Industrial process monitoring method

Abstract

The invention discloses an industrial process monitoring method, and belongs to the technical field of machine learning, and the method comprises the steps: collecting production process data in a historical industrial process, and constructing a sample data set; time and space information is considered at the same time, a dynamic space-time neighborhood relation is established to apply constraints to input data and features of the auto-encoder, and an adaptive space-time neighborhood feature learning auto-encoder is constructed; training an adaptive space-time neighborhood feature learning auto-encoder by using the sample data set; and industrial process monitoring is realized based on the trained adaptive space-time neighborhood feature learning auto-encoder. The method can adapt to the uncertainty of real industrial process data, and the industrial process monitoring accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and particularly to an industrial process monitoring method. Background Art

[0002] Industry is a symbol of a country's development. It determines the development of the national economic modernization and is the leading force in the national economic development. In modern industry, the scale of the process flow is expanding day by day, which puts forward higher requirements for process safety and product quality. The occurrence of faults may impose a huge burden on the economy and the environment, and the monitoring system is crucial for timely identifying faults and abnormalities. Therefore, building a monitoring system that can identify faults timely and reliably has become a key task.

[0003] With the progress of distributed control systems and intelligent sensors, a large amount of process data has been recorded, thus greatly promoting the development of data-driven monitoring methods. Compared with the first-principle-based methods, data-driven methods do not require accurately establishing physical models, but describe the relationships in complex industrial systems through historical data. As an important branch of data-driven methods, Multivariate Statistical Process Monitoring (MSPM) has been widely studied, including techniques such as Principal Component Analysis (PCA), Partial Least Squares (PLS), and Independent Component Analysis (ICA).

[0004] To further explore the deep information of data, deep learning methods are introduced to handle complex non-linear process data. Related models include Convolutional Neural Network (CNN), Deep Belief Network (DBN), Recurrent Neural Network (RNN), and Autoencoder (AE). The autoencoder realizes feature extraction and data reconstruction through its unsupervised architecture. Due to its superior monitoring performance and easily adjustable network structure, various variants have been derived, including Sparse Autoencoder (SAE), Denoising Autoencoder (DAE), and Variational Autoencoder (VAE). By constructing statistical metrics in the latent layer, faults can be effectively detected. However, these methods do not consider the neighborhood structure of process data. In the encoding and decoding process, only imposing constraints on the Euclidean distance between the input data and the reconstructed data may lead to the destruction of the data topology. In addition, the extracted features may also lose neighborhood information, which will have a negative impact on the accuracy of process monitoring.

[0005] Process variables in industrial data usually exhibit strong topological correlations. Maintaining the data topology during model training helps with feature extraction and reconstruction. However, existing methods mainly perform feature extraction by minimizing the Euclidean distance between data points, which may destroy the neighborhood topology of the data. Therefore, the extracted features lack neighborhood structure information, resulting in inaccurate monitoring results.

[0006] Manifold learning methods, such as Locally Linear Embedding (LLE), Isometric Mapping (ISOMAP), Locality Preserving Projections (LPP), and Neighborhood Preserving Embedding (NPE), etc., can effectively capture the manifold structure of data. Methods that apply manifold learning to autoencoders have gradually emerged, aiming to capture the local and global structures of data.

[0007] Wang et al. proposed a new stacked local preserving autoencoder to more effectively maintain the local data structure in latent features. To make the extracted features more comprehensive, both local and global neighborhood information must be considered simultaneously. Liu et al. proposed a non-local and local structure preserving stacked autoencoder, which combines a regularizer to capture non-local and local data structure information. Yang et al. proposed a new local, non-local, and global preserving stacked autoencoder to extract more comprehensive key structure-related features. The above methods determine neighborhood relationships by calculating the distances between spatially proximate points. However, when addressing the inherent uncertainty in industrial process data, the deterministic structures derived in this way face challenges. Probabilistic manifold methods, such as t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP), describe local and global neighborhood structures in a non-linear probabilistic manner, making them more suitable for real-world industrial processes. Additionally, constructing an adjacency graph in the spatial domain only captures the static information of process data. Due to process inertia and feedback control in industrial systems, temporal dynamics are often overlooked. Considering adjacency relationships in the temporal domain can extract neighborhood features more comprehensively.

[0008] Recent research has applied the concept of preserving spatio-temporal features to the extraction of dynamic features. Song et al. integrated local preserving projection with sequence information, thus achieving spatio-temporal position preserving coordination. Additionally, some research has incorporated spatio-temporal structure constraints into the loss function to extract spatio-temporal features. Liu et al. proposed a spatio-temporal neighborhood preserving stacked autoencoder to learn deep dynamic features. Sampling time distance is used to construct the temporal neighborhood structure. Wang et al. proposed a spatio-temporal neighborhood learning network combined with an average teacher to learn the spatio-temporal neighborhood structure of data. The weights of the temporal neighborhood are measured by the ratio of the time differences between samples. Both methods focus on the influence of the sampling time on the data neighborhood structure, where the closer the sampling times of two samples are, the greater the weight. Wang et al. proposed a new spatio-temporal feature extraction method by developing a local spatio-temporal structure preserving stacked semi-supervised autoencoder, which considers both temporal and spatial correlations when constructing the neighborhood structure. However, the above methods only consider the spatial structure on the temporal neighborhood and lack temporal information in the description of the neighborhood structure. Considering both spatial and temporal information when constructing the neighborhood structure can enable the network to capture more comprehensive spatio-temporal features. Additionally, industrial process data is a slowly changing continuous time series, and different time periods usually exhibit different spatio-temporal features. This may cause the fixed neighborhood relationship to be affected by noise and fluctuations, thereby affecting the monitoring results and leading to inaccurate final monitoring results. Summary of the Invention

[0009] The present invention provides an industrial process monitoring method to solve the technical problem that the features extracted by existing industrial process monitoring methods lack neighborhood structure information, resulting in inaccurate monitoring results.

[0010] To solve the above technical problem, the present invention provides the following technical solutions:

[0011] On the one hand, the present invention provides an industrial process monitoring method, including:

[0012] Collect production process data in the historical industrial process to construct a sample data set;

[0013] Considering both time and space information, establish a dynamic spatio-temporal neighborhood relationship to impose constraints on the input data and features of the autoencoder, and construct an adaptive spatio-temporal neighborhood feature learning autoencoder;

[0014] Use the sample data set to train the adaptive spatio-temporal neighborhood feature learning autoencoder;

[0015] Based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder, realize industrial process monitoring.

[0016] Further, collecting production process data in the historical industrial process to construct a sample data set includes:

[0017] Collect production process data during normal production in the historical industrial process in chronological order, and perform standardization processing on the collected production process data to obtain raw data with zero mean and unit variance, which constitute the sample data set.

[0018] Further, the adaptive spatio-temporal neighborhood feature learning autoencoder first uses the autoencoder model to learn the basic features of the input data through reconstruction constraints. Subsequently, the uniform manifold approximation and projection method is used to calculate the spatial topological structure between adjacent data points in the time neighborhood of the input layer and the hidden layer, and an attention mechanism is introduced to learn the influence of time distance on the topological structure, so as to realize the dynamic adjustment of the spatial topological structure, thereby generating a topological structure containing spatio-temporal information and converting it into a joint probability distribution; by using a cross-entropy variant to measure the similarity of the two joint probability distributions and introducing it into the loss function to constrain the autoencoder, enabling it to capture dynamic spatio-temporal features.

[0019] Further, considering both time and space information, establish a dynamic spatio-temporal neighborhood relationship to impose constraints on the input data and features of the autoencoder, and construct an adaptive spatio-temporal neighborhood feature learning autoencoder, including:

[0020] Construct an autoencoder; wherein, the autoencoder learns the features of the data by minimizing the reconstruction error;

[0021] Consider both time and space information, establish a dynamic spatio-temporal neighborhood relationship, and construct an adaptive spatio-temporal neighborhood feature learning loss function to constrain the autoencoder so that it can capture dynamic spatio-temporal features.

[0022] Furthermore, the calculation process of the adaptive spatio-temporal neighborhood feature learning loss function includes:

[0023] Calculate the adaptive time distance weight of the data;

[0024] Calculate the spatial distance of the data;

[0025] Based on the adaptive time distance weight and spatial distance of the data, calculate the adaptive spatio-temporal manifold loss, and combine it with the loss function of the autoencoder to calculate the adaptive spatio-temporal neighborhood feature learning loss function.

[0026] Furthermore, the calculation of the adaptive time distance weight of the data includes:

[0027] Construct a time distance matrix T d :

[0028]

[0029] where N represents the number of samples in the sample dataset; t represents the current moment, T t represents the sampling time at the current moment, T i represents the sampling time at the i-th moment, i ∈ (t, N), T t+1 represents the sampling time at the next moment, T N represents the sampling time at the N-th moment.

[0030] Determine the local maxima and local minima in the current range time series, calculate the distances from adjacent maxima to maxima, from adjacent maxima to minima, from adjacent minima to maxima, and from adjacent minima to minima respectively, calculate the Shannon entropy for each of the four groups of extreme value distance data, and find the average of the Shannon entropies corresponding to the four groups of extreme value distance data to obtain the attention entropy H attention ;

[0031] Calculate the time weight T w :

[0032]

[0033] Take T w as the input of the self-attention network to obtain the key value k and query value q respectively:

[0034] k = W k T w + bk

[0035] q = W q T w + b q

[0036] where W k and W q represent the weight matrix of the network; b k and b q represent the bias vector of the network;

[0037] Calculate the attention score sim:

[0038]

[0039] where T represents the transpose of the matrix; D is the dimension of k;

[0040] Calculate the attention weight β = softmax(sim); where softmax is an activation function, also known as the normalized exponential function, and its specific formula is

[0041] Calculate the adaptive time distance weight S T = βT d .

[0042] Furthermore, the formula for calculating the spatial distance of the data is expressed as:

[0043]

[0044] where represents the spatial distance between x i and x ij ; x i represents the i-th sample point in the sample dataset; x ij represents the j-th sample point within the temporal neighborhood of x i ; the temporal neighborhood of x i is composed of the K sample points before x i and the K sample points after x i , where K is a preset integer value.

[0045] Furthermore, based on the adaptive time distance weight and spatial distance of the data, calculate the adaptive spatio-temporal manifold loss, and in combination with the loss function of the autoencoder, calculate the adaptive spatio-temporal neighborhood feature learning loss function, including:

[0046] In the original space constituted by the input data, use the heat kernel function to calculate the conditional probability of the spatio-temporal relationship between data neighboring points:

[0047]

[0048] Among them, p i|j represents the relationship between the point j and the point i within the neighborhood centered at the point i; represents x i and x ij 's spatial distance; x i represents the i-th sample point in the sample dataset; x ij represents x i 's j-th sample point within the temporal neighborhood; the temporal neighborhood of x i is composed of the K sample points before x i and the K sample points after x i , where K is a preset integer value; represents the temporal distance weight between x i and x ij ; σ i is a normalization factor, determined in the following way:

[0049]

[0050] Define the joint probability p ij as:

[0051] p ij =(p j|i +p i|j )-p j|i p i|j

[0052] Among them, p j|i represents the relationship between the point i and the point j within the neighborhood centered at the point j;

[0053] Calculate the spatio-temporal relationship probability between the sample points in the target embedding space composed of features and the sample points within their temporal neighborhoods; among them, for the i-th sample point z i in the target embedding space composed of features, the spatio-temporal relationship probability q ij between it and the j-th sample point z ij within its temporal neighborhood is expressed as:

[0054]

[0055] Among them, and are respectively the spatial distance and the adaptive temporal adjustment weight between the two points z i and z ij ; a and b are parameters used to fit the function; the temporal neighborhood of z i is composed of the K sample points before z i and the K sample points after z iIt consists of the subsequent K sample points, where K is a preset integer value;

[0056] Construct an adaptive manifold regularization loss function Loss ASMR :

[0057]

[0058] Among them, N represents the number of samples in the sample dataset; p ij (x) represents the spatio-temporal relationship calculated for the original input data x; q ij (z) represents the spatio-temporal relationship calculated for the feature z;

[0059] Calculate the adaptive spatio-temporal neighborhood feature learning loss function Loss ASMRAE ;

[0060] Loss ASMRAE = Loss MSE + αLoss ASMR

[0061] Among them, Loss MSE is the loss function of the autoencoder, and α is a preset weight parameter.

[0062] Furthermore, the training of the adaptive spatio-temporal neighborhood feature learning autoencoder using the sample dataset includes:

[0063] Divide the training data in the sample dataset into multiple batches and do not shuffle them during the training process, where each batch is sorted by time;

[0064] Initialize the network parameters of the adaptive spatio-temporal neighborhood feature learning autoencoder, input each batch of data into the adaptive spatio-temporal neighborhood feature learning autoencoder in turn to obtain feature data and reconstructed data, calculate the adaptive spatio-temporal neighborhood feature learning loss function, and update the network parameters through backpropagation.

[0065] Furthermore, the implementation of industrial process monitoring based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder includes:

[0066] Based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder, create the T 2 statistic for monitoring the feature space and the SPE statistic for monitoring the residual space through the sample dataset;

[0067] Use the method of kernel density estimation to calculate the control limits of the T 2 statistic and the SPE statistic respectively;

[0068] Collect the production process data in the current industrial process, input the production process data in the current industrial process into the trained adaptive spatio-temporal neighborhood feature learning autoencoder, and calculate the T corresponding to the production process data in the current industrial process based on the output of the adaptive spatio-temporal neighborhood feature learning autoencoder 2 statistic and SPE statistic;

[0069] If the T 2 statistic or SPE statistic corresponding to the production process data in the current industrial process exceeds the corresponding control limit, it is determined that the current industrial process is abnormal; otherwise, it is determined that the current industrial process is normal.

[0070] On the other hand, the present invention also provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the above method.

[0071] On another aspect, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the instruction is loaded and executed by the processor to implement the above method.

[0072] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0073] The solution of the present invention can maintain the spatio-temporal topological structure of data during the process of feature extraction by the autoencoder. The adaptive spatio-temporal neighborhood structure calculation method can more effectively extract the spatio-temporal neighborhood information in the data. At the same time, the probability-based method can adapt to the uncertainty of real industrial process data, improving the accuracy of industrial process monitoring. Description of the Drawings

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0075] Figure 1 is the structural diagram of the adaptive spatio-temporal neighborhood feature learning autoencoder provided by the embodiment of the present invention;

[0076] Figure 2 is the flow chart of realizing industrial process monitoring by using the adaptive spatio-temporal neighborhood feature learning autoencoder provided by the embodiment of the present invention;

[0077] Figure 3 is the system block diagram of the electronic device provided by the embodiment of the present invention. Detailed Embodiments

[0078] To make the objectives, technical solutions, and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0079] First of all, it should be noted that in the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" in the present invention should not be construed as being more preferred or more advantageous than other embodiments or design solutions. Exactly, the use of the word "exemplarily" is intended to present concepts in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two can be selected.

[0080] First Embodiment

[0081] Regarding the problem that the features extracted by the existing industrial process monitoring methods lack neighborhood structure information, resulting in inaccurate monitoring results, this embodiment simultaneously considers time and space information to construct a dynamic spatio-temporal neighborhood relationship, imposes constraints on the input data and features to capture spatio-temporal neighborhood information, and proposes an adaptive spatio-temporal neighborhood feature learning autoencoder, which describes the topological structure between data points in a probabilistic manner, enhances the tolerance to process uncertainty, and at the same time retains the spatio-temporal neighborhood structure to ensure the consistency of data topology and improve the monitoring ability of the model. And considering the dynamic time weight, the spatial topology structure is adaptively adjusted using time information to cope with the influence of noise and fluctuations, and more accurately describe the spatio-temporal neighborhood structure. The generated elements contain comprehensive spatio-temporal neighborhood information, thereby improving the monitoring performance of the model. On this basis, this embodiment proposes an industrial process monitoring method based on an adaptive spatio-temporal neighborhood feature learning autoencoder, which includes:

[0082] S1. Collect the production process data in the historical industrial process and construct a sample data set;

[0083] Specifically, in this embodiment, the implementation process of the above S1 is as follows: Collect the production process data during normal production in the historical industrial process in chronological order, and perform standardization processing on the collected production process data to obtain the original data with zero mean and unit variance, and form a sample data set. This method can be widely applied to a variety of large industrial processes, such as chemical processes, aluminum electrolysis production processes, and blast furnace ironmaking processes. Taking the production process of vinyl acetate as an example, usually, data of 29 monitoring variables such as feed rate, flow rate, liquid level, pressure, temperature, and concentration of each instrument at each position are collected.

[0084] S2. Simultaneously consider time and space information, establish a dynamic spatio-temporal neighborhood relationship to impose constraints on the input data and features of the autoencoder, and construct an adaptive spatio-temporal neighborhood feature learning autoencoder;

[0085] Specifically, the structure of the adaptive spatio-temporal neighborhood feature learning autoencoder in this embodiment is as follows Figure 1 as shown, and the process of learning the probabilistic neighborhood features is as follows:

[0086] First, using the autoencoder model, the basic features of the input data are learned through reconstruction constraints; subsequently, the unified manifold approximation and projection method is used to calculate the spatial topological structure between adjacent data points in the time neighborhood of the input layer and the hidden layer, and an attention mechanism is introduced to learn the influence of time distance on the topological structure. This mechanism can realize the dynamic adjustment of the spatial topological structure, thereby generating a topological structure containing spatio-temporal information and converting it into a joint probability distribution. By using a cross-entropy variant to measure the similarity between the two joint probability distributions and introducing it into the loss function, the autoencoder is constrained to enable it to capture dynamic spatio-temporal features.

[0087] Specifically, the construction process of the adaptive spatio-temporal neighborhood feature learning autoencoder is as follows:

[0088] S21, Time neighborhood selection. For the dataset sorted by time, its time neighborhood is defined as:

[0089]

[0090] where represents the time neighborhood of sample x i . 2K represents the length of the neighborhood, and its value needs to be set manually according to the data characteristics of different industrial objects, and its optimal value needs to be obtained through multiple experimental tests.

[0091] S22, Construct the autoencoder network;

[0092] Among them, the autoencoder usually consists of an encoder and a decoder, and the two have the same number of network layers. For the input data x i , the feature z i can be obtained through encoder calculation. is the hidden layer of the encoder. Then the reconstructed data is obtained through the decoder. The hidden layer of the decoder. The autoencoder learns the characteristics of the data by minimizing the reconstruction error. For a dataset of N samples, the loss function can be written as:

[0093]

[0094] S23, Calculate the adaptive time distance weight of the data; specifically including:

[0095] S231, Construct the time distance matrix T d :

[0096]

[0097] This matrix describes the sampling intervals between sampling points. When the data fluctuates greatly in a time period, the data within a short time becomes irrelevant, and the influence of adjacent points should be reduced, that is, the time distance of adjacent points is enlarged, and vice versa. Among them, N represents the number of samples in the sample dataset; t represents the current moment, T t represents the sampling time at the current moment, T i represents the sampling time at the i-th moment, i ∈ (t, N), T t+1 represents the sampling time at the next moment, T N represents the sampling time at the N-th moment.

[0098] S232. To measure the influence of data volatility in different time periods on the time weight, attention entropy is used for calculation. The specific implementation method is as follows: First, determine the local maximum and local minimum values in the current range time series (data within the time neighborhood DT in the above text) as key points, and calculate the distances from adjacent maximum values to maximum values, from adjacent maximum values to minimum values, from adjacent minimum values to maximum values, and from adjacent minimum values to minimum values respectively. Among them, the distance between two extreme values refers to the difference in sampling time between two points. Each variable calculates the Shannon entropy for the four groups of data respectively and takes the average to obtain the attention entropy H attention . Specifically: For a variable of all sample points within the range, calculate the four groups of time difference distance data, calculate the corresponding four Shannon entropies, take an average once to obtain the attention entropy of this variable, and perform the above operations for each variable of the sample points, and then take the average to obtain the overall attention entropy.

[0099] S233. Calculate the time weight T w :

[0100]

[0101] S234. Take T w as the input of the self-attention network, and obtain the key value k and the query value q respectively:

[0102] k = W k T w + b k (5)

[0103] q = W q T w + b q (6)

[0104] Among them, {W k , b k} and {W q , b q} represent the weight matrix and bias vector of the network respectively;

[0105] S235, obtain the attention score sim by calculating the similarity between the key value and the query value:

[0106]

[0107] where T represents the transpose of the matrix; D is the dimension of k;

[0108] S236, calculate the attention weight β:

[0109] β = softmax(sim) (8)

[0110] where softmax is an activation function, also known as the normalized exponential function, and its specific formula is

[0111] S237, calculate the adaptive time distance weight S T :

[0112] S T = βT d (9)

[0113] S24, calculate the spatial distance of the data: Points that are closer in time will be given more importance, while points that are far apart in time, even if they are close in space, should be relatively ignored. So for the sample point x i , is the sample point in the time neighborhood of x i , and the spatial distance calculation method is as follows:

[0114]

[0115] S25, calculate the adaptive spatio-temporal manifold loss; specifically including:

[0116] S251, in the original space composed of the input data, use the heat kernel function to calculate the conditional probability p of the spatio-temporal relationship between data neighboring points i|j :

[0117]

[0118] where represents the time distance weight between x i and x ij ; σ i is the normalization factor, determined in the following way:

[0119]

[0120] S252, to ensure the symmetry of the probability, define the joint probability p ij as:

[0121] p ij = (p j|i + p i|j ) - p j|i p i|j (13)

[0122] Among them, p i|j represents the relationship between the point j and the point i within the neighborhood when taking the point i as the center; p j|i represents the relationship between the point i and the point j within the neighborhood when taking the point j as the center; in order to ensure that the relationship between two points is a unique value and the result will not be inconsistent due to the calculation order, the joint probability p ij is designed.

[0123] S253. Correspondingly, for a point z i in the target embedding space composed of features, is the point within its temporal neighborhood, and the spatio-temporal relationship probability between the two points is:

[0124]

[0125] Among them, and are respectively the spatial distance and the adaptive time adjustment weight between the two points z i and z ij ; a and b are parameters used to fit the function; The calculation method is the same as formula (10), and the calculation method of

[0126] S254. In order to measure the consistency between two probability distributions, an adaptive manifold regularization loss function Loss ASMR is constructed:

[0127]

[0128] The first term of the above loss function can make the neighboring points in the original space closer in the target embedding space, and the second term can make the non-neighboring points in the original space farther in the target embedding space.

[0129] Among them, p ij (x) represents the spatio-temporal relationship calculated for the original input data x; q ij (z) represents the spatio-temporal relationship calculated for the feature z;

[0130] S255. Calculate the adaptive spatio-temporal neighborhood feature learning loss function Loss ASMRAE ;

[0131] Loss ASMRAE = Loss MSE+α Loss ASMR (16)

[0132] Among them, α is a parameter used to balance the sample reconstruction loss and the manifold constraint loss.

[0133] S3. Use the sample data set to train the adaptive spatio-temporal neighborhood feature learning autoencoder;

[0134] Specifically, in this embodiment, the implementation process of the above S3 is as follows: First, divide the training data into multiple batches and do not shuffle them during the training process. Among them, the data in each batch is sorted by time. Establish the adaptive spatio-temporal neighborhood feature learning autoencoder in step 2 and initialize the network parameters. Input the normalized data x = [x1, x2, …, x N T into the network to obtain the feature z = [z1, z2, …, z N T and the reconstructed data Calculate the loss function according to Equation (16) and update the network parameters through backpropagation.

[0135] S4. Implement industrial process monitoring based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder;

[0136] Specifically, in this embodiment, as Figure 2 shown, the implementation process of the above S4 is as follows:

[0137] S41. Based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder, create the T 2 statistic and the SPE statistic through the sample data set; Among them,

[0138] The T 2 statistic is used to monitor the feature space, and the calculation method is:

[0139]

[0140] Among them, Σ z represents the variance of z.

[0141] The SPE statistic is used to monitor the residual space, and the calculation method is:

[0142]

[0143] S42. Use the method of kernel density estimation to calculate the control limits of the T 2 statistic and the SPE statistic; During the online monitoring process, if the statistic exceeds the control limit, it indicates that a fault has occurred in the industrial process.

[0144] ​​S43. Collect the production process data in the current industrial process, input it into the trained adaptive spatio-temporal neighborhood feature learning autoencoder, and calculate the T statistic and SPE statistic corresponding to the production process data in the current industrial process based on the output of the adaptive spatio-temporal neighborhood feature learning autoencoder. 2 Statistic and SPE statistic;

[0145] S44. If the T statistic or SPE statistic corresponding to the production process data in the current industrial process exceeds the corresponding control limit, it is determined that the current industrial process is abnormal; otherwise, it is determined that the current industrial process is normal. 2 Statistic or SPE statistic exceeds the corresponding control limit, then it is determined that the current industrial process is abnormal; otherwise, it is determined that the current industrial process is normal.

[0146] In summary, this embodiment provides an adaptive spatio-temporal manifold regularization autoencoder, designs a new loss function, describes the topological structure between data points in a probabilistic manner, enhances the tolerance to process uncertainty, retains the spatio-temporal neighborhood structure at the same time, and ensures the consistency of data topology; and designs a new dynamic time distance weight, which can adaptively adjust the spatial topological structure by using time information, so as to cope with the influence of noise and fluctuations, more accurately describe the spatio-temporal neighborhood structure, and improve the monitoring performance of the model.

[0147] Second Embodiment

[0148] This embodiment provides an electronic device. As shown, the electronic device includes: a processor and a memory; wherein, the processor and the memory can be connected through a communication bus; at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the method of the first embodiment above. In addition, the electronic device may further include a transceiver, and the processor and the transceiver can be connected through a communication bus, and the transceiver is used for communicating with other devices. Figure 3

[0149] Figure 3 Next, the specific components of the electronic device will be introduced in combination with

[0150] ​​Among them, the processor is the control center of the electronic device. The electronic device may include multiple processors, and each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may be a single processor or a collective term for multiple processing elements. For example, the processor may be one or more central processing units (CPUs), or other general-purpose processors, application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement the embodiments of the present invention. For example: one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor can execute various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0151] In a specific implementation, as an embodiment, the processor may include one or more CPUs. For example Figure 3 CPU0 and CPU1 shown in, of course, this is only an illustrative example.

[0152] The memory is used to store the software program for executing the solution of the present invention and is controlled by the processor for execution. The specific implementation manner can refer to the above method embodiments and will not be elaborated here.

[0153] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or may also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and be coupled to the processor through the interface circuit ( Figure 3 not shown) of the electronic device. The embodiments of the present invention do not make specific limitations thereto.

[0154] The transceiver may include a receiver and a transmitter ( Figure 3 not shown separately). Among them, the receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver may be integrated with the processor or exist independently and be coupled to the processor through the interface circuit ( Figure 3 not shown) of the electronic device. The embodiments of the present invention do not make specific limitations thereto.

[0155] In addition, it should be noted that Figure 3 the structure of the electronic device shown in does not constitute a limitation on the device. The actual device may include more or fewer components than shown in the figure, or combine certain components, or have a different component layout. In addition, the technical effects achieved by the electronic device when executing the method of the first embodiment above may refer to the technical effects described in the first embodiment above, so they will not be elaborated here.

[0156] Third Embodiment

[0157] This embodiment provides a computer-readable storage medium, in which at least one instruction is stored. The instruction is loaded and executed by a processor to implement the method of the first embodiment above. Among them, the computer-readable storage medium may be ROM, random access memory, CD-ROM, magnetic tape, floppy disk, and optical data storage devices, etc. The instructions stored therein can be loaded and executed by the processor in the terminal to implement the above method.

[0158] In addition, it should be noted that the present invention can be provided as a method, an apparatus, or a computer program product. Therefore, the embodiments of the present invention can take the form of all or part of a hardware embodiment, all or part of a software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented using software, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center containing one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0159] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing terminal device generate a device for implementing the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0160] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1The functions specified in one or more boxes. These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one or more processes and / or boxes. Figure 1 One process or more processes and / or boxes Figure 1 Steps for implementing the functions specified in one or more boxes.

[0161] It should also be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising the said element. In addition, the term "and / or" is only a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Among them, A and B can be singular or plural. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context. "At least one" means one or more, and "a plurality" means two or more. "At least one of the following (items)" or similar expressions refer to any combination of these items, including any combination of single items (items) or plural items (items). For example, at least one of a, b or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, c can be single or multiple.

[0162] In addition, it can be understood that, in various embodiments of the present invention, the magnitude of the sequence numbers of the above processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0163] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0164] 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 functional modules / units is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. One can select some or all of the units according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0165] If the method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. The foregoing storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0166] Finally, it should be noted that the above description is only the preferred embodiment of the present invention. It should be pointed out that although the preferred embodiments of the present invention have been described, for those of ordinary skill in the art, once the basic creative concept of the present invention is known, several improvements and refinements can be made without departing from the principle described in the present invention. These improvements and refinements should also be regarded as the protection scope of the present invention. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. An industrial process monitoring method, characterized in that, Including: Collecting production process data in historical industrial processes to construct a sample data set; Considering both time and space information simultaneously, establishing a dynamic spatio-temporal neighborhood relationship to impose constraints on the input data and features of the autoencoder, and constructing an adaptive spatio-temporal neighborhood feature learning autoencoder; Using the sample data set to train the adaptive spatio-temporal neighborhood feature learning autoencoder; Based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder, realizing industrial process monitoring.

2. The industrial process monitoring method according to claim 1, characterized in that The collecting production process data in historical industrial processes to construct a sample data set includes: Collecting production process data during normal production in historical industrial processes in chronological order, and performing standardization processing on the collected production process data to obtain raw data with zero mean and unit variance, which form a sample data set.

3. The industrial process monitoring method according to claim 1, characterized in that, The adaptive spatio-temporal neighborhood feature learning autoencoder first uses the autoencoder to learn the basic features of the input data through reconstruction constraints. Subsequently, it uses the uniform manifold approximation and projection method to calculate the spatial topological structure between adjacent data points in the time neighborhood of the input layer and the hidden layer, and introduces an attention mechanism to learn the influence of time distance on the topological structure, so as to realize the dynamic adjustment of the spatial topological structure, thereby generating a topological structure containing spatio-temporal information and converting it into a joint probability distribution; by using a cross-entropy variant to measure the similarity of the two joint probability distributions and introducing it into the loss function, the autoencoder is constrained to enable it to capture dynamic spatio-temporal features.

4. The industrial process monitoring method according to claim 1, characterized in that, The considering both time and space information simultaneously, establishing a dynamic spatio-temporal neighborhood relationship to impose constraints on the input data and features of the autoencoder, and constructing an adaptive spatio-temporal neighborhood feature learning autoencoder includes: Constructing an autoencoder; among them, the autoencoder learns the features of the data by minimizing the reconstruction error; Considering both time and space information simultaneously, establishing a dynamic spatio-temporal neighborhood relationship to construct an adaptive spatio-temporal neighborhood feature learning loss function to constrain the autoencoder to enable it to capture dynamic spatio-temporal features.

5. The industrial process monitoring method according to claim 4, characterized in that, The calculation process of the adaptive spatio-temporal neighborhood feature learning loss function includes: Calculating the adaptive time distance weight of the data; Calculating the spatial distance of the data; Based on the adaptive time distance weight and spatial distance of the data, calculating the adaptive spatio-temporal manifold loss, and combining it with the loss function of the autoencoder to calculate the adaptive spatio-temporal neighborhood feature learning loss function.

6. The industrial process monitoring method according to claim 5, characterized in that, The calculating the adaptive time distance weight of the data includes: Construct the time distance matrix T d : Among them, N represents the number of samples in the sample dataset; t represents the current moment, and T t represents the sampling time at the current moment, T i represents the sampling time at the i-th moment, i ∈ (t, N), T t+1 represents the sampling time at the next moment, T N represents the sampling time at the N-th moment; Determine the local maxima and local minima in the current range time series, calculate the distances from adjacent maxima to maxima, from adjacent maxima to minima, from adjacent minima to maxima, and from adjacent minima to minima respectively. Calculate the Shannon entropy for each of the four sets of extreme value distance data obtained, and find the average of the Shannon entropies corresponding to the four sets of extreme value distance data to obtain the attention entropy H attention ; Calculate the time weight T w : Take T w as the input of the self-attention network, and obtain the key value k and the query value q respectively: k = W k T w + b k q = W q T w + b q Among them, W k and W q represent the weight matrix of the network; b k and b q represent the bias vector of the network; Calculating the attention score sim: where, T represents the transpose of the matrix; D is the dimension of k; Calculating the attention weight β = softmax(sim); where, softmax(.) represents the activation function; Calculate the adaptive time-distance weight S T = βT d .

7. The industrial process monitoring method according to claim 5, characterized in that The formula for calculating the spatial distance of the data is expressed as: Among them, represents the spatial distance between x i and x ij ; x i represents the i-th sample point in the sample dataset; x ij represents the j-th sample point within the time neighborhood of x i ; the time neighborhood of x i consists of the K sample points before x i and the K sample points after x i , where K is a preset integer value.

8. The industrial process monitoring method according to claim 5, characterized in that, The based on the adaptive time distance weight and spatial distance of the data, calculating the adaptive spatio-temporal manifold loss, and combining it with the loss function of the autoencoder to calculate the adaptive spatio-temporal neighborhood feature learning loss function includes: In the original space composed of input data, the heat kernel function is used to calculate the conditional probability p of the spatio-temporal relationship between the neighboring points of the data i|j : Among them, p i|j represents the relationship between the point j and the point i within the neighborhood centered at the point i; represents x i and x ij 's spatial distance; x i represents the i-th sample point in the sample dataset; x ij represents the j-th sample point within the time neighborhood of x i ; the time neighborhood of x i is composed of the first K sample points before x i and the K sample points after x i , where K is a preset integer value; represents the time distance weight between x i and x ij ; σ i is a normalization factor, determined in the following manner: where K is a preset integer value; define the joint probability p ij as: p ij = (p j|i + p i|j ) - p j|i p i|j where p j|i represents the relationship between the point i and the point j within the neighborhood centered at the point j Calculate the spatio-temporal relationship probability between the sample points in the target embedding space composed of features and the sample points within their temporal neighborhoods; where, for the $i$-th sample point $z$ in the target embedding space composed of features i , and the $j$-th sample point $z$ ij within its temporal neighborhood, the spatio-temporal relationship probability $q$ ij is expressed as: wherein, and are the spatial distance between two points z i and z ij respectively, and a and b are parameters for fitting the function; the time neighborhood of z i is composed of the first K sample points located before z i and the last K sample points located after z i , where K is a preset integer value; Construct the Adaptive Manifold Regularization Loss Function Loss ASMR : Among them, N represents the number of samples in the sample dataset; p ij (x) represents the spatio-temporal relationship calculated for the original input data x; q ij (z) represents the spatio-temporal relationship calculated for the feature z; Calculate the adaptive spatio-temporal neighborhood feature learning loss function Loss ASMRAE ; Loss ASMRAE = Loss MSE + αLoss ASMR Among them, Loss MSE is the loss function of the autoencoder, and α is a preset weight parameter.

9. The industrial process monitoring method according to claim 1, characterized in that The using the sample data set to train the adaptive spatio-temporal neighborhood feature learning autoencoder includes: Dividing the training data in the sample data set into multiple batches, and not shuffling them during the training process, where each batch is sorted by time; Initialize the parameters of the adaptive spatio-temporal neighborhood feature learning autoencoder network, sequentially input each batch of data into the adaptive spatio-temporal neighborhood feature learning autoencoder to obtain feature data and reconstructed data, calculate the adaptive spatio-temporal neighborhood feature learning loss function, and update the network parameters through backpropagation.

10. The industrial process monitoring method according to claim 1, characterized in that, Implement industrial process monitoring based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder, including: Based on the trained adaptive spatio-temporal neighborhood feature learning autoencoder, create the T statistic for monitoring the feature space and the SPE statistic for monitoring the residual space through the sample data set. 2 Statistic and the SPE statistic for monitoring the residual space; Calculate the control limits of the T 2 statistic and the SPE statistic respectively using the kernel density estimation method; Collect the production process data in the current industrial process, input the production process data in the current industrial process into the trained adaptive spatio-temporal neighborhood feature learning autoencoder, and calculate the T 2 statistic and SPE statistic corresponding to the production process data in the current industrial process; If the T 2 statistic or the SPE statistic corresponding to the production process data in the current industrial process exceeds the corresponding control limit, it is determined that the current industrial process is abnormal; otherwise, it is determined that the current industrial process is normal.

Images