Purification industrial control system for carbon dioxide-based polyols

Abstract

The invention relates to the technical field of polymer purification, and particularly discloses a purification industrial control system for carbon dioxide-based polyols, in the removal process of phenolic compounds in the carbon dioxide-based polyols, the removal rates of the phenolic compounds are detected at two different moments respectively, and the removal rate of the phenolic compounds in the carbon dioxide-based polyols is calculated. Meanwhile, a data processing technology based on deep learning is used for conducting fine-grained local time domain interaction analysis on the stirring speed and the stirring temperature between the two time points to reveal the combined effect of the stirring speed and the stirring temperature on the removal efficiency of the phenolic compounds, removal measurement is conducted based on the combined effect, an effective removal measurement coefficient is generated, and the removal efficiency of the phenolic compounds is improved. And intelligently estimating the still required stirring duration by combining the removal rate change between the two time points. Through the mode, the reaction endpoint can be predicted in advance, and incomplete reaction or excessive reaction can be effectively avoided, so that the purification efficiency and the product quality are improved, and the automation level of industrial production is improved.

Description

Technical Field

[0001] The present application relates to the technical field of polymer purification, and more specifically, to an industrial control system for purifying carbon dioxide-based polyols. Background Art

[0002] Carbon dioxide-based polyols, which are polymerized by carbon dioxide, propylene oxide and phenolic compound initiators under the action of bimetallic catalysts, are one of the main raw materials for synthesizing polyurethane resins and are widely used in polyurethane foam plastics, polyurethane coatings, polyurethane adhesives, polyurethane elastomers and other fields.

[0003] However, in the production process of CO2-based polyols, phenolic compounds are often inevitably introduced as by-products or raw material residues. These phenolic compounds not only affect the purity of the product, but may also have adverse effects on product performance, such as reducing the heat resistance, chemical corrosion resistance and biocompatibility of the material. Therefore, the development of efficient and environmentally friendly methods for removing phenolic compounds has become an important research direction in this field.

[0004] Traditional methods for removing phenolic compounds include physical separation (such as distillation and extraction), chemical conversion (such as oxidation and esterification), and adsorption treatment. Among them, the adsorption method is generally considered to be superior to other methods because of its simple operation, low cost, rapid and effective removal of pollutants, and low risk of secondary pollution. In the adsorption method, factors such as the selection of adsorbent types and the control of adsorption process parameters have a decisive influence on the final removal effect. However, the current control of process parameters for the purification process of carbon dioxide-based polyols mainly relies on empirical judgment, lacks precise quantitative analysis and real-time monitoring, and may result in poor stability and repeatability of the removal process.

[0005] Therefore, an industrial control system for the purification of carbon dioxide-based polyols is desired. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a purification industrial control system for carbon dioxide-based polyols, which detects the removal rate of phenolic compounds at two different times in the removal process of phenolic compounds in carbon dioxide-based polyols, and uses a data processing technology based on deep learning to perform a fine-grained local time domain interactive analysis of the stirring speed and stirring temperature between the two time points to reveal the combined effect of the stirring speed and stirring temperature on the removal efficiency of phenolic compounds, and based on this, remove the measurement, generate an effective removal measurement coefficient, and then combine the change in the removal rate between the two time points to intelligently estimate the stirring time still required. In this way, the reaction endpoint can be predicted in advance, effectively avoiding incomplete reaction or overreaction, thereby improving purification efficiency and product quality, and improving the automation level of industrial production.

[0007] Accordingly, according to one aspect of the present application, there is provided an industrial control system for purifying carbon dioxide-based polyols, comprising: A removal rate data receiving module, used to receive the removal rates of phenolic compounds in the carbon dioxide-based polyol at a first time point and a second time point to obtain a first removal rate and a second removal rate; A stirring state parameter receiving module, used for receiving a time queue of stirring speed values ​​and a time queue of stirring temperature values ​​collected by the sensor group between a first time point and a second time point; A local time domain feature extraction module, used for performing local time domain feature extraction on the time queue of the stirring speed value and the time queue of the stirring temperature value respectively to obtain a sequence of local time series associated feature vectors of the stirring temperature and a sequence of local time series associated feature vectors of the stirring speed; A multi-modal stirring state parameter interactive analysis module, used for interactively coupling each group of corresponding stirring temperature local time series correlation feature vectors and stirring speed local time series correlation feature vectors in the sequence of stirring temperature local time series correlation feature vectors and the sequence of stirring speed local time series correlation feature vectors to obtain a sequence of stirring state multi-modal local time series coupling feature vectors; An effective removal metric module, used to generate an effective removal metric coefficient based on the global time-domain aggregation characteristics of the sequence of multi-modal local temporal coupling feature vectors of the stirring state; The module for estimating the remaining stirring time is used to generate an estimated remaining stirring time based on the first removal rate, the second removal rate and the effective removal metric coefficient.

[0008] In the above-mentioned industrial control system for the purification of carbon dioxide-based polyols, the multimodal stirring state parameter interactive analysis module includes: a principal component analysis unit, which is used to perform principal component analysis on the stirring temperature local time series association feature vector and the stirring speed local time series association feature vector respectively to obtain a set of stirring temperature local time series principal component feature components and a set of stirring speed local time series principal component feature components; an optimal pairing interactive coupling unit, which is used to perform optimal pairing multi-scale interactive coupling on the set of stirring temperature local time series principal component feature components and the set of stirring speed local time series principal component feature components to obtain the stirring state multimodal local time series coupling feature vector.

[0009] In the above-mentioned industrial control system for purification of carbon dioxide-based polyols, the local time domain feature extraction module is used to: input the time queue of the stirring speed value and the time queue of the stirring temperature value into the parameter local time series feature extractor based on the 1D-CNN model respectively to obtain the sequence of the stirring temperature local time series associated feature vectors and the sequence of the stirring speed local time series associated feature vectors.

[0010] In the above-mentioned industrial control system for purifying carbon dioxide-based polyols, the optimal pairing interactive coupling unit includes: an optimal matching screening subunit, which is used to use each stirring temperature local time series principal component characteristic component in the set of stirring temperature local time series principal component characteristic components as a query vector, and use the set of stirring speed local time series principal component characteristic components as a query library, and match the stirring speed local time series principal component characteristic components that best match each query vector from the query library to obtain a set of optimal matching pairs of stirring temperature local time series principal component characteristic components and stirring speed local time series principal component characteristic components; a multi-scale A degree interactive coupling subunit is used to input the best matching pairs of the local time series principal component characteristic components of the stirring temperature and the local time series principal component characteristic components of the stirring speed into the multi-scale interactive response coupling module to obtain a set of interactive coupling representation vectors of the best matching pairs of the local time series principal component characteristic components of the stirring temperature and the stirring speed; a feature cascade subunit is used to cascade the set of interactive coupling representation vectors of the best matching pairs of the local time series characteristics of the stirring temperature and the speed to obtain the multi-modal local time series coupling feature vector of the stirring state.

[0011] In the above-mentioned industrial control system for the purification of carbon dioxide-based polyols, the best matching screening subunit is used to: calculate the Mahalanobis distance between the query vector and each stirring speed local time series principal component characteristic component in the query library to obtain a set of stirring temperature-speed matching difference coefficients; select the stirring speed local time series principal component characteristic component corresponding to the minimum value in the set of stirring temperature-speed matching difference coefficients as the optimal matching result of the query vector to obtain the best matching pair of the stirring temperature local time series principal component characteristic component and the stirring speed local time series principal component characteristic component.

[0012] In the above-mentioned industrial control system for purifying carbon dioxide-based polyols, the multi-scale interactive coupling subunit is used to: respectively calculate the best matching pairs of the local time series principal component characteristic components of the stirring temperature and the local time series principal component characteristic components of the stirring speed by adding, subtracting and multiplying by positions to obtain the first stirring temperature-speed principal component best matching interactive representation vector, the second stirring temperature-speed principal component best matching interactive representation vector and the third stirring temperature-speed principal component best matching interactive representation vector; and calculate the first stirring temperature-speed principal component best matching interactive representation vector, the second stirring temperature-speed principal component best matching interactive representation vector and the third stirring temperature-speed principal component best matching interactive representation vector. The second stirring temperature-speed principal component best matching interaction representation vector and the third stirring temperature-speed principal component best matching interaction representation vector are cascaded and fused, and then input into a one-dimensional convolutional layer including a maximum pooling layer to obtain a stirring temperature-speed principal component best matching multi-scale interaction representation vector; the maximum value among the second norm of the stirring temperature local time series principal component feature component, the second norm of the stirring speed local time series principal component feature component and a preset scale adjustment parameter is selected as a scale scaling factor, and each eigenvalue in the stirring temperature-speed principal component best matching multi-scale interaction representation vector is divided by the scale scaling factor to obtain the stirring temperature-speed local time series feature best matching pair interaction coupling representation vector.

[0013] In the above-mentioned industrial control system for the purification of carbon dioxide-based polyols, the effective removal measurement module includes: a global propagation aggregation unit, which is used to perform characteristic timing dynamic propagation on the sequence of the stirring state multimodal local timing coupling characteristic vectors to obtain the stirring state timing action aggregation representation vector; a removal measurement unit, which is used to input the stirring state timing action aggregation representation vector into the decoder-based removal measurement module to obtain the effective removal measurement coefficient.

[0014] In the above-mentioned purification industrial control system for carbon dioxide-based polyols, the global propagation aggregation unit includes: a current stirring state feature principal component extraction subunit, which is used to extract the current stirring state multimodal local temporal coupling feature vector from the sequence of the stirring state multimodal local temporal coupling feature vectors, and perform principal component extraction based on eigenvalues ​​to obtain the current stirring state multimodal local temporal feature principal component representation vector; a historical stirring state feature principal component extraction subunit, which is used to define other stirring state multimodal local temporal coupling feature vectors in the sequence of the stirring state multimodal local temporal coupling feature vectors as historical stirring state multimodal local temporal coupling feature vectors to obtain a sequence of historical stirring state multimodal local temporal coupling feature vectors, and perform principal component extraction based on eigenvalues ​​on each of the historical stirring state multimodal local temporal coupling feature vectors to obtain a sequence of historical stirring state multimodal local temporal coupling feature vectors; a temporal gradient attenuation weight calculation subunit, which is used to extract the current stirring state multimodal local temporal coupling feature vector from the sequence of the stirring state multimodal local temporal coupling feature vectors. The time span of each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of historical stirring state multimodal local temporal feature principal component representation vectors relative to the current stirring state multimodal local temporal feature principal component representation vector is calculated to obtain a sequence of historical stirring state local feature temporal gradient attenuation weights; a historical information dynamic propagation aggregation subunit is used to calculate the weighted sum of the sequence of historical stirring state multimodal local temporal feature principal component representation vectors based on the sequence of historical stirring state local feature temporal gradient attenuation weights to obtain a historical stirring state multimodal local feature principal component significant transfer aggregation representation vector; a principal component aggregation subunit is used to calculate the positional sum of the historical stirring state multimodal local feature principal component significant transfer aggregation representation vector and the current stirring state multimodal local temporal feature principal component representation vector to obtain the stirring state temporal action aggregation representation vector.

[0015] In the above-mentioned industrial control system for purifying carbon dioxide-based polyols, the temporal gradient attenuation weight calculation subunit includes: a temporal propagation absolute attenuation factor calculation secondary subunit, which is used to calculate the temporal propagation absolute attenuation factor of each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of the historical stirring state multimodal local temporal feature principal component representation vector relative to the current stirring state multimodal local temporal feature principal component representation vector to obtain a sequence of temporal propagation absolute attenuation factors; a time span modulation secondary subunit, which is used to modulate the temporal dimension of each temporal propagation absolute attenuation factor in the sequence of the temporal propagation absolute attenuation factor based on the time span between each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of the historical stirring state multimodal local temporal feature principal component representation vector and the current stirring state multimodal local temporal feature principal component representation vector to obtain a sequence of time span modulated propagation attenuation factors; a weight screening secondary subunit, which is used to input the sequence of time span modulated propagation attenuation factors into the information transfer screening module based on the gating function to obtain a sequence of temporal gradient attenuation weights of the historical stirring state local feature.

[0016] In the above-mentioned industrial control system for the purification of carbon dioxide-based polyols, the time span modulation secondary subunit is used to: calculate the timestamp difference between the historical stirring state multimodal local time series feature principal component representation vector and the current stirring state multimodal local time series feature principal component representation vector, round the timestamp difference down to obtain the time span factor; divide the time propagation absolute attenuation factor of the historical stirring state multimodal local time series feature principal component representation vector by the natural exponential function value of the time span factor to obtain the time span modulation propagation attenuation factor.

[0017] In the above-mentioned industrial control system for purification of carbon dioxide-based polyols, the remaining stirring time estimation module is used to: process the first removal rate, the second removal rate and the effective removal measurement coefficient according to the following estimated remaining stirring time calculation formula to obtain the estimated remaining stirring time, wherein the estimated remaining stirring time calculation formula is:

[0018] Compared with the prior art, the purification industrial control system for carbon dioxide-based polyols provided by the present application detects the removal rate of phenolic compounds at two different times during the removal process of phenolic compounds in carbon dioxide-based polyols, and uses data processing technology based on deep learning to perform fine-grained local time-domain interactive analysis of the stirring speed and stirring temperature between the two time points to reveal the combined effect of the stirring speed and stirring temperature on the removal efficiency of phenolic compounds, and based on this, remove the measurement, generate an effective removal measurement coefficient, and then combine the change in the removal rate between the two time points to intelligently estimate the stirring time still required. In this way, the reaction endpoint can be predicted in advance, effectively avoiding incomplete reaction or overreaction, thereby improving purification efficiency and product quality, and improving the level of automation in industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 The block diagram is a control system for purifying carbon dioxide-based polyols according to an embodiment of the present application.

[0021] Figure 2 The figure is a data flow diagram of an industrial control system for purifying carbon dioxide-based polyols according to an embodiment of the present application.

[0022] Figure 3 It is a block diagram of a multi-modal stirring state parameter interactive analysis module in an industrial control system for purification of carbon dioxide-based polyols according to an embodiment of the present application.

[0023] Figure 4 It is a block diagram of an effective removal measurement module in an industrial control system for purifying carbon dioxide-based polyols according to an embodiment of the present application.

[0024] Figure 5 The block diagram is a global propagation polymerization unit in an industrial control system for purification of carbon dioxide-based polyols according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0026] The present application provides a method for removing phenolic compounds and metal ions from carbon dioxide-based polyols, which specifically comprises: (1) Dynamic adsorption: (1-1) adding a modified adsorbent to a carbon dioxide-based polyol, wherein the modified adsorbent accounts for 1-26 wt% of the mass of the carbon dioxide-based polyol, and pre-adsorbing the adsorbent using low-frequency ultrasound, wherein the modified adsorbent is a six-membered nitrogen heterocyclic modified adsorbent; (1-2) heating and mechanically stirring to achieve adsorption equilibrium, wherein the adsorption equilibrium refers to dynamic monitoring of the removal rate of phenolic compounds in the carbon dioxide-based polyol to reach more than 99%, and the dynamic monitoring refers to sampling and testing every 20 minutes; (2) Transformer solution: (2-1) transferring the adsorption equilibrium mixture into a pressurized filter tank, and filtering the mixture with an inert gas at a low pressure to form a filter cake, wherein the low pressure is 0.1-0.6 MPa; (2-2) Then, the product is rapidly filtered under medium pressure of an inert gas, wherein the medium pressure is 1.5-3 MPa, and the filtrate is collected at the bottom of the pressure filter tank to obtain the purified carbon dioxide-based polyol.

[0027] Specifically, the dynamic adsorption process includes low-frequency ultrasonic pre-adsorption and heating mechanical stirring process. Low-frequency ultrasonic waves can form tiny bubbles and generate violent eddies and vortices in the bubbles, which can accelerate the transmission and diffusion of substances, so that the carbon dioxide-based polyols in the adsorption process can reach the adsorbent surface faster, thereby improving the adsorption rate and adsorption amount; and ultrasonic waves can also cause vibration and rotation of carbon dioxide-based polyols, thereby enhancing the contact frequency and time between substances in the adsorption process, which helps to improve the adsorption efficiency. The heating mechanical stirring process can increase the temperature of carbon dioxide-based polyols, thereby enhancing the movement speed and activity of carbon dioxide-based polyol molecules, which is more conducive to their rapid adsorption by activated carbon; and the heating stirring process can also trigger the interaction between the hydrogen bonds and π-π bonds of phenolic compounds in carbon dioxide-based polyols and six-membered nitrogen heterocyclic activated carbon, making it easier to chemically adsorb, which helps to enhance the adsorption effect and stabilize the adsorption of carbon dioxide-based polyols. At the same time, by performing dynamic adsorption monitoring during the heating mechanical stirring process, when the removal rate of phenolic compounds reaches more than 99%, it is considered that adsorption equilibrium has been reached. At this time, the mixture is transferred in time for pressure change desorption to avoid adsorption for too long time leading to adsorbent saturation and affecting the purification effect.

[0028] In the process of variable pressure desorption, a method combining low pressure to form a filter cake and then medium pressure rapid filtration is adopted. Under low pressure conditions, the carbon dioxide-based polyol adsorbed on the activated carbon can be initially desorbed by adjusting the low pressure to form a desorption filter cake, which is helpful for the preliminary separation of carbon dioxide-based polyols and lays the foundation for subsequent medium pressure rapid filtration. Subsequently, rapid filtration under medium pressure conditions can further improve the separation efficiency of the desorption product. The medium pressure can enhance its fluidity and make the desorption product pass through the filter cake faster, thereby achieving efficient separation. The removal method described in the present application is simple and convenient to operate, and can make the removal rate of phenolic compounds in the purified carbon dioxide-based polyols as high as 99% or more, and the metal ion concentration is less than 10ppm. Carbon dioxide-based polyols with quality that meets the requirements of downstream production can be obtained efficiently and quickly, with less metal residue, low chroma and high transparency.

[0029] In one embodiment of the present application, the modified adsorbent is prepared by the steps of pre-dispersion, six-membered nitrogen heterocycle modification, high-temperature calcination, and carbon dioxide gas modification, specifically comprising: (1) pre-dispersion: dispersing activated carbon in deionized water to prepare an activated carbon dispersion to form an activated and stable activated carbon precursor; (2) six-membered nitrogen heterocycle modification: adding the six-membered nitrogen heterocycle modifier to the activated carbon dispersion at a mass ratio of 1:0.5-20, heating and refluxing to prepare the six-membered nitrogen heterocycle modified activated carbon, wherein the purity of the six-membered nitrogen heterocycle modifier is above 99.99%; (3) high-temperature calcination: activating the six-membered nitrogen heterocycle modified activated carbon by high-temperature calcination to obtain an activated adsorbent; (4) carbon dioxide gas modification: preparing a solution of the activated adsorbent using water as a solvent, dispersing and soaking it, and blowing it with carbon dioxide gas until the solution pH is 6-8, filtering, and vacuum drying to obtain the modified adsorbent.

[0030] In one embodiment of the present application, in step (1-1), the low-frequency ultrasonic wave has a frequency of 10-40 kHz; In one embodiment of the present application, the pre-adsorption temperature is 30-90 °C; In one embodiment of the present application, the pre-adsorption time is 1.5-6 h; In one embodiment of the present application, the low-pressure filtration time is 0.5-3.5 h, and the temperature is 25±10°C; In one embodiment of the present application, the medium pressure rapid filtration time is 3-12 h and the temperature is 10-120°C.

[0031] In one embodiment of the present invention, the method for detecting and calculating the removal rate of the phenolic compounds is as follows: using an ultraviolet spectrophotometer to detect the absorbance corresponding to the maximum absorption wavelength (Amax) of the phenolic compounds, and then calculating a standard curve according to the Lambert-Beer law, and finally obtaining the concentration of the phenolic compounds in the carbon dioxide-based polyol, and then calculating the removal rate of the phenolic compounds according to the following formula;

[0032] ŋ represents the removal rate of phenolic compounds (%), C0 represents the initial concentration of phenolic compounds in CO2-based polyols (mg / L), C t It represents the concentration of phenolic compounds in CO2-based polyols after adsorption by the modified adsorbent (mg / L).

[0033] In particular, considering that during the heating mechanical stirring process, the manual sampling test every 20min can effectively monitor the removal rate, but it is limited by the influence of the sampling interval, and the reaction process cannot be understood in real time, and the optimal adsorption equilibrium state may be missed, thereby affecting the purification effect. In response to this technical problem, the present application proposes a purification industrial control system for carbon dioxide-based polyols, which is used to intelligently predict the stirring time in the dynamic adsorption process. Specifically, in the removal process of phenolic compounds in carbon dioxide-based polyols, the removal rate of phenolic compounds is detected at two different times, and the stirring speed and stirring temperature between the two time points are fine-grained. The local time domain interaction analysis is performed to reveal the combined effect of the stirring speed and stirring temperature on the removal efficiency of phenolic compounds, and based on this, the removal measurement is performed to generate an effective removal measurement coefficient, and then combined with the change in the removal rate between the two time points, the stirring time still required is intelligently estimated. In this way, the reaction end point can be predicted in advance, and incomplete reaction or overreaction can be effectively avoided, thereby improving the purification efficiency and product quality, and improving the automation level of industrial production.

[0034] Figure 1 The block diagram is a control system for purifying carbon dioxide-based polyols according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of an industrial control system for purifying carbon dioxide-based polyols according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to an embodiment of the present application, an industrial control system 100 for purifying carbon dioxide-based polyols includes: a removal rate data receiving module 110, which is used to receive the removal rates of phenolic compounds in carbon dioxide-based polyols at a first time point and a second time point to obtain a first removal rate and a second removal rate; a stirring state parameter receiving module 120, which is used to receive a time queue of stirring speed values ​​and a time queue of stirring temperature values ​​collected by a sensor group between the first time point and the second time point; a local time domain feature extraction module 130, which is used to perform local time domain feature extraction on the time queue of stirring speed values ​​and the time queue of stirring temperature values, respectively, to obtain a sequence of local time series correlation feature vectors of stirring temperature and a sequence of local time series correlation feature vectors of stirring speed. A multimodal stirring state parameter interactive analysis module 140 is used to interactively couple each group of corresponding stirring temperature local time series correlation feature vectors and stirring speed local time series correlation feature vectors in the sequence of stirring temperature local time series correlation feature vectors and the sequence of stirring speed local time series correlation feature vectors to obtain a sequence of stirring state multimodal local time series coupling feature vectors; an effective removal metric module 150 is used to generate an effective removal metric coefficient based on the global time domain aggregation features of the sequence of stirring state multimodal local time series coupling feature vectors; a remaining stirring time estimation module 160 is used to generate an estimated remaining stirring time based on the first removal rate, the second removal rate and the effective removal metric coefficient.

[0035] In the above-mentioned industrial control system 100 for purification of carbon dioxide-based polyols, the removal rate data receiving module 110 is used to receive the removal rates of phenolic compounds in carbon dioxide-based polyols at a first time point and a second time point to obtain a first removal rate and a second removal rate. It should be understood that based on the comparison between the first removal rate and the second removal rate, the progress of the removal of phenolic compounds in this time period can be understood, and a preliminary evaluation of the removal efficiency of phenolic compounds can be achieved.

[0036] In the above-mentioned purification industrial control system 100 for carbon dioxide-based polyols, the stirring state parameter receiving module 120 is used to receive the time queue of the stirring speed value and the time queue of the stirring temperature value between the first time point and the second time point collected by the sensor group. It should be understood that the stirring speed and the stirring temperature are important process parameters that affect the removal efficiency of phenolic compounds. Therefore, in order to fully understand the comprehensive influence of the stirring speed and the stirring temperature on the removal efficiency of phenolic compounds during this time period, the time queue of the stirring speed value and the time queue of the stirring temperature value between the first time point and the second time point are further obtained, and the combined effect of the stirring speed and the stirring temperature on the removal efficiency of phenolic compounds is revealed by the time series joint analysis of the stirring speed and the stirring temperature.

[0037] In the above-mentioned purification industrial control system 100 for carbon dioxide-based polyols, the local time domain feature extraction module 130 is used to perform local time domain feature extraction on the time queue of the stirring speed value and the time queue of the stirring temperature value to obtain a sequence of local time series associated feature vectors of the stirring temperature and a sequence of local time series associated feature vectors of the stirring speed. In a specific example of the present application, the local time domain feature extraction module 130 is used to: input the time queue of the stirring speed value and the time queue of the stirring temperature value into a parameter local time series feature extractor based on a 1D-CNN model to obtain a sequence of local time series associated feature vectors of the stirring temperature and a sequence of local time series associated feature vectors of the stirring speed. It should be understood that considering that directly analyzing the stirring speed and temperature data over a long time span may cause the calculation to be too complicated and cause the loss of local detail information. Therefore, in order to analyze the temporal variation trend of the stirring speed and temperature in more detail, the present application further performs a temporal analysis based on the local time scale on the time queue of the stirring speed value and the time queue of the stirring temperature value. Specifically, the time queue of the stirring speed value and the time queue of the stirring temperature value are first divided into local time domains, and the long time span data is divided into multiple shorter time periods. Then, a parameter local time series feature extractor based on the 1D-CNN model is used to extract features of the stirring speed and stirring temperature in each local time domain. A one-dimensional sliding convolution operation along the time dimension is performed to capture the local time series change pattern of the stirring speed and temperature changes, thereby obtaining a sequence of local time series associated feature vectors of the stirring temperature and a sequence of local time series associated feature vectors of the stirring speed.

[0038] In the above-mentioned purification industrial control system 100 for carbon dioxide-based polyols, the multimodal stirring state parameter interaction analysis module 140 is used to interactively couple each group of corresponding stirring temperature local time series association feature vectors and stirring speed local time series association feature vectors in the sequence of the stirring temperature local time series association feature vectors and the sequence of the stirring speed local time series association feature vectors to obtain a sequence of multimodal local time series coupling feature vectors of the stirring state. That is, in order to dig out the temporal joint effect of the stirring speed and the stirring temperature, it is necessary to further interactively fuse and analyze the stirring temperature characteristics and the stirring speed characteristics in the corresponding local time domain. It is worth mentioning that in order to improve the representativeness of feature interaction, the present application proposes a feature optimal matching multi-scale interactive coupling method, which is based on principal component analysis and feature matching technology, and can effectively identify and match the important correlation interaction effects between the stirring speed characteristics and the stirring temperature characteristics in the same local time domain, and realize a deeper feature interaction analysis, thereby improving the distinguishability and representativeness of the features.

[0039] Figure 3FIG. 1 is a block diagram of a multi-modal stirring state parameter interactive analysis module in an industrial control system for purification of carbon dioxide-based polyols according to an embodiment of the present application. Figure 3 As shown, the multimodal stirring state parameter interaction analysis module 140 includes: a principal component analysis unit 141, which is used to perform principal component analysis on the stirring temperature local time series correlation feature vector and the stirring speed local time series correlation feature vector respectively to obtain a set of stirring temperature local time series principal component feature components and a set of stirring speed local time series principal component feature components; an optimal pairing interaction coupling unit 142, which is used to perform optimal pairing multi-scale interaction coupling on the set of stirring temperature local time series principal component feature components and the set of stirring speed local time series principal component feature components to obtain the stirring state multimodal local time series coupling feature vector.

[0040] Specifically, the optimal pairing interactive coupling unit 142 includes: an optimal matching screening subunit, which is used to use each stirring temperature local time series principal component characteristic component in the set of stirring temperature local time series principal component characteristic components as a query vector, and use the set of stirring speed local time series principal component characteristic components as a query library, and match the stirring speed local time series principal component characteristic components that best match each query vector from the query library to obtain a set of optimal matching pairs of stirring temperature local time series principal component characteristic components and stirring speed local time series principal component characteristic components; a multi-scale interactive coupling subunit , used to input the best matching pairs of the local timing principal component characteristic components of the stirring temperature and the local timing principal component characteristic components of the stirring speed into the multi-scale interactive response coupling module respectively to obtain a set of interactive coupling representation vectors of the best matching pairs of the stirring temperature-speed local timing characteristics; a feature cascade subunit, used to cascade the set of interactive coupling representation vectors of the best matching pairs of the stirring temperature-speed local timing characteristics to obtain the multi-modal local timing coupling characteristic vector of the stirring state.

[0041] More specifically, the best match screening subunit is used to: calculate the Mahalanobis distance between the query vector and each stirring speed local timing principal component feature component in the query library to obtain a set of stirring temperature-speed matching difference coefficients; select the stirring speed local timing principal component feature component corresponding to the minimum value in the set of stirring temperature-speed matching difference coefficients as the optimal matching result of the query vector to obtain the best matching pair of the stirring temperature local timing principal component feature component and the stirring speed local timing principal component feature component.

[0042] More specifically, the multi-scale interactive coupling subunit is used to: respectively calculate the best matching pair of the local time series principal component characteristic component of the stirring temperature and the local time series principal component characteristic component of the stirring speed by adding, subtracting and multiplying by position points between the local time series principal component characteristic component of the stirring temperature and the local time series principal component characteristic component of the stirring speed to obtain the first stirring temperature-speed principal component component best matching interactive representation vector, the second stirring temperature-speed principal component component best matching interactive representation vector and the third stirring temperature-speed principal component component best matching interactive representation vector; the first stirring temperature-speed principal component component best matching interactive representation vector, the second stirring temperature-speed principal component component best matching interactive representation vector and the third stirring temperature-speed principal component component best matching interactive representation vector are added; The best matching interaction representation vector of the component component and the best matching interaction representation vector of the third stirring temperature-speed principal component component are cascaded and fused, and then input into a one-dimensional convolutional layer including a maximum pooling layer to obtain a multi-scale interaction representation vector of the best matching of the stirring temperature-speed principal component component; the maximum value among the second norm of the local time series principal component feature component of the stirring temperature, the second norm of the local time series principal component feature component of the stirring speed and a preset scale adjustment parameter is selected as a scale scaling factor, and each eigenvalue in the multi-scale interaction representation vector of the best matching of the stirring temperature-speed principal component component is divided by the scale scaling factor to obtain the interaction coupling representation vector of the best matching pair of the stirring temperature-speed local time series characteristics.

[0043] That is, first, the principal component analysis (PCA) technique is used to process the local time series correlation feature vector of the stirring temperature and the local time series correlation feature vector of the stirring speed in the same local time domain to reduce the complexity of the data and extract the main features therein, so that the subsequent interactive response analysis process focuses on the main feature parts, thereby obtaining a set of the local time series principal component feature components of the stirring temperature and a set of the local time series principal component feature components of the stirring speed. Subsequently, the optimal matching algorithm is used to perform feature query matching on the sets of the two principal component feature components, and the Mahalanobis distance is used as a metric to query the most relevant feature pairs between the two, and a one-to-one correspondence between the local time series principal component feature components of the stirring temperature and the local time series principal component feature components of the stirring speed is established, thereby achieving an accurate match of the strong correlation interaction mode between the stirring speed and the stirring temperature. Next, each best matching pair is input into the multi-scale interactive response coupling module for interactive response analysis of the best pairing features. The multi-scale interactive response coupling module can effectively capture the interaction patterns of stirring speed and stirring temperature at different abstract levels by performing multi-level feature interaction and deep learning on the local time series principal component feature components of stirring temperature and the local time series principal component feature components of stirring speed in the best matching pair, and automatically learn and generate the interactive feature representation of each best matching pair. Finally, by performing a cascade operation on the interactive feature representations of all best matching pairs to integrate global information, a multi-modal local time series coupling feature vector of the stirring state is obtained, which comprehensively reflects the interaction pattern between stirring temperature and stirring speed in the local time domain, thereby providing a more comprehensive and in-depth analysis basis for the prediction of the remaining stirring time.

[0044] Correspondingly, the multimodal stirring state parameter interaction analysis module 140 is used to process the stirring temperature local time series correlation feature vector and the stirring speed local time series correlation feature vector using the following principal component matching interaction fusion formula to obtain the stirring state multimodal local time series coupling feature vector, wherein the principal component matching interaction fusion formula is:

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] in, and Respectively represent the covariance matrix of the local time series correlation feature vector of the stirring temperature and the local time series correlation feature vector of the stirring speed, is a diagonal matrix consisting of the set of principal component eigenvalues ​​of the local time series of stirring temperature obtained by eigenvalue decomposition of the covariance matrix of the local time series characteristics of the stirring temperature, and Respectively represent the first and the second in the set of principal component eigenvalues ​​of the local time series of the stirring temperature The principal component eigenvalues ​​of the local time series of stirring temperature, where is the number of principal component eigenvalues ​​of the local time series of the stirring temperature, is a matrix composed of the set of principal component characteristic components of the local time series of stirring temperature, , , and respectively represent the first, second, and third components of the set of main component characteristic components of the local time series of the stirring temperature. and The principal component characteristic components of the local time series of stirring temperature, represents the transpose of a matrix, is a diagonal matrix consisting of the set of principal component eigenvalues ​​of the local time series of stirring speed obtained by eigenvalue decomposition of the local time series characteristic covariance matrix of stirring speed, and Respectively represent the first and the second in the set of principal component eigenvalues ​​of the local time series of the stirring speed The principal component eigenvalues ​​of the local time series of stirring speed, is the matrix composed of the set of characteristic components of the local time series of the stirring speed. , , , and Respectively represent the first, second, and third principal component characteristic components of the local time series of the stirring speed. , and The principal component characteristic components of the local time series of stirring speed, Indicates The principal component characteristic components of the local time series of the stirring temperature and the The inverse matrix of the covariance matrix between the principal component characteristic components of the local time series of stirring speed, Indicates the index corresponding to the minimum value. Indicates that The index of the local time series principal component feature component of the stirring speed with the smallest semantic difference among the local time series principal component feature components of the stirring temperature, represents dot product, Indicates point addition, Indicates point reduction, represents a one-dimensional convolution operation, represents the maximum pooling operation, represents the bi-norm of the eigenvector, To preset the scale adjustment parameters, is the maximum value function, , , and Respectively represent the first, second, and and The best matching vector of the local time series characteristics of stirring temperature and speed for interaction coupling is represented by: Indicates cascade operation, Represents the multi-modal local temporal coupling eigenvector of the stirring state.

[0054] In the above industrial control system 100 for purification of carbon dioxide-based polyols, the effective removal metric module 150 is used to generate an effective removal metric coefficient based on the global time-domain aggregation characteristics of the sequence of the multi-modal local temporal coupling feature vectors of the stirring state. Figure 4 FIG. 1 is a block diagram of an effective removal measurement module in an industrial control system for purifying carbon dioxide-based polyols according to an embodiment of the present application. Figure 4 As shown, the effective removal metric module 150 includes: a global propagation aggregation unit 151, which is used to perform feature timing dynamic propagation on the sequence of the stirring state multimodal local timing coupling feature vectors to obtain the stirring state timing effect aggregation representation vector; a removal metric unit 152, which is used to input the stirring state timing effect aggregation representation vector into the decoder-based removal metric module to obtain the effective removal metric coefficient.

[0055] Specifically, the global propagation aggregation unit 151 is used to perform characteristic time series dynamic propagation on the sequence of the multimodal local time series coupling feature vectors of the stirring state to obtain the stirring state time series action aggregation representation vector. It should be understood that after the above processing, each stirring state multimodal local time series coupling feature vector in the sequence of the multimodal local time series coupling feature vectors of the stirring state respectively represents the joint action characteristics of the multimodal stirring parameters in each local time domain. Further, in order to comprehensively consider the joint action information of the multimodal stirring parameters in the global time domain, the present application uses the time series analysis technology to recursively fuse the sequence of the multimodal local time series coupling feature vectors of the stirring state to achieve a global understanding of the stirring state of the entire reaction process. In particular, considering that in the processing of time series data, each local time domain feature representation will have different dependencies due to different time intervals. Therefore, in order to better simulate the stirring state information association under long time intervals, the present application further introduces the information time series attenuation mechanism in the information transmission fusion process, so as to dynamically adjust the time series propagation weight of the feature, so as to carry out more effective information propagation and aggregation.

[0056] Figure 5 FIG. 1 is a block diagram of a global propagation polymerization unit in an industrial control system for purifying carbon dioxide-based polyols according to an embodiment of the present application. Figure 5As shown, the global propagation aggregation unit 151 includes: a current stirring state feature principal component extraction subunit 1511, which is used to extract the current stirring state multimodal local temporal coupling feature vector from the sequence of the stirring state multimodal local temporal coupling feature vector, and perform principal component extraction based on eigenvalues ​​to obtain the current stirring state multimodal local temporal feature principal component representation vector; a historical stirring state feature principal component extraction subunit 1512, which is used to define other stirring state multimodal local temporal coupling feature vectors in the sequence of the stirring state multimodal local temporal coupling feature vector as historical stirring state multimodal local temporal coupling feature vectors to obtain a sequence of historical stirring state multimodal local temporal coupling feature vectors, and perform principal component extraction based on eigenvalues ​​on each of the historical stirring state multimodal local temporal coupling feature vectors to obtain a sequence of historical stirring state multimodal local temporal coupling feature principal component representation vectors; a temporal gradient attenuation weight calculation subunit 1513, which is used to extract the current stirring state multimodal local temporal coupling feature vector based on the historical stirring state. The time span of each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of multimodal local temporal feature principal component representation vectors relative to the current stirring state multimodal local temporal feature principal component representation vector is calculated to obtain a sequence of historical stirring state local feature temporal gradient attenuation weights; a historical information dynamic propagation aggregation subunit 1514 is used to calculate the weighted sum of the sequence of historical stirring state multimodal local temporal feature principal component representation vectors based on the sequence of historical stirring state local feature temporal gradient attenuation weights to obtain a historical stirring state multimodal local feature principal component significant transfer aggregation representation vector; a principal component aggregation subunit 1515 is used to calculate the positional sum of the historical stirring state multimodal local feature principal component significant transfer aggregation representation vector and the current stirring state multimodal local temporal feature principal component representation vector to obtain the stirring state temporal action aggregation representation vector.

[0057] Specifically, the current stirring state feature principal component extraction subunit 1511 is used to: calculate the covariance matrix of the current stirring state multimodal local temporal coupling feature vector to obtain the current stirring state multimodal local temporal feature covariance matrix; perform eigenvalue decomposition on the current stirring state multimodal local temporal feature covariance matrix to obtain a set of current stirring state multimodal local temporal feature principal component eigenvalues ​​and a corresponding set of current stirring state multimodal local temporal feature principal component eigenvectors; select the current stirring state multimodal local temporal feature principal component eigenvector corresponding to the maximum value in the set of current stirring state multimodal local temporal feature principal component eigenvalues ​​as the current stirring state multimodal local temporal feature principal component representation vector.

[0058] Specifically, the temporal gradient attenuation weight calculation subunit 1513 includes: a temporal propagation absolute attenuation factor calculation secondary subunit, which is used to calculate the temporal propagation absolute attenuation factor of each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of the historical stirring state multimodal local temporal feature principal component representation vector relative to the current stirring state multimodal local temporal feature principal component representation vector to obtain a sequence of temporal propagation absolute attenuation factors; a time span modulation secondary subunit, which is used to modulate the temporal dimension of each temporal propagation absolute attenuation factor in the sequence of temporal propagation absolute attenuation factors based on the time span between each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of the historical stirring state multimodal local temporal feature principal component representation vector and the current stirring state multimodal local temporal feature principal component representation vector to obtain a sequence of time span modulated propagation attenuation factors; a weight screening secondary subunit, which is used to input the sequence of time span modulated propagation attenuation factors into the information transfer screening module based on the gating function to obtain a sequence of temporal gradient attenuation weights of the historical stirring state local feature.

[0059] In a specific example of the present application, the secondary subunit for calculating the absolute attenuation factor of time propagation is used to: calculate the dot division vector between the principal component representation vector of the historical stirring state multimodal local time series characteristics and the principal component representation vector of the current stirring state multimodal local time series characteristics, and calculate the base-two logarithmic value of the absolute value of each eigenvalue in the dot division vector to obtain the absolute standard representation vector of time propagation attenuation; calculate the dot product vector between the absolute standard representation vector of time propagation attenuation and the principal component representation vector of the historical stirring state multimodal local time series characteristics, and calculate the exponential function value with base e and the sum of the eigenvalues ​​of the dot product vector as the exponent to obtain the absolute attenuation factor of time propagation.

[0060] In a specific example of the present application, the time span modulation secondary subunit is used to: calculate the timestamp difference between the historical stirring state multimodal local time series feature principal component representation vector and the current stirring state multimodal local time series feature principal component representation vector, round the timestamp difference down to obtain a time span factor; divide the time propagation absolute attenuation factor of the historical stirring state multimodal local time series feature principal component representation vector by the natural exponential function value of the time span factor to obtain the time span modulation propagation attenuation factor.

[0061] In a specific example of the present application, the weight screening secondary sub-unit is used to: based on a preset threshold, mask the sequence of time span modulated propagation attenuation factors to obtain a sequence of masked time span modulated propagation attenuation factors; input the sequence of masked time span modulated propagation attenuation factors into a Sigmoid function for normalization processing to obtain a sequence of temporal gradient attenuation weights of the local characteristics of the historical stirring state.

[0062] That is, first, the last time node in the sequence of the multimodal local temporal coupling feature vector of the stirring state is regarded as the current time node, and the other time nodes are regarded as historical nodes. PCA is performed on the multimodal local temporal coupling feature vector of the stirring state at each time node to remove noise interference, highlight key information, and extract the most representative principal component, thereby obtaining a sequence of the principal component representation vector of the current multimodal local temporal feature of the stirring state and the principal component representation vector of the historical multimodal local temporal feature of the stirring state. Then, in order to capture the trend of feature evolution over time and the information loss, the characteristic principal component change rate of each historical time node relative to the current time node and its corresponding absolute attenuation factor of temporal propagation are further calculated to enhance the model's ability to understand information transmission under long time intervals. Then, the above-mentioned absolute attenuation factor of temporal propagation is adjusted by introducing the time span factor to simulate the gradual forgetting process of data in time, relatively weaken the dependence on long time interval information, and the absolute attenuation factor of temporal propagation modulated by the time span is gated and screened to identify the most influential propagation path and assign corresponding weights accordingly. Finally, the principal component features of each time node are aggregated by calculating the position weighted sum to integrate the stirring action information in the global time domain and generate an aggregate representation vector of the stirring state temporal action, thus providing a solid information basis for the final stirring duration prediction.

[0063] Accordingly, the global propagation aggregation unit 151 is used to process the sequence of the stirring state multimodal local temporal coupling feature vectors using the following information temporal dynamic propagation formula to obtain a stirring state temporal action aggregation representation vector, wherein the information temporal dynamic propagation formula is:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070] in, represents the sequence of multi-modal local temporal coupling eigenvectors of the stirring state, is the number of vectors in the sequence of the multi-modal local temporal coupling characteristic vectors of the stirring state, represents the multi-modal local temporal coupling eigenvector of the current stirring state, , and The first, second and third in the sequence of multi-modal local temporal coupling eigenvectors of the historical stirring state are represented respectively. The multi-modal local temporal coupling eigenvectors of the historical stirring states, represents the principal component analysis function, Indicates The principal component representation vector of the multi-modal local time series characteristics of the historical stirring state, Represents the principal component representation vector of the multimodal local time series characteristics of the current stirring state, represents the transpose of a vector, represents the length of the multi-modal local temporal coupling characteristic vector of the historical stirring state, Indicates the The sample covariance matrix of the multi-modal local time series coupling feature vectors of the historical stirring state, that is, the covariance matrix of the multi-modal local time series features of the historical stirring state, Indicates that through The matrix is ​​composed of the set of principal component eigenvectors of the historical stirring state multimodal local time series characteristics obtained by eigenvalue decomposition of the covariance matrix of the historical stirring state multimodal local time series characteristics. Indicates that by The diagonal matrix composed of the set of principal component eigenvalues ​​of the historical stirring state multimodal local time series characteristics obtained by eigenvalue decomposition of the covariance matrix of the historical stirring state multimodal local time series characteristics is: is the diag function, which is used to construct a diagonal matrix. , , and Respectively represent The first, second, and third principal component eigenvalues ​​of the multi-modal local time series characteristics of the historical stirring state are and The principal component eigenvalues ​​of the multi-modal local time series characteristics of the historical stirring state, is the number of eigenvalues ​​in the set of principal component eigenvalues ​​of the multi-modal local time series characteristics of the historical stirring state, , , and Respectively represent The first, second, and third principal component feature vectors of the multi-modal local time series features of the historical stirring state are and The principal component eigenvectors of the multi-modal local time series characteristics of the historical stirring state, Indicates the index corresponding to the maximum value. represents the index of the maximum value in the set of principal component eigenvalues ​​of the multi-modal local time series characteristics of the historical stirring state, Indicates The principal component representation vector of the multimodal local time series characteristics of the historical stirring state eigenvalues, The first principal component representation vector of the multimodal local temporal characteristics of the current stirring state is represented by eigenvalues, represents the absolute value, represents the logarithmic function with base 2, represents the length of the principal component representation vector of the multimodal local time series characteristics of the current stirring state, represents the exponential function with base e, Indicates The absolute attenuation factor of the time series propagation of the principal component representation vector of the multi-modal local time series characteristics of the historical stirring state, and Respectively represent the multi-modal local temporal coupling eigenvector of the current stirring state and the The timestamp of the multi-modal local temporal coupling eigenvector of the historical stirring state, Indicates rounding down. Indicates The time span modulates the propagation attenuation factor, Indicates mask processing, represents the sigmoid function, Indicates the preset threshold value, Indicates The local feature temporal gradient attenuation weight of the historical stirring state, A vector representing the temporal action aggregation of the stirring state.

[0071] Specifically, the removal measurement unit 152 is used to input the stirring state temporal action aggregation representation vector into the decoder-based removal measurement module to obtain the effective removal measurement coefficient. In the technical solution of the present application, the decoder can effectively learn the temporal joint action information of the multimodal stirring parameters contained in the stirring state temporal action aggregation representation vector through training optimization, so as to intelligently predict the removal degree of phenolic compounds under such stirring action, generate corresponding effective removal measurement coefficients, and realize the quantitative representation of the stirring action.

[0072] Preferably, the stirring state sequential action aggregation representation vector is passed through a decoder-based removal measurement module to obtain an effective removal measurement coefficient, including: determining a stirring state sequential action aggregation association response matrix and a stirring state sequential action aggregation distance response matrix based on the square root of the association value of the eigenvalue of the stirring state sequential action aggregation representation vector and the L2 distance value, that is: matrix multiplying the stirring state sequential action aggregation representation vector with the stirring state sequential action aggregation association response matrix to obtain a stirring state sequential action aggregation association response vector; matrix multiplying the stirring state sequential action aggregation distance response matrix with the transposed vector of the stirring state sequential action aggregation representation vector to obtain to the stirring state sequential action aggregation distance response vector; after dot-multiplying the stirring state sequential action aggregation association response matrix and the stirring state sequential action aggregation distance response matrix, perform matrix multiplication with the transposed vector of the stirring state sequential action aggregation representation vector to obtain the stirring state sequential action aggregation bias vector; perform dot-multiplication on the stirring state sequential action aggregation association response vector, the stirring state sequential action aggregation distance response vector and the stirring state sequential action aggregation bias vector to obtain an optimized stirring state sequential action aggregation representation vector; pass the optimized stirring state sequential action aggregation representation vector through a decoder-based removal metric module to obtain an effective removal metric coefficient.

[0073] Here, the optimization process of the stirring state temporal action aggregation representation vector can be expressed by the following optimization formula:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] in, is the aggregation correlation response matrix of the stirring state temporal action, is the aggregate distance response matrix of the stirring state temporal effect, The stirring state time series aggregation correlation response matrix The eigenvalues ​​of the positions, The distance response matrix for the stirring state timing effect aggregation The eigenvalues ​​of the positions, and are the first and eigenvalues, It is the point-add vector between the stirring state time series action aggregate association response vector and the stirring state time series action aggregate distance response vector, is the aggregate bias vector for the stirring state timing action, is the set of real numbers, is the length of the aggregation vector of the stirring state time series action, It means point multiplication by position. represents matrix multiplication, Indicates adding by position point. is the aggregation representation vector of the stirring state timing action, Aggregate representation vector for optimized stirring state timing action.

[0081] That is, each stirring state multimodal local time series coupling feature vector in the sequence of the stirring state multimodal local time series coupling feature vectors respectively represents the interactive response coupling characteristics of the stirring temperature and stirring speed in each local time domain window based on the optimal matching of the characteristic principal component granularity. When the characteristic time series dynamic propagation aggregation is performed, the uneven distribution of the response coupling characteristics corresponding to the inconsistent parameter feature distribution in the local time domain will lead to the lack of local sparsity aggregation in the global time domain, that is, it will lead to the simple repetition of the time series aggregation distribution in the local time domain, which affects the logical dependency of the set time series features and reduces the decoding accuracy of the effective removal of the metric coefficient.

[0082] Therefore, by taking the stirring state temporal action aggregate representation vector as its statistical reference-free distribution response framework based on the autocorrelation matrix and the self-distance matrix, and constructing it based on the retrieval-enhanced reverse response of the stirring state temporal action aggregate representation vector, simple repetition of feature distribution is avoided, and by ensuring the intrinsic mapping logic of the stirring state temporal action aggregate representation vector based on the retrieval-response context, its surface combinatorial mapping is avoided, so as to achieve the logical dependency mapping of the stirring state temporal action aggregate representation vector to the classification target domain while maintaining the intuitive response correlation, thereby improving the accuracy of the effective removal metric coefficient of the stirring state temporal action aggregate representation vector obtained by the decoder-based removal metric module.

[0083] In the above-mentioned purification industrial control system 100 for carbon dioxide-based polyols, the remaining stirring time estimation module 160 is used to generate an estimated remaining stirring time based on the first removal rate, the second removal rate and the effective removal measurement coefficient. It should be understood that the effective removal measurement coefficient is a quantitative indicator for measuring the degree of removal of phenolic compounds within a corresponding time range. By calculating the ratio between the relative change of the first removal rate and the second removal rate and the effective removal measurement coefficient, the removal efficiency and removal stage of the phenolic compounds can be effectively revealed, which can be used as a measure of the time required for removal completion, and can provide effective guidance and reference for the removal control of phenolic compounds in industrial production processes.

[0084] Accordingly, the remaining stirring time estimation module 160 is used to process the first removal rate, the second removal rate and the effective removal measurement coefficient according to the following estimated remaining stirring time calculation formula to obtain the estimated remaining stirring time, wherein the estimated remaining stirring time calculation formula is:

[0085] In summary, the purification industrial control system for carbon dioxide-based polyols described in the present application is explained, which detects the removal rate of phenolic compounds at two different times during the removal process of phenolic compounds in carbon dioxide-based polyols, and uses data processing technology based on deep learning to perform fine-grained local time-domain interactive analysis of the stirring speed and stirring temperature between the two time points to reveal the combined effect of stirring speed and stirring temperature on the removal efficiency of phenolic compounds, and based on this, removes the measurement, generates an effective removal measurement coefficient, and then combines the change in the removal rate between the two time points to intelligently estimate the stirring time that is still required. In this way, the reaction endpoint can be predicted in advance, effectively avoiding incomplete reaction or overreaction, thereby improving purification efficiency and product quality, and improving the level of automation in industrial production.

[0086] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present application are merely examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are necessarily possessed by each embodiment of the present application.

Claims

1. An industrial control system for the purification of carbon dioxide-based polyols, characterized in that: include: A removal rate data receiving module, used to receive the removal rates of phenolic compounds in the carbon dioxide-based polyol at a first time point and a second time point to obtain a first removal rate and a second removal rate; A stirring state parameter receiving module, used for receiving a time queue of stirring speed values ​​and a time queue of stirring temperature values ​​collected by the sensor group between a first time point and a second time point; A local time domain feature extraction module, used for performing local time domain feature extraction on the time queue of the stirring speed value and the time queue of the stirring temperature value respectively to obtain a sequence of local time series associated feature vectors of the stirring temperature and a sequence of local time series associated feature vectors of the stirring speed; A multi-modal stirring state parameter interactive analysis module, used for interactively coupling each group of corresponding stirring temperature local time series correlation feature vectors and stirring speed local time series correlation feature vectors in the sequence of stirring temperature local time series correlation feature vectors and the sequence of stirring speed local time series correlation feature vectors to obtain a sequence of stirring state multi-modal local time series coupling feature vectors; An effective removal metric module, used to generate an effective removal metric coefficient based on the global time-domain aggregation characteristics of the sequence of multi-modal local temporal coupling feature vectors of the stirring state; A module for estimating a remaining stirring time, configured to generate an estimated remaining stirring time based on the first removal rate, the second removal rate and the effective removal metric coefficient; Among them, the multimodal stirring state parameter interactive analysis module includes: a principal component analysis unit, which is used to perform principal component analysis on the local time series correlation feature vector of the stirring temperature and the local time series correlation feature vector of the stirring speed respectively to obtain a set of local time series principal component feature components of the stirring temperature and a set of local time series principal component feature components of the stirring speed; an optimal pairing interactive coupling unit, which is used to perform optimal pairing multi-scale interactive coupling on the set of local time series principal component feature components of the stirring temperature and the set of local time series principal component feature components of the stirring speed to obtain the multimodal local time series coupling feature vector of the stirring state.

2. The industrial control system for purification of carbon dioxide-based polyols according to claim 1, characterized in that: The local time domain feature extraction module is used to: The time queue of the stirring speed values ​​and the time queue of the stirring temperature values ​​are respectively input into the parameter local temporal feature extractor based on the 1D-CNN model to obtain a sequence of the stirring temperature local temporal association feature vectors and a sequence of the stirring speed local temporal association feature vectors.

3. The industrial control system for purification of carbon dioxide-based polyols according to claim 2, characterized in that: The optimal pairing interaction coupling unit comprises: The best matching screening subunit is used to use each stirring temperature local time series principal component characteristic component in the set of stirring temperature local time series principal component characteristic components as a query vector, and use the set of stirring speed local time series principal component characteristic components as a query library, and match the stirring speed local time series principal component characteristic components that best match each query vector from the query library to obtain a set of best matching pairs of stirring temperature local time series principal component characteristic components and stirring speed local time series principal component characteristic components; A multi-scale interactive coupling subunit, used for inputting each best matching pair of the local time series principal component characteristic component of the stirring temperature and the local time series principal component characteristic component of the stirring speed in the set of the best matching pairs of the local time series principal component characteristic component of the stirring temperature and the local time series principal component characteristic component of the stirring speed into the multi-scale interactive response coupling module to obtain a set of interactive coupling representation vectors of the best matching pairs of the local time series characteristics of the stirring temperature and the speed; The feature cascade subunit is used to cascade the set of interactive coupling representation vectors of the best matching pair of the stirring temperature-speed local time series features to obtain the stirring state multi-modal local time series coupling feature vector.

4. The industrial control system for purification of carbon dioxide-based polyols according to claim 3, characterized in that: The best matching screening subunit is used for: Calculating the Mahalanobis distance between the query vector and each stirring speed local time series principal component characteristic component in the query library to obtain a set of stirring temperature-speed matching difference coefficients; The local time series principal component characteristic component of the stirring speed corresponding to the minimum value in the set of the stirring temperature-speed matching difference coefficients is selected as the optimal matching result of the query vector to obtain the best matching pair of the local time series principal component characteristic component of the stirring temperature and the local time series principal component characteristic component of the stirring speed.

5. The industrial control system for purification of carbon dioxide-based polyols according to claim 4, characterized in that: The multi-scale interactive coupling subunit is used to: Respectively calculate the position point addition, position point subtraction and position point multiplication between the local time series principal component characteristic components of the stirring temperature and the local time series principal component characteristic components of the stirring speed in the best matching pair to obtain the first stirring temperature-speed principal component component best matching interaction representation vector, the second stirring temperature-speed principal component component best matching interaction representation vector and the third stirring temperature-speed principal component component best matching interaction representation vector; The first stirring temperature-speed principal component best matching interaction representation vector, the second stirring temperature-speed principal component best matching interaction representation vector and the third stirring temperature-speed principal component best matching interaction representation vector are cascaded and fused, and then input into a one-dimensional convolutional layer including a maximum pooling layer to obtain a stirring temperature-speed principal component best matching multi-scale interaction representation vector; The second norm of the local time series principal component characteristic component of the stirring temperature, the second norm of the local time series principal component characteristic component of the stirring speed and the maximum value of the preset scale adjustment parameter are selected as the scale scaling factor, and each eigenvalue in the best matching multi-scale interaction representation vector of the stirring temperature-speed principal component component is divided by the scale scaling factor to obtain the best matching interaction coupling representation vector of the stirring temperature-speed local time series characteristics.

6. The industrial control system for purification of carbon dioxide-based polyols according to claim 5, characterized in that: The effective removal measurement module comprises: A global propagation aggregation unit, used for performing characteristic time series dynamic propagation on the sequence of the stirring state multi-modal local time series coupling characteristic vectors to obtain a stirring state time series action aggregation representation vector; The removal metric unit is used to input the stirring state timing action aggregation representation vector into the decoder-based removal metric module to obtain the effective removal metric coefficient.

7. The industrial control system for purification of carbon dioxide-based polyols according to claim 6, characterized in that: The global propagation aggregation unit includes: The current stirring state characteristic principal component extraction subunit is used to extract the current stirring state multimodal local temporal coupling characteristic vector from the sequence of the stirring state multimodal local temporal coupling characteristic vector, and perform principal component extraction based on the eigenvalue to obtain the current stirring state multimodal local temporal characteristic principal component representation vector; A historical stirring state feature principal component extraction subunit is used to define other stirring state multimodal local temporal coupling feature vectors in the sequence of stirring state multimodal local temporal coupling feature vectors as historical stirring state multimodal local temporal coupling feature vectors to obtain a sequence of historical stirring state multimodal local temporal coupling feature vectors, and perform principal component extraction based on eigenvalues ​​on each of the historical stirring state multimodal local temporal coupling feature vectors to obtain a sequence of historical stirring state multimodal local temporal coupling feature principal component representation vectors; A temporal gradient attenuation weight calculation subunit is used to calculate the temporal gradient attenuation weight of each historical stirring state multimodal local temporal feature principal component representation vector in the sequence of the historical stirring state multimodal local temporal feature principal component representation vectors relative to the current stirring state multimodal local temporal feature principal component representation vector to obtain a sequence of historical stirring state local feature temporal gradient attenuation weights; A historical information dynamic propagation aggregation subunit is used to calculate the weighted sum of the sequence of principal component representation vectors of the historical stirring state multimodal local temporal feature based on the sequence of temporal gradient attenuation weights of the historical stirring state local feature to obtain a historical stirring state multimodal local feature principal component significant transfer aggregation representation vector; The principal component aggregation subunit is used to calculate the positional sum of the principal component significant transfer aggregation representation vector of the historical stirring state multimodal local feature and the principal component representation vector of the current stirring state multimodal local temporal feature to obtain the stirring state temporal action aggregation representation vector.

8. The industrial control system for purification of carbon dioxide-based polyols according to claim 7, characterized in that: The temporal gradient attenuation weight calculation subunit includes: A secondary subunit for calculating the time series propagation absolute attenuation factor is used to calculate the time series propagation absolute attenuation factor of each historical stirring state multimodal local time series feature principal component representation vector in the sequence of the historical stirring state multimodal local time series feature principal component representation vector relative to the current stirring state multimodal local time series feature principal component representation vector to obtain a sequence of time series propagation absolute attenuation factors; A time span modulation secondary subunit is used to modulate the time dimension of each time series propagation absolute attenuation factor in the sequence of time series propagation absolute attenuation factors based on the time span between each historical stirring state multimodal local time series feature principal component representation vector in the sequence of historical stirring state multimodal local time series feature principal component representation vector and the current stirring state multimodal local time series feature principal component representation vector to obtain a sequence of time span modulated propagation attenuation factors; The weight screening secondary subunit is used to input the sequence of time span modulated propagation attenuation factors into the information transfer screening module based on the gating function to obtain the sequence of temporal gradient attenuation weights of the local characteristics of the historical stirring state.

9. The industrial control system for purification of carbon dioxide-based polyols according to claim 8, characterized in that: The time span modulation secondary subunit is used for: Calculating the timestamp difference between the principal component representation vector of the historical stirring state multimodal local time series feature and the principal component representation vector of the current stirring state multimodal local time series feature, and rounding down the timestamp difference to obtain a time span factor; The time span modulation propagation attenuation factor is obtained by dividing the time span absolute attenuation factor of the principal component representation vector of the multi-modal local time series characteristics of the historical stirring state by the natural exponential function value of the time span factor.

10. The industrial control system for purification of carbon dioxide-based polyols according to claim 9, characterized in that: The remaining stirring time estimation module is used to: The first removal rate, the second removal rate and the effective removal metric coefficient are processed by the following calculation formula for estimating the remaining stirring time to obtain the estimated remaining stirring time, wherein the calculation formula for estimating the remaining stirring time is: Estimated remaining mixing time = .

Images