A method and system for restoring corrosion detection signals of an oil storage tank wall

Abstract

The present invention belongs to the fields of fault diagnosis and artificial intelligence, and discloses a method and system for restoring a tank wall corrosion detection signal of an oil storage tank, including the following steps: obtaining the tank wall corrosion detection signal to be restored of the oil storage tank to be detected, and dividing the tank wall corrosion detection signal to be restored into a plurality of tank wall corrosion detection sub-signals to be restored; sequentially inputting the plurality of tank wall corrosion detection sub-signals to be restored into a preset abnormal signal restoration model to obtain a plurality of tank wall corrosion detection restoration signals; wherein, the abnormal signal restoration model is obtained by training a preset denoising autoencoder model based on a storage module with the tank wall corrosion detection signal of a healthy oil storage tank as training data and the tank wall corrosion detection signal of the oil storage tank to be detected as test data; integrating the plurality of tank wall corrosion detection restoration signals to obtain the tank wall corrosion detection signal after restoring the abnormal signal of the oil storage tank to be detected. It can effectively prevent the restoration of the original abnormal signal and improve the accuracy of restoring the tank wall corrosion detection signal.

Description

Technical Field

[0001] The present invention belongs to the fields of fault diagnosis and artificial intelligence, and relates to a method and system for restoring corrosion detection signals of a storage tank wall. Background Art

[0002] With the rapid development of the economy, the demand for energy is also growing rapidly. One of the most important energy sources is oil, which is a recognized dangerous good. If an oil leak occurs, it will cause great harm to the environment on which we depend for survival and also result in huge economic losses. The safe storage and transportation of oil require very strict conditions. Storage tanks are one of the most common storage and transportation facilities. A storage tank includes multiple parts such as a tank wall, a tank bottom, and a floating roof. The tank wall is easily corroded under the daily immersion of oil, which may lead to oil leakage problems. At present, the commonly used non-destructive testing method for the tank wall is the magnetic flux leakage testing method. Under the normal use of the storage tank, the magnetic flux leakage signals of the tank wall are collected, and the tank wall is judged whether there is corrosion by analyzing the magnetic flux leakage signals. Therefore, the accuracy of the magnetic flux leakage signals directly affects the corrosion detection results of the tank wall. However, in actual measurement, due to reasons such as human operation, the collected magnetic flux leakage signals often have abnormal signals such as oscillations and mutations. The existence of these abnormal signals seriously affects the detection accuracy of the tank wall corrosion signals. Therefore, the processing of the magnetic flux leakage abnormal signals of the tank wall is crucial. Commonly used processing methods include statistical-based methods and clustering-based methods, etc.

[0003] Among them, there are two statistical-based processing methods: the 3Sigma criterion and the Histogram-based OutlierScore. Among them, the 3Sigma criterion is a data processing method, and the specific steps are as follows: (1) Determine that only random errors appear in the sample data; (2) Calculate the data to obtain the standard deviation; (3) Use the probability value as the boundary for setting the interval; (4) Judge whether the error is within this interval; (5) The error belonging to this interval is a random error, and the corresponding sample data is retained, and the error not belonging to this interval is a gross error, and the corresponding sample data is excluded. The 3Sigma criterion has corresponding limitations, which are specifically as follows: (1) The sample data needs to satisfy a normal distribution or be close to a normal distribution; (2) A large number of measurement times are required; (3) Insufficient measurement times will lead to inaccurate measurement.

[0004] The processing methods based on clustering methods include K-Means clustering and DBSCAN smiling face clustering. Among them, K-Means clustering is a clustering analysis algorithm that solves problems based on iteration. The specific steps are as follows: (1) Group the data and determine the clustering centers; (2) Calculate the distances between each object and the clustering centers; (3) Each object belongs to the clustering center closest to it; (4) Thus, a clustering is formed, which consists of the objects and the clustering center closest to it; (5) According to the different sample data, the clustering centers will be repeatedly obtained; (6) Stop the iteration when the termination condition is met. K-Means clustering has certain limitations, specifically: (1) It is very difficult to determine the value of k; (2) If two categories are too close, it is extremely easy to reduce the accuracy of the results; (3) The final result is not the global optimal solution. DBSCAN smiling face clustering is a density-based clustering algorithm. The steps are as follows: (1) Determine two values, the scanning radius and the minimum number of points included; (2) Randomly start from an unprocessed point and determine all the points whose distances from it are less than the scanning radius; (3) Judge the numerical size relationship between the number of all points within the scanning radius and the minimum number of points included; (4) If the number of points is less than the minimum number of points included, the starting point is a noise point; if the former is greater than or equal to the latter, the points within the scanning radius form a cluster, and the cluster is further expanded according to recursion. DBSCAN smiling face clustering has certain disadvantages, specifically as follows: (1) The effect of this method applied to high-dimensional data is not ideal; (2) The corresponding density change of the data set cannot be well reflected; (3) Uneven data will lead to poor clustering results. Summary of the Invention

[0005] The object of the present invention is to overcome the above-mentioned disadvantages of the prior art and provide a method, system, device and medium for restoring the corrosion detection signal of the tank wall of an oil storage tank.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In the first aspect of the present invention, a method for restoring the corrosion detection signal of the tank wall of an oil storage tank includes the following steps:

[0008] Obtain the tank wall corrosion detection signal to be restored of the oil storage tank to be detected, and divide the tank wall corrosion detection signal to be restored into several tank wall corrosion detection sub-signals to be restored;

[0009] Input several tank wall corrosion detection sub-signals to be restored into a preset abnormal signal restoration model in sequence to obtain several tank wall corrosion detection restoration signals; among them, the abnormal signal restoration model is obtained by training a preset noise reduction autoencoder model based on a storage module with the tank wall corrosion detection signal of a healthy oil storage tank as training data and the tank wall corrosion detection signal of the oil storage tank to be detected as test data.

[0010] Integrate several tank wall corrosion detection recovery signals to obtain the tank wall corrosion detection signal of the storage tank to be detected after the recovery abnormal signal.

[0011] A further improvement of the method for recovering the tank wall corrosion detection signal of the storage tank of the present invention lies in:

[0012] The specific method for dividing the tank wall corrosion detection signal to be recovered into several tank wall corrosion detection sub-signals to be recovered is: through a sliding window with a preset size and a preset sliding step size, divide the tank wall corrosion detection signal to be recovered into several tank wall corrosion detection sub-signals to be recovered.

[0013] Before dividing the tank wall corrosion detection signal to be recovered into several tank wall corrosion detection sub-signals to be recovered, perform baseline correction on the tank wall corrosion detection signal to be recovered.

[0014] The noise reduction autoencoder model based on the storage module includes an autoencoder, a storage module, and a noise reduction autoencoder;

[0015] Among them, the storage module is used to provide storage entry data for the autoencoder and the noise reduction autoencoder;

[0016] The autoencoder is used to obtain the features of the autoencoder input data, obtain the first feature vector, and reconstruct the autoencoder input data according to the storage entry data and the second feature vector;

[0017] The noise reduction autoencoder is used to obtain the features of the noise reduction autoencoder input data, obtain the second feature vector, and reconstruct the noise reduction autoencoder input data according to the second feature vector and the storage entry data.

[0018] The loss function Loss of the noise reduction autoencoder model based on the storage module is:

[0019]

[0020] Among them, L AE is the autoencoder loss function, is the autoencoder output data, d i is the autoencoder input data; L DAE is the noise reduction autoencoder loss function, is the noise reduction autoencoder output data, is the noise reduction autoencoder input data, and α is a hyperparameter.

[0021] The method for training a preset noise reduction autoencoder model based on the storage module with the tank wall corrosion detection signal of a healthy storage tank as training data and the tank wall corrosion detection signal of the storage tank to be detected as test data is:

[0022] After performing baseline correction on the tank wall corrosion detection signal of a healthy storage tank, it is segmented into several first tank wall corrosion detection sub-signals, and a preset abnormal type is added to each first tank wall corrosion detection sub-signal to obtain several abnormal signals of each first tank wall corrosion detection sub-signal. The several abnormal signals of each first tank wall corrosion detection sub-signal and each first tank wall corrosion detection sub-signal are combined into a group in sequence to obtain several training samples;

[0023] After performing baseline correction on the tank wall corrosion detection signal of the storage tank to be detected, it is segmented into several second tank wall corrosion detection sub-signals to obtain several test samples;

[0024] Perform a training step, and the training step includes: training a preset denoising autoencoder model based on a storage module through several training samples, and testing the trained denoising autoencoder model based on the storage module through test samples to obtain a test result. When the test result meets the preset test conditions, the currently trained denoising autoencoder model based on the storage module is used as an abnormal signal recovery model; otherwise, repeat the training step.

[0025] The preset abnormal types include adding white noise, channel signal loss, and block signal loss.

[0026] The specific method for training the preset denoising autoencoder model based on the storage module is:

[0027] Initialize the autoencoder parameters of the denoising autoencoder model based on the storage module, the stored entry data in the storage module, and the denoising autoencoder parameters;

[0028] Input the first tank wall corrosion detection sub-signal in the training sample into the encoder of the autoencoder to obtain the first feature vector γ 1 , and calculate the cosine similarity between the first feature vector γ 1 and each stored entry data in the storage module through the following formula:

[0029]

[0030] where ξ k is the k-th stored entry data, v k is the cosine similarity between the first feature vector γ 1 and the k-th stored entry data in the storage module, and m is the number of stored entry data;

[0031] Normalize the cosine similarity between the first feature vector γ 1 and each stored entry data in the storage module to obtain the normalized cosine similarity v = {v' 1 , v' 2 ,..., v' m};

[0032] Obtain the first reconstructed feature information γ through the following formula 2 :

[0033] γ 2 = v' 1 ξ 1 + v' 2 ξ 2 +... + v' m ξ m

[0034] Input the first reconstructed feature information γ 2 into the decoder of the autoencoder to obtain the reconstructed signal of the first tank wall corrosion detection sub-signal;

[0035] Input the abnormal signal of the first tank wall corrosion detection sub-signal in the training sample into the encoder of the denoising autoencoder to obtain the second feature vector Calculate the cosine similarity between the second feature vector and the data of each storage entry in the storage module through the following formula:

[0036]

[0037] where, w k is the cosine similarity between the first feature vector and the data of the k-th storage entry in the storage module;

[0038] Normalize the cosine similarities between the second feature vector and the data of each storage entry in the storage module to obtain the normalized cosine similarity w = {w' 1 , w' 2 ,..., w' m};

[0039] Obtain the second reconstructed feature information through the following formula

[0040]

[0041] Input the second reconstructed feature information into the decoder of the denoising autoencoder to obtain the reconstructed signal of the abnormal signal of the first tank wall corrosion detection sub-signal;

[0042] The autoencoder and the denoising autoencoder perform backpropagation simultaneously. Among them, when the autoencoder performs backpropagation, aiming to minimize the loss function Loss of the denoising autoencoder model based on the storage module, it updates the autoencoder parameters and the stored entry data in the storage module according to the first tank wall corrosion detection sub-signal and the reconstructed signal of the first tank wall corrosion detection sub-signal; when the denoising autoencoder performs backpropagation, aiming to minimize the loss function Loss of the denoising autoencoder model based on the storage module, it updates the denoising autoencoder parameters according to the abnormal signal of the first tank wall corrosion detection sub-signal and the reconstructed signal of the abnormal signal of the first tank wall corrosion detection sub-signal.

[0043] When updating the autoencoder parameters and the stored entry data in the storage module, and updating the denoising autoencoder parameters, the stochastic gradient descent method is used for updating.

[0044] In the second aspect of the present invention, a system for recovering a tank wall corrosion detection signal of an oil storage tank includes:

[0045] An acquisition module, configured to acquire the tank wall corrosion detection signal to be recovered of the oil storage tank to be detected, and divide the tank wall corrosion detection signal to be recovered into several tank wall corrosion detection sub-signals to be recovered;

[0046] A recovery module, configured to sequentially input several tank wall corrosion detection sub-signals to be recovered into a preset abnormal signal recovery model to obtain several tank wall corrosion detection recovery signals; wherein, the abnormal signal recovery model is obtained by training a preset denoising autoencoder model based on the storage module with the tank wall corrosion detection signal of a healthy oil storage tank as training data and the tank wall corrosion detection signal of the oil storage tank to be detected as test data;

[0047] An integration module, configured to integrate several tank wall corrosion detection recovery signals to obtain the tank wall corrosion detection signal of the oil storage tank to be detected after recovering the abnormal signal.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The method for restoring the corrosion detection signal of the storage tank wall of the present invention uses the corrosion detection signal of the healthy storage tank wall as training data and the corrosion detection signal of the storage tank wall to be detected as test data to train a preset denoising autoencoder model based on the storage module to obtain an abnormal signal restoration model. A number of sub-signals of the corrosion detection of the tank wall to be restored are sequentially restored into a number of restored signals of the corrosion detection of the tank wall, completing the reconstruction of the sub-signals of the corrosion detection of the tank wall to be restored, and realizing the restoration of the abnormal signal of the corrosion detection signal of the storage tank wall to be detected. Among them, a storage module for storing the characteristics of normal signals is added to the denoising autoencoder model based on the storage module. The stored entry data in the storage module is obtained by training a convolutional autoencoder. After the denoising autoencoder extracts the features through the encoding part, the corresponding stored entry data is selected from the storage module to reconstruct the features, and the reconstructed features are input into the decoder to complete the reconstruction. By reconstructing the stored entry data in the storage module, the generalization of the denoising autoencoder is reduced, preventing the restoration of the original abnormal signal and improving the accuracy of signal restoration. At the same time, this method is also an unsupervised learning method, and the training process does not require actual abnormal samples, solving the problem of few training samples of abnormal signals. Brief Description of the Drawings

[0050] Figure 1 It is a flowchart of the method for restoring the corrosion detection signal of the storage tank wall of the present invention;

[0051] Figure 2 It is a schematic structural diagram of the denoising autoencoder model based on the storage module of the present invention. Detailed Embodiments

[0052] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0054] The present invention will be further described in detail below with reference to the accompanying drawings:

[0055] See Figure 1 , in an embodiment of the present invention, a method for restoring the corrosion detection signal of the tank wall of an oil storage tank is provided, which greatly improves the accuracy of restoring the corrosion detection signal of the tank wall of the oil storage tank. Specifically, the method for restoring the corrosion detection signal of the tank wall of the oil storage tank includes the following steps.

[0056] S1: Obtain the to-be-restored corrosion detection signal of the tank wall of the to-be-detected oil storage tank, and divide the to-be-restored corrosion detection signal of the tank wall into several to-be-restored corrosion detection sub-signals.

[0057] In this embodiment, preferably, the specific method for dividing the to-be-restored corrosion detection signal of the tank wall into several to-be-restored corrosion detection sub-signals is: through a sliding window with a preset size, at a preset sliding step size, divide the to-be-restored corrosion detection signal of the tank wall into several to-be-restored corrosion detection sub-signals. Among them, the size of the preset sliding window and the preset sliding step size can be set according to actual needs.

[0058] In this embodiment, preferably, before dividing the to-be-restored corrosion detection signal of the tank wall into several to-be-restored corrosion detection sub-signals, perform baseline correction on the to-be-restored corrosion detection signal of the tank wall. By performing baseline correction on the to-be-restored corrosion detection signal of the tank wall, it is ensured that the signal base values at different positions of different sensors are the same.

[0059] Specifically, the baseline correction is performed by the following formula:

[0060]

[0061] where: L is the number of channels of the signal; k is the number of signal counting points; x ij is the original value of the j-th channel at the i-th counting point position; x ij ′ is the corrected value of the j-th channel at the i-th counting point position; s is the base value.

[0062] S2: Input a number of sub-signals of tank wall corrosion detection to be restored into a preset abnormal signal restoration model in sequence to obtain a number of restored signals of tank wall corrosion detection. Among them, the abnormal signal restoration model is obtained by training a preset denoising autoencoder model based on a storage module using the tank wall corrosion detection signals of healthy oil storage tanks as training data and the tank wall corrosion detection signals of the oil storage tank to be detected as test data.

[0063] Among them, the denoising autoencoder model based on the storage module includes an autoencoder, a storage module, and a denoising autoencoder. Among them, the storage module is used to provide storage entry data for the autoencoder and the denoising autoencoder. The autoencoder is used to obtain the features of the input data of the autoencoder to obtain a first feature vector, and reconstruct the input data of the autoencoder according to the storage entry data and the second feature vector. The denoising autoencoder is used to obtain the features of the input data of the denoising autoencoder to obtain a second feature vector, and reconstruct the input data of the denoising autoencoder according to the second feature vector and the storage entry data.

[0064] Among them, the storage module was first proposed for answering questions in semantic segmentation. It stores known information in a storage barcode and indexes relevant answers from the storage entries when a question comes in. With the in-depth research, the storage module is no longer limited to QA questions. It combines with other networks and updates the storage entry data in the storage module through the update of the main network to complete information storage; then extracts the storage entry data therein through indexing to complete the reading process. In this embodiment, the storage module is set as a matrix M of size λ×m (M = [ξ 1 , ξ 2 ,..., ξ m ), which contains m storage entries, and the dimension of each storage entry data ξ λ×m is λ-dimensional. The storage module is updated depending on the update of the autoencoder, and at the same time provides the features of the normal signal, that is, the storage entry data, for the autoencoder and the denoising autoencoder. i

[0065] The input data of the autoencoder is also used for the supervision of the autoencoder and the denoising autoencoder. The encoder part of the autoencoder contains c convolutional layers, each convolutional layer contains an activation function, and the encoding part is used to extract the features of the input signal, and the features output by the encoder are λ-dimensional. The decoder part of the autoencoder contains c deconvolutional layers, each deconvolutional layer contains an activation function, and the decoding part is used to reconstruct the features.

[0066] The denoising autoencoder is an autoencoder that can process corrupted data and has a feature extractor with denoising effects. The input data contains noise, and the output data is clean data without noise. It is mainly applied to data denoising and visualization dimensionality reduction and has high robustness. The encoding part of the denoising autoencoder contains c convolutional layers, and each convolutional layer contains an activation function. The encoding part is used to extract the features of the input signal, and the features output by the encoder are λ-dimensional. The decoder part of the denoising autoencoder contains c deconvolutional layers, and each deconvolutional layer contains an activation function. The decoding part is used to reconstruct the input signal according to the stored entry data read from the storage module.

[0067] In this embodiment, preferably, the loss function Loss of the denoising autoencoder model based on the storage module is as follows:

[0068]

[0069] where L AE is the autoencoder loss function, is the output data of the autoencoder, and d i is the input data of the autoencoder; L DAE is the denoising autoencoder loss function, is the output data of the denoising autoencoder, is the input data of the denoising autoencoder, and α is a hyperparameter.

[0070] Among them, the method for training a preset denoising autoencoder model based on the storage module with the tank wall corrosion detection signal of a healthy storage tank as the training data and the tank wall corrosion detection signal of the storage tank to be detected as the test data is as follows: After baseline correction of the tank wall corrosion detection signal of the healthy storage tank, it is segmented into several first tank wall corrosion detection sub-signals, and a preset abnormal type is added to each first tank wall corrosion detection sub-signal to obtain several abnormal signals of each first tank wall corrosion detection sub-signal. The several abnormal signals of each first tank wall corrosion detection sub-signal and each first tank wall corrosion detection sub-signal are combined into a group in sequence to obtain several training samples; After baseline correction of the tank wall corrosion detection signal of the storage tank to be detected, it is segmented into several second tank wall corrosion detection sub-signals to obtain several test samples; Perform the training step, and the training step includes: training a preset denoising autoencoder model based on the storage module through several training samples, and testing the trained denoising autoencoder model based on the storage module through the test samples to obtain a test result. When the test result meets the preset test conditions, the currently trained denoising autoencoder model based on the storage module is used as the abnormal signal recovery model; Otherwise, repeat the training step.

[0071] Specifically, baseline correction is performed on all signals to ensure that the signal base values at different positions of different sensors are the same. Then, a sliding window of size a×b is used to segment the signals after baseline correction, and the step size of the sliding movement is v. The dataset after training signal segmentation is D = {d 1 , d 2 ,..., d n}, which contains n training samples. The dataset after testing signal segmentation is T = {t 1 , t 2 ,..., t m}, which contains m test samples.

[0072] Abnormal information is added to the training set D. Various types of abnormalities are randomly added to the segmented dataset, such as white noise, missing channel signals, missing block signals, etc., to obtain the training set D * , as shown in the following formula:

[0073]

[0074] where represents that the j-th type of abnormality is added to the i-th training sample.

[0075] The specific method for training the preset denoising autoencoder model based on the storage module is as follows:

[0076] Initialize the autoencoder parameters of the denoising autoencoder model based on the storage module, the stored entry data in the storage module, and the denoising autoencoder parameters.

[0077] Input the first tank wall corrosion detection sub-signal in the training sample into the encoder of the autoencoder to obtain the first feature vector γ 1 , and calculate the cosine similarity between the first feature vector γ 1 and each stored entry data in the storage module through the following formula:

[0078]

[0079] where ξ k is the k-th stored entry data, v k is the cosine similarity between the first feature vector γ 1 and the k-th stored entry data in the storage module, and m is the number of stored entry data.

[0080] Among them, the encoding operation of the encoder of the autoencoder is shown in the following formula:

[0081]

[0082] where: x' i is the output after one encoding, σ is the activation function ReLU, is a convolution operation, is a convolution kernel of size c.

[0083] Normalize the cosine similarity between the first eigenvector γ 1 and the data of each storage entry in the storage module to obtain the normalized cosine similarity v = {v' 1 , v' 2 ,..., v' m}.

[0084] Obtain the first reconstructed feature information γ through the following formula 2 :

[0085] γ 2 = v' 1 ξ 1 + v' 2 ξ 2 +... + v' m ξ m

[0086] Input the first reconstructed feature information γ 2 into the decoder of the autoencoder to obtain the reconstructed signal of the first tank wall corrosion detection sub-signal.

[0087] Among them, the decoding operation of the decoder of the autoencoder is shown by the following formula:

[0088]

[0089] Among them: is the output after one decoding, σ is the activation function ReLU, ⊙ is the transposed convolution operation, is a transposed convolution kernel of size d.

[0090] Input the abnormal signal of the first tank wall corrosion detection sub-signal in the training sample into the encoder of the denoising autoencoder to obtain the second eigenvector Calculate the cosine similarity between the second eigenvector and the data of each storage entry in the storage module through the following formula:

[0091]

[0092] Among them, w k is the cosine similarity between the first eigenvector and the data of the k-th storage entry in the storage module.

[0093] Normalize the cosine similarity between the second eigenvector and the data of each storage entry in the storage module to obtain the normalized cosine similarity w = {w' 1 , w'2 ,..., w' m}。

[0094] Obtain the second reconstructed feature information through the following formula

[0095]

[0096] Input the second reconstructed feature information into the decoder of the denoising autoencoder to obtain the reconstructed signal of the abnormal signal of the first tank wall corrosion detection sub-signal.

[0097] The autoencoder and the denoising autoencoder perform backpropagation simultaneously. Among them, when the autoencoder performs backpropagation, based on the first tank wall corrosion detection sub-signal and the reconstructed signal of the first tank wall corrosion detection sub-signal, with the goal of minimizing the loss function Loss of the denoising autoencoder model based on the storage module, update the autoencoder parameters and the stored entry data in the storage module; when the denoising autoencoder performs backpropagation, based on the abnormal signal of the first tank wall corrosion detection sub-signal and the reconstructed signal of the abnormal signal of the first tank wall corrosion detection sub-signal, with the goal of minimizing the loss function Loss of the denoising autoencoder model based on the storage module, update the denoising autoencoder parameters.

[0098] Among them, when updating the autoencoder parameters and the stored entry data in the storage module, and when updating the denoising autoencoder parameters, the stochastic gradient descent method is used for updating.

[0099] S3: Integrate several tank wall corrosion detection recovery signals to obtain the tank wall corrosion detection signal of the storage tank to be detected after recovering the abnormal signal. Specifically, according to the adopted segmentation method, integrate several tank wall corrosion detection recovery signals in reverse to obtain the tank wall corrosion detection signal of the storage tank to be detected after recovering the abnormal signal.

[0100] In summary, for the method for recovering the corrosion detection signal of the oil storage tank wall in the present invention, by using the corrosion detection signal of the healthy oil storage tank wall as the training data and the corrosion detection signal of the oil storage tank wall to be detected as the test data, an abnormal signal recovery model obtained by training a preset denoising autoencoder model based on a storage module is used to sequentially recover a plurality of sub-signals of the corrosion detection of the tank wall to be recovered into a plurality of corrosion detection recovery signals of the tank wall, completing the reconstruction of the sub-signals of the corrosion detection of the tank wall to be recovered, and realizing the abnormal signal recovery of the corrosion detection signal of the tank wall to be detected. Among them, a storage module for storing the characteristics of normal signals is added to the denoising autoencoder model based on the storage module. The stored entry data in the storage module is obtained by training a convolutional autoencoder. After the denoising autoencoder extracts the features through the encoding part, the corresponding stored entry data is selected from the storage module to reconstruct the features, and the reconstructed features are input into the decoder to complete the reconstruction. By reconstructing the stored entry data in the storage module, the generalization of the denoising autoencoder is reduced, preventing the recovery of the original abnormal signal and improving the accuracy of signal recovery. At the same time, this method is also an unsupervised learning method, and the training process does not require actual abnormal samples, solving the problem of few training samples of abnormal signals.

[0101] The following is an apparatus embodiment of the present invention, which can be used to execute the method embodiment of the present invention. For the details not disclosed in the apparatus embodiment, please refer to the method embodiment of the present invention.

[0102] In another embodiment of the present invention, a system for recovering the corrosion detection signal of the oil storage tank wall is provided, which can be used to implement the above method for recovering the corrosion detection signal of the oil storage tank wall. Specifically, the system for recovering the corrosion detection signal of the oil storage tank wall includes: an acquisition module, a recovery module, and an integration module.

[0103] Among them, the acquisition module is used to acquire the sub-signals of the corrosion detection of the tank wall to be recovered of the oil storage tank to be detected and divide the sub-signals of the corrosion detection of the tank wall to be recovered into a plurality of sub-signals of the corrosion detection of the tank wall to be recovered; the recovery module is used to sequentially input the plurality of sub-signals of the corrosion detection of the tank wall to be recovered into a preset abnormal signal recovery model to obtain a plurality of corrosion detection recovery signals of the tank wall; wherein, the abnormal signal recovery model is obtained by using the corrosion detection signal of the healthy oil storage tank wall as the training data and the corrosion detection signal of the oil storage tank wall to be detected as the test data to train a preset denoising autoencoder model based on a storage module; the integration module is used to integrate the plurality of corrosion detection recovery signals to obtain the corrosion detection signal of the oil storage tank to be detected after the abnormal signal is recovered.

[0104] In another embodiment of the present invention, a computer device is provided. The computer device includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function. The processor described in the embodiment of the present invention can be used for the operation of the method for recovering the corrosion detection signal of the oil storage tank wall.

[0105] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the computer device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, and the operating system of the terminal is stored in this storage space. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the method for recovering the corrosion detection signal of the oil storage tank wall in the above embodiments.

[0106] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0107] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0108] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that realizes the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, so that the instructions executed on the computer or other programmable devices provide steps for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still can modify the specific embodiments of the present invention or make equivalent substitutions, and any modification or equivalent substitution that does not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. A method for recovering the corrosion detection signal of the storage tank wall, characterized in that, it includes the following steps: Obtain the to-be-recovered corrosion detection signal of the storage tank to be detected, and divide the to-be-recovered corrosion detection signal of the storage tank wall into several to-be-recovered corrosion detection sub-signals of the storage tank wall; Input several to-be-recovered corrosion detection sub-signals of the storage tank wall into a preset abnormal signal recovery model in sequence to obtain several corrosion detection recovery signals of the storage tank wall; among them, the abnormal signal recovery model is obtained by using the corrosion detection signal of the healthy storage tank as training data and the corrosion detection signal of the storage tank to be detected as test data to train a preset denoising autoencoder model based on the storage module; Integrate several corrosion detection recovery signals of the storage tank wall to obtain the corrosion detection signal of the storage tank wall after recovering the abnormal signal to be detected; The denoising autoencoder model based on the storage module includes an autoencoder, a storage module, and a denoising autoencoder; Among them, the storage module is used to provide storage entry data for the autoencoder and the denoising autoencoder; The autoencoder is used to obtain the features of the autoencoder input data to obtain the first feature vector, and reconstruct the autoencoder input data according to the storage entry data and the second feature vector; The denoising autoencoder is used to obtain the features of the denoising autoencoder input data to obtain the second feature vector, and reconstruct the denoising autoencoder input data according to the second feature vector and the storage entry data; Among them, the storage entry data is obtained by training a convolutional autoencoder according to the normal signal features.

2. The method for recovering the corrosion detection signal of the storage tank wall according to claim 1, characterized in that, The specific method for dividing the to-be-recovered corrosion detection signal of the storage tank wall into several to-be-recovered corrosion detection sub-signals of the storage tank wall is: through a sliding window of a preset size and with a preset sliding step, divide the to-be-recovered corrosion detection signal of the storage tank wall into several to-be-recovered corrosion detection sub-signals of the storage tank wall.

3. The method for recovering the corrosion detection signal of the storage tank wall according to claim 1, characterized in that, Before dividing the to-be-recovered corrosion detection signal of the storage tank wall into several to-be-recovered corrosion detection sub-signals of the storage tank wall, perform baseline correction on the to-be-recovered corrosion detection signal of the storage tank wall.

4. The method for recovering the corrosion detection signal of the storage tank wall according to claim 1, characterized in that, The loss function Loss of the denoising autoencoder model based on the storage module is: Among them, L AE is the autoencoder loss function, is the output data of the autoencoder, and d i is the input data of the autoencoder; L DAE is the denoising autoencoder loss function, is the output data of the denoising autoencoder, is the input data of the denoising autoencoder, and α is a hyperparameter.

5. The method for recovering the corrosion detection signal of the storage tank wall according to claim 4, characterized in that, The method for using the corrosion detection signal of the healthy storage tank as training data and the corrosion detection signal of the storage tank to be detected as test data to train a preset denoising autoencoder model based on the storage module is: After performing baseline correction on the corrosion detection signal of the healthy storage tank wall, divide it into several first corrosion detection sub-signals of the storage tank wall, add a preset abnormal type to each first corrosion detection sub-signal of the storage tank wall to obtain several abnormal signals of each first corrosion detection sub-signal of the storage tank wall, and combine the several abnormal signals of each first corrosion detection sub-signal of the storage tank wall and each first corrosion detection sub-signal of the storage tank wall into a group in sequence to obtain several training samples; After performing baseline correction on the tank wall corrosion detection signal of the storage tank to be detected, it is segmented into several second tank wall corrosion detection sub-signals to obtain several test samples; Perform a training step, which includes: training a preset denoising autoencoder model based on a storage module through several training samples, and testing the trained denoising autoencoder model based on the storage module with the test samples to obtain a test result. When the test result meets the preset test conditions, use the currently trained denoising autoencoder model based on the storage module as the abnormal signal recovery model; Otherwise, repeat the training step.

6. The method for recovering the tank wall corrosion detection signal of a storage tank according to claim 5, characterized in that, the preset abnormal types include adding white noise, channel signal loss, and block signal loss.

7. The method for recovering the tank wall corrosion detection signal of a storage tank according to claim 5, characterized in that, the specific method for training the preset denoising autoencoder model based on the storage module is: Initialize the autoencoder parameters of the denoising autoencoder model based on the storage module, the stored entry data in the storage module, and the denoising autoencoder parameters; Input the first tank wall corrosion detection sub-signal in the training samples into the encoder of the autoencoder to obtain the first feature vector γ 1 , and calculate the first feature vector γ through the following formula 1 The cosine similarity with the data of each storage entry in the storage module: Among them, ξ k is the data of the k-th storage entry, v k is the cosine similarity between the first feature vector γ 1 and the data of the k-th storage entry in the storage module, and m is the number of storage entry data; Normalize the cosine similarity between the first feature vector γ 1 and the data of each storage entry in the storage module to obtain the normalized cosine similarity v = {v' 1 , v' 2 ,..., v' m}; Obtain the first reconstructed feature information γ through the following formula 2 :[[]]END]] γ 2 = v' 1 ξ 1 + v' 2 ξ 2 +... + v' m ξ m Input the first reconstructed feature information γ 2 into the decoder of the autoencoder to obtain the reconstructed signal of the first tank wall corrosion detection sub-signal; Input the abnormal signal of the first tank wall corrosion detection sub-signal in the training sample into the encoder of the denoising autoencoder to obtain a second feature vector Calculate the second feature vector through the following formula The cosine similarity with the data of each storage entry in the storage module: where, w k is the cosine similarity of the first eigenvector to the data of the k-th storage entry in the storage module; Normalize the cosine similarity between the second feature vector and the data of each storage entry in the storage module to obtain the normalized cosine similarity w = {w' 1 , w' 2 ,..., w' m}; Obtain the second reconstructed feature information through the following formula Input the second reconstructed feature information into the decoder of the denoising autoencoder to obtain the reconstructed signal of the abnormal signal of the first tank wall corrosion detection sub-signal; The autoencoder and the denoising autoencoder perform backpropagation simultaneously. Among them, when the autoencoder performs backpropagation, based on the first tank wall corrosion detection sub-signal and the reconstructed signal of the first tank wall corrosion detection sub-signal, with the goal of minimizing the loss function Loss of the denoising autoencoder model based on the storage module, update the autoencoder parameters and the stored entry data in the storage module; when the denoising autoencoder performs backpropagation, based on the abnormal signal of the first tank wall corrosion detection sub-signal and the reconstructed signal of the abnormal signal of the first tank wall corrosion detection sub-signal, with the goal of minimizing the loss function Loss of the denoising autoencoder model based on the storage module, update the denoising autoencoder parameters.

8. The method for recovering the tank wall corrosion detection signal of a storage tank according to claim 7, characterized in that, When updating the autoencoder parameters and the stored entry data in the storage module, and updating the denoising autoencoder parameters, the stochastic gradient descent method is used for updating.

9. A system for recovering the tank wall corrosion detection signal of a storage tank, characterized in that, includes: An acquisition module for acquiring the tank wall corrosion detection signal to be recovered of the storage tank to be detected, and segmenting the tank wall corrosion detection signal to be recovered into several tank wall corrosion detection sub-signals to be recovered; A recovery module for sequentially inputting several tank wall corrosion detection sub-signals to be recovered into a preset abnormal signal recovery model to obtain several tank wall corrosion detection recovery signals; among them, the abnormal signal recovery model is obtained by training a preset denoising autoencoder model based on the storage module with the tank wall corrosion detection signal of a healthy storage tank as the training data and the tank wall corrosion detection signal of the storage tank to be detected as the test data; An integration module for integrating several tank wall corrosion detection recovery signals to obtain the tank wall corrosion detection signal after recovering the abnormal signal of the storage tank to be detected; The denoising autoencoder model based on the storage module includes an autoencoder, a storage module, and a denoising autoencoder; Among them, the storage module is used to provide stored entry data for the autoencoder and the denoising autoencoder; An autoencoder is used to obtain the features of the autoencoder input data, obtain the first feature vector, and reconstruct the autoencoder input data according to the stored entry data and the second feature vector; A denoising autoencoder is used to obtain the features of the denoising autoencoder input data, obtain the second feature vector, and reconstruct the denoising autoencoder input data according to the second feature vector and the stored entry data; Among them, the stored entry data is trained by a convolutional autoencoder according to the normal signal features.

Images