WO2024114835A2 - 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法 - Google Patents

用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法 Download PDF

Info

Publication number
WO2024114835A2
WO2024114835A2 PCT/CN2024/074348 CN2024074348W WO2024114835A2 WO 2024114835 A2 WO2024114835 A2 WO 2024114835A2 CN 2024074348 W CN2024074348 W CN 2024074348W WO 2024114835 A2 WO2024114835 A2 WO 2024114835A2
Authority
WO
WIPO (PCT)
Prior art keywords
slurry
training
feature
ceramic composite
composite fiber
Prior art date
Application number
PCT/CN2024/074348
Other languages
English (en)
French (fr)
Other versions
WO2024114835A3 (zh
Inventor
李惠林
褚奇奇
邓国敢
朱继保
安国栋
史少军
沈才方
潘建法
金秀峰
王丽娟
Original Assignee
浙江致远环境科技股份有限公司
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 浙江致远环境科技股份有限公司 filed Critical 浙江致远环境科技股份有限公司
Publication of WO2024114835A2 publication Critical patent/WO2024114835A2/zh
Publication of WO2024114835A3 publication Critical patent/WO2024114835A3/zh

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
    • B01D53/34—Chemical or biological purification of waste gases
    • B01D53/46—Removing components of defined structure
    • B01D53/48—Sulfur compounds
    • B01D53/50—Sulfur oxides
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
    • B01D53/34—Chemical or biological purification of waste gases
    • B01D53/46—Removing components of defined structure
    • B01D53/54—Nitrogen compounds
    • B01D53/56—Nitrogen oxides
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
    • B01D53/34—Chemical or biological purification of waste gases
    • B01D53/46—Removing components of defined structure
    • B01D53/68—Halogens or halogen compounds
    • B01D53/70—Organic halogen compounds
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D53/00—Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases, aerosols
    • B01D53/34—Chemical or biological purification of waste gases
    • B01D53/74—General processes for purification of waste gases; Apparatus or devices specially adapted therefor
    • B01D53/86—Catalytic processes
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D67/00—Processes specially adapted for manufacturing semi-permeable membranes for separation processes or apparatus
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D69/00—Semi-permeable membranes for separation processes or apparatus characterised by their form, structure or properties; Manufacturing processes specially adapted therefor
    • B01D69/02—Semi-permeable membranes for separation processes or apparatus characterised by their form, structure or properties; Manufacturing processes specially adapted therefor characterised by their properties
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D69/00—Semi-permeable membranes for separation processes or apparatus characterised by their form, structure or properties; Manufacturing processes specially adapted therefor
    • B01D69/04—Tubular membranes
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
    • B01D—SEPARATION
    • B01D69/00—Semi-permeable membranes for separation processes or apparatus characterised by their form, structure or properties; Manufacturing processes specially adapted therefor
    • B01D69/12—Composite membranes; Ultra-thin membranes

Definitions

  • the present application relates to the technical field of catalytic filter tubes, and more specifically, to a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins and a preparation method thereof.
  • the embodiment of the present application provides a ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins and a preparation method thereof.
  • the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins comprises: a ceramic composite fiber tube body and a catalyst formed on the surface of the ceramic composite fiber tube body, and the catalyst is a vanadium-titanium composite oxide.
  • the preparation method thereof includes using an artificial intelligence control algorithm based on deep learning to extract global multi-scale implicit association features of multiple groups of reference data of the slurry as a reference feature matrix, and performing feature query from the reference feature matrix based on the global implicit association features of the actual detection data of the slurry, so as to determine the reference value of the slurry for the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins.
  • the reference value for cutting can be accurately determined according to the actual parameters of the slurry, thereby improving the efficiency of preparation.
  • a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins comprising:
  • a catalyst is formed on the surface of the ceramic composite fiber tube, and the catalyst is a vanadium-titanium composite oxide.
  • the components of the vanadium-titanium composite oxide are: 0.5-1.5% platinum, 2-5% vanadium pentoxide, 0.5%-3% tungsten trioxide, 1%-3% rare earth, 0.2%-0.5% titanium dioxide, 0.2%-0.6% thiourea, 0.3%-0.5% Tween 60, 0.1%-0.5% dispersant and 90-93.4% pure water.
  • a method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins comprises:
  • Step S1 pretreating the fiber to obtain a slurry
  • Step S2 injecting the slurry into the length and thickness automatic adjustment mold of the ceramic composite fiber filter tube through the pressure grouting port above the mold to obtain a blank tube of the ceramic composite fiber filter tube;
  • Step S3 the central control unit controls the closing of the first solenoid valve disposed above the pressure grouting port, and the central control unit controls the vacuum pump to suck the blank tube of the ceramic composite fiber filter tube to obtain a shaped ceramic composite fiber filter tube blank tube;
  • Step S4 placing the shaped ceramic composite fiber filter tube blank in a catalyst sol, and soaking, airing, drying and sintering under vacuum conditions to obtain a ceramic composite fiber filter tube with catalytic function;
  • Step S5 drying the ceramic composite fiber filter tube with catalytic function in a drying room to obtain a ceramic composite fiber catalytic filter tube.
  • the step S2 comprises:
  • each group of reference data includes the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry, and the actual reference value of the reference slurry;
  • S22 passing each group of reference data in the multiple groups of reference data through a context encoder including an embedding layer to obtain multiple context reference data item feature vectors, and cascading the multiple context reference data item feature vectors into a reference feature vector to obtain multiple reference feature vectors;
  • test data of the slurry includes the fiber length of the slurry, the pH value of the slurry and the solid content of the slurry;
  • the step S22 comprises:
  • the multiple context reference data item feature vectors are concatenated into a reference feature vector to obtain multiple reference feature vectors.
  • the step S23 comprises: using each mixed convolution layer of the convolutional neural network model to perform the following operations on the input data in the forward transfer of the layer:
  • the output of the last mixed convolution layer of the convolutional neural network model is the reference feature matrix.
  • the multi-scale convolution processing of the input data to obtain a multi-scale convolution feature map includes:
  • the first feature map, the second feature map, the third feature map and the fourth feature map are cascaded and aggregated to obtain the multi-scale convolutional feature map.
  • the method further includes step S100: training the context encoder including the embedding layer and the decoder;
  • step S100 includes:
  • each set of training reference data includes a training fiber length of a reference slurry, a training pH value of the reference slurry, a training solid content of the reference slurry, and a real reference value of the reference slurry;
  • S120 Pass each group of training reference data in the multiple groups of training reference data through the context encoder including the embedding layer to obtain multiple training context reference data item feature vectors, and cascade the multiple training context reference data item feature vectors into a training reference feature vector to obtain multiple training reference feature vectors;
  • S130 Arranging the multiple training reference feature vectors in two dimensions into a training reference feature matrix and then passing the matrix through the convolutional neural network model including multiple hybrid convolutional layers to obtain a training reference feature matrix;
  • S140 Acquire training detection data of the slurry, wherein the training detection data of the slurry includes training fiber length of the slurry, training pH value of the slurry, training solid content of the slurry, and a real value of a reference value of the slurry;
  • S170 Passing the training decoding feature vector through the decoder to obtain a decoding loss function value
  • S190 Calculate a weighted sum of the decoding loss function value and the multi-distribution binary regression quality loss function value as a loss function value to train the context encoder including the embedding layer and the decoder.
  • the present application provides a ceramic composite fiber catalytic filter tube for desulfurization, denitration and dioxin removal and a preparation method thereof, wherein the ceramic composite fiber catalytic filter tube for desulfurization, denitration and dioxin removal comprises: a ceramic composite fiber tube body and a catalyst formed on the surface of the ceramic composite fiber tube body, wherein the catalyst is a vanadium-titanium composite oxide.
  • the preparation method thereof includes using an artificial intelligence control algorithm based on deep learning to extract global multi-scale implicit association features of multiple groups of reference data of the slurry as a reference feature matrix, and performing feature query from the reference feature matrix based on the global implicit association features of the actual detection data of the slurry, so as to determine the reference value of the slurry for the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and dioxin removal.
  • the reference value for the material can be accurately determined according to the actual parameters of the slurry, thereby improving the efficiency of the preparation.
  • FIG1 is a schematic diagram of the structure of a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG. 2 is a flow chart of a method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG3 is a schematic diagram of the scene of step S2 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG. 4 is a flow chart of sub-steps of step S2 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG. 5 is a schematic diagram of the sub-step architecture of step S2 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG. 6 is a flow chart of sub-steps of step S22 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG. 7 is a flow chart of sub-steps of step S100 further included in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • FIG. 8 is a block diagram of a system for preparing a blank tube of a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • Figure 1 is a structural schematic diagram of a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • the ceramic composite fiber catalytic filter tube 10 for desulfurization, denitrification and removal of dioxins includes: a ceramic composite fiber tube body 12; and a catalyst 13 formed on the surface of the ceramic composite fiber tube body, wherein the catalyst is a vanadium-titanium composite oxide.
  • the components of the vanadium-titanium composite oxide are: platinum 0.5-1.5%, vanadium pentoxide 2-5%, tungsten trioxide 0.5%-3%, rare earth 1%-3%, titanium dioxide 0.2%-0.5%, thiourea 0.2%-0.6%, Tween 600.3%-0.5%, dispersant 0.1%-0.5% and pure water 90-93.4%.
  • the ceramic composite fiber catalytic filter tube 10 for desulfurization, denitrification and removal of dioxins further includes an end body 11 fixed to the ceramic composite fiber tube body 12 , and the diameter of the end body 11 is greater than the diameter of the ceramic composite fiber tube body 12 .
  • FIG. 2 is a flow chart of a method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitration and dioxin removal according to an embodiment of the present application.
  • the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitration and dioxin removal comprises: step S1: pretreating the fiber to obtain a slurry; step S2: injecting the slurry into a mold of the ceramic composite fiber filter tube through a pressure grouting port above the mold to obtain a blank tube of the ceramic composite fiber filter tube, the length of the ceramic composite fiber filter tube is 0.5 to 9 m, the thickness is 5 to 50 mm, and the diameter is 5 0 ⁇ 200mm;
  • Step S3 The central control unit controls the closing of the first solenoid valve arranged above the pressure grouting port, and controls the vacuum pump through the central control unit to suck the blank tube of the ceramic composite fiber filter tube to obtain a shaped ceramic composite composite
  • the above-mentioned ceramic composite fiber catalytic filter tube for desulfurization and denitrification to remove dioxins also has a dust removal function.
  • the ceramic composite fiber catalytic filter tube has a high porosity and a small air resistance. Dioxins, dust particles, etc. are trapped on the surface of the ceramic composite fiber catalytic filter tube. The dust filtration efficiency can reach more than 99%. After dust removal, the gas passes through the porous element of the ceramic composite fiber catalytic filter tube and is discharged under the action of the catalyst, and the dust removal effect is significant.
  • step S2 before grouting, the obtained fiber length of the slurry, the pH value of the slurry and the solid content of the slurry need to be pre-input into the central control unit, and the central control unit determines the slurry feeding reference value according to the fiber length of the slurry, the pH value of the slurry and the solid content of the slurry.
  • the central control unit controls the grouting pressure and the grouting pressure holding time of the grouting unit connected to the pressure grouting port on the left side of the mold, as well as the suction pressure and suction time of the vacuum pump arranged under the mold according to the feeding reference value, and then performs grouting.
  • the central control unit determines the reference value for feeding the slurry based on the fiber length of the slurry, the pH value of the slurry and the solid content of the slurry, since the various parameters of the slurry are correlated with each other, it is difficult to determine the reference value for feeding according to the actual parameters of the slurry during the actual operation process, and it is difficult to adjust the preparation parameters of the ceramic composite fiber filter tube to be prepared, resulting in low preparation efficiency.
  • an artificial intelligence control algorithm based on deep learning is used to extract global multi-scale implicit correlation features of multiple groups of reference data of the slurry as a reference feature matrix, and a feature query is performed from the reference feature matrix based on the global implicit correlation features of the actual detection data of the slurry, so as to determine the reference value for feeding the slurry.
  • the reference value for feeding can be accurately determined according to the actual parameters of the slurry to improve the efficiency of preparation.
  • each group of reference data includes the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry, and the actual reference value of the reference slurry.
  • each group of reference data in the multiple groups of reference data is further encoded through a context encoder including an embedding layer to extract the global context correlation features between each data item in each group of reference data to be more suitable for characterizing the implicit feature information of the slurry, thereby obtaining multiple context reference data item feature vectors.
  • the multiple context reference data item feature vectors are further cascaded into reference feature vectors to integrate the global feature information of each data item in each group of reference data in the multiple groups of reference data to obtain multiple reference feature vectors.
  • the multiple reference feature vectors are arranged in two dimensions into a reference feature matrix and then processed in a convolutional neural network model including multiple hybrid convolutional layers to extract the multi-scale implicit association features of the global features of each data item of each group of reference data in the reference feature matrix, thereby obtaining multiple multi-scale time-frequency feature vectors.
  • the design of this module includes four branches in parallel, which are composed of a common convolutional layer with a convolution kernel size of 3 ⁇ 3 and three dilated convolutional layers with a convolution kernel size of 3 ⁇ 3.
  • the reference feature matrix is operated respectively, and the expansion rates of the three branches of the dilated convolution are set to 2, 3, and 4 respectively.
  • image information of different receptive domains can be obtained, and feature maps of different scales can be obtained. While expanding the receptive field, downsampling loss information is avoided. Then, the 4 branch feature maps are fused to make the sampling more intensive, with both high-level features and no additional parameters. In this way, the data association feature library of the reference slurry can be constructed to facilitate subsequent feature queries.
  • a feature query can be performed in the constructed data association feature library of the reference slurry based on the global feature association information between the various data items of the actual detection data of the slurry, so as to determine the reference value for the actual slurry. That is, specifically, first, the detection data of the slurry is obtained, and the detection data of the slurry includes the fiber length of the slurry, the pH value of the slurry, and the solid content of the slurry. Then, the detection data of the slurry is encoded through the context encoder containing the embedding layer to extract the globally implicit association features between the various data items of the detection data of the slurry, thereby obtaining a detection feature vector.
  • the detection feature vector is used as a query feature vector and multiplied with the reference feature matrix to query the slurry feed reference value feature corresponding to the associated feature information of each data item under the actual detection data of the slurry, and decoding regression is performed on this to obtain a decoded value for representing the slurry feed reference value.
  • the feed reference value can be accurately determined according to the actual parameters of the slurry to improve the efficiency of preparation.
  • the convolutional neural network model including multiple mixed convolutional layers can extract multi-scale local correlation features between the multiple reference feature vectors.
  • the present application provides a method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins, wherein the slurry is injected into the mold of the ceramic composite fiber filter tube through the pressure grouting port above the mold to obtain a blank tube of the ceramic composite fiber filter tube, the length of the ceramic composite fiber filter tube is 0.5 ⁇ 9m, the thickness is 5 ⁇ 50mm, and the diameter is 50 ⁇ 200mm, which includes: obtaining multiple groups of reference data, each group of reference data includes the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry and the actual reference value of the reference slurry; each group of reference data in the multiple groups of reference data is passed through a context encoder including an embedding layer to obtain multiple context reference data item feature vectors, and the multiple context The feature vectors of reference data items are cascaded into reference feature vectors to obtain multiple reference feature vectors; the multiple reference feature vectors are two-dimensionally arranged into
  • FIG3 is a schematic diagram of the scenario of step S2 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and dioxin removal according to an embodiment of the present application.
  • multiple sets of reference data for example, D1 as shown in FIG3) and slurry detection data (for example, D2 as shown in FIG3) are first obtained respectively, wherein each set of reference data includes the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry and the actual reference value of the reference slurry, and the slurry detection data includes the fiber length of the slurry, the pH value of the slurry and the solid content of the slurry.
  • the multiple sets of reference data and the slurry detection data are input into the central control unit (for example, S as shown in FIG3), wherein the central control unit can generate a decoding value for representing the reference value of the slurry.
  • FIG4 is a flow chart of the sub-steps of step S2 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and dioxin removal according to an embodiment of the present application.
  • step S2 of the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and dioxin removal according to an embodiment of the present application includes the following steps: S21: obtaining multiple groups of reference data, each group of reference data including the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry and the actual reference value of the reference slurry; S22: passing each group of reference data in the multiple groups of reference data through a context encoder including an embedding layer to obtain multiple context reference data item feature vectors, and cascading the multiple context reference data item feature vectors into a reference feature vector to obtain multiple reference feature vectors; S23: converting the multiple reference feature vectors into a reference feature vector; S21: obtaining
  • FIG5 is a schematic diagram of the sub-step architecture of step S2 in the method for preparing a ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins according to an embodiment of the present application.
  • each group of reference data includes the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry and the actual reference value of the reference slurry; then, each group of reference data in the multiple groups of reference data is passed through a context encoder including an embedding layer to obtain multiple context reference data item feature vectors, and the multiple context reference data item feature vectors are cascaded into reference feature vectors to obtain multiple reference feature vectors; then, the multiple reference feature vectors are arranged in two dimensions into a reference feature matrix and then passed through A convolutional neural network model including multiple mixed convolutional layers is used to obtain a reference feature matrix; then, the detection data of the slurry is obtained, and the
  • each group of reference data includes the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry, and the actual reference value of the reference slurry.
  • an artificial intelligence control algorithm based on deep learning is used to extract the global multi-scale implicit correlation features of multiple groups of reference data of the slurry as a reference feature matrix, and based on the global implicit correlation features of the actual detection data of the slurry, feature query is performed from the reference feature matrix to determine the reference value of the slurry.
  • the reference value of the slurry can be accurately determined according to the actual parameters of the slurry to improve the efficiency of preparation.
  • each group of reference data in the plurality of groups of reference data is passed through a context encoder including an embedding layer to obtain a plurality of context reference data item feature vectors, and the plurality of context reference data item feature vectors are cascaded into a reference feature vector to obtain a plurality of reference feature vectors.
  • each group of reference data in the plurality of groups of reference data is further encoded through a context encoder including an embedding layer to extract the global context correlation features between each data item in each group of reference data to be more suitable for characterizing the implicit feature information of the slurry, thereby obtaining a plurality of context reference data item feature vectors.
  • the plurality of context reference data item feature vectors are further cascaded into a reference feature vector to integrate the global feature information of each data item in each group of reference data in the plurality of groups of reference data to obtain a plurality of reference feature vectors.
  • the step S22 includes: S221, passing each group of reference data in the multiple groups of reference data through the embedding layer of the context encoder to convert each group of reference data in the multiple groups of reference data into an embedding vector to obtain a sequence of reference data embedding vectors; S222, inputting the sequence of reference data embedding vectors into the context encoder including the embedding layer to obtain multiple context reference data item feature vectors; and, S223, cascading the multiple context reference data item feature vectors into a reference feature vector to obtain multiple reference feature vectors.
  • the reference feature matrix is obtained by arranging the multiple reference feature vectors in two dimensions into a reference feature matrix and then passing through a convolutional neural network model including multiple hybrid convolutional layers.
  • the reference feature matrix is arranged in two dimensions into a reference feature matrix and then processed in a convolutional neural network model including multiple hybrid convolutional layers to extract multi-scale implicit correlation features of global features of each data item of each group of reference data in the reference feature matrix, thereby obtaining multiple multi-scale time-frequency feature vectors.
  • the step S23 includes: using each mixed convolution layer of the convolutional neural network model to perform the following on the input data in the forward pass of the layer: performing multi-scale convolution processing on the input data to obtain a multi-scale convolution feature map; performing pooling processing on the multi-scale convolution feature map along the local channel dimension to obtain a pooled feature map; and performing nonlinear activation processing on the pooled feature map to obtain an activated feature map; wherein the output of the last mixed convolution layer of the convolutional neural network model is the reference feature matrix.
  • the multi-scale convolution processing of the input data to obtain a multi-scale convolution feature map includes: using a first convolution kernel with a first size to perform convolution processing on the input data to obtain a first feature map; using a second convolution kernel with a first void ratio to perform convolution processing on the input data to obtain a second feature map; using a third convolution kernel with a second void ratio to perform convolution processing on the input data to obtain a third feature map; using a fourth convolution kernel with a third void ratio to perform convolution processing on the input data to obtain a fourth feature map; cascading the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the multi-scale convolution feature map.
  • the design of this module includes four parallel branches, which are composed of a common convolution layer with a convolution kernel size of 3 ⁇ 3 and three dilated convolution layers with a convolution kernel size of 3 ⁇ 3.
  • the reference feature matrix is operated respectively, and the expansion rates of the three branches of dilated convolution are set to 2, 3, and 4 respectively.
  • image information of different receptive fields can be obtained, and feature maps of different scales can be obtained.
  • While expanding the receptive field downsampling loss information is avoided.
  • the 4 branch feature maps are fused to make the sampling more intensive, with both high-level features and no additional parameters. In this way, the data association feature library of the reference slurry can be constructed to facilitate subsequent feature queries.
  • step S24 the detection data of the slurry is acquired, and the detection data of the slurry includes the fiber length of the slurry, the pH value of the slurry and the solid content of the slurry.
  • the detection data of the slurry is passed through the context encoder including the embedding layer to obtain a detection feature vector.
  • a feature query can be performed in the constructed data association feature library of the reference slurry based on the global feature association information between the various data items of the actual detection data of the slurry, so as to determine the reference value for the actual slurry. That is, specifically, first, the detection data of the slurry is obtained, and the detection data of the slurry includes the fiber length of the slurry, the pH value of the slurry, and the solid content of the slurry. Then, the detection data of the slurry is encoded through the context encoder including the embedding layer to extract the globally implicit association features between the various data items of the detection data of the slurry, thereby obtaining a detection feature vector.
  • the detection feature vector is used as a query feature vector and multiplied with the reference feature matrix to obtain a decoding feature vector.
  • the detection feature vector is then used as a query feature vector and multiplied with the reference feature matrix to query the slurry blanking reference value characteristics corresponding to the associated feature information of each data item under the actual detection data of the slurry, and decoding regression is performed on this to obtain a decoding value for representing the blanking reference value of the slurry.
  • the blanking reference value can be accurately determined according to the actual parameters of the slurry to improve the efficiency of preparation.
  • step S27 the decoded feature vector is decoded and regressed by a decoder to obtain a decoded value representing a reference value for feeding the slurry.
  • step S100 is also included: training the context encoder and the decoder including the embedding layer; wherein the step S100 includes: S110: obtaining multiple groups of training reference data, each group of training reference data including the training fiber length of the reference slurry, the training pH value of the reference slurry, the training solid content of the reference slurry and the actual reference value of the reference slurry; S120: passing each group of training reference data in the multiple groups of training reference data through the context encoder including the embedding layer to obtain multiple training context reference data item feature vectors, and cascading the multiple training context reference data item feature vectors into training reference feature vectors to obtain multiple training reference feature vectors; S130: arranging the multiple training reference feature vectors in two dimensions into a training reference feature matrix and then passing through the multiple mixed convolutional layers convolutional neural network model to obtain
  • the convolutional neural network model including multiple mixed convolutional layers can extract multi-scale local correlation features between the multiple reference feature vectors, but it is still expected to improve the expression effect of the reference feature matrix for the global correlation features of the multiple reference feature vectors. That is, considering that the global feature distribution of the reference feature matrix is obtained through their respective local feature distributions after the multiple reference feature vectors are arranged in two dimensions, it is necessary to improve the correlation between the local feature distribution and the global feature distribution in the decoding target domain.
  • a ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins includes: a ceramic composite fiber tube body and a catalyst formed on the surface of the ceramic composite fiber tube body, and the catalyst is a vanadium-titanium composite oxide.
  • Its preparation method includes using an artificial intelligence control algorithm based on deep learning to extract global multi-scale implicit correlation features of multiple groups of reference data of the slurry as a reference feature matrix, and performing feature query from the reference feature matrix based on the global implicit correlation features of the actual detection data of the slurry, so as to determine the reference value of the slurry for the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins.
  • the reference value for cutting can be accurately determined according to the actual parameters of the slurry, thereby improving the efficiency of preparation.
  • FIG8 is a block diagram of a system for preparing a blank tube of a ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins according to an embodiment of the present application.
  • a system 20 for preparing a blank tube of a ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins includes: a reference data acquisition module 21 for acquiring multiple groups of reference data, each group of reference data including the fiber length of the reference slurry, the pH value of the reference slurry, the solid content of the reference slurry and the actual reference value of the reference slurry; a reference context encoding module 22 for passing each group of reference data in the multiple groups of reference data through a context encoder including an embedding layer to obtain multiple context reference data item feature vectors, and cascading the multiple context reference data item feature vectors into a reference feature vector to obtain multiple reference feature vectors; a mixed convolution module 23 for two
  • the reference context encoding module 22 is further used to: pass each group of reference data in the multiple groups of reference data through the embedding layer of the context encoder to convert each group of reference data in the multiple groups of reference data into an embedding vector to obtain a sequence of reference data embedding vectors; input the sequence of reference data embedding vectors into the context encoder including the embedding layer to obtain multiple context reference data item feature vectors; and cascade the multiple context reference data item feature vectors into a reference feature vector to obtain multiple reference feature vectors.
  • the hybrid convolution module 23 is further used to: use the various hybrid convolution layers of the convolutional neural network model to perform the following on the input data in the forward pass of the layer: perform multi-scale convolution processing on the input data to obtain a multi-scale convolution feature map; perform pooling processing on the multi-scale convolution feature map along the local channel dimension to obtain a pooling feature map; and perform nonlinear activation processing on the pooling feature map to obtain an activation feature map; wherein the output of the last hybrid convolution layer of the convolutional neural network model is the reference feature matrix.
  • the multi-scale convolution processing of the input data to obtain a multi-scale convolution feature map includes: using a first convolution kernel having a first size to perform convolution processing on the input data to obtain a first feature map; using a second convolution kernel having a first void ratio to perform convolution processing on the input data to obtain a second feature map; using a third convolution kernel having a second void ratio to perform convolution processing on the input data to obtain a third feature map; using a fourth convolution kernel having a third void ratio to perform convolution processing on the input data to obtain a fourth feature map; cascading the first feature map, the second feature map, the third feature map and the fourth feature map to obtain the multi-scale convolution feature map.
  • the training module for training the context encoder containing the embedding layer and the decoder; wherein the training module includes: a training reference data acquisition module, used to obtain multiple groups of training reference data, each group of training reference data includes the training fiber length of the reference slurry, the training pH value of the reference slurry, the training solid content of the reference slurry and the actual reference value of the reference slurry; a training reference context encoding module, used to pass each group of training reference data in the multiple groups of training reference data through the context encoder containing the embedding layer to obtain multiple training context reference data item feature vectors, and cascade the multiple training context reference data item feature vectors into training reference feature vectors to obtain multiple training reference feature vectors; a training mixed convolution module, used to arrange the multiple training reference feature vectors in two dimensions into a training reference
  • the system 20 for preparing the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins can be implemented in various wireless terminals, such as a server for preparing the blank tube algorithm of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins, etc.
  • the system 20 for preparing the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins according to the embodiment of the present application can be integrated into the wireless terminal as a software module and/or a hardware module.
  • the system 20 for preparing the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the system 20 for preparing the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitration and removal of dioxins can also be one of the many hardware modules of the wireless terminal.
  • the system 20 for preparing the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins and the wireless terminal may also be separate devices, and the system 20 for preparing the blank tube of the ceramic composite fiber catalytic filter tube for desulfurization, denitrification and removal of dioxins may be connected to the wireless terminal through a wired and/or wireless network, and transmit interactive information in accordance with an agreed data format.

Landscapes

  • Chemical & Material Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Environmental & Geological Engineering (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Analytical Chemistry (AREA)
  • General Chemical & Material Sciences (AREA)
  • Oil, Petroleum & Natural Gas (AREA)
  • Manufacturing & Machinery (AREA)
  • Catalysts (AREA)
  • Exhaust Gas Treatment By Means Of Catalyst (AREA)

Abstract

本申请公开了一种用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法。所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管包括：陶瓷复合纤维管体以及形成于所述陶瓷复合纤维管体的表面的催化剂，所述催化剂为钒钛复合氧化物。其制备方法中包括采用基于深度学习的人工智能控制算法来提取出浆料的多组参考数据的基于全局的多尺度隐含关联特征作为参考特征矩阵，并基于浆料实际检测数据的全局隐含关联特征来从所述参考特征矩阵中进行特征查询，以此来进行所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的浆料的下料参考值的确定。这样，可以根据浆料的实际参数精准地确定下料参考值，提高制备的效率。

Description

用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法 技术领域
本申请涉及催化滤管技术领域，且更为具体地，涉及一种用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法。
 背景技术
近年来随着国家环保要求日趋严格，烟气脱硫脱硝去除二噁英一体化工艺成为国内烟气治理的研发热点。经济方面，因陶瓷纤维耐高温性能好，在高温过滤过程中，省去降温成本且热能回收创造利益；运行平稳，节约大量的人力维护与运行成本；安全方面，陶瓷纤维管坚固、耐温不燃烧，避免布袋破损与意外烧毁的危险；长效方面，耐腐蚀、耐磨损，催化剂不易中毒，使用寿命长达八年以上；回收利用方面，可水洗再生。符合国家超净排放要求，对烟气治理领域具有重要的经济、环保和社会效益。
 技术问题
但是，目前在对于用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管制备过程中，难以根据浆料的实际参数确定下料参考值，进而难以对需要制备的陶瓷复合纤维滤管的制备参数进行调整，导致制备的效率较低。
因此，期望一种优化的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方案。
 技术解决方案
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脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法。所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管包括：陶瓷复合纤维管体以及形成于所述陶瓷复合纤维管体的表面的催化剂，所述催化剂为钒钛复合氧化物。其制备方法中包括采用基于深度学习的人工智能控制算法来提取出浆料的多组参考数据的基于全局的多尺度隐含关联特征作为参考特征矩阵，并基于浆料实际检测数据的全局隐含关联特征来从所述参考特征矩阵中进行特征查询，以此来进行所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的浆料的下料参考值的确定。这样，可以根据浆料的实际参数精准地确定下料参考值，提高制备的效率。
根据本申请的一个方面，提供了一种用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管，其包括：
陶瓷复合纤维管体；以及
形成于所述陶瓷复合纤维管体的表面的催化剂，所述催化剂为钒钛复合氧化物。
在上述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管中，所述钒钛复合氧化物的组分为：铂0.5-1.5%，五氧化二钒2-5%，三氧化钨0.5%-3%，稀土1%-3%，钛白粉0.2%-0.5%，硫脲 0.2%-0.6%，吐温60 0.3%-0.5%，分散剂0.1%-0.5%以及纯水90-93.4%。
根据本申请的另一个方面，提供了一种用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其包括：
步骤S1：对纤维进行预处理以得到浆料；
步骤S2：将所述浆料通过模具上方的压力注浆口注入到陶瓷复合纤维滤管的长度、厚度自动调节模具中以得到陶瓷复合纤维滤管的坯管；
步骤S3：中控单元控制关闭所述压力注浆口上方设置的第一电磁阀，并通过所述中控单元控制真空泵对所述陶瓷复合纤维滤管的坯管进行抽吸，以得到定型陶瓷复合纤维滤管坯管；
步骤S4：将所述定型陶瓷复合纤维滤管坯管放置在催化剂溶胶中，并在真空条件下进行浸泡、晾干、干燥及烧结以得到有催化功能的陶瓷复合纤维滤管；以及
步骤S5：将所述有催化功能的陶瓷复合纤维滤管在烘房烘干后以得到陶瓷复合纤维催化滤管。
在上述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述步骤S2，包括：
S21：获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值；
S22：将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量；
S23：将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵；
S24：获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量；
S25：将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量；
S26：以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量；以及
S27：将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值。
在上述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述步骤S22，包括：
将所述多组参考数据中各组参考数据分别通过所述上下文编码器的嵌入层以将所述多组参考数据中各组参考数据分别转化为嵌入向量以得到参考数据嵌入向量的序列；
将所述参考数据嵌入向量的序列输入所述包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量；以及
将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量。
在上述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述步骤S23，包括：使用所述卷积神经网络模型的各个混合卷积层在层的正向传递中分别对输入数据进行：
对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图；
对所述多尺度卷积特征图进行沿局部通道维度的池化处理以得到池化特征图；以及
对所述池化特征图进行非线性激活处理以得到激活特征图；
其中，所述卷积神经网络模型的最后一个混合卷积层的输出为所述参考特征矩阵。
在上述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图，包括：
使用具有第一尺寸的第一卷积核对所述输入数据进行卷积处理以得到第一特征图；
使用具有第一空洞率的第二卷积核对所述输入数据进行卷积处理以得到第二特征图；
使用具有第二空洞率的第三卷积核对所述输入数据进行卷积处理以得到第三特征图；
使用具有第三空洞率的第四卷积核对所述输入数据进行卷积处理以得到第四特征图；
将所述第一特征图、所述第二特征图、所述第三特征图和所述第四特征图进行级联聚合以得到所述多尺度卷积特征图。
在上述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，还包括步骤S100：对所述包含嵌入层的上下文编码器和所述解码器进行训练；
其中，所述步骤S100，包括：
S110：获取多组训练参考数据，每组训练参考数据包括参考浆料的训练纤维长度、所述参考浆料的训练PH值、所述参考浆料的训练固相含量以及所述参考浆料的下料真实参考值；
S120：将所述多组训练参考数据中各组训练参考数据通过所述包含嵌入层的上下文编码器以得到多个训练上下文参考数据项特征向量，并将所述多个训练上下文参考数据项特征向量进行级联为训练参考特征向量以得到多个训练参考特征向量；
S130：将所述多个训练参考特征向量进行二维排列为训练参考特征矩阵后通过所述包含多个混合卷积层的卷积神经网络模型以得到训练参考特征矩阵；
S140：获取所述浆料的训练检测数据，所述浆料的训练检测数据包括浆料的训练纤维长度、浆料的训练PH值和浆料的训练固相含量，以及，所述浆料的下料参考值的真实值；
S150：将所述浆料的训练检测数据通过所述包含嵌入层的上下文编码器以得到训练检测特征向量；
S160：以所述训练检测特征向量作为查询特征向量与所述训练参考特征矩阵进行相乘以得到训练解码特征向量；
S170：将所述训练解码特征向量通过所述解码器以得到解码损失函数值；
S180：计算所述多个训练参考特征向量的多分布二元回归质量损失函数值；以及
S190：计算所述解码损失函数值和所述多分布二元回归质量损失函数值的加权和作为损失函数值来对所述包含嵌入层的上下文编码器和所述解码器进行训练。
 有益效果
与现有技术相比，本申请提供的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法，所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管包括：陶瓷复合纤维管体以及形成于所述陶瓷复合纤维管体的表面的催化剂，所述催化剂为钒钛复合氧化物。其制备方法中包括采用基于深度学习的人工智能控制算法来提取出浆料的多组参考数据的基于全局的多尺度隐含关联特征作为参考特征矩阵，并基于浆料实际检测数据的全局隐含关联特征来从所述参考特征矩阵中进行特征查询，以此来进行所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的浆料的下料参考值的确定。这样，可以根据浆料的实际参数精准地确定下料参考值，提高制备的效率。
 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脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的结构示意图。
图2为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法的流程图。
图3为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S2的场景示意图。
图4为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S2的子步骤流程图。
图5为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S2的子步骤架构示意图。
图6为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S22的子步骤流程图。
图7为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中进一步包括的步骤S100的子步骤流程图。
图8为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统的框图。
 本发明的实施方式
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
如上所述，近年来随着国家环保要求日趋严格，烟气脱硫脱硝去除二噁英一体化工艺成为国内烟气治理的研发热点。经济方面，因陶瓷纤维耐高温性能好，在高温过滤过程中，省去降温成本且热能回收创造利益；运行平稳，节约大量的人力维护与运行成本；安全方面，陶瓷纤维管坚固、耐温不燃烧，避免布袋破损与意外烧毁的危险；长效方面，耐腐蚀、耐磨损，催化剂不易中毒，使用寿命长达八年以上；回收利用方面，可水洗再生。符合国家超净排放要求，对烟气治理领域具有重要的经济、环保和社会效益。
但是，目前在对于用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管制备过程中，难以根据浆料的实际参数确定下料参考值，进而难以对需要制备的陶瓷复合纤维滤管的制备参数进行调整，导致制备的效率较低。因此，期望一种优化的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方案。
具体地，图1为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的结构示意图，如图1所示，所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管10，包括：陶瓷复合纤维管体12；以及，形成于所述陶瓷复合纤维管体的表面的催化剂13，所述催化剂为钒钛复合氧化物，特别地，所述钒钛复合氧化物的组分为：铂 0.5-1.5%，五氧化二钒 2-5%，三氧化钨 0.5%-3%，稀土1%-3%，钛白粉0.2%-0.5%，硫脲 0.2%-0.6%，吐温600.3%-0.5%，分散剂0.1%-0.5%以及纯水90-93.4%。在一个示例中，所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管10还包括固定于所述陶瓷复合纤维管体12上的端体11，所述端体11的直径大于所述陶瓷复合纤维管体12的直径。
更具体地，图2为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法的流程图，如图2所示，在本申请的技术方案中，所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，包括：步骤S1：对纤维进行预处理以得到浆料；步骤S2：将所述浆料通过模具上方的压力注浆口注入到陶瓷复合纤维滤管的模具中以得到陶瓷复合纤维滤管的坯管，陶瓷复合纤维滤管的长度0.5~9m，厚度5~50mm，直径50~200mm；步骤S3：中控单元控制关闭所述压力注浆口上方设置的第一电磁阀，并通过所述中控单元控制真空泵对所述陶瓷复合纤维滤管的坯管进行抽吸，以得到定型陶瓷复合纤维滤管坯管；步骤S4：将所述定型陶瓷复合纤维滤管坯管放置在催化剂溶胶中，并在真空条件下进行浸泡、晾干、干燥及烧结以得到有催化功能的陶瓷复合纤维滤管；以及，步骤S5：将所述有催化功能的陶瓷复合纤维滤管在烘房烘干后以得到陶瓷复合纤维催化滤管。在本申请实施例中，上述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管还具备除尘功能，具体地，该陶瓷复合纤维催化滤管的孔隙率高，气阻小，二噁英、粉尘颗粒物等被截留在陶瓷复合纤维催化滤管的表面，粉尘过滤效率可以达到99%以上，经过除尘后气体经过陶瓷复合纤维催化滤管的多孔元件，在催化剂的作用下排出，除尘效果显著。
相应地，在所述步骤S2中，在注浆之前，需要将获取到的浆料的纤维长度、浆料的PH值和浆料的固相含量预先输入至中控单元，所述中控单元根据浆料的纤维长度、浆料的PH值和浆料的固相含量确定浆料的下料参考值，所述中控单元再根据下料参考值对设置在模具左侧与压力注浆口连接的注浆单元的注浆压力和注浆压力保持时间，以及设置在模具下方的真空泵的抽吸压力和抽吸时间，进而再进行注浆。
基于此，考虑到在中控单元根据浆料的纤维长度、浆料的PH值和浆料的固相含量确定浆料的下料参考值的过程中，由于浆料的各个参数之间具有着关联性的关系，因此，在实际的操作过程中难以根据浆料的实际参数确定下料参考值，进而难以对需要制备的陶瓷复合纤维滤管的制备参数进行调整，导致制备的效率较低。因此，在本申请的技术方案中，采用基于深度学习的人工智能控制算法来提取出浆料的多组参考数据的基于全局的多尺度隐含关联特征作为参考特征矩阵，并基于浆料实际检测数据的全局隐含关联特征来从所述参考特征矩阵中进行特征查询，以此来进行浆料的下料参考值的确定。这样，能够根据浆料的实际参数精准地确定下料参考值，以提高制备的效率。
具体地，在本申请的技术方案中，首先，获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值。然后，考虑到在所述多组参考数据中的各组参考数据的各个数据项之间都具有着关联性关系，为了能够充分提取出所述各组参考数据中的各个数据项与所述参考浆料的下料真实参考值间的隐含关联特征，以此来准确地进行浆料的下料参考值确定，在本申请的技术方案中，进一步将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器中进行编码，以提取出所述各组参考数据中各个数据项间的基于全局的上下文关联特征以更适于表征所述浆料的隐含特征信息，从而得到多个上下文参考数据项特征向量。接着，进一步再将所述多个上下文参考数据项特征向量进行级联为参考特征向量，以整合所述多组参考数据中的各组参考数据的各个数据项的全局特征信息以得到多个参考特征向量。
进一步地，将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型中进行处理，以提取出所述参考特征矩阵中关于所述各组参考数据的各个数据项的全局特征的多尺度隐含关联特征，从而得到多个多尺度时频特征向量。也就是，在本申请的一个具体示例中，在所述混合卷积层中，此模块的设计包括并联的四个分支，由一个卷积核大小为3×3的普通卷积层以及三个卷积核大小为3×3的空洞卷积层构成，分别对所述参考特征矩阵进行操作，将空洞卷积三个分支的扩张率分别设置为2、3、4，通过不同扩张率的设置可获得不同感受域的图像信息，即可得到不同尺度的特征图，在扩大感受野的同时，又避免了下采样损失信息，接着将 4 个分支特征图进行融合，使得采样更为密集，既拥有了高层特征，也没有增加额外的参数量。这样，能够以此来构建所述参考浆料的数据关联特征库以便于后续的特征查询。
然后，在对于实际浆料的下料参考值进行确定时，可以基于所述浆料的实际检测数据的各个数据项间的全局特征关联信息来在构建的所述参考浆料的数据关联特征库中进行特征查询，以此来进行实际浆料的下料参考值确定。也就是，具体地，首先，获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量。然后，将所述浆料的检测数据通过所述包含嵌入层的上下文编码器中进行编码，以提取出所述浆料的检测数据的各个数据项间的基于全局的隐含关联特征，从而得到检测特征向量。
进一步地，再以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘，以查询出对应于所述浆料的实际检测数据下的各个数据项的关联特征信息的浆料的下料参考值特征，并以此来进行解码回归来得到用于表示所述浆料的下料参考值的解码值。这样，能够根据浆料的实际参数精准地确定下料参考值，以提高制备的效率。
特别地，在本申请的技术方案中，将所述多个参考特征向量进行二维排列后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵时，所述包含多个混合卷积层的卷积神经网络模型可以提取所述多个参考特征向量之间的多尺度局域的关联特征，但是，仍然期望提高所述参考特征矩阵对于所述多个参考特征向量的全局关联特征的表达效果。
也就是，考虑到所述多个参考特征向量在进行二维排列后，是通过其各自的局部特征分布来得到所述参考特征矩阵的全局特征分布，因此，需要提升所述局部特征分布在解码目标域下与所述全局特征分布的相关性。
基于此，本申请提供了一种用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，将所述浆料通过模具上方的压力注浆口注入到陶瓷复合纤维滤管的模具中以得到陶瓷复合纤维滤管的坯管，陶瓷复合纤维滤管的长度0.5~9m，厚度5~50mm，直径50~200mm，其包括：获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值；将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量；将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵；获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量；将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量；以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量；以及，将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值。
图3为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S2的场景示意图。如图3所示，在该应用场景中，先分别获取多组参考数据（例如，如图3中所示意的D1）和浆料的检测数据（例如，如图3中所示意的D2），其中，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量。然后，将所述多组参考数据和所述浆料的检测数据输入至中控单元（例如，如图3中所示意的S）中，其中，所述中控单元能够生成用于表示所述浆料的下料参考值的解码值。
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。
示例性方法
图4为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S2的子步骤流程图。如图4所示，根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法的步骤S2，包括步骤：S21：获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值；S22：将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量；S23：将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵；S24：获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量；S25：将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量；S26：以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量；以及，S27：将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值。
图5为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中步骤S2的子步骤架构示意图。如图5所示，在该网络架构中，首先，获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值；接着，将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量；然后，将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵；接着，获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量；然后，将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量；接着，以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量；最后，将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值。
更具体地，在步骤S21中，获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值。考虑到在中控单元根据浆料的纤维长度、浆料的PH值和浆料的固相含量确定浆料的下料参考值的过程中，由于浆料的各个参数之间具有着关联性的关系，因此，在实际的操作过程中难以根据浆料的实际参数确定下料参考值，进而难以对需要制备的陶瓷复合纤维滤管的制备参数进行调整，导致制备的效率较低。因此，在本申请的技术方案中，采用基于深度学习的人工智能控制算法来提取出浆料的多组参考数据的基于全局的多尺度隐含关联特征作为参考特征矩阵，并基于浆料实际检测数据的全局隐含关联特征来从所述参考特征矩阵中进行特征查询，以此来进行浆料的下料参考值的确定。这样，能够根据浆料的实际参数精准地确定下料参考值，以提高制备的效率。
更具体地，在步骤S22中，将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量。考虑到在所述多组参考数据中的各组参考数据的各个数据项之间都具有着关联性关系，为了能够充分提取出所述各组参考数据中的各个数据项与所述参考浆料的下料真实参考值间的隐含关联特征，以此来准确地进行浆料的下料参考值确定，在本申请的技术方案中，进一步将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器中进行编码，以提取出所述各组参考数据中各个数据项间的基于全局的上下文关联特征以更适于表征所述浆料的隐含特征信息，从而得到多个上下文参考数据项特征向量。接着，进一步再将所述多个上下文参考数据项特征向量进行级联为参考特征向量，以整合所述多组参考数据中的各组参考数据的各个数据项的全局特征信息以得到多个参考特征向量。
相应地，在一个具体示例中，如图6所示，在所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述步骤S22，包括：S221，将所述多组参考数据中各组参考数据分别通过所述上下文编码器的嵌入层以将所述多组参考数据中各组参考数据分别转化为嵌入向量以得到参考数据嵌入向量的序列；S222，将所述参考数据嵌入向量的序列输入所述包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量；以及，S223，将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量。
更具体地，在步骤S23中，将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵。将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型中进行处理，以提取出所述参考特征矩阵中关于所述各组参考数据的各个数据项的全局特征的多尺度隐含关联特征，从而得到多个多尺度时频特征向量。
相应地，在一个具体示例中，在所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述步骤S23，包括：使用所述卷积神经网络模型的各个混合卷积层在层的正向传递中分别对输入数据进行：对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图；对所述多尺度卷积特征图进行沿局部通道维度的池化处理以得到池化特征图；以及，对所述池化特征图进行非线性激活处理以得到激活特征图；其中，所述卷积神经网络模型的最后一个混合卷积层的输出为所述参考特征矩阵。
相应地，在一个具体示例中，在所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，所述对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图，包括：使用具有第一尺寸的第一卷积核对所述输入数据进行卷积处理以得到第一特征图；使用具有第一空洞率的第二卷积核对所述输入数据进行卷积处理以得到第二特征图；使用具有第二空洞率的第三卷积核对所述输入数据进行卷积处理以得到第三特征图；使用具有第三空洞率的第四卷积核对所述输入数据进行卷积处理以得到第四特征图；将所述第一特征图、所述第二特征图、所述第三特征图和所述第四特征图进行级联聚合以得到所述多尺度卷积特征图。
也就是，在本申请的一个具体示例中，在所述混合卷积层中，此模块的设计包括并联的四个分支，由一个卷积核大小为3×3的普通卷积层以及三个卷积核大小为3×3的空洞卷积层构成，分别对所述参考特征矩阵进行操作，将空洞卷积三个分支的扩张率分别设置为2、3、4，通过不同扩张率的设置可获得不同感受域的图像信息，即可得到不同尺度的特征图，在扩大感受野的同时，又避免了下采样损失信息，接着将 4 个分支特征图进行融合，使得采样更为密集，既拥有了高层特征，也没有增加额外的参数量。这样，能够以此来构建所述参考浆料的数据关联特征库以便于后续的特征查询。
更具体地，在步骤S24中，获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量。
更具体地，在步骤S25中，将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量。在对于实际浆料的下料参考值进行确定时，可以基于所述浆料的实际检测数据的各个数据项间的全局特征关联信息来在构建的所述参考浆料的数据关联特征库中进行特征查询，以此来进行实际浆料的下料参考值确定。也就是，具体地，首先，获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量。然后，将所述浆料的检测数据通过所述包含嵌入层的上下文编码器中进行编码，以提取出所述浆料的检测数据的各个数据项间的基于全局的隐含关联特征，从而得到检测特征向量。
更具体地，在步骤S26中，以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量。再以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘，以查询出对应于所述浆料的实际检测数据下的各个数据项的关联特征信息的浆料的下料参考值特征，并以此来进行解码回归来得到用于表示所述浆料的下料参考值的解码值。这样，能够根据浆料的实际参数精准地确定下料参考值，以提高制备的效率。
更具体地，在步骤S27中，将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值。
相应地，在一个具体示例中，如图7所示，在所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法中，还包括步骤S100：对所述包含嵌入层的上下文编码器和所述解码器进行训练；其中，所述步骤S100，包括：S110：获取多组训练参考数据，每组训练参考数据包括参考浆料的训练纤维长度、所述参考浆料的训练PH值、所述参考浆料的训练固相含量以及所述参考浆料的下料真实参考值；S120：将所述多组训练参考数据中各组训练参考数据通过所述包含嵌入层的上下文编码器以得到多个训练上下文参考数据项特征向量，并将所述多个训练上下文参考数据项特征向量进行级联为训练参考特征向量以得到多个训练参考特征向量；S130：将所述多个训练参考特征向量进行二维排列为训练参考特征矩阵后通过所述包含多个混合卷积层的卷积神经网络模型以得到训练参考特征矩阵；S140：获取所述浆料的训练检测数据，所述浆料的训练检测数据包括浆料的训练纤维长度、浆料的训练PH值和浆料的训练固相含量，以及，所述浆料的下料参考值的真实值；S150：将所述浆料的训练检测数据通过所述包含嵌入层的上下文编码器以得到训练检测特征向量；S160：以所述训练检测特征向量作为查询特征向量与所述训练参考特征矩阵进行相乘以得到训练解码特征向量；S170：将所述训练解码特征向量通过所述解码器以得到解码损失函数值；S180：计算所述多个训练参考特征向量的多分布二元回归质量损失函数值；以及，S190：计算所述解码损失函数值和所述多分布二元回归质量损失函数值的加权和作为损失函数值来对所述包含嵌入层的上下文编码器和所述解码器进行训练。
特别地，在本申请的技术方案中，将所述多个参考特征向量进行二维排列后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵时，所述包含多个混合卷积层的卷积神经网络模型可以提取所述多个参考特征向量之间的多尺度局域的关联特征，但是，仍然期望提高所述参考特征矩阵对于所述多个参考特征向量的全局关联特征的表达效果。也就是，考虑到所述多个参考特征向量在进行二维排列后，是通过其各自的局部特征分布来得到所述参考特征矩阵的全局特征分布，因此，需要提升所述局部特征分布在解码目标域下与所述全局特征分布的相关性。这通常可以通过对每个参考特征向量引入作为超参数的加权因数来解决，但另一方面，超参数的设置将会增大模型的训练负担。因此，本申请的申请人考虑引入多分布二元回归质量损失函数。
综上，基于本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法。所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管包括：陶瓷复合纤维管体以及形成于所述陶瓷复合纤维管体的表面的催化剂，所述催化剂为钒钛复合氧化物。其制备方法中包括采用基于深度学习的人工智能控制算法来提取出浆料的多组参考数据的基于全局的多尺度隐含关联特征作为参考特征矩阵，并基于浆料实际检测数据的全局隐含关联特征来从所述参考特征矩阵中进行特征查询，以此来进行所述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的浆料的下料参考值的确定。这样，可以根据浆料的实际参数精准地确定下料参考值，提高制备的效率。
示例性系统
图8为根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统的框图。如图8所示，根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20，其包括：参考数据获取模块21，用于获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值；参考上下文编码模块22，用于将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量；混合卷积模块23，用于将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵；检测数据获取模块24，用于获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量；检测上下文编码模块25，用于将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量；解码特征向量计算模块26，用于以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量；以及，解码回归模块27，用于将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值。
在一个示例中，在上述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20中，所述参考上下文编码模块22，进一步用于：将所述多组参考数据中各组参考数据分别通过所述上下文编码器的嵌入层以将所述多组参考数据中各组参考数据分别转化为嵌入向量以得到参考数据嵌入向量的序列；将所述参考数据嵌入向量的序列输入所述包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量；以及，将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量。
在一个示例中，在上述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20中，所述混合卷积模块23，进一步用于：使用所述卷积神经网络模型的各个混合卷积层在层的正向传递中分别对输入数据进行：对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图；对所述多尺度卷积特征图进行沿局部通道维度的池化处理以得到池化特征图；以及，对所述池化特征图进行非线性激活处理以得到激活特征图；其中，所述卷积神经网络模型的最后一个混合卷积层的输出为所述参考特征矩阵。
在一个示例中，在上述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20中，所述对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图，包括：使用具有第一尺寸的第一卷积核对所述输入数据进行卷积处理以得到第一特征图；使用具有第一空洞率的第二卷积核对所述输入数据进行卷积处理以得到第二特征图；使用具有第二空洞率的第三卷积核对所述输入数据进行卷积处理以得到第三特征图；使用具有第三空洞率的第四卷积核对所述输入数据进行卷积处理以得到第四特征图；将所述第一特征图、所述第二特征图、所述第三特征图和所述第四特征图进行级联聚合以得到所述多尺度卷积特征图。
在一个示例中，在上述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20中，还包括：对所述包含嵌入层的上下文编码器和所述解码器进行训练的训练模块；其中，所述训练模块，包括：训练参考数据获取模块，用于获取多组训练参考数据，每组训练参考数据包括参考浆料的训练纤维长度、所述参考浆料的训练PH值、所述参考浆料的训练固相含量以及所述参考浆料的下料真实参考值；训练参考上下文编码模块，用于将所述多组训练参考数据中各组训练参考数据通过所述包含嵌入层的上下文编码器以得到多个训练上下文参考数据项特征向量，并将所述多个训练上下文参考数据项特征向量进行级联为训练参考特征向量以得到多个训练参考特征向量；训练混合卷积模块，用于将所述多个训练参考特征向量进行二维排列为训练参考特征矩阵后通过所述包含多个混合卷积层的卷积神经网络模型以得到训练参考特征矩阵；训练检测数据获取模块，用于获取所述浆料的训练检测数据，所述浆料的训练检测数据包括浆料的训练纤维长度、浆料的训练PH值和浆料的训练固相含量，以及，所述浆料的下料参考值的真实值；训练检测上下文编码模块，用于将所述浆料的训练检测数据通过所述包含嵌入层的上下文编码器以得到训练检测特征向量；训练解码特征向量计算模块，用于以所述训练检测特征向量作为查询特征向量与所述训练参考特征矩阵进行相乘以得到训练解码特征向量；训练解码损失计算模块，用于将所述训练解码特征向量通过所述解码器以得到解码损失函数值；多分布二元回归质量损失计算模块，用于计算所述多个训练参考特征向量的多分布二元回归质量损失函数值；以及，上下文编码器和解码器训练模块，用于计算所述解码损失函数值和所述多分布二元回归质量损失函数值的加权和作为损失函数值来对所述包含嵌入层的上下文编码器和所述解码器进行训练。
这里，本领域技术人员可以理解，上述用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20中的各个单元和模块的具体功能和操作已经在上面参考图3到图7的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法的步骤S2的描述中得到了详细介绍，并因此，将省略其重复描述。
如上所述，根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20可以实现在各种无线终端中，例如用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备算法的服务器等。在一个示例中，根据本申请实施例的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20可以作为一个软件模块和/或硬件模块而集成到无线终端中。例如，该用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20可以是该无线终端的操作系统中的一个软件模块，或者可以是针对于该无线终端所开发的一个应用程序；当然，该用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20同样可以是该无线终端的众多硬件模块之一。
替换地，在另一示例中，该用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20与该无线终端也可以是分立的设备，并且该用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的坯管的制备系统20可以通过有线和/或无线网络连接到该无线终端，并且按照约定的数据格式来传输交互信息。
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。
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 (6)

  1. 一种用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其特征在于，包括：
    步骤S1：对纤维进行预处理以得到浆料；
    步骤S2：将所述浆料通过模具上方的压力注浆口注入到陶瓷复合纤维滤管的模具中以得到陶瓷复合纤维滤管的坯管，其中陶瓷复合纤维滤管的长度0.5～9m，厚度5～50mm，直径50～200mm；
    步骤S3：中控单元控制关闭所述压力注浆口上方设置的第一电磁阀，并通过所述中控单元控制真空泵对所述陶瓷复合纤维滤管的坯管进行抽吸，以得到定型陶瓷复合纤维滤管坯管；
    步骤S4：将所述定型陶瓷复合纤维滤管坯管放置在催化剂溶胶中，并在真空条件下进行浸泡、晾干、干燥及烧结以得到有催化功能的陶瓷复合纤维滤管；以及
    步骤S5：将所述有催化功能的陶瓷复合纤维滤管在烘房烘干后以得到陶瓷复合纤维催化滤管；
    所述步骤S2，包括：
    S21：获取多组参考数据，每组参考数据包括参考浆料的纤维长度、所述参考浆料的PH值、所述参考浆料的固相含量以及所述参考浆料的下料真实参考值；
    S22：将所述多组参考数据中各组参考数据通过包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量，并将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量；
    S23：将所述多个参考特征向量进行二维排列为参考特征矩阵后通过包含多个混合卷积层的卷积神经网络模型以得到参考特征矩阵；
    S24：获取所述浆料的检测数据，所述浆料的检测数据包括浆料的纤维长度、浆料的PH值和浆料的固相含量；
    S25：将所述浆料的检测数据通过所述包含嵌入层的上下文编码器以得到检测特征向量；
    S26：以所述检测特征向量作为查询特征向量与所述参考特征矩阵进行相乘以得到解码特征向量；以及
    S27：将所述解码特征向量通过解码器进行解码回归以得到用于表示所述浆料的下料参考值的解码值；
    所述方法还包括步骤S100：对所述包含嵌入层的上下文编码器和所述解码器进行训练；
    其中，所述步骤S100，包括：
    S110：获取多组训练参考数据，每组训练参考数据包括参考浆料的训练纤维长度、所述参考浆料的训练PH值、所述参考浆料的训练固相含量以及所述参考浆料的下料真实参考值；
    S120：将所述多组训练参考数据中各组训练参考数据通过所述包含嵌入层的上下文编码器以得到多个训练上下文参考数据项特征向量，并将所述多个训练上下文参考数据项特征向量进行级联为训练参考特征向量以得到多个训练参考特征向量；
    S130：将所述多个训练参考特征向量进行二维排列为训练参考特征矩阵后通过所述包含多个混合卷积层的卷积神经网络模型以得到训练参考特征矩阵；
    S140：获取所述浆料的训练检测数据，所述浆料的训练检测数据包括浆料的训练纤维长度、浆料的训练PH值和浆料的训练固相含量，以及，所述浆料的下料参考值的真实值；
    S150：将所述浆料的训练检测数据通过所述包含嵌入层的上下文编码器以得到训练检测特征向量；
    S160：以所述训练检测特征向量作为查询特征向量与所述训练参考特征矩阵进行相乘以得到训练解码特征向量；
    S170：将所述训练解码特征向量通过所述解码器以得到解码损失函数值；
    S180：计算所述多个训练参考特征向量的多分布二元回归质量损失函数值；以及
    S190：计算所述解码损失函数值和所述多分布二元回归质量损失函数值的加权和作为损失函数值来对所述包含嵌入层的上下文编码器和所述解码器进行训练。
  2. 根据权利要求1所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其特征在于，所述陶瓷复合纤维催化滤管包括：
    陶瓷复合纤维管体；以及
    形成于所述陶瓷复合纤维管体的表面的催化剂，所述催化剂为钒钛复合氧化物。
  3. 根据权利要求2所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其特征在于，所述钒钛复合氧化物的组分为：铂0.5-1.5%，五氧化二钒2-5%，三氧化钨0.5%-3%，稀土1%-3%，钛白粉0.2%-0.5%，硫脲 0.2%-0.6%，吐温60 0.3%-0.5%，分散剂0.1%-0.5%以及纯水90-93.4%。
  4. 根据权利要求1所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其特征在于，所述步骤S22，包括：
    将所述多组参考数据中各组参考数据分别通过所述上下文编码器的嵌入层以将所述多组参考数据中各组参考数据分别转化为嵌入向量以得到参考数据嵌入向量的序列；
    将所述参考数据嵌入向量的序列输入所述包含嵌入层的上下文编码器以得到多个上下文参考数据项特征向量；以及
    将所述多个上下文参考数据项特征向量进行级联为参考特征向量以得到多个参考特征向量。
  5. 根据权利要求4所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其特征在于，所述步骤S23，包括：使用所述卷积神经网络模型的各个混合卷积层在层的正向传递中分别对输入数据进行：
    对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图；
    对所述多尺度卷积特征图进行沿局部通道维度的池化处理以得到池化特征图；以及
    对所述池化特征图进行非线性激活处理以得到激活特征图；
    其中，所述卷积神经网络模型的最后一个混合卷积层的输出为所述参考特征矩阵。
  6. 根据权利要求5所述的用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管的制备方法，其特征在于，所述对所述输入数据进行多尺度卷积处理以得到多尺度卷积特征图，包括：
    使用具有第一尺寸的第一卷积核对所述输入数据进行卷积处理以得到第一特征图；
    使用具有第一空洞率的第二卷积核对所述输入数据进行卷积处理以得到第二特征图；
    使用具有第二空洞率的第三卷积核对所述输入数据进行卷积处理以得到第三特征图；
    使用具有第三空洞率的第四卷积核对所述输入数据进行卷积处理以得到第四特征图；
    将所述第一特征图、所述第二特征图、所述第三特征图和所述第四特征图进行级联聚合以得到所述多尺度卷积特征图。
PCT/CN2024/074348 2022-11-29 2024-01-27 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法 WO2024114835A2 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211507981.X 2022-11-29
CN202211507981.XA CN115738747B (zh) 2022-11-29 2022-11-29 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法

Publications (2)

Publication Number Publication Date
WO2024114835A2 true WO2024114835A2 (zh) 2024-06-06
WO2024114835A3 WO2024114835A3 (zh) 2024-08-02

Family

ID=85340715

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/074348 WO2024114835A2 (zh) 2022-11-29 2024-01-27 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法

Country Status (3)

Country Link
CN (1) CN115738747B (zh)
WO (1) WO2024114835A2 (zh)
ZA (1) ZA202402155B (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115738747B (zh) * 2022-11-29 2024-01-23 浙江致远环境科技股份有限公司 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4528237B2 (ja) * 2005-05-12 2010-08-18 株式会社日立製作所 製品設計パラメータ決定支援システム
US7289859B2 (en) * 2005-09-30 2007-10-30 Hitachi, Ltd. Method for determining parameter of product design and its supporting system
US20130178383A1 (en) * 2008-11-12 2013-07-11 David Spetzler Vesicle isolation methods
CN106097355A (zh) * 2016-06-14 2016-11-09 山东大学 基于卷积神经网络的胃肠道肿瘤显微高光谱图像处理方法
CN110383299B (zh) * 2017-02-06 2023-11-17 渊慧科技有限公司 记忆增强的生成时间模型
KR102237189B1 (ko) * 2019-11-18 2021-04-07 유비엘바이오 주식회사 약물 효과 추정을 위한 기계학습 기반의 단백질-리간드 결합력 예측 모델 생성 장치 및 그 동작 방법
CN111582126B (zh) * 2020-04-30 2024-02-27 浙江工商大学 基于多尺度行人轮廓分割融合的行人重识别方法
CN111930732B (zh) * 2020-08-03 2022-06-28 浙江大学 基于级联卷积自编码器的缺失电力负荷数据修复方法及装置
CN113577916B (zh) * 2021-07-12 2022-04-12 浙江致远环境科技有限公司 一种陶瓷复合纤维催化滤管的制备工艺
CN115018068B (zh) * 2022-05-30 2023-02-17 福建天甫电子材料有限公司 用于光阻洗净液生产的自动配料系统及其配料方法
CN114974477B (zh) * 2022-06-07 2023-09-19 扬州华铁机电集团有限公司 一种用于双流向防雷器的镍铁电极材料优化方法及系统
CN115212790B (zh) * 2022-06-30 2023-04-07 福建天甫电子材料有限公司 用于光阻剥离液生产的自动配料系统及其配料方法
CN114900126B (zh) * 2022-07-13 2022-11-04 江苏福明太阳能有限公司 太阳能电池组件的接地测试设备及其接地测试方法
CN115738747B (zh) * 2022-11-29 2024-01-23 浙江致远环境科技股份有限公司 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法

Also Published As

Publication number Publication date
CN115738747A (zh) 2023-03-07
ZA202402155B (en) 2024-04-24
CN115738747B (zh) 2024-01-23
WO2024114835A3 (zh) 2024-08-02

Similar Documents

Publication Publication Date Title
WO2024114835A2 (zh) 用于脱硫脱硝去除二噁英的陶瓷复合纤维催化滤管及其制备方法
CN101822931B (zh) 富二氧化碳吸收剂溶液中空纤维膜接触器减压再生系统及方法
CN201684613U (zh) 微波活性炭再生装置
CN105688566A (zh) 一种烧结烟气脱硫脱硝装置及方法
JPS5827480B2 (ja) 希ガスホ−ルドアツプ装置の脱湿塔再生方法
CN110205427B (zh) 一种智能热风炉优化控制系统及方法
CN107715160B (zh) 实验室尾气生物气溶胶中气载内毒素的微波辐射处理方法
CN111675194A (zh) 一种稳定制氢气体质量的纯化装置及其纯化方法
CN203244899U (zh) 钨合金烧结炉的气体回收净化循环再利用系统
CN116020235A (zh) 一种制备超低露点超低二氧化碳含量空气的吸附干燥装置
CN216785727U (zh) 一种循环式制氧机
CN113230816A (zh) 吸附VOCs的活性炭微波脱附再生设备及再生方法
CN113720667A (zh) 一种环境监测用水样品处理系统及方法
JPS5515657A (en) Method of removing mal-odorous component in exhaust gas
CN215463090U (zh) 吸附VOCs的活性炭微波脱附再生设备
CN112179866B (zh) 基于光电分析的烟气检测装置和方法
CN208212850U (zh) 一种低浓度有机废气的浓缩装置
CN116127729B (zh) 基于线性动态模型的二氧化碳捕集精准预测方法及系统
CN108499308A (zh) 一种空气净化装置的进气量控制系统及方法
CN111085105A (zh) 丙烯腈催化剂焙烧尾气的处理方法和处理装置
CN219244373U (zh) 一种蒸汽冷凝液余热再利用系统
CN218495620U (zh) 一种一氧化碳回收利用系统
CN218653720U (zh) 一种有机溶剂回收处理装置
CN203699898U (zh) 可循环过滤的制氧机
CN213853131U (zh) 一种陶粉干燥尾气再循环系统