Adaptive timestamp coding enhanced complex equipment incomplete state monitoring data time sequence interpolation method

Abstract

The invention discloses an adaptive timestamp coding enhanced complex equipment incomplete state monitoring data time sequence interpolation method, and relates to the technical field of industrial sensing data processing, and the method comprises the steps: carrying out the standardization of obtained online incomplete state monitoring time sequence data, carrying out the normalization processing of online timestamp sequence data, and carrying out the normalization processing of the online timestamp sequence data; then, a timestamp prior feature coding module is used for outputting online timestamp prior coding features, and effective utilization of timestamp information in the monitoring data in the incomplete state in an interpolation task is achieved; using the trained timestamp-local interpolation model joint framework to obtain an online interpolation result of the monitoring data in the incomplete state; according to the scheme, advantage complementation between a locality time sequence interpolation model and a timestamp probabilistic interpolation model is achieved, the accuracy and robustness of complex equipment state monitoring data time sequence interpolation are effectively improved, and the method can be widely applied to a data enhancement link in various complex equipment operation state analysis tasks.

Description

Technical Field

[0001] The present application relates to the technical field of industrial sensor data processing, and in particular to a method for time series interpolation of complex equipment incomplete state monitoring data enhanced by adaptive timestamp coding. Background Art

[0002] In the field of industrial intelligent operation and maintenance in the era of big data, the condition monitoring data of complex equipment is widely used in key tasks such as predicting equipment operating performance, fault diagnosis, and life prediction. Among them, the temporal integrity of condition monitoring data is the key prerequisite to ensure the accurate execution of these tasks. However, in actual industrial scenarios, due to the combined influence of multiple factors such as sensor failure, communication interference, and equipment failure shutdown, the collected complex equipment condition monitoring data is often partially missing, which interferes with the accurate perception and judgment of the equipment status, and thus affects the equipment operation decision-making and predictive maintenance effect. Therefore, effective interpolation of the incomplete condition monitoring data of complex equipment can not only restore the temporal continuity of the data, but also improve the data quality, thereby enhancing the reliability of fault diagnosis and performance prediction, improving equipment safety, and ensuring the efficient operation of complex equipment. Among the many interpolation methods, the interpolation method based on deep learning can more effectively model and make full use of the nonlinear correlation characteristics in multivariate condition monitoring data. It has the advantages of strong representation ability and suitability for high-dimensional data. In recent years, it has been particularly widely used in the temporal interpolation of complex equipment condition monitoring data.

[0003] The existing mainstream deep learning time series interpolation methods usually continuously intercept a section of incomplete state monitoring data of a specified local time window size in the time dimension as input, infer the missing data based on the correlation characteristics between the visible data and the missing data, and complete the time series interpolation. Although this type of local interpolation method has achieved good results, its use of timestamp series data without missing problems in complex equipment status monitoring data is still not sufficient. On the one hand, most interpolation methods are limited to using timestamps to perform position encoding or calculate missing time lag matrices within the local time window corresponding to a single sample to indicate the relative position relationship of input data features in the time dimension. There is a lack of mining of the time series features contained in the timestamp series data itself, and it is impossible to directly model the macro-scale periodic time series association pattern beyond the local window range. On the other hand, some existing methods embed and encode timestamp information based on several dominant cycles of complex equipment time series data obtained based on prior knowledge, and fuse it with the input time series data features to improve the interpolation performance, but it is highly dependent on domain prior knowledge and is difficult to generalize to different industrial fields. In summary, the lack of automatic and effective use of timestamp information makes it difficult for existing interpolation methods to adapt to the time series interpolation problem of complex equipment status monitoring data in scenarios where there is a lack of prior knowledge, uneven data missing, and severe lack of observable data in some samples. Summary of the invention

[0004] The purpose of this application is to provide a time series interpolation method for complex equipment incomplete status monitoring data enhanced by adaptive timestamp coding, which can effectively utilize the timestamp information in the complex equipment status monitoring data and improve the accuracy and robustness of the interpolation method.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] The present application provides a method for time series interpolation of complex equipment incomplete state monitoring data enhanced by adaptive timestamp coding, comprising the following steps:

[0007] Obtain online incomplete state monitoring data of complex equipment; the online incomplete state monitoring data includes online incomplete state monitoring time series data of complex equipment and corresponding online timestamp sequence data; the online incomplete state monitoring time series data is the monitoring values ​​of several state monitoring variables of complex equipment under multiple continuous sampling steps.

[0008] The online defect state monitoring time series data is standardized to obtain preprocessed online defect state monitoring time series data.

[0009] After normalizing the online timestamp sequence data, the online timestamp prior coding features are outputted using the timestamp prior feature coding module; the parameters of the timestamp prior feature coding module are obtained based on an unsupervised frequency domain prior mining model.

[0010] The preprocessed online incomplete state monitoring time series data and online timestamp prior encoding features are input into the trained timestamp-local interpolation model joint framework to obtain the online interpolation results of the incomplete state monitoring data; the timestamp-local interpolation model joint framework is a joint framework based on the timestamp probabilistic interpolation model and the local time series interpolation model.

[0011] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0012] The present application provides a method for time series interpolation of complex equipment incomplete state monitoring data enhanced by adaptive timestamp coding. In the method, for the acquired online incomplete state monitoring data of the complex equipment, the online incomplete state monitoring time series data is first standardized to obtain preprocessed online incomplete state monitoring time series data, and then the online timestamp sequence data is normalized, and then the online timestamp prior feature coding module is used to output the online timestamp prior coding feature; the parameters of the timestamp prior feature coding module here are obtained based on the unsupervised frequency domain prior mining model, which realizes the timestamp information in the incomplete state monitoring data in the interpolation task Effective utilization of the preprocessed online incomplete state monitoring time series data and online timestamp prior coding features are then input into the trained timestamp-local interpolation model joint framework to obtain the online interpolation results of the incomplete state monitoring data; the timestamp-local interpolation model joint framework here is a joint framework based on the timestamp probabilistic interpolation model and the local time series interpolation model, which realizes the complementary advantages between the local time series interpolation model and the timestamp probabilistic interpolation model, effectively improving the accuracy and robustness of the time series interpolation of complex equipment state monitoring data, and can be widely used in the data enhancement link of various complex equipment operation status analysis tasks. The above-mentioned scheme of the present application realizes the adaptive feature encoding of the macro-scale inherent periodic time series pattern information of complex equipment corresponding to the timestamp information without introducing any external prior knowledge, and effectively enhances the interpolation performance of the existing local time series interpolation model through the design of the joint framework, which is suitable for the time series interpolation problem of complex equipment state monitoring data in various scenarios where complex equipment state monitoring data is missing. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A flowchart of a method for temporal interpolation of complex equipment incomplete status monitoring data enhanced by adaptive timestamp coding provided in one embodiment of the present application.

[0015] Figure 2 A flowchart of training and optimizing the timestamp-local interpolation model joint framework parameters in a complex equipment defect status monitoring data time series interpolation method enhanced by adaptive timestamp coding provided in one embodiment of the present application.

[0016] Figure 3This is a flowchart of step B1 in a method for temporal interpolation of complex equipment incomplete status monitoring data enhanced by adaptive timestamp coding provided in one embodiment of the present application.

[0017] Figure 4 This is a flowchart of step B4 in a method for temporal interpolation of complex equipment incomplete status monitoring data enhanced by adaptive timestamp coding provided in one embodiment of the present application.

[0018] Figure 5 This is a flowchart of step B42 in a method for temporal interpolation of complex equipment incomplete status monitoring data enhanced by adaptive timestamp coding provided in one embodiment of the present application.

[0019] Figure 6 A schematic diagram of the technical concept of a method for timing interpolation of complex equipment incomplete status monitoring data enhanced by adaptive timestamp coding provided in one embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0022] The embodiment of the present application provides a method for time series interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding, such as Figure 1 The method comprises the following steps:

[0023] A1. Obtain online incomplete state monitoring data of complex equipment; the online incomplete state monitoring data includes online incomplete state monitoring time series data of complex equipment and corresponding online timestamp sequence data; the online incomplete state monitoring time series data is the monitoring values ​​of several state monitoring variables of complex equipment under multiple continuous sampling steps.

[0024] A2. Standardize the online defect state monitoring time series data to obtain preprocessed online defect state monitoring time series data.

[0025] A3. After normalizing the online timestamp sequence data, the online timestamp prior coding features are outputted using the timestamp prior feature coding module; the parameters of the timestamp prior feature coding module are obtained based on an unsupervised frequency domain prior mining model.

[0026] A4. Input the preprocessed online incomplete state monitoring time series data and the online timestamp prior coding features into the trained timestamp-local interpolation model joint framework to obtain the online interpolation results of the incomplete state monitoring data; the timestamp-local interpolation model joint framework is a joint framework obtained based on the timestamp probabilistic interpolation model and the local time series interpolation model.

[0027] Specifically in this embodiment, before step A3, the adaptive timestamp coding enhanced complex equipment defective state monitoring data time series interpolation method provided in this embodiment also includes a process of training and optimizing the parameters of the timestamp-local interpolation model joint framework, such as Figure 2 As shown, the following steps are included:

[0028] B1. Based on the offline monitoring data of the historical scenario of complex equipment, construct an interpolation data training set, an interpolation data verification set, and an interpolation data test set. The interpolation data training set is used to train the timestamp-local interpolation model joint framework; the interpolation data verification set and the interpolation data test set are used to verify, optimize, and evaluate the performance of the trained timestamp-local interpolation model joint framework. In this embodiment, Figure 3 As shown, step B1 specifically includes the following steps:

[0029] B11. Obtain the initial incomplete state monitoring data of complex equipment; the initial incomplete state monitoring data includes the offline incomplete state monitoring time series data of the complex equipment and the corresponding offline timestamp sequence data; the offline incomplete state monitoring time series data is the monitoring values ​​of several state monitoring variables of the complex equipment under multiple continuous sampling steps in historical scenarios.

[0030] B12. After normalizing the offline timestamp sequence data, non-overlapping data division is performed based on the initial incomplete state monitoring data to obtain initial training data, initial verification data, and initial test data.

[0031] B13. Standardize the offline incomplete state monitoring time series data in the initial training data, the initial verification data and the initial test data to obtain preprocessed initial training data, preprocessed initial verification data and preprocessed initial test data.

[0032] B14. Using local time windows of the same size, sliding window operations are performed on the preprocessed initial training data, the preprocessed initial verification data, and the preprocessed initial test data to construct interpolated data training sets, interpolated data verification sets, and interpolated data test sets; any sample in the interpolated data training sets, interpolated data verification sets, and interpolated data test sets includes offline incomplete local state monitoring time series data, incomplete local state monitoring time series data with high missing rate obtained by artificial masking operations on offline incomplete local state monitoring time series data, and local timestamp series data.

[0033] B2. Input the high-quality offline monitoring data in the historical scenario into the frequency domain priori mining model, and output the prior core frequency information and prior phase information corresponding to each state monitoring variable based on the cluster labels of the state monitoring variables. Specifically in this embodiment, the offline incomplete state monitoring time series data in the preprocessed initial verification data is used as the high-quality offline monitoring data in the historical scenario, and input into the frequency domain priori mining model, and output the prior core frequency information and prior phase information corresponding to each state monitoring variable based on the cluster labels of the state monitoring variables.

[0034] B3. Determine the parameters of the timestamp prior feature encoding module based on the prior core frequency information and prior phase information corresponding to each state monitoring variable.

[0035] B4. Use the interpolation data training set to train the timestamp-local interpolation model joint framework to obtain a trained timestamp-local interpolation model joint framework. Figure 4 As shown, step B4 specifically includes the following steps:

[0036] B41. Input the local timestamp sequence data in the interpolation data training set into the timestamp prior feature encoding module to obtain the offline timestamp prior encoding feature.

[0037] B42. Input the offline timestamp prior coding features into the timestamp probabilistic interpolation model to obtain the local prior distribution mean estimate, the local prior distribution standard deviation estimate and the timestamp interpolation result.

[0038] B43, inputting the incomplete local state monitoring time series data with high missing rate in the interpolation data training set into the local time series interpolation model to obtain the local interpolation result. In this embodiment, step B43 specifically includes the following steps:

[0039] B431. Construct missing value mask by detecting the missing data location information of time series data with high missing rate and incomplete local state monitoring.

[0040] B432. Input the incomplete local state monitoring time series data with high missing rate and the missing value mask into the local time series interpolation model to obtain the local interpolation result.

[0041] B44. Determine the final interpolation result based on the timestamp interpolation result and the local interpolation result.

[0042] B45. Construct a joint loss function based on the incomplete local state monitoring time series data with high missing rate, the estimated value of the local prior distribution mean, the estimated value of the local prior distribution standard deviation and the final interpolation result.

[0043] B46. According to the joint loss function, the timestamp-local interpolation model joint framework is trained based on gradient back propagation to obtain a trained timestamp-local interpolation model joint framework.

[0044] In an exemplary embodiment, a timestamp probabilistic interpolation model includes: a distribution coding module, a distribution aggregation module and a reparameter sampling module; the distribution coding module includes a plurality of parameter-independent distribution encoders; the distribution encoder is used to process state monitoring variables corresponding to cluster labels; the distribution encoder includes a plurality of parameter-independent distribution estimation networks, which correspond to each core frequency component of the state monitoring variable respectively.

[0045] In this embodiment, if Figure 5 As shown, step B42 specifically includes the following steps:

[0046] B421. Input the offline timestamp prior encoding features into the corresponding distribution encoder, and output the distribution parameters corresponding to each core frequency component; the distribution parameters include the mean estimate and the logarithmic variance estimate.

[0047] B422. Use the distribution aggregation module to aggregate the mean estimate and log variance estimate corresponding to each core frequency component, and output the local prior distribution mean estimate and the local prior distribution log variance estimate.

[0048] B423. Transform the estimated value of the logarithmic variance of the local prior distribution into an estimated value of the standard deviation of the local prior distribution, and obtain an auxiliary noise quantity by random sampling from the standard normal distribution.

[0049] B424. Based on the local prior distribution mean estimate, the local prior distribution standard deviation estimate and the auxiliary noise amount, sampling is performed to obtain the timestamp interpolation result.

[0050] After obtaining the trained timestamp-local interpolation model joint framework through the above process, it also includes:

[0051] B5. Use the interpolation data validation set and the interpolation data test set to verify and optimize the trained timestamp-local interpolation model joint framework and evaluate its performance. The way in which each sample in the interpolation data validation set and the interpolation data test set is input into the trained timestamp-local interpolation model joint framework is the same as that of each sample in the interpolation data training set. The final interpolation results output are compared with the offline incomplete local state monitoring time series data in each sample, and the interpolation error is calculated to achieve verification optimization and performance evaluation of the proposed interpolation model joint framework.

[0052] The above embodiment illustrates a method for time series interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding provided by the present application from the perspective of practical application. The following is a specific example to introduce the above method provided by the present application from the perspective of theoretical derivation. The overall technical concept is as follows: Figure 6 shown.

[0053] As a complete set of power machinery with continuously flowing gas as the working fluid, the gas turbine is composed of multiple components such as the compressor, combustion chamber, and gas turbine. It has the characteristics of complex structure and highly coupled performance between sub-components. It is known as the "crown jewel" of the manufacturing industry and is a very representative complex equipment. With the development of advanced sensing technology, the Internet of Things and other technologies, a large amount of gas turbine status monitoring data including output power, blade channel temperature, exhaust pressure, combustion chamber casing internal pressure, spindle speed, etc. can be collected and analyzed. However, due to the influence of factors such as the extreme operating environment of high temperature and high pressure and complex and changeable operating conditions, the collected gas turbine status monitoring data is generally missing. Time series interpolation of the incomplete status monitoring data of the gas turbine is conducive to improving the integrity and continuity of the gas turbine operating status data, and providing a reliable data basis for the equipment's operating performance analysis and predictive maintenance.

[0054] In this embodiment, by cooperating with relevant enterprises, historical status monitoring data of a certain type of gas turbine in real service status within a period of time is obtained, including the value of each data point itself and the corresponding timestamp information. Due to sensor failure, data transmission and storage failure, etc., some status monitoring data is missing. First, the following data preprocessing steps are required for the incomplete status monitoring data of the gas turbine:

[0055] The gas turbine defective state monitoring data are sorted in order in terms of time and variable dimensions to obtain the initial defective state monitoring data Z of complex equipment. raw ={X raw ,TS raw}. X raw Represents the initial value of each state monitoring time series data, some of which are missing, and the size is T full ×C, where Tfull represents the total time steps of the continuous uniform sampling process, C = 199 represents the total number of state monitoring variables, TS raw The timestamp sequence data representing the initial incomplete state monitoring time series data, whose size is T full ×1, there is no missing phenomenon.

[0056] Since in this embodiment, the encoding of the timestamp sequence data is adaptively determined by a learnable model, and the prior time period information such as year, month, and day is not directly used, in order to unify the input of the timestamp prior feature encoding model and facilitate subsequent calculations, the timestamp sequence data in the initial incomplete state monitoring data of complex equipment needs to be normalized. That is, after converting each timestamp into a long integer original timestamp value, a uniform translation and scaling is performed according to the following formula, so that the normalized sampling frequency f s becomes 1:

[0057]

[0058] in, Represents normalized timestamp information, and They represent the minimum and second minimum values ​​of all original timestamp values ​​respectively, and s represents the difference between the original timestamp values ​​corresponding to adjacent time steps.

[0059] After completing the normalization of the timestamp series data, the initial value X of the state monitoring time series data is processed in the time dimension according to the ratio of 5:1:1 among the training set, validation set and test set. raw and normalized timestamp information Divide the data in order without overlap to get the initial training data Initial Validation Data Initial test data The basic information of various types of status monitoring data after statistical division is shown in Table 1 below.

[0060] Table 1 Basic information of gas turbine initial defective state monitoring data

[0061]

[0062] In addition, according to statistics, the initial missing rate of various types of data is less than 5%, and the data integrity is relatively high. In particular, the relative integrity of the verification data is conducive to ensuring the accuracy of the core frequency-related information of each state monitoring variable obtained by the subsequent frequency domain prior mining model learning.

[0063] Then calculate the initial training data feature sequence The mean μ and standard deviation σ of the time series data corresponding to each state monitoring variable in (both sizes are 1×C), and at the same time Perform independent standardization operations on each channel in the time dimension, convert the monitoring data of each channel into a standard normal distribution with a mean of 0 and a standard deviation of 1, and obtain three sets of preprocessed time series data feature sequences

[0064] Then T s =15 is the local time window size, L=15 is the step size, and the three sets of preprocessed time series data feature sequences are respectively And three sets of normalized timestamp sequences Perform sliding window operation simultaneously to construct interpolation data training set Z train , interpolation data validation set Z val and the imputed data test set Z test At this time, each sample in the above three sample sets contains a set of actual, i.e., offline incomplete local state monitoring time series data. (Size is T s ×C) and a set of local time stamp series data (Size is T s ×1).

[0065] In order to solve the problem of model verification and evaluation, it is necessary to specify appropriate artificial missing types and perform artificial masking operations on the actual incomplete local state monitoring time series data in each sample set. The results of the visualization analysis show that the actual missing pattern of the gas turbine state monitoring time series data in this example is mainly random discrete point missing, and occasionally local continuous block missing in the variable and time dimensions. In order to comprehensively evaluate the performance of the model under different missing rates and missing modes, four missing types are designed here, and their relevant information is summarized in Table 2 below. The subsequent training, verification, and testing of the interpolation model are carried out separately under each group of missing types to comprehensively evaluate the time series interpolation method proposed in this application.

[0066] Table 2 Basic information of four missing types

[0067]

[0068] A set of missing types is selected from Table 2, and the actual incomplete local state monitoring time series data of each sample in the sample set is analyzed according to its missing mode and missing parameters. Perform artificial masking operations to obtain high-missing rate incomplete local state monitoring time series data X local (Size is T s ×C), and also regard it as part of the sample. Finally, the interpolation data training set samples , imputation data validation set samples , imputation data test set samples All are included in the time dimension alignment Three parts of data.

[0069] After completing the preprocessing of the above-mentioned gas turbine incomplete state monitoring data, the effectiveness of the time series interpolation method for complex equipment incomplete state monitoring data enhanced by adaptive timestamp coding provided in this application can be verified.

[0070] First, for a section of high-quality historical state monitoring data (as defined in step B2 of the above embodiment, the offline incomplete state monitoring time series data in the pre-processed initial verification data can be used as the high-quality offline monitoring data in the historical scenario), a fixed large-size time window W is used along the time dimension. L = 256 non-overlapping random sampling, and obtain L = 5 long sequence time series samples , where a single sample The size is W L ×C. Then, the data of each channel in each sample is discretely transformed to calculate the amplitude spectrum and phase spectrum of each time series data in the frequency domain. Since the high-quality initial verification data may still be incomplete data, the above Fourier transform problem can be regarded as a non-uniform Fourier transform with non-uniform time domain and uniform frequency domain, as shown in the following formula:

[0071]

[0072] t n =n o / W L ,(0≤n o <N o ).

[0073] Among them, X(k) represents the discrete Fourier transform result of the incomplete time series data x(n), N o Represents the total number of non-missing data in the incomplete time series data, n o Represents the relative position index of a specific non-missing data point in the specified sequence x(n) in the time dimension.

[0074] The interpolation-based non-uniform fast Fourier transform algorithm can be used to solve the frequency domain expression of each incomplete time series data, and then calculate the normalized single-sided amplitude spectrum c k and the one-sided phase spectrum φ k , as shown below:

[0075] X(k)=NUFFT(x(n)).

[0076]

[0077]

[0078] Among them, c k and φ kBoth discard negative frequencies and DC components, and the size is L×W L / 2 ×C,W L / 2 =W L / 2-1,X re and X im Represent the real and imaginary parts of the frequency domain features, respectively, X(k|1:(W L / 2-1)) means to intercept the number 1 in X(k) until the number (W L / 2-1) is a subsequence of elements.

[0079] The normalized single-sided amplitude spectrum c of each sample k Take the average and get the average unilateral amplitude spectrum feature c k , whose size is W L / 2 ×C. Based on the average unilateral amplitude spectrum features, construct the amplitude spectrum sparse mask M A , size W L / 2 ×C, where the corresponding position of the Z=32 frequency components with the largest amplitude in each variable takes the value 1, and the other positions take the value 0. The sparse amplitude feature H is obtained by multiplying the amplitude spectrum sparse mask and the average unilateral amplitude spectrum feature element by element. A-s , in order to achieve zeroing of the noise frequency amplitude, as shown in the following formula:

[0080]

[0081] For the sparse amplitude feature H A-s After performing Z-Score normalization in the frequency dimension, principal component analysis (PCA) is used to reduce the dimension and obtain the compressed amplitude feature. , its size is M L ×C, where M L =4 is the number of low-dimensional main features retained in principal component analysis. Then, the number of clusters G=10 is specified, based on the compressed amplitude feature Perform K-means clustering on each state monitoring variable to obtain the state monitoring variable cluster label Y L (size is C) and the index of each cluster center status monitoring variable I C (Size G).

[0082] Considering that the total number of variables for complex equipment condition monitoring is large, and there are significant local shared correlation patterns between many variables, in order to simplify the analysis process, it is assumed here that all condition monitoring variables under the same cluster share the same set of core frequency components, and the periodic a priori time feature differences between these variables can be expressed as the phase difference corresponding to each core frequency. Under this assumption, the TOP-K selection strategy is used to select the K = 4 frequencies with the largest amplitude in the sparse amplitude features corresponding to each cluster center variable as the cluster center frequency. And share it among the variables in the same cluster, and finally get the prior core frequency information F c , whose size is K×C. At the same time, the phase of each core frequency component in each cluster center variable is regarded as the phase origin, and the unilateral phase spectrum φ corresponding to each variable and its corresponding cluster center variable is calculated in each sample. k The difference between the samples is averaged to obtain the average phase feature . Filter out the phase value corresponding to the class center frequency position and obtain the prior phase information Φ c , whose size is K×C.

[0083] The timestamp prior feature encoding module uses the normalized timestamp data at any time step T (size is T×1) as input, and the prior core frequency information F corresponding to each state monitoring variable is c and the prior phase information Φ c As auxiliary parameters, the encoding functions are constructed to perform sine position encoding and cosine position encoding respectively, and the two encoding results are stacked to output the timestamp prior encoding feature H TS , whose size is T×2×K×C, and the above encoding stacking process is shown in the following formula.

[0084]

[0085] H TS (i,:,j,k)=[H sin (i,j,k),H cos (i,j,k)]

[0086] where 0≤i<T,0≤j<K,0≤k<C

[0087] Among them, [·] is the concatenation operation.

[0088] Timestamp prior feature encoding module f s (·) can be expressed in the following forms:

[0089]

[0090] In the scheme of the present application, assuming that the prior core frequency and the corresponding phase information are known, the probability of the value of a specific state monitoring variable c at a single timestamp TS corresponding to a time point is recorded as the state random variable X TS,c , which obeys the normal distribution with independent variables and independent time series:

[0091]

[0092] in, They respectively represent the prior core frequency information and prior phase information corresponding to the state monitoring variable c.

[0093] Timestamp probabilistic interpolation model f d (·) Taking the timestamp prior encoding feature as input, output the probability distribution parameter estimate of the value of the specified state monitoring variable at the specified time point, that is, the single point prior distribution mean estimate is obtained and a single-point prior distribution standard deviation estimate , as shown below:

[0094]

[0095] Here Represents normalized timestamp data at a single time step.

[0096] According to the estimated distribution parameters, variable-independent and time-series-independent sampling is performed, and then the sampling results of all variables within the specified time step range are merged to obtain the single interpolation result of the timestamp probabilistic interpolation model.

[0097] To achieve the above functions, the timestamp probabilistic interpolation model consists of three parts: distribution encoding module, distribution aggregation module and reparameter sampling module.

[0098] (1) Distributed coding module:

[0099] The distribution coding module contains a total of G groups of independent distribution encoders, each of which processes a variable under a specific cluster label. A group of distribution encoders contains K independent distribution estimation networks, corresponding to each core frequency component. Each distribution estimation network is a single hidden layer multilayer perceptron structure with an input dimension i=2, a hidden layer dimension h=4, an output dimension o=2, and a hidden layer activation function, which selects the GELU function. Its function is to transform the sine and cosine position encoding results corresponding to a core frequency component into distribution parameter component estimates. The estimated distribution parameters include component means. and the logarithmic variance of the components

[0100]

[0101] in, It represents the timestamp prior encoding feature at a specific time step t and a specific variable c, and its size is 2 × K. g is the cluster label value corresponding to the variable c. Represents the concatenated vector of estimated values ​​of the component distribution parameters corresponding to the k-th core frequency component. (g),1 (·) and Linear (g),2 (·) represents two independent linear layers corresponding to the cluster label value g, GELU(·) represents the use of GELU nonlinear activation function, Represents interception of a two-dimensional matrix The second dimension index is k for all data.

[0102] The component distribution parameter estimation vectors obtained based on each core frequency component are stacked to obtain the distribution parameter component estimation feature As the output of the module, its size is 2×K.

[0103] (2) Distribution aggregation module:

[0104] The distribution encoding module assumes that the single-point component value of the state monitoring variable under the condition of known prior information of a single core frequency follows a normal distribution, and estimates its distribution parameters. In the distribution aggregation module, it is assumed that the state monitoring variable value under the condition of known prior information of all core frequencies can be obtained by the state monitoring variable component value corresponding to each core frequency according to the combination operation g(·,...,·) including addition or multiplication, as shown in the following formula, and its distribution still maintains a normal distribution.

[0105]

[0106] in, They respectively represent the prior core frequency information and prior phase information corresponding to the i-th core frequency in the state monitoring variable c.

[0107] Under idealized assumptions, the estimated mean of the single-point prior distribution of the state monitoring variable with all core frequency prior information known is and the single-point prior distribution log variance estimate It can be expressed as a nonlinear combination of the component mean estimate and the component logarithmic variance estimate corresponding to each core frequency component. Here, two parameter-independent single-output single-layer fully connected neural networks are used to approximate the aggregation operation of the estimated values ​​of each distribution parameter.

[0108]

[0109] Where g is the cluster label value corresponding to variable c. In the distribution aggregation module, the parameters of the neural network layer are also independent under different variable clusters. Aggregation(·,...,·,) represents the aggregation operation, Linear (g),μ (·) and Linear (g),-2logσ (·) represents two parameter-independent linear layers corresponding to the mean estimate and log-variance estimate under the cluster label value g, and GELU(·) indicates the use of the GELU nonlinear activation function.

[0110] (3) Re-parameter sampling module:

[0111] In addition to outputting the estimated values ​​of the distribution parameters of a specific variable at a specific time point, the timestamp probabilistic interpolation model also needs to output a specific interpolation value for constructing the final interpolation result.

[0112] Here, the estimated distribution parameters are used to estimate the state random variable X TS,c A single random sampling is performed to generate the numerical result of timestamp interpolation. In order to ensure the correct back propagation when optimizing the timestamp probabilistic interpolation model, the reparameterization technique is introduced to convert the random sampling process into a deterministic operation. First, the single-point prior distribution logarithmic variance estimate Transform to obtain the standard deviation estimate of the single-point prior distribution Then, an auxiliary noise quantity ε is randomly sampled from the standard normal distribution to calculate the timestamp interpolation result obtained by a single sampling: As shown below:

[0113]

[0114] Input normalized timestamp data at the specified time step T Based on the timestamp prior feature encoding module and the timestamp probabilistic interpolation model, the estimated value of the prior distribution mean can be calculated and the log variance estimated by the prior distribution The estimated standard deviation of the prior distribution obtained by transformation The sizes of both are T×C. Based on each set of distribution parameter estimates, T·C independent reparameter sampling is performed, and then the sampling results are stacked in sequence to obtain the timestamp interpolation results of multivariate time series data. , whose dimensions are T×C.

[0115] The local time series interpolation model of this embodiment can directly select any one of the existing technologies to specify the local time window size T s Incomplete local state monitoring time series data X local As input, the local interpolation result of the state monitoring time series data is The deep learning time series interpolation model with output that can be optimized using the gradient back propagation algorithm, where X local and The size is T s ×C.

[0116] In this embodiment, the self-attention-based time series completion model (SAITS) is selected as the local time series interpolation model. This model performs additional artificial missing operations on the input data during the training process, and considers the time series completion and time series reconstruction tasks at the same time, and has good performance in interpolation accuracy and efficiency. It is worth noting that although the time series features are also positionally encoded in this model, it is only a relative position encoding within the local window range, rather than a feature encoding that considers the global unified position information in the timestamp.

[0117] Both the timestamp probabilistic interpolation model and the local time series interpolation model can be used to interpolate complex equipment status monitoring data. However, the characteristics of the two models are complementary. Specifically, the timestamp probabilistic interpolation model can make full use of the prior periodic time series pattern characteristics obtained by the frequency domain prior mining model analysis. It is suitable for modeling stable, large-scale, and highly periodic time series change laws. It can provide a basic estimate for the value of each status monitoring variable at a specific time point. It has strong globality and is less affected by the missing pattern and the size of the local missing rate. However, due to the limited representation ability of the model and the use of multiple idealized assumptions, when faced with complex equipment incomplete status monitoring data with complex multivariate local dependencies, the interpolation accuracy it can achieve is often difficult to directly meet the actual requirements.

[0118] The local time series interpolation model can fully capture the various complex time series dependencies and cross-variable dependencies of complex equipment status monitoring data within a specific local time window. It is suitable for modeling local, dynamic, and fine-grained time series change laws, and has the advantages of strong model representation ability and high upper limit of interpolation accuracy. However, the interpolation effect of this type of model is greatly affected by the local missing rate and missing pattern of the input data, and there is a problem of insufficient generalization performance. For example, when dealing with samples with too high local missing rate or abnormal missing pattern, it is often difficult to fully and accurately capture local dependencies, resulting in unsatisfactory interpolation effect.

[0119] The role of the timestamp-local interpolation model joint framework is to adaptively fuse the interpolation results of the two interpolation models in a suitable manner, so as to improve the stability of the interpolation performance while ensuring the interpolation accuracy.

[0120] The input of the joint framework is the incomplete local state monitoring time series data X local And the corresponding local timestamp prior encoding feature , where the local timestamp prior encodes the feature By local timestamp sequence It is obtained by transforming the timestamp prior feature encoding module. In addition, by detecting the incomplete local state monitoring time series data X localThe missing data location information can be used to construct the missing value mask M local (Size is T s ×C), where the element corresponding to the missing value is taken as 0, and the remaining positions are taken as 1.

[0121] The process of completing interpolation in the above joint framework is as follows:

[0122] 1) Convert the incomplete local state monitoring time series data X local With missing value mask M local Input into the local time series interpolation model to obtain the local interpolation result of the condition monitoring time series data Encoding timestamp prior features Input into the timestamp probabilistic interpolation model to obtain the local prior distribution mean estimate Estimation of log variance from local prior distribution The estimated standard deviation of the local prior distribution obtained by transformation And the timestamp interpolation results of the condition monitoring time series data

[0123] 2) Introduce the adaptive weighted fusion module and calculate the adaptive fusion threshold α fusion , size T s ×C. Final interpolation result of condition monitoring time series data Interpolation results of timestamps based on adaptive fusion threshold And the local interpolation results The weighted fusion is obtained as shown in the following formula:

[0124]

[0125] Among them, ⊙ represents element-by-element multiplication.

[0126] In the above process, the adaptive weighted fusion module aims to adaptively analyze the relative difference of the output values ​​of the two types of models, the missing state information and the uncertainty information of the timestamp probabilistic interpolation model, and assign appropriate fusion weights to each time point of each variable. Since for a random variable that obeys a normal distribution, the width of its specific confidence interval is positively correlated with the size of the standard deviation, the uncertainty information generated by the timestamp probabilistic interpolation model can be estimated using the local prior distribution standard deviation In addition, the missing state information can be represented by the missing value mask M local Therefore, the local interpolation results of the state monitoring time series data can be , Time stamp interpolation results of status monitoring time series data , local prior distribution standard deviation estimate , missing value mask M localStacked to construct the adaptive weighted fusion module input feature H fusion , whose size is 4×T s ×C, as shown in the following formula:

[0127]

[0128] Among them, [·,·] represents the concatenation operation.

[0129] Input feature H to the adaptive weighted fusion module fusion Perform deformation operation to obtain transposed fusion input features Its dimensions are C×4×T s , and then P = 2 one-dimensional convolutional layers with different convolution kernel sizes are used in parallel to extract and fuse the multi-scale temporal dependencies of the input features within each state monitoring variable. To simplify the model structure, the parameters of each convolution kernel are shared among different variables.

[0130]

[0131] in, represents the output of the pth convolutional layer, whose size is C×D×T s , D = 32 is the fusion hidden layer representation dimension. Conv1d(·) represents a one-dimensional convolution operation, ω (0) ,ω (1) ,...,ω (P-1) Represents the learnable convolution kernel weights of different sizes, whose sizes are; D×4×K (0) ,D×4×K (1) ,...,D×4×K (P-1) In this example, [K (0) ,K (1) ]=[3,5].b (0) ,b (1) ,...,b (P-1) It indicates a learnable bias. Each convolutional layer fills the boundary with 0 to ensure that the length of the input and output features in the time dimension remains unchanged.

[0132] The outputs of the multi-scale convolutional layers are summed, and then the features of each hidden layer are fused using a fusion convolutional layer with a convolution kernel size of 1. The Sigmoid function is activated to limit the range of the output value to [0, 1], and the fusion weighted output feature H is obtained. fusion_out , whose dimensions are C×1×T s The fusion weight output features are rearranged and dimensionally compressed to obtain the final adaptive fusion threshold α fusion (Size is T s ×C).

[0133]

[0134] Among them, ω F represents the learnable weights of the fused convolutional layer, whose size is 1×D×1, b F is the learnable bias of the fused convolutional layer.

[0135] The above introduces the respective processing flows of the frequency domain prior mining model, the timestamp prior feature encoding module and the timestamp-local interpolation model joint framework of the present application. Before the above scheme is actually applied online, corresponding training is required before online application can be performed. The application process of the above scheme provided in this embodiment can be divided into two links, offline and online. The offline link is only used to realize the adaptive analysis of some prior parameters and the offline training of each model, and there is no need to execute it again after the offline training is completed; and each module in the online link participates in the online application of the interpolation method.

[0136] Specifically, the above offline training process is divided into two stages: the timestamp prior information adaptive unsupervised learning stage and the timestamp-local interpolation model joint framework parameter optimization training stage.

[0137] In the phase of adaptive unsupervised learning of timestamp prior information, the preprocessed complex equipment verification time series data feature sequence is transformed into It is regarded as high-quality historical condition monitoring data and input into the frequency domain priori mining model. After random sampling, spectrum analysis, data dimension reduction, unsupervised feature clustering, TOP-K selection and other links, the condition monitoring variable clustering label Y is obtained. L Based on the output of the prior core frequency information F corresponding to each variable c and the prior phase information Φ c , and then determine the parameters of the timestamp prior feature encoding module. This stage is mainly based on the adaptive implementation of the unsupervised machine learning algorithm. Once the obtained parameters are determined, they remain unchanged and do not participate in the gradient back propagation process of the next stage.

[0138] In the parameter optimization training phase of the timestamp-local interpolation model joint framework, the interpolation data training set Z is firstly train The samples in are taken as input, and the high missing rate incomplete local state monitoring time series data X local and based on local timestamp sequence The local timestamp prior encoding feature obtained by transformation Input into the timestamp-local interpolation model joint framework to calculate the final interpolation results of the condition monitoring time series data , local prior distribution mean estimate , local prior distribution standard deviation estimate Then, the joint loss function is calculated by

[0139]

[0140] The joint loss function is composed of the interpolated main loss and prior distribution loss The interpolation main loss in this example uses a form similar to that in the SAITS algorithm. The only difference is that the final interpolation result after fusion is Replaces the original local interpolation result As the final output of the algorithm.

[0141] The prior distribution loss The negative log-likelihood distribution loss and distribution standard deviation regularization loss Based on the distribution weight β, we get:

[0142]

[0143] The negative log-likelihood distribution loss Based on the maximum likelihood estimation principle, its goal is to improve the rationality of the distribution parameter estimation in the timestamp probabilistic interpolation model. Assume that the observation data corresponding to a specific time step t in a specific variable c is At the same time, a set of distribution parameter estimates are given based on the timestamp probabilistic interpolation model and The corresponding likelihood function can be expressed as:

[0144]

[0145] Among them TS t,c Represents the timestamp information corresponding to the data. Then the negative log-likelihood function can be expressed as:

[0146]

[0147] Discarding the last constant term, we get the negative log-likelihood loss for a single data point, as shown below:

[0148]

[0149] The negative log-likelihood loss on the entire input data scale is obtained by calculating the negative log-likelihood loss on the non-missing positions of the condition monitoring time series data and taking the average:

[0150]

[0151] Among them, M local Indicates input status monitoring time series data X local The corresponding missing value mask.

[0152] The standard deviation of the distribution regularizes the loss Then, by further restricting the local prior distribution standard deviation estimate The L1 norm of t is used to reduce the uncertainty of the timestamp probabilistic interpolation model. Assume that the standard deviation estimate of the single-point prior distribution corresponding to a specific time step t in a specific variable c is , then the distribution standard deviation regularization loss under a single data point is:

[0153]

[0154] In order to make the distribution standard deviation regularization loss play a joint optimization effect with the negative log-likelihood distribution loss, the distribution standard deviation regularization loss calculation method on the entire input data scale is similar to the negative log-likelihood distribution loss:

[0155]

[0156] In the above loss function, the hyperparameters α and β are optimized based on the hyperparameter tuning algorithm.

[0157] After determining the loss function, the joint framework can be trained based on gradient back propagation. The learner in the training process uses Adams, and the basic learning rate is optimized based on the hyperparameter tuning algorithm.

[0158] The joint framework adopts a multi-round training mode. After each round of training, the interpolation data validation set Z val The high missing rate of each sample in the incomplete local state monitoring time series data X local and local timestamp sequence After transformation, it is input into the joint framework to obtain the final time series interpolation result At the same time, based on the actual incomplete local state monitoring time series data and high missing rate incomplete local state monitoring time series data X local Constructing the indicator mask , whose size is T s ×C, where where is the observable value and in X local The corresponding position elements of the missing values ​​in are taken as 1, and the remaining position elements are taken as 0. On this basis, the mean square error (MSE) is used to calculate the interpolation error as the validation loss

[0159]

[0160] The validation loss is used to analyze the degree of overfitting of the model. The early stopping strategy with a Patience of 10 is used to determine the termination condition of the training round to optimize the training effect. To prevent the early stopping condition from failing to be met, the maximum number of training rounds is set to 100.

[0161] Next, based on the interpolation data test set Z test , the interpolation model joint framework based on adaptive timestamp coding obtained by training in this embodiment can be tested and evaluated. Interpolation data test set Z test The way each sample is input and output in the model is the same as the interpolation data validation set Z val The evaluation indicators used in the test are both mean absolute error (MAE) and mean relative error (MRE). The calculation process of the indicators is similar to that of the verification loss. Table 3 compares the test results of the interpolation method proposed in this example with the interpolation method using only the local time series interpolation model (SAITS) under four groups of artificial missing types.

[0162] Table 3 Comparison of interpolation effects between this method and the local interpolation model under four missing types

[0163]

[0164] In each experimental condition in Table 3, the model training and testing were repeated 5 times independently using different random number seeds, and the average value of the test results was recorded. In addition, the hyperparameters of both models were automatically determined based on the training set and the validation set using the random search hyperparameter tuning algorithm under the P1 missing type, and the number of tuning trials was set to 100. The hyperparameters of the same model under different missing types remained consistent.

[0165] After completing the above offline training and evaluation process, when performing the online interpolation task, it is necessary to first obtain the new C state monitoring variables at T s Online incomplete condition monitoring data of complex equipment with continuous sampling steps and the corresponding online time-stamped series data , and then monitor the online defect status data Perform standardization operations to obtain pre-processed online defective state monitoring data , for online time-stamped series data After normalization and transformation by the timestamp prior feature encoding module, the online timestamp prior encoding feature is obtained. , and then input both into the trained timestamp-local interpolation model joint framework to output the online interpolation results of the incomplete state monitoring data

[0166] Analyzing the experimental data, it can be seen that the average interpolation error of the proposed method under different missing modes and missing rates is lower than that of the method using only the local time series interpolation model, which shows that the proposed method has the advantages of high interpolation accuracy and strong adaptability. At the same time, the interpolation accuracy of the proposed method in high missing rate scenarios is significantly improved compared with the local interpolation model, which shows that the proposed method is more robust and can better adapt to the scenario where the condition monitoring data of complex equipment is highly incomplete under sudden and severe working conditions. In summary, the proposed method designs adaptive timestamp coding and timestamp probabilistic interpolation models, and organically combines them with the existing advanced local time series interpolation models, achieving a comprehensive improvement in the interpolation performance of gas turbine incomplete condition monitoring data.

[0167] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0168] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for temporal interpolation of complex equipment incomplete state monitoring data enhanced by adaptive timestamp coding, characterized in that: include: Obtain online defect status monitoring data of complex equipment; The online defective state monitoring data includes online defective state monitoring time series data of complex equipment and corresponding online timestamp sequence data; The online incomplete state monitoring time series data is the monitoring values ​​of several state monitoring variables of complex equipment under multiple continuous sampling steps; Standardizing the online defect state monitoring time series data to obtain preprocessed online defect state monitoring time series data; After normalizing the online timestamp sequence data, a timestamp priori feature encoding module is used to output online timestamp priori encoding features; The parameters of the timestamp prior feature encoding module are obtained based on an unsupervised frequency domain prior mining model; The preprocessed online incomplete state monitoring time series data and the online timestamp prior coding features are input into the trained timestamp-local interpolation model joint framework to obtain the online interpolation result of the incomplete state monitoring data; the timestamp-local interpolation model joint framework is a joint framework obtained based on the timestamp probabilistic interpolation model and the local time series interpolation model.

2. The method for time series interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 1 is characterized in that: Before using the time stamp prior feature encoding module to output the online time stamp prior encoding feature, the adaptive time stamp encoding enhanced complex equipment defective state monitoring data time series interpolation method further includes: According to the offline monitoring data in the historical scenario of complex equipment, an interpolation data training set, an interpolation data verification set and an interpolation data test set are constructed; the interpolation data training set is used to train the timestamp-local interpolation model joint framework; the interpolation data verification set and the interpolation data test set are used to verify, optimize and evaluate the performance of the trained timestamp-local interpolation model joint framework; The high-quality offline monitoring data in historical scenarios is input into the frequency domain prior mining model, and the prior core frequency information and prior phase information corresponding to each state monitoring variable are output based on the cluster labels of the state monitoring variables. Determine the parameters of the timestamp prior feature encoding module based on the prior core frequency information and the prior phase information corresponding to each state monitoring variable; The interpolation data training set is used to train the timestamp-local interpolation model joint framework to obtain a trained timestamp-local interpolation model joint framework.

3. The method for time series interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 2 is characterized in that: According to the offline monitoring data of complex equipment in historical scenarios, the interpolation data training set, interpolation data verification set and interpolation data test set are constructed, including: Acquire initial incomplete state monitoring data of complex equipment; the initial incomplete state monitoring data includes offline incomplete state monitoring time series data of complex equipment and corresponding offline timestamp sequence data; the offline incomplete state monitoring time series data is the monitoring values ​​of several state monitoring variables of complex equipment under multiple continuous sampling steps in historical scenarios; After normalizing the offline timestamp sequence data, non-overlapping data division is performed according to the initial incomplete state monitoring data to obtain initial training data, initial verification data and initial test data; Performing standardization processing on the offline defective state monitoring time series data in the initial training data, the initial verification data and the initial test data, respectively, to obtain preprocessed initial training data, preprocessed initial verification data and preprocessed initial test data; The preprocessed initial training data, the preprocessed initial verification data and the preprocessed initial test data are respectively subjected to sliding window operations at the same time with local time windows of the same size to construct an interpolated data training set, an interpolated data verification set and an interpolated data test set; any sample in the interpolated data training set, the interpolated data verification set and the interpolated data test set includes offline incomplete local state monitoring time series data, high missing rate incomplete local state monitoring time series data obtained by performing artificial masking operations on the offline incomplete local state monitoring time series data, and local timestamp sequence data.

4. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 3 is characterized in that: The offline incomplete state monitoring time series data in the preprocessed initial verification data is used as high-quality offline monitoring data in the historical scenario and input into the frequency domain prior mining model. Based on the clustering labels of the state monitoring variables, the prior core frequency information and prior phase information corresponding to each state monitoring variable are output.

5. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 3 is characterized in that: The timestamp-local interpolation model joint framework is trained using the interpolation data training set to obtain a trained timestamp-local interpolation model joint framework, specifically including: Inputting the local timestamp sequence data in the interpolation data training set into the timestamp prior feature encoding module to obtain offline timestamp prior encoding features; Inputting the offline timestamp prior coding features into a timestamp probabilistic interpolation model to obtain a local prior distribution mean estimate, a local prior distribution standard deviation estimate, and a timestamp interpolation result; Inputting the incomplete local state monitoring time series data with high missing rate in the interpolation data training set into the local time series interpolation model to obtain a local interpolation result; Determining a final interpolation result based on the timestamp interpolation result and the locality interpolation result; Constructing a joint loss function based on the incomplete local state monitoring time series data with high missing rate, the local prior distribution mean estimate, the local prior distribution standard deviation estimate and the final interpolation result; According to the joint loss function, the timestamp-local interpolation model joint framework is trained based on gradient back propagation to obtain a trained timestamp-local interpolation model joint framework.

6. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 5 is characterized in that: The joint loss function is shown as follows: in, is the joint loss function, X local For high missing rate incomplete local state monitoring time series data, is the final interpolation result, is the local prior distribution mean estimate, is the local prior distribution standard deviation estimate, is the interpolation main loss function, is the prior distribution loss function.

7. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 5 is characterized in that: The timestamp probabilistic interpolation model includes: a distribution coding module, a distribution aggregation module and a re-parameter sampling module; the distribution coding module includes a plurality of parameter-independent distribution encoders; the distribution encoder is used to process the state monitoring variables corresponding to the clustering labels; the distribution encoder includes a plurality of parameter-independent distribution estimation networks, which correspond to each core frequency component of the state monitoring variables respectively; The offline timestamp prior coding feature is input into the timestamp probabilistic interpolation model to obtain the local prior distribution mean estimate, the local prior distribution standard deviation estimate and the timestamp interpolation result, which specifically includes: Inputting the offline timestamp priori coding feature into the corresponding distribution encoder, and outputting the distribution parameters corresponding to each core frequency component; the distribution parameters include the mean estimate and the logarithmic variance estimate; Utilizing the distribution aggregation module, respectively aggregating the mean estimate and the logarithmic variance estimate corresponding to each core frequency component, and outputting the local prior distribution mean estimate and the local prior distribution logarithmic variance estimate; The local prior distribution logarithmic variance estimate is transformed into a local prior distribution standard deviation estimate, and an auxiliary noise amount is obtained by randomly sampling from a standard normal distribution; Based on the local prior distribution mean estimate, the local prior distribution standard deviation estimate and the auxiliary noise amount, sampling is performed to obtain a timestamp interpolation result.

8. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 5 is characterized in that: Inputting the incomplete local state monitoring time series data with high missing rate in the interpolation data training set into the local time series interpolation model to obtain the local interpolation results, specifically including: By detecting the missing data location information of the time series data with high missing rate and incomplete local state monitoring, a missing value mask is constructed; The high missing rate incomplete local state monitoring time series data and the missing value mask are input into a local time series interpolation model to obtain a local interpolation result.

9. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 5 is characterized in that: The final interpolation result is determined according to the following formula: in, is the final interpolation result, is the local interpolation result, α fusion is the adaptive fusion threshold, is the timestamp interpolation result, and ⊙ represents element-by-element multiplication.

10. The method for temporal interpolation of complex equipment defective state monitoring data enhanced by adaptive timestamp coding according to claim 9 is characterized in that: The adaptive fusion threshold is obtained through the following process: The local interpolation results, timestamp interpolation results, local prior distribution standard deviation estimation and missing value mask are stacked to construct the input features of the adaptive weighted fusion module; The input features of the adaptive weighted fusion module are deformed to obtain the transposed fused input features, and then the multi-scale convolutional layer is used to extract and fuse the multi-scale temporal dependencies of the input features within each state monitoring variable. The multi-scale convolution layer is a one-dimensional convolution layer with multiple convolution kernels of different sizes; The outputs of the multi-scale convolutional layers are summed, and the summed results are sequentially passed through the fusion convolutional layer and the activation function layer to output the fusion weight output features; The fusion weight output features are rearranged and dimensionally compressed to obtain an adaptive fusion threshold.

Images