Missing data filling method for industrial soft measurement
By using Bi-LSTM and graph convolution network technologies during nickel flash furnace smelting, the problem of sensor data is solved, high-precision and fast data filling are achieved, data integrity and computing efficiency are ensured, and data integrity and calculation efficiency are improved. The level of intelligence in the production process.
Patent Information
- Application Number
- CN202411121413.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-15
AI Technical Summary
The data loss caused by high temperature, corrosive gases, vibration or wear during the nickel flash furnace smelting process, affecting process parameter optimization and product quality, and insufficient model training results in the prediction and control strategy being unable to achieve its maximum potential. .
Using missing data filling method for industrial soft measurement, the auxiliary variables of the target variable are determined, and predicted values are generated based on time correlation and spatial dependencies. Bi-LSTM, random span sampling and graph convolution network are used to fill data to build relationships. Topology graphs and adjacency matrix perform feature decomposition and graph convolution operations to realize intelligent filling of missing data.
It realizes high-precision and rapid data generation, reduces the complexity of model calculations, ensures data integrity and accuracy, and provides solid data support for production process optimization and product quality improvement.
Smart Images

Figure CN119025838A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of sensor data anomaly processing, and more particularly to a missing data filling method for industrial soft measurement. Background Art
[0002] In the modern metallurgical industry, especially in the nickel flash furnace smelting process, real-time monitoring and data analysis of the smelting process are crucial. The nickel flash furnace uses a high-temperature rapid oxidation reaction. Through the precise ratio of concentrate ore, flux and oxygen-enriched air, the concentrate can be oxidized in a very short time within the temperature range of 1450 to 1550°C. The molten material then falls into the sedimentation tank for sedimentation and separation, thereby achieving efficient extraction of nickel. This process requires not only precise control of materials, but also continuous monitoring of the internal conditions of the reaction tower, including key parameters such as temperature, pressure, and gas composition, to ensure reaction efficiency and product quality.
[0003] However, traditional sensor monitoring systems face many challenges, especially reliability issues in extreme working environments. Sensors may fail due to factors such as high temperature, corrosive gases, vibration or wear, resulting in data loss.
[0004] These missing data not only affect the accurate analysis of the actual working conditions of the smelting process, but also hinder the optimization of process parameters based on real-time data, which may have a negative impact on production efficiency and product quality. In addition, the missing data in the smelting process may also lead to insufficient model training, making it impossible for data-driven prediction and control strategies to achieve their maximum potential.
[0005] Therefore, developing a method that can effectively process and fill missing data is of great significance for maintaining the stability and reliability of the smelting process monitoring system and improving the intelligence level of the entire production process. Summary of the invention
[0006] In view of this, the present invention proposes a missing data filling method for industrial soft measurement, aiming to solve the problem of missing data caused by sensor failure or system shutdown in nickel flash furnace smelting process monitoring.
[0007] In order to achieve the above object, the present invention adopts the following technical solution:
[0008] A missing data filling method for industrial soft measurement, comprising:
[0009] Determine an auxiliary variable of the target variable, generate a first prediction value according to the time correlation of existing data in the auxiliary variable, and fill in missing data in the auxiliary variable using the first prediction value;
[0010] The target variable is predicted based on the spatial dependency between the filled auxiliary variables and the target variable.
[0011] Preferably, the temporal correlation of the data in the auxiliary variables is extracted by Bi-LSTM.
[0012] Preferably, generating the first prediction value according to the time correlation of the existing data in the auxiliary variable comprises:
[0013] According to the distribution of existing data, the noise is gradually diffused forward;
[0014] Based on the diffusion-noised data, random span sampling is adopted, and the first prediction value is generated by reverse denoising considering the time correlation.
[0015] As a preferred method, the data distribution when diffused to the nth step is:
[0016]
[0017] In the formula, L represents the length of the existing data, t represents the data at the tth moment, and α is defined as t =1-β t , β t is the noise added at time t. is the variance, and I is the identity matrix.
[0018] Preferably, the step of adopting random span sampling and taking into account time correlation and performing inverse denoising to generate the first prediction value comprises:
[0019] Considering the time correlation to fit the noise,
[0020] From the current data Remove the fitting noise, sample with random span s, and get the data distribution of the nsth step; and so on, get the overall data distribution after random span sampling of the diffusion and noise data;
[0021] The overall data distribution is input into the linear layer to obtain the first prediction value.
[0022] Preferably, the noise fitting data is:
[0023]
[0024] In the formula, ε and ε θ are the original noise and the fitted noise respectively, n,H t Where n is the number of steps and Ht is the time correlation; it is used to enable the model to generate missing data with time orientation.
[0025] Preferably, the formula for inputting the overall data distribution into the linear layer is expressed as:
[0026] x t ' +1 =Linear(X t ' +1 )+bias
[0027] Where, X t ' +1 It represents the overall data distribution, and bias represents the deviation rate.
[0028] Preferably, the target variable is predicted based on the spatial dependency relationship between the filled auxiliary variable and the target variable; including:
[0029] Construct a relationship topology graph based on auxiliary variables and target variables, and construct an adjacency matrix based on the correlation between different variables;
[0030] The adjacency matrix is decomposed by symmetric normalized Laplace matrix to obtain the eigenvector matrix;
[0031] The target variable is predicted based on the feature vector matrix based on graph convolution.
[0032] Preferably, the process of performing a graph convolution operation on the feature vector matrix to obtain the second prediction value includes:
[0033] The convolution kernel h and feature vector V of the graph convolution are Fourier transformed respectively, and then multiplied to obtain the new feature representation X t , the public notice is:
[0034]
[0035] U represents the Fourier transform of the graph, U t h、U t V represents the result of Fourier transform of convolution kernels h and V respectively;
[0036] For feature X t Perform graph convolution operation to obtain the second prediction value; the public expression is:
[0037]
[0038] In the formula, y t ' +1 represents the second predicted value, X t is the auxiliary variable input at time t, A represents the adjacency matrix, W t represents the GCN network parameters, and σ is the activation function.
[0039] Preferably, the adjacency matrix is constructed based on the correlation between variables according to the following formula:
[0040]
[0041] In the formula, e ij represents the correlation between variables, which is calculated by KL divergence; μ represents the correlation threshold, a ij ∈A represents an element in the adjacency matrix, and i and j represent the types of auxiliary variables.
[0042] Preferably, when the loss of generating the first predicted value and the loss of predicting the target variable are about half of their respective initial losses, the total loss of the prediction process is introduced;
[0043] The loss function for generating the first prediction is:
[0044]
[0045] The loss function for predicting the target variable is:
[0046]
[0047] In the formula, loss noise Represents the noise fitting error, loss FGCDM represents the total loss of the prediction process, x′ t+1 is the first predicted value, x t+1 Indicates the target value corresponding to the first predicted value, y′ t+1 represents the predicted value of the target variable, y t+1 Indicates the target value corresponding to the predicted value of the target variable.
[0048] As a preference, the total loss of the prediction process FGCDM for:
[0049]
[0050] Preferably, the isolation forest algorithm is used to detect the data, determine the location of missing data and the amount of missing data, and fill in the data according to the missing location and the amount of missing data.
[0051] It can be seen from the above technical solutions that the present invention discloses a missing data filling method for industrial soft measurement, which uses an intelligent algorithm and model to predict and fill missing values using an existing and complete data set; compared with the prior art, the present application has the following technical effects:
[0052] (1) FCDM not only has high-precision data generation performance, but also has a faster calculation process, which can effectively reduce the model calculation complexity. At the same time, it also shows good prediction performance in the GCN-based soft sensor model;
[0053] (2) The mild loss feedback strategy proposed in the present invention can make the soft measurement model of the target variable have higher accuracy and achieve the goal of more reasonable prediction;
[0054] (3) The present invention can ensure the integrity and accuracy of data, thereby providing solid data support for the optimization of the production process and the improvement of product quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0056] Figure 1 This is a flow chart of the missing data filling method for industrial soft measurement of the present invention;
[0057] Figure 2 It is a data processing flow chart of the fast conditional diffusion model (FCDM) of the present invention;
[0058] Figure 3 This is a flow chart of the present invention for extracting time-related information through Bi-LSTM;
[0059] Figure 4 Schematic diagram of data diffusion process in the fast conditional diffusion model of the present invention;
[0060] Figure 5 This is a one-dimensional convolutional neural network structure diagram of the present invention;
[0061] Figure 6 This is a data processing flow chart of the graph convolution (GCN) of the present invention. DETAILED DESCRIPTION
[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0063] The embodiment of the present invention discloses a missing data filling method for industrial soft measurement, referring to Figure 1 , mainly includes two parts, namely
[0064] Determine an auxiliary variable of the target variable, generate a first prediction value according to the time correlation of existing data in the auxiliary variable, and fill in missing data in the auxiliary variable using the first prediction value;
[0065] The target variable is predicted based on the spatial dependency between the filled auxiliary variables and the target variable.
[0066] In an exemplary embodiment,
[0067] Part I, reference Figure 2 ;
[0068] This embodiment uses Bi-LSTM to extract the time correlation of existing data in the auxiliary variable and generates a first prediction value based on the fast conditional diffusion model (FCDM);
[0069] Specifically, the structure of Bi-LSTM is as follows: Figure 3 As shown, it includes an input layer, a forward layer, a reverse layer and an output layer. The process of extracting time correlation based on existing data is a prior art and will not be repeated here. It should be noted that in this application, the existing data is dynamically acquired through a moving window.
[0070] Get the time correlation H t After that, it is further input into FCDM together with the data of the moving window to realize the generation of the first prediction value; assuming the moving window length is L, the local data in the window is X t-L:t ={x t-L ,x t-L+1 ,x t-L+2 ,L,x t-1 ,x t};
[0071] This application uses Bi-LSTM network training and parameter optimization to make the hidden variable H at time t t Contains the historical information of local data, and finally obtains the time dependency relationship at time t+1, by obtaining the hidden variable H t , providing guidance information for FCDM to achieve the time-oriented missing data generation task.
[0072] Reference Figure 4 , set the total number of FCDM diffusion steps to N, n is the current diffusion step, and the data processing process is:
[0073] First, the noise is added by forward diffusion step by step according to the distribution of the existing data in the auxiliary variables; when the missing data at time t+1 is predicted and filled in time-dependently, the data distribution at the time of diffusion to the nth step can be calculated by the initial data distribution in the moving window, and the calculation formula is:
[0074]
[0075] In the formula, L represents the length of the existing data, t represents the data at the tth moment, and α is defined as t =1-β t , β t is the noise added at time t. is the variance, and I is the identity matrix.
[0076] Secondly, based on the forward diffusion noisy data, random span sampling is adopted to reversely generate the first prediction value considering the time correlation; in this application, the random span, that is, the step length is a random value s and is not fixed, so that the network performs sampling at the ns-th diffusion step, which can effectively reduce the amount of calculation and training time in the reverse process, speed up the model training speed and reduce the computational complexity, thereby realizing fast generation of missing data.
[0077] In the training phase of the reverse process, the time-dependent information H about the t+1th moment is introduced into the FCDM. t As additional information, the model is enabled to generate missing data in a time-oriented manner, including:
[0078] Considering the time correlation, the noise is fitted. In this embodiment, a one-dimensional convolutional neural network is used for noise fitting. Its structure is as follows: Figure 5 As shown, it includes input layer, convolution layer, pooling layer and output layer; and the fitted noise data is:
[0079]
[0080] In the formula, ε and ε θ are the original noise and the fitted noise respectively. n,H t Where n is the number of steps and Ht is the time correlation, which enables the model to generate missing data in a time-oriented manner.
[0081] Then, from the current data Remove the fitting noise, sample with random span s, and get the data distribution of the nsth step, that is, By analogy, the overall data distribution after random span sampling of diffuse and noisy data is obtained;
[0082] Finally, the overall data distribution is input into the linear layer to obtain the first prediction value, that is,
[0083] x t ' +1 =Linear(X t ' +1 )+bias
[0084] Where, X t ' +1It represents the overall data distribution, and bias represents the deviation rate.
[0085] In this embodiment, when the moving window is on the existing data sequence, FCDM uses the actual next moment data as labels and performs parameter optimization, and the obtained data is the prediction data guided by the time-dependent information of the latent variables. When the moving window reaches the missing area, the trained FCDM is used to generate the time-dependent missing data. The prediction results are then used to fill the missing values.
[0086] Part II,
[0087] Based on the spatial dependency relationship between the filled auxiliary variables and the target variable, the target variable is predicted; in this embodiment, Figure 6 As shown, including:
[0088] (1) A relationship topology diagram is constructed based on the auxiliary variables and the target variable, and an adjacency matrix is constructed based on the correlation between different variables; that is, the data matrix F composed of the auxiliary variables in the window and the target variable data Y = {y t-L ,y t-L+1 ,L,y t}, we can construct the adjacency matrix A and obtain the spatial dependencies between nodes (variables).
[0089] In this application, the auxiliary variable sensor X={X1,X2,L,X k} are regarded as nodes of the graph, and the spatial coupling relationship between sensors is regarded as the edge of the graph; and the KL divergence is used to calculate the correlation between variables e ij , and construct the adjacency matrix based on the correlation between auxiliary variables according to the following formula;
[0090]
[0091] In the formula, e ij represents the correlation between auxiliary variables, which is calculated by KL divergence; μ represents the correlation threshold, a ij ∈A represents an element in the adjacency matrix, and i and j represent the types of auxiliary variables.
[0092] When e ij When >μ, it means that there is a strong correlation between the i-th auxiliary variable and the j-th auxiliary variable and there is an edge connection, e ij <μ indicates weak association and no edge connection.
[0093] (2) Using the symmetric normalized Laplace matrix to perform eigendecomposition on the adjacency matrix to obtain the eigenvector matrix;
[0094] The decomposition process is:
[0095]
[0096] Where: L represents the Laplace matrix, I is the unit matrix, D is the degree matrix, is the eigenvector matrix.
[0097] (3) Predict the target variable based on the feature vector matrix based on graph convolution; the process includes:
[0098] The convolution kernel h and feature vector V of the graph convolution are Fourier transformed respectively, and then multiplied to obtain the new feature representation X t , the public notice is:
[0099]
[0100] U represents the Fourier transform of the graph, U t h、U t V represents the result of Fourier transform of convolution kernels h and V respectively;
[0101] For feature X t Perform graph convolution operation to obtain the predicted value of the target variable; the public expression is:
[0102]
[0103] In the formula, y t ' +1 represents the second predicted value, X t is the auxiliary variable input at time t, A represents the adjacency matrix, W t represents the GCN network parameters, and σ is the activation function.
[0104] This application uses Fourier transform to convert convolution operations into point multiplication operations, thereby simplifying the calculation process and improving calculation efficiency. At the same time, Fourier transform also helps to better understand the structure and characteristics of the graph, thereby improving the performance of the model.
[0105] This application can achieve the next moment target variable y by learning the graph structure and optimizing the network parameters. t ' +1 GCN soft-sensing spatial correlation prediction.
[0106] Through the first and second parts, a missing data prediction model for industrial soft sensing can be obtained, denoted as FGCDM.
[0107] In a preferred embodiment, a mild loss feedback mechanism is constructed for the FCDM and GCN modules to improve the accuracy of customized missing data generation based on downstream soft measurement tasks. Specifically, the total loss of the FGCDM prediction process is introduced into the GCN and FCDM modules respectively, so that the overall model can achieve milder data generation and soft measurement modeling. That is, the FCDM accepts feedback from the downstream GCN module and fine-tunes its own parameters to achieve data generation for specific soft measurement tasks. The GCN accepts information from the upstream FCDM and can achieve time-space related soft measurement predictions.
[0108] Among them, the predicted value of FCDM is x t ' +1 , prediction error
[0109] The predicted value of GCN is y t ' +1 , prediction error
[0110] Therefore, the loss of FGCDM is expressed as loss FGCDM , as shown below:
[0111]
[0112] When the loss FGCDM When introduced into GCN and FCDM respectively to participate in and guide the module optimization process, the loss improvements of the two modules are as follows:
[0113]
[0114]
[0115] In the formula, loss noise Represents the noise fitting error, loss FGCDM represents the total loss of the prediction process, x′ t+1 is the first predicted value, x t+1 Indicates the target value corresponding to the first predicted value, y′ t+1 represents the predicted value of the target variable, y t+1 Indicates the target value corresponding to the predicted value of the target variable.
[0116] The improved FCDM loss is composed of the noise fitting error loss noise , prediction bias And GCN's prediction error loss FGCDM The loss of GCN consists of three parts. And the prediction error loss of FCDM FGCDM It consists of two parts.
[0117] By introducing the downstream GCN prediction loss into the upstream FCDM, the FCDM can achieve the goal of generating missing data customized for downstream tasks. Introducing the time-dependent prediction loss of the FCDM into the GCN can also enable the GCN to take into account the time-space dependency of the data and achieve the purpose of soft-sensing prediction based on time-space correlation.
[0118] To further optimize the above technical solution, when the loss of FCDM and GCN is reduced to half of their initial loss values, loss FGCDM ;
[0119] In the actual industrial process, the sensor is an independent detection individual, and its detection results exist objectively and are not affected by the detection results of other sensors. If the prediction of the target variable is too dependent on the prediction results of other variables, it will not conform to the actual industrial process. Therefore, try to avoid the negative impact of the prediction results of downstream applications on the optimization of upstream models.
[0120] As when and , indicating that the generated data of FCDM has a smaller deviation from the original data, while the prediction error of GCN is larger. If the GCN prediction error is introduced into FCDM and guides its optimization process, the generated data will have a greater deviation, which will reduce the accuracy of missing data filling.
[0121] In the initial training phase, the two modules are set to be optimized separately instead of adding another loss directly to the model. Assume that the network is trained at time t0, t n After time t, the FCDM generation loss is That is reduced to the initial loss Similarly, GCN loss Also reduced to the initial loss When the above conditions are met at the same time, loss FGCDM They are introduced into FCDM and GCN respectively to avoid transferring large errors into GCN and FCDM modules, which will have a negative impact on module optimization and further improve the soft sensor prediction accuracy.
[0122] In another embodiment, the isolation forest algorithm (IIF) is preferentially used to detect the data to determine the location of missing data and the amount of missing data, and data is filled in for the missing location and the amount of missing data.
[0123] The following example verifies the data filling method of this application.
[0124] All data in the present invention come from sensor monitoring data of a nickel smelting plant in China and gas turbine sensor monitoring data of a power plant in northwestern Turkey in the UCI public database. By analyzing the nickel flash smelting process, four auxiliary variables with the highest correlation with furnace temperature, namely, concentrate humidity (CH), concentrate particle size (CPS), sulfur dioxide content (SDC) of sulfur-containing flue gas, and water content (WC) of sulfur-containing flue gas, are selected for soft measurement modeling of furnace temperature (T). Four variables, namely, ambient temperature (AT), atmospheric pressure (AP), relative humidity (RH), and exhaust pressure (V), are selected as auxiliary variables of soft measurement for soft measurement modeling of equipment full load output power (PE).
[0125] The variables in different datasets are shown in Table 1 ;
[0126] Table 1
[0127]
[0128]
[0129] The missing region information in different datasets is shown in Table 2;
[0130] Table 2
[0131]
[0132] The time consumption of different comparison methods on different data sets is shown in Table 3;
[0133] Table 3
[0134]
[0135] According to the calculation time of the four different methods in the two data sets,
[0136] FGCDM and its three extension methods all use moving windows to iteratively generate missing data. In FGCDM-2, the acceleration structure is cancelled, that is, this method uses a common DDPM for data generation, and a structure with a step span of 1 is used in the reverse process. It can be seen that FGCDM-2 takes the longest time to calculate and has the highest computational complexity.
[0137] FGCDM, FGCDM-1, and FGCDM-3 all use the acceleration structure proposed in the present invention, that is, the sampling strategy of random step span is adopted in the reverse process, which takes less time than FGCDM-2. Therefore, by recording the running time of each model, it is indirectly shown that the FGCDM method proposed in the present invention can effectively reduce the computational complexity of the model and reduce the computational time.
[0138] MSE and R of missing data generated by different methods 2As shown in Table 4;
[0139] Table 4
[0140]
[0141] MSE and R of missing data and original data generated by different methods 2 The error shows that even under different data sets, different missing locations, and different amounts of missing data, the missing data MSE and R of the three models FGCDM-1, FGCDM-2, and FGCDM are higher than those of the other six models. 2 The errors are lower.
[0142] This result shows that compared with the method of treating missing data generation and soft measurement prediction as two independent tasks, the missing data generated with customized downstream tasks has lower deviation and higher generation accuracy. Therefore, the data generation strategy for specific needs of downstream tasks in actual industrial processes proposed in the present invention is reasonable, effective and superior.
[0143] In addition, it can be seen from Table 4 that FGCDM and FGCDM-2 have lower data generation errors and higher generation accuracy than FGCDM-1. Therefore, the mild feedback strategy proposed in the present invention is more reasonable and superior.
[0144] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0145] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A missing data filling method for industrial soft measurement, It is characterized in that Determine an auxiliary variable of the target variable, generate a first prediction value according to the time correlation of existing data in the auxiliary variable, and fill in missing data in the auxiliary variable using the first prediction value; The target variable is predicted based on the spatial dependency between the filled auxiliary variables and the target variable.
2. A missing data filling method for industrial soft measurement according to claim 1, It is characterized in that The temporal correlation of data in auxiliary variables is extracted through Bi-LSTM.
3. A missing data filling method for industrial soft measurement according to claim 1, It is characterized in that Generating a first prediction value based on the time correlation of the existing data in the auxiliary variable includes: According to the distribution of existing data, the noise is gradually diffused forward; Based on the diffusion-noised data, random span sampling is adopted, and the first prediction value is generated by reverse denoising considering the time correlation.
4. A missing data filling method for industrial soft measurement according to claim 3, It is characterized in that The data distribution when diffused to the nth step is: In the formula, L represents the length of the existing data, t represents the data at the tth moment, and α is defined as t =1-β t , β t is the noise added at time t, is the variance, and I is the identity matrix.
5. A missing data filling method for industrial soft measurement according to claim 3, It is characterized in that The steps of adopting random span sampling and taking into account time correlation and performing inverse denoising to generate a first prediction value include: Considering the time correlation to fit the noise, From the current data Remove the fitting noise, sample with random span s, and get the data distribution of the nsth step; and so on, get the overall data distribution after random span sampling of the diffusion and noise data; The overall data distribution is input into the linear layer to obtain the first prediction value.
6. A missing data filling method for industrial soft measurement according to claim 1, It is characterized in that Based on the spatial dependency between the filled auxiliary variables and the target variable, the target variable is predicted; including: Construct a relationship topology graph based on auxiliary variables and target variables, and construct an adjacency matrix based on the correlation between different variables; The adjacency matrix is decomposed by symmetric normalized Laplace matrix to obtain the eigenvector matrix; The target variable is predicted based on the feature vector matrix based on graph convolution.
7. A missing data filling method for industrial soft measurement according to claim 6, It is characterized in that Based on the correlation between variables, the adjacency matrix is constructed according to the following formula; In the formula, e ij represents the correlation between variables, which is calculated by KL divergence; μ represents the correlation threshold, a ij ∈A represents an element in the adjacency matrix, and i and j represent the types of auxiliary variables.
8. A missing data filling method for industrial soft measurement according to claim 1, It is characterized in that When the loss of generating the first prediction value and the loss of predicting the target variable are both half of their initial losses, the total loss of the prediction process is introduced; The loss function for generating the first prediction is: The loss function for predicting the target variable is: In the formula, loss noise Represents the noise fitting error, loss FGCDM represents the total loss of the prediction process, x′ t+1 is the first predicted value, x t+1 Indicates the target value corresponding to the first predicted value, y′ t+1 represents the predicted value of the target variable, y t+1 Indicates the target value corresponding to the predicted value of the target variable.
9. A missing data filling method for industrial soft measurement according to claim 8, It is characterized in that The total loss of the prediction process FGCDM for:
10. A missing data filling method for industrial soft measurement according to claim 1, It is characterized in that The isolation forest algorithm is used to detect data, determine the location and amount of missing data, and fill in data according to the missing location and amount of missing data.
Citation Information
Patent Citations
A method for completing and predicting an attribute missing data set based on generative adversarial network
CN109165664A
Sampling-free method used in target detector model training process
CN111291796A
Traffic speed prediction method based on space-time diagram convolution-generative adversarial network
CN113450561A
Long-time-sequence traffic flow prediction method based on graph convolution-Informer model
CN113487061A
Science and technology paper reference relation representation learning method and system and storage medium
CN114817578A
