Semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene rectification

Through a combination of timing consistency and semi-supervised learning, the real-time accuracy of diethylbenzene concentration measurement during ethylbenzene distillation is solved, and the accurate prediction of diethylbenzene concentration is achieved, which is better than the accuracy and application of traditional methods. sex.

CN119007872AActive Publication Date: 2024-11-22ZHEJIANG UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410988531.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-22
Estimated Expiration
2044-07-23

AI Technical Summary

Technical Problem

The existing technology is difficult to accurately measure diethylbenzene concentration in real time during ethylbenzene distillation. The traditional measurement methods have problems such as expensive equipment, high maintenance costs and data lag. The timing prediction method based on machine learning cannot adapt to the industry with dual-rate sampling. process.

Method used

Combining timing consistency and semi-supervised learning, by obtaining labeled and unlabeled sample data, calculating mixed mutual information for feature selection, building a dual-channel prediction network, using gradient descent method to train the network, extract trend terms and fluctuation terms features, and design timing Consistency loss function, realizes accurate prediction of diethylbenzene concentration.

Benefits of technology

By reducing the data dimension, improving prediction accuracy and reducing the computational complexity, accurate and real-time prediction of diethylbenzene concentration is achieved, which is better than the accuracy and applicability of traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007872A_ABST
    Figure CN119007872A_ABST
Patent Text Reader

Abstract

The invention discloses a semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene rectification, which comprises the following steps: acquiring ethylbenzene distillation process variables in labeled and unlabeled ethylbenzene towers, and calculating mixed mutual information for feature selection to obtain auxiliary variables of soft measurement; performing data filling and time sequence decomposition on each auxiliary variable to obtain a fluctuation item and a trend item of the auxiliary variable; a dual-channel prediction network is constructed, the dual-channel prediction network comprises a fluctuation item prediction channel and a trend item prediction channel, the two channels are finally subjected to regression prediction through a full connection layer, and output of the two channels is superposed to obtain a predicted value of the diethylbenzene concentration; network training is carried out, and semi-supervised loss functions are constructed for samples with labels and samples without labels; and inputting the auxiliary variables after time sequence decomposition into the trained network to obtain a soft measurement result of the diethylbenzene concentration. By applying the method, accurate soft measurement of the diethylbenzene concentration in the ethylbenzene rectification process can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to soft measurement of chemical process, and in particular to a semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation. Background Art

[0002] Ethylbenzene is an important petrochemical raw material. Using benzene and ethylene as raw materials, the production of ethylene through alkylation reaction is a widely used method in the industry, but it will also produce byproducts such as diethylbenzene. The concentration of diethylbenzene in ethylbenzene not only affects the quality of the product, but is also a key indicator for the optimization of production process control. Therefore, accurate measurement of diethylbenzene concentration is crucial to the ethylbenzene production process. In order to measure the concentration of diethylbenzene in the product, most factories use online analyzers or laboratory sampling and analysis. Online analyzers are expensive, and their accuracy will decrease with time. Although laboratory analysis has high accuracy, the analysis cycle is long, usually in days, and it is impossible to guide production in real time.

[0003] In modern industrial processes, real-time monitoring of key quality variables plays an important role in maintaining the safety and effective control and optimization of production processes. However, due to poor measurement environment, expensive maintenance costs of measurement equipment and time lag of laboratory test data, traditional measurement methods can no longer meet the needs of modern industry. Therefore, in order to make up for the shortcomings of traditional measurement methods, soft measurement came into being. Soft measurement uses easy-to-measure process variables to indirectly estimate key quality variables that are not easy to measure.

[0004] In real industrial production, there are a large number of process variables, such as temperature, pressure, and flow, which act together on production and affect quality variables such as concentration and acidity. However, due to the characteristics of measurement technology and industrial processes, it is impossible to ensure that all variables maintain the same sampling frequency. For example, in the ethylbenzene distillation process, variables such as flow need to be controlled in seconds, while the concentration of diethylbenzene changes slowly, and the sampling time interval is in minutes or even hours. This sampling process at different rates brings challenges to the design of industrial soft sensors, which is characterized by a large number of missing values ​​in the collected data, especially the lack of quality variables, that is, unlabeled samples. In recent years, although time series prediction methods based on machine learning and deep learning are very popular, most methods assume that all variables are sampled at the same rate, that is, the data set is complete. This is obviously inconsistent with the actual industrial situation, making these methods inapplicable to industrial processes with dual-rate sampling.

[0005] Semi-supervised learning is a method in machine learning that can add a large number of unlabeled samples to a limited number of labeled samples for training and learning, thereby improving learning performance. Based on the above analysis, in order to achieve low-cost and accurate measurement of diethylbenzene concentration, it is necessary to establish a semi-supervised soft measurement method. Summary of the invention

[0006] The purpose of the present invention is to provide a semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation in view of the deficiencies in the prior art. The present invention combines "temporal consistency" with semi-supervised learning, and simultaneously extracts the characteristics and potential associations of labeled and unlabeled sample data in the ethylbenzene distillation process, thereby establishing an accurate diethylbenzene soft measurement prediction model.

[0007] The objective of the present invention is achieved through the following technical scheme: a semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation, the method comprising:

[0008] S1. Obtain the process variables of ethylbenzene distillation in the ethylbenzene tower with and without labels, calculate the mixed mutual information for feature selection, and obtain the auxiliary variables of soft measurement;

[0009] S2. Perform data filling and time series decomposition on each auxiliary variable to obtain the fluctuation term and trend term of the auxiliary variable;

[0010] S3. Construct a dual-channel prediction network, including a fluctuation item prediction channel and a trend item prediction channel. The two channels are respectively subjected to regression prediction using a fully connected layer, and the outputs of the two channels are superimposed to obtain the predicted value of diethylbenzene concentration.

[0011] S4. Use the gradient descent method to train the network, construct a semi-supervised loss function for labeled and unlabeled samples, and obtain the optimal parameters after training;

[0012] S5. Input the auxiliary variables after time series decomposition into the trained network to obtain the soft measurement results of diethylbenzene concentration.

[0013] Furthermore, the process variables include: reflux flow, heating steam flow, upper temperature of the distillation tower, feed tower plate temperature, middle temperature of the distillation tower, distillation tower kettle temperature, top outlet temperature, reflux tank outlet temperature, benzene tower kettle liquid level, feed flow, sensitive plate temperature, heating steam condensate tank level, tower kettle liquid level, tower kettle production flow, top heat exchanger level, reflux tank level, product production flow and top outlet pressure.

[0014] Furthermore, the feature selection process is specifically as follows: calculating the mutual information of all process variables and diethylbenzene concentration to obtain the correlation coefficient MI, and multiplying the production experience matrix with the mutual information correlation coefficient matrix of all process variables to calculate the hybrid mutual information HMI:

[0015] Rank all mixed mutual information values ​​from large to small and take the top k variables.

[0016] Furthermore, the data filling is to fill the front and back ends of the time series respectively.

[0017] Furthermore, the process of time series decomposition is specifically as follows:

[0018] The time series after data filling of each auxiliary variable is processed by moving average to obtain the medium- and long-term trends in the signal. The trend term is then subtracted from the original time series to obtain the time series of the fluctuation term.

[0019] Furthermore, in the dual-channel prediction network, the prediction channel of the fluctuation term uses a sequentially stacked LSTM network, a fully connected layer, and a ReLU function to extract features;

[0020] The prediction channel of the trend item uses the fully connected layer, Dropout layer and ReLU activation function to extract features in sequence.

[0021] Furthermore, the gradient descent method is used in the network training to solve the optimal model parameters, and the loss function is:

[0022] Loss=Loss l +αLoss u

[0023] Where α is a hyperparameter used to balance the two loss functions;

[0024] Loss l is the loss function of the labeled sample, specifically, the loss function is expressed in the second-order form of the deviation between the predicted output and the true value:

[0025]

[0026] where y i is the label output by the model, Output values ​​for the model;

[0027] Loss l is the loss function of the unlabeled sample, and the time-consistent regularization is used to obtain:

[0028]

[0029] The first difference represents the deviation between the output estimate and the nearest labeled sample, and ΔY is the threshold hyperparameter.

[0030] According to another aspect of the specification, a semi-supervised soft measurement device for diethylbenzene concentration in ethylbenzene distillation is also provided, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, the semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation is implemented.

[0031] According to another aspect of the specification, a computer-readable storage medium is provided, on which a program is stored. When the program is executed by a processor, the semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation is implemented.

[0032] According to another aspect of the specification, a computer program product is also provided, including a computer program, wherein when the computer program is executed by a processor, the semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation is implemented.

[0033] The beneficial effect of the present invention is that auxiliary variables are screened out from process variables by hybrid mutual information, irrelevant variables are proposed and the dimension of data is reduced, thereby improving prediction accuracy and reducing computational complexity. Through time series decomposition and dual-channel prediction network, the characteristics of trend items and fluctuation items are extracted respectively, and a loss function considering time series consistency is designed. At the same time, the information of labeled samples and unlabeled samples is used, and the relationship between the two types of data is explored to guide network training, and finally accurate prediction of diethylbenzene concentration is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is an overall flow chart of a soft measurement algorithm provided by an embodiment of the present invention;

[0035] Figure 2 is a schematic diagram of timing consistency provided by an embodiment of the present invention;

[0036] Figure 3 1 is a diagram of diethylbenzene prediction results of different models provided in an embodiment of the present invention;

[0037] Figure 4 is a performance comparison chart of feature selection algorithms provided by an embodiment of the present invention;

[0038] Figure 5 It is a schematic diagram of a semi-supervised soft measurement device for diethylbenzene concentration in ethylbenzene distillation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention proposes a semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation. The present invention is further described in detail below in conjunction with the accompanying drawings and the drawings in the specific embodiments:

[0040] In this implementation example, actual plant operation data of an annual production of 500,000 tons of ethylbenzene is used, such as Figure 1 As shown in the figure, the task of soft sensing is to use the process variables in the distillation tower to establish a deep learning prediction model to predict the concentration of diethylbenzene.

[0041] Step 1, feature selection;

[0042] Obtain the reflux flow, heating steam flow, distillation tower upper temperature (21st tower plate temperature), feed tower plate temperature (43rd tower plate temperature), distillation tower middle temperature (48th tower plate temperature), distillation tower bottom temperature (53rd tower plate temperature), tower top outlet temperature, reflux tank outlet temperature, benzene tower bottom liquid level, feed flow, sensitive plate temperature, heating steam condensate tank level, tower bottom liquid level, tower bottom production flow, tower top heat exchanger level, reflux tank level, product production flow, tower top outlet pressure from the ethylbenzene distillation unit, calculate the mutual information of all process variables and diethylbenzene concentration, obtain the correlation coefficient MI, and multiply the production experience matrix by the mutual information correlation coefficient matrix to calculate the hybrid mutual information HMI:

[0043]

[0044] Where p(x, y) is the joint distribution of each process variable x and diethylbenzene concentration y, p(x) and p(y) are the marginal distributions of variable x and variable y, respectively. A is a production experience matrix. According to the historical experience provided by process engineers and production operators, weighted parameters are set for the correlation coefficients between each process variable and diethylbenzene concentration. The experience matrix in the present invention is:

[0045] A=[0.9,0.8,0.9,1,1,0.7,0.9,1,1,0.8,1,0.8,0.8,0.9,1,0.9,0.9,0.9]

[0046] According to the HMI value of each process variable and diethylbenzene concentration, sort them from high to low and select them as auxiliary process variables for soft measurement:

[0047] x1,…,x k =argTopk(HMI(X,Y))

[0048] argTopk means ranking all mixed mutual information values ​​from large to small and taking the top k variables. Using feature selection, 9 variables are selected as auxiliary variables for diethylbenzene concentration prediction, namely, reflux flow, heating steam flow, reflux tank outlet temperature, feed flow, sensitive plate temperature, tower bottom liquid level, tower bottom extraction flow, product extraction flow and tower top pressure.

[0049] Step 2, timing decomposition:

[0050] Perform moving average processing on the 9 auxiliary variables to obtain the medium- and long-term trends in the signal, and then subtract the trend term from the original time series to obtain the time series of the fluctuation term. Use historical data with a length of 120 seconds for prediction. The specific decomposition method is as follows:

[0051] X t =AvgPool(Padding(X))

[0052] X f =XX t

[0053] in They represent the fluctuation term and trend term respectively obtained by decomposing the time series. In order to ensure that the time series obtained after decomposition has the same length as the original data, the data padding method Padding(·) is first used to fill the front and back ends of the time series, and then AvgPool(·) is used for moving average processing.

[0054] Step 3: Build a dual-channel prediction network:

[0055] The prediction channel of the fluctuation term uses a stacked 2-layer 256-dimensional LSTM network, a 256-dimensional fully connected layer, and a ReLU function to extract features. The prediction channel of the trend term uses a 256-dimensional fully connected layer, a Dropout layer, and a ReLU activation function to extract features. At the end of the two channels, a 256-dimensional fully connected layer is used for regression prediction, and the outputs of the two channels are superimposed to obtain the predicted value of diethylbenzene concentration.

[0056] Step 4, design the loss function:

[0057] use Represents the mapping relationship, where y is the label output by the model, that is, the true value of diethylbenzene concentration, θ t and θ f Represent the model parameters of trend term and fluctuation term respectively.

[0058] For labeled samples, since the true value of the output is known, the loss function can be expressed as a second-order norm of the deviation between the predicted output and the true value, as shown below:

[0059]

[0060] Where N l Represents the number of labeled samples.

[0061] For unlabeled samples, due to the lack of actual output values, it is impossible to directly calculate the difference between the predicted results and the actual results. Considering the characteristics of continuous production in process industries, variables generally have temporal continuity and rarely have large jumps, such as Figure 2 As shown in the figure, the loss function of the unlabeled sample is calculated based on the actual value of the adjacent labeled sample as a reference, which is called time-consistent regularization. The calculation method is as follows:

[0062]

[0063] The first difference represents the deviation between the output estimate and the nearest labeled sample, and ΔY is the threshold hyperparameter, which is designed differently according to the variation range of industrial data.

[0064] Combining the loss functions of the above labeled samples and unlabeled samples, the overall loss function is as follows:

[0065] Loss=Loss l +αLoss u

[0066] Among them, α is a hyperparameter used to balance the two loss functions. In the present invention, it is taken as 0.5. Combined with the two regularization terms, it can simultaneously utilize the information of labeled and unlabeled samples, and also combined with the actual characteristics of the industrial process, the temporal consistency of the process variables is taken into consideration.

[0067] Step 5, model offline training phase, first collect process variables and quality variables from the actual industrial site, obtain the soft measurement data set after feature selection, and divide 70% of the data set into training data X train and Y train Set the historical data window length and the prediction time window length to 120, and create a training data block. Through the time series decomposition module, the process variable is decomposed into trend items X t,train and the fluctuation term X f,train . Establish the overall network model structure and randomly initialize the network model parameters θ={θ t ,θ f}, respectively representing the parameters of the prediction network of the trend branch and the fluctuation branch. Use the loss function defined in step 4 to calculate the Loss, and use the gradient descent method to solve the optimal model parameters:

[0068]

[0069] In the online application stage, only two prediction channels are called for forward propagation to give predictions, as shown in the following formula:

[0070]

[0071] Where X test Contains nine auxiliary variables in the test set, is the predicted value of diethylbenzene concentration for the test data.

[0072] The following experiments illustrate that the semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation of the present invention is effective and feasible, and has certain advantages:

[0073] Based on the industrial example of ethylbenzene distillation, the test set data is input into the soft sensor model established by the present invention, and the prediction results of diethylbenzene concentration obtained by other soft sensor methods and the method of the present invention are compared to analyze the effectiveness and superiority of the modeling method of the present invention. The specific steps are as follows:

[0074] The serialized ethylbenzene distillation process data is divided into a training set and a test set. The data set contains the operation data of the ethylbenzene tower for ten days. The sampling interval of process variables such as flow is 30 seconds, with a total of 28,800 samples, of which 70% are used for model training and 30% are used for testing. The quality variable is the diethylbenzene concentration in the product, and the measurement cycle is 30 minutes. Since the sampling rate of the quality variable is much smaller than that of the process variable, there are only 336 and 144 labeled samples in the training set and test set. Complete the model training according to steps 1 to 5.

[0075] The validity and superiority of the prediction ability of the soft sensor model were tested using the test set, and 8 soft sensor models were established for comparative test experiments. These 8 soft sensor models include CoReg based on manifold regularization regression algorithm and ICTREG based on consistency regularization, U-FCN and U-LSTM that combine upsampling technology with deep learning, where upsampling uses cubic spline difference to fill unlabeled samples in the original data, deep learning methods D-FCN and D-LSTM combined with downsampling technology, L-FCN using data lifting techniques, and the method of the present invention, temporal consistency semi-supervised network (TCS-Net). The input features of all methods are the 9 variables screened in step 1, and the mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation indicators for the prediction performance of dynamic soft sensor models.

[0076] Table 1 shows the test results of comparative experiments using 8 soft sensor models. From the data in the table, it can be concluded that the prediction errors of the proposed method under both evaluation indicators are smaller than those of other methods. Figure 3 The prediction results of 8 soft sensing methods are compared with the actual value of diethylbenzene. Compared with other methods, the prediction accuracy of the method of the present invention is better than that of other methods, especially when the diethylbenzene concentration fluctuates rapidly near the 10th and 70th sampling points. The results show that after introducing dual-channel prediction and combining the loss function of temporal consistency, the proposed semi-supervised soft sensing model is superior to other methods.

[0077] In order to further compare the effectiveness of the feature selection method using hybrid mutual information (HMI) in the present invention, the prediction results after prediction using all process variables and feature screening using random forest (RF) and mutual information (MI) are compared using the TCS-Net model as a benchmark. Figure 4 As shown, after using hybrid mutual information to filter features, the prediction accuracy of the model can be improved, which proves the effectiveness of the hybrid mutual information feature screening method in the invention.

[0078] Table 1 shows the average error of 10 independent experiments for all soft sensor models.

[0079]

[0080] Table 2 shows the prediction results using different feature screening methods

[0081]

[0082] The present invention utilizes easily measurable variables such as flow rate and temperature in the ethylbenzene distillation process, selects auxiliary variables from process variables through mixed mutual information, proposes irrelevant variables and reduces the dimension of data, thereby improving prediction accuracy and reducing computational complexity. Through time series decomposition and dual-channel prediction network, the characteristics of trend items and fluctuation items are extracted respectively, and a loss function considering time series consistency is designed. At the same time, the information of labeled samples and unlabeled samples is utilized, and the relationship between the two types of data is explored to guide network training, and finally accurate prediction of diethylbenzene concentration is achieved.

[0083] Corresponding to the above-mentioned embodiment of a semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation, the present invention also provides an embodiment of a semi-supervised soft measurement device for diethylbenzene concentration in ethylbenzene distillation.

[0084] See also Figure 5 A semi-supervised soft measurement device for diethylbenzene concentration in ethylbenzene distillation provided in an embodiment of the present invention comprises a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement a semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation in the above embodiment.

[0085] An embodiment of a semi-supervised soft-sensing device for diethylbenzene concentration in ethylbenzene distillation provided by the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, the processor of any device with data processing capability reads the corresponding computer program instructions in the non-volatile memory into the internal memory and runs them. From the hardware level, if Figure 5 As shown, it is a hardware structure diagram of any device with data processing capability where a semi-supervised soft measurement device for diethylbenzene concentration in ethylbenzene distillation provided by the present invention is located. Figure 5 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0086] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0087] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein 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 may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0088] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation in the above embodiment is implemented.

[0089] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0090] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the semi-supervised soft measurement method for diethylbenzene concentration in ethylbenzene distillation is implemented.

[0091] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0092] It should be understood that the above general description and the detailed description below are only exemplary and explanatory and cannot limit the present application. The present application is not limited to the precise structure described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present application is limited only by the attached claims.

Claims

1. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation. It is characterized in that The method includes: S1. Obtain the process variables of ethylbenzene distillation in the ethylbenzene tower with and without labels, calculate the mixed mutual information for feature selection, and obtain the auxiliary variables of soft measurement; S2. Perform data filling and time series decomposition on each auxiliary variable to obtain the fluctuation term and trend term of the auxiliary variable; S3. Construct a dual-channel prediction network, including a fluctuation item prediction channel and a trend item prediction channel. The two channels are respectively subjected to regression prediction using a fully connected layer, and the outputs of the two channels are superimposed to obtain the predicted value of diethylbenzene concentration. S4. Use the gradient descent method to train the network, construct a semi-supervised loss function for labeled and unlabeled samples, and obtain the optimal parameters after training; S5. Input the auxiliary variables after time series decomposition into the trained network to obtain the soft measurement results of diethylbenzene concentration.

2. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to claim 1, It is characterized in that The process variables include: reflux flow, heating steam flow, upper temperature of the distillation tower, feed tower plate temperature, middle temperature of the distillation tower, distillation tower kettle temperature, top outlet temperature, reflux tank outlet temperature, benzene tower kettle liquid level, feed flow, sensitive plate temperature, heating steam condensate tank level, tower kettle liquid level, tower kettle production flow, tower top heat exchanger level, reflux tank level, product production flow and tower top outlet pressure.

3. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to claim 1, It is characterized in that The feature selection process is specifically as follows: the mutual information of all process variables and diethylbenzene concentration is calculated to obtain the correlation coefficient MI, and the production experience matrix is ​​multiplied by the mutual information correlation coefficient matrix of all process variables to calculate the hybrid mutual information HMI: Rank all mixed mutual information values ​​from large to small and take the top k variables.

4. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to claim 1, It is characterized in that The data filling is to fill the front and back ends of the time series respectively.

5. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to claim 1, It is characterized in that The process of timing decomposition is specifically as follows: The time series after data filling of each auxiliary variable is processed by moving average to obtain the medium- and long-term trends in the signal. The trend term is then subtracted from the original time series to obtain the time series of the fluctuation term.

6. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to claim 1, It is characterized in that In the dual-channel prediction network, the prediction channel of the fluctuation term uses a sequentially stacked LSTM network, a fully connected layer, and a ReLU function to extract features; The prediction channel of the trend item uses the fully connected layer, Dropout layer and ReLU activation function to extract features in sequence.

7. A semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to claim 1, It is characterized in that The gradient descent method is used in the network training to solve the optimal model parameters, and the loss function is: Loss=Loss l +αLoss u Where α is a hyperparameter used to balance the two loss functions; Loss l is the loss function of the labeled sample, specifically, the loss function is expressed in the second-order form of the deviation between the predicted output and the true value: where y i is the label output by the model, Output values ​​for the model; Loss l is the loss function of the unlabeled sample, and the time-consistent regularization is used to obtain: The first difference represents the deviation between the output estimate and the nearest labeled sample, and ΔY is the threshold hyperparameter.

8. A semi-supervised soft-sensing device for diethylbenzene concentration in ethylbenzene distillation, comprising a memory and one or more processors, wherein the memory stores executable code, It is characterized in that When the processor executes the executable code, a semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a program stored thereon, It is characterized in that When the program is executed by a processor, a semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, a semi-supervised soft sensing method for diethylbenzene concentration in ethylbenzene distillation according to any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Soft measurement modeling method of semi-supervised dynamic soft measurement network

    CN112001115A

  • Semi-supervised soft measurement modeling method and device and electronic equipment

    CN117272823A

  • Soft sensing of a nonlinear and multimode processes based on semi-supervised weighted Gaussian regression

    US10678196B1