit provided by the embodiment of the present disclosure. Figure 1 As shown, the fault diagnosis method of the hydroelectric unit may include the following steps:

[0068] Step 101, obtaining time series data of the hydropower unit.

[0069] The time series data of the hydropower unit may be the operation data of the hydropower unit, which has the characteristics of time sequence. For example, it may be the operation data of the hydropower unit at each moment in a certain period of time, or the operation data corresponding to the hydropower unit in different periods of time, etc., which is not limited in the present disclosure.

[0070] It can be understood that the timing data of the hydropower unit, that is, the operating data with time sequence characteristics, can be obtained by collecting the original vibration signal of a certain channel through a sensor; or it can be collected by other means during the operation of the hydropower unit, etc., and the present disclosure does not limit this.

[0071] Step 102: Process the time series data to obtain corresponding input data.

[0072] There are many ways to process time series data, and this disclosure does not limit this.

[0073] Optionally, the time series data can be segmented into overlapping samples to determine multiple groups of sample sequences, and then each group of sample sequences can be subjected to complementary ensemble empirical mode decomposition (CEEMD) to obtain multiple intrinsic mode functions IMFs corresponding to each group of sample sequences, and then the input data can be determined based on the correlation coefficient between each IMF and the corresponding sample sequence.

[0074] For example, there are 100 time series data, and a sliding window with a length of 10 is used to divide the time series data with a length of 100 at a time. The overlapping part of the next sliding window and the previous sliding window accounts for 40% of the sliding window length, that is, 4 sampling points are overlapped, and the group of time series data can obtain 16 groups of sample sequences. Then, the 16 groups of sample sequences can be processed by CEEMD to obtain multiple intrinsic mode functions (IMFs) corresponding to the 16 groups of sample sequences. Then, for each group of sample sequences, the correlation coefficients between each IMF component corresponding to the group of sample sequences and the group of sample sequences can be determined respectively, and then the first M IMFs with the largest correlation coefficients can be determined as input data, for example, the first 5 IMFs can be determined as input data corresponding to the group of sample sequences, or the first 7 IMFs can be determined as input data corresponding to the group of sample sequences, etc., where M can be a value set in advance, and the present disclosure does not limit this.

[0075] It can be understood that if the first 7 IMFs corresponding to a certain sample sequence are input data, the input data may be an input matrix including 7 channels and each channel having a size of 1*10, etc.

[0076] Optionally, there may be multiple ways to determine the correlation coefficient between each IMF and the corresponding sample sequence, such as Euclidean distance, Manhattan distance, covariance formula, etc., which is not limited in the present disclosure.

[0077] It should be noted that the above examples are merely illustrative and cannot be used as limitations on the manner in which time series data is processed in the embodiments of the present disclosure.

[0078] It can be understood that in the embodiments of the present disclosure, by processing the time series data, more comprehensive and reliable input data can be obtained, thereby providing conditions for improving the accuracy of diagnosis of the hydropower unit.

[0079] Step 103: Use the trained convolutional neural network model to process the input data to obtain the corresponding type label.

[0080] The convolutional neural network model may be a model that has been trained in advance, and may include an input layer, a convolutional layer, a pooling layer, etc. In addition, it may also be a one-dimensional convolutional neural network (1D-CNN), etc., which is not limited in the present disclosure.

[0081] It can be understood that when the input data is input into the convolutional neural network model, the type label corresponding to the input data can be obtained after being processed by each network layer in the convolutional neural network model. This is not limited in the present disclosure.

[0082] In addition, the style or presentation form of the type label may be pre-set, such as 0, 1, 10, 11, etc., which is not limited in the present disclosure. Step 104: Determine the operating status of the hydropower unit according to the type label.

[0083] It is understandable that the relationship between each type of label and the operating status of the hydropower unit can be set in advance, so that after the type label is obtained, the operating status of the hydropower unit can be determined.

[0084] For example, if "01" is set in advance to indicate that the hydropower unit is running "normally", "10" indicates that the hydropower unit has an "early warning", and "11" indicates that the hydropower unit has an "error". If the input data is input into the convolutional neural network model and the determined type label is "10", it can be determined that the hydropower unit has an "early warning". Alternatively, if a certain input data is input into the convolutional neural network model and the determined type label is "11", it can be determined that the hydropower unit has an "error".

[0085] It should be noted that the above examples are only illustrative and cannot be used as limitations on the type labels in the present disclosure and their relationship with the operating conditions of the hydropower units.

[0086] It is understandable that if the hydropower unit is determined to be in the "warning" state based on the time series data of the hydropower unit in a certain period of time, the hydropower unit can be checked in time according to the "warning" situation to restore it to a normal state. Alternatively, if the hydropower unit is determined to be in the "fault" state based on the time series data of the hydropower unit in a certain period of time, the fault location and degree of each component in the hydropower unit can be discovered in time according to the "fault" situation so as to repair and replace it in time.

[0087] It can be understood that in the embodiments of the present disclosure, the diagnostic accuracy of hydropower unit fault diagnosis is effectively improved by combining the advantages of CEEMD in analyzing nonlinear and non-stationary signals with the superiority of one-dimensional convolutional neural network in processing time series and its powerful feature extraction capabilities.

[0088] In the disclosed embodiment, the time series data of the hydropower unit can be first obtained, and then the time series data can be processed to obtain corresponding input data, and then the input data can be processed using the trained convolutional neural network model to obtain the corresponding type label, and then the operating status of the hydropower unit can be determined according to the type label. Therefore, by using the trained convolutional neural network model and combining the time series data of the hydropower unit, the operating status of the hydropower unit can be determined more accurately, and the diagnostic accuracy of the fault diagnosis of the hydropower unit can be effectively improved.

[0089] Figure 2 The figure is a flow chart of the fault diagnosis method of the hydropower unit provided by the embodiment of the present disclosure. Figure 2 As shown, the fault diagnosis method of the hydroelectric unit may include the following steps:

[0090] Step 201 : obtaining R group training sample sequences of the hydropower generating units, wherein the training sample sequences include operation data of the hydropower generating units and corresponding operation labels.

[0091] Here, R can be any positive number, such as 3, 5, 100, etc., and the present disclosure does not limit this.

[0092] Optionally, T groups of time series data of the hydropower unit may be obtained first, and then overlapping samples may be segmented on the T groups of time series data to generate R groups of sample sequences.

[0093] Wherein, T can be any positive number, for example, it can be 1, 2, etc., and the present disclosure does not limit this.

[0094] For example, three groups of time series data of a single signal channel in a hydropower unit are collected by sensors, which are the operating states of the hydropower unit in normal working condition, warning working condition and fault working condition, respectively. The length of each group of time series data is 100,000 sampling points. Afterwards, the time series data of length 100,000 can be segmented using a sliding window of length 2048. For example, the overlapping part of the next sliding window and the previous sliding window can be set to account for 20% of the sliding window length, that is, 410 sampling points overlap. Thus, by performing overlapping sample segmentation on each group of time series data, 40 groups of sample sequences can be obtained for each group of time series data. Then, 120 groups of sample sequences can be obtained from 3 groups of time series data, 84 of which can be divided into training sets, 36 groups can be divided into test sets, and so on. The present disclosure does not limit this.

[0095] Alternatively, R group time series data of the hydropower unit may also be obtained, and the time series data is the corresponding training sample sequence, etc., which is not limited in the present disclosure.

[0096] It should be noted that the above examples are merely illustrative and cannot be used as limitations on the method of generating a sample sequence in the embodiments of the present disclosure.

[0097] It can be understood that in the embodiment of the present disclosure, by segmenting the acquired time series data of the hydropower unit into overlapping samples, a relatively complete and comprehensive sample sequence can be generated, thereby improving the accuracy and reliability of the sample sequence.

[0098] Step 202: Process each group of training sample sequences to generate an input matrix corresponding to each group of training sample sequences.

[0099] Optionally, CEEMD processing may be performed on each group of training sample sequences to generate an input matrix corresponding to the group of training sample sequences.

[0100] Optionally, paired positive and negative white noises with a mean of zero can be introduced into each group of training sample sequences to obtain a first sequence after the positive white noise is introduced into the training sample sequence and a second sequence after the negative white noise is introduced into the training sample sequence. Then, the first sequence and the second sequence corresponding to each group of training sample sequences can be subjected to empirical mode decomposition (EMD) to obtain each IMF component corresponding to each group of training sample sequences. Then, according to the correlation coefficient between each IMF component and the corresponding training sample sequence, the top M IMFs with the largest correlation coefficients in each group of training sample sequences can be determined as the corresponding input matrix.

[0101] For example, n pairs of positive and negative white noises with a mean of zero can be introduced into any training sample sequence to obtain a first sequence of the training sample sequence after the positive white noise is introduced and a second sequence of the training sample sequence after the negative white noise is introduced, which can be expressed in the form described in the following formula (1):

[0102]

[0103] Among them, f1(t) is the first sequence after adding positive white noise, f2(t) is the second sequence after adding negative white noise, y(t) is any training sample sequence, and g(t) is the white noise sequence.

[0104] Afterwards, each sequence can be processed by EMD to obtain a series of IMF components, where the jth IMF component of the i-th sequence can be: ij For example, by performing EMD processing on f1(t) and f2(t), a series of IMF components can be obtained.

[0105] Then, the multi-component weights can be averaged, which can satisfy the form shown in the following formula (2):

[0106]

[0107] Among them, IMF j represents the jth IMF component.

[0108] Afterwards, the correlation coefficient between each IMF component and any training sample sequence y(t) can be determined, and then the first M IMF components with the largest correlation coefficients can be determined as the input data corresponding to y(t).

[0109] Here, M may be a pre-set value, such as 3, 5, etc., which is not limited in the present disclosure.

[0110] In addition, the correlation coefficient can be determined by the relationship shown in the following formula (3):

[0111]

[0112] Among them, r(IMF j ,y(t)) can be the correlation coefficient between the jth IMF component and the training sample sequence, Cov(IMF j ,y(t)) can be the covariance between the jth IMF component and the sample sequence, Var(IMF j ) can be the variance of the j-th IMF component, and Var(y(t)) can be the variance of the training sample sequence.

[0113] Step 203: Process the input matrix using the initial model to obtain the predicted label corresponding to the input matrix.

[0114] Among them, the initial model can be a model set in advance. By inputting the input matrix into the initial model and processing it by the initial model, the corresponding prediction label can be obtained, and the present disclosure does not limit this.

[0115] Optionally, the initial model may include a convolutional layer, a pooling layer, a flatten layer, a fully connected layer, etc., which is not limited in the present disclosure.

[0116] In addition, the initial model may include one convolutional layer or multiple convolutional layers, or may also include one pooling layer or multiple pooling layers, etc., which is not limited in the present disclosure.

[0117] For example, the structure of the initial model may be: input layer L1 - convolution layer L2 - pooling layer L3 - convolution layer L4 - pooling layer L5 - convolution layer L6 - pooling layer L7 - flatten layer L8 - fully connected layer L9 - output layer L10, etc. For example, the output layer may be a softmax layer, etc. This disclosure does not limit this.

[0118] In addition, if the input matrix size is an IMF matrix of M*1*N, it can be input as input data into the input layer of the one-dimensional convolutional neural network.

[0119] Optionally, if the data is input to the convolution layer, the convolution operation can be performed first, and then the output data can be obtained by the nonlinear operation of the activation function, which can be shown in the following formula (4):

[0120]

[0121] Among them, L can represent the number of layers of the convolutional neural network, G L can be the Lth convolutional layer, A L-1 is the output of the L-1th layer, is the convolution operator, W L The convolution kernel weight matrix for the Lth layer convolution operation, b L is the bias vector, and f is the ReLu activation function.

[0122] Among them, the ReLu activation function expression can be shown as the following formula (5):

[0123]

[0124] Among them, x is the input value of the activation function, corresponding to It is understandable that the activation function f may also be other forms of activation functions, etc., which is not limited in the present disclosure.

[0125] Optionally, each convolution layer can be set with K groups of convolution kernels, the step size of the convolution operation can be set to 1, and the input features of the convolution layer can be filled with all 0s, so that the dimensions of the input data and output data of the convolution layer can be the same, and then the convolution operation can be performed from top to bottom according to the vertical coordinates of the input features of the convolution layer.

[0126] It is understandable that, while ensuring that the dimensions of the input data and output data of the convolution layer are consistent, the step size of the convolution operation can also be set to any other feasible value, or the input features of the base layer can be replaced by other padding, etc., and the present disclosure does not limit this.

[0127] Optionally, the pooling layer may be maximum pooling, minimum pooling, average pooling, etc., which is not limited in the present disclosure.

[0128] For example, the pooling layer may be a maximum pooling layer, and the step size of the filter of the pooling layer may be set to 4, so that the output features of the previous convolutional layer connected to the pooling layer can be reduced in dimension and compressed, etc. The step size may also be other values, etc., which is not limited in the present disclosure.

[0129] In addition, the output data of the pooling layer L7 can be converted into a one-dimensional data matrix through the flatten layer, which is not limited in the present disclosure.

[0130] In addition, the softmax layer can map the output of the fully connected layer to the interval (0, 1), which can satisfy the relationship shown in the following formula (6):

[0131]

[0132] Among them, S i is the softmax mapping value corresponding to the i-th type label, j represents the j-th type label, and T is the total number of type labels, which can correspond to T groups of different fault categories, thereby realizing the classification of samples.

[0133] For example, the size of the input matrix is ​​7*1*2048, the convolution layer is set to 64 convolution kernels, the step size of the convolution operation is 1, and all 0s are filled. The pooling layer uses maximum pooling, and the step size of the filter of the pooling layer is 4. If the structure of the initial model is as follows Figure 2A As shown, the 7*1*2048 IMF component is input to the input layer L1 in the initial model, and then processed by the convolution layer L2, the size of the output data is: 64*1*2048; after processing by the pooling layer L3, the size of the output data is: 64*1*512; after processing by the convolution layer L4, the size of the output data is 64*1*512; after processing by the pooling layer L5, the size of the output data is 64*1*128; after processing by the convolution layer L6, the size of the output data is 64*1*128 ; After being processed by the pooling layer L7, the size of the output data is: 64*1*32; after being processed by the Flatten layer L8, the output data of the pooling layer L7 can be converted into a one-dimensional data matrix with a length of 2048; after being processed by the fully connected layer L9, the one-dimensional data matrix can be input into the softmax layer L10. After being processed by the softmax layer L10, the output of the fully connected layer L9 can be mapped to the (0,1) interval, so that the corresponding type label can be obtained to realize the classification of the sample.

[0134] It should be noted that the above examples are merely illustrative and cannot be used as limitations on the structure of the initial model in the embodiments of the present disclosure.

[0135] Step 204 , based on the difference between the predicted label and the running label corresponding to each input matrix, determine the corresponding loss value.

[0136] It can be understood that the prediction label and the running label corresponding to any training sample are also the prediction label and the running label corresponding to the input matrix corresponding to the training sample.

[0137] For example, the actual running labels corresponding to the training sample sequence and the predicted labels generated by the initial model can be as follows Figure 2B As shown in the figure, o is the actual running label and * is the predicted label.

[0138] Depend on Figure 2B It can be seen that for the training sample sequence with serial number "2", the operating status corresponding to the predicted label obtained after processing by the initial model is the "warning" category, while the operating status corresponding to the actual operating label is the "normal" category. There is a difference between the two. Based on this difference, the initial model can be corrected to improve the performance of the initial model.

[0139] It should be noted that the above examples are merely illustrative and cannot be used as limitations on the training sample sequences, prediction labels, operation labels, etc. in the embodiments of the present disclosure.

[0140] Optionally, the difference between the predicted label and the running label corresponding to each input matrix can be determined first, and then the corresponding loss value can be determined based on the square loss function.

[0141] For example, when determining the loss value corresponding to any training sample, the relationship shown in the following formula (7) can be used:

[0142]

[0143] Among them, L(x) is the square loss function, u i is the predicted label of the i-th training sample sequence, Y i is the expected output value corresponding to the i-th training sample sequence, that is, the actual running label, R i is the total number of training sample sequences, that is, the total number of input matrices. For example, there are 84 sets of training sample sequences, then R i The value of is 84. This disclosure does not limit this.

[0144] Optionally, the initial model can be trained and optimized based on the Adagrad gradient descent method, where the Adagrad gradient descent method can satisfy the relationship shown in the following formula (8):

[0145]

[0146] Among them, W t represents the updated weight, W t-1 is the weight before update, g t-1 is the derivative of the square loss function before updating, η is the learning rate, and e is a very small constant.

[0147] Step 205: Based on the loss value, reverse correction is performed on the initial model to generate a trained convolutional neural network model.

[0148] It is understandable that when the difference between the predicted label generated by the initial model and the actual operating label does not meet the model accuracy requirements, or the model training times do not reach the training times, the initial model can be reversely corrected according to the loss value of each training sample sequence to enhance the learning ability of the initial model, thereby generating a convolutional neural network model with strong learning ability, so that when the convolutional neural network model is used to process the time series data of the hydropower unit, a more accurate type label can be obtained, so that the operating status of the hydropower unit can be accurately judged, thereby improving the accuracy of fault diagnosis of the hydropower unit.

[0149] It is understandable that the structure of the convolutional neural network model generated after training can be the same as the initial model, for example, Figure 2A The network structure shown may also be other forms of network structures, etc., which are not limited in the present disclosure.

[0150] In the disclosed embodiment, the R group training sample sequences of the hydropower unit can be first obtained, and then each group of training sample sequences can be processed to generate an input matrix corresponding to each group of training sample sequences, and then the input matrix can be processed using the initial model to obtain the prediction label corresponding to the input matrix, and then the corresponding loss value can be determined based on the difference between the prediction label and the operation label corresponding to each input matrix, and then the initial model can be reversely corrected based on the loss value to generate a trained convolutional neural network model. Thus, by correcting the initial model through the difference between the prediction label and the operation label corresponding to the training sample sequence, a convolutional neural network model with better performance can be generated, and then the convolutional neural network model generated by the training can be used to predict the operating status of the hydropower unit, thereby providing conditions for improving the diagnostic accuracy of the fault diagnosis of the hydropower unit.

[0151] In order to implement the above embodiment, the present disclosure also proposes a fault diagnosis device for a hydropower unit.

[0152] Figure 3 A schematic diagram of the structure of a fault diagnosis device for a hydropower unit provided in an embodiment of the present disclosure.

[0153] like Figure 3 As shown, the fault diagnosis device 100 of the hydroelectric unit may include: an acquisition module 110 , a first processing module 120 , a second processing module 130 and a determination module 140 .

[0154] The acquisition module 110 is used to acquire the time series data of the hydropower unit.

[0155] The first processing module 120 is used to process the time series data to obtain corresponding input data.

[0156] The second processing module 130 is used to process the input data using the trained convolutional neural network model to obtain a corresponding type label.

[0157] The determination module 140 is used to determine the operating status of the hydropower unit according to the type label.

[0158] Optionally, the first processing module 120 is specifically configured to:

[0159] Perform overlapping sample segmentation on the time series data to determine multiple groups of sample sequences;

[0160] Perform complementary set empirical mode decomposition (CEEMD) on each group of sample sequences to obtain each intrinsic mode function (IMF) corresponding to each group of sample sequences;

[0161] Input data is determined according to the correlation coefficient between each of the IMFs and the corresponding sample sequence.

[0162] Optionally, the acquisition module 110 is further used to acquire R groups of training sample sequences of the hydropower unit, wherein the training sample sequences include operation data of the hydropower unit and corresponding operation labels, and R is an arbitrary positive number.

[0163] The first processing module 120 is further configured to process each group of the training sample sequences to generate an input matrix corresponding to each group of the training sample sequences.

[0164] The second processing module 130 is further configured to process the input matrix using the initial model to obtain a predicted label corresponding to the input matrix.

[0165] The determination module 140 is further configured to determine a corresponding loss value based on a difference between a predicted label corresponding to each of the input matrices and the running label;

[0166] The device further includes a generating module, which is used for:

[0167] Based on the loss value, the initial model is reversely modified to generate a trained convolutional neural network model.

[0168] Optionally, the acquisition module 110 is specifically configured to:

[0169] Obtaining T groups of time series data of the hydropower unit, where T is any positive number;

[0170] The T groups of time series data are segmented into overlapping samples to generate R groups of sample sequences.

[0171] Optionally, the first processing module 120 is specifically configured to:

[0172] Introducing paired positive and negative white noises with a mean of zero into each group of the training sample sequences to obtain a first sequence of the training sample sequences after the positive white noise is introduced and a second sequence of the training sample sequences after the negative white noise is introduced;

[0173] Performing empirical mode decomposition (EMD) on the first sequence and the second sequence corresponding to each group of the training sample sequences to obtain corresponding IMF components;

[0174] According to the correlation coefficient between each of the IMF components and the corresponding training sample sequence, the first M IMF components with the largest correlation coefficient in each group of the training sample sequences are determined as the corresponding input matrix.

[0175] Optionally, the second processing module 130 is specifically configured to:

[0176] The input matrix is ​​input into the initial model so that the corresponding prediction label is obtained after being processed by the initial model, wherein the initial model includes an input layer, a convolution layer, a pooling layer, a flatten layer, a fully connected layer and a softmax layer.

[0177] The functions and specific implementation principles of the above modules in the embodiments of the present disclosure can be referred to the above method embodiments, and will not be repeated here.

[0178] The fault diagnosis device for a hydroelectric unit provided by the embodiment of the present disclosure can first obtain the time series data of the hydroelectric unit, and then process the time series data to obtain corresponding input data, and then use the trained convolutional neural network model to process the input data to obtain the corresponding type label, and then determine the operating status of the hydroelectric unit according to the type label. Thus, by using the trained convolutional neural network model and combining the time series data of the hydroelectric unit, the operating status of the hydroelectric unit can be determined more accurately, thereby effectively improving the diagnostic accuracy of the fault diagnosis of the hydroelectric unit.

[0179] In order to implement the above embodiments, the present disclosure also proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the fault diagnosis method for the hydropower unit proposed in the above embodiments of the present disclosure is implemented.

[0180] In order to implement the above embodiments, the present disclosure further proposes a non-temporary computer-readable storage medium storing a computer program, which, when executed by a processor, implements the fault diagnosis method for a hydropower unit proposed in the above embodiments of the present disclosure.

[0181] In order to implement the above embodiments, the present disclosure further proposes a computer program product. When the instruction processor in the computer program product is executed, the fault diagnosis method of the hydropower unit proposed in the above embodiments of the present disclosure is executed.

[0182] Figure 4 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present disclosure is shown. Figure 4 The electronic device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0183] like Figure 4 As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0184] The bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnection (PCI) bus.

[0185] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0186] The memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 4 not shown, usually called a "hard drive"). Although Figure 4 Not shown in the figure, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the various embodiments of the present disclosure.

[0187] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described in the present disclosure.

[0188] The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 12, and / or may communicate with any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the electronic device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0189] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the methods mentioned in the above embodiments.

[0190] According to the technical solution of the embodiment of the present disclosure, the time series data of the hydropower unit can be obtained first, and then the time series data can be processed to obtain the corresponding input data. Then, the input data can be processed using the trained convolutional neural network model to obtain the corresponding type label, and then the operating status of the hydropower unit can be determined according to the type label. Therefore, by using the trained convolutional neural network model and combining the time series data of the hydropower unit, the operating status of the hydropower unit can be determined more accurately, and the diagnostic accuracy of the fault diagnosis of the hydropower unit can be effectively improved.

[0191] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0192] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, "plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0193] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present disclosure includes additional implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present disclosure belong.

[0194] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0195] It should be understood that the various parts of the present disclosure can be implemented in hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0196] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0197] In addition, each functional unit in each embodiment of the present disclosure may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0198] The storage medium mentioned above may be a read-only memory, a disk or an optical disk, etc. Although the embodiments of the present disclosure have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present disclosure. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present disclosure.

Claims

1. A fault diagnosis method for a hydropower unit, characterized in that: include: Obtain time series data of hydropower units; Processing the time series data to obtain corresponding input data; Processing the input data using the trained convolutional neural network model to obtain corresponding type labels; Determining the operating status of the hydropower unit according to the type label; The method for training the convolutional neural network model includes: Obtain R group training sample sequences of the hydropower unit, wherein the training sample sequences include operation data of the hydropower unit and corresponding operation labels, and R is an arbitrary positive number; Introduce paired positive and negative white noises with a mean of zero into each group of the training sample sequences to obtain a first sequence after the positive white noise is introduced into the training sample sequences and a second sequence after the negative white noise is introduced into the training sample sequences, perform empirical mode decomposition (EMD) on the first sequence and the second sequence corresponding to each group of the training sample sequences to obtain corresponding IMF components, and determine the first M IMF components with the largest correlation coefficients between the IMF components in each group of the training sample sequences and the corresponding training sample sequences as the input matrix corresponding to each group of the training sample sequences; wherein: The following formula is used to express the first sequence after the positive white noise is introduced into the training sample sequence and the second sequence after the negative white noise is introduced into the training sample sequence: Where, f1(t) is the first sequence after adding positive white noise, f2(t) is the second sequence after adding negative white noise, y(t) is any training sample sequence, and g(t) is the white noise sequence; The correlation coefficient is determined by the following formula: Among them, r(IMF j ,y(t)) is the correlation coefficient between the jth IMF component and the training sample sequence, Cov(IMF j ,y(t)) is the covariance between the jth IMF component and the sample sequence, Var(IMF j ) is the variance of the jth IMF component, Var(y(t)) is the variance of the training sample sequence; The input matrix is ​​processed using the initial model to obtain the predicted label corresponding to the input matrix, wherein the initial model includes a fully connected layer and a softmax layer, and the softmax layer maps the output of the fully connected layer to the interval (0, 1), which needs to satisfy the relationship shown in the following formula: Among them, S i is the softmax mapping value corresponding to the i-th type label, j represents the j-th type label, T is the total number of type labels, which corresponds to T groups of different fault categories; Based on the predicted label corresponding to each of the input matrices, the running label and the square loss function, the corresponding loss value is determined, wherein the expression of the square loss function is: Among them, L(x) is the square loss function, u i is the predicted label of the i-th training sample sequence, Y i is the running label corresponding to the i-th training sample sequence, R i is the total number of training sample sequences; Based on the loss value, the initial model is reversely modified to generate a trained convolutional neural network model.

2. The method according to claim 1, characterized in that The processing of the time series data to obtain corresponding input data includes: Perform overlapping sample segmentation on the time series data to determine multiple groups of sample sequences; Perform complementary set empirical mode decomposition (CEEMD) on each group of sample sequences to obtain each intrinsic mode function (IMF) corresponding to each group of sample sequences; Input data is determined according to the correlation coefficient between each of the IMFs and the corresponding sample sequence.

3. The method according to claim 1, characterized in that The step of obtaining a training sample sequence of a group R of hydropower generating units comprises: Obtaining T groups of time series data of the hydropower unit, where T is any positive number; The T groups of time series data are segmented into overlapping samples to generate R groups of sample sequences.

4. The method according to claim 1, characterized in that The using the initial model to process the input matrix to obtain a prediction label corresponding to the input matrix includes: The input matrix is ​​input into the initial model so that the corresponding prediction label is obtained after being processed by the initial model, wherein the initial model also includes an input layer, a convolution layer, a pooling layer and a flatten layer.

5. A fault diagnosis device for a hydroelectric unit, characterized in that: include: An acquisition module, used to acquire the time series data of the hydropower unit; A first processing module, used for processing the time series data to obtain corresponding input data; A second processing module is used to process the input data using the trained convolutional neural network model to obtain a corresponding type label; A determination module, used to determine the operating status of the hydropower unit according to the type label; The method for training the convolutional neural network model includes: Obtain R group training sample sequences of the hydropower unit, wherein the training sample sequences include operation data of the hydropower unit and corresponding operation labels, and R is an arbitrary positive number; Introduce paired positive and negative white noises with a mean of zero into each group of the training sample sequences to obtain a first sequence after the positive white noise is introduced into the training sample sequences and a second sequence after the negative white noise is introduced into the training sample sequences, perform empirical mode decomposition (EMD) on the first sequence and the second sequence corresponding to each group of the training sample sequences to obtain corresponding IMF components, and determine the first M IMF components with the largest correlation coefficients between the IMF components in each group of the training sample sequences and the corresponding training sample sequences as the input matrix corresponding to each group of the training sample sequences; wherein: The following formula is used to express the first sequence after the positive white noise is introduced into the training sample sequence and the second sequence after the negative white noise is introduced into the training sample sequence: Where, f1(t) is the first sequence after adding positive white noise, f2(t) is the second sequence after adding negative white noise, y(t) is any training sample sequence, and g(t) is the white noise sequence; The correlation coefficient is determined by the following formula: Among them, r(IMF j ,y(t)) is the correlation coefficient between the jth IMF component and the training sample sequence, Cov(IMF j ,y(t)) is the covariance between the jth IMF component and the sample sequence, Var(IMF j ) is the variance of the jth IMF component, Var(y(t)) is the variance of the training sample sequence; The input matrix is ​​processed using the initial model to obtain the predicted label corresponding to the input matrix, wherein the initial model includes a fully connected layer and a softmax layer, and the softmax layer maps the output of the fully connected layer to the interval (0, 1), which needs to satisfy the relationship shown in the following formula: Among them, S i is the softmax mapping value corresponding to the i-th type label, j represents the j-th type label, T is the total number of type labels, which corresponds to T groups of different fault categories; Based on the predicted label corresponding to each of the input matrices, the running label and the square loss function, the corresponding loss value is determined, wherein the expression of the square loss function is: Among them, L(x) is the square loss function, u i is the predicted label of the i-th training sample sequence, Y i is the running label corresponding to the i-th training sample sequence, R i is the total number of training sample sequences; Based on the loss value, the initial model is reversely modified to generate a trained convolutional neural network model.

6. The device according to claim 5, characterized in that The first processing module is specifically used for: Perform overlapping sample segmentation on the time series data to determine multiple groups of sample sequences; Perform complementary set empirical mode decomposition (CEEMD) on each group of sample sequences to obtain each intrinsic mode function (IMF) corresponding to each group of sample sequences; Input data is determined according to the correlation coefficient between each of the IMFs and the corresponding sample sequence.

7. An electronic device, characterized in that: include: processor; a memory for storing executable instructions for the processor; The processor is configured to call and execute the executable instructions stored in the memory to implement the method according to any one of claims 1 to 4.

8. A non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the method as claimed in any one of claims 1 to 4.