WO2024000798A1 - 用于电子级氢氟酸制备的生产管理控制系统及其控制方法 - Google Patents

用于电子级氢氟酸制备的生产管理控制系统及其控制方法 Download PDF

Info

Publication number
WO2024000798A1
WO2024000798A1 PCT/CN2022/116539 CN2022116539W WO2024000798A1 WO 2024000798 A1 WO2024000798 A1 WO 2024000798A1 CN 2022116539 W CN2022116539 W CN 2022116539W WO 2024000798 A1 WO2024000798 A1 WO 2024000798A1
Authority
WO
WIPO (PCT)
Prior art keywords
tower
feature vector
absorption tower
parameter
parameters
Prior art date
Application number
PCT/CN2022/116539
Other languages
English (en)
French (fr)
Inventor
丘添明
雷游生
黄明新
廖鸿辉
邱汉林
郑宏慎
丘赞文
Original Assignee
福建省龙氟新材料有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 福建省龙氟新材料有限公司 filed Critical 福建省龙氟新材料有限公司
Publication of WO2024000798A1 publication Critical patent/WO2024000798A1/zh

Links

Images

Classifications

    • C—CHEMISTRY; METALLURGY
    • C01—INORGANIC CHEMISTRY
    • C01B—NON-METALLIC ELEMENTS; COMPOUNDS THEREOF; METALLOIDS OR COMPOUNDS THEREOF NOT COVERED BY SUBCLASS C01C
    • C01B7/00—Halogens; Halogen acids
    • C01B7/19—Fluorine; Hydrogen fluoride
    • C01B7/191—Hydrogen fluoride
    • C—CHEMISTRY; METALLURGY
    • C01—INORGANIC CHEMISTRY
    • C01B—NON-METALLIC ELEMENTS; COMPOUNDS THEREOF; METALLOIDS OR COMPOUNDS THEREOF NOT COVERED BY SUBCLASS C01C
    • C01B7/00—Halogens; Halogen acids
    • C01B7/19—Fluorine; Hydrogen fluoride
    • C01B7/191—Hydrogen fluoride
    • C01B7/195—Separation; Purification
    • C—CHEMISTRY; METALLURGY
    • C01—INORGANIC CHEMISTRY
    • C01B—NON-METALLIC ELEMENTS; COMPOUNDS THEREOF; METALLOIDS OR COMPOUNDS THEREOF NOT COVERED BY SUBCLASS C01C
    • C01B7/00—Halogens; Halogen acids
    • C01B7/19—Fluorine; Hydrogen fluoride
    • C01B7/191—Hydrogen fluoride
    • C01B7/195—Separation; Purification
    • C01B7/196—Separation; Purification by distillation
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G05B19/41885—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Definitions

  • the present invention relates to the field of intelligent equipment monitoring, and more specifically, to a production management control system for the preparation of electronic grade hydrofluoric acid and a control method thereof.
  • Electronic-grade hydrofluoric acid is a strong acidic cleaning agent and corrosive agent, mainly used in the production of very large-scale integrated circuits.
  • the main production method of electronic grade hydrofluoric acid is: first chemically pretreat industrial anhydrous hydrofluoric acid, then perform distillation, then cool the obtained hydrogen fluoride gas, absorb it with pure water, and finally filter and fill it. Since the presence of impurity arsenic has a serious impact on the performance of electronic devices, arsenic removal is a key issue in the purification process of hydrofluoric acid.
  • the commonly used method is to use an oxidant to convert trivalent arsenic impurities into high boiling point pentavalent arsenic. Compounds, commonly used oxidants include KMnO4, CrO3, persulfate, etc.
  • Patent 103991847 discloses a method for preparing electronic grade hydrofluoric acid, which uses hydrogen peroxide as an oxidant to convert trivalent arsenic impurities into high-boiling point pentavalent arsenic compounds without introducing additional impurities, thus producing high yield and good quality products. , low-cost electronic grade hydrofluoric acid products, and recover hydrogen fluoride gas in the exhaust gas.
  • an equipment production management control system for the preparation of electronic grade hydrofluoric acid is expected to provide online monitoring of the performance of each equipment in the production line.
  • Embodiments of the present application provide a production management control system and a control method for the preparation of electronic grade hydrofluoric acid, which adopts artificial intelligence-based monitoring technology to control the distillation tower and Various parameters of the absorption tower are monitored to ensure the production yield of the final electronic grade hydrofluoric acid product, and during this process, the dynamic change characteristics of the product are also used to estimate the responsiveness to more accurately evaluate the distillation tower. and monitor whether the equipment performance of the absorption tower is normal.
  • a production management control system for the preparation of electronic grade hydrofluoric acid which includes:
  • the equipment parameter acquisition module is used to obtain multiple first parameters of the rectification tower and multiple second parameters of the absorption tower at multiple predetermined time points, wherein the multiple first parameters include the towers of the rectification tower. Still temperature, the tower body temperature of the rectification tower, the top temperature of the rectification tower, the bottom pressure of the rectification tower, the top pressure of the rectification tower, the Reflux temperature and reflux ratio in the tower, the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower;
  • a product status acquisition module configured to obtain multiple liquid chromatograms of the hydrogen fluoride liquid before first filtration at the multiple predetermined time points through a liquid chromatograph
  • a product tracking encoding module configured to pass multiple liquid chromatograms of the hydrogen fluoride liquid before first filtration at the multiple predetermined time points through a first convolutional neural network using a three-dimensional convolution kernel to obtain a product tracking feature vector.
  • the rectification tower parameter encoding module is used to arrange multiple first parameters of the rectification tower at the plurality of predetermined time points into a rectification tower parameter input matrix according to the time dimension and the parameter sample dimension and then pass it through the second parameter as a filter.
  • Convolutional neural network to obtain distillation column feature vector;
  • the absorption tower parameter encoding module is used to arrange multiple second parameters of the absorption tower at the plurality of predetermined time points into an absorption tower parameter input matrix according to the time dimension and the parameter sample dimension and then pass it through the third convolutional neural network as a filter. network to obtain the absorption tower eigenvector;
  • a parameter feature fusion module used to fuse the distillation tower feature vector and the absorption tower feature vector to obtain a collaborative feature vector
  • a responsiveness estimation module configured to calculate a responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector to obtain a classification feature vector
  • the equipment management result generation module is used to pass the classification feature vector through a classifier to obtain a classification result.
  • the classification result is used to indicate whether the equipment performance of the distillation tower and the absorption tower is normal.
  • the product tracking encoding module includes: a feature dynamic capture unit for using each layer of the first convolutional neural network in the forward direction of the layer.
  • the input data are respectively subjected to convolution processing, pooling processing and non-linear activation processing based on the three-dimensional convolution kernel to output a tracking feature map from the last layer of the first convolutional neural network, wherein,
  • the input of the first convolutional neural network is a plurality of liquid chromatograms of the hydrogen fluoride liquid before the first filtering at the plurality of predetermined time points; and, a global pooling unit is used to perform the tracking feature map on the Global mean pooling based on the feature matrix is used to obtain the tracking feature vector.
  • the distillation tower parameter encoding module includes: a first matrix construction unit, used to combine multiple parameters of the distillation tower at the plurality of predetermined time points. First parameters are arranged as multiple first parameter row vectors corresponding to the distillation tower according to the time dimension, and the first parameter row vectors are arranged as the parameter input matrix of the distillation tower according to the parameter sample dimension.
  • the last layer of the second convolutional neural network generates the distillation tower feature vector, wherein the input of the first layer of the second convolutional neural network is the distillation tower parameter input matrix.
  • the absorption tower parameter encoding module includes: a second matrix construction unit for converting multiple first parameters of the absorption tower at the plurality of predetermined time points. The two parameters are arranged into multiple second parameter row vectors corresponding to the absorption tower according to the time dimension, and the multiple second parameter row vectors are arranged into the absorption tower parameter input matrix according to the parameter sample dimension; second filtering Converter unit for performing convolution processing, pooling processing along the feature matrix and activation processing on the input data in the forward pass of the layer using each layer of the third convolutional neural network to be processed by the third convolution The last layer of the neural network generates the absorption tower feature vector, wherein the input of the first layer of the third convolutional neural network is the absorption tower parameter input matrix.
  • the parameter feature fusion module is used to calculate the position-weighted sum of the distillation tower feature vector and the absorption tower feature vector to obtain the Collaborative feature vectors.
  • the weighted weight of the distillation tower eigenvector is calculated according to the following formula, wherein the formula is:
  • w 1 represents the weighting coefficient of the rectification tower eigenvector when the rectification tower eigenvector and the absorption tower eigenvector are fused.
  • the weight of the absorption tower eigenvector is calculated according to the following formula, where the formula is:
  • w 2 represents the weighting coefficient of the absorption tower eigenvector when the rectification tower eigenvector and the absorption tower eigenvector are fused.
  • the responsiveness estimation module is further used to calculate the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector using the following formula to obtain The classification feature vector;
  • s 1 represents the product tracking feature vector
  • s 2 represents the collaboration feature vector
  • s 3 represents the classification feature vector
  • ⁇ represents the vector dot product
  • ⁇ -1 represents the reciprocal of the value of each position of the vector.
  • the equipment management result generation module is further used to: use the classifier to process the classification feature vector with the following formula to obtain the classification
  • the formula is: softmax ⁇ (W n ,B n ):...:(W 1 ,B 1 )
  • a control method of a production management control system for the preparation of electronic grade hydrofluoric acid includes:
  • the multiple first parameters include the bottom temperature of the rectification tower, and the rectification tower
  • the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower
  • the multiple first parameters of the distillation tower at the multiple predetermined time points are arranged into a distillation tower parameter input matrix according to the time dimension and the parameter sample dimension, and then passed through the second convolutional neural network as a filter to obtain the distillation tower.
  • Feature vector
  • the classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate whether the equipment performance of the distillation tower and the absorption tower is normal.
  • multiple liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points are obtained through a liquid chromatograph, It includes: using each layer of the first convolutional neural network to perform convolution processing, pooling processing and non-linear activation processing based on the three-dimensional convolution kernel on the input data during the forward transmission process of the layer to obtain the desired result from the input data.
  • the last layer of the first convolutional neural network outputs a tracking feature map, wherein the input of the first convolutional neural network is a plurality of liquid phases of the hydrogen fluoride liquid before the first filtration at the plurality of predetermined time points. a chromatogram; and performing global mean pooling based on a feature matrix on the tracking feature map to obtain the tracking feature vector.
  • a plurality of first parameters of the distillation tower at the plurality of predetermined time points are arranged into a distillation tower according to the time dimension and the parameter sample dimension.
  • the second convolutional neural network is used as a filter to obtain the distillation tower feature vector, which includes: arranging multiple first parameters of the distillation tower at the multiple predetermined time points according to the time dimension to correspond to Multiple first parameter row vectors of the distillation tower, and the multiple first parameter row vectors are arranged into the distillation tower parameter input matrix according to parameter sample dimensions; a first filter unit is used to use all the first parameter row vectors of the distillation tower.
  • Each layer of the second convolutional neural network performs convolution, pooling and activation along the feature matrix on the input data in the forward pass of the layer to be generated by the last layer of the second convolutional neural network.
  • the distillation tower feature vector wherein the input of the first layer of the second convolutional neural network is the distillation tower parameter input matrix.
  • a plurality of second parameters of the absorption tower at the plurality of predetermined time points are arranged according to the time dimension and the parameter sample dimension as the absorption tower parameter input
  • the matrix is then passed through the third convolutional neural network as a filter to obtain the absorption tower feature vector, which includes: arranging multiple second parameters of the absorption tower at the multiple predetermined time points according to the time dimension to correspond to the absorption tower multiple second parameter row vectors, and arrange the multiple second parameter row vectors into the absorption tower parameter input matrix according to parameter sample dimensions; use each layer of the third convolutional neural network to
  • the input data is convolved, pooled along the feature matrix, and activated in a pass to generate the absorption tower feature vector by the last layer of the third convolutional neural network, where the third convolution
  • the input of the first layer of the product neural network is the parameter input matrix of the absorption tower.
  • fusing the distillation tower feature vector and the absorption tower feature vector to obtain a synergistic feature vector includes: calculating the distillation tower feature The position-weighted sum of the vector and the absorption tower eigenvector is used to obtain the collaborative eigenvector.
  • the weighted weight of the distillation tower eigenvector is calculated according to the following formula, wherein the formula is:
  • w 1 represents the weighting coefficient of the rectification tower eigenvector when the rectification tower eigenvector and the absorption tower eigenvector are fused.
  • the weighted weight of the absorption tower eigenvector is calculated according to the following formula, wherein the formula is:
  • w 2 represents the weighting coefficient of the absorption tower eigenvector when the rectification tower eigenvector and the absorption tower eigenvector are fused.
  • calculating the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector to obtain the classification feature vector includes: calculating with the following formula Responsiveness estimation of the product tracking feature vector relative to the collaborative feature vector to obtain the classification feature vector;
  • s 1 represents the product tracking feature vector
  • s 2 represents the collaboration feature vector
  • s 3 represents the classification feature vector
  • ⁇ represents the vector dot product
  • ⁇ -1 represents the reciprocal of the value of each position of the vector.
  • passing the classification feature vector through a classifier to obtain a classification result includes: using the classifier to classify the classification feature vector according to the following formula Processing is performed to obtain the classification result, where the formula is: softmax ⁇ (W n ,B n ):...:(W 1 ,B 1 )
  • a computer-readable medium is provided with computer program instructions stored thereon, which when executed by a processor cause the processor to perform the above-described method for electronics. Control method of production management control system for the preparation of high-grade hydrofluoric acid.
  • the production management control system and its control method for the preparation of electronic grade hydrofluoric acid adopt monitoring technology based on artificial intelligence, and perform distillation during the preparation process of electronic grade hydrofluoric acid.
  • Various parameters of the tower and absorption tower are monitored to ensure the production yield of the final electronic grade hydrofluoric acid product, and during this process, the dynamic change characteristics of the product are also used to estimate the responsiveness to more accurately estimate the precision. Monitor whether the equipment performance of the distillation tower and the absorption tower is normal.
  • Figure 1 is an application scenario diagram of a production management control system for the preparation of electronic grade hydrofluoric acid according to an embodiment of the present application.
  • Figure 2 is a block diagram of a production management control system for the preparation of electronic grade hydrofluoric acid according to an embodiment of the present application.
  • Figure 3 is a block diagram of a product tracking coding module in a production management control system for electronic grade hydrofluoric acid preparation according to an embodiment of the present application.
  • Figure 4 is a block diagram of a distillation tower parameter encoding module in a production management control system for the preparation of electronic grade hydrofluoric acid according to an embodiment of the present application.
  • Figure 5 is a flow chart of a control method of a production management control system for electronic grade hydrofluoric acid preparation according to an embodiment of the present application.
  • Figure 6 is a schematic structural diagram of a control method of a production management control system for electronic grade hydrofluoric acid preparation according to an embodiment of the present application.
  • electronic grade hydrofluoric acid is a strong acidic cleaning agent and corrosive agent, mainly used in the production of very large-scale integrated circuits.
  • the main production method of electronic grade hydrofluoric acid is: first chemically pretreat industrial anhydrous hydrofluoric acid, then perform distillation, then cool the obtained hydrogen fluoride gas, absorb it with pure water, and finally filter and fill it. Since the presence of impurity arsenic has a serious impact on the performance of electronic devices, arsenic removal is a key issue in the purification process of hydrofluoric acid.
  • the commonly used method is to use an oxidant to convert trivalent arsenic impurities into high boiling point pentavalent arsenic. Compounds, commonly used oxidants include KMnO4, CrO3, persulfate, etc.
  • Patent 103991847 discloses a method for preparing electronic grade hydrofluoric acid, which uses hydrogen peroxide as an oxidant to convert trivalent arsenic impurities into high-boiling point pentavalent arsenic compounds without introducing additional impurities, thus producing high yield and good quality products. , low-cost electronic grade hydrofluoric acid products, and recover hydrogen fluoride gas in the exhaust gas.
  • an equipment production management control system for the preparation of electronic grade hydrofluoric acid is expected to provide online monitoring of the performance of each equipment in the production line.
  • the preparation method of the electronic grade hydrofluoric acid is as follows:
  • the inventor of the present application considered that if he wanted to ensure the production yield of the final electronic-grade hydrofluoric acid product, it was necessary to conduct online monitoring of the performance of each equipment for preparing electronic-grade hydrofluoric acid on the production line.
  • the inventor of the present application found that the main part to monitor the performance of each equipment is to monitor the rectification tower and absorption tower, and when monitoring the rectification tower and absorption tower, , using the liquid chromatogram of the hydrogen fluoride liquid before the first filtration as a reference for the responsiveness of the product can further improve the accuracy of judging the equipment performance of the distillation tower and absorption tower.
  • first, multiple first parameters of the rectification tower and multiple second parameters of the absorption tower at multiple predetermined time points are obtained through various sensors deployed in the rectification tower and the absorption tower.
  • the equipment parts that mainly need to be monitored are the temperature of the distillation tower still, the temperature of the tower body, the tower top temperature, the tower bottom pressure, the tower top pressure, the reflux temperature and reflux ratio in the rectification tower, and all The temperature of the absorption tower and the pressure of the absorption tower.
  • the plurality of first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, and the top temperature of the rectification tower.
  • the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower.
  • liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points are obtained through a liquid chromatograph as dynamic change responsiveness characteristics of the product. Then, the liquid chromatogram of the hydrogen fluoride liquid before the first filtration is processed using a convolutional neural network model that has excellent performance in extracting local implicit features of the image.
  • the multiple first parameters and the multiple second parameters are arranged into a distillation tower parameter input matrix and an absorption tower parameter input matrix according to the time dimension and parameter sample dimension respectively, and then passed through convolution as a filter.
  • the neural network is used to obtain the distillation tower eigenvector and the absorption tower eigenvector. In this way, the multiple items in the distillation tower parameter matrix and the absorption tower parameter matrix that integrate the time dimension and sample dimension can be mined respectively.
  • the coefficient performs explicit generalization of semantic reasoning information through bottom-up explicit generalization of the semantic concept corresponding to the feature value, and thereby performs bottom-up information reasoning on the feature semantics.
  • the dynamic change characteristics of the product over time are responsive characteristics related to the control of the rectification tower and the absorption tower, therefore, in order to more accurately Monitor and control the distillation tower and the absorption tower, and calculate the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector to obtain a classification feature vector.
  • the classification feature vector is then classified through a classifier to obtain a classification result indicating whether the equipment performance of the distillation tower and the absorption tower is normal.
  • this application proposes a production management control system for the preparation of electronic grade hydrofluoric acid, which includes: an equipment parameter acquisition module, used to obtain multiple first parameters and parameters of the distillation tower at multiple predetermined time points.
  • Multiple second parameters of the absorption tower wherein the multiple first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, and the tower top temperature of the rectification tower, so The bottom pressure of the rectification tower, the top pressure of the rectification tower, the reflux temperature and reflux ratio in the tower of the rectification tower, and the plurality of second parameters include the temperature of the absorption tower and the The pressure of the absorption tower; a product status acquisition module, used to obtain multiple liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points through a liquid chromatograph; a product tracking encoding module, used to Pass the multiple liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points through the first convolutional neural network
  • Figure 1 illustrates an application scenario diagram of a production management control system for the preparation of electronic grade hydrofluoric acid according to an embodiment of the present application.
  • the distillation tower for example, R as shown in Figure 1
  • the absorption tower for example, A as shown in Figure 1
  • Sensors for example, T1-Tn as illustrated in Figure 1 acquire multiple first parameters of the distillation tower and multiple second parameters of the absorption tower at multiple predetermined time points, and use liquid chromatography to
  • the instrument for example, L as shown in Figure 1) acquires multiple liquid chromatograms of the hydrogen fluoride liquid (for example, H as shown in Figure 1) before the first filtration at the plurality of predetermined time points.
  • the plurality of first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, the top temperature of the rectification tower, the bottom pressure of the rectification tower, The top pressure of the rectification tower, the reflux temperature and reflux ratio in the rectification tower, and the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower.
  • a plurality of liquid chromatograms are input into a server deployed with a production management control algorithm for the preparation of electronic grade hydrofluoric acid (for example, the cloud server S illustrated in Figure 1), wherein the server can be used for
  • the production management control algorithm for the preparation of electronic grade hydrofluoric acid controls a plurality of first parameters of the distillation tower and a plurality of second parameters of the absorption tower at the plurality of predetermined time points and the plurality of predetermined time points.
  • Multiple liquid chromatograms of hydrogen fluoride liquid before first filtration are processed to generate classification results indicating whether the equipment performance of the distillation tower and absorption tower is normal.
  • Figure 2 illustrates a block diagram of a production management control system for electronic grade hydrofluoric acid preparation according to an embodiment of the present application.
  • the production management control system 200 for the preparation of electronic grade hydrofluoric acid according to the embodiment of the present application includes: an equipment parameter acquisition module 210, used to obtain multiple parameters of the distillation tower at multiple predetermined time points.
  • the first parameters and multiple second parameters of the absorption tower, wherein the multiple first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, the tower temperature of the rectification tower.
  • the top temperature, the bottom pressure of the rectification tower, the top pressure of the rectification tower, the reflux temperature and reflux ratio of the rectification tower, and the plurality of second parameters include the absorption tower The temperature and the pressure of the absorption tower; the product status acquisition module 220 is used to obtain multiple liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points through a liquid chromatograph; product tracking The encoding module 230 is configured to pass the multiple liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points through a first convolutional neural network using a three-dimensional convolution kernel to obtain a product tracking feature vector; The distillation tower parameter encoding module 240 is used to arrange the multiple first parameters of the distillation tower at the plurality of predetermined time points into a distillation tower parameter input matrix according to the time dimension and the parameter sample dimension and then pass it through the first parameter as a filter.
  • the absorption tower parameter encoding module 250 is used to arrange multiple second parameters of the absorption tower at the multiple predetermined time points into absorption towers according to the time dimension and parameter sample dimension.
  • the absorption tower feature vector is obtained through the third convolutional neural network as a filter;
  • the parameter feature fusion module 260 is used to fuse the distillation tower feature vector and the absorption tower feature vector to obtain a collaborative feature vector ;
  • Responsiveness estimation module 270 used to calculate the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector to obtain a classification feature vector;
  • equipment management result generation module 280 used to convert the classification feature vector into Classification results are obtained through the classifier, and the classification results are used to indicate whether the equipment performance of the distillation tower and absorption tower is normal.
  • the equipment parameter acquisition module 210, the product status acquisition module 220 and the product tracking encoding module 230 are used to obtain multiple first parameters of the distillation tower at multiple predetermined time points.
  • One parameter and a plurality of second parameters of the absorption tower wherein the plurality of first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, the top of the tower of the rectification tower temperature, the bottom pressure of the rectification tower, the top pressure of the rectification tower, the reflux temperature and reflux ratio of the rectification tower, and the plurality of second parameters include the reflux temperature of the absorption tower.
  • the main part of monitoring the performance of each equipment is to monitor the distillation tower and the absorption tower.
  • the liquid chromatogram of the hydrogen fluoride liquid before the first filtration is used as a reference for the responsiveness of the product, which can further improve the monitoring of the rectification tower and the absorption tower. Accuracy of equipment performance judgment of the absorption tower.
  • the equipment parts that mainly need to be monitored are the temperature of the distillation tower still, the temperature of the tower body, the tower top temperature, the tower bottom pressure, the tower top pressure, the reflux temperature and reflux ratio in the rectification tower, and all The temperature of the absorption tower and the pressure of the absorption tower. Therefore, the plurality of first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, and the top temperature of the rectification tower. , the bottom pressure of the rectification tower, the top pressure of the rectification tower, the reflux temperature and reflux ratio in the rectification tower, and the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower.
  • liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points are obtained through a liquid chromatograph as dynamic change responsiveness characteristics of the product. Then, the liquid chromatogram of the hydrogen fluoride liquid before the first filtration is processed using a convolutional neural network model that has excellent performance in extracting local implicit features of the image.
  • a first convolutional neural network with a three-dimensional convolution kernel is further used to perform multiple analysis of the hydrogen fluoride liquid before the first filtering at the multiple predetermined time points.
  • the liquid chromatogram is processed to obtain product tracking feature vectors.
  • the product tracking encoding module includes: first, using each layer of the first convolutional neural network to separately perform the input data based on the forward transmission process of the layer.
  • the convolution processing, pooling processing and nonlinear activation processing of the three-dimensional convolution kernel are used to output the tracking feature map from the last layer of the first convolutional neural network, wherein the input of the first convolutional neural network is A plurality of liquid chromatograms of the hydrogen fluoride liquid before first filtration at the plurality of predetermined time points.
  • global mean pooling based on the feature matrix is performed on the tracking feature map to obtain the tracking feature vector.
  • the pooling operation can be used to reduce feature dimensionality, alleviate the risk of over-fitting, and reduce the excessive sensitivity of the convolutional layer to detection information, and the global mean pooling can retain the important information of each tracking feature map. , to highlight the most important part of the response in the tracking feature map.
  • the product tracking encoding module 230 includes: a feature dynamic capture unit 231, which is used to use each layer of the first convolutional neural network to respectively perform based on input data during the forward transmission process of the layer.
  • the convolution processing, pooling processing and nonlinear activation processing of the three-dimensional convolution kernel are used to output the tracking feature map from the last layer of the first convolutional neural network, wherein the input of the first convolutional neural network A plurality of liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the plurality of predetermined time points; and a global pooling unit 232 for performing global mean pooling based on a feature matrix on the tracking feature map. to obtain the tracking feature vector.
  • the rectification tower parameter encoding module 240 and the absorption tower parameter encoding module 250 are used to encode multiple first parameters of the rectification tower at multiple predetermined time points according to
  • the time dimension and the parameter sample dimension are arranged into a distillation tower parameter input matrix and then passed through the second convolutional neural network as a filter to obtain the distillation tower feature vector, and the multi-term parameters of the absorption tower at the multiple predetermined time points are
  • the two parameters are arranged according to the time dimension and parameter sample dimension into the absorption tower parameter input matrix, and then passed through the third convolutional neural network as a filter to obtain the absorption tower feature vector.
  • the multiple first parameters and the multiple second parameters are arranged into a distillation tower parameter input matrix and an absorption tower parameter according to the time dimension and the parameter sample dimension respectively.
  • the convolutional neural network as a filter is used to obtain the distillation tower eigenvector and the absorption tower eigenvector.
  • the distillation tower parameter matrix and the rectification tower parameter matrix that integrate the time dimension and sample dimension can be mined respectively.
  • the distillation tower parameter encoding module includes: first, arranging multiple first parameters of the distillation tower at the multiple predetermined time points according to the time dimension to correspond to the Multiple first parameter row vectors of the distillation tower, and arrange the multiple first parameter row vectors into the distillation tower parameter input matrix according to parameter sample dimensions; then, use the second convolutional neural network
  • Each layer of the layer performs convolution processing, pooling processing along the feature matrix and activation processing on the input data in the forward pass of the layer to generate the distillation column feature vector by the last layer of the second convolutional neural network , wherein the input of the first layer of the second convolutional neural network is the distillation tower parameter input matrix.
  • the absorption tower parameter encoding module includes: first, arranging multiple second parameters of the absorption tower at the multiple predetermined time points according to the time dimension to correspond to the absorption tower parameters. Multiple second parameter row vectors of the tower, and arrange the multiple second parameter row vectors into the absorption tower parameter input matrix according to parameter sample dimensions; then, use each layer of the third convolutional neural network to In the forward pass of the layer, the input data is convolved, pooled along the feature matrix, and activated to generate the absorption tower feature vector by the last layer of the third convolutional neural network, wherein, The input of the first layer of the third convolutional neural network is the absorption tower parameter input matrix.
  • Figure 4 illustrates a block diagram of a distillation column parameter encoding module in a production management control system for electronic grade hydrofluoric acid preparation according to an embodiment of the present application.
  • the distillation tower parameter encoding module 240 includes: a first matrix construction unit 241 for arranging multiple first parameters of the distillation tower at multiple predetermined time points according to the time dimension as Corresponding to the multiple first parameter row vectors of the rectification tower, and arranging the multiple first parameter row vectors according to parameter sample dimensions into the rectification tower parameter input matrix; the first filter unit 242 uses Each layer of the second convolutional neural network is used to perform convolution processing, pooling processing and activation processing along the feature matrix on the input data in the forward pass of the layer to obtain the final layer of the second convolutional neural network.
  • One layer generates the distillation column feature vector, wherein the input of the first layer of the second convolutional neural network is the distillation column parameter input matrix.
  • the parameter feature fusion module 260 is used to fuse the distillation tower feature vector and the absorption tower feature vector to obtain a collaborative feature vector. It should be understood that when using the convolutional neural network as a filter to perform feature extraction on the data matrix in the time and data sample dimensions to obtain the distillation tower feature vector V 1 and the absorption tower feature vector V 2 , because the filter of the convolutional neural network is for feature extraction of local feature dimensions, but because when the feature vectors are fused, it is based on the rectification tower feature vector V 1 and the absorption tower feature vector V 2 The global feature distribution is fused.
  • the collaborative feature vector can be obtained by calculating the position-weighted sum of the distillation tower feature vector and the absorption tower feature vector.
  • the parameter feature fusion module is further used to calculate the weight of the distillation column feature vector according to the following formula, where the formula is:
  • w 1 represents the weighting coefficient of the rectification tower eigenvector when the rectification tower eigenvector and the absorption tower eigenvector are fused.
  • the weight of the absorption tower feature vector is calculated by the following formula, where the formula is:
  • w 2 represents the weighting coefficient of the absorption tower eigenvector when the rectification tower eigenvector and the absorption tower eigenvector are fused. It should be understood that the coefficient performs explicit generalization of semantic reasoning information through bottom-up explicit generalization of the semantic concept corresponding to the feature value, and thereby performs bottom-up generalization of the feature semantics.
  • Informational reasoning is used to obtain the information plasticity of the spatial complexity of the high-dimensional manifold corresponding to the feature in the high-dimensional semantic space, thereby promoting the distillation tower feature vector V 1 and the absorption tower feature vector V 2 to be based on global features Distributed features are fused to obtain the collaborative feature vector, which can improve the accuracy of subsequent classification.
  • the responsiveness estimation module 270 and the device management result generation module 280 are used to calculate the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector to obtain classification feature vector, and pass the classification feature vector through a classifier to obtain a classification result, which is used to indicate whether the equipment performance of the distillation tower and absorption tower is normal. It should be understood that considering that in the process of preparing the electronic grade hydrofluoric acid, the dynamic change characteristics of the product over time are responsive characteristics related to the control of the rectification tower and the absorption tower, therefore, In order to monitor and control the distillation tower and the absorption tower more accurately, the responsiveness estimate of the product tracking feature vector relative to the collaboration feature vector is calculated to obtain a classification feature vector.
  • the classification feature vector is then classified through a classifier to obtain a classification result indicating whether the equipment performance of the distillation tower and the absorption tower is normal.
  • the classifier is used to process the classification feature vector with the following formula to obtain the classification result, where the formula is: softmax ⁇ (W n ,B n ):... :(W 1 ,B 1 )
  • the responsiveness estimation module is further configured to calculate the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector using the following formula to obtain the classification feature vector;
  • s 1 represents the product tracking feature vector
  • s 2 represents the collaboration feature vector
  • s 3 represents the classification feature vector
  • ⁇ represents the vector dot product
  • ⁇ -1 represents the reciprocal of the value of each position of the vector.
  • the production management control system 200 for the preparation of electronic grade hydrofluoric acid is clarified based on the embodiment of the present application, which uses artificial intelligence-based monitoring technology to control the rectification tower in the preparation process of electronic grade hydrofluoric acid. and monitoring various parameters of the absorption tower to ensure the production yield of the final electronic grade hydrofluoric acid product, and in this process, the dynamic change characteristics of the product are also used to estimate the responsiveness to more accurately estimate the distillation Monitor whether the equipment performance of the tower and the absorption tower is normal.
  • the production management control system 200 for the preparation of electronic grade hydrofluoric acid according to the embodiment of the present application can be implemented in various terminal devices, such as a server for the production management control algorithm of the preparation of electronic grade hydrofluoric acid, etc. .
  • the production management control system 200 for electronic grade hydrofluoric acid preparation according to the embodiment of the present application can be integrated into the terminal device as a software module and/or hardware module.
  • the production management control system 200 for the preparation of electronic grade hydrofluoric acid can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the user
  • the production management control system 200 for the preparation of electronic grade hydrofluoric acid can also be one of the many hardware modules of the terminal equipment.
  • the production management control system 200 for the preparation of electronic grade hydrofluoric acid and the terminal device may also be separate devices, and the production management control system 200 for the preparation of electronic grade hydrofluoric acid 200 can be connected to the terminal device through a wired and/or wireless network, and transmit interactive information according to an agreed data format.
  • Figure 5 illustrates a flow chart of a control method of a production management control system for electronic grade hydrofluoric acid preparation.
  • the control method of the production management control system for the preparation of electronic grade hydrofluoric acid according to the embodiment of the present application includes the step: S110, obtaining multiple first parameters of the distillation tower at multiple predetermined time points.
  • the plurality of first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, and the tower top temperature of the rectification tower
  • the bottom pressure of the rectification tower, the top pressure of the rectification tower, the reflux temperature and reflux ratio in the tower of the rectification tower, and the multiple second parameters include the temperature and reflux ratio of the absorption tower.
  • the multiple liquid chromatograms of the hydrogen fluoride liquid before the first filtration are passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the product tracking feature vector; S140, combine the multiple liquid chromatograms of the rectification tower at the multiple predetermined time points.
  • the multiple first parameters are arranged into a distillation tower parameter input matrix according to the time dimension and the parameter sample dimension, and then passed through the second convolutional neural network as a filter to obtain the distillation tower feature vector; S150, combine the multiple predetermined time points
  • the multiple second parameters of the absorption tower are arranged according to the time dimension and parameter sample dimension into the absorption tower parameter input matrix and then passed through the third convolutional neural network as a filter to obtain the absorption tower feature vector;
  • S160 fuse the distillation tower feature vector and the absorption tower feature vector to obtain a collaborative feature vector;
  • S170 calculate the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector to obtain a classification feature vector; and, S180, combine the classification feature
  • the vector passes through the classifier to obtain a classification result, which is used to indicate whether the equipment performance of the distillation tower and absorption tower is normal.
  • FIG. 6 illustrates an architectural schematic diagram of a control method of a production management control system for electronic grade hydrofluoric acid preparation according to an embodiment of the present application.
  • the obtained hydrogen fluoride before the first filtration at the plurality of predetermined time points is Multiple liquid chromatograms of the liquid (eg, P as illustrated in Figure 6) are passed through a first convolutional neural network (eg, CNN1 as illustrated in Figure 6) using a three-dimensional convolution kernel to obtain product tracking features vector (for example, VF as shown in Figure 6); then, the multiple first parameters of the distillation tower at the plurality of predetermined time points (for example, Q1 as shown in Figure 6) are obtained according to time Dimensions and parameters
  • the sample dimensions are arranged as the distillation column parameter input matrix (for example, M1 as shown in Figure 6) and then passed through the second convolutional neural network as a filter (for example, CNN2 as shown in Figure
  • steps S110, S120 and S130 multiple first parameters of the distillation tower and multiple second parameters of the absorption tower at multiple predetermined time points are obtained, wherein the multiple first parameters are Including the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, the top temperature of the rectification tower, the bottom pressure of the rectification tower, and the top pressure of the rectification tower , the reflux temperature and reflux ratio in the distillation tower, the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower, and the plurality of predetermined parameters are obtained through a liquid chromatograph. Multiple liquid chromatograms of hydrogen fluoride liquid before first filtration at time points.
  • the main part of monitoring the performance of each equipment is to monitor the distillation tower and the absorption tower.
  • the liquid chromatogram of the hydrogen fluoride liquid before the first filtration is used as a reference for the responsiveness of the product, which can further improve the monitoring of the rectification tower and the absorption tower. Accuracy of equipment performance judgment of the absorption tower.
  • the equipment parts that mainly need to be monitored are the temperature of the distillation tower still, the temperature of the tower body, the tower top temperature, the tower bottom pressure, the tower top pressure, the reflux temperature and reflux ratio in the rectification tower, and all The temperature of the absorption tower and the pressure of the absorption tower. Therefore, the plurality of first parameters include the bottom temperature of the rectification tower, the tower body temperature of the rectification tower, and the top temperature of the rectification tower. , the bottom pressure of the rectification tower, the top pressure of the rectification tower, the reflux temperature and reflux ratio in the rectification tower, and the plurality of second parameters include the temperature of the absorption tower and the pressure of the absorption tower.
  • liquid chromatograms of the hydrogen fluoride liquid before the first filtration at the multiple predetermined time points are obtained through a liquid chromatograph as dynamic change responsiveness characteristics of the product. Then, the liquid chromatogram of the hydrogen fluoride liquid before the first filtration is processed using a convolutional neural network model that has excellent performance in extracting local implicit features of the image.
  • a first convolutional neural network with a three-dimensional convolution kernel is further used to perform multiple analysis of the hydrogen fluoride liquid before the first filtering at the multiple predetermined time points.
  • the liquid chromatogram is processed to obtain product tracking feature vectors.
  • steps S140 and S150 multiple first parameters of the distillation tower at the plurality of predetermined time points are arranged into a distillation tower parameter input matrix according to the time dimension and the parameter sample dimension and then passed as a filter.
  • the second convolutional neural network is used to obtain the distillation tower feature vector
  • the multiple second parameters of the absorption tower at the multiple predetermined time points are arranged into an absorption tower parameter input matrix according to the time dimension and parameter sample dimension, and then passed as Filter the third convolutional neural network to obtain the absorption tower feature vector. That is to say, in the technical solution of the present application, the multiple first parameters and the multiple second parameters are arranged into a distillation tower parameter input matrix and an absorption tower parameter according to the time dimension and the parameter sample dimension respectively.
  • the convolutional neural network as a filter is used to obtain the distillation tower eigenvector and the absorption tower eigenvector.
  • the distillation tower parameter matrix and the rectification tower parameter matrix that integrate the time dimension and sample dimension can be mined respectively.
  • step S160 the distillation tower feature vector and the absorption tower feature vector are fused to obtain a collaborative feature vector.
  • the convolutional neural network as a filter is used to perform feature extraction on the data matrix in the time and data sample dimensions to obtain the distillation tower feature vector V 1 and the absorption tower feature.
  • the filter of the convolutional neural network is based on the feature extraction of the local feature dimension, but when the feature vector is fused, it is based on the distillation tower feature vector V 1 and the absorption tower feature. The global feature distribution of the vector V 2 is fused.
  • the collaborative feature vector can be obtained by calculating the position-weighted sum of the distillation tower feature vector and the absorption tower feature vector.
  • the responsiveness estimate of the product tracking feature vector relative to the collaborative feature vector is calculated to obtain a classification feature vector
  • the classification feature vector is passed through a classifier to obtain a classification result
  • the classification results are used to indicate whether the equipment performance of the distillation tower and absorption tower is normal. It should be understood that considering that in the process of preparing the electronic grade hydrofluoric acid, the dynamic change characteristics of the product over time are responsive characteristics related to the control of the rectification tower and the absorption tower, therefore, In order to monitor and control the distillation tower and the absorption tower more accurately, the responsiveness estimate of the product tracking feature vector relative to the collaboration feature vector is calculated to obtain a classification feature vector.
  • the classification feature vector is then classified through a classifier to obtain a classification result indicating whether the equipment performance of the distillation tower and the absorption tower is normal.
  • the classifier is used to process the classification feature vector with the following formula to obtain the classification result, where the formula is: softmax ⁇ (W n ,B n ):... :(W 1 ,B 1 )
  • the control method of the production management control system for the preparation of electronic grade hydrofluoric acid is clarified, which uses artificial intelligence-based monitoring technology to accurately monitor the production process of electronic grade hydrofluoric acid.
  • Various parameters of the distillation tower and absorption tower are monitored to ensure the production yield of the final electronic grade hydrofluoric acid product, and during this process, the dynamic change characteristics of the product are also used to estimate the responsiveness to more accurately estimate the described Monitor whether the equipment performance of the distillation tower and the absorption tower is normal.
  • embodiments of the present application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to execute the “exemplary method” described above in this specification.
  • the steps in the functions of the control method of the production management control system for electronic grade hydrofluoric acid preparation according to various embodiments of the present application are described in the section.
  • the computer program product can be used to write program codes for performing the operations of the embodiments of the present application in any combination of one or more programming languages, including object-oriented programming languages, such as Java, C++, etc. , also includes conventional procedural programming languages, such as the "C" language or similar programming languages.
  • the program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server execute on.
  • embodiments of the present application may also be a computer-readable storage medium having computer program instructions stored thereon.
  • the computer program instructions When the computer program instructions are run by a processor, the computer program instructions cause the processor to execute the above-mentioned "example method" part of this specification. The steps in the control method of the production management control system for the preparation of electronic grade hydrofluoric acid are described in.
  • the computer-readable storage medium may be any combination of one or more readable media.
  • the readable medium may be a readable signal medium or a readable storage medium.
  • Readable storage media may include, for example, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: electrical connection with one or more conductors, portable disk, hard disk, random access memory (RAM), read only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
  • each component or each step can be decomposed and/or recombined. These decompositions and/or recombinations shall be considered equivalent versions of this application.

Landscapes

  • Chemical & Material Sciences (AREA)
  • Organic Chemistry (AREA)
  • Inorganic Chemistry (AREA)
  • Engineering & Computer Science (AREA)
  • Manufacturing & Machinery (AREA)
  • General Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Organic Low-Molecular-Weight Compounds And Preparation Thereof (AREA)
  • Feedback Control In General (AREA)

Abstract

一种用于电子级氢氟酸制备的生产管理控制系统及其控制方法，其采用基于人工智能的监测技术，通过对于电子级氢氟酸制备过程中精馏塔（R）和吸收塔（A）的各个参数进行监测来保证最终电子级氢氟酸产品的制备良率，并且在此过程中，还使用产物的动态变化特征来进行响应性估计，以更准确地对所述精馏塔（R）和所述吸收塔（A）的设备性能是否正常进行监测。

Description

用于电子级氢氟酸制备的生产管理控制系统及其控制方法 技术领域
本发明涉及设备智能监测的领域，且更为具体地，涉及一种用于电子级氢氟酸制备的生产管理控制系统及其控制方法。
背景技术
电子级氢氟酸为强酸性清洗剂、腐蚀剂，主要用于超大规模集成电路生产。目前，电子级氢氟酸的主要生产方法为：首先将工业无水氢氟酸进行化学预处理，接着进行精馏，再将得到的氟化氢气体冷却，用纯水吸收，最后过滤，灌装。由于杂质砷的存在对电子器件的性能有严重影响，因此砷的脱除是氢氟酸提纯过程中的关键问题，通常采用的方法是使用氧化剂把三价砷杂质转化成高沸点的五价砷化合物，常用的氧化剂有KMnO4、CrO3、过硫酸盐等。
专利103991847揭露了一种电子级氢氟酸的制备方法，其使用过氧化氢作为氧化剂把三价砷杂质转化成高沸点的五价砷化合物，且不引入额外杂质，制备出产量高、品质好、成本低的电子级氢氟酸产品，并对尾气中的氟化氢气体进行回收。
但在实际制备过程中发现，虽然通过专利103991847所揭露的技术方案能够制得纯度较高的电子级氢氟酸产品，但是由于制备电子级氢氟酸的产线包括多个设备，一旦其中某个设备出现故障或者某个设备的性能下降，都会影响最终电子级氢氟酸产品的制备良率。
因此，期待一种用于电子级氢氟酸制备的设备生产管理控制系统，以对产线中各个设备的性能进行在线监测。
发明内容
为了解决上述技术问题，提出了本申请。本申请的实施例提供了一种用于电子级氢氟酸制备的生产管理控制系统及其控制方法，其采用基于人工智能的监测技术，通过对于电子级氢氟酸制备过程中精馏塔和吸收塔的各个参数进行监测来保证最终电子级氢氟酸产品的制备良率，并且在此过程中，还使用产物的动态变化特征来进行响应性估计，以更准确地对所述精馏塔和所 述吸收塔的设备性能是否正常进行监测。
根据本申请的一个方面，提供了一种用于电子级氢氟酸制备的生产管理控制系统，其包括：
设备参数采集模块，用于获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；
产物状态采集模块，用于通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；
产物跟踪编码模块，用于将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；
精馏塔参数编码模块，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量；
吸收塔参数编码模块，用于将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量；
参数特征融合模块，用于融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；
响应性估计模块，用于计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及
设备管理结果生成模块，用于将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述产物跟踪编码模块，包括：特征动态捕捉单元，用于使用所述第一卷积神经网络的各层在层的正向传递过程中对输入数据分别进行基于所述三维卷积核的卷积处理、池化处理和非线性激活处理以由所述第一卷积神经网络的最后一层输出跟踪特征图，其中，所述第一卷积神经网络的输入为所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；以及，全局池化单元，用于 对所述跟踪特征图进行基于特征矩阵的全局均值池化以得到所述跟踪特征向量。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述精馏塔参数编码模块，包括：第一矩阵构造单元，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度排列为对应于所述精馏塔的多项第一参数行向量，并将所述多项第一参数行向量按照参数样本维度排列为所述精馏塔参数输入矩阵；第一过滤器单元，用于使用所述第二卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第二卷积神经网络的最后一层生成所述精馏塔特征向量，其中，所述第二卷积神经网络的第一层的输入为所述精馏塔参数输入矩阵。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述吸收塔参数编码模块，包括：第二矩阵构造单元，用于将所述多个预定时间点的吸收塔的多项第二参数按照时间维度排列为对应于所述吸收塔的多项第二参数行向量，并将所述多项第二参数行向量按照参数样本维度排列为所述吸收塔参数输入矩阵；第二过滤器单元，用于使用所述第三卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第三卷积神经网络的最后一层生成所述吸收塔特征向量，其中，所述第三卷积神经网络的第一层的输入为所述吸收塔参数输入矩阵。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述参数特征融合模块，用于计算所述精馏塔特征向量和所述吸收塔特征向量的按位置加权和以得到所述协同特征向量。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述精馏塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
Figure PCTCN2022116539-appb-000001
其中，
Figure PCTCN2022116539-appb-000002
表示所述精馏塔特征向量的各个位置的特征值，w 1表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述精馏塔特征向量的加权系数。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述吸收塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
Figure PCTCN2022116539-appb-000003
其中，
Figure PCTCN2022116539-appb-000004
表示所述吸收塔特征向量的各个位置的特征值，w 2表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述吸收塔特征向量的加权系数。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述响应性估计模块，进一步用于以如下公式计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到所述分类特征向量；
其中，所述公式为：
s 3＝s 2⊙s 1 ⊙-1
其中s 1表示所述产物跟踪特征向量，s 2表示所述协同特征向量，s 3表示所述分类特征向量，⊙表示向量点乘，⊙-1表示对向量的每个位置的值取倒数。
在上述用于电子级氢氟酸制备的生产管理控制系统中，所述设备管理结果生成模块，进一步用于：使用所述分类器以如下公式对所述分类特征向量进行处理以获得所述分类结果，其中，所述公式为：softmax{(W n,B n):…:(W 1,B 1)|X}，其中，W 1到W n为权重矩阵，B 1到B n为偏置向量，X为所述分类特征向量。
根据本申请的另一方面，一种用于电子级氢氟酸制备的生产管理控制系统的控制方法，其包括：
获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；
通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；
将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；
将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网 络以得到精馏塔特征向量；
将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量；
融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；
计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及
将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图，包括：使用所述第一卷积神经网络的各层在层的正向传递过程中对输入数据分别进行基于所述三维卷积核的卷积处理、池化处理和非线性激活处理以由所述第一卷积神经网络的最后一层输出跟踪特征图，其中，所述第一卷积神经网络的输入为所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；以及，对所述跟踪特征图进行基于特征矩阵的全局均值池化以得到所述跟踪特征向量。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量，包括：将所述多个预定时间点的精馏塔的多项第一参数按照时间维度排列为对应于所述精馏塔的多项第一参数行向量，并将所述多项第一参数行向量按照参数样本维度排列为所述精馏塔参数输入矩阵；第一过滤器单元，用于使用所述第二卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第二卷积神经网络的最后一层生成所述精馏塔特征向量，其中，所述第二卷积神经网络的第一层的输入为所述精馏塔参数输入矩阵。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量，包括：将所述多个预定时间点的吸收塔的多项第二参数按 照时间维度排列为对应于所述吸收塔的多项第二参数行向量，并将所述多项第二参数行向量按照参数样本维度排列为所述吸收塔参数输入矩阵；使用所述第三卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第三卷积神经网络的最后一层生成所述吸收塔特征向量，其中，所述第三卷积神经网络的第一层的输入为所述吸收塔参数输入矩阵。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量，包括：计算所述精馏塔特征向量和所述吸收塔特征向量的按位置加权和以得到所述协同特征向量。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，所述精馏塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
Figure PCTCN2022116539-appb-000005
其中，
Figure PCTCN2022116539-appb-000006
表示所述精馏塔特征向量的各个位置的特征值，w 1表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述精馏塔特征向量的加权系数。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，所述吸收塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
Figure PCTCN2022116539-appb-000007
其中，
Figure PCTCN2022116539-appb-000008
表示所述吸收塔特征向量的各个位置的特征值，w 2表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述吸收塔特征向量的加权系数。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量，包括：以如下公式计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到所述分类特征向量；
其中，所述公式为：
s 3＝s 2⊙s 1 ⊙-1
其中s 1表示所述产物跟踪特征向量，s 2表示所述协同特征向量，s 3表示 所述分类特征向量，⊙表示向量点乘，⊙-1表示对向量的每个位置的值取倒数。
在上述用于电子级氢氟酸制备的生产管理控制系统的控制方法中，将所述分类特征向量通过分类器以得到分类结果，包括：使用所述分类器以如下公式对所述分类特征向量进行处理以获得所述分类结果，其中，所述公式为：softmax{(W n,B n):…:(W 1,B 1)|X}，其中，W 1到W n为权重矩阵，B 1到B n为偏置向量，X为所述分类特征向量。
根据本申请的再又一方面，提供了一种计算机可读介质，其上存储有计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行如上所述的用于电子级氢氟酸制备的生产管理控制系统的控制方法。
与现有技术相比，本申请提供的用于电子级氢氟酸制备的生产管理控制系统及其控制方法，其采用基于人工智能的监测技术，通过对于电子级氢氟酸制备过程中精馏塔和吸收塔的各个参数进行监测来保证最终电子级氢氟酸产品的制备良率，并且在此过程中，还使用产物的动态变化特征来进行响应性估计，以更准确地对所述精馏塔和所述吸收塔的设备性能是否正常进行监测。
附图说明
通过结合附图对本申请实施例进行更详细的描述，本申请的上述以及其他目的、特征和优势将变得更加明显。附图用来提供对本申请实施例的进一步理解，并且构成说明书的一部分，与本申请实施例一起用于解释本申请，并不构成对本申请的限制。在附图中，相同的参考标号通常代表相同部件或步骤。
图1为根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的应用场景图。
图2为根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的框图。
图3为根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统中产物跟踪编码模块的框图。
图4为根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统中精馏塔参数编码模块的框图。
图5为根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的控制方法的流程图。
图6为根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的控制方法的架构示意图。
具体实施方式
下面，将参考附图详细地描述根据本申请的示例实施例。显然，所描述的实施例仅仅是本申请的一部分实施例，而不是本申请的全部实施例，应理解，本申请不受这里描述的示例实施例的限制。
场景概述
如前所述，电子级氢氟酸为强酸性清洗剂、腐蚀剂，主要用于超大规模集成电路生产。目前，电子级氢氟酸的主要生产方法为：首先将工业无水氢氟酸进行化学预处理，接着进行精馏，再将得到的氟化氢气体冷却，用纯水吸收，最后过滤，灌装。由于杂质砷的存在对电子器件的性能有严重影响，因此砷的脱除是氢氟酸提纯过程中的关键问题，通常采用的方法是使用氧化剂把三价砷杂质转化成高沸点的五价砷化合物，常用的氧化剂有KMnO4、CrO3、过硫酸盐等。
专利103991847揭露了一种电子级氢氟酸的制备方法，其使用过氧化氢作为氧化剂把三价砷杂质转化成高沸点的五价砷化合物，且不引入额外杂质，制备出产量高、品质好、成本低的电子级氢氟酸产品，并对尾气中的氟化氢气体进行回收。
但在实际制备过程中发现，虽然通过专利103991847所揭露的技术方案能够制得纯度较高的电子级氢氟酸产品，但是由于制备电子级氢氟酸的产线包括多个设备，一旦其中某个设备出现故障或者某个设备的性能下降，都会影响最终电子级氢氟酸产品的制备良率。
因此，期待一种用于电子级氢氟酸制备的设备生产管理控制系统，以对产线中各个设备的性能进行在线监测。
相应地，所述电子级氢氟酸的制备方法如下：
(1)将工业无水氟化氢液体和纯水通入精馏塔中，形成第一浓度的氢氟酸；在第一浓度的氢氟酸中，加入过氧化氢溶液，进行氧化处理，氧化其 中的砷、硅杂质；接着在精馏塔中进行精馏，杂质留在精馏塔的塔釜，在精馏塔塔顶得到纯化的氟化氢气体；
(2)将纯化的氟化氢气体通入吸收塔中，进行冷凝处理，得到氟化氢液体；
(3)将步骤(2)中得到的氟化氢液体进行第一次过滤，以除去大颗粒杂质；
(4)将第一次过滤后除去大颗粒杂质的氟化氢液体从成品中间槽底部通入成品中间槽，在成品中间槽中用纯水吸收，得到第二浓度的氢氟酸；
(5)将第二浓度的氢氟酸进行第二次过滤，以除去小颗粒杂质，然后进行无尘罐装，得到电子级氢氟酸产品；
(6)将制备电子级氢氟酸过程中产生的尾气用纯水吸收，以制成工业级氢氟酸。
基于此，本申请发明人考虑到若想保证最终电子级氢氟酸产品的制备良率，就需要对产线上制备电子级氢氟酸的各个设备的性能进行在线监测，而在所述电子级氢氟酸制备的方法中，本申请发明人发现对于所述各个设备的性能进行监测的主要部分是对于精馏塔和吸收塔进行监测，并且，在对于精馏塔和吸收塔进行监测时，将第一次过滤前的氟化氢液体的液相色谱图来作为产物的响应性参照，能够进一步提高对于精馏塔和吸收塔的设备性能判断的准确性。
具体地，在本申请的技术方案中，首先，通过部署于精馏塔和吸收塔的各个传感器获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数。这里，考虑到其主要需要监控的设备部分为所述精馏塔塔釜的温度、塔身的温度、塔顶温度、塔釜压力、塔顶压力、精馏塔中回流温度和回流比以及所述吸收塔的温度和吸收塔的压力，因此，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力。
同时，通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图，来作为产物的动态变化响应性特征。然后，使用在图像的局部隐含特征提取方面具有优异表现的卷积神经网络模型来对所 述第一次过滤前的氟化氢液体的液相色谱图进行处理。应可以理解，由于在对所述第一次过滤前的氟化氢液体的液相色谱图进行特征挖掘时，考虑到所述液相色谱图在时间维度上具有动态地特征信息，因此，为了提取到这种动态的变化特征信息，进一步使用三维卷积核的第一卷积神经网络来对所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图进行处理以得到产物跟踪特征向量。
而对于所述多项第一参数和所述多项第二参数，分别将其按照时间维度和参数样本维度排列为精馏塔参数输入矩阵和吸收塔参数输入矩阵后通过作为过滤器的卷积神经网络以得到精馏塔特征向量和吸收塔特征向量，这样，就可以分别挖掘出整合了时间维度和样本维度的所述精馏塔参数矩阵和所述吸收塔参数矩阵中的所述多项第一参数以及所述多项第二参数的在时序上的高维隐含关联特征信息。
应可以理解，在使用作为过滤器的卷积神经网络对时间和数据样本维度上的数据矩阵进行特征提取以获得精馏塔特征向量V 1和吸收塔特征向量V 2时，由于卷积神经网络的过滤器是针对局部特征维度进行的特征提取，但是由于在特征向量融合时，是基于精馏塔特征向量V 1和吸收塔特征向量V 2的全局特征分布进行融合，因此对于精馏塔特征向量V 1和吸收塔特征向量V 2在融合时计算系数为：
Figure PCTCN2022116539-appb-000009
Figure PCTCN2022116539-appb-000010
Figure PCTCN2022116539-appb-000011
和
Figure PCTCN2022116539-appb-000012
分别是精馏塔特征向量V 1和吸收塔特征向量V 2的每个位置的特征值。
这里，该系数通过对特征值所对应的语义概念自下而上地进行显式泛化，来进行语义推理信息的显式泛化，从而通过对特征语义的自下而上地信息化推理来获得特征所对应的高维流形在高维语义空间内的空间复杂度的信息可塑性，从而促进精馏塔特征向量V 1和吸收塔特征向量V 2基于全局特征分布的特征融合以获得协同特征向量，从而能够提高后续分类的准确性。
考虑到在对于所述电子级氢氟酸进行制备的过程中，所述产物随时间的动态变化特征是关于所述精馏塔和所述吸收塔控制的响应性特征，因此，为了更准确地对于所述精馏塔和所述吸收塔进行监测控制，计算所述产物跟踪 特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量。再将所述分类特征向量通过分类器进行分类处理，以获得用于表示精馏塔和吸收塔的设备性能是否正常的分类结果。
基于此，本申请提出了一种用于电子级氢氟酸制备的生产管理控制系统，其包括：设备参数采集模块，用于获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；产物状态采集模块，用于通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；产物跟踪编码模块，用于将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；精馏塔参数编码模块，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量；吸收塔参数编码模块，用于将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量；参数特征融合模块，用于融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；响应性估计模块，用于计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及，设备管理结果生成模块，用于将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
图1图示了根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的应用场景图。如图1所示，在该应用场景中，首先，通过部署于精馏塔(例如，如图1中所示意的R)和吸收塔(例如，如图1中所示意的A)的多个传感器(例如，如图1中所示意的T1-Tn)获取多个预定时间点的所述精馏塔的多项第一参数和所述吸收塔的多项第二参数，并且通过液相色谱仪(例如，如图1中所示意的L)获取所述多个预定时间点的第一次过滤前的氟化氢液体(例如，如图1中所示意的H)的多个液相色谱图。这里，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述 精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力。然后，将所述多个预定时间点的所述精馏塔的多项第一参数和所述吸收塔的多项第二参数以及所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图输入至部署有用于电子级氢氟酸制备的生产管理控制算法的服务器中(例如，如图1中所示意的云服务器S)，其中，所述服务器能够以用于电子级氢氟酸制备的生产管理控制算法对所述多个预定时间点的所述精馏塔的多项第一参数和所述吸收塔的多项第二参数以及所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图进行处理，以生成用于表示精馏塔和吸收塔的设备性能是否正常的分类结果。
在介绍了本申请的基本原理之后，下面将参考附图来具体介绍本申请的各种非限制性实施例。
示例性系统
图2图示了根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的框图。如图2所示，根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统200，包括：设备参数采集模块210，用于获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；产物状态采集模块220，用于通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；产物跟踪编码模块230，用于将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；精馏塔参数编码模块240，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量；吸收塔参数编码模块250，用于将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷 积神经网络以得到吸收塔特征向量；参数特征融合模块260，用于融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；响应性估计模块270，用于计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及，设备管理结果生成模块280，用于将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
具体地，在本申请实施例中，所述设备参数采集模块210、所述产物状态采集模块220和所述产物跟踪编码模块230，用于获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力，并通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图，再将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量。如前所述，由于考虑到若想保证最终电子级氢氟酸产品的制备良率，就需要对产线上制备电子级氢氟酸的各个设备的性能进行在线监测。而在所述电子级氢氟酸制备的方法中，由于对于所述各个设备的性能进行监测的主要部分是对于精馏塔和吸收塔进行监测。并且，在对于所述精馏塔和所述吸收塔进行监测时，将第一次过滤前的氟化氢液体的液相色谱图来作为产物的响应性参照，能够进一步提高对于所述精馏塔和所述吸收塔的设备性能判断的准确性。
也就是，具体地，在本申请的技术方案中，首先，通过部署于精馏塔和吸收塔的各个传感器获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数。这里，考虑到其主要需要监控的设备部分为所述精馏塔塔釜的温度、塔身的温度、塔顶温度、塔釜压力、塔顶压力、精馏塔中回流温度和回流比以及所述吸收塔的温度和吸收塔的压力，因此，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力。
同时，通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图，来作为产物的动态变化响应性特征。然后，使用在图像的局部隐含特征提取方面具有优异表现的卷积神经网络模型来对所述第一次过滤前的氟化氢液体的液相色谱图进行处理。应可以理解，由于在对所述第一次过滤前的氟化氢液体的液相色谱图进行特征挖掘时，考虑到所述液相色谱图在时间维度上具有动态地特征信息，因此，在本申请的技术方案中，为了提取到这种动态的变化特征信息，进一步使用三维卷积核的第一卷积神经网络来对所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图进行处理以得到产物跟踪特征向量。
更具体地，在本申请实施例中，所述产物跟踪编码模块，包括：首先，使用所述第一卷积神经网络的各层在层的正向传递过程中对输入数据分别进行基于所述三维卷积核的卷积处理、池化处理和非线性激活处理以由所述第一卷积神经网络的最后一层输出跟踪特征图，其中，所述第一卷积神经网络的输入为所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图。然后，对所述跟踪特征图进行基于特征矩阵的全局均值池化以得到所述跟踪特征向量。应可以理解，池化操作可用于特征降维，缓解过拟合风险，降低卷积层对检测信息的过度敏感性，而所述全局均值池化能够保留每个所述跟踪特征图的重要信息，以用于突出所述跟踪特征图中响应最重要的部分。
图3图示了根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统中产物跟踪编码模块的框图。如图3所示，所述产物跟踪编码模块230，包括：特征动态捕捉单元231，用于使用所述第一卷积神经网络的各层在层的正向传递过程中对输入数据分别进行基于所述三维卷积核的卷积处理、池化处理和非线性激活处理以由所述第一卷积神经网络的最后一层输出跟踪特征图，其中，所述第一卷积神经网络的输入为所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；以及，全局池化单元232，用于对所述跟踪特征图进行基于特征矩阵的全局均值池化以得到所述跟踪特征向量。
具体地，在本申请实施例中，所述精馏塔参数编码模块240和所述吸收塔参数编码模块250，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤 器的第二卷积神经网络以得到精馏塔特征向量，并将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量。也就是，在本申请的技术方案中，对于所述多项第一参数和所述多项第二参数，分别将其按照时间维度和参数样本维度排列为精馏塔参数输入矩阵和吸收塔参数输入矩阵后通过作为过滤器的卷积神经网络以得到精馏塔特征向量和吸收塔特征向量，这样，就可以分别挖掘出整合了时间维度和样本维度的所述精馏塔参数矩阵和所述吸收塔参数矩阵中的所述多项第一参数以及所述多项第二参数的在时序上的高维隐含关联特征信息。
更具体地，在本申请实施例中，所述精馏塔参数编码模块，包括：首先，将所述多个预定时间点的精馏塔的多项第一参数按照时间维度排列为对应于所述精馏塔的多项第一参数行向量，并将所述多项第一参数行向量按照参数样本维度排列为所述精馏塔参数输入矩阵；然后，使用所述第二卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第二卷积神经网络的最后一层生成所述精馏塔特征向量，其中，所述第二卷积神经网络的第一层的输入为所述精馏塔参数输入矩阵。
更具体地，在本申请实施例中，所述吸收塔参数编码模块，包括：首先，将所述多个预定时间点的吸收塔的多项第二参数按照时间维度排列为对应于所述吸收塔的多项第二参数行向量，并将所述多项第二参数行向量按照参数样本维度排列为所述吸收塔参数输入矩阵；然后，使用所述第三卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第三卷积神经网络的最后一层生成所述吸收塔特征向量，其中，所述第三卷积神经网络的第一层的输入为所述吸收塔参数输入矩阵。
图4图示了根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统中精馏塔参数编码模块的框图。如图4所示，所述精馏塔参数编码模块240，包括：第一矩阵构造单元241，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度排列为对应于所述精馏塔的多项第一参数行向量，并将所述多项第一参数行向量按照参数样本维度排列为所述精馏塔参数输入矩阵；第一过滤器单元242，用于使用所述第二卷积神经网络的各层 在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第二卷积神经网络的最后一层生成所述精馏塔特征向量，其中，所述第二卷积神经网络的第一层的输入为所述精馏塔参数输入矩阵。
具体地，在本申请实施例中，所述参数特征融合模块260，用于融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量。应可以理解，在使用作为过滤器的所述卷积神经网络对时间和数据样本维度上的数据矩阵进行特征提取以获得所述精馏塔特征向量V 1和所述吸收塔特征向量V 2时，由于所述卷积神经网络的过滤器是针对局部特征维度进行的特征提取，但是由于在特征向量融合时，是基于所述精馏塔特征向量V 1和所述吸收塔特征向量V 2的全局特征分布进行融合，因此，在本申请的技术方案中，需要在对于所述精馏塔特征向量V 1和所述吸收塔特征向量V 2在融合时计算各自的系数对其各自本身进行加权修正。然后，计算所述精馏塔特征向量和所述吸收塔特征向量的按位置加权和就可以得到所述协同特征向量。
更具体地，在本申请实施例中，所述参数特征融合模块，进一步用于：所述精馏塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
Figure PCTCN2022116539-appb-000013
其中，
Figure PCTCN2022116539-appb-000014
表示所述精馏塔特征向量的各个位置的特征值，w 1表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述精馏塔特征向量的加权系数。所述吸收塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
Figure PCTCN2022116539-appb-000015
其中，
Figure PCTCN2022116539-appb-000016
表示所述吸收塔特征向量的各个位置的特征值，w 2表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述吸收塔特征向量的加权系数。应可以理解，该所述系数通过对特征值所对应的语义概念自下而上地进行显式泛化，来进行语义推理信息的显式泛化，从而通过对特征语义的自下而上地信息化推理来获得特征所对应的高维流形在高维语义空间内的空间复杂度的信息可塑性，从而促进所述精馏塔特征向量V 1和所述吸收塔特征向量V 2基于全局特征分布的特征融合以获得所述协同特征向量，从而能够提高后续分类的准确性。
具体地，在本申请实施例中，所述响应性估计模块270和所述设备管理结果生成模块280，用于计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量，并将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。应可以理解，考虑到在对于所述电子级氢氟酸进行制备的过程中，所述产物随时间的动态变化特征是关于所述精馏塔和所述吸收塔控制的响应性特征，因此，为了更准确地对于所述精馏塔和所述吸收塔进行监测控制，计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量。再将所述分类特征向量通过分类器进行分类处理，以获得用于表示精馏塔和吸收塔的设备性能是否正常的分类结果。相应地，在一个具体示例中，使用所述分类器以如下公式对所述分类特征向量进行处理以获得所述分类结果，其中，所述公式为：softmax{(W n,B n):…:(W 1,B 1)|X}，其中，W 1到W n为权重矩阵，B 1到B n为偏置向量，X为所述分类特征向量。
更具体地，在本申请实施例中，所述响应性估计模块，进一步用于以如下公式计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到所述分类特征向量；
其中，所述公式为：
s 3＝s 2⊙s 1 ⊙-1
其中s 1表示所述产物跟踪特征向量，s 2表示所述协同特征向量，s 3表示所述分类特征向量，⊙表示向量点乘，⊙-1表示对向量的每个位置的值取倒数。
综上，基于本申请实施例的所述用于电子级氢氟酸制备的生产管理控制系统200被阐明，其采用基于人工智能的监测技术，通过对于电子级氢氟酸制备过程中精馏塔和吸收塔的各个参数进行监测来保证最终电子级氢氟酸产品的制备良率，并且在此过程中，还使用产物的动态变化特征来进行响应性估计，以更准确地对所述精馏塔和所述吸收塔的设备性能是否正常进行监测。
如上所述，根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统200可以实现在各种终端设备中，例如用于电子级氢氟酸制备的生产管理控制算法的服务器等。在一个示例中，根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统200可以作为一个软件模块和/或硬件模块而 集成到终端设备中。例如，该用于电子级氢氟酸制备的生产管理控制系统200可以是该终端设备的操作系统中的一个软件模块，或者可以是针对于该终端设备所开发的一个应用程序；当然，该用于电子级氢氟酸制备的生产管理控制系统200同样可以是该终端设备的众多硬件模块之一。
替换地，在另一示例中，该用于电子级氢氟酸制备的生产管理控制系统200与该终端设备也可以是分立的设备，并且该用于电子级氢氟酸制备的生产管理控制系统200可以通过有线和/或无线网络连接到该终端设备，并且按照约定的数据格式来传输交互信息。
示例性方法
图5图示了用于电子级氢氟酸制备的生产管理控制系统的控制方法的流程图。如图5所示，根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的控制方法，包括步骤：S110，获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；S120，通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；S130，将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；S140，将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量；S150，将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量；S160，融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；S170，计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及，S180，将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
图6图示了根据本申请实施例的用于电子级氢氟酸制备的生产管理控制系统的控制方法的架构示意图。如图6所示，在所述用于电子级氢氟酸制备 的生产管理控制系统的控制方法的网络架构中，首先，将获得的所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图(例如，如图6中所示意的P)通过使用三维卷积核的第一卷积神经网络(例如，如图6中所示意的CNN1)以得到产物跟踪特征向量(例如，如图6中所示意的VF)；接着，将获得的所述多个预定时间点的精馏塔的多项第一参数(例如，如图6中所示意的Q1)按照时间维度和参数样本维度排列为精馏塔参数输入矩阵(例如，如图6中所示意的M1)后通过作为过滤器的第二卷积神经网络(例如，如图6中所示意的CNN2)以得到精馏塔特征向量(例如，如图6中所示意的VF1)；然后，将获得的所述多个预定时间点的吸收塔的多项第二参数(例如，如图6中所示意的Q2)按照时间维度和参数样本维度排列为吸收塔参数输入矩阵(例如，如图6中所示意的M2)后通过作为过滤器的第三卷积神经网络(例如，如图6中所示意的CNN3)以得到吸收塔特征向量(例如，如图6中所示意的VF2)；接着，融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量(例如，如图6中所示意的VF3)；然后，计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量(例如，如图6中所示意的V)；以及，最后，将所述分类特征向量通过分类器(例如，如图6中所示意的圈S)以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
更具体地，在步骤S110、步骤S120和步骤S130中，获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力，并通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图。应可以理解，由于考虑到若想保证最终电子级氢氟酸产品的制备良率，就需要对产线上制备电子级氢氟酸的各个设备的性能进行在线监测。而在所述电子级氢氟酸制备的方法中，由于对于所述各个设备的性能进行监测的主要部分是对于精馏塔和吸收塔进行监测。并且，在对于所述精馏塔和所述吸收塔进行监测时，将第一次过滤前的氟化氢液体的液相色谱图来作为产物的响应性参照，能够进一步提高对于所述精馏塔和所述吸收塔的设备性能判断的准确性。
也就是，具体地，在本申请的技术方案中，首先，通过部署于精馏塔和吸收塔的各个传感器获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数。这里，考虑到其主要需要监控的设备部分为所述精馏塔塔釜的温度、塔身的温度、塔顶温度、塔釜压力、塔顶压力、精馏塔中回流温度和回流比以及所述吸收塔的温度和吸收塔的压力，因此，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力。
同时，通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图，来作为产物的动态变化响应性特征。然后，使用在图像的局部隐含特征提取方面具有优异表现的卷积神经网络模型来对所述第一次过滤前的氟化氢液体的液相色谱图进行处理。应可以理解，由于在对所述第一次过滤前的氟化氢液体的液相色谱图进行特征挖掘时，考虑到所述液相色谱图在时间维度上具有动态地特征信息，因此，在本申请的技术方案中，为了提取到这种动态的变化特征信息，进一步使用三维卷积核的第一卷积神经网络来对所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图进行处理以得到产物跟踪特征向量。
更具体地，在步骤S140和步骤S150中，将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量，并将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量。也就是，在本申请的技术方案中，对于所述多项第一参数和所述多项第二参数，分别将其按照时间维度和参数样本维度排列为精馏塔参数输入矩阵和吸收塔参数输入矩阵后通过作为过滤器的卷积神经网络以得到精馏塔特征向量和吸收塔特征向量，这样，就可以分别挖掘出整合了时间维度和样本维度的所述精馏塔参数矩阵和所述吸收塔参数矩阵中的所述多项第一参数以及所述多项第二参数的在时序上的高维隐含关联特征信息。
更具体地，在步骤S160中，融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量。应可以理解，应可以理解，在使用作为过滤器 的所述卷积神经网络对时间和数据样本维度上的数据矩阵进行特征提取以获得所述精馏塔特征向量V 1和所述吸收塔特征向量V 2时，由于所述卷积神经网络的过滤器是针对局部特征维度进行的特征提取，但是由于在特征向量融合时，是基于所述精馏塔特征向量V 1和所述吸收塔特征向量V 2的全局特征分布进行融合，因此，在本申请的技术方案中，需要在对于所述精馏塔特征向量V 1和所述吸收塔特征向量V 2在融合时计算各自的系数对其各自本身进行加权修正。然后，计算所述精馏塔特征向量和所述吸收塔特征向量的按位置加权和就可以得到所述协同特征向量。
更具体地，在步骤S170和步骤S180中，计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量，并将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。应可以理解，考虑到在对于所述电子级氢氟酸进行制备的过程中，所述产物随时间的动态变化特征是关于所述精馏塔和所述吸收塔控制的响应性特征，因此，为了更准确地对于所述精馏塔和所述吸收塔进行监测控制，计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量。再将所述分类特征向量通过分类器进行分类处理，以获得用于表示精馏塔和吸收塔的设备性能是否正常的分类结果。相应地，在一个具体示例中，使用所述分类器以如下公式对所述分类特征向量进行处理以获得所述分类结果，其中，所述公式为：softmax{(W n,B n):…:(W 1,B 1)|X}，其中，W 1到W n为权重矩阵，B 1到B n为偏置向量，X为所述分类特征向量。
综上，基于本申请实施例的所述用于电子级氢氟酸制备的生产管理控制系统的控制方法被阐明，其采用基于人工智能的监测技术，通过对于电子级氢氟酸制备过程中精馏塔和吸收塔的各个参数进行监测来保证最终电子级氢氟酸产品的制备良率，并且在此过程中，还使用产物的动态变化特征来进行响应性估计，以更准确地对所述精馏塔和所述吸收塔的设备性能是否正常进行监测。
示例性计算机程序产品和计算机可读存储介质
除了上述方法和设备以外，本申请的实施例还可以是计算机程序产品，其包括计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处 理器执行本说明书上述“示例性方法”部分中描述的根据本申请各种实施例的用于电子级氢氟酸制备的生产管理控制系统的控制方法中的功能中的步骤。
所述计算机程序产品可以以一种或多种程序设计语言的任意组合来编写用于执行本申请实施例操作的程序代码，所述程序设计语言包括面向对象的程序设计语言，诸如Java、C++等，还包括常规的过程式程序设计语言，诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。
此外，本申请的实施例还可以是计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的用于电子级氢氟酸制备的生产管理控制系统的控制方法中的步骤。
所述计算机可读存储介质可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以包括但不限于电、磁、光、电磁、红外线、或半导体的系统、系统或器件，或者任意以上的组合。可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
以上结合具体实施例描述了本申请的基本原理，但是，需要指出的是，在本申请中提及的优点、优势、效果等仅是示例而非限制，不能认为这些优点、优势、效果等是本申请的各个实施例必须具备的。另外，上述公开的具体细节仅是为了示例的作用和便于理解的作用，而非限制，上述细节并不限制本申请为必须采用上述具体的细节来实现。
本申请中涉及的器件、装置、设备、系统的方框图仅作为例示性的例子并且不意图要求或暗示必须按照方框图示出的方式进行连接、布置、配置。如本领域技术人员将认识到的，可以按任意方式连接、布置、配置这些器件、 装置、设备、系统。诸如“包括”、“包含”、“具有”等等的词语是开放性词汇，指“包括但不限于”，且可与其互换使用。这里所使用的词汇“或”和“和”指词汇“和/或”，且可与其互换使用，除非上下文明确指示不是如此。这里所使用的词汇“诸如”指词组“诸如但不限于”，且可与其互换使用。
还需要指出的是，在本申请的装置、设备和方法中，各部件或各步骤是可以分解和/或重新组合的。这些分解和/或重新组合应视为本申请的等效方案。
提供所公开的方面的以上描述以使本领域的任何技术人员能够做出或者使用本申请。对这些方面的各种修改对于本领域技术人员而言是非常显而易见的，并且在此定义的一般原理可以应用于其他方面而不脱离本申请的范围。因此，本申请不意图被限制到在此示出的方面，而是按照与在此公开的原理和新颖的特征一致的最宽范围。
为了例示和描述的目的已经给出了以上描述。此外，此描述不意图将本申请的实施例限制到在此公开的形式。尽管以上已经讨论了多个示例方面和实施例，但是本领域技术人员将认识到其某些变型、修改、改变、添加和子组合。

Claims (10)

  1. 一种用于电子级氢氟酸制备的生产管理控制系统，其特征在于，包括：
    设备参数采集模块，用于获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；
    产物状态采集模块，用于通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；
    产物跟踪编码模块，用于将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；
    精馏塔参数编码模块，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量；
    吸收塔参数编码模块，用于将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量；
    参数特征融合模块，用于融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；
    响应性估计模块，用于计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及
    设备管理结果生成模块，用于将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
  2. 根据权利要求1所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述产物跟踪编码模块，包括：
    特征动态捕捉单元，用于使用所述第一卷积神经网络的各层在层的正向传递过程中对输入数据分别进行基于所述三维卷积核的卷积处理、池化处理和非线性激活处理以由所述第一卷积神经网络的最后一层输出跟踪特征图， 其中，所述第一卷积神经网络的输入为所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；以及
    全局池化单元，用于对所述跟踪特征图进行基于特征矩阵的全局均值池化以得到所述跟踪特征向量。
  3. 根据权利要求2所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述精馏塔参数编码模块，包括：
    第一矩阵构造单元，用于将所述多个预定时间点的精馏塔的多项第一参数按照时间维度排列为对应于所述精馏塔的多项第一参数行向量，并将所述多项第一参数行向量按照参数样本维度排列为所述精馏塔参数输入矩阵；
    第一过滤器单元，用于使用所述第二卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第二卷积神经网络的最后一层生成所述精馏塔特征向量，其中，所述第二卷积神经网络的第一层的输入为所述精馏塔参数输入矩阵。
  4. 根据权利要求3所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述吸收塔参数编码模块，包括：
    第二矩阵构造单元，用于将所述多个预定时间点的吸收塔的多项第二参数按照时间维度排列为对应于所述吸收塔的多项第二参数行向量，并将所述多项第二参数行向量按照参数样本维度排列为所述吸收塔参数输入矩阵；
    第二过滤器单元，用于使用所述第三卷积神经网络的各层在层的正向传递中对输入数据进行卷积处理、沿特征矩阵的池化处理和激活处理以由所述第三卷积神经网络的最后一层生成所述吸收塔特征向量，其中，所述第三卷积神经网络的第一层的输入为所述吸收塔参数输入矩阵。
  5. 根据权利要求4所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述参数特征融合模块，用于计算所述精馏塔特征向量和所述吸收塔特征向量的按位置加权和以得到所述协同特征向量。
  6. 根据权利要求5所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述精馏塔特征向量的加权权重以如下公式计算而得，其中，所述公 式为：
    Figure PCTCN2022116539-appb-100001
    其中，
    Figure PCTCN2022116539-appb-100002
    表示所述精馏塔特征向量的各个位置的特征值，w 1表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述精馏塔特征向量的加权系数。
  7. 根据权利要求6所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述吸收塔特征向量的加权权重以如下公式计算而得，其中，所述公式为：
    Figure PCTCN2022116539-appb-100003
    其中，
    Figure PCTCN2022116539-appb-100004
    表示所述吸收塔特征向量的各个位置的特征值，w 2表示所述精馏塔特征向量和所述吸收塔特征向量在融合时所述吸收塔特征向量的加权系数。
  8. 根据权利要求7所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述响应性估计模块，进一步用于以如下公式计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到所述分类特征向量；
    其中，所述公式为：
    s 3＝s 2⊙s 1 ⊙-1
    其中s 1表示所述产物跟踪特征向量，s 2表示所述协同特征向量，s 3表示所述分类特征向量，⊙表示向量点乘，⊙-1表示对向量的每个位置的值取倒数。
  9. 根据权利要求8所述的用于电子级氢氟酸制备的生产管理控制系统，其中，所述设备管理结果生成模块，进一步用于：使用所述分类器以如下公式对所述分类特征向量进行处理以获得所述分类结果，其中，所述公式为：softmax{(W n,B n):…:(W 1,B 1)|X}，其中，W 1到W n为权重矩阵，B 1到B n为偏置向量，X为所述分类特征向量。
  10. 一种用于电子级氢氟酸制备的生产管理控制系统的控制方法，其特征在于，包括：
    获取多个预定时间点的精馏塔的多项第一参数和吸收塔的多项第二参数，其中，所述多项第一参数包括所述精馏塔的塔釜温度，所述精馏塔的塔身温度，所述精馏塔的塔顶温度，所述精馏塔的塔釜压力，所述精馏塔的塔顶压力，所述精馏塔的的塔中回流温度和回流比，所述多项第二参数包括所述吸收塔的温度和所述吸收塔的压力；
    通过液相色谱仪获取所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图；
    将所述多个预定时间点的第一次过滤前的氟化氢液体的多个液相色谱图通过使用三维卷积核的第一卷积神经网络以得到产物跟踪特征向量；
    将所述多个预定时间点的精馏塔的多项第一参数按照时间维度和参数样本维度排列为精馏塔参数输入矩阵后通过作为过滤器的第二卷积神经网络以得到精馏塔特征向量；
    将所述多个预定时间点的吸收塔的多项第二参数按照时间维度和参数样本维度排列为吸收塔参数输入矩阵后通过作为过滤器的第三卷积神经网络以得到吸收塔特征向量；
    融合所述精馏塔特征向量和所述吸收塔特征向量以得到协同特征向量；
    计算所述产物跟踪特征向量相对于所述协同特征向量的响应性估计以得到分类特征向量；以及
    将所述分类特征向量通过分类器以得到分类结果，所述分类结果用于表示精馏塔和吸收塔的设备性能是否正常。
PCT/CN2022/116539 2022-07-01 2022-09-01 用于电子级氢氟酸制备的生产管理控制系统及其控制方法 WO2024000798A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210767053.0 2022-07-01
CN202210767053.0A CN115259089B (zh) 2022-07-01 2022-07-01 用于电子级氢氟酸制备的生产管理控制系统及其控制方法

Publications (1)

Publication Number Publication Date
WO2024000798A1 true WO2024000798A1 (zh) 2024-01-04

Family

ID=83763234

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/116539 WO2024000798A1 (zh) 2022-07-01 2022-09-01 用于电子级氢氟酸制备的生产管理控制系统及其控制方法

Country Status (2)

Country Link
CN (1) CN115259089B (zh)
WO (1) WO2024000798A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115688592B (zh) * 2022-11-09 2023-05-09 福建德尔科技股份有限公司 用于电子级四氟化碳制备的精馏控制系统及其方法
CN115903705B (zh) * 2022-11-30 2023-07-14 福建省杭氟电子材料有限公司 用于电子级六氟丁二烯制备的生产管理控制系统
CN117046140B (zh) * 2023-02-28 2024-03-08 福建天甫电子材料有限公司 用于电子级氢氟酸精馏的装置
CN116819957B (zh) * 2023-03-29 2024-03-08 福建省龙德新能源有限公司 电子级六氟磷酸锂的尾气处理系统及其方法
CN117208852B (zh) * 2023-10-07 2024-02-13 福建天甫电子材料有限公司 用于电子级氢氟酸制备的纯化控制系统及方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH11149302A (ja) * 1997-11-17 1999-06-02 Mitsubishi Chemical Corp 製造または処理プロセスの操作制御方法
CN102183892A (zh) * 2011-05-10 2011-09-14 上海交通大学 甲醇三塔精馏系统的变负荷能耗优化控制方法
CN103901130A (zh) * 2014-04-10 2014-07-02 中国科学院福建物质结构研究所 一种生产碳酸二甲酯的催化剂在线评价装置、评价方法和生产方法
CN103991847A (zh) * 2013-02-18 2014-08-20 福建省邵武市永晶化工有限公司 电子级氢氟酸制备方法
CN207210295U (zh) * 2017-08-29 2018-04-10 湖北工程学院 一种乙醇纯化系统以及色谱纯乙醇生产系统
JP2021163207A (ja) * 2020-03-31 2021-10-11 月島機械株式会社 制御装置、制御方法、及びプログラム
CN113703314A (zh) * 2020-05-20 2021-11-26 中国石油化工股份有限公司 碳三液相加氢反应器在线预测自动控制方法及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106215447A (zh) * 2016-08-29 2016-12-14 成都灵感元素科技有限公司 一种电子级氢氟酸生产用精馏塔
CN212503988U (zh) * 2020-07-13 2021-02-09 南京德源环保科技发展有限公司 连续法制备电子级氢氟酸的系统
CN111704109A (zh) * 2020-07-13 2020-09-25 南京德源环保科技发展有限公司 连续法制备电子级氢氟酸的方法及系统
CN113651294B (zh) * 2021-07-02 2024-01-02 北京化工大学 电子级氢氟酸的连续化制备系统及方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH11149302A (ja) * 1997-11-17 1999-06-02 Mitsubishi Chemical Corp 製造または処理プロセスの操作制御方法
CN102183892A (zh) * 2011-05-10 2011-09-14 上海交通大学 甲醇三塔精馏系统的变负荷能耗优化控制方法
CN103991847A (zh) * 2013-02-18 2014-08-20 福建省邵武市永晶化工有限公司 电子级氢氟酸制备方法
CN103901130A (zh) * 2014-04-10 2014-07-02 中国科学院福建物质结构研究所 一种生产碳酸二甲酯的催化剂在线评价装置、评价方法和生产方法
CN207210295U (zh) * 2017-08-29 2018-04-10 湖北工程学院 一种乙醇纯化系统以及色谱纯乙醇生产系统
JP2021163207A (ja) * 2020-03-31 2021-10-11 月島機械株式会社 制御装置、制御方法、及びプログラム
CN113703314A (zh) * 2020-05-20 2021-11-26 中国石油化工股份有限公司 碳三液相加氢反应器在线预测自动控制方法及系统

Also Published As

Publication number Publication date
CN115259089B (zh) 2023-02-10
CN115259089A (zh) 2022-11-01

Similar Documents

Publication Publication Date Title
WO2024000798A1 (zh) 用于电子级氢氟酸制备的生产管理控制系统及其控制方法
WO2024045247A1 (zh) 用于氟化铵生产的生产管理控制系统及其控制方法
WO2024021254A1 (zh) 电子级三氟化氯的智能分离纯化系统
WO2024021259A1 (zh) 用于缓冲氧化物蚀刻液生产的自动配料系统及其配料方法
WO2024026993A1 (zh) 氟化铵制备用的自动配料控制系统及其控制方法
WO2023206724A1 (zh) 用于电子级二氟甲烷制备的精馏控制系统及其控制方法
CN115079572A (zh) 用于六氟磷酸锂制备的能源管理控制系统及其控制方法
WO2024045244A1 (zh) 用于无水氟化氢生产的能源管理控制系统及其控制方法
WO2024011747A1 (zh) 一种用于六氟丁二烯制备的智能化冷却液循环控制系统
CN115235612B (zh) 伺服电机的故障智能诊断系统及其诊断方法
CN115013298B (zh) 污水泵的实时性能在线监测系统及其监测方法
CN115099684B (zh) 企业安全生产管理系统及其管理方法
WO2023226236A1 (zh) 用于电子级氢氟酸制备的能源管理控制系统及其控制方法
WO2024113599A1 (zh) 用于电子级六氟丁二烯制备的生产管理控制系统
CN110991504A (zh) 基于motcn-ae的多维时间序列异常值检测方法及系统
CN114241515A (zh) 一种基于时空上下文特征感知的三维人体姿态估计方法
CN108573232B (zh) 基于卷积神经网络的人体动作识别方法
CN115146676A (zh) 电路故障检测方法及其系统
CN117455451A (zh) 一种草甘膦水解溶剂的回收方法
CN114610488A (zh) 基于负载均衡的数据采集方法、系统和电子设备
CN116644865A (zh) 商用车油耗预测方法、电子设备和存储介质
CN116401528A (zh) 一种基于全局-局部散度的多元时间序列无监督降维方法
CN115309895A (zh) 基于研发项目文本描述的进度管理系统及其管理方法
CN115410064A (zh) 智能制造下的设备状态监测系统及其监测方法
CN115131735A (zh) 五金工具热处理后冷却媒介温度控制系统及温度控制方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22948888

Country of ref document: EP

Kind code of ref document: A1