Industrial Internet of Things equipment operation state time series data missing value filling method

Abstract

The invention relates to a big data processing technology, and discloses an industrial Internet of Things equipment operation state time sequence data missing value filling method comprising the following steps: obtaining industrial Internet of Things equipment operation state time sequence data, carrying out mask processing on a missing value position in the data, and carrying out iteration reconstruction to obtain iteration update data; extracting data features of the iterative update data, and performing multi-task processing on the data features by using a missing value interpolation model to obtain a multi-task processing result; and reversely updating model parameters in the missing value interpolation model by using a multi-task processing result to obtain a target missing value interpolation model, and performing data interpolation on the operation state time sequence data of the industrial Internet of Things equipment by using the target missing value interpolation model to obtain operation state time sequence data of the target industrial Internet of Things equipment. The invention further provides an industrial Internet of Things equipment operation state time series data missing value filling device, electronic equipment and a storage medium. According to the method, the accuracy of the missing value interpolation of the running state time sequence data of the industrial Internet of Things equipment can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to a method for filling missing values in time series data of the operating state of industrial Internet of Things devices. Background Art

[0002] In the industrial Internet of Things (IIoT), due to phenomena such as sensor failures and unstable data transmission in industrial Internet of Things devices, and in addition, reasons such as no data being recorded during equipment maintenance, the situation of missing values in the time series of device operating state data is very common. And a complete data set is the basis for training a reliable machine learning model. Incomplete data will lead to inaccurate subsequent analysis and decision-making. Therefore, filling missing values is particularly important for enhancing the efficiency and reliability of the industrial Internet of Things system.

[0003] Existing technologies usually use interpolation algorithms to handle missing values. However, existing technologies mainly rely on supervised learning methods, that is, by using the remaining complete data to build relevant prediction models to achieve the interpolation of missing values. But in this method, when a data set is very incomplete, the existing technologies will result in relatively low correlation and accuracy of the filled missing values. Summary of the Invention

[0004] The present invention provides a method for filling missing values in time series data of the operating state of industrial Internet of Things devices, which can improve the accuracy of interpolating missing values in time series data of the operating state of industrial Internet of Things devices.

[0005] To achieve the above object, a method for filling missing values in time series data of the operating state of industrial Internet of Things devices provided by the present invention includes:

[0006] Obtain the time series data of the operating state of industrial Internet of Things devices, and perform masking processing on the positions of missing values in the time series data of the operating state of industrial Internet of Things devices to obtain masked time series data of the operating state of industrial Internet of Things devices;

[0007] Use a weighted median adaptive learning strategy to iteratively reconstruct the masked time series data of the operating state of industrial Internet of Things devices to obtain iteratively updated data;

[0008] Extract the data features in the iteratively updated data, and use a pre-constructed missing value interpolation model to perform multi-task processing on the data features to obtain a multi-task processing result;

[0009] Use the multi-task processing result to reversely update the model parameters in the pre-constructed missing value interpolation model. After the update is completed, obtain a target missing value interpolation model, and use the target missing value interpolation model to perform data interpolation on the time series data of the operating state of industrial Internet of Things devices to obtain target time series data of the operating state of industrial Internet of Things devices.

[0010] Optionally, the masking process for the positions of missing values in the time-series data of the operating status of industrial Internet of Things devices includes:

[0011] The masking process for the positions of missing values in the time-series data of the operating status of industrial Internet of Things devices to obtain masked time-series data of the operating status of industrial Internet of Things devices includes:

[0012] Performing data standardization processing on the time-series data of the operating status of industrial Internet of Things devices to obtain standardized time-series data of the operating status of industrial Internet of Things devices;

[0013] Querying the positions of missing values in the standardized time-series data of the operating status of industrial Internet of Things devices and performing position annotation on the positions of missing values to obtain missing value position annotation;

[0014] Constructing a masking matrix at the missing value position annotation to obtain masked time-series data of the operating status of industrial Internet of Things devices.

[0015] Optionally, the

[0016] Using a weighted median adaptive learning strategy to iteratively reconstruct the masked time-series data of the operating status of industrial Internet of Things devices to obtain iteratively updated data, including:

[0017] Generating initial values for the missing values in the masked time-series data of the operating status of industrial Internet of Things devices to obtain initial input data;

[0018] Each iterative update performs:

[0019] Using the initial input data to obtain reconstructed data through forward propagation in the weighted median adaptive learning strategy;

[0020] Calculating the weighted median of the reconstructed data and the initial input data and dynamically updating the initial input data using the weighted median to obtain updated initial input data;

[0021] If the number of iterative updates reaches the preset number of updates, the updated initial input data obtained in this iterative update is used as the iteratively updated data. If the number of iterative updates does not reach the preset number of updates, the updated initial input data obtained in this iterative update is used as the initial input data for the next iterative update.

[0022] Optionally, after the iterative update reaches the preset number of updates to obtain the iteratively updated data, it further includes dynamically updating the network parameters in the weighted median adaptive learning strategy as the number of iterations progresses.

[0023] Optionally, the dynamic update of the network parameters in the weighted median adaptive learning strategy as the number of iterations progresses includes:

[0024] The following update formula is adopted for the dynamic update of network parameters in the weighted median adaptive learning strategy:

[0025]

[0026] where L is the adaptive loss function, N is the number of data samples, k is the number of iterations, and α (k) is the first decay coefficient at the k-th iteration, β (k) is the second decay coefficient at the k-th iteration, M is an N×N diagonal matrix, the diagonal is the mean of the position annotation vectors composed of the position annotations in the masked time-series data of the industrial IoT device operating status, ⊙ represents the Hadamard product, I is the identity matrix, is the reconstructed data, is the initialized input data.

[0027] Optionally, the multi-task processing of data features using the pre-constructed missing value imputation model to obtain the multi-task processing result includes:

[0028] Extracting the data features in the iteratively updated data and decoding the data features using multiple task decoders in the pre-constructed missing value imputation model. After the decoding is completed, the multi-task processing result is obtained, where the multiple task decoders include a missing value imputation task decoder, an anomaly detection task decoder, and a trend prediction task decoder.

[0029] Optionally, the reverse update of the model parameters in the pre-constructed missing value imputation model using the multi-task processing result includes: updating the model parameters using the adaptive weight decay mechanism in the pre-constructed optimizer.

[0030] To solve the above problems, the present invention also provides an industrial IoT device operating status time-series data missing value filling device, and the device includes:

[0031] A data preprocessing module, configured to obtain the industrial IoT device operating status time-series data and perform masking processing on the positions of the missing values in the industrial IoT device operating status time-series data to obtain the masked industrial IoT device operating status time-series data;

[0032] A multi-task processing module, configured to iteratively reconstruct the masked industrial IoT device operating status time-series data using the weighted median adaptive learning strategy to obtain iteratively updated data; extracting the data features in the iteratively updated data and performing multi-task processing on the data features using the pre-constructed missing value imputation model to obtain the multi-task processing result;

[0033] A target industrial Internet of Things device operation status time series data generation module is used to inversely update the model parameters in a pre-built missing value imputation model by using the multi-task processing results. After the update is completed, a target missing value imputation model is obtained, and the target missing value imputation model is used to perform data imputation on the industrial Internet of Things device operation status time series data to obtain the target industrial Internet of Things device operation status time series data.

[0034] To solve the above problems, the present invention also provides an electronic device, which includes:

[0035] At least one processor; and,

[0036] A memory communicatively connected to the at least one processor; wherein,

[0037] The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned method for filling missing values in the industrial Internet of Things device operation status time series data.

[0038] To solve the above problems, the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores at least one computer program, and the at least one computer program is executed by a processor in an electronic device to implement the above-mentioned method for filling missing values in the industrial Internet of Things device operation status time series data.

[0039] By performing mask processing on the positions of missing values in the industrial Internet of Things device operation status time series data, the present invention can accurately locate the missing areas, effectively distinguish valid data from the missing areas, avoid interference from invalid data, and improve the accuracy of missing value imputation. In addition, by using a weighted median adaptive learning strategy to iteratively reconstruct the masked industrial Internet of Things device operation status time series data, the initial estimated values of the data in the missing areas can be dynamically optimized. Moreover, by inversely updating the model parameters in the pre-built missing value imputation model by using the multi-task processing results, global features can be obtained through multi-task learning, improving the generalization ability of the target missing value imputation model and the accuracy of missing value imputation for the industrial Internet of Things device operation status time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flowchart of the method for filling missing values in the industrial Internet of Things device operation status time series data provided by an embodiment of the present invention;

[0041] Figure 2 It is a schematic diagram of the implementation of the imputation algorithm for the method for filling missing values in the industrial Internet of Things device operation status time series data provided by an embodiment of the present invention;

[0042] Figure 3Schematic diagram of multi - task learning for the method of filling missing values in time - series data of the operating state of industrial Internet of Things devices provided by an embodiment of the present invention;

[0043] Figure 4 Adaptive learning flowchart of the method of filling missing values in time - series data of the operating state of industrial Internet of Things devices provided by an embodiment of the present invention;

[0044] Figure 5 Functional module diagram of a device for filling missing values in time - series data of the operating state of industrial Internet of Things devices provided by an embodiment of the present invention;

[0045] Figure 6 Schematic diagram of the structure of an electronic device for implementing the method of filling missing values in time - series data of the operating state of the industrial Internet of Things devices provided by an embodiment of the present invention.

[0046] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0047] It should be understood that the specific embodiments described herein are only used to explain the present invention, not to limit the present invention.

[0048] The embodiments of the present application provide a method for filling missing values in time - series data of the operating state of industrial Internet of Things devices. The execution subject of the method for filling missing values in time - series data of the operating state of industrial Internet of Things devices includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiments of the present application. In other words, the method for filling missing values in time - series data of the operating state of industrial Internet of Things devices can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0049] Referring to Figure 1 As shown, it is a flowchart of the method for filling missing values in time - series data of the operating state of industrial Internet of Things devices provided by an embodiment of the present invention. In this embodiment, the method for filling missing values in time - series data of the operating state of industrial Internet of Things devices includes:

[0050] S1. Obtain the time-series data of the operating status of industrial IoT devices, and perform masking processing on the positions of missing values in the time-series data of the operating status of industrial IoT devices to obtain the masked time-series data of the operating status of industrial IoT devices.

[0051] In the embodiments of the present invention, the time-series data of the operating status of industrial IoT devices refers to the device operating status data collected by industrial IoT sensors in the industrial Internet of Things for a period of time. For example, in the manufacturing field, the time-series data of the operating status of industrial IoT devices includes, but is not limited to, the vibration change data, temperature change data, and current change data of the device collected within a period of time.

[0052] In the embodiments of the present invention, the position of the missing value refers to a certain period or time point in the time-series data of the operating status of industrial IoT devices where no data is recorded.

[0053] As an embodiment of the present invention, performing masking processing on the positions of missing values in the time-series data of the operating status of industrial IoT devices to obtain the masked time-series data of the operating status of industrial IoT devices includes:

[0054] Perform data standardization processing on the time-series data of the operating status of industrial IoT devices to obtain the standardized time-series data of the operating status of industrial IoT devices;

[0055] Query the positions of missing values in the standardized time-series data of the operating status of industrial IoT devices, and perform position marking on the positions of missing values to obtain the missing value position marking;

[0056] Construct a masking matrix at the missing value position marking to obtain the masked time-series data of the operating status of industrial IoT devices.

[0057] Exemplarily, to perform data standardization processing on the time-series data of the operating status of industrial IoT devices, the following implementation steps can be adopted:

[0058] First, standardize the time-series data so that the data mean is 0 and the variance is 1, thereby improving the convergence speed of the model.

[0059]

[0060] Among them, x is the original data value, x′ is the standardized data value, μ is the mean of the data set, and σ is the standard deviation of the data set.

[0061] Then, perform input data filling: for each missing value, find K nearest complete samples. By considering the neighboring data, the dynamic changes of the data can be captured more effectively. Fill the missing value with the mean of the K neighbors so that the model can be trained using the complete data.

[0062]

[0063] Next, create a missing value marker by creating a mask matrix M ij to indicate the positions of missing values in the data.

[0064]

[0065] where x ij is the missing data in the i-th row and j-th column.

[0066] S2. Use the weighted median adaptive learning strategy to iteratively reconstruct the masked time-series data of the industrial IoT device operating status to obtain iteratively updated data.

[0067] In the embodiment of the present invention, the weighted median adaptive learning strategy refers to an optimization method that combines the statistical characteristics of the weighted median and a dynamic adjustment mechanism, aiming to improve the robustness and adaptability of the model through the adaptive weight allocation of data features.

[0068] As an embodiment of the present invention,

[0069] using the weighted median adaptive learning strategy to iteratively reconstruct the masked time-series data of the industrial IoT device operating status to obtain iteratively updated data, including:

[0070] Generate initial values for the missing values in the masked time-series data of the industrial IoT device operating status to obtain initial input data;

[0071] Each iteration update performs:

[0072] Use the initial input data to obtain reconstructed data through forward propagation in the weighted median adaptive learning strategy;

[0073] Calculate the weighted median of the reconstructed data and the initial input data, and dynamically update the initial input data using the weighted median to obtain updated initial input data;

[0074] If the number of iteration updates reaches the preset number of updates, then use the updated initial input data obtained in this iteration update as the iteratively updated data. If the number of iteration updates does not reach the preset number of updates, then use the updated initial input data obtained in this iteration update as the initial input data for the next iteration update.

[0075] Furthermore, in each iteration update, it also performs: Dynamically update the network parameters in the weighted median adaptive learning strategy based on the adaptive loss function, including:

[0076] The adaptive loss function is:

[0077]

[0078] where L is the adaptive loss function, N is the number of data samples, k is the index of the iteration update times, and α(k) is the first decay coefficient at the k-th iteration, β (k) is the second decay coefficient at the k-th iteration, M is an N×N diagonal matrix, the diagonal is the mean of the position annotation vectors composed of the position annotations in the time series data of the running state of the masked industrial IoT devices, ⊙ represents the Hadamard product, and I is the identity matrix. is the reconstructed data, is the initialized input data.

[0079] In the embodiments of the present invention, when α (k) is large and β (k) is small, the model will pay more attention to the reconstruction of non-missing values. As k increases (i.e., as the number of training rounds progresses), if the gap between α (k) and β (k) narrows, the model will gradually balance the attention to non-missing values and missing values.

[0080] In the embodiments of the present invention, the update formula of the decay coefficient adopts the following formula:

[0081] α (k) = 2(1 - decay(k, epoch))

[0082] β (k) = decay(k, epoch)

[0083] where decay(k, epoch) is a decay function that can be adjusted according to the training progress and model performance. It is possible to dynamically adjust the values of α (k) and β (k) to respond to the performance of the model on the validation set. For example, if the model performs poorly in reconstructing non-missing values, the value of α (k) can be temporarily increased. In terms of the selection of the decay function, the following function can be selected.

[0084]

[0085] where performance_feedback(k) is a feedback function based on the performance of the model on the validation set, and it can increase or decrease the rate of decay according to the performance of the model. By introducing a dynamic adjustment mechanism into the adaptive learning strategy, the loss function can more flexibly adapt to different stages and data characteristics in the training process. The coefficients α (k) and β (k) change as the number of iterations k increases, allowing the loss function to adjust its focus during the training process: in the early training stage (when k is small), α (k) is large and β (k) is small, meaning that the loss function pays more attention to the reconstruction of non-missing values; as the training progresses (when k increases), β(k) increases while α (k) decreases, causing the loss function to focus more on the imputation of missing values.

[0086] In the embodiments of the present invention, in the weighted median adaptive learning strategy, all non-missing values from complete data and incomplete data are used to learn model parameters, making full use of data information. In addition, the input of missing values and the training of model parameters are carried out synchronously during the training process. And after each forward propagation, the missing values are recursively replaced by the input weighted median and its reconstruction. Once the model is trained, the missing values are completed in its input at the same time. Furthermore, during the entire training, the change of the initial value generates a model bias, while the parameter update promotes the reconstruction bias. Finally, the parameter update is completed when convergence reaches an equilibrium state.

[0087] Exemplarily, to obtain iteratively updated data by iteratively reconstructing the masked industrial IoT device operation state time series data using the weighted median adaptive learning strategy, the following implementation steps can be adopted:

[0088] Initializing the input:

[0089] init(X) = init(X where m = 1)

[0090]

[0091] where is the initialization input for the first iteration, X is the original data, m is the position annotation vector indicating which values are missing (missing when the value is 1) or non-missing (non-missing when the value is 0), init(X) is a function that generates initial values for missing values, such as initial values of zero, Gaussian noise, or the mean of variables, and the operation ⊙ represents the Hadamard product (element-wise multiplication).

[0092] Forward propagation:

[0093]

[0094] In each iteration k, the weighted median adaptive learning strategy generates reconstructed data by taking the initialized input data

[0095] as the input. Then, the weighted median of the initialized input data and the reconstructed data

[0096]

[0097] Next, update the missing values, keeping the non-missing values unchanged and replacing the missing values with the weighted median through the operations shown below. To update the data:

[0098] Among them, is the input data that keeps the data unchanged when it is a non-missing value in the initialized input data.

[0099] And prepare the input for the next iteration:

[0100]

[0101] Among them, the updated data is used as the initialization input for the next iteration

[0102] S3. Extract the data features in the iteratively updated data, and use the pre-built missing value imputation model to perform multi-task processing on the data features to obtain the multi-task processing result.

[0103] As an embodiment of the present invention, using the pre-built missing value imputation model to perform multi-task processing on the data features to obtain the multi-task processing result includes:

[0104] Extract the data features in the iteratively updated data, and use multiple task decoders in the pre-built missing value imputation model to decode the data features. After the decoding is completed, the multi-task processing result is obtained. Among them, the multiple task decoders include a missing value imputation task decoder, an anomaly detection task decoder, and a trend prediction task decoder.

[0105] Exemplarily, to extract the data features in the iteratively updated data and use the pre-built missing value imputation model to perform multi-task processing on the data features to obtain the multi-task processing result, the following implementation steps can be adopted:

[0106] (1) Shared encoder: Compress the input data into a low-dimensional latent representation h.

[0107] h = σ(W e1 x + b e1 )

[0108] Among them, W e1 is the weight matrix, b e1 is the bias vector, σ is the activation function, and the activation function used in this patent is the rectified linear unit ReLU that can help solve the vanishing gradient problem and accelerate the training process of the neural network.

[0109] ReLU(x) = max(0, x)

[0110] The output of this function is the maximum value between the input value x and 0. Specifically, if the input x is greater than 0, the ReLU function directly outputs x; if the input x is less than or equal to 0, the ReLU function outputs 0.

[0111] Progression of the hidden layer:

[0112] h (i) = σ(W ei h (i-1) + b ei )

[0113] (2) Decoder: Reconstruct the original data from the low-dimensional latent space.

[0114] x reconstructed = σ(W d1 h (L) + b d1 )

[0115] where W d1 is the weight matrix of the decoder, and b d1 is the bias.

[0116] Progression of the hidden layer:

[0117] h (j) = σ(W dj h (j-1) + b dj )

[0118] Extract features from the input data, and then these features are fed into three different decoders to handle the tasks of missing value imputation, anomaly detection, and trend prediction for time series data respectively. For the decoder of Task 1, it is required to reconstruct the missing values and output data with the same shape as the input. For the decoder of Task 2, it outputs an anomaly score for anomaly detection. For the decoder of Task 3, it outputs the predicted values at future time points for trend prediction.

[0119] In the embodiments of the present invention, anomaly detection can help identify anomaly points in industrial Internet of Things data. Anomaly points may be caused by sensor failures or other reasons, which may lead to data missing. Therefore, the results of anomaly detection can be used to guide missing value imputation, especially when deciding whether a certain value needs to be imputed. Conversely, missing value imputation can provide a more complete data view for anomaly detection, helping to more accurately identify anomalies. Similarly, missing value imputation provides a more complete data sequence, which is a better input for the trend prediction model because the model can learn more continuous data patterns. And, anomaly detection can identify data points that deviate from the normal trend, which is an important signal for the trend prediction model because the abnormal data points may be those that the prediction model needs to handle specifically.

[0120] In the embodiment of the present invention, the depth and width of the network can be adjusted according to the actual task, and the Bayesian global optimization strategy is used to obtain the optimal number of layers and the number of neurons in each layer in the pre-built missing value imputation model.

[0121] S4. Use the multi-task processing results to inversely update the model parameters in the pre-built missing value imputation model. After the update is completed, the target missing value imputation model is obtained, and the target industrial Internet of Things device operation status time series data is obtained by using the target missing value imputation model to perform data imputation on the industrial Internet of Things device operation status time series data.

[0122] In the embodiment of the present invention, in order to improve the generalization ability and efficiency of the model, at this time, in addition to the imputation task, that is, predicting the missing values in the time series, an anomaly detection and a trend prediction task are also added to identify anomalies or outliers in the time series.

[0123] In the anomaly detection task, the binary cross-entropy loss is used to measure the accuracy of anomaly detection. The goal of the model is to identify outliers. The model output p i represents the probability that a given sample is an anomaly. The binary cross-entropy loss trains the model by minimizing the difference between the true label and the predicted probability, enabling it to more accurately distinguish normal and abnormal samples. During training, it is desired to minimize the binary cross-entropy loss over the entire dataset

[0124]

[0125] where N is the number of data samples, y i is the true label of the i-th sample. For the anomaly detection task, y i is usually 0 (normal) or 1 (abnormal), p i is the probability that the model predicts the i-th sample as an anomaly, log is the natural logarithm, and θ represents the model parameters. By adjusting θ, the binary cross-entropy loss is reduced The model learns how to more accurately predict the anomaly status of each sample.

[0126] In the trend prediction task, the Quantile loss is used. For a single data point prediction and the true value y and the quantile τ, it is defined as follows:

[0127]

[0128] τ is the quantile (from 0 to 1), specifying the target quantile you want to predict. For example, τ = 0.5 corresponds to the median loss. When is higher than the true value, the penalty coefficient is (1 - τ), and when is lower than the true value, the penalty coefficient is τ.

[0129] So far, combining the losses of the three tasks, a comprehensive loss function is designed. The total loss is the weighted sum of the three task losses, and the weights can be adjusted according to the importance of the tasks.

[0130] L total = λ1L imputation + λ2L anomaly + λ3L τ

[0131] Among them, λ1 is the weight of the missing value imputation task, λ2 is the weight of the anomaly detection task, and λ3 is the weight of the trend prediction task. The comprehensive loss function enables information sharing between related tasks. The knowledge learned by one task can promote the learning of another task and improve the performance of each task.

[0132] As an embodiment of the present invention, the model parameters in the pre-constructed missing value imputation model are updated in reverse using the multi-task processing results, including: updating the model parameters using the adaptive weight decay mechanism in the pre-constructed optimizer.

[0133] Exemplarily, updating the model parameters using the adaptive weight decay mechanism in the pre-constructed optimizer includes the following steps:

[0134] The parameter update formula of the adaptive weight decay mechanism in the pre-constructed optimizer is:

[0135]

[0136] Among them, η t is the learning rate dynamically adjusted at the t-th iteration, and ∈ is a constant for numerical stability. λ i is the adaptive weight decay coefficient of the i-th parameter, and θ t-1,i is the value of the i-th parameter at the (t - 1)-th iteration.

[0137] In the embodiment of the present invention, the adaptive weight decay mechanism selects the ReduceLROnPlateau strategy. For example, when the loss on the validation set has not improved for several epochs, the learning rate is reduced to half of the original.

[0138] In the embodiment of the present invention, the adaptive weight decay mechanism helps the model to converge more stably during training and may improve the final performance. Then, the weight decay coefficient is updated according to the gradient of the parameter. If the gradient of a parameter is large, its weight decay coefficient can be increased to reduce its influence during update.

[0139] Furthermore, the update formula of the adaptive weight decay coefficient in the adaptive weight decay mechanism is as follows:

[0140] λ i = λ i ·(1 + ω·|g t,i |)

[0141] where g t,i is the gradient of the i-th parameter at the t-th iteration, and ω is a hyperparameter used to control the adjustment speed of the weight decay coefficient.

[0142] In the embodiments of the present invention, a target missing value imputation model is used to perform data imputation on the time-series data of the operating state of industrial Internet of Things devices, and the target time-series data of the operating state of industrial Internet of Things devices is obtained through a dynamic sample stacking technique. The following implementation steps can be adopted:

[0143] Definition of dynamic samples: It is formed by stacking t samples with consecutive sampling intervals together. Mathematically, it is expressed as

[0144] X Δt (t) = [X T (t + 1 - Δt) X T (t + 2 - Δt) … X T (t)] T

[0145] where X Δt (t) represents the sample vector at time t, and Δt is the number of stacked samples.

[0146] Formation of dynamic samples: The dynamic sample set with a batch size of N can be expressed as

[0147] X N = {x Δt (t1), x Δt (t2), …, x Δt (t N )}

[0148] where t1, t2, …, t n are indices randomly selected from the total number of samples T.

[0149] Therefore, the entire process of the adaptive learning weighted median filling depth self-supervised learning model for processing missing values in time-series data and improving the accuracy and efficiency of imputation through dynamic sample stacking and an adaptive loss function can be described as follows: First, complete the initialization work of the data, and then for k, perform iterations from 1 to the maximum number of iterations epoch, calculate the adaptive coefficient for adjusting the loss function to balance the emphasis on reconstruction and imputation during training, then perform random shuffling and batch selection, generate a random integer sequence S from Δt to T, and extract N integers from it as batch indices. Then, according to the obtained batch indices, obtain dynamic samples from and

[0150]

[0151] Then, the reconstructed data is calculated by an adaptive learning weighted median filling depth self-supervised learning model Median and the imputed data Then, the loss function is calculated and backpropagation is performed to update the model parameters. When updating the input for the next iteration, for each dynamic sample, update

[0152]

[0153] Then, update the index pointer to output the imputed time-series data of the target industrial Internet of Things device operating status

[0154] Embodiments of the present invention use the dynamic characteristics of time series and improve the data imputation frequency and learning efficiency by dynamically stacking samples.

[0155] By masking the positions of missing values in the time-series data of the operating status of industrial Internet of Things devices, the present invention can accurately locate the missing areas, effectively distinguish valid data from the missing areas, avoid interference from invalid data, and improve the accuracy of missing value imputation. In addition, by using a weighted median adaptive learning strategy to iteratively reconstruct the masked time-series data of the operating status of industrial Internet of Things devices, the initial estimated values of the data in the missing areas can be dynamically optimized. Moreover, by using the results of multi-task processing to reversely update the model parameters in the pre-constructed missing value imputation model, global features can be obtained through multi-task learning, improving the generalization ability of the target missing value imputation model and the accuracy of missing value imputation for the time-series data of the operating status of industrial Internet of Things devices.

[0156] As Figure 2 shown, it is a schematic diagram of the implementation of the imputation algorithm for the method of filling missing values in the time-series data of the operating status of industrial Internet of Things devices provided by an embodiment of the present invention.

[0157] As Figure 3 shown, it is a schematic diagram of multi-task learning for the method of filling missing values in the time-series data of the operating status of industrial Internet of Things devices provided by an embodiment of the present invention.

[0158] As Figure 4 shown, it is a flowchart of adaptive learning for the method of filling missing values in the time-series data of the operating status of industrial Internet of Things devices provided by an embodiment of the present invention.

[0159] As Figure 5 shown, it is a functional module diagram of a device for filling missing values in the time-series data of the operating status of industrial Internet of Things devices provided by an embodiment of the present invention.

[0160] The device 100 for filling missing values in the time-series data of the operating state of industrial Internet of Things devices according to the present invention can be installed in an electronic device. According to the functions achieved, the device 100 for filling missing values in the time-series data of the operating state of industrial Internet of Things devices can include a data preprocessing module 101, a multi-task processing module 102, and a target industrial Internet of Things device operating state time-series data generation module 103.

[0161] The module in the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0162] In this embodiment, the functions of each module / unit are as follows:

[0163] The data preprocessing module 101 is used to obtain the time-series data of the operating state of industrial Internet of Things devices and perform masking processing on the positions of missing values in the time-series data of the operating state of industrial Internet of Things devices to obtain masked time-series data of the operating state of industrial Internet of Things devices.

[0164] In the embodiment of the present invention, the time-series data of the operating state of industrial Internet of Things devices refers to the device operating state data collected by industrial Internet of Things sensors in the industrial Internet of Things for a period of time. For example, in the manufacturing field, the time-series data of the operating state of industrial Internet of Things devices includes, but is not limited to, the vibration change data, temperature change data, and current change data of the device collected within a period of time.

[0165] In the embodiment of the present invention, the position of the missing value refers to a certain period or time point in the time-series data of the operating state of industrial Internet of Things devices where no data is recorded.

[0166] As an embodiment of the present invention, performing masking processing on the positions of missing values in the time-series data of the operating state of industrial Internet of Things devices to obtain masked time-series data of the operating state of industrial Internet of Things devices includes:

[0167] Performing data standardization processing on the time-series data of the operating state of industrial Internet of Things devices to obtain standardized time-series data of the operating state of industrial Internet of Things devices;

[0168] Querying the positions of missing values in the standardized time-series data of the operating state of industrial Internet of Things devices and performing position marking on the positions of missing values to obtain missing value position markings;

[0169] Constructing a masking matrix at the missing value position markings to obtain masked time-series data of the operating state of industrial Internet of Things devices.

[0170] Exemplarily, the following implementation steps can be adopted for performing data standardization processing on the time-series data of the operating state of industrial Internet of Things devices:

[0171] First, standardize the time series data so that the data mean is 0 and the variance is 1, thereby improving the convergence speed of the model.

[0172]

[0173] Among them, x is the original data value, and x ′ is the standardized data value, μ is the mean of the data set, and σ is the standard deviation of the data set.

[0174] Then, perform input data filling: for each missing value, find K nearest complete samples. By considering the neighboring data, the dynamic changes of the data can be captured more effectively. Fill the missing value with the mean of the K neighbors so that the model can be trained using the complete data.

[0175]

[0176] Next, create missing value markers by creating a mask matrix M ij to indicate the positions of the missing values in the data.

[0177]

[0178] Among them, x ij is the missing data in the i-th row and j-th column.

[0179] The multi-task processing module 102 is used to iteratively reconstruct the masked industrial IoT device operation state time series data using a weighted median adaptive learning strategy to obtain iteratively updated data; extract the data features in the iteratively updated data, and perform multi-task processing on the data features using a pre-constructed missing value imputation model to obtain multi-task processing results.

[0180] In the embodiment of the present invention, the weighted median adaptive learning strategy refers to an optimization method that combines the statistical characteristics of the weighted median and a dynamic adjustment mechanism, aiming to improve the robustness and adaptability of the model through the adaptive weight allocation of data features.

[0181] As an embodiment of the present invention, using the weighted median adaptive learning strategy to iteratively reconstruct the masked industrial IoT device operation state time series data to obtain iteratively updated data includes:

[0182] Generate an initial value for the missing values in the masked industrial IoT device operation state time series data to obtain initial input data;

[0183] Use the forward propagation in the weighted median adaptive learning strategy for the initial input data to obtain reconstructed data;

[0184] Calculate the weighted median of the reconstructed data and the initial input data, and use the weighted median to dynamically update the initial input data to obtain updated initial input data;

[0185] Use the updated initial input data as the initial input data for the next data update process, and obtain the iterative update data after the iterative update reaches the preset number of update times.

[0186] Furthermore, when the number of iterations progresses, dynamically update the network parameters in the weighted median adaptive learning strategy, including:

[0187] Dynamically update the network parameters in the weighted median adaptive learning strategy using the following update formula:

[0188]

[0189] where L is the adaptive loss function, N is the number of data samples, k is the number of iterations, α (k) is the first decay coefficient at the k-th iteration, β (k) is the second decay coefficient at the k-th iteration, M is an N×N diagonal matrix, the diagonal is composed of the mean of the position annotation vectors formed by the position annotations in the masked industrial IoT device operation status time series data, ⊙ represents the Hadamard product, and I is the identity matrix, is the reconstructed data, is the initialized input data.

[0190] In the embodiments of the present invention, when α (k) is large and β (k) is small, the model will pay more attention to the reconstruction of non-missing values. As k increases (i.e., as the number of training rounds progresses), if the gap between α (k) and β (k) narrows, the model will gradually balance the attention to non-missing values and missing values.

[0191] In the embodiments of the present invention, the update formula of the decay coefficient adopts the following formula:

[0192] α (k) = 2(1 - decay(k, epoch))

[0193] β (k) = decay(k, epoch)

[0194] where decay(k, epoch) is a decay function that can be adjusted according to the training progress and model performance. The values of α (k) and β (k) can be dynamically adjusted to respond to the performance of the model on the validation set. For example, if the model performs poorly in reconstructing non-missing values, the value of α (k) can be temporarily increased. In the selection of the decay function, the following function can be selected.

[0195]

[0196] Among them, performance_feedback(k) is a feedback function based on the performance of the model on the validation set, which can increase or decrease the attenuation rate according to the performance of the model. By introducing a dynamic adjustment mechanism into the adaptive learning strategy, the loss function can more flexibly adapt to different stages and data characteristics in the training process. Coefficient α (k) and β (k) change with the increase of the iteration number k, allowing the loss function to adjust its focus during the training process: in the early training stage (when k is small), α (k) is large and β (k) is small, meaning that the loss function pays more attention to the reconstruction of non-missing values; as the training progresses (when k increases), β (k) increases and α (k) decreases, resulting in the loss function paying more attention to the imputation of missing values.

[0197] In the embodiment of the present invention, in the weighted median adaptive learning strategy, all non-missing values from complete data and incomplete data will be used to learn model parameters, making full use of data information. In addition, the input of missing values and the training of model parameters are carried out synchronously during the training process. And after each forward propagation, the missing values are recursively replaced by the input weighted median and its reconstruction. Once the model is trained, the missing values are completed in its input at the same time. Furthermore, during the whole training, the change of the initial value will generate model bias, while the parameter update will promote the reconstruction bias. Finally, the parameter update is completed when convergence reaches the equilibrium state.

[0198] Exemplarily, to obtain iterative update data by iteratively reconstructing the time-series data of the running state of masked industrial IoT devices using the weighted median adaptive learning strategy, the following implementation steps can be adopted:

[0199] Initialize the input:

[0200] init(X) = init(X where m = 1)

[0201]

[0202] Among them, is the initialization input for the first iteration, X is the original data, m is the position annotation vector indicating which values are missing (missing when the value is 1) or non-missing (non-missing when the value is 0), init(X) is a function that generates initial values for missing values, such as initial values of zero, Gaussian noise, or the average value of variables, and the operation ⊙ represents the Hadamard product (element-wise multiplication).

[0203] Forward propagation:

[0204]

[0205] In each iteration k, the weighted median adaptive learning strategy generates reconstructed data by taking the initialized input data as the input

[0206] Then, calculate the weighted median of the initialized input data the reconstructed data (1 - α) is used to measure the importance of the input value. The larger α is, the more important the input value is, paying more attention to the trend of adjacent data points and reflecting the dynamic changes of the data, as shown below:

[0207]

[0208] Next, update the missing values, keep the non - missing values unchanged and replace the missing values with the weighted median through the following operations to update the data:

[0209] where is the input data that keeps the data unchanged when it is a non - missing value in the initialized input data.

[0210] And prepare the input for the next iteration:

[0211]

[0212] where the updated data is used as the initialized input for the next iteration

[0213] As an embodiment of the present invention, a pre - constructed missing value imputation model is used to perform multi - task processing on data features to obtain multi - task processing results, including:

[0214] Extract the data features in the iteratively updated data, and use multiple task decoders in the pre - constructed missing value imputation model to decode the data features. After the decoding is completed, multi - task processing results are obtained, where the multiple task decoders include a missing value imputation task decoder, an anomaly detection task decoder, and a trend prediction task decoder.

[0215] Exemplarily, to extract the data features in the iteratively updated data and use the pre - constructed missing value imputation model to perform multi - task processing on the data features to obtain multi - task processing results, the following implementation steps can be adopted:

[0216] (1) Shared encoder: Compress the input data into a low - dimensional latent representation h.

[0217] h = σ(W e1 x + b e1 )

[0218] where W e1 is the weight matrix, b e1 is the bias vector, and σ is the activation function. The activation function used in this patent is the rectified linear unit (ReLU) which can help solve the vanishing gradient problem and accelerate the training process of the neural network.

[0219] ReLU(x) = max(0, x)

[0220] The output of this function is the maximum value between the input value x and 0. Specifically, if the input x is greater than 0, the ReLU function directly outputs x; if the input x is less than or equal to 0, the ReLU function outputs 0.

[0221] Progression of the hidden layer:

[0222] h (i) = σ(W ei h (i-1) + b ei )

[0223] (2) Decoder: Reconstruct the original data from the low-dimensional latent space.

[0224] x reconstructed = σ(W d1 h (L) + b d1 )

[0225] where W d1 is the weight matrix of the decoder, and b d1 is the bias.

[0226] Progression of the hidden layer:

[0227] h (j) = σ(W dj h (j-1) + b dj )

[0228] Extract features from the input data, and then these features are fed into three different decoders to handle the tasks of missing value imputation, anomaly detection, and trend prediction for time series data respectively. For the decoder of Task 1, it is required to reconstruct the missing values and output data with the same shape as the input. For the decoder of Task 2, it outputs the anomaly score for anomaly detection. For the decoder of Task 3, it outputs the predicted values at future time points for trend prediction.

[0229] In the embodiments of the present invention, anomaly detection can help identify anomaly points in industrial Internet of Things (IIoT) data. Anomaly points may be caused by sensor failures or other reasons, which may lead to missing data. Therefore, the results of anomaly detection can be used to guide missing value imputation, especially when deciding whether a value needs to be imputed. Conversely, missing value imputation can provide a more complete data view for anomaly detection, helping to more accurately identify anomalies. Similarly, missing value imputation provides a more complete data sequence, which is a better input for trend prediction models because the models can learn more continuous data patterns. And, anomaly detection can identify data points that deviate from the normal trend, which is an important signal for trend prediction models because the anomalous data points may require special handling by the prediction models.

[0230] In the embodiments of the present invention, the depth and width of the network can be adjusted according to the actual task, and the Bayesian global optimization strategy is used to obtain the optimal number of layers and the number of neurons in each layer in the pre-built missing value imputation model.

[0231] The target industrial IoT device operation status time series data generation module 103 is used to inversely update the model parameters in the pre-built missing value imputation model by using the multi-task processing results. After the update is completed, the target missing value imputation model is obtained, and the industrial IoT device operation status time series data is imputed by using the target missing value imputation model to obtain the target industrial IoT device operation status time series data.

[0232] In the embodiments of the present invention, in order to improve the generalization ability and efficiency of the model, at this time, in addition to the imputation task, that is, predicting the missing values in the time series, anomaly detection and trend prediction tasks are also added to identify anomalies or outliers in the time series.

[0233] In the anomaly detection task, binary cross-entropy loss is used to measure the accuracy of anomaly detection. The goal of the model is to identify outliers. The model outputs p i representing the probability that a given sample is an anomaly. The binary cross-entropy loss trains the model by minimizing the difference between the true label and the predicted probability, enabling it to more accurately distinguish normal and anomalous samples. During training, it is desired to minimize the binary cross-entropy loss over the entire dataset

[0234]

[0235] where N is the number of data samples, and y i is the true label of the i-th sample. For the anomaly detection task, y i is usually 0 (normal) or 1 (anomalous), and p iis the probability that the model predicts the i-th sample as an anomaly. log is the natural logarithm, and θ represents the model parameters. By adjusting θ, the binary cross-entropy loss is reduced. The model learns how to more accurately predict the anomaly status of each sample.

[0236] In the trend prediction task, the Quantile loss is used for the prediction of a single data point. For the true value y and the quantile τ, it is defined as follows:

[0237]

[0238] τ is the quantile (from 0 to 1), specifying the target quantile you want to predict. For example, τ = 0.5 corresponds to the median loss. When is higher than the true value, the penalty coefficient is (1 - τ), and when is lower than the true value, the penalty coefficient is τ.

[0239] So far, combining the losses of the three tasks, a comprehensive loss function is designed. The total loss is the weighted sum of the losses of the three tasks, and the weights can be adjusted according to the importance of the tasks.

[0240] L total = λ1L imputation + λ2L anomaly + λ3L τ

[0241] Among them, λ1 is the weight of the missing value imputation task, λ2 is the weight of the anomaly detection task, and λ3 is the weight of the trend prediction task. The comprehensive loss function enables information sharing between related tasks. The knowledge learned by one task can promote the learning of another task and improve the performance of each task.

[0242] As an embodiment of the present invention, the model parameters in the pre-constructed missing value imputation model are updated in reverse using the multi-task processing results, including: updating the model parameters using the adaptive weight decay mechanism in the pre-constructed optimizer.

[0243] Exemplarily, updating the model parameters using the adaptive weight decay mechanism in the pre-constructed optimizer includes the following steps:

[0244] The parameter update formula of the adaptive weight decay mechanism in the pre-constructed optimizer is:

[0245]

[0246] Among them, η t is the learning rate dynamically adjusted at the t-th iteration, and ∈ is a constant for numerical stability. λ i is the adaptive weight decay coefficient of the i-th parameter, and θt-1,i is the value of the i-th parameter at the (t-1)-th iteration.

[0247] In the embodiments of the present invention, the adaptive weight decay mechanism selects to use the ReduceLROnPlateau strategy. For example, when the loss on the validation set has not improved for several epochs, the learning rate is reduced to half of the original.

[0248] In the embodiments of the present invention, the adaptive weight decay mechanism helps the model to converge more stably during training and may improve the final performance. Then, the weight decay coefficient is updated according to the gradient of the parameter. If the gradient of a parameter is large, its weight decay coefficient can be increased to reduce its influence during the update.

[0249] Furthermore, the update formula of the adaptive weight decay coefficient in the adaptive weight decay mechanism is as follows:

[0250] λ i = λ i ·(1 + ω·|g t,i |)

[0251] where g t,i is the gradient of the i-th parameter at the t-th iteration, and ω is a hyperparameter used to control the adjustment speed of the weight decay coefficient.

[0252] In the embodiments of the present invention, the target missing value imputation model is used to perform data imputation on the time series data of the operation status of industrial IoT devices. The time series data of the operation status of the target industrial IoT devices can be obtained through the dynamic sample stacking technology, and the following implementation steps can be adopted:

[0253] Definition of dynamic samples: It is formed by stacking t samples with consecutive sampling intervals. Mathematically, it is expressed as

[0254] X Δt (t) = [X T (t + 1 - Δt) x T (t + 2 - Δt) … X T (t)] T

[0255] where X Δt (t) represents the sample vector at time t, and Δt is the number of stacked samples.

[0256] Formation of dynamic samples: The dynamic sample set with a batch size of N can be expressed as

[0257] X N = {x Δt (t1), x Δt (t2), …, x Δt (tN )}

[0258] Among them, t1, t2, ..., t n are indices randomly selected from the total number of samples T.

[0259] Therefore, the whole process of the adaptive learning weighted median filling depth self-supervised learning model for processing missing values in time series data and improving the accuracy and efficiency of imputation through dynamic sample stacking and adaptive loss function can be described as follows: First, complete the initialization of the data, then for k, iterate from 1 to the maximum number of iterations epoch, calculate the adaptive coefficient for adjusting the loss function to balance the emphasis on reconstruction and imputation during training, then perform random shuffling and batch selection to generate a random integer sequence S from Δt to T, and extract N integers from it as batch indices. Then, according to the obtained batch indices, obtain dynamic samples from and

[0260]

[0261] Then, calculate the reconstructed data through the adaptive learning weighted median filling depth self-supervised learning model median and the imputed data Then calculate the loss function and perform backpropagation to update the model parameters. When updating the input for the next iteration, for each dynamic sample, update

[0262]

[0263] Then update the index pointer to output the imputed target industrial Internet of Things device operation status time series data

[0264] The embodiments of the present invention utilize the dynamic characteristics of time series to improve the data imputation frequency and learning efficiency through dynamic sample stacking.

[0265] Referring to Figure 6 shown in

[0266] is a schematic structural diagram of an electronic device for implementing the method for filling missing values in the operation status time series data of an industrial Internet of Things device provided by an embodiment of the present invention.

[0267] Among them, in some embodiments, the processor 10 may be composed of an integrated circuit. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple packaged integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. By running or executing programs or modules stored in the memory 11 (such as executing a program for filling missing values in time-series data of the operating state of an industrial Internet of Things device), and calling data stored in the memory 11, it performs various functions of the electronic device and processes data.

[0268] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disks, multimedia cards, card-type memories (such as SD or DX memories, etc.), magnetic memories, magnetic disks, optical disks, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device, such as the mobile hard disk of the electronic device. In some other embodiments, the memory 11 may also be an external storage device of the electronic device, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., equipped on the electronic device. Further, the memory 11 may also include both an internal storage unit and an external storage device of the electronic device. The memory 11 can be used not only to store application software installed on the electronic device and various types of data, such as the code of a program for filling missing values in time-series data of the operating state of an industrial Internet of Things device, but also to temporarily store data that has been output or will be output.

[0269] The communication bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is set to enable connection communication between the memory 11 and at least one processor 10, etc.

[0270] The communication interface 13 is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used to establish a communication connection between this electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device and to display a visual user interface.

[0271] Figure 6 Only the electronic device with components is shown. Those skilled in the art can understand that Figure 6 the shown structure does not constitute a limitation on the electronic device, and it may include fewer or more components than shown, or combine certain components, or have a different component arrangement.

[0272] For example, although not shown, the electronic device may further include a power source (such as a battery) for powering each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so as to implement functions such as charge management, discharge management, and power consumption management through the power management device. The power source may also include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or an inverter, and a power status indicator. The electronic device may also include a variety of sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0273] It should be understood that the above embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0274] The program of the method for filling missing values in the time series data of the operation status of industrial Internet of Things devices stored in the memory 11 in the electronic device is a combination of multiple instructions. When running in the processor 10, it can implement:

[0275] Obtain the time series data of the operation status of industrial Internet of Things devices, and perform masking processing on the positions of the missing values in the time series data of the operation status of industrial Internet of Things devices to obtain masked time series data of the operation status of industrial Internet of Things devices;

[0276] Use a weighted median adaptive learning strategy to iteratively reconstruct the masked time series data of the operation status of industrial Internet of Things devices to obtain iteratively updated data;

[0277] Extract the data features in the iteratively updated data, and use a pre-constructed missing value imputation model to perform multi-task processing on the data features to obtain a multi-task processing result;

[0278] Use the multi-task processing result to inversely update the model parameters in the pre-constructed missing value imputation model. After the update is completed, obtain a target missing value imputation model, and use the target missing value imputation model to perform data imputation on the time-series data of the operating state of the industrial Internet of Things device to obtain the target time-series data of the operating state of the industrial Internet of Things device.

[0279] Specifically, the specific implementation method of the above instructions by the processor 10 can refer to the description of the relevant steps in the corresponding embodiments of the accompanying drawings, which will not be elaborated here.

[0280] Further, if the modules / units integrated in the electronic device 1 are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM, Read-Only Memory).

[0281] The present invention also provides a computer-readable storage medium. The readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, it can implement:

[0282] Obtain the time-series data of the operating state of the industrial Internet of Things device, and perform masking processing on the positions of the missing values in the time-series data of the operating state of the industrial Internet of Things device to obtain masked time-series data of the operating state of the industrial Internet of Things device;

[0283] Use a weighted median adaptive learning strategy to iteratively reconstruct the masked time-series data of the operating state of the industrial Internet of Things device to obtain iteratively updated data;

[0284] Extract the data features in the iteratively updated data, and use a pre-constructed missing value imputation model to perform multi-task processing on the data features to obtain a multi-task processing result;

[0285] Use the multi-task processing result to inversely update the model parameters in the pre-constructed missing value imputation model. After the update is completed, obtain a target missing value imputation model, and use the target missing value imputation model to perform data imputation on the time-series data of the operating state of the industrial Internet of Things device to obtain the target time-series data of the operating state of the industrial Internet of Things device.

[0286] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0287] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0288] In addition, in each embodiment of the present invention, the functional modules can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of a combination of hardware and software functional modules.

[0289] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0290] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0291] The blockchain referred to in the present invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithms. Blockchain, essentially a decentralized database, is a string of data blocks generated by using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of the information (anti-counterfeiting) and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer, etc.

[0292] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use the knowledge to obtain the best results of theory, method, technology, and application systems.

[0293] In addition, it is obvious that the term "including" does not exclude other units or steps, and the singular does not exclude the plural. A plurality of units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. Terms such as first, second, etc. are used to denote names and do not denote any particular order.

[0294] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for filling missing values in time-series data of the operating status of industrial Internet of Things devices, characterized in that, The method includes: Obtaining the time-series data of the operation status of industrial Internet of Things devices, and performing masking processing on the positions of missing values in the time-series data of the operation status of industrial Internet of Things devices to obtain masked time-series data of the operation status of industrial Internet of Things devices; Iteratively reconstructing the masked time-series data of the operation status of industrial Internet of Things devices by using a weighted median adaptive learning strategy to obtain iteratively updated data; Extracting data features from the iteratively updated data, and performing multi-task processing on the data features by using a pre-constructed missing value imputation model to obtain a multi-task processing result; Using the multi-task processing result to reversely update the model parameters in the pre-constructed missing value imputation model. After the update is completed, a target missing value imputation model is obtained, and the target missing value imputation model is used to perform data imputation on the time-series data of the operation status of industrial Internet of Things devices to obtain target time-series data of the operation status of industrial Internet of Things devices.

2. The method for filling missing values in the time-series data of the operating state of industrial Internet of Things devices according to claim 1, wherein The performing masking processing on the positions of missing values in the time-series data of the operation status of industrial Internet of Things devices to obtain masked time-series data of the operation status of industrial Internet of Things devices includes: Performing data standardization processing on the time-series data of the operation status of industrial Internet of Things devices to obtain standardized time-series data of the operation status of industrial Internet of Things devices; Querying the positions of missing values in the standardized time-series data of the operation status of industrial Internet of Things devices, and performing position marking on the positions of missing values to obtain missing value position marking; Constructing a masking matrix at the missing value position marking to obtain masked time-series data of the operation status of industrial Internet of Things devices.

3. The method for filling missing values in the time-series data of the operating state of industrial Internet of Things devices according to claim 1, wherein, The iteratively reconstructing the masked time-series data of the operation status of industrial Internet of Things devices by using a weighted median adaptive learning strategy to obtain iteratively updated data includes: Generating initial values for the missing values in the masked time-series data of the operation status of industrial Internet of Things devices to obtain initial input data; Each iterative update performs: Using the initial input data to obtain reconstructed data through forward propagation in the weighted median adaptive learning strategy; Calculating the weighted median of the reconstructed data and the initial input data, and dynamically updating the initial input data by using the weighted median to obtain updated initial input data; If the number of iterative updates reaches the preset number of updates, the updated initial input data obtained in this iterative update is used as the iteratively updated data. If the number of iterative updates does not reach the preset number of updates, the updated initial input data obtained in this iterative update is used as the initial input data for the next iterative update.

4. The method for filling missing values in the time-series data of the operating state of industrial Internet of Things devices according to claim 3, wherein, In each iterative update, it also performs: dynamically updating the network parameters in the weighted median adaptive learning strategy based on an adaptive loss function.

5. The method for filling missing values in the time-series data of the operation status of industrial Internet of Things devices according to claim 4, wherein The adaptive loss function is: Among them, L is the adaptive loss function, N is the number of data samples, k is the index of the iterative update times, and α ( k ) is the first attenuation coefficient at the k-th iteration, β ( k ) is the second attenuation coefficient at the k-th iteration, M is an N×N diagonal matrix, the diagonal is the mean of the position annotation vectors composed of the position annotations in the masked industrial IoT device operation status time series data, ⊙ represents the Hadamard product, and I is the identity matrix, is the reconstructed data, is the initialized input data.

6. The method for filling missing values in the time series data of the operating state of industrial Internet of Things devices according to claim 1, wherein The performing multi-task processing on the data features by using a pre-constructed missing value imputation model to obtain a multi-task processing result includes: Extracting data features from the iteratively updated data, and performing decoding processing on the data features by using multiple task decoders in the pre-constructed missing value imputation model. After the decoding is completed, a multi-task processing result is obtained, wherein the multiple task decoders include a missing value imputation task decoder, an anomaly detection task decoder, and a trend prediction task decoder.

7. The method for filling missing values in the time-series data of the operating status of industrial Internet of Things devices according to claim 1, wherein Updating the model parameters in the pre-constructed missing value imputation model using the multi-task processing results includes: updating the model parameters using the adaptive weight decay mechanism in the pre-constructed optimizer.

8. An apparatus for filling missing values in time-series data of the operating state of an industrial Internet of Things device, characterized in that, The device implements the method for filling missing values in the time-series data of the operating state of industrial Internet of Things devices as described in any one of claims 1-7. The device includes: A data preprocessing module, configured to obtain the time-series data of the operating state of industrial Internet of Things devices, and perform masking processing on the positions of missing values in the time-series data of the operating state of industrial Internet of Things devices to obtain masked time-series data of the operating state of industrial Internet of Things devices; A multi-task processing module, configured to iteratively reconstruct the masked time-series data of the operating state of industrial Internet of Things devices using a weighted median adaptive learning strategy to obtain iteratively updated data; extract data features from the iteratively updated data, and perform multi-task processing on the data features using the pre-constructed missing value imputation model to obtain multi-task processing results; A target time-series data generation module for the operating state of industrial Internet of Things devices, configured to update the model parameters in the pre-constructed missing value imputation model using the multi-task processing results. After the update is completed, a target missing value imputation model is obtained, and the target missing value imputation model is used to perform data imputation on the time-series data of the operating state of industrial Internet of Things devices to obtain target time-series data of the operating state of industrial Internet of Things devices.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; And a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the method for filling missing values in the time-series data of the operating state of industrial Internet of Things devices as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for filling missing values in the time-series data of the operating state of industrial Internet of Things devices as described in any one of claims 1-7.

Images