Method for predicting the future working status of scheduled equipment

Abstract

The present disclosure relates to the technical field of electrical signal processing methods, and specifically provides a method for predicting the future working state of a predetermined device, which can be applied to the fields of mechanical equipment status monitoring and artificial intelligence. The method comprises: S10: establishing a data set; S20: processing the signals in the data set to determine the health value of the signals; S30: processing the health value of the signals to determine a future signal that is a predetermined time interval from the current time interval; S40: monitoring and obtaining a future signal that is a predetermined time interval from the current time interval; S50: comparing the future signal determined in step S30 with the future signal obtained by monitoring in step S40 to determine the difference between the two; S60: correcting steps S20 and S30 based on the difference determined in step S50; S70: determining an updated health value of the signal; S80: determining the future working state of the monitored predetermined device based on the updated health value, thereby effectively improving the accuracy and reliability of predicting the future working state of the predetermined device.

Description

Technical Field

[0001] The present disclosure relates to the fields of mechanical equipment status monitoring and artificial intelligence, and in particular to a method for predicting the future working status of predetermined equipment. Background Art

[0002] In the industrial field, real-time monitoring and analysis of signals are crucial to ensure the normal operation of equipment and assets. Due to the complexity of equipment and the dynamic changes in the environment, it is difficult to accurately predict the behavior and trends of signals.

[0003] Current mechanical equipment signal monitoring typically relies on commercial software or deep learning-based network models. Commercial software relies on high-precision modeling of the equipment, resulting in slow computational speeds and inability to meet the real-time demands of industrial equipment monitoring. Traditional deep learning-based models also have limited accuracy in modeling complex systems. Therefore, a method to improve the accuracy of equipment signal prediction is needed. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a method for predicting the future working status of a predetermined device.

[0005] According to one aspect of the present disclosure, a method for predicting the future working status of a predetermined device is provided, comprising the following steps: S10: establishing a data set, wherein the data in the data set represent signals of the predetermined device monitored in real time, and new signals obtained by monitoring the predetermined device are added to the data set in real time; S20: processing the signals in the data set to determine a health value of the signal; S30: processing the health value of the signal to determine a future signal that is a predetermined time interval from the current time interval; S40: monitoring to obtain a future signal that is a predetermined time interval from the current time interval; S50: comparing the future signal determined in step S30 with the future signal obtained by monitoring in step S40 to determine the difference between the two; S60: correcting steps S20 and S30 based on the difference determined in step S50; S70: determining an updated health value of the signal; S80: determining the future working status of the monitored predetermined device based on the updated health value.

[0006] According to an embodiment of the present disclosure, in step S10, establishing a data set further includes: S11: using a sensor to obtain multiple real-time signals of a predetermined device in a predetermined time period; S12: segmenting each of the multiple real-time signals; S13: the segmented signals form a data set.

[0007] According to an embodiment of the present disclosure, in step S20, the health value of the signal satisfies the following expression:

[0008]

[0009] Where HI represents the health value of the signal, t represents the time, x(t) represents the signal at the current time, and z(t) satisfies the following expression:

[0010]

[0011] represents the Fourier transform of z(t); W f Represents a filter, To filter the predetermined frequency in, σ satisfies the following expression:

[0012]

[0013] W hl is the dimension reduction matrix; W P is the signal matrix; c represents the signal channel, l represents the signal length; α is the sparsity factor; b is the bias factor.

[0014] According to an embodiment of the present disclosure, in step S30, processing the health value of the signal includes inverse processing the health value of the signal, which satisfies the following expression:

[0015] x(t+Δt)=W Q ×[z(t+Δt),t+Δt]

[0016] x(t+Δt) represents the predicted signal at the future time, W Q is the projection matrix, z(t+Δt) is determined by the following expression:

[0017]

[0018] W lh is a dimension-raising matrix, Indicates W lh Inverse Fourier transform of HI, W s is the transformation matrix.

[0019] According to an embodiment of the present disclosure, in step S50, the difference satisfies the following expression:

[0020]

[0021] RSE represents the difference, y i represents the signal obtained in step S40, represents the signal obtained in step S30, and ε represents a numerical correction constant.

[0022] According to an embodiment of the present disclosure, in step S60, the correction process conforms to the following expression:

[0023]

[0024] θ represents the set of parameters that need to be corrected in steps S20 and S30, m represents the number of iterations, λ represents the step size of the parameter update that needs to be corrected, and the parameters that need to be corrected include: dimensionality reduction matrix, signal matrix, sparse factor, bias factor, projection matrix, dimensionality increase matrix, and transformation matrix.

[0025] According to an embodiment of the present disclosure, the dimension reduction matrix includes a matrix of elements with h rows and l columns, and the initial value of each element is a random value of Gaussian distribution.

[0026] According to an embodiment of the present disclosure, the signal matrix includes a matrix of elements with c+1 rows and n columns, and the initial value of each element is a random value of Gaussian distribution.

[0027] According to an embodiment of the present disclosure, the initial value of the sparsity factor is a predetermined value.

[0028] According to an embodiment of the present disclosure, the initial value of the bias factor is a predetermined value.

[0029] According to an embodiment of the present disclosure, the projection matrix includes a matrix of elements with n+1 rows and c columns, and the initial value of each element is a random value of Gaussian distribution.

[0030] According to an embodiment of the present disclosure, the dimension-increased matrix includes a matrix of elements with l rows and h columns, and the initial value of each element is a random value of Gaussian distribution.

[0031] According to an embodiment of the present disclosure, the transformation matrix includes a matrix of elements with c rows and c columns, and an initial value of each element is a random value of Gaussian distribution.

[0032] According to the method for predicting the future operating state of a predetermined device provided by the present disclosure, signals in a data set are processed to determine the signal's health value, thereby determining a future signal at a predetermined time interval from the current time. Based on the difference between this signal and the future monitored signal at a predetermined time interval from the current time interval, the health value is corrected and updated to determine the future operating state of the monitored predetermined device. By correcting and updating the health value and future state based on the difference between the determined and monitored future signals, the accuracy and reliability of the predicted future operating state of the predetermined device are effectively improved, solving the problem of low prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0034] Figure 1 Schematically shows a flow chart of a method for predicting the future working state of a predetermined device according to an embodiment of the present disclosure;

[0035] Figure 2A Schematically illustrates an example diagram of mean square error under different noise levels according to an embodiment of the present disclosure;

[0036] Figure 2B Schematically illustrates an example diagram of root mean square error at different noise levels according to an embodiment of the present disclosure;

[0037] Figure 2C Schematically illustrates an example diagram of mean absolute error at different noise levels according to an embodiment of the present disclosure;

[0038] Figure 2D An example diagram schematically illustrates relative square errors at different noise levels according to an embodiment of the present disclosure; and

[0039] Figure 3 The flowchart of the dynamic update process of the prediction model according to the embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0040] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0041] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0042] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0043] When expressions such as "at least one of A, B and C, etc." are used, they should generally be interpreted in accordance with the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0044] During the implementation of this disclosure, it was discovered that current mechanical equipment signal monitoring typically relies on commercial software or deep learning-based network models. Commercial software relies on high-precision modeling of the equipment, resulting in slow computational speeds and inability to meet the real-time requirements of industrial equipment monitoring. Traditional deep learning-based models also have limited accuracy in modeling complex systems. Therefore, a method to improve the accuracy of equipment signal prediction is needed.

[0045] In view of this, embodiments of the present disclosure provide a method for predicting the future operating state of a predetermined device. The method includes: establishing a data set, wherein the data in the data set represents signals from real-time monitoring of the predetermined device, and adding new signals obtained from monitoring the predetermined device to the data set in real time; processing the signals in the data set to determine a health value of the signals; processing the health value of the signals to determine a future signal that will be a predetermined time interval from the current time interval; monitoring and obtaining a future signal that will be a predetermined time interval from the current time interval; comparing the determined future signal with the monitored future signal to determine a difference between the two; performing corrections based on the determined difference; determining an updated health value of the signal; and determining the future operating state of the monitored predetermined device based on the updated health value.

[0046] The following will be passed Figures 1 to 3 A method for predicting the future operating status of a predetermined device according to the disclosed embodiment is described in detail.

[0047] Figure 1 The flowchart of the method for predicting the future working status of a predetermined device according to an embodiment of the present disclosure is schematically shown.

[0048] like Figure 1 As shown, the method 100 includes operations S10 to S80.

[0049] In operation S10 , a data set is established, wherein the data in the data set represents signals of a predetermined device monitored in real time, and new signals obtained from monitoring the predetermined device are added to the data set in real time.

[0050] In operation S20 , the signals in the data set are processed to determine health values of the signals.

[0051] In operation S30 , the health value of the signal is processed to determine a future signal that is a predetermined time interval from the current signal.

[0052] In operation S40, a signal in the future that is a predetermined time interval from the current time is monitored.

[0053] In operation S50, the future signal determined in step S30 is compared with the future signal obtained by monitoring in step S40 to determine a difference between the two.

[0054] In operation S60 , steps S20 and S30 are modified according to the difference determined in step S50 .

[0055] In operation S70, an updated health value of the signal is determined.

[0056] In operation S80 , a future operating state of the monitored predetermined device is determined based on the updated health value.

[0057] According to an embodiment of the present disclosure, the data set represents a dynamic data set, and the data in the data set represents signals of a predetermined device monitored in real time, and new signals obtained from monitoring the predetermined device are added to the data set in real time.

[0058] According to the embodiments of the present disclosure, by collecting monitoring signals of predetermined devices and establishing a dynamic data set, it can be used to learn the health status of the system and dynamically monitor the future status.

[0059] According to an embodiment of the present disclosure, a signal in a data set is processed to determine a health value of the signal. The signal in the data set represents a monitoring signal at the current time. For example, the signal in the data set can be filtered and subjected to matrix dimensionality reduction processing to determine the health value of the signal.

[0060] According to an embodiment of the present disclosure, the health value of the signal is processed to determine a future signal at a predetermined time interval from the current time interval. This signal represents a predicted value of the future signal at a predetermined time interval from the current time interval.

[0061] According to an embodiment of the present disclosure, a signal of a predetermined device in the future at a predetermined time interval from the current time is monitored, and the signal represents an actual monitored value of the signal in the future at a predetermined time interval from the current time interval.

[0062] According to an embodiment of the present disclosure, by determining the difference between the predicted value of a future signal predetermined time interval from the current time interval and the actual monitored value, the process of determining the health value and predicting the future signal is corrected, and then the updated health value is determined, thereby determining the future working status of the monitored predetermined equipment.

[0063] According to the embodiments of the present disclosure, by correcting and updating the health value and future status based on the difference between the determined and monitored future signals, the accuracy and reliability of predicting the future working status of the predetermined equipment are effectively improved, and the problem of low prediction accuracy is solved.

[0064] According to an embodiment of the present disclosure, in step S10, establishing a data set further includes: in operation S11, using a sensor to obtain multiple real-time signals of a predetermined device over a predetermined time period; in operation S12, segmenting each of the multiple real-time signals; and in operation S13, assembling the segmented signals into a data set.

[0065] According to an embodiment of the present disclosure, a plurality of real-time signals of a predetermined device in a predetermined time period are collected by a sensor, and the predetermined time period may represent a time period from time t=0 to time T.

[0066] According to an embodiment of the present disclosure, for example, each of the multiple real-time signals is segmented, and the segmented signals form a data set D. Δt is the time interval. Each signal after segmentation includes N sampling points. t =x t1,t2,…,tN .

[0067] According to the embodiment of the present disclosure, the data set represents a dynamic data set, and the monitoring signal at the current time T is continuously collected by the sensor. in, The length of the dynamically updated monitoring signal is represented, and the predetermined time period from time t=0 to time T represents the dynamically updated monitoring signal. Each of the multiple real-time signals is segmented, and the segmented signals form a data set.

[0068] According to an embodiment of the present disclosure, in step S20 , a health value is generated according to the signal.

[0069] According to an embodiment of the present disclosure, the frequency domain of the signal at the current moment is thinned out by a soft threshold σ, and thinning out means that most elements of the frequency domain signal are zero.

[0070] In one embodiment, the soft threshold function performs frequency domain sparsification As shown in formula (1):

[0071]

[0072] Among them, α represents the sparse factor, b represents the bias factor, and W f Characterize the filter, represents the frequency domain signal, and z(t) represents the projection signal at the current time t.

[0073] According to an embodiment of the present disclosure, the sparsity factor is used to measure the sharpness of the threshold function, and the bias factor is used to characterize the learning bias.

[0074] According to an embodiment of the present disclosure, the signal x(t) at the current time t is projected at a corresponding time point according to the signal projection matrix to obtain a projection signal z(t).

[0075] In one embodiment, the projection signal z(t) is expressed as formula (2):

[0076]

[0077] Among them, W Prepresents the signal matrix, x(t) represents the signal at the current time t, c represents the signal channel, l represents the signal length, z(t) represents the projection signal at the current time t, × represents matrix multiplication, and [x(t), t] represents the block vectors of x(t) and (t).

[0078] According to an embodiment of the present disclosure, a sequence having the same dimension as the signal is constructed at a corresponding time point t by signal channel replication.

[0079] According to an embodiment of the present disclosure, Fourier transform is performed on z(t) to obtain a frequency domain signal In order to capture signals in a preset frequency range.

[0080] In one embodiment, the frequency domain signal As shown in formula (3):

[0081]

[0082] Wherein, ω represents the angular frequency of the projection signal z(t).

[0083] According to an embodiment of the present disclosure, the frequency domain signal is filtered Filter the predetermined frequencies in the

[0084] In one embodiment, the health value HI of the signal is as shown in formula (4):

[0085]

[0086] Among them, W hl Representation dimensionality reduction matrix, represents the frequency domain for the soft thresholding function to perform sparsification, and × represents the matrix multiplication.

[0087] According to an embodiment of the present disclosure, the health value representation of the signal is generated based on sparse frequency domain construction.

[0088] According to the embodiments of the present disclosure, a dynamic Fourier filter layer can be established to facilitate dynamic learning of a dynamic signal data set of a predetermined device, and the dynamic Fourier filter layer can be stacked to construct a complete signal monitoring operator network.

[0089] According to an embodiment of the present disclosure, in step S30 , processing the health value of the signal includes inversely processing the health value of the signal.

[0090] According to an embodiment of the present disclosure, the health value HI of the signal is converted into a time domain signal using inverse Fourier transform.

[0091] In one embodiment, the time domain signal As shown in formula (5):

[0092]

[0093] Where ω represents the angular frequency of the projection signal z(t), W lh Representation of the dimension-raising matrix, W lh HI represents the reconstructed frequency domain of the health value of the signal.

[0094] According to an embodiment of the present disclosure, a residual connection unit is established between dynamic Fourier filter layers to ensure that signals between network layers are not lost.

[0095] In one embodiment, the dynamic Fourier filter layer learns the monitoring signal to obtain a future learning signal z(t+Δt) with a predetermined time interval from the current time as shown in formula (6):

[0096]

[0097] Among them, W s Representation transformation matrix, W lh represents the dimension-raising matrix, [z(t),t] represents the block vectors of z(t) and (t), and b represents the bias parameter.

[0098] According to an embodiment of the present disclosure, a back-projection layer is established to project the learned signal z(t+Δt) into the predicted signal space to obtain the predicted signal x(t+Δt) at the future moment.

[0099] In one embodiment, the prediction signal x(t+Δt) is as shown in formula (7):

[0100] x(t+Δt)=W Q ×[z(t+Δt),t+Δt] (7)

[0101] Among them, W Q represents the projection matrix, and [z(t+Δt),t+Δt] represents the block vectors of z(t+Δt) and (t+Δt).

[0102] According to an embodiment of the present disclosure, the overall learning to prediction process of the dynamic Fourier filter layer is shown in formula (8):

[0103]

[0104] According to the embodiment of the present disclosure, the time interval Δt can be discretized more precisely, and the output signal process is obtained as shown in formula (9):

[0105]

[0106] Among them, δt represents the small time step, and ι represents the number of dynamic Fourier filter layers.

[0107] According to an embodiment of the present disclosure, the intermediate health value HI l ,HI l-1 ,...,HI 1 ,HI 0 It reflects the changes in health values at different time steps during Δt.

[0108] According to an embodiment of the present disclosure, the signal is processed by multiple Fourier filter layers. P Project the signal x(t) at the current time t to the corresponding time point t to obtain the projection signal z(t); perform Fourier transform on z(t) to obtain the frequency domain signal Through the filter W f For frequency domain signals The predetermined frequency in the filter is filtered, and then the filtered frequency domain signal is filtered by the soft threshold function σ Perform sparseness and finally according to the dimension reduction matrix W hl Get the health value of the signal.

[0109] According to an embodiment of the present disclosure, the health value HI of the signal is converted into a time domain signal using inverse Fourier transform. According to the transformation matrix W s , get the future learning signal z(t+Δt) with a predetermined time interval from the current time interval; in the back projection layer Q according to the projection matrix W Q , project the learned signal z(t+Δt) into the predicted signal space to obtain the predicted future signal x(t+Δt).

[0110] According to an embodiment of the present disclosure, in step S50, the relative square error is used as the loss function to obtain the difference RSE as shown in formula (10):

[0111]

[0112] Among them, y i Characterize the signal obtained in step S40, Characterizes the signal obtained in step S30, and ε represents a numerical correction constant.

[0113] According to an embodiment of the present disclosure, a numerical correction constant may be used to avoid the situation where the denominator is entirely zero.

[0114] According to an embodiment of the present disclosure, a future monitoring signal and a prediction signal with a predetermined time interval from the current time interval are processed according to a loss function to obtain a real-time dynamically updated signal difference prediction model.

[0115] Figure 2A An example diagram of mean square error under different noise levels according to an embodiment of the present disclosure is schematically shown.

[0116] like Figure 2A As shown, the horizontal axis represents the noise level, and the vertical axis represents the mean square error. Signals from predetermined devices are predicted using three models: a signal operator model 23 (SON), a long short-term memory neural network model 21 (LSTMN), and a U-net model 22. The mean square error (MSE) is calculated for each model under different noise levels.

[0117] Figure 2B An example diagram schematically illustrates root mean square error (RMSE) at different noise levels according to an embodiment of the present disclosure.

[0118] like Figure 2B As shown in the figure, the horizontal axis represents the noise level, and the vertical axis represents the root mean square error. The signals of the predetermined devices are predicted using three models: the signal operator model 23 (SON), the long short-term memory neural network model 21 (LSTMN), and the U-net model 22. The root mean square error (RMSE) is calculated for each model under different noise levels.

[0119] Figure 2C An example diagram of mean absolute errors at different noise levels according to an embodiment of the present disclosure is schematically shown.

[0120] like Figure 2C As shown in the figure, the horizontal axis represents the noise level, and the vertical axis represents the mean absolute error. The signal of the predetermined device is predicted using three models: the signal operator model 23 (SON), the long short-term memory neural network model 21 (LSTMN), and the U-net model 22. The mean absolute error (MAE) is calculated for each model under different noise levels.

[0121] Figure 2D An example diagram of relative square errors at different noise levels according to an embodiment of the present disclosure is schematically shown.

[0122] like Figure 2D As shown in the figure, the horizontal axis represents the noise level, and the vertical axis represents the relative squared error. The signals of the predetermined devices are predicted using three models: the signal operator model 23 (SON), the long short-term memory neural network model 21 (LSTMN), and the U-net model 22. The relative squared error (RSE) is calculated for each model under different noise levels.

[0123] According to the embodiments of the present disclosure, the signals of the predetermined equipment are predicted respectively by three models, and then different results are obtained by processing the predicted values and the monitored values through different error functions. It can be seen that the processing results of the signal operator model and the relative square error RSE are better. Therefore, the signal operator model is selected as the prediction model for predicting the signals of the predetermined equipment.

[0124] According to an embodiment of the present disclosure, in step S60, correction is performed based on the determined signal difference, and the correction process is shown in formula (11):

[0125]

[0126] Here, m represents the number of iterations, λ represents the step size of the parameter update that needs to be corrected, represents the numerical correction constant, and θ represents the parameter set that needs to be corrected in steps S20 and S30.

[0127] According to an embodiment of the present disclosure, the parameters that need to be corrected include: a dimensionality reduction matrix, a signal matrix, a sparse factor, a bias factor, a projection matrix, a dimensionality increase matrix, and a transformation matrix.

[0128] According to the embodiment of the present disclosure, by dynamically updating the data set D and the task of training the signal prediction model, a real-time dynamically updated signal difference prediction model RSE is obtained, and then it is corrected according to the determined signal difference, and continuously iterated and updated until the optimization objective function converges.

[0129] In one embodiment, the signal prediction model G is shown in formula (12):

[0130]

[0131] in, Characterization mapping symbol, t p Represent the prediction moment.

[0132] In one embodiment, the optimization objective function is shown in formula (13):

[0133]

[0134] According to an embodiment of the present disclosure, Characterize the signal prediction model with optimal parameters.

[0135] According to an embodiment of the present disclosure, x~u represents an independent and identically distributed sequence generated from a probability measure on X.

[0136] Figure 3 The flowchart of the dynamic update of the prediction model according to the embodiment of the present disclosure is schematically shown.

[0137] like Figure 3As shown, in step S310, a dynamic dataset of monitoring data is established by dynamically updating dataset D. In step S320, a predicted signal for a predetermined time interval relative to the current time interval is obtained using the signal prediction model. Based on the difference between the monitored signal and the predetermined time interval relative to the current time interval, a real-time, dynamically updated signal difference prediction model RSE is obtained. In step S330, corrections are made based on the determined signal difference, and the parameters of the signal prediction model are continuously iteratively updated until the optimization objective function converges. In this process, the dataset is continuously and dynamically updated, and the parameters of the prediction model are iteratively updated, thereby dynamically predicting the future operating status of the predetermined equipment.

[0138] According to an embodiment of the present disclosure, the dimension reduction matrix includes a matrix of elements with h rows and l columns, and the initial value of each element is a random value of Gaussian distribution.

[0139] According to an embodiment of the present disclosure, the dimension reduction matrix may be a matrix from a high dimension to a low dimension, and the initial value of each element in the dimension reduction matrix is a random value of a Gaussian distribution.

[0140] According to an embodiment of the present disclosure, the signal matrix includes a matrix of elements with c+1 rows and n columns, and the initial value of each element is a random value of Gaussian distribution.

[0141] According to an embodiment of the present disclosure, the initial value of the sparsity factor is a predetermined value.

[0142] According to an embodiment of the present disclosure, the initial value of the bias factor is a predetermined value.

[0143] According to an embodiment of the present disclosure, the projection matrix includes a matrix of elements with n+1 rows and c columns, and the initial value of each element is a random value of Gaussian distribution.

[0144] According to an embodiment of the present disclosure, the dimension-increased matrix includes a matrix of elements with l rows and h columns, and the initial value of each element is a random value of Gaussian distribution.

[0145] According to an embodiment of the present disclosure, the dimension-raising matrix may be a matrix from a low dimension to a high dimension, and the initial value of each element in the dimension-raising matrix is a random value of a Gaussian distribution.

[0146] According to an embodiment of the present disclosure, the transformation matrix includes a matrix of elements with c rows and c columns, and an initial value of each element is a random value of Gaussian distribution.

[0147] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways, even if such combinations and / or couplings are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or couplings are intended to fall within the scope of this disclosure.

[0148] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for predicting the future working state of a predetermined device, characterized in that: The following steps are involved: S10: Establishing a data set, wherein the data in the data set represents the signals of the predetermined device monitored in real time, and new signals obtained from monitoring the predetermined device are added to the data set in real time; S20: Processing the signals in the data set to determine the health value of the signals; S30: Processing the health value of the signal to determine a future signal with a predetermined time interval from the current signal; S40: monitoring and obtaining a future signal with a predetermined time interval from the current time; S50: comparing the future signal determined in step S30 with the future signal monitored in step S40 to determine a difference between the two; S60: According to the difference determined in step S50, step S20 and step S30 are modified; S70: Determine an updated health value of the signal; S80: Determine a future operating state of the monitored predetermined device based on the updated health value; Among them, in step S20, The health value of the signal satisfies the following expression: Where HI represents the health value of the signal, t represents the time, x(t) represents the signal at the current time, and z(t) satisfies the following expression: represents the Fourier transform of z(t); W f Represents a filter, Filter the predetermined frequencies in σ satisfies the following expression: W hl is the dimension reduction matrix; W P is the signal matrix; c represents the signal channel, l represents the signal length; α is the sparsity factor; b is the bias factor.

2. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: In step S10, the following is also included: S11: using a sensor to obtain a plurality of real-time signals of the predetermined device in a predetermined time period; S12: Segmenting each of the multiple real-time signals; S13: The segmented signals form the data set.

3. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: In step S30, The processing of the health value of the signal includes performing inverse processing on the health value of the signal, which satisfies the following expression: x(t+Δt)=W Q ×[z(t+Δt),t+Δt] x(t+Δt) represents the predicted signal at the future time, W Q is the projection matrix, z(t+Δt) is determined by the following expression: W lh is a dimension-raising matrix, Indicates W lh Inverse Fourier transform of HI, W s is the transformation matrix.

4. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: In step S50, The difference satisfies the following expression: RSE represents the difference, y i represents the signal obtained in step S40, represents the signal obtained in step S30, and ε represents a numerical correction constant.

5. The method for predicting the future working state of a predetermined device according to claim 4, characterized in that: In step S60, the correction process conforms to the following expression: θ represents the set of parameters that need to be corrected in steps S20 and S30, m represents the number of iterations, and λ represents the step size of the parameter update that needs to be corrected. The parameters that need to be corrected include: dimensionality reduction matrix, signal matrix, sparse factor, bias factor, projection matrix, dimensionality increase matrix, and transformation matrix.

6. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: The dimension reduction matrix includes a matrix of elements with h rows and l columns, and the initial value of each element is a random value of Gaussian distribution.

7. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: The signal matrix includes a matrix of elements with c+1 rows and n columns, and an initial value of each element is a random value of Gaussian distribution.

8. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: The initial value of the sparse factor is a predetermined value.

9. The method for predicting the future working state of a predetermined device according to claim 1, characterized in that: The initial value of the bias factor is a predetermined value.

10. The method for predicting the future working state of a predetermined device according to claim 3, characterized in that: The projection matrix includes a matrix of elements with n+1 rows and c columns, and an initial value of each element is a random value of Gaussian distribution.

11. The method for predicting the future working state of a predetermined device according to claim 3, characterized in that: The dimension-raising matrix includes a matrix of elements with l rows and h columns, and the initial value of each element is a random value of Gaussian distribution.

12. The method for predicting the future working state of a predetermined device according to claim 3, characterized in that: The transformation matrix includes a matrix of elements with c rows and c columns, and the initial value of each element is a random value of Gaussian distribution.

Images