Data processing method and device for long-time-sequence industrial real-time data, equipment and medium

Abstract

The invention relates to an artificial intelligence technology, and discloses a data processing method and device for long-time-sequence industrial real-time data, equipment and a storage medium. The method comprises the following steps: constructing a modal decomposition optimization formula, introducing industrial time sequence data and a pre-constructed Lagrange multiplier to obtain an augmented Lagrange formula, and solving to obtain a frequency component set; performing convolution operation on the frequency component set to obtain a convolution feature set, and performing function activation and residual connection operation based on preset times on the convolution feature set to obtain an industrial data feature matrix set; performing normalized feature full connection operation on the data feature sequence set to obtain an industrial data feature set; performing feature recognition operation based on time sequence features on the industrial data feature set to obtain a time sequence feature sequence set; and according to the service scene, calling a downstream prediction network corresponding to the service scene for prediction. According to the method, the data availability of the long-time-sequence industrial real-time data can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a data processing method, device, equipment and computer-readable storage medium for long-time-series industrial real-time data. Background Art

[0002] In the modern industrial environment, with the rapid development of Industry 4.0 and Internet of Things technologies, the number of industrial devices and sensors has increased sharply. These devices and sensors generate a large amount of time-series data in real time for monitoring various physical parameters such as temperature, pressure, vibration, current, etc. These data provide a rich source of information for the optimization and management of industrial processes. However, due to the complexity of the industrial environment, these time-series data are often accompanied by high noise and non-linear complex patterns, posing great challenges to data analysis.

[0003] In this context, how to effectively process and analyze these data to extract valuable information has become the focus of attention in the industrial community. Accurately predicting equipment performance changes, reasonably arranging maintenance plans, and timely detecting faults are the keys to improving production efficiency and reducing operating costs. Current data processing methods are difficult to cope with the problems of high dimensionality, heterogeneity, and dynamics of data, so more advanced data fusion methods need to be introduced. Summary of the Invention

[0004] The present invention provides a data processing method, device, equipment and storage medium for long-time-series industrial real-time data, and its main purpose is to improve the data availability of long-time-series industrial real-time data.

[0005] To achieve the above object, a data processing method for long-time-series industrial real-time data provided by the present invention includes:

[0006] Obtain industrial time-series data and obtain a pre-trained time fusion network, wherein the time fusion network includes a variational mode decomposition network, a temporal convolutional network model, and a gated recurrent unit;

[0007] Use the variational mode decomposition network to construct a modal decomposition optimization formula according to a pre-configured number of modes, and use a pre-constructed alternating direction algorithm to introduce and utilize the industrial time-series data and a pre-constructed Lagrange multiplier in the modal decomposition optimization formula to obtain an augmented Lagrangian formula;

[0008] Solve the augmented Lagrangian formula to obtain a frequency component set, where the frequency component set includes a plurality of modal functions and the central frequencies corresponding to each modal function;

[0009] Obtain the temporal convolutional network model, wherein the temporal convolutional network model includes a causal convolutional layer, a residual layer, and a fully connected layer;

[0010] Perform a convolution operation on the set of frequency components using the causal convolution layer to obtain a set of convolution features, and perform a function activation and residual connection operation on the set of convolution features based on a preset number of times using the residual layer to obtain a set of industrial data feature matrices;

[0011] Perform a normalized feature fully connected operation on the set of data feature sequences using the fully connected layer to obtain a set of industrial data features;

[0012] Perform a feature recognition operation based on temporal features on the set of industrial data features using the gated recurrent unit to obtain a set of temporal feature sequences;

[0013] According to a preset business scenario, call the downstream prediction network corresponding to the business scenario to perform a prediction on the set of temporal feature sequences to obtain a business prediction result.

[0014] Optionally, the obtaining of the industrial time series data includes:

[0015] Use a pre-built sensor cluster to obtain real-time acquisition data within a preset time period;

[0016] Use a database management tool to perform data cleaning operations on the real-time acquisition data based on missing values and outliers to obtain clean time series data;

[0017] Perform a noise reduction operation on the clean time series data based on Gaussian filtering, and perform data format standardization processing on the primary noise-reduced time series data according to a preset data format standard table to obtain industrial time series data.

[0018] Optionally, the using of the variational mode decomposition network to construct a modal decomposition optimization formula according to a pre-configured number of modes includes:

[0019] Obtain the initialization optimization formula of the pre-built variational mode decomposition algorithm and obtain the pre-configured number of modes;

[0020] Substitute the initialization optimization formula according to the number of modes to obtain a modal decomposition optimization formula, where the modal decomposition optimization formula is expressed as:

[0021]

[0022] In the formula, K represents the number of modes, min represents the minimum value function, u k represents the k-th modal function to be solved, ω k represents the central frequency of the k-th modal function, t represents time, δ(t) represents the unit impulse function, and j represents the imaginary unit. Denotes the Hilbert transform of the k-th mode function. Denotes the conversion of the center frequency from the modulus spectrum to the baseband.

[0023] Optionally, solving the augmented Lagrangian formula to obtain a set of frequency components, including:

[0024] Obtaining the augmented Lagrangian formula, where the augmented Lagrangian formula is expressed as:

[0025]

[0026] In the formula, Denotes the augmented Lagrangian formula, λ and λ(t) denote Lagrange multipliers, α denotes the balance parameter, and f(t) denotes the industrial time series data;

[0027] Obtaining the updated Lagrange multiplier constraint, and iteratively solving the augmented Lagrangian formula according to the updated Lagrange multiplier constraint to obtain the updated function of each mode function in the frequency domain and the center frequency corresponding to each updated function, where the updated function of the mode function in the frequency domain is expressed as:

[0028]

[0029] In the formula, Denotes the Fourier transform of the updated k-th mode function. Denotes the Fourier transform of the industrial time series data. Denotes the Fourier transform of the i-th mode function. Denotes the Fourier transform of the Lagrange multiplier;

[0030] Among them, the center frequency corresponding to each updated function is expressed as:

[0031]

[0032] In the formula, Denotes the updated center frequency;

[0033] Among them, the updated Lagrange multiplier constraint is expressed as:

[0034]

[0035] In the formula, the Denotes the updated Lagrange multiplier, and τ denotes the step size.

[0036] Optionally, obtaining the temporal convolutional network model includes:

[0037] Obtain a causal convolutional layer, where the layer function of the causal convolutional layer is expressed as:

[0038]

[0039] In the formula, y t represents the output result of the causal convolutional layer at time t, k′ represents the filter size, w i is the i-th filter weight, and x t-i represents the input of the i-th filter at time t;

[0040] Obtain an activation function and a residual block, and perform a preset number of cyclic connection operations on the activation function and the residual block in sequence to obtain a residual layer. Among them, the convolution process of the activation function is expressed as:

[0041] (x * df)(t) = ∑i = 0 k′-1 f(i) · x t-d，i

[0042] In the formula, x represents the input sequence, d represents the dilation factor, f represents the filter, and x t-d,i represents the input of the i-th filter at time t after being affected by the dilation factor d;

[0043] Among them, the residual block is expressed as:

[0044] o = Activation(x + F(x))

[0045] In the formula, o represents the residual result, Activation represents the activation function, and F(x) represents the function learned by the stacked layer in the residual block;

[0046] Obtain a fully connected layer, where the fully connected layer is expressed as:

[0047]

[0048] In the formula, represents the output result of the l-th layer network at time t in the fully connected layer, is the output result of the (l - 1)-th layer network at time t in the fully connected layer, ReLU represents an activation function used to introduce non-linearity, LN represents a normalization function, represents the weight of the l-th layer network, represents the bias of the l-th layer network;

[0049] Connect the causal convolutional layer, the residual layer, and the fully connected layer in sequence to obtain a primary temporal convolutional network model;

[0050] Obtain a training sample set, and according to the cross-entropy loss algorithm and the gradient descent algorithm, use the training sample set to train the primary temporal convolutional network model to obtain a trained primary temporal convolutional network model.

[0051] Optionally, the gated recurrent unit is expressed as:

[0052]

[0053] In the formula, r t represents the reset gate at time t, z t represents the update gate at time t, represents the candidate hidden state at time t, h t represents the final hidden state at time t, x t represents the input at time t, W r 、W z and W represent weight matrices, h t-1 represents the final hidden state at time t-1, b r 、b z and b represent biases, σ represents the sigmoid function, and ⊙ represents element-wise multiplication.

[0054] Optionally, according to the preset business scenario, call the downstream prediction network corresponding to the business scenario to predict the set of temporal feature sequences to obtain a business prediction result, including:

[0055] According to the preset business scenario, call the downstream prediction network corresponding to the business scenario;

[0056] Obtain the key parameter data set of the business scenario, and according to the preset allocation ratio, allocate the key parameter data set into a training set and a test set;

[0057] Use the training set to fine-tune and train the downstream prediction network to obtain an updated downstream prediction network, and use the test set to test the updated downstream prediction network to obtain test results, where the test results include accuracy, precision, recall rate, and F1 score;

[0058] When the test results meet the preset standard threshold, use the updated downstream prediction network to predict the set of temporal feature sequences to obtain a business prediction result.

[0059] To solve the above problems, the present invention also provides a data processing device for long-term industrial real-time data, and the device includes:

[0060] A modal decomposition noise reduction module, which is used to obtain industrial time series data, obtain a pre-trained time fusion network, use the variational modal decomposition network in the time fusion network to construct a modal decomposition optimization formula according to a pre-configured number of modes, and use a pre-constructed alternating direction algorithm to introduce and utilize the industrial time series data and a pre-constructed Lagrange multiplier in the modal decomposition optimization formula to obtain an augmented Lagrangian formula, and solve the augmented Lagrangian formula to obtain a frequency component set, where the frequency component set includes a plurality of modal functions and the central frequency corresponding to each modal function;

[0061] A causal feature recognition module, which is used to obtain the temporal convolutional network model in the time fusion network, where the temporal convolutional network model includes a causal convolutional layer, a residual layer, and a fully connected layer, and use the causal convolutional layer to perform a convolutional operation on the frequency component set to obtain a convolutional feature set, use the residual layer to perform a function activation and residual connection operation on the convolutional feature set based on a preset number of times to obtain an industrial data feature matrix set, and use the fully connected layer to perform a normalized feature full connection operation on the data feature sequence set to obtain an industrial data feature set;

[0062] A temporal feature extraction module, which is used to perform a feature recognition operation based on temporal features on the industrial data feature set by using the gated recurrent unit in the time fusion network to obtain a temporal feature sequence set;

[0063] A downstream service application module, which is used to call the downstream prediction network corresponding to the service scenario according to a preset service scenario to predict the temporal feature sequence set to obtain a service prediction result.

[0064] To solve the above problems, the present invention also provides an electronic device, which includes:

[0065] At least one processor; and,

[0066] A memory communicatively connected to the at least one processor; wherein,

[0067] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned data processing method for long-time series industrial real-time data.

[0068] To solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored, and the at least one computer program is executed by a processor in an electronic device to implement the above-mentioned data processing method for long-time series industrial real-time data.

[0069] In the embodiment of the present invention, industrial time series data is first obtained, and then according to the variational mode decomposition technology, a modal decomposition optimization formula is constructed. The processing step of denoising the industrial time series data is expressed as a constrained variational problem, and then a Lagrange multiplier is introduced to transform the constrained variational problem into an augmented Lagrangian problem, so as to better solve and obtain the set of denoised frequency components. Among them, the set of frequency components includes a plurality of modal functions and the corresponding central frequencies of each modal function; then, the present invention uses a temporal convolutional network model to extract causal features from the set of frequency components to obtain an industrial data feature set, ensuring that the prediction of the time step only depends on the time and earlier inputs, preventing future information leakage; the present invention then performs a feature recognition operation based on temporal features on the industrial data feature set through a pre-constructed gated recurrent unit to obtain a set of temporal feature sequences. Thus, the set of temporal feature sequences meets the data processing requirements of denoising, extracting long-term dependence relationships, and extracting temporal features, and thus downstream services can be called for calculation. Therefore, a data processing method, device, equipment, and storage medium for long-time series industrial real-time data provided by the embodiment of the present invention can improve the data availability of long-time series industrial real-time data. Description of the Drawings

[0070] Figure 1 It is a schematic flowchart of a data processing method for long-time series industrial real-time data provided by an embodiment of the present invention;

[0071] Figure 2 It is a detailed flowchart of a step in the data processing method for long-time series industrial real-time data provided by an embodiment of the present invention;

[0072] Figure 3 It is a detailed structural diagram of a time fusion network in the data processing method for long-time series industrial real-time data provided by an embodiment of the present invention;

[0073] Figure 4 It is the experimental data of the time fusion network in the data processing method for long-time series industrial real-time data provided by an embodiment of the present invention;

[0074] Figure 5 It is a detailed flowchart of a step in the data processing method for long-time series industrial real-time data provided by an embodiment of the present invention;

[0075] Figure 6 It is a functional module diagram of a data processing device for long-time series industrial real-time data provided by an embodiment of the present invention;

[0076] Figure 7 It is a structural diagram of an electronic device for implementing the data processing method for long-time series industrial real-time data provided by an embodiment of the present invention.

[0077] The implementation, functional features, and advantages of the present invention will be further described in conjunction with embodiments with reference to the accompanying drawings. Detailed implementation manners

[0078] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0079] The embodiments of the present application provide a data processing method for long-time industrial real-time data. In the embodiments of the present application, the execution subject of the data processing method for long-time industrial real-time data 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 in the embodiments of the present application. In other words, the data processing method for long-time industrial real-time data 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.

[0080] Refer to Figure 1 As shown, it is a flowchart of the data processing method for long-time industrial real-time data provided by an embodiment of the present invention. In this embodiment, the data processing method for long-time industrial real-time data includes:

[0081] S1. Obtain industrial time series data and obtain a pre-trained time fusion network, where the time fusion network includes a variational mode decomposition network, a temporal convolutional network model, and a gated recurrent unit.

[0082] In the embodiments of the present invention, the industrial time series data is multi-source data collected by sensors at different regular time periods and has the characteristics of high noise and complex data structure.

[0083] Specifically, the present invention is described by taking an industrial equipment monitoring system as an example. In the embodiments of the present invention, first, real-time collected data is obtained, and the real-time collected data includes multi-dimensional time series such as temperature, pressure, and vibration.

[0084] In detail, refer to Figure 2 As shown, in the embodiments of the present invention, the obtaining of the industrial time series data includes:

[0085] S11. Use a pre-constructed sensor cluster to obtain real-time collected data within a preset time period;

[0086] S12. Use a database management tool to perform data cleaning operations on the real-time collected data based on missing values and outliers to obtain clean time-series data;

[0087] S13. Perform noise reduction operations on the clean time-series data based on Gaussian filtering to obtain primary noise-reduced time-series data, and perform data format standardization processing on the primary noise-reduced time-series data according to a preset standard data format table to obtain industrial time-series data.

[0088] Among them, the sensor cluster includes a temperature sensor, a vibration sensor, a humidity sensor, and so on.

[0089] Further, the standard data format table is for enterprises to input, calculate, and store data according to a fixed format to improve data availability. For example, the temperature is uniformly in degrees Celsius, the time format is uniformly in the "month / day / hour / minute" format, and the text number or digital data format is the same, etc.

[0090] In the embodiment of the present invention, due to the influence of factors such as high temperature, vibration, unstable voltage, and poor contact on the working environment of the equipment motor, the real-time collected data has large data missing and noise. The present invention first uses a database management tool to process abnormal data and missing data, then uses the Gaussian filtering algorithm to preliminarily process the noise to obtain primary noise-reduced time-series data, and then standardizes the data format according to the standard data format table to obtain available industrial time-series data.

[0091] Further, referring to Figure 3 As shown, in the embodiment of the present invention, the Time Fusion Network (TFN) includes three modules: a Variational Mode Decomposition Network (VMD), a Temporal Convolutional Network (TCN) model, and a Gated Recurrent Unit (GRU).

[0092] When processing industrial real-time data, the Time Fusion Network (TFN) has the characteristics of fast response, low latency, high computational efficiency, and deployment flexibility compared with the publicly disclosed cases. Because the Temporal Convolutional Network (TCN) module in the Time Fusion Network (TFN) has a simple structure, a small number of parameters, and fast training and inference speeds, the present invention is easy to deploy in industrial fields, especially suitable for edge devices with limited computing resources. Compared with the Temporal Convolutional Network (TCN), although the Transformer can capture the global characteristics of data, the computational complexity of the Transformer is high, the requirements for hardware resources are high, and it brings certain difficulties to deployment.

[0093] Specifically, the time fusion network is used to: perform preprocessing using variational mode decomposition (VMD) to separate different frequency components of the signal. Secondly, a temporal convolutional network (TCN) model is constructed to capture long-term dependencies in time series data. Finally, a gated recurrent unit (GRU) is used to process the features and input them into the final linear layer.

[0094] In the embodiment of the present invention, referring to Figure 4 As shown, through experiments on real datasets of industrial component degradation prediction and industrial motor predictive maintenance, Table 1 is obtained, which proves the effectiveness of the TFN and demonstrates its potential to enhance prediction capabilities in industrial applications. Among them, Table 1:

[0095]

[0096] S2. Using the variational mode decomposition network, according to the preconfigured number of modes, construct a modal decomposition optimization formula, and use the pre-constructed alternating direction algorithm to introduce the industrial time series data and the pre-constructed Lagrange multiplier into the modal decomposition optimization formula to obtain an augmented Lagrangian formula.

[0097] Among them, the number of modes refers to the number of different modes into which the user expects to decompose the data. In the embodiment of the present invention, the number of modes K is configured to be 3.

[0098] Furthermore, the modal decomposition optimization formula is the optimization formula in the variational mode decomposition algorithm, which can transform the data denoising problem into a constrained variational problem. Among them, the variational mode decomposition (VMD) is an advanced technology for signal processing, aiming to extract meaningful modes (or frequency components) from complex signals. It is a non-linear and adaptive signal decomposition method that decomposes the signal into a set of modal functions with limited narrow bandwidths through variational optimization methods.

[0099] Specifically, in the embodiment of the present invention, constructing the modal decomposition optimization formula according to the preconfigured number of modes includes:

[0100] Obtain the initialization optimization formula of the pre-constructed variational mode decomposition algorithm and obtain the preconfigured number of modes;

[0101] Substitute into the initialization optimization formula according to the number of modes to obtain a modal decomposition optimization formula, where the modal decomposition optimization formula is expressed as:

[0102]

[0103] In the formula, K represents the number of modes, min represents the minimum value function, u k represents the k-th modal function to be solved, ω kdenotes the central frequency of the k-th mode function, t denotes time, δ(t) denotes the unit impulse function, and j denotes the imaginary unit. denotes the Hilbert transform of the k-th mode function. denotes converting the central frequency from the modulus spectrum to the baseband.

[0104] Wherein, the modulus spectrum refers to the representation of the modulated signal in the frequency domain, which describes the amplitude and phase information of different frequency components in the signal, and the baseband refers to the unmodulated signal, usually the original form of the signal.

[0105] It should be known that traditional variational mode decomposition constructs an optimization problem to seek a set of mode functions such that the sum of these functions is closest to the original signal and their bandwidths are minimized. Then, an iterative algorithm is used to solve the optimization problem, and the mode functions and corresponding central frequencies are updated step by step until convergence.

[0106] Furthermore, in the embodiments of the present invention, for the convenience of calculation, by introducing Lagrange multipliers, the constrained variational problem is transformed into an augmented Lagrangian problem, and the augmented Lagrangian algorithm is obtained.

[0107] Wherein, the augmented Lagrangian formula is expressed as:

[0108]

[0109] In the formula, denotes the augmented Lagrangian formula, both λ and λ(t) denote Lagrange multipliers, α denotes the balance parameter, and f(t) denotes the industrial time series data;

[0110] S3. Solve the augmented Lagrangian formula to obtain a frequency component set, wherein the frequency component set includes multiple mode functions and the central frequencies corresponding to each mode function.

[0111] Specifically, in the embodiments of the present invention, the solving of the augmented Lagrangian formula to obtain a frequency component set includes:

[0112] Obtain the augmented Lagrangian formula;

[0113] Obtain the updated Lagrange multiplier constraint, and based on the updated Lagrange multiplier constraint, perform iterative solution on the augmented Lagrangian formula to obtain the updated functions of each mode function in the frequency domain and the central frequencies corresponding to each updated function, wherein the updated function of the mode function in the frequency domain is expressed as:

[0114]

[0115] In the formula, Denote the Fourier transform of the updated k-th mode function, Denote the Fourier transform of the industrial time series data, Denote the Fourier transform of the i-th mode function, Denote the Fourier transform of the Lagrange multiplier;

[0116] Among them, the center frequencies corresponding to the respective update functions are expressed as:

[0117]

[0118] In the formula, Denote the updated center frequency;

[0119] Among them, the updated Lagrange multiplier constraint is expressed as:

[0120]

[0121] In the formula, the Denote the updated Lagrange multiplier, and τ denotes the step size.

[0122] Specifically, in the embodiments of the present invention, through methods such as the Lagrange multiplier method, regularization constraints on the modes are introduced to ensure the stability and physical meaning of the solution, and this optimization problem is solved by an iterative algorithm (such as the alternating direction algorithm) to obtain each mode and its corresponding frequency.

[0123] In the embodiments of the present invention, according to the above calculation process, a frequency component set is obtained, where the frequency component set includes a plurality of mode functions and the center frequencies corresponding to each mode function.

[0124] S4. Obtain the temporal convolutional network model, where the temporal convolutional network model includes a causal convolutional layer, a residual layer, and a fully connected layer.

[0125] Among them, the temporal convolutional network model is a neural network model including a causal convolutional layer, a residual layer, and a fully connected layer, and is used to further extract causal features from the denoised frequency component set.

[0126] Specifically, in the embodiments of the present invention, the obtaining of the temporal convolutional network model includes:

[0127] Obtain a causal convolutional layer, where the layer function of the causal convolutional layer is expressed as:

[0128]

[0129] In the formula, y t Denote the output result of the causal convolutional layer at time t, k′ denotes the filter size, w iis the i-th filter weight, x t-i represents the input of the i-th filter at time t;

[0130] Obtain an activation function and a residual block, and perform a preset number of cyclic connection operations on the activation function and the residual block in sequence to obtain a residual layer. Among them, the convolution process of the activation function is expressed as:

[0131] (x * df)(t) = ∑i = 0 k′-1 f(i) · x t-d,i

[0132] In the formula, x represents the input sequence, d represents the dilation factor, f represents the filter, and x t-d,i represents the input of the i-th filter at time t after being affected by the dilation factor d;

[0133] Among them, the residual block is expressed as:

[0134] o = Activation(x + F(x))

[0135] In the formula, o represents the residual result, Activation represents the activation function, and F(x) represents the function learned by the stacked layers in the residual block;

[0136] Obtain a fully connected layer. Among them, the fully connected layer is expressed as:

[0137]

[0138] In the formula, represents the output result of the l-th layer network at time t in the fully connected layer, the output result of the (l - 1)-th layer network at time t in the fully connected layer, ReLU represents an activation function for introducing non-linearity, and LN represents a normalization function, represents the weight of the l-th layer network, represents the bias of the l-th layer network;

[0139] Connect the causal convolution layer, the residual layer, and the fully connected layer in sequence to obtain a primary temporal convolutional network model;

[0140] Obtain a training sample set, and train the primary temporal convolutional network model using the training sample set according to the cross-entropy loss algorithm and the gradient descent algorithm to obtain a trained primary temporal convolutional network model.

[0141] Specifically, in the embodiments of the present invention, causal convolution ensures that the prediction of the time step only depends on the time and earlier inputs, preventing future information leakage. The causal convolution layer is constructed to extract features and can better capture long-term dependencies.

[0142] Furthermore, in the embodiments of the present invention, a residual layer is constructed to enhance the expressive power of the model and help avoid the problem of gradient disappearance.

[0143] In addition, layer normalization is used in each residual block in the embodiments of the present invention to stabilize training:

[0144]

[0145] where μ and σ are the mean and standard deviation respectively, and γ and β are learnable parameters.

[0146] In the embodiments of the present invention, using layer normalization to stabilize the training process can help the model converge faster and perform better by normalizing the input.

[0147] Furthermore, the present invention is based on the above to clarify the forward propagation process of the network of the solution.

[0148] Specifically, in the embodiments of the present invention, the final output layer of the fully connected layer is also equipped with the last hidden state:

[0149]

[0150] Based on the hidden state of the fully connected layer and its output layer, the present invention can obtain the operation process of the temporal convolutional network model.

[0151] Specifically, in the embodiments of the present invention, it is also necessary to connect the causal convolutional layer, the residual layer, and the fully connected layer in sequence to obtain a primary temporal convolutional network model, and then use the cross-entropy loss algorithm and the gradient descent algorithm to train the primary temporal convolutional network model with the training sample set to obtain a trained primary temporal convolutional network model.

[0152] Based on the construction and training of the above model, a temporal convolutional network model that can be put into use is obtained

[0153] S5. Use the causal convolutional layer to perform a convolution operation on the frequency component set to obtain a convolution feature set, and use the residual layer to perform a function activation and residual connection operation on the convolution feature set based on a preset number of times to obtain an industrial data feature matrix set.

[0154] In the embodiments of the present invention, based on the causal convolutional layer and the residual layer, the frequency component set can be further subjected to a feature extraction operation based on temporal features to obtain an industrial data feature matrix set.

[0155] Specifically, the convolution process of the present invention is performed by traversing convolution calculations, and the preset number of residual connections can be controlled according to the data dimension, so that the data dimension of the finally obtained industrial data feature matrix set meets the calculation dimension of subsequent services. In the embodiment of the present invention, the residual connection helps the model maintain the original information flow by directly adding the input to the output. This connection method not only alleviates the problem of gradient disappearance, but also makes it easier for the model to learn effective features.

[0156] S6. Use the fully connected layer to perform normalized feature full connection operations on the data feature sequence set to obtain an industrial data feature set.

[0157] In the embodiment of the present invention, the industrial data feature set is obtained according to the forward propagation calculation of the fully connected layer. Among them, the use of an activation function (such as ReLU) in the fully connected layer of the present invention can introduce non-linear features, enabling the model to learn more complex patterns.

[0158] S7. Use the gated recurrent unit to perform feature recognition operations based on temporal features on the industrial data feature set to obtain a temporal feature sequence set.

[0159] Specifically, in the embodiment of the present invention, the gated recurrent unit is a simplified version of the long short-term memory network (LSTM), with fewer parameters and higher computational efficiency, suitable for processing sequence data, especially in temporal prediction tasks.

[0160] Specifically, in the embodiment of the present invention, the gated recurrent unit is expressed as:

[0161]

[0162] In the formula, r t represents the reset gate at time t, z t represents the update gate at time t, represents the candidate hidden state at time t, h t represents the final hidden state at time t, x t represents the input at time t, W r 、W z and W represent weight matrices, h t-1 represents the final hidden state at time t-1, b r 、b z and b represent biases, σ represents the sigmoid function, and ⊙ represents element-wise multiplication.

[0163] Among them, the reset gate: used to control how to combine the new input information x t with the past final hidden state h t-1 Combined. Update gate: used to control the final hidden state h t-1Degree of update. Candidate hidden state: Calculate the candidate hidden state at the current moment Final hidden state Combine the past final hidden state h t-1 and the candidate hidden state Generate the final hidden state h at the current moment t .

[0164] In the embodiments of the present invention, the gated recurrent unit is first used to process the input features to generate the final hidden state h t , and then the final prediction result is output by calling the linear layer of the subsequent service. This structure is very effective in processing time series data, and can capture the time series features and make accurate predictions.

[0165] S8. According to the preset business scenario, call the downstream prediction network corresponding to the business scenario to predict the time series feature sequence set, and obtain the business prediction result.

[0166] In the embodiments of the present invention, it is expected to process long-time series industrial real-time data. After obtaining the time series feature sequence set, any downstream prediction network that meets the scenario can be called to use the time series feature sequence set.

[0167] In the embodiments of the present invention, the downstream prediction network can be experiments on industrial component degradation prediction and industrial motor predictive maintenance.

[0168] After calling the downstream prediction network, some real data sets need to be obtained for fine-tuning training.

[0169] Specifically, referring to Figure 5 shown, in the embodiments of the present invention, the step of calling the downstream prediction network corresponding to the business scenario according to the preset business scenario to predict the time series feature sequence set and obtain the business prediction result includes:

[0170] S81. According to the preset business scenario, call the downstream prediction network corresponding to the business scenario;

[0171] S82. Obtain the key parameter data set of the business scenario, and allocate the key parameter data set into a training set and a test set according to the preset allocation ratio;

[0172] S83. Use the training set to fine-tune the downstream prediction network to obtain an updated downstream prediction network, and use the test set to test the updated downstream prediction network to obtain a test result, where the test result includes accuracy, precision, recall rate and F1 score;

[0173] S84. When the test result meets the preset standard threshold, use the updated downstream prediction network to predict the set of time series feature sequences to obtain a service prediction result.

[0174] Specifically, in the embodiment of the present invention, the Adam optimizer is used for training, and the learning rate is 0.001. If there is no improvement for 5 consecutive epochs, the learning rate will be reduced by 0.1. The L2 regularization for improving generalization and data augmentation techniques of random time warping and size scaling are adopted. Then, the key parameters of the motor are collected, including temperature, humidity, vibration, and current. Then the data set is divided into a training set (70%) and a test set (30%). Finally, the following metrics are used to evaluate the performance of the model on the test set: accuracy, precision, recall, and F1 score. Among them, accuracy refers to the proportion of the number of samples correctly predicted by the model in the total number of samples, reflecting the overall performance of the model; precision refers to the proportion of actual positive samples among all samples predicted as positive by the model, used to focus on the prediction quality of the model for the positive class; recall refers to the proportion of samples correctly predicted as positive by the model among all actual positive samples, used to focus on the recognition ability of the model for the positive class; the F1 score is the harmonic mean of precision and recall, taking both into account, used to synthesize precision and recall, and is suitable for use on unbalanced data sets.

[0175] In the embodiment of the present invention, when all test results meet the preset standard threshold, it can be put into use.

[0176] In the present invention, a comparison is made using the same downstream prediction network, and the accuracy of the model using the set of time series feature sequences is significantly better than the calculation result using the original industrial time series data.

[0177] In the embodiment of the present invention, industrial time series data is first obtained, and then according to the variational mode decomposition technology, a modal decomposition optimization formula is constructed. The processing steps of denoising the industrial time series data are expressed as a constrained variational problem, and then a Lagrange multiplier is introduced to transform the constrained variational problem into an augmented Lagrangian problem, so as to better solve and obtain the set of frequency components after denoising. Among them, the set of frequency components includes multiple modal functions and the central frequency corresponding to each modal function. Then, the present invention uses a temporal convolutional network model to extract causal features from the set of frequency components to obtain an industrial data feature set, ensuring that the prediction of the time step only depends on the time and earlier inputs, and preventing future information leakage. The present invention then performs a feature recognition operation based on temporal features on the industrial data feature set through a pre-constructed gated recurrent unit to obtain a set of temporal feature sequences. Thus, the set of temporal feature sequences meets the data processing requirements of denoising, long-term dependence extraction, and temporal feature extraction, and thus downstream services can be called for calculation. Therefore, a data processing method for long-time industrial real-time data provided by the embodiment of the present invention can improve the data availability of long-time industrial real-time data.

[0178] As Figure 6 shown, it is a functional module diagram of a data processing device for long-time industrial real-time data provided by an embodiment of the present invention.

[0179] The data processing device 100 for long-time industrial real-time data according to the present invention can be installed in an electronic device. According to the implemented functions, the data processing device 100 for long-time industrial real-time data can include a modal decomposition denoising module 101, a causal feature recognition module 102, a temporal feature extraction module 103, and a downstream service application module 104. The modules in the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, and are stored in the memory of the electronic device.

[0180] In this embodiment, the functions of each module / unit are as follows:

[0181] The modal decomposition denoising module 101 is used to obtain industrial time series data, obtain a pre-trained time fusion network, use the variational mode decomposition network in the time fusion network to construct a modal decomposition optimization formula according to the pre-configured number of modes, and use the pre-constructed alternating direction algorithm to introduce and utilize the industrial time series data and the pre-constructed Lagrange multiplier in the modal decomposition optimization formula to obtain an augmented Lagrangian formula, and solve the augmented Lagrangian formula to obtain a set of frequency components, where the set of frequency components includes multiple modal functions and the central frequency corresponding to each modal function;

[0182] The causal feature recognition module 102 is configured to obtain the temporal convolutional network model in the temporal fusion network, where the temporal convolutional network model includes a causal convolutional layer, a residual layer, and a fully connected layer, and to perform a convolutional operation on the frequency component set by using the causal convolutional layer to obtain a convolutional feature set, perform a function activation and residual connection operation on the convolutional feature set based on a preset number of times by using the residual layer to obtain an industrial data feature matrix set, and perform a normalized feature fully connected operation on the data feature sequence set by using the fully connected layer to obtain an industrial data feature set;

[0183] The temporal feature extraction module 103 is configured to perform a feature recognition operation based on temporal features on the industrial data feature set by using the gated recurrent unit in the temporal fusion network to obtain a temporal feature sequence set;

[0184] The downstream service application module 104 is configured to call the downstream prediction network corresponding to the service scenario according to a preset service scenario, and perform a prediction on the temporal feature sequence set to obtain a service prediction result.

[0185] Specifically, each module in the long temporal industrial real-time data processing device 100 in the embodiment of the present application adopts the same technical means as those in the Figures 1 to 5 long temporal industrial real-time data processing method described above, and can produce the same technical effects, which will not be elaborated here.

[0186] As Figure 7 shown, it is a schematic structural diagram of an electronic device 1 for implementing the long temporal industrial real-time data processing method provided by an embodiment of the present invention.

[0187] The electronic device 1 may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13, and may further include a computer program stored in the memory 11 and executable on the processor 10, such as a long temporal industrial real-time data processing program.

[0188] 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 1, connecting various components of the entire electronic device through various interfaces and circuits, and by running or executing programs or modules stored in the memory 11 (such as data processing programs for long-time industrial real-time data, etc.), and calling data stored in the memory 11, to perform various functions of the electronic device and process data.

[0189] 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 not only be used to store application software installed on the electronic device and various types of data, such as the code of a data processing program for long-time industrial real-time data, etc., but can also be used to temporarily store data that has been output or will be output.

[0190] 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 realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0191] The communication interface 13 is used for communication between the above-mentioned electronic device 1 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.), and is generally 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, and is used to display the information processed in the electronic device and to display a visual user interface.

[0192] Figure 7 Only the electronic device with components is shown. Those skilled in the art can understand that Figure 7 the shown structure does not constitute a limitation on the electronic device 1, and it may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0193] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for supplying power to 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 1 may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0194] 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.

[0195] The data processing program for long-time industrial real-time data stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can implement:

[0196] Obtain industrial time series data and obtain a pre-trained time fusion network;

[0197] Using the variational mode decomposition network in the time fusion network, construct a modal decomposition optimization formula according to a pre-configured number of modes, and use a pre-constructed alternating direction algorithm to introduce the industrial time series data and a pre-constructed Lagrange multiplier into the modal decomposition optimization formula to obtain an augmented Lagrangian formula;

[0198] Solve the augmented Lagrangian formula to obtain a frequency component set, where the frequency component set includes multiple modal functions and the corresponding central frequencies of each modal function;

[0199] Obtain the temporal convolutional network model in the time fusion network, where the temporal convolutional network model includes a causal convolutional layer, a residual layer, and a fully connected layer;

[0200] Perform a convolution operation on the frequency component set using the causal convolutional layer to obtain a convolution feature set, and perform function activation and residual connection operations on the convolution feature set based on a preset number of times using the residual layer to obtain an industrial data feature matrix set;

[0201] Perform a normalized feature fully connected operation on the data feature sequence set using the fully connected layer to obtain an industrial data feature set;

[0202] Perform a feature recognition operation based on temporal features on the industrial data feature set using the gated recurrent unit in the time fusion network to obtain a temporal feature sequence set;

[0203] According to a preset business scenario, call the downstream prediction network corresponding to the business scenario to predict the temporal feature sequence set to obtain a business prediction result.

[0204] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiments of the accompanying drawings, which will not be elaborated here.

[0205] Furthermore, 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).

[0206] The present invention also provides a computer-readable storage medium, where 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:

[0207] Obtain industrial time series data and obtain a pre-trained time fusion network;

[0208] Using the variational mode decomposition network in the time fusion network, construct a modal decomposition optimization formula according to the pre-configured number of modes, and use the pre-constructed alternating direction algorithm to introduce the industrial time series data and the pre-constructed Lagrange multiplier into the modal decomposition optimization formula to obtain an augmented Lagrangian formula;

[0209] Solve the augmented Lagrangian formula to obtain a frequency component set, where the frequency component set includes multiple modal functions and the central frequency corresponding to each modal function;

[0210] Obtain the temporal convolutional network model in the time fusion network, where the temporal convolutional network model includes a causal convolutional layer, a residual layer, and a fully connected layer;

[0211] Use the causal convolutional layer to perform a convolution operation on the frequency component set to obtain a convolution feature set, and use the residual layer to perform function activation and residual connection operations on the convolution feature set based on a preset number of times to obtain an industrial data feature matrix set;

[0212] Use the fully connected layer to perform a normalized feature full connection operation on the data feature sequence set to obtain an industrial data feature set;

[0213] Use the gated recurrent unit in the time fusion network to perform a feature recognition operation based on temporal features on the industrial data feature set to obtain a temporal feature sequence set;

[0214] According to the preset business scenario, call the downstream prediction network corresponding to the business scenario to predict the temporal feature sequence set to obtain a business prediction result.

[0215] 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 device 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.

[0216] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place, or they may be 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.

[0217] In addition, in each embodiment of the present invention, each functional module may be integrated into a processing unit, may exist separately as individual units physically, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware, or in the form of a combination of hardware and software functional modules.

[0218] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above 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.

[0219] Therefore, from any perspective, 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 encompassed by the present invention. Any associated drawing reference signs in the claims should not be regarded as limiting the claims involved.

[0220] 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. A 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. A blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, etc.

[0221] The embodiments of the present application may 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.

[0222] In addition, obviously, the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices stated in the system claims may also be implemented by one unit or device through software or hardware. Words such as first, second, etc. are used to represent names and do not represent any specific order.

[0223] 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 data processing method for long-time industrial real-time data, characterized in that: The method comprises: Acquire industrial time series data and a pre-trained time fusion network, wherein the time fusion network includes a variational mode decomposition network, a temporal convolutional network model, and a gated recurrent unit; Using the variational mode decomposition network, constructing a mode decomposition optimization formula according to a preconfigured mode number, and using a pre-constructed alternating direction algorithm, introducing the industrial time series data and pre-constructed Lagrangian multipliers into the mode decomposition optimization formula to obtain an augmented Lagrangian formula; Solving the augmented Lagrangian formula to obtain a frequency component set, wherein the frequency component set includes a plurality of modal functions and a center frequency corresponding to each modal function; Acquire the temporal convolutional network model, wherein the temporal convolutional network model includes a causal convolutional layer, a residual layer and a fully connected layer; Using the causal convolution layer to perform a convolution operation on the frequency component set to obtain a convolution feature set, and using the residual layer to perform a function activation and residual connection operation based on a preset number of times on the convolution feature set to obtain an industrial data feature matrix set; Using the fully connected layer to perform a normalized feature full connection operation on the data feature sequence set to obtain an industrial data feature set; Using the gated recurrent unit to perform a feature recognition operation based on time series features on the industrial data feature set to obtain a time series feature sequence set; According to the preset business scenario, the downstream prediction network corresponding to the business scenario is called to predict the time series feature sequence set to obtain a business prediction result.

2. The data processing method for long-time-series industrial real-time data according to claim 1, characterized in that: The obtaining of industrial time series data includes: Use pre-built sensor clusters to obtain real-time collected data within a preset time period; Using a database management tool, performing a data cleaning operation based on missing values ​​and outliers on the real-time collected data to obtain clean time series data; The clean time series data is subjected to a noise reduction operation based on Gaussian filtering to obtain primary noise-reduced time series data, and the primary noise-reduced time series data is subjected to data format standardization processing according to a preset data format standard table to obtain industrial time series data.

3. The data processing method for long-time industrial real-time data according to claim 2, characterized in that: The method of constructing a modal decomposition optimization formula according to the preconfigured modal number includes: Obtain the initialization optimization formula of the pre-built variational mode decomposition algorithm and obtain the pre-configured number of modes; According to the modal number, the initialization optimization formula is substituted to obtain the modal decomposition optimization formula, wherein the modal decomposition optimization formula is expressed as: Where K represents the modal number, min represents the minimum function, and u k represents the kth mode function of the required solution, ω k represents the center frequency of the kth mode function, t represents time, δ(t) represents the unit pulse function, j represents the imaginary unit, represents the Hilbert transform of the kth mode function, Represents the shift of the center frequency from the mode spectrum to the baseband.

4. The data processing method for long-time-series industrial real-time data according to claim 3, characterized in that: The augmented Lagrangian formula is solved to obtain a set of frequency components, including: The augmented Lagrangian formula is obtained, wherein the augmented Lagrangian formula is expressed as: In the formula, represents the augmented Lagrangian formula, λ and λ(t) represent Lagrangian multipliers, α represents the equilibrium parameter, and f(t) represents the industrial time series data; An updated Lagrangian multiplier constraint is obtained, and the augmented Lagrangian formula is iteratively solved according to the updated Lagrangian multiplier constraint to obtain an updated function of each modal function in the frequency domain and a center frequency corresponding to each updated function, wherein the updated function of the modal function in the frequency domain is expressed as: In the formula, represents the updated Fourier transform of the kth mode function, represents the Fourier transform of the industrial time series data, represents the Fourier transform of the ith mode function, represents the Fourier transform of the Lagrange multiplier; The center frequencies corresponding to the update functions are expressed as: In the formula, represents the updated center frequency; Wherein, the updated Lagrange multiplier constraint is expressed as: In the formula, represents the updated Lagrange multiplier, and τ represents the step size.

5. The data processing method for long-time-series industrial real-time data according to claim 4, characterized in that: The obtaining of the temporal convolutional network model comprises: A causal convolutional layer is obtained, wherein the layer function of the causal convolutional layer is expressed as: In the formula, y t represents the output of the causal convolutional layer at time t, k′ represents the filter size, and w i is the i-th filter weight, x t-i represents the input of the i-th filter at time t; Obtain an activation function and a residual block, and perform a preset number of cyclic connection operations on the activation function and the residual block in turn to obtain a residual layer, wherein the convolution process of the activation function is expressed as: (x*df)(t)=∑i=0 k′-1 f(i)·x t-d,i In the formula, x represents the input sequence, d represents the expansion factor, f represents the filter, and x t-d,i It represents the input of the i-th filter after being affected by the expansion factor d at time t; Among them, the residual block is expressed as: o=Activation(x+F(x)) Wherein, o represents the residual result, Activation represents the activation function, and F(x) represents the function learned by the stacked layer in the residual block; Get a fully connected layer, where the fully connected layer represents: In the formula, represents the output result of the lth layer network at time t in the fully connected layer, The output result of the l-1th layer network at time t in the fully connected layer. ReLU represents an activation function used to introduce nonlinearity, and LN represents a normalization function. represents the weight of the lth layer network, Represents the bias of the l-th layer network; The causal convolutional layer, the residual layer and the fully connected layer are connected in sequence to obtain a primary temporal convolutional network model; A training sample set is obtained, and the primary temporal convolutional network model is trained using the training sample set according to a cross entropy loss algorithm and a gradient descent algorithm to obtain a trained primary temporal convolutional network model.

6. The data processing method for long-time-series industrial real-time data according to claim 5, characterized in that: The gated recurrent unit is expressed as: In the formula, r t represents the reset gate at time t, z t represents the update gate at time t, represents the candidate hidden state at time t, h t represents the final hidden state at time t, x t represents the input at time t, W r , W z and W represents the weight matrix, h t-1 represents the final hidden state at time t-1, b r , b z And b represents bias, σ represents sigmoid function, and ⊙ represents element-wise multiplication.

7. The data processing method for long-time-series industrial real-time data according to claim 6, characterized in that: The method of calling a downstream prediction network corresponding to the business scenario according to a preset business scenario, predicting the time series feature sequence set, and obtaining a business prediction result includes: According to the preset business scenario, calling the downstream prediction network corresponding to the business scenario; Acquire a key parameter data set of the business scenario, and allocate the key parameter data set into a training set and a test set according to a preset allocation ratio; Fine-tune the downstream prediction network using the training set to obtain an updated downstream prediction network, and test the updated downstream prediction network using the test set to obtain a test result, wherein the test result includes accuracy, precision, recall rate, and F1 score; When the test result meets the preset standard threshold, the updated downstream prediction network is used to predict the time series feature sequence set to obtain a business prediction result.

8. A data processing device for long-time industrial real-time data, characterized in that: The device comprises: A modal decomposition denoising module is used to obtain industrial time series data, obtain a pre-trained time fusion network, and use a variational modal decomposition network in the time fusion network to construct a modal decomposition optimization formula according to a pre-configured modal number, and use a pre-constructed alternating direction algorithm to introduce the industrial time series data and pre-constructed Lagrangian multipliers into the modal decomposition optimization formula to obtain an augmented Lagrangian formula, and solve the augmented Lagrangian formula to obtain a frequency component set, wherein the frequency component set includes multiple modal functions and the center frequency corresponding to each modal function; A causal feature recognition module, used to obtain a temporal convolutional network model in the time fusion network, wherein the temporal convolutional network model includes a causal convolutional layer, a residual layer and a fully connected layer, and the causal convolutional layer is used to perform a convolution operation on the frequency component set to obtain a convolution feature set, the residual layer is used to perform a function activation and residual connection operation based on a preset number of times on the convolution feature set to obtain an industrial data feature matrix set, and the fully connected layer is used to perform a normalized feature full connection operation on the data feature sequence set to obtain an industrial data feature set; A time series feature extraction module, used to perform a feature recognition operation based on time series features on the industrial data feature set using a gated recurrent unit in the time fusion network to obtain a time series feature sequence set; The downstream service application module is used to call the downstream prediction network corresponding to the business scenario according to the preset business scenario, predict the time series feature sequence set, and obtain the business prediction result.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the data processing method for long-time industrial real-time data as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for processing long-time-series industrial real-time data as claimed in any one of claims 1 to 7 is implemented.

Images