Multi-sensor outlier detection method and device, computer device and storage medium

Abstract

Embodiments of the present application disclose a multi-sensor outlier detection method and device, computer equipment and a storage medium. The method comprises: acquiring a multi-element original signal; training a first autoencoder using reconstruction error according to the multi-element original signal, and decomposing a hidden layer in the first autoencoder by VMD to obtain a decomposed signal; sampling data features in a bagging manner using the decomposed signal to obtain feature values; inputting the feature values into a second autoencoder and optimizing two auto-encoding decoders in the second autoencoder to determine an upper bound estimate of reconstruction features and a lower bound estimate of reconstruction features; determining a sample label corresponding to the feature values according to the decomposed signal, the upper bound estimate of reconstruction features and the lower bound estimate of reconstruction features; and determining an outlier according to the sample label corresponding to the feature values. The method of the embodiments of the present application can meet the demand of large-scale data, and the robustness of the detection algorithm is good.

Description

Technical Field

[0001] The present invention relates to an outlier detection method, and more specifically to a multi-sensor outlier detection method, apparatus, computer equipment, and storage medium. Background Art

[0002] An autoencoder is a neural network trained through unsupervised learning that is trained to learn a reconstruction close to its original input. An autoencoder consists of two parts: an encoder and a decoder. A neural network with a single hidden layer has Z = σ(W1*Y+b1) and Y * =σ(W2*Z+b2) encoder and decoder, W and b are the weights and biases of the neural network, and σ is the nonlinear transformation function. The optimization goal of AE is || Y—Y * || min ; The encoder in Z = σ(W1*Y+b1) maps the input vector Y to the latent space Z through an affine mapping after nonlinearity. * =σ(W2*Z+b2) The decoder in σ2 maps the hidden representation h back to the original input space as the reconstructed signal through the same transformation as the encoder. The original input vector Y and the reconstructed Y * The difference between them is called the reconstruction error, such as ||Y—Y * || min As shown in . The autoencoder learns to minimize the reconstruction error. The autoencoder decoder belongs to the generative model, which requires that the difference between the reconstructed data and the original signal is as small as possible. For the network model trained with normal data, the reconstruction error || Y—Y is input. * || is very small. If abnormal data is input, the reconstruction error will be very different.

[0003] When the traditional autoencoder decoder performs outlier detection, the original information Y is input into the encoder, and after hidden layer processing, the decoder performs decoding operation. The training data is used for network training, and the loss function is || Y - Y * ||,Y * It can be regarded as f(Y), where f represents the trained model. Input the training data, perform model training, obtain the weight parameters of the encoder and decoder, input the data to be tested into the model, and compare the reconstruction error || Y - Y * ||, set the threshold. If the difference between the original signal and the reconstructed signal is greater than the threshold, it is judged as an outlier; otherwise, it is a normal value.

[0004] However, for sensor data, as the detection time increases, the data scale grows exponentially. The current outlier detection method cannot meet the needs of processing large-scale data, and the current algorithm has poor robustness.

[0005] Therefore, it is necessary to design a new method to meet the needs of large-scale data and ensure the robustness of the detection algorithm. Summary of the Invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a multi-sensor outlier detection method, device, computer equipment and storage medium.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a multi-sensor outlier detection method, comprising:

[0008] Obtaining multivariate original signals;

[0009] The first autoencoder is trained using the reconstruction error according to the multivariate original signal, and the hidden layer in the first autoencoder is subjected to VMD decomposition to obtain a decomposed signal;

[0010] The decomposed signal is sampled by bagging to obtain a feature value;

[0011] Inputting the feature value into a second autoencoder and optimizing two autoencoder decoders in the second autoencoder to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature;

[0012] Determining a sample label corresponding to the feature value according to the decomposed signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature;

[0013] Determine outliers based on sample labels corresponding to the feature values.

[0014] Its further technical solution is: the reconstruction loss function of the first autoencoder is ||YY′||, and the optimization target is the 50th percentile of the reconstruction loss function; wherein the optimization target is loss q＝0.5 =0.5([∈ i ]), Y=[y1,y2,…,y i ];∈ i is the reconstruction error, loss q＝0.5 is the 50th percentile; i represents the i-th column in the multivariate time series Y.

[0015] A further technical solution is: inputting the feature value into the second autoencoder and optimizing the two autoencoder decoders in the second autoencoder to determine the upper bound estimate value of the reconstructed feature and the lower bound estimate value of the reconstructed feature, including:

[0016] The feature value is input to the second autoencoder and the two autoencoder decoders in the second autoencoder are optimized. Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where Represents the lower bound of the i-th feature, U s Represents the sth component after VMD decomposition, y i,t Represents the next monitoring value of the i-th sensor (feature) at time t.

[0017] A further technical solution is: determining the sample label corresponding to the feature value according to the decomposed signal, the upper bound estimated value of the reconstructed feature, and the lower bound estimated value of the reconstructed feature, including:

[0018] Determining whether the decomposed signal is between an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature;

[0019] If the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determining that the sample label corresponding to the feature value is an abnormal label;

[0020] If the decomposed signal is not between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, it is determined that the sample label corresponding to the feature value is a non-abnormal label.

[0021] A further technical solution is: determining the outlier based on the sample label corresponding to the feature value includes:

[0022] A voting mechanism is used to determine outliers by taking the majority of the sample labels corresponding to all feature values.

[0023] A further technical solution is: sampling the data features of the decomposed signal using a bagging method to obtain feature values, including:

[0024] The anomaly score of the decomposed signal is calculated using a bagging method to obtain a eigenvalue.

[0025] The present invention also provides a multi-sensor outlier detection device, comprising:

[0026] A signal acquisition unit, used for acquiring a multivariate original signal;

[0027] a decomposition unit, configured to train a first autoencoder using a reconstruction error according to the multivariate original signal, and perform VMD decomposition on a hidden layer in the first autoencoder to obtain a decomposed signal;

[0028] A sampling unit, configured to sample data features of the decomposed signal using a bagging method to obtain feature values;

[0029] an optimization unit, configured to input the feature value into a second autoencoder and optimize two autoencoder decoders in the second autoencoder to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature;

[0030] a label determination unit, configured to determine a sample label corresponding to the feature value based on the decomposed signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature;

[0031] An outlier determination unit is used to determine an outlier based on a sample label corresponding to the feature value.

[0032] Its further technical solution is: the optimization unit is used to input the feature value into the second autoencoder and optimize the two autoencoder decoders in the second autoencoder, using Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where represents the lower bound of the i-th feature, U9 represents the s-th component after VMD decomposition, and y i,t Represents the next monitoring value of the i-th sensor (feature) at time t.

[0033] The present invention further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0034] The present invention also provides a storage medium storing a computer program, which can implement the above method when executed by a processor.

[0035] The beneficial effects of the present invention compared with the prior art are as follows: the present invention obtains a multivariate original signal, processes it using a first autoencoder, performs VMD decomposition on the hidden layer, samples it in a bagging manner, and inputs it into a second autoencoder to determine the upper bound estimate of the reconstructed feature and the lower bound estimate of the reconstructed feature, and then determines the outlier. The low-dimensional information of the hidden layer replaces the high-dimensional original information, and an integrated method is used to detect outliers on the data, thereby meeting the needs of large-scale data and achieving good robustness of the detection algorithm.

[0036] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A schematic diagram of an application scenario of the multi-sensor outlier detection method provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of a multi-sensor outlier detection method according to an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of a sub-process of a multi-sensor outlier detection method provided by an embodiment of the present invention;

[0041] Figure 4 A schematic block diagram of a multi-sensor outlier detection device provided by an embodiment of the present invention;

[0042] Figure 5 A schematic block diagram of a label determination unit of a multi-sensor outlier detection device provided by an embodiment of the present invention;

[0043] Figure 6 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0046] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0047] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0048] See also Figure 1 and Figure 2 , Figure 1 Schematic diagram of an application scenario of the multi-sensor outlier method provided in an embodiment of the present invention. Figure 2 This is a schematic flow chart of a multi-sensor outlier method provided in an embodiment of the present invention. This Zookeeper monitoring method is applied to a server. The server interacts with sensors, acquires multivariate raw signals from the sensors, and performs VMD decomposition on the hidden layer, replacing the high-dimensional raw information with the hidden layer's low-dimensional information to address the algorithm's need to process large amounts of data. A bagging approach is used to sample features and calculate anomaly scores. The anomaly score for each sample is ultimately averaged. This approach uses an integrated approach to detect outliers in the data, making the algorithm more robust.

[0049] Figure 2 FIG. 1 is a flow chart of a multi-sensor outlier detection method according to an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S150.

[0050] S110: Acquire a multivariate original signal.

[0051] In this embodiment, the multi-element original signal refers to a detection signal generated by multiple sensors.

[0052] S120 , training a first autoencoder using a reconstruction error according to the multivariate original signal, and performing VMD decomposition on a hidden layer in the first autoencoder to obtain a decomposed signal.

[0053] In this embodiment, the decomposed signal refers to a signal obtained by inputting the multivariate original signal into the first autoencoder and performing VMD (Variational Mode Decomposition) decomposition on the hidden layer.

[0054] VMD decomposition is performed using the following formula: Where, K is the number of modes to be decomposed (positive integer); {μ k},{ω k} respectively correspond to the kth modal component and center frequency after decomposition; j represents the amplitude; f represents the sum of the signal values; δ(t) is the Dirac function, and * is the convolution operator.

[0055] Specifically, Y represents the multivariate time series of multiple sensors, Y = [y1,y2,…,y i ], i represents the ii-th sensor; Y′ represents the time series reconstructed after the self-encoder decoder, Y′=[y1′,y2′,…,y i ′]. The reconstruction loss function of the first autoencoder is ||YY′||, and the optimization target is the 50th percentile of the reconstruction loss function; wherein the optimization target is loss q＝0.5 =0.5([∈ i ]), Y=[y1,y2,…,y i ];∈ i is the reconstruction error, loss q＝0.5 is the 50th percentile; i represents the i-th column in the multivariate time series Y.

[0056] S130: Sample the data features of the decomposed signal using a bagging method to obtain feature values.

[0057] In this embodiment, the characteristic value refers to the abnormality score value corresponding to the decomposed signal.

[0058] Specifically, the anomaly score is calculated on the decomposed signal using a bagging method to obtain a eigenvalue.

[0059] The bagging method is used to sample the features and calculate the anomaly score. The anomaly score of each sample is finally obtained by averaging the scores. This method uses an integrated method to detect outliers in the data, making the algorithm more robust.

[0060] S140: Input the feature value into the second autoencoder and optimize two autoencoder decoders in the second autoencoder to determine an upper bound estimate value of the reconstructed feature and a lower bound estimate value of the reconstructed feature.

[0061] In this embodiment, the upper bound estimate value and the lower bound estimate value of the reconstructed feature refer to the upper and lower limits of the scores corresponding to the abnormal labels. The reconstructed feature refers to the feature obtained after bagging, that is, the feature value.

[0062] Specifically, the feature value is input to the second autoencoder and the two autoencoders in the second autoencoder are optimized, using Determine the upper bound estimate of the reconstructed feature and the lower bound estimate of the reconstructed feature, where q represents the qth quantile, σ represents the input quantile parameter, the range is [0,1], i represents the number of features, and y i,t represents the next monitoring value of the i-th sensor (feature) at time t, represents the lower bound of the i-th feature at time t, and U9 represents the s-th modal component after VMD decomposition.

[0063] Specifically, the sample to be tested is input into the trained model to obtain the upper and lower bounds of its confidence interval. The discriminant formula is as follows:

[0064] N represents the number of samples, t represents the time; I features are extracted through bagging, i represents the i-th feature, the corresponding abnormal labels of I features are calculated, and the final label of each sample is obtained by voting.

[0065] S150 , determining a sample label corresponding to the feature value according to the decomposed signal, the upper bound estimate value of the reconstructed feature, and the lower bound estimate value of the reconstructed feature.

[0066] In this embodiment, the sample label refers to the label corresponding to each signal in the multivariate original signal, including an abnormal label and a non-abnormal label.

[0067] In one embodiment, see Figure 3 , the above-mentioned step S150 may include steps S151 to S153.

[0068] S151, determining whether the decomposed signal is between an upper bound estimate value of the reconstructed feature and a lower bound estimate value of the reconstructed feature;

[0069] S152: If the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determining that the sample label corresponding to the feature value is an abnormal label;

[0070] S153: If the decomposed signal is not between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determine that the sample label corresponding to the feature value is a non-abnormal label.

[0071] The data features are sampled using bagging and input into the autoencoder 2. The two autoencoders are optimized by the formula, and we get The lower bound estimate of the upper reconstruction feature is obtained; finally, the sample label of each sampling is obtained by judging the conditions, and the final label is obtained by voting.

[0072] S160: Determine an outlier based on a sample label corresponding to the feature value.

[0073] In this embodiment, a voting mechanism is used to determine outliers by taking the majority of the sample labels corresponding to all feature values.

[0074] The final decision is determined by voting on the results of multiple bagging. For example, if there are ten eigenvalues ​​and the model has ten labels, the label is determined by taking the mode, and the probability is determined by the mean probability of all labels.

[0075] The above-mentioned multi-sensor outlier detection method obtains multivariate original signals, processes them using the first autoencoder, performs VMD decomposition on the hidden layer, and then samples them in a bagging manner and inputs them into the second autoencoder to determine the upper bound estimate of the reconstructed features and the lower bound estimate of the reconstructed features. Then, the outlier is determined, and the high-dimensional original information is replaced by the low-dimensional information of the hidden layer. An integrated method is used to detect outliers on the data, thereby meeting the needs of large-scale data and achieving good robustness of the detection algorithm.

[0076] Figure 4 FIG is a schematic block diagram of a multi-sensor abnormal value detection device 300 provided by an embodiment of the present invention. Figure 4 As shown, corresponding to the above multi-sensor outlier detection method, the present invention also provides a multi-sensor outlier detection device 300. The multi-sensor outlier detection device 300 includes a unit for executing the above multi-sensor outlier detection method, and the device can be configured in a server. Figure 4 The multi-sensor outlier detection device 300 includes a signal acquisition unit 301 , a decomposition unit 302 , a sampling unit 303 , an optimization unit 304 , a label determination unit 305 and an outlier determination unit 306 .

[0077] The signal acquisition unit 301 is used to acquire a multivariate original signal; the decomposition unit 302 is used to train a first autoencoder using a reconstruction error according to the multivariate original signal, and perform VMD decomposition on the hidden layer in the first autoencoder to obtain a decomposed signal; the sampling unit 303 is used to sample data features using a bagging method on the decomposed signal to obtain a eigenvalue; the optimization unit 304 is used to input the eigenvalue into a second autoencoder and optimize two autoencoder decoders in the second autoencoder to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature; the label determination unit 305 is used to determine a sample label corresponding to the eigenvalue according to the decomposition signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature; the outlier determination unit 306 is used to determine an outlier based on the sample label corresponding to the eigenvalue.

[0078] In one embodiment, the optimization unit 304 is configured to input the feature value into the second autoencoder and optimize the two autoencoders in the second autoencoder, using Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where represents the lower bound of the i-th feature, and U9 represents the s-th component after VMD decomposition.

[0079] In one embodiment, if Figure 5 As shown, the label determination unit 305 includes a signal determination subunit 3051 , a first determination subunit 3052 and a second determination subunit 3053 .

[0080] The signal judgment subunit 3051 is used to judge whether the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature; the first determination subunit 3052 is used to determine that the sample label corresponding to the feature value is an abnormal label if the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature; the second determination subunit 3053 is used to determine that the sample label corresponding to the feature value is a non-abnormal label if the decomposed signal is not between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature.

[0081] In one embodiment, the outlier determination unit 306 is configured to determine an outlier by taking the majority of the sample labels corresponding to all feature values ​​using a voting mechanism.

[0082] In one embodiment, the sampling unit 303 is configured to calculate an anomaly score on the decomposed signal using a bagging method to obtain a feature value.

[0083] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned multi-sensor outlier detection device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and brevity of the description, it will not be repeated here.

[0084] The multi-sensor outlier detection device 300 can be implemented as a computer program. Figure 6 Runs on the computer equipment shown.

[0085] See also Figure 6 , Figure 6 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0086] See Figure 6The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0087] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, may enable the processor 502 to perform a multi-sensor outlier detection method.

[0088] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0089] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a multi-sensor abnormal value detection method.

[0090] The network interface 505 is used to communicate with other devices through the network. Figure 6 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0091] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0092] A multivariate original signal is obtained; a first autoencoder is trained using a reconstruction error according to the multivariate original signal, and a hidden layer in the first autoencoder is subjected to VMD decomposition to obtain a decomposition signal; data features are sampled using a bagging method using the decomposition signal to obtain a eigenvalue; the eigenvalue is input into a second autoencoder and two autoencoder decoders in the second autoencoder are optimized to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature; a sample label corresponding to the eigenvalue is determined according to the decomposition signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature; and an outlier is determined according to the sample label corresponding to the eigenvalue.

[0093] The reconstruction loss function of the first autoencoder is ||YY′||, and the optimization target is the 50th percentile of the reconstruction loss function; the optimization target is loss q＝0.5 =0.5([∈ i ]), Y=[y1,y2,…,yi ];∈ i is the reconstruction error, loss q＝0.5 is the 50th percentile; i represents the i-th column in the multivariate time series Y.

[0094] In one embodiment, when the processor 502 implements the step of inputting the feature value into the second autoencoder and optimizing the two autoencoder decoders in the second autoencoder to determine the upper bound estimate value of the reconstructed feature and the lower bound estimate value of the reconstructed feature, the processor 502 specifically implements the following steps:

[0095] The feature value is input to the second autoencoder and the two autoencoder decoders in the second autoencoder are optimized. Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where Represents the lower bound of the i-th feature, U s represents the sth component after VMD decomposition, and σ represents the input quantile parameter.

[0096] In one embodiment, when the processor 502 implements the step of determining the sample label corresponding to the feature value according to the decomposed signal, the upper bound estimated value of the reconstructed feature, and the lower bound estimated value of the reconstructed feature, the processor 502 specifically implements the following steps:

[0097] Determine whether the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature; if the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determine that the sample label corresponding to the feature value is an abnormal label; if the decomposed signal is not between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determine that the sample label corresponding to the feature value is a non-abnormal label.

[0098] In one embodiment, when the processor 502 implements the step of determining an outlier based on the sample label corresponding to the feature value, the processor 502 specifically implements the following steps:

[0099] A voting mechanism is used to determine outliers by taking the majority of the sample labels corresponding to all feature values.

[0100] In one embodiment, when the processor 502 implements the step of sampling the data features of the decomposed signal using a bagging method to obtain a feature value, the processor 502 specifically implements the following steps:

[0101] The anomaly score of the decomposed signal is calculated using a bagging method to obtain a eigenvalue.

[0102] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0103] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0104] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0105] A multivariate original signal is obtained; a first autoencoder is trained using a reconstruction error according to the multivariate original signal, and a hidden layer in the first autoencoder is subjected to VMD decomposition to obtain a decomposition signal; data features are sampled using a bagging method using the decomposition signal to obtain a eigenvalue; the eigenvalue is input into a second autoencoder and two autoencoder decoders in the second autoencoder are optimized to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature; a sample label corresponding to the eigenvalue is determined according to the decomposition signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature; and an outlier is determined according to the sample label corresponding to the eigenvalue.

[0106] The reconstruction loss function of the first autoencoder is ||YY′||, and the optimization target is the 50th percentile of the reconstruction loss function; the optimization target is loss q＝0.5 =0.5([∈ i ]), Y=[y1,y2,…,y i ];∈ i is the reconstruction error, loss q＝0.5is the 50th percentile; i represents the i-th column in the multivariate time series Y.

[0107] In one embodiment, when the processor executes the computer program to implement the step of inputting the feature value into the second autoencoder and optimizing the two autoencoder decoders in the second autoencoder to determine the upper bound estimate value of the reconstructed feature and the lower bound estimate value of the reconstructed feature, the processor specifically implements the following steps:

[0108] The feature value is input to the second autoencoder and the two autoencoder decoders in the second autoencoder are optimized. Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where represents the lower bound of the i-th feature, and U9 represents the s-th component after VMD decomposition.

[0109] In one embodiment, when the processor executes the computer program to implement the step of determining the sample label corresponding to the feature value based on the decomposed signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature, the processor specifically implements the following steps:

[0110] Determine whether the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature; if the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determine that the sample label corresponding to the feature value is an abnormal label; if the decomposed signal is not between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determine that the sample label corresponding to the feature value is a non-abnormal label.

[0111] In one embodiment, when the processor executes the computer program to implement the step of determining an outlier based on the sample label corresponding to the feature value, the processor specifically implements the following steps:

[0112] A voting mechanism is used to determine outliers by taking the majority of the sample labels corresponding to all feature values.

[0113] In one embodiment, when the processor executes the computer program to implement the step of sampling the data features of the decomposed signal using a bagging method to obtain a feature value, the processor specifically implements the following steps:

[0114] The anomaly score of the decomposed signal is calculated using a bagging method to obtain a eigenvalue.

[0115] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0116] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0117] In the several embodiments provided herein, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementation may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0118] The steps in the methods of the embodiments of the present invention may be adjusted in order, combined, or deleted as needed. The units in the devices of the embodiments of the present invention may be combined, divided, or deleted as needed. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0119] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0120] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A multi-sensor outlier detection method, characterized in that: include: Obtaining multivariate original signals; The first autoencoder is trained using the reconstruction error according to the multivariate original signal, and the hidden layer in the first autoencoder is subjected to VMD decomposition to obtain a decomposed signal; The decomposed signal is sampled by bagging to obtain a data feature; the feature is input into a second autoencoder and the second autoencoder is optimized to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature; Determining a sample label corresponding to the feature according to the decomposed signal, an upper bound estimate of the reconstructed feature, and a lower bound estimate of the reconstructed feature; Determine outliers based on sample labels corresponding to the features; Inputting the feature into a second autoencoder and optimizing the second autoencoder to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature, comprising: The features are input to the second autoencoder and the second autoencoder is optimized, using Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where represents the lower bound of the i-th feature at time t, Us represents the s-th component after VMD decomposition; where q represents the quantile, σ represents the input quantile parameter, and the value range is [0,1]; i represents the i-th feature; yi,t represents the monitoring value of the i-th sensor at time t; I represents the total number of features; Among them, the reconstructed features refer to the features obtained after processing by bagging.

2. The multi-sensor outlier detection method according to claim 1, wherein: The reconstruction loss function of the first autoencoder is ||YY′||, and the optimization target is the 50th percentile of the reconstruction loss function; wherein the optimization target is loss q =0.5=0.5([∈i]), Y=[y1,y2,…,yi];∈i is the reconstruction error, loss q =0.5 is the 50th percentile; yi represents the i-th column data in the multivariate time series Y; Y represents the multivariate time series of multiple sensors, i represents the i-th sensor; Y′ represents the time series reconstructed after the first autoencoder.

3. The multi-sensor outlier detection method according to claim 1, wherein: The determining, according to the decomposed signal, the upper bound estimate of the reconstructed feature, and the lower bound estimate of the reconstructed feature, of the sample label corresponding to the feature includes: Determining whether the decomposed signal is between an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature; If the decomposed signal is between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, determining that the sample label corresponding to the feature is an abnormal label; If the decomposed signal is not between the upper bound estimated value of the reconstructed feature and the lower bound estimated value of the reconstructed feature, it is determined that the sample label corresponding to the feature is a non-abnormal label.

4. The multi-sensor outlier detection method according to claim 1, wherein: The determining of outliers according to the sample labels corresponding to the features includes: A voting mechanism is used to determine outliers by taking the majority of the sample labels corresponding to all features.

5. The multi-sensor outlier detection method according to claim 1, wherein: The step of sampling the data features of the decomposed signal in a bagging manner to obtain the features includes: calculating an anomaly score of the decomposed signal in a bagging manner to obtain the features.

6. A multi-sensor outlier detection device, characterized in that: include: A signal acquisition unit, used for acquiring a multivariate original signal; a decomposition unit, configured to train a first autoencoder using a reconstruction error according to the multivariate original signal, and perform VMD decomposition on a hidden layer in the first autoencoder to obtain a decomposed signal; A sampling unit, configured to sample the data features of the decomposed signal using a bagging method to obtain features; an optimization unit, configured to input the feature into a second autoencoder and optimize the second autoencoder to determine an upper bound estimate of the reconstructed feature and a lower bound estimate of the reconstructed feature; a label determination unit, configured to determine a sample label corresponding to the feature based on the decomposed signal, an upper bound estimate of the reconstructed feature, and a lower bound estimate of the reconstructed feature; an outlier determination unit, configured to determine an outlier based on a sample label corresponding to the feature; The optimization unit is used to input the features into the second autoencoder and optimize the second autoencoder, using Determine an upper bound estimate of the reconstructed features and a lower bound estimate of the reconstructed features, where represents the lower bound of the i-th feature at time t, Us represents the s-th component after VMD decomposition; where q represents the quantile, σ represents the input quantile parameter, and the value range is [0,1]; i represents the i-th feature; yi,t represents the monitoring value of the i-th sensor at time t; I represents the total number of features; Among them, the reconstructed features refer to the features obtained after processing by bagging.

7. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 5 when executing the computer program.

8. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 can be implemented.

Images