WO2024007445A1 - 用于氟化锂制备的能源管理控制系统及其控制方法 - Google Patents

用于氟化锂制备的能源管理控制系统及其控制方法 Download PDF

Info

Publication number
WO2024007445A1
WO2024007445A1 PCT/CN2022/117768 CN2022117768W WO2024007445A1 WO 2024007445 A1 WO2024007445 A1 WO 2024007445A1 CN 2022117768 W CN2022117768 W CN 2022117768W WO 2024007445 A1 WO2024007445 A1 WO 2024007445A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature
vector
feature vector
feature map
calciner
Prior art date
Application number
PCT/CN2022/117768
Other languages
English (en)
French (fr)
Inventor
华博文
雷炎芳
廖育能
黄吉华
陈三凤
吴仕显
胡新
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 WO2024007445A1 publication Critical patent/WO2024007445A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D23/00—Control of temperature
    • G05D23/19—Control of temperature characterised by the use of electric means
    • G05D23/20—Control of temperature characterised by the use of electric means with sensing elements having variation of electric or magnetic properties with change of temperature

Definitions

  • the present invention relates to the field of intelligent preparation control, and more specifically, to an energy management control system for lithium fluoride preparation and a control method thereof.
  • Lithium fluoride is an important lithium-based material that appears as white non-hygroscopic cubic crystals at room temperature.
  • high-purity lithium fluoride is widely used in the preparation of fluorinated glass and optical fibers.
  • high-purity lithium fluoride is also an important raw material for electrolyte materials for lithium-ion batteries.
  • Commonly used preparation methods of lithium fluoride mainly include direct preparation method, ion exchange preparation method and extraction preparation method. No matter which preparation scheme is adopted, energy management during the preparation process is a very important concern, especially in the current context of energy conservation, environmental protection, and carbon neutrality.
  • the preparation process of lithium fluoride mostly involves temperature control such as calcination and cooling. It should be understood that temperature control not only involves energy management, but also involves the quality of the final lithium fluoride product. Most of the existing lithium fluoride preparation solutions use preset temperature technical solutions, which is not only detrimental to energy optimization, but also detrimental to controlling the quality of the final lithium fluoride product.
  • an energy management control system for lithium fluoride preparation is expected to intelligently control the temperature of the calciner to optimize energy while ensuring the quality of the final lithium fluoride product.
  • Embodiments of the present application provide an energy management control system and a control method for the preparation of lithium fluoride, which use a convolutional neural network model based on deep learning to dynamically extract temperature-related features inside the calciner in real time. , and conduct in-depth exploration of the structural change characteristics and internal heat distribution characteristics of the calcined product, and then combine the time series characteristic information of these three to intelligently adjust the temperature of the calciner to ensure energy optimization while ensuring The quality of the final lithium fluoride product.
  • an energy management control system for lithium fluoride preparation which includes:
  • the calcining temperature acquisition module is used to obtain the calcining temperature values of the calciner at multiple predetermined time points through a temperature sensor deployed in the calciner, wherein the calciner is used to calcine crystals that form lithium fluoride finished products;
  • a product data acquisition module configured to acquire thermal infrared images and X-Ray scans of the calcined product at multiple predetermined time points through a thermal infrared camera and an X-Ray scanner deployed in the calciner;
  • a thermal infrared encoding module configured to pass the thermal infrared images of the calcined product at the plurality of predetermined time points through a first convolutional neural network using a spatial attention mechanism to obtain a thermal infrared feature map;
  • a perspective image encoding module configured to pass the X-Ray scan images of the calcined product at the plurality of predetermined time points through a second convolutional neural network using a three-dimensional convolution kernel to obtain an X-Ray feature map;
  • a correlation coding module used to cascade the thermal infrared feature map and the X-Ray feature map and then pass them through a third convolutional neural network as a filter to obtain a product feature vector;
  • An energy data encoding module used to pass the calcining temperature values of the calciner at multiple predetermined time points through a time series encoder including a one-dimensional convolution layer and a fully connected layer to obtain a temperature feature vector;
  • a feature fusion module used to fuse the product feature vector and the temperature feature vector to obtain a classification feature vector
  • An energy 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 that the temperature of the calciner at the current time point should be increased or decreased.
  • the thermal infrared encoding module is further used to use each layer of the first convolutional neural network to perform: on the input data in the forward transmission of the layer.
  • the input data is subjected to convolution processing based on a two-dimensional convolution kernel to generate a convolution feature map; the convolution feature map is pooled to generate a pooled feature map; and the pooled feature map is activated.
  • the generated feature map output by the last layer of the first convolutional neural network model is The thermal infrared characteristic map.
  • the perspective encoding module is further used for the second convolutional neural network using a three-dimensional convolution kernel to separately process the input data in the forward transmission of the layer.
  • the feature map is nonlinearly activated to obtain an activation feature map; wherein, the output of the last layer of the second convolutional neural network is the X-Ray feature map, and the output of the first layer of the second convolutional neural network is The input is an X-Ray scan of the calcined product at the multiple predetermined time points.
  • the correlation coding module the cascade unit, is used to cascade the thermal infrared feature map and the X-Ray feature map to obtain cascade features Figure; and.
  • a feature extraction unit configured to perform 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 as a filter to obtain the desired result from the input data.
  • the last layer of the third convolutional neural network generates the product feature vector, wherein the input of the first layer of the third convolutional neural network is the cascade feature map.
  • the energy data encoding module includes: an input vector construction unit for arranging the calcining temperature values of the calciner at multiple predetermined time points according to the time dimension. is a one-dimensional input vector; a fully connected coding unit, used to use the fully connected layer of the temporal encoder to fully connect the input vector with the following formula to extract the feature values of each position in the input vector High-dimensional latent features, where the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, Represents matrix multiplication; a one-dimensional convolution coding unit, used to use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution coding on the input vector according to the following formula to extract the position of each position in the input vector High-dimensional implicit correlation features between eigenvalues, where the formula is:
  • a is the width of the convolution kernel in the x direction
  • F is the convolution kernel parameter vector
  • G is the local vector matrix that operates with the convolution kernel function
  • w is the size of the convolution kernel.
  • the feature fusion module includes: a vector difference unit for calculating the difference feature vector between the product feature vector and the temperature feature vector; logarithm An arithmetic unit, used to calculate the logarithmic function value of the eigenvalues at each position in the differential eigenvector to obtain a logarithmic differential eigenvector; a depth information characteristic value calculation unit, used to calculate a norm of the differential eigenvector; A full-scene homography correlation matrix construction unit, used to calculate the product between the product feature vector and the transposed vector of the temperature feature vector to obtain a full-scene homography correlation matrix; a depth sensing unit, used to calculate the full-scene homography correlation matrix Frobenius norm of the scene homography correlation matrix; an alignment unit, used to use a norm of the differential feature vector as a weighting coefficient and use the Frobenius norm of the full scene homography correlation matrix as a bias term to align the pair
  • the differential feature vector is processed to obtain a corrected temperature feature vector
  • the alignment unit is used to use a norm of the differential feature vector as a weighting coefficient and a Frobenius norm of the full-scene homography correlation matrix as a
  • the offset term processes the logarithmic difference feature vector according to the following formula to obtain the corrected temperature feature vector;
  • the energy management result generation module is further configured to: use the classifier to process the classification feature vector with the following formula to obtain the classification result, Among them, the formula is: softmax ⁇ (W n ,B n ):...:(W 1 ,B 1 )
  • the above-mentioned energy management control system for lithium fluoride preparation also includes an energy control module for adjusting the temperature of the calciner based on the classification results.
  • a control method for an energy management control system for lithium fluoride preparation includes:
  • the calcining temperature values of the calciner at multiple predetermined time points are obtained through a temperature sensor deployed in the calciner, wherein the calciner is used to calcine crystals to form lithium fluoride finished products;
  • Thermal infrared images and X-Ray scans of the calcined product at the plurality of predetermined time points are obtained through a thermal infrared camera and an X-Ray scanner deployed in the calciner;
  • the thermal infrared feature map and the X-Ray feature map are cascaded and passed through a third convolutional neural network as a filter to obtain a product feature vector;
  • the classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate that the temperature of the calciner at the current time point should be increased or decreased.
  • the thermal infrared images of the calcined product at the plurality of predetermined time points are passed through the first convolutional neural network using the spatial attention mechanism to obtain the thermal infrared
  • the feature map includes: using each layer of the first convolutional neural network to perform a convolution process on the input data in the forward pass of the layer based on a two-dimensional convolution kernel to generate a convolution feature.
  • the X-Ray scanning images of the calcined products at multiple predetermined time points are passed through a second convolutional neural network using a three-dimensional convolution kernel to obtain
  • the X-Ray feature map includes: the second convolutional neural network using a three-dimensional convolution kernel performs three-dimensional convolution on the input data in the forward pass of the layer: three-dimensional convolution on the input data based on the three-dimensional convolution kernel Process to obtain a convolution feature map; perform pooling processing on the convolution feature map to obtain a pooled feature map; and perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein, the third The output of the last layer of the second convolutional neural network is the X-Ray feature map, and the input of the first layer of the second convolutional neural network is the X-Ray scan of the calcined product at the multiple predetermined time points. picture.
  • the thermal infrared feature map and the X-Ray feature map are cascaded and then passed through the third convolutional neural network as a filter to obtain
  • the product feature vector includes: cascading the thermal infrared feature map and the X-Ray feature map to obtain a cascade feature map; and using each layer of the third convolutional neural network as a filter in In the forward pass of the layer, the input data is convolved, pooled along the feature matrix, and activated to generate the product feature vector by the last layer of the third convolutional neural network, where the third convolutional neural network generates the product feature vector.
  • the input of the first layer of the three-convolutional neural network is the cascaded feature map.
  • the calcining temperature values of the calciner at multiple predetermined time points are passed through a time series encoder including a one-dimensional convolution layer and a fully connected layer to obtain
  • the temperature feature vector includes: arranging the calcining temperature values of the calciner at multiple predetermined time points into a one-dimensional input vector according to the time dimension; using the fully connected layer of the temporal encoder to calculate the input vector using the following formula Fully connected encoding is performed to extract high-dimensional hidden features of the eigenvalues at each position in the input vector, where the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, represents matrix multiplication; use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the input vector with the following formula to extract high-dimensional implicit correlation features between the eigenvalues of each position in the input vector, Among them, the formula is:
  • a is the width of the convolution kernel in the x direction
  • F is the convolution kernel parameter vector
  • G is the local vector matrix that operates with the convolution kernel function
  • w is the size of the convolution kernel.
  • fusing the product feature vector and the temperature feature vector to obtain a classification feature vector includes: calculating the product feature vector and the temperature feature vector differential eigenvectors between them; calculate the logarithmic function value of the eigenvalues at each position in the differential eigenvector to obtain a logarithmic differential eigenvector; calculate the norm of the differential eigenvector; calculate the product eigenvector and The product between the transposed vectors of the temperature feature vectors is used to obtain the full-scene homography correlation matrix; the Frobenius norm of the full-scene homography correlation matrix is calculated; a norm of the differential feature vector is used as a weighting coefficient; The logarithmic difference feature vector is processed using the Frobenius norm of the full-scene homography correlation matrix as a bias term to obtain a corrected temperature feature vector that is homogeneously aligned with the product feature vector in depth; calculate the The position-weighted sum of the corrected temperature feature vector and the
  • fusing the product feature vector and the temperature feature vector to obtain a classification feature vector includes: using a norm of the differential feature vector as a weight
  • the coefficients and the Frobenius norm of the full-scene homography correlation matrix are used as bias items to process the logarithmic difference eigenvector with the following formula to obtain the corrected temperature eigenvector;
  • passing the classification feature vector through a classifier to obtain a classification result includes: using the classifier to process the classification feature vector with the following formula To obtain the classification result, the formula is: softmax ⁇ (W n ,B n ):...:(W 1 ,B 1 )
  • the classification feature vector is passed through a classifier to obtain a classification result, and further includes: adjusting the temperature of the calciner based on the classification result.
  • the energy management control system and its control method provided by this application for the preparation of lithium fluoride use a convolutional neural network model based on deep learning to conduct real-time dynamic analysis of the temperature-related characteristics inside the calciner. extract, and conduct in-depth exploration of the structural change characteristics and internal heat distribution characteristics of the calcined product, and then combine the time series characteristic information of these three to intelligently adjust the temperature of the calciner to optimize energy. At the same time, the quality of the final lithium fluoride product is guaranteed.
  • Figure 1 is an application scenario diagram of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • FIG. 2 is a block diagram of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • Figure 3 is a block diagram of a feature fusion module in an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • Figure 4 is a flow chart of a control method of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • Figure 5 is a schematic structural diagram of a control method of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • lithium fluoride is an important lithium-based material that appears as white, non-hygroscopic cubic crystals at room temperature.
  • high-purity lithium fluoride is widely used in the preparation of fluorinated glass and optical fibers.
  • high-purity lithium fluoride is also an important raw material for electrolyte materials for lithium-ion batteries.
  • Commonly used preparation methods of lithium fluoride mainly include direct preparation method, ion exchange preparation method and extraction preparation method. No matter which preparation scheme is adopted, energy management during the preparation process is a very important concern, especially in the current context of energy conservation, environmental protection, and carbon neutrality.
  • the preparation process of lithium fluoride mostly involves temperature control such as calcination and cooling. It should be understood that temperature control not only involves energy management, but also involves the quality of the final lithium fluoride product. Most of the existing lithium fluoride preparation solutions use preset temperature technical solutions, which is not only detrimental to energy optimization, but also detrimental to controlling the quality of the final lithium fluoride product. Therefore, an energy management control system for lithium fluoride preparation is expected to intelligently control the temperature of the calciner to optimize energy while ensuring the quality of the final lithium fluoride product.
  • deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields.
  • deep learning and neural networks have also shown that they are close to or even beyond human performance in areas such as image classification, object detection, semantic segmentation, and text translation.
  • the inventor of the present application considered that if he wanted to intelligently and accurately control the temperature of the calciner to improve the quality of the finished lithium fluoride product, he would need to measure the temperature inside the calciner dynamically in real time, and also need to measure the calcining temperature.
  • the structural characteristics and heat distribution of the product were analyzed. Therefore, in this application, a temperature sensor is used to collect the heat information in the calciner, and a thermal infrared camera and an X-Ray scanner are used to collect the internal heat distribution and internal structural information of the calcined product. Further, through The convolutional neural network model based on deep learning is used to deeply mine the implicit correlation feature distribution, and then during classification, it can ensure that the temperature adjustment of the calciner at the current time point is more suitable for the preparation of lithium fluoride.
  • the calcining temperature values of the calciner at multiple predetermined time points are obtained through a temperature sensor deployed in the calciner, wherein the calciner is used to calcine the finished product of lithium fluoride. crystal.
  • the thermal infrared camera can collect to the internal heat distribution of the calcined product to more accurately control the temperature inside the calciner
  • the X-Ray scanner can collect the internal structure and shape change information of the calcined product, to better monitor the formation of lithium fluoride product.
  • a convolutional neural network model with excellent performance in extracting local latent features of the image is used to perform feature mining on the thermal infrared image of the calcined product.
  • a convolutional neural network model with excellent performance in extracting local latent features of the image is used to perform feature mining on the thermal infrared image of the calcined product.
  • the thermal infrared image of the calcined product When mining features, more attention should be paid to the heat change characteristics of the calcined product, and the interfering heat features around it need to be removed.
  • the first convolutional neural network of the spatial attention mechanism is used to Feature extraction is performed on the thermal infrared images of the calcined product at multiple predetermined time points to extract the local implicit heat feature distribution of the thermal infrared images of the calcined product at multiple predetermined time points, thereby obtaining a thermal infrared feature map.
  • a second convolutional neural network with a three-dimensional convolution kernel is used to process the X-Ray scans of the calcined product at the multiple predetermined time points, In order to obtain the X-Ray characteristic diagram with the dynamic change characteristics of the calcined product.
  • the thermal infrared feature map and the X-Ray feature map are further cascaded to integrate the heat characteristic distribution and the characteristic information of internal structural changes in the calcined product, and then the obtained cascade feature map is used as a filter Feature extraction is performed in the third convolutional neural network of the device to obtain the product feature vector.
  • a context encoder including an embedding layer is used to evaluate the calciner at The calcination temperature values at multiple predetermined time points to extract global-based high-dimensional semantic features between the calcination temperature values at the multiple predetermined time points to be more suitable for characterizing the essential features of the internal temperature correlation of the calciner, Thus the temperature characteristic vector is obtained.
  • the product feature vector is cascaded through the first convolutional neural network and the third convolutional neural network and the three-dimensional convolutional neural network and the third convolutional neural network Obtained, which will be larger in terms of feature depth than the temperature feature vector obtained through the timing encoder, therefore, the product feature vector, for example, marked as V 1 and the temperature feature vector, for example, marked as V 2 Before fusion, first perform deep monopair alignment, expressed as:
  • 1 represents the one norm of the vector
  • F represents the Frobenius norm of the matrix
  • the depth homography alignment performs homography alignment between vectors based on the scene depth flow based on the feature depth information characteristics represented by vector differences, and performs depth perception through the full-scene homography correlation matrix between vectors, whereby, dense depth fitting between vectors is performed on the basis of depth differences in feature distributions between vectors to obtain the modified temperature feature vector V 2 that is monotonically aligned in depth with the product feature vector V 1 ', thereby improving the accuracy of subsequent classification.
  • a position-weighted sum of the modified temperature feature vector and the product feature vector is calculated to fuse the two feature information, thereby obtaining the classification feature vector.
  • the classification feature vector is then classified through a classifier to obtain a classification result indicating that the temperature of the calciner at the current time point should be increased or decreased.
  • this application proposes an energy management control system for the preparation of lithium fluoride, which includes: a calcination temperature acquisition module, used to obtain the temperature of the calciner at multiple predetermined time points through a temperature sensor deployed in the calciner. Calcination temperature value, wherein the calciner is used to calcine crystals to form lithium fluoride finished products; a product data acquisition module is used to acquire the multiple The thermal infrared image and X-Ray scan of the calcined product at a predetermined time point; a thermal infrared encoding module for converting the thermal infrared images of the calcined product at multiple predetermined time points through the first convolution using a spatial attention mechanism neural network to obtain thermal infrared feature maps; a perspective encoding module for passing the X-Ray scans of the calcined products at multiple predetermined time points through a second convolutional neural network using a three-dimensional convolution kernel to obtain X-ray Ray feature map; correlation coding module, used to cascade
  • FIG 1 illustrates an application scenario diagram of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • the calciner is obtained through a temperature sensor (for example, T as shown in Figure 1) deployed in the calciner (for example, R as shown in Figure 1) Calcination temperature values at multiple predetermined time points, wherein the calciner is used to calcine crystals to form lithium fluoride finished products, and is measured by a thermal infrared camera deployed within the calciner (for example, as illustrated in Figure 1 C) and an X-Ray scanner (eg, E as illustrated in Figure 1) acquires thermal infrared images and X-rays of the calcined product (eg, as illustrated in Figure 1) at the plurality of predetermined time points Ray scan.
  • a temperature sensor for example, T as shown in Figure 1
  • R as shown in Figure 1
  • a thermal infrared camera deployed within the calciner
  • an X-Ray scanner eg, E as illustrated in Figure 1 acquires thermal infrare
  • the obtained calcining temperature values of the calciner at multiple predetermined time points and the thermal infrared images and X-Ray scans of the calcined product are input to a server deployed with an energy management control algorithm for lithium fluoride preparation.
  • a server deployed with an energy management control algorithm for lithium fluoride preparation for example, the server S as shown in Figure 1
  • the server can control the calcining temperature values of the calciner at multiple predetermined time points and the energy management control algorithm for lithium fluoride preparation.
  • Thermal infrared images and X-Ray scans of the calcined product are processed to generate classification results indicating whether the temperature of the calciner at the current point in time should be increased or decreased. Furthermore, based on the classification result, the temperature of the calciner is adjusted.
  • FIG. 2 illustrates a block diagram of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • the energy management control system 200 for lithium fluoride preparation according to the embodiment of the present application includes: a calcination temperature acquisition module 210, which is used to acquire multiple temperatures of the calciner through temperature sensors deployed in the calciner.
  • the calcining temperature value at a predetermined time point wherein the calciner is used to calcine the crystal to form the finished product of lithium fluoride; the product data acquisition module 220 is used to pass the thermal infrared camera and X-Ray scanner deployed in the calciner Acquire the thermal infrared images and X-Ray scans of the calcined product at the multiple predetermined time points; the thermal infrared encoding module 230 is used to convert the thermal infrared images of the calcined product at the multiple predetermined time points by using spatial attention.
  • the first convolutional neural network of the mechanism is used to obtain the thermal infrared feature map; the perspective image encoding module 240 is used to pass the X-Ray scan images of the calcined product at the plurality of predetermined time points through the second volume using a three-dimensional convolution kernel.
  • the correlation coding module 250 is used to cascade the thermal infrared feature map and the X-Ray feature map and then pass it through the third convolutional neural network as a filter to obtain Product feature vector; energy data encoding module 260, used to pass the calcining temperature values of the calciner at multiple predetermined time points through a time series encoder including a one-dimensional convolution layer and a fully connected layer to obtain a temperature feature vector; feature fusion Module 270 is used to fuse the product feature vector and the temperature feature vector to obtain a classification feature vector; and an energy management result generation module 280 is used to pass the classification feature vector through a classifier to obtain a classification result, the The classification result is used to indicate whether the temperature of the calciner at the current point in time should be increased or decreased.
  • the calcination temperature acquisition module 210 and the product data acquisition module 220 are used to acquire the calcination temperature values of the calciner at multiple predetermined time points through temperature sensors deployed in the calciner.
  • the calciner is used to calcine crystals to form lithium fluoride finished products, and obtain thermal infrared images of the calcined products at multiple predetermined time points through thermal infrared cameras and X-Ray scanners deployed in the calciner. Images and X-Ray scans.
  • a temperature sensor is used to collect the heat information in the calciner, and a thermal infrared camera and an X-Ray scanner are used to collect the internal heat distribution and internal structural information of the calcined product.
  • a convolutional neural network model based on deep learning to deeply mine its implicit correlation feature distribution, thereby ensuring that the temperature adjustment of the calciner at the current time point is more suitable for the preparation of lithium fluoride during classification.
  • the calcining temperature values of the calciner at multiple predetermined time points are obtained through a temperature sensor deployed in the calciner, wherein the calciner is used for calcining to form fluorinated Lithium finished crystal. And obtain thermal infrared images and X-Ray scans of the calcined product at multiple predetermined time points through a thermal infrared camera and an X-Ray scanner deployed in the calciner.
  • the thermal infrared camera can collect to the internal heat distribution of the calcined product to more accurately control the temperature inside the calciner, and the X-Ray scanner can collect the internal structure and shape change information of the calcined product, to better monitor the formation of lithium fluoride product.
  • the thermal infrared encoding module 230 is used to pass the thermal infrared images of the calcined product at the plurality of predetermined time points through the first convolutional neural network using the spatial attention mechanism to obtain Thermal infrared signature map. That is, in the technical solution of the present application, a convolutional neural network model with excellent performance in extracting local implicit features of the image is used to perform feature mining on the thermal infrared image of the calcined product. However, considering that for When performing feature mining on the thermal infrared image of the calcined product, more attention should be paid to the heat change characteristics of the calcined product, and the interfering heat features around it need to be removed.
  • the first convolutional neural network of the spatial attention mechanism is used to perform feature extraction on the thermal infrared images of the calcined products at the multiple predetermined time points to extract the multiple predetermined time points.
  • the local implicit heat characteristic distribution of the thermal infrared image of the calcined product at the time point is obtained, thereby obtaining the thermal infrared characteristic map.
  • the thermal infrared encoding module is further configured to use each layer of the first convolutional neural network to perform on the input data in the forward pass of the layer: Perform convolution processing based on a two-dimensional convolution kernel to generate a convolution feature map; perform pooling processing on the convolution feature map to generate a pooled feature map; perform activation processing on the pooled feature map to generate an activation feature Figure; perform global average pooling along the channel dimension on the activation feature map to obtain a spatial feature matrix; perform convolution processing and activation processing on the spatial feature matrix to generate a weight vector; and, with each of the weight vectors The weight value of the position weights each feature matrix of the activation feature map to obtain a generated feature map; wherein, the generated feature map output by the last layer of the first convolutional neural network model is the thermal infrared Feature map.
  • the perspective image encoding module 240 and the correlation encoding module 250 are used to convert the X-Ray scan images of the calcined product at the multiple predetermined time points by using a three-dimensional convolution kernel.
  • the second convolutional neural network is used to obtain the X-Ray feature map, and the thermal infrared feature map and the X-Ray feature map are cascaded and passed through the third convolutional neural network as a filter to obtain product features. vector.
  • a second convolutional neural network with a three-dimensional convolution kernel is used to process the X-Ray scanning images of the calcined product at the multiple predetermined time points to obtain the calcined product with the X-Ray feature map of dynamically changing features.
  • the thermal infrared characteristic map and the X-Ray characteristic map are further cascaded to fuse the thermal characteristic distribution and the characteristic information of internal structural changes in the calcined product to obtain a cascade characteristic map, and then the obtained cascaded characteristic map is
  • the feature map is extracted through the third convolutional neural network as a filter to obtain the product feature vector.
  • the input data are convolved, pooled along the feature matrix, and activated in the forward pass of the layer using each layer of the third convolutional neural network as a filter.
  • the product feature vector is generated 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 cascade feature map.
  • the perspective encoding module is further used to perform the second convolutional neural network using a three-dimensional convolution kernel on the input data in the forward pass of the layer: based on the The three-dimensional convolution kernel performs three-dimensional convolution processing on the input data to obtain a convolution feature map; performs pooling processing on the convolution feature map to obtain a pooled feature map; and performs non-convolution processing on the pooled feature map.
  • the energy data encoding module 260 is used to pass the calcination temperature values of the calciner at multiple predetermined time points through a temporal encoder including a one-dimensional convolution layer and a fully connected layer. to get the temperature feature vector.
  • the context encoder of the embedding layer encodes the calcination temperature values of the calciner at multiple predetermined time points to extract global-based high-dimensional semantic features between the calcination temperature values of the multiple predetermined time points to be more suitable for Characterize the essential characteristics of the temperature correlation inside the calciner, thereby obtaining a temperature feature vector.
  • the energy data encoding module includes: an input vector construction unit for arranging the calcining temperature values of the calciner at multiple predetermined time points into a one-dimensional structure according to the time dimension.
  • Input vector a fully connected encoding unit, used to use the fully connected layer of the temporal encoder to fully connect the input vector with the following formula to extract the high-dimensional implicit features of the eigenvalues of each position in the input vector, Among them, the formula is: where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector, Represents matrix multiplication; a one-dimensional convolution coding unit, used to use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution coding on the input vector according to the following formula to extract the position of each position in the input vector High-dimensional implicit correlation features between eigenvalues, where the formula is:
  • a is the width of the convolution kernel in the x direction
  • F is the convolution kernel parameter vector
  • G is the local vector matrix that operates with the convolution kernel function
  • w is the size of the convolution kernel.
  • the feature fusion module 270 is used to fuse the product feature vector and the temperature feature vector to obtain a classification feature vector.
  • the product feature vector is cascaded through the first convolutional neural network and the third convolutional neural network and the three-dimensional convolutional neural network and the third convolutional neural network obtained, which in terms of feature depth will be greater than the temperature feature vector obtained through the timing encoder. Therefore, in the technical solution of this application, the product feature vector, for example, is recorded as V 1 and the Before merging the temperature feature vector, for example, denoted as V 2 , the temperature feature vector first needs to be aligned to a deep single pair.
  • the difference eigenvector between the product eigenvector and the temperature eigenvector is calculated.
  • calculate the logarithmic function value of the eigenvalues at each position in the differential feature vector to obtain a logarithmic differential feature vector.
  • a norm of the differential feature vector is calculated.
  • the product between the product feature vector and the transpose vector of the temperature feature vector is calculated to obtain a full scene homography correlation matrix.
  • the logarithmic difference eigenvector is processed using a norm of the difference eigenvector as a weighting coefficient and the Frobenius norm of the full-scene homography correlation matrix as an offset term to obtain the product feature.
  • Corrected temperature feature vectors with vectors aligned in depth is calculated using the following formula: The vector is processed to obtain the corrected temperature feature vector;
  • the depth homography alignment performs homography alignment between vectors based on the scene depth flow based on the feature depth information characteristics represented by vector differences, and performs depth homography through the full-scene homography correlation matrix between vectors. Perception, thereby performing dense depth fitting between vectors on the basis of depth differences in feature distributions between vectors to obtain the modified temperature feature vector that is monopair-aligned in depth with the product feature vector V 1 V 2 ', thus improving the accuracy of subsequent classification.
  • Figure 3 illustrates a block diagram of a feature fusion module in an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • the feature fusion module 270 includes: a vector difference unit 271, used to calculate the difference feature vector between the product feature vector and the temperature feature vector; a logarithmic operation unit 272, used to calculate The logarithmic function value of the feature value at each position in the differential feature vector is used to obtain the logarithmic differential feature vector; the depth information feature value calculation unit 273 is used to calculate a norm of the differential feature vector; full-scene homography correlation
  • the matrix construction unit 274 is used to calculate the product between the product feature vector and the transpose vector of the temperature feature vector to obtain the full-scene homography correlation matrix; the depth sensing unit 275 is used to calculate the full-scene homography
  • the Frobenius norm of the correlation matrix; the alignment unit 276 is used to use a norm of the differential feature vector as a weighting coefficient and the Frobenius norm of the full-scene homography correlation matrix as
  • the energy management result generation module 280 is used to pass the classification feature vector through a classifier to obtain a classification result.
  • the classification result is used to represent the temperature of the calciner at the current point in time. should be increased or should be decreased. That is to say, in the technical solution of the present application, after fusing the modified temperature feature vector and the product feature vector, the classification feature vector is further classified through a classifier to obtain a parameter representing the current time. Classification results of points where the temperature of the calciner should be increased or should be decreased.
  • the energy management result generation module is further configured to: use the classifier to process the classification feature vector with the following formula to obtain the classification result, wherein: The formula is: softmax ⁇ (W n ,B n ):...:(W 1 ,B 1 )
  • the energy management control system 200 for lithium fluoride preparation based on the embodiment of the present application is clarified, which uses a convolutional neural network model based on deep learning to dynamically conduct real-time and dynamic analysis of the temperature-related characteristics inside the calciner. Extract, conduct in-depth exploration of the structural change characteristics and internal heat distribution characteristics of the calcined product, and then combine the time series characteristic information of these three to intelligently adjust the temperature of the calciner to optimize energy while Ensure the quality of the final lithium fluoride product.
  • the energy management control system 200 for lithium fluoride preparation according to the embodiment of the present application can be implemented in various terminal devices, such as a server of the energy management control algorithm for lithium fluoride preparation, etc.
  • the energy management control system 200 for lithium fluoride 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 energy management control system 200 for lithium fluoride production can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the energy management control system 200 for lithium fluoride production
  • the lithium chemical energy management control system 200 can also be one of the many hardware modules of the terminal equipment.
  • the energy management control system 200 for lithium fluoride preparation and the terminal device may also be separate devices, and the energy management control system 200 for lithium fluoride preparation may be connected via a wired And/or a wireless network is connected to the terminal device, and the interactive information is transmitted according to the agreed data format.
  • Figure 4 illustrates a flow chart of a control method of an energy management control system for lithium fluoride preparation.
  • the control method of the energy management control system for lithium fluoride preparation includes the step of: S110, obtaining the temperature of the calciner at multiple predetermined time points through a temperature sensor deployed in the calciner.
  • the calcining temperature value wherein the calciner is used to calcine crystals to form lithium fluoride finished products;
  • S120 obtain the multiple predetermined time points through a thermal infrared camera and an X-Ray scanner deployed in the calciner.
  • the thermal infrared image and X-Ray scan of the calcined product S130, pass the thermal infrared images of the calcined product at multiple predetermined time points through the first convolutional neural network using the spatial attention mechanism to obtain a thermal infrared feature map; S140, pass the X-Ray scans of the calcined product at multiple predetermined time points through a second convolutional neural network using a three-dimensional convolution kernel to obtain an X-Ray feature map; S150, combine the thermal infrared feature map with The X-Ray feature map is cascaded and passed through the third convolutional neural network as a filter to obtain the product feature vector; S160, the calcining temperature values of the calciner at multiple predetermined time points are passed through a one-dimensional volume containing The temporal encoder of the cumulative layer and the fully connected layer is used to obtain the temperature feature vector; S170, fuse the product feature vector and the temperature feature vector to obtain a classification feature vector; and, S180, pass the classification feature vector through
  • FIG. 5 illustrates a schematic architectural diagram of a control method of an energy management control system for lithium fluoride preparation according to an embodiment of the present application.
  • the thermal infrared feature map (for example, as shown in Figure 5) is obtained by using the first convolutional neural network (for example, CNN1 as shown in Figure 5) of the spatial attention mechanism (for example, as shown in Figure 5 F1);
  • the obtained X-Ray scans of the calcined product at the plurality of predetermined time points for example, P2 as shown in Figure 5
  • a second convolutional neural network using a three-dimensional convolution kernel ( For example, CNN2 as shown in Figure 5) to obtain the X-Ray feature map (for example, F2 as shown in Figure 5); then, the thermal in
  • the calcining temperature values of the calciner at multiple predetermined time points are obtained through a temperature sensor deployed in the calciner, wherein the calciner is used to calcine crystals that form lithium fluoride finished products
  • Thermal infrared images and X-Ray scans of the calcined product at multiple predetermined time points are obtained through a thermal infrared camera and an X-Ray scanner deployed in the calciner.
  • a temperature sensor is used to collect the heat information in the calciner
  • a thermal infrared camera and an X-Ray scanner are used to collect the internal heat distribution and internal structural information of the calcined product.
  • the convolutional neural network model based on deep learning to deeply mine its implicit correlation feature distribution, thereby ensuring that the temperature adjustment of the calciner at the current time point is more suitable for the preparation of lithium fluoride during classification.
  • the calcining temperature values of the calciner at multiple predetermined time points are obtained through a temperature sensor deployed in the calciner, wherein the calciner is used for calcining to form fluorinated Lithium finished crystal. And obtain thermal infrared images and X-Ray scans of the calcined product at multiple predetermined time points through a thermal infrared camera and an X-Ray scanner deployed in the calciner.
  • the thermal infrared camera can collect to the internal heat distribution of the calcined product to more accurately control the temperature inside the calciner, and the X-Ray scanner can collect information on the internal structure and shape changes of the calcined product, to better monitor the formation of lithium fluoride product.
  • step S130 the thermal infrared images of the calcined product at the plurality of predetermined time points are passed through a first convolutional neural network using a spatial attention mechanism to obtain a thermal infrared feature map. That is, in the technical solution of the present application, a convolutional neural network model with excellent performance in extracting local implicit features of the image is used to perform feature mining on the thermal infrared image of the calcined product. However, considering that for When performing feature mining on the thermal infrared image of the calcined product, more attention should be paid to the heat change characteristics of the calcined product, and the interfering heat features around it need to be removed.
  • the first convolutional neural network of the spatial attention mechanism is used to perform feature extraction on the thermal infrared images of the calcined products at the multiple predetermined time points to extract the multiple predetermined time points.
  • the local implicit heat characteristic distribution of the thermal infrared image of the calcined product at the time point is obtained, thereby obtaining the thermal infrared characteristic map.
  • step S140 and step S150 the X-Ray scan images of the calcined product at the plurality of predetermined time points are passed through a second convolutional neural network using a three-dimensional convolution kernel to obtain the X-Ray feature map,
  • the thermal infrared feature map and the X-Ray feature map are cascaded and passed through a third convolutional neural network as a filter to obtain a product feature vector.
  • a second convolutional neural network with a three-dimensional convolution kernel is used to process the X-Ray scanning images of the calcined product at the multiple predetermined time points to obtain the calcined product with the X-Ray feature map of dynamically changing features.
  • the thermal infrared characteristic map and the X-Ray characteristic map are further cascaded to fuse the thermal characteristic distribution and the characteristic information of internal structural changes in the calcined product to obtain a cascade characteristic map, and then the obtained cascaded characteristic map is
  • the feature map is extracted through the third convolutional neural network as a filter to obtain the product feature vector.
  • the input data are convolved, pooled along the feature matrix, and activated in the forward pass of the layer using each layer of the third convolutional neural network as a filter.
  • the product feature vector is generated 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 cascade feature map.
  • step S160 the calcining temperature values of the calciner at multiple predetermined time points are passed through a temporal encoder including a one-dimensional convolution layer and a fully connected layer to obtain a temperature feature vector.
  • a temporal encoder including a one-dimensional convolution layer and a fully connected layer to obtain a temperature feature vector.
  • step S170 the product feature vector and the temperature feature vector are fused to obtain a classification feature vector.
  • the product feature vector is cascaded through the first convolutional neural network and the third convolutional neural network and the three-dimensional convolutional neural network and the third convolutional neural network obtained, which in terms of feature depth will be greater than the temperature feature vector obtained through the timing encoder. Therefore, in the technical solution of this application, the product feature vector, for example, is recorded as V 1 and the Before merging the temperature feature vector, for example, denoted as V 2 , the temperature feature vector first needs to be aligned to a deep single pair.
  • step S180 the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate that the temperature of the calciner at the current time point should be increased or decreased. That is to say, in the technical solution of the present application, after fusing the modified temperature feature vector and the product feature vector, the classification feature vector is further classified through a classifier to obtain a parameter representing the current time. Classification results of points where the temperature of the calciner should be increased or should be decreased.
  • control method of the energy management control system for lithium fluoride preparation based on the embodiment of the present application is clarified, which uses a convolutional neural network model based on deep learning to conduct real-time analysis of the temperature-related characteristics inside the calciner. Dynamically extract and deeply explore the structural change characteristics and internal heat distribution characteristics of the calcined product, and then combine the time series characteristic information of these three to intelligently adjust the temperature of the calciner to optimize energy while ensuring the quality of the final lithium fluoride product.
  • 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

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)
  • Battery Electrode And Active Subsutance (AREA)

Abstract

本申请涉及制备智能控制的领域，其具体地公开了一种用于氟化锂制备的能源管理控制系统及其控制方法，其通过基于深度学习的卷积神经网络模型来对于煅烧器内部的温度关联特征进行实时动态地提取，以及对于煅烧产物的结构变化特征以及内部的热量分布特征进行深层次地挖掘，进而结合这三者在时序上的特征信息来智能地调整所述煅烧器的温度，以在优化能源的同时保证最终氟化锂的成品质量。

Description

用于氟化锂制备的能源管理控制系统及其控制方法 技术领域
本发明涉及制备智能控制的领域，且更为具体地，涉及一种用于氟化锂制备的能源管理控制系统及其控制方法。
背景技术
氟化锂是一种重要的锂基材料，在常温下为白色非吸潮性立方晶体。作为一种重要的无机氟化物，高纯氟化锂大量用于氟化玻璃和光学纤维的制备。同时，高纯氟化锂还是锂离子电池用电解质材料的重要原料。
常用的氟化锂的制备方法主要包括直接制备法、离子交换制备法和萃取制备方法。不管采用哪种制备方案，制备过程中能源管理都是非常重要的关注点，尤其是当下在节能环保、碳中和的大背景下。
具体地，在氟化锂的制备过程中，大都涉及到煅烧、冷却等温度控制，应可以理解，温度控制不仅仅涉及到能源管理，还涉及到最终氟化锂的成品质量。在现有的氟化锂的制备方案中，大多采用预设温度的技术方案，这不仅不利于能源优化，且不利于把控最终氟化锂的成品质量。
因此，期望一种用于氟化锂制备的能源管理控制系统来对于煅烧器的温度进行智能控制，以在优化能源的同时保证最终氟化锂的成品质量。
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氟化锂制备的能源管理控制系统及其控制方法，其通过基于深度学习的卷积神经网络模型来对于煅烧器内部的温度关联特征进行实时动态地提取，以及对于煅烧产物的结构变化特征以及内部的热量分布特征进行深层次地挖掘，进而结合这三者在时序上的特征信息来智能地调整所述煅烧器的温度，以在优化能源的同时保证最终氟化锂的成品质量。
根据本申请的一个方面，提供了一种用于氟化锂制备的能源管理控制系统，其包括：
煅烧温度采集模块，用于通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成 品的晶体；
产物数据采集模块，用于通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；
热红外编码模块，用于将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；
透视图编码模块，用于将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；
关联编码模块，用于将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；
能源数据编码模块，用于将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；
特征融合模块，用于融合所述产物特征向量和所述温度特征向量以得到分类特征向量；以及
能源管理结果生成模块，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当前时间点的煅烧器的温度应增大或应减小。
在上述用于氟化锂制备的能源管理控制系统中，所述热红外编码模块，进一步用于使用所述第一卷积神经网络的各层在层的正向传递中对输入数据进行：对所述输入数据进行基于二维卷积核的卷积处理以生成卷积特征图；对所述卷积特征图进行池化处理以生成池化特征图；对所述池化特征图进行激活处理以生成激活特征图；对所述激活特征图进行沿通道维度的全局平均池化以获得空间特征矩阵；对所述空间特征矩阵进行卷积处理和激活处理以生成权重向量；以及，以所述权重向量中各个位置的权重值分别对所述激活特征图的各个特征矩阵进行加权以获得生成特征图；其中，所述第一卷积神经网络模型的最后一层输出的所述生成特征图为所述热红外特征图。
在上述用于氟化锂制备的能源管理控制系统中，所述透视图编码模块，进一步用于所述使用三维卷积核的第二卷积神经网络在层的正向传递中对输入数据分别进行：基于所述三维卷积核对所述输入数据进行三维卷积处理以得到卷积特征图；对所述卷积特征图进行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以得到激活特征图；其中，所述第二卷积神经网络的最后一层的输出为所述X-Ray特征图，所述第二卷积神经网 络的第一层的输入为所述多个预定时间点的煅烧产物的X-Ray扫描图。
在上述用于氟化锂制备的能源管理控制系统中，所述关联编码模块，级联单元，用于将所述热红外特征图和所述X-Ray特征图进行级联以获得级联特征图；以及。特征提取单元，用于使用作为过滤器的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产物特征向量，其中，所述第三卷积神经网络的第一层的输入为所述级联特征图。
在上述用于氟化锂制备的能源管理控制系统中，所述能源数据编码模块，包括：输入向量构造单元，用于将所述煅烧器在多个预定时间点的煅烧温度值按照时间维度排列为一维的输入向量；全连接编码单元，用于使用所述时序编码器的全连接层以如下公式对所述输入向量进行全连接编码以提取出所述输入向量中各个位置的特征值的高维隐含特征，其中，所述公式为：
Figure PCTCN2022117768-appb-000001
Figure PCTCN2022117768-appb-000002
其中X是所述输入向量，Y是输出向量，W是权重矩阵，B是偏置向量，
Figure PCTCN2022117768-appb-000003
表示矩阵乘；一维卷积编码单元，用于使用所述时序编码器的一维卷积层以如下公式对所述输入向量进行一维卷积编码以提取出所述输入向量中各个位置的特征值间的高维隐含关联特征，其中，所述公式为：
Figure PCTCN2022117768-appb-000004
其中，a为卷积核在x方向上的宽度、F为卷积核参数向量、G为与卷积核函数运算的局部向量矩阵，w为卷积核的尺寸。
在上述用于氟化锂制备的能源管理控制系统中，所述特征融合模块，包括：向量差分单元，用于计算所述产物特征向量和所述温度特征向量之间的差分特征向量；对数运算单元，用于计算所述差分特征向量中各个位置的特征值的对数函数值以得到对数差分特征向量；深度信息特性值计算单元，用于计算所述差分特征向量的一范数；全场景单应关联矩阵构造单元，用于计算所述产物特征向量与所述温度特征向量的转置向量之间的乘积以得到全场景单应关联矩阵；深度感知单元，用于计算所述全场景单应关联矩阵的Frobenius范数；对齐单元，用于以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项对所述对数差分特征向量进行处理以得到与所述产物特征向量在深度上单应对齐的修正后温度特征向量；融合单元，用于计算所述修正后温度特征向量和所述产物 特征向量的按位置加权和以得到所述分类特征向量。
在上述用于氟化锂制备的能源管理控制系统中，所述对齐单元，用于以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项以如下公式对所述对数差分特征向量进行处理以得到所述修正后温度特征向量；
其中，所述公式为：
Figure PCTCN2022117768-appb-000005
其中，
Figure PCTCN2022117768-appb-000006
表示所述差分特征向量的一范数，且||V 1 TV 2|| F表示所述全场景单应关联矩阵的Frobenius范数，⊙表示按位置点乘、
Figure PCTCN2022117768-appb-000007
表示按位置相加，
Figure PCTCN2022117768-appb-000008
表示按位置相减。
在上述用于氟化锂制备的能源管理控制系统中，所述能源管理结果生成模块，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。
在上述用于氟化锂制备的能源管理控制系统中，还包括能源控制模块，用于基于所述分类结果，调整所述煅烧器的温度。
根据本申请的另一方面，一种用于氟化锂制备的能源管理控制系统的控制方法，其包括：
通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体；
通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；
将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；
将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；
将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；
将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；
融合所述产物特征向量和所述温度特征向量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当前时间点的煅烧器的温度应增大或应减小。
在上述用于氟化锂制备的能源管理控制系统的控制方法中，将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图，包括：使用所述第一卷积神经网络的各层在层的正向传递中对输入数据进行：对所述输入数据进行基于二维卷积核的卷积处理以生成卷积特征图；对所述卷积特征图进行池化处理以生成池化特征图；对所述池化特征图进行激活处理以生成激活特征图；对所述激活特征图进行沿通道维度的全局平均池化以获得空间特征矩阵；对所述空间特征矩阵进行卷积处理和激活处理以生成权重向量；以及，以所述权重向量中各个位置的权重值分别对所述激活特征图的各个特征矩阵进行加权以获得生成特征图；其中，所述第一卷积神经网络模型的最后一层输出的所述生成特征图为所述热红外特征图。
在上述用于氟化锂制备的能源管理控制系统的控制方法中，将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图，包括：所述使用三维卷积核的第二卷积神经网络在层的正向传递中对输入数据分别进行：基于所述三维卷积核对所述输入数据进行三维卷积处理以得到卷积特征图；对所述卷积特征图进行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以得到激活特征图；其中，所述第二卷积神经网络的最后一层的输出为所述X-Ray特征图，所述第二卷积神经网络的第一层的输入为所述多个预定时间点的煅烧产物的X-Ray扫描图。
在上述用于氟化锂制备的能源管理控制系统的控制方法中，将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量，包括：将所述热红外特征图和所述X-Ray特征图进行级联以获得级联特征图；以及，使用作为过滤器的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产物特征向量，其中，所述第三卷积神经网络的第一层的输入为所述级联特征图。
在上述用于氟化锂制备的能源管理控制系统的控制方法中，将所述煅烧 器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量，包括：将所述煅烧器在多个预定时间点的煅烧温度值按照时间维度排列为一维的输入向量；使用所述时序编码器的全连接层以如下公式对所述输入向量进行全连接编码以提取出所述输入向量中各个位置的特征值的高维隐含特征，其中，所述公式为：
Figure PCTCN2022117768-appb-000009
其中X是所述输入向量，Y是输出向量，W是权重矩阵，B是偏置向量，
Figure PCTCN2022117768-appb-000010
表示矩阵乘；使用所述时序编码器的一维卷积层以如下公式对所述输入向量进行一维卷积编码以提取出所述输入向量中各个位置的特征值间的高维隐含关联特征，其中，所述公式为：
Figure PCTCN2022117768-appb-000011
其中，a为卷积核在x方向上的宽度、F为卷积核参数向量、G为与卷积核函数运算的局部向量矩阵，w为卷积核的尺寸。
在上述用于氟化锂制备的能源管理控制系统的控制方法中，融合所述产物特征向量和所述温度特征向量以得到分类特征向量，包括：计算所述产物特征向量和所述温度特征向量之间的差分特征向量；计算所述差分特征向量中各个位置的特征值的对数函数值以得到对数差分特征向量；计算所述差分特征向量的一范数；计算所述产物特征向量与所述温度特征向量的转置向量之间的乘积以得到全场景单应关联矩阵；计算所述全场景单应关联矩阵的Frobenius范数；以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项对所述对数差分特征向量进行处理以得到与所述产物特征向量在深度上单应对齐的修正后温度特征向量；计算所述修正后温度特征向量和所述产物特征向量的按位置加权和以得到所述分类特征向量。
在上述用于氟化锂制备的能源管理控制系统的控制方法中，融合所述产物特征向量和所述温度特征向量以得到分类特征向量，包括：以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项以如下公式对所述对数差分特征向量进行处理以得到所述修正后温度特征向量；
其中，所述公式为：
Figure PCTCN2022117768-appb-000012
其中，
Figure PCTCN2022117768-appb-000013
表示所述差分特征向量的一范数，且||V 1 TV 2|| F表示所述全场景单应关联矩阵的Frobenius范数，⊙表示按位置点乘、
Figure PCTCN2022117768-appb-000014
表示按位置相加，
Figure PCTCN2022117768-appb-000015
表示按位置相减。
在上述用于氟化锂制备的能源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。
在上述用于氟化锂制备的能源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，还包括：基于所述分类结果，调整所述煅烧器的温度。
与现有技术相比，本申请提供的用于氟化锂制备的能源管理控制系统及其控制方法，其通过基于深度学习的卷积神经网络模型来对于煅烧器内部的温度关联特征进行实时动态地提取，以及对于煅烧产物的结构变化特征以及内部的热量分布特征进行深层次地挖掘，进而结合这三者在时序上的特征信息来智能地调整所述煅烧器的温度，以在优化能源的同时保证最终氟化锂的成品质量。
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氟化锂制备的能源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氟化锂制备的能源管理控制系统的框图。
图3为根据本申请实施例的用于氟化锂制备的能源管理控制系统中特征融合模块的框图。
图4为根据本申请实施例的用于氟化锂制备的能源管理控制系统的控制方法的流程图。
图5为根据本申请实施例的用于氟化锂制备的能源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，氟化锂是一种重要的锂基材料，在常温下为白色非吸潮性立方晶体。作为一种重要的无机氟化物，高纯氟化锂大量用于氟化玻璃和光学纤维的制备。同时，高纯氟化锂还是锂离子电池用电解质材料的重要原料。
常用的氟化锂的制备方法主要包括直接制备法、离子交换制备法和萃取制备方法。不管采用哪种制备方案，制备过程中能源管理都是非常重要的关注点，尤其是当下在节能环保、碳中和的大背景下。
具体地，在氟化锂的制备过程中，大都涉及到煅烧、冷却等温度控制，应可以理解，温度控制不仅仅涉及到能源管理，还涉及到最终氟化锂的成品质量。在现有的氟化锂的制备方案中，大多采用预设温度的技术方案，这不仅不利于能源优化，且不利于把控最终氟化锂的成品质量。因此，期望一种用于氟化锂制备的能源管理控制系统来对于煅烧器的温度进行智能控制，以在优化能源的同时保证最终氟化锂的成品质量。
目前，深度学习以及神经网络已经广泛应用于计算机视觉、自然语言处理、文本信号处理等领域。此外，深度学习以及神经网络在图像分类、物体检测、语义分割、文本翻译等领域，也展现出了接近甚至超越人类的水平。
近年来，深度学习以及神经网络的发展，为煅烧器的温度控制提供了解决思路和方案。
基于此，本申请发明人考虑到若想对于煅烧器的温度进行智能准确地控制，以提高氟化锂成品的质量，就需要对于煅烧器内部的温度进行实时动态地测量，并且还需要对于煅烧产物的结构特征以及热量分布进行分析。因此，在本申请中，采用温度传感器来采集所述煅烧器中的热量信息，以及通过热红外相机和X-Ray扫描仪来采集所述煅烧产物的内部热量分布以及内部的结 构信息，进一步通过基于深度学习的卷积神经网络模型来对其隐含的关联特征分布进行深层挖掘，进而在分类时能够保证对于当前时间点的煅烧器的温度调节更适于氟化锂的制备。
具体地，在本申请的技术方案中，首先，通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体。并且通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图，这里，所述热红外相机能够采集到所述煅烧产物的内部热量分布情况，以更准确地对于所述煅烧器内部的温度进行控制，且所述X-Ray扫描仪能够采集到所述煅烧产物的内部结构以及形状的变化信息，以对于氟化锂产物的形成制备进行更好地监控。
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煅烧产物的热红外图像进行特征挖掘，但是，考虑到在对于所述煅烧产物的热红外图像进行特征挖掘时，应该更加关注于所述煅烧产物的热量变化特征，而需要去除其周围的干扰热量特征，因此，在本申请的技术方案中，使用空间注意力机制的第一卷积神经网络来对所述多个预定时间点的煅烧产物的热红外图像进行特征提取，以提取出所述多个预定时间点的煅烧产物的热红外图像的局部隐含热量特征分布，从而得到热红外特征图。
同时，考虑到在利用X-Ray扫描仪来挖掘所述煅烧产物的内部空间结构特征时，需要关注到所述煅烧产物的动态变化特征，以防止反应过快而导致氟化锂成品的质量达不到应有的要求，因此，在本申请的技术方案中，使用三维卷积核的第二卷积神经网络来对所述多个预定时间点的煅烧产物的X-Ray扫描图进行处理，以得到具有所述煅烧产物动态变化特征的X-Ray特征图。这样，进一步将所述热红外特征图和所述X-Ray特征图进行级联以融合所述煅烧产物中热量特征分布以及内部结构变化的特征信息，再将得到的级联特征图通过作为过滤器的第三卷积神经网络中进行特征提取，以得到产物特征向量。
对于所述煅烧器在多个预定时间点的煅烧温度值，考虑到所述各个时间点的煅烧温度值之间存在着关联性，因此，使用包含嵌入层的上下文编码器对所述煅烧器在多个预定时间点的煅烧温度值进行编码以提取所述多个预定时间点的煅烧温度值之间的基于全局的高维语义特征以更适于表征所述 煅烧器内部温度关联的本质特征，从而得到温度特征向量。
应可以理解，考虑到所述产物特征向量是通过级联的所述第一卷积神经网络和所述第三卷积神经网络以及所述三维卷积神经网络和所述第三卷积神经网络获得的，其在特征深度方面会大于通过所述时序编码器获得的所述温度特征向量，因此，在将所述产物特征向量，例如记为V 1与所述温度特征向量，例如记为V 2融合之前，首先将其进行深度单应对齐，表示为：
Figure PCTCN2022117768-appb-000016
其中，||·|| 1表示向量的一范数，且||·|| F表示矩阵的Frobenius范数。
这里，该所述深度单应对齐通过根据向量差分表征的特征深度信息特性，来进行向量间的基于场景深度流的单应性对齐，并通过向量间的全场景单应关联矩阵进行深度感知，从而在向量之间存在特征分布的深度差异的基础上进行向量之间的稠密深度拟合，以获得与所述产物特征向量V 1在深度上单应对齐的所述修正的温度特征向量V 2'，进而提高后续分类的准确性。
进一步地，计算所述修正的温度特征向量和所述产物特征向量的按位置加权和以融合这两者的特征信息，从而得到所述分类特征向量。再将所述分类特征向量通过分类器中进行分类处理，以获得用于表示当前时间点的煅烧器的温度应增大或应减小的分类结果。
基于此，本申请提出了一种用于氟化锂制备的能源管理控制系统，其包括：煅烧温度采集模块，用于通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体；产物数据采集模块，用于通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；热红外编码模块，用于将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；透视图编码模块，用于将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；关联编码模块，用于将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；能源数据编码模块，用于将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；特征融合模块，用于融合所述产物特征向量和所述温度特征向量以得到分类特征向量；以及，能源管理结果生成模块， 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当前时间点的煅烧器的温度应增大或应减小。
图1图示了根据本申请实施例的用于氟化锂制备的能源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，通过部署于煅烧器(例如，如图1中所示意的R)内的温度传感器(例如，如图1中所示意的T)获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体，并且通过部署于所述煅烧器内的热红外相机(例如，如图1中所示意的C)和X-Ray扫描仪(例如，如图1中所示意的E)获取所述多个预定时间点的煅烧产物(例如，如图1中所示意的P)的热红外图像和X-Ray扫描图。然后，将获得的所述煅烧器在多个预定时间点的煅烧温度值以及所述煅烧产物的热红外图像和X-Ray扫描图输入至部署有用于氟化锂制备的能源管理控制算法的服务器中(例如，如图1中所示意的服务器S)，其中，所述服务器能够以用于氟化锂制备的能源管理控制算法对所述煅烧器在多个预定时间点的煅烧温度值以及所述煅烧产物的热红外图像和X-Ray扫描图进行处理，以生成用于表示当前时间点的煅烧器的温度应增大或应减小的分类结果。进而，基于所述分类结果，调整所述煅烧器的温度。
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氟化锂制备的能源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氟化锂制备的能源管理控制系统200，包括：煅烧温度采集模块210，用于通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体；产物数据采集模块220，用于通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；热红外编码模块230，用于将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；透视图编码模块240，用于将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；关联编码模块250，用于将所述热红外特征图和所述 X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；能源数据编码模块260，用于将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；特征融合模块270，用于融合所述产物特征向量和所述温度特征向量以得到分类特征向量；以及，能源管理结果生成模块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当前时间点的煅烧器的温度应增大或应减小。
具体地，在本申请实施例中，所述煅烧温度采集模块210和所述产物数据采集模块220，用于通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体，并通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图。如前所述，应可以理解，考虑到若想对于煅烧器的温度进行智能准确地控制，以提高氟化锂成品的质量，就需要对于所述煅烧器内部的温度进行实时动态地测量，并且还需要对于煅烧产物的结构特征以及热量分布进行分析。因此，在本申请的技术方案中，采用温度传感器来采集所述煅烧器中的热量信息，以及通过热红外相机和X-Ray扫描仪来采集所述煅烧产物的内部热量分布以及内部的结构信息，进一步再通过基于深度学习的卷积神经网络模型来对其隐含的关联特征分布进行深层挖掘，进而在分类时能够保证对于当前时间点的煅烧器的温度调节更适于氟化锂的制备。
也就是，具体地，在本申请的技术方案中，首先，通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体。并且通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图，这里，所述热红外相机能够采集到所述煅烧产物的内部热量分布情况，以更准确地对于所述煅烧器内部的温度进行控制，且所述X-Ray扫描仪能够采集到所述煅烧产物的内部结构以及形状的变化信息，以对于氟化锂产物的形成制备进行更好地监控。
具体地，在本申请实施例中，所述热红外编码模块230，用于将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图。也就是，在本申请的技术方案中，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煅烧产物的热红外图像进行特征挖掘，但是，考虑到在对于所述煅烧产物的热红外图像进行特征挖掘时，应该更加关注于所述煅烧产物的热量变化特征，而需要去除其周围的干扰热量特征。因此，在本申请的技术方案中，使用空间注意力机制的第一卷积神经网络来对所述多个预定时间点的煅烧产物的热红外图像进行特征提取，以提取出所述多个预定时间点的煅烧产物的热红外图像的局部隐含热量特征分布，从而得到热红外特征图。
更具体地，在本申请实施例中，所述热红外编码模块，进一步用于使用所述第一卷积神经网络的各层在层的正向传递中对输入数据进行：对所述输入数据进行基于二维卷积核的卷积处理以生成卷积特征图；对所述卷积特征图进行池化处理以生成池化特征图；对所述池化特征图进行激活处理以生成激活特征图；对所述激活特征图进行沿通道维度的全局平均池化以获得空间特征矩阵；对所述空间特征矩阵进行卷积处理和激活处理以生成权重向量；以及，以所述权重向量中各个位置的权重值分别对所述激活特征图的各个特征矩阵进行加权以获得生成特征图；其中，所述第一卷积神经网络模型的最后一层输出的所述生成特征图为所述热红外特征图。
具体地，在本申请实施例中，所述透视图编码模块240和所述关联编码模块250，用于将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图，并将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量。应可以理解，考虑到在利用所述X-Ray扫描仪来挖掘所述煅烧产物的内部空间结构特征时，需要关注到所述煅烧产物的动态变化特征，以防止反应过快而导致氟化锂成品的质量达不到应有的要求。因此，在本申请的技术方案中，使用三维卷积核的第二卷积神经网络来对所述多个预定时间点的煅烧产物的X-Ray扫描图进行处理，以得到具有所述煅烧产物动态变化特征的X-Ray特征图。这样，进一步将所述热红外特征图和所述X-Ray特征图进行级联以融合所述煅烧产物中热量特征分布以及内部结构变化的特征信息以获得级联特征图，再将得到的级联特征图通过作为过滤器的第三卷积神经网络中进行特征提取，以得到产物特征向量。相应地，在一个具体示例中，使用作为过滤器的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产物特征向量，其中，所述第三卷积神经网络的第一层的输入为所述级联特征图。
更具体地，在本申请实施例中，所述透视图编码模块，进一步用于所述使用三维卷积核的第二卷积神经网络在层的正向传递中对输入数据分别进行：基于所述三维卷积核对所述输入数据进行三维卷积处理以得到卷积特征图；对所述卷积特征图进行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以得到激活特征图；其中，所述第二卷积神经网络的最后一层的输出为所述X-Ray特征图，所述第二卷积神经网络的第一层的输入为所述多个预定时间点的煅烧产物的X-Ray扫描图。
具体地，在本申请实施例中，所述能源数据编码模块260，用于将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量。应可以理解，对于所述煅烧器在多个预定时间点的煅烧温度值，考虑到所述各个时间点的煅烧温度值之间存在着关联性，因此，在本申请的技术方案中，使用包含嵌入层的上下文编码器对所述煅烧器在多个预定时间点的煅烧温度值进行编码以提取所述多个预定时间点的煅烧温度值之间的基于全局的高维语义特征以更适于表征所述煅烧器内部温度关联的本质特征，从而得到温度特征向量。
更具体地，在本申请实施例中，所述能源数据编码模块，包括：输入向量构造单元，用于将所述煅烧器在多个预定时间点的煅烧温度值按照时间维度排列为一维的输入向量；全连接编码单元，用于使用所述时序编码器的全连接层以如下公式对所述输入向量进行全连接编码以提取出所述输入向量中各个位置的特征值的高维隐含特征，其中，所述公式为：
Figure PCTCN2022117768-appb-000017
其中X是所述输入向量，Y是输出向量，W是权重矩阵，B是偏置向量，
Figure PCTCN2022117768-appb-000018
表示矩阵乘；一维卷积编码单元，用于使用所述时序编码器的一维卷积层以如下公式对所述输入向量进行一维卷积编码以提取出所述输入向量中各个位置的特征值间的高维隐含关联特征，其中，所述公式为：
Figure PCTCN2022117768-appb-000019
其中，a为卷积核在x方向上的宽度、F为卷积核参数向量、G为与卷积核函数运算的局部向量矩阵，w为卷积核的尺寸。
具体地，在本申请实施例中，所述特征融合模块270，用于融合所述产 物特征向量和所述温度特征向量以得到分类特征向量。应可以理解，考虑到所述产物特征向量是通过级联的所述第一卷积神经网络和所述第三卷积神经网络以及所述三维卷积神经网络和所述第三卷积神经网络获得的，其在特征深度方面会大于通过所述时序编码器获得的所述温度特征向量，因此，在本申请的技术方案中，在将所述产物特征向量，例如记为V 1与所述温度特征向量，例如记为V 2融合之前，首先需要将所述温度特征向量进行深度单应对齐。
也就是，具体地，在本申请实施例中，首先，计算所述产物特征向量和所述温度特征向量之间的差分特征向量。接着，计算所述差分特征向量中各个位置的特征值的对数函数值以得到对数差分特征向量。然后，计算所述差分特征向量的一范数。接着，计算所述产物特征向量与所述温度特征向量的转置向量之间的乘积以得到全场景单应关联矩阵。然后，计算所述全场景单应关联矩阵的Frobenius范数。接着，以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项对所述对数差分特征向量进行处理以得到与所述产物特征向量在深度上单应对齐的修正后温度特征向量。相应地，在一个具体示例中，以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项以如下公式对所述对数差分特征向量进行处理以得到所述修正后温度特征向量；
其中，所述公式为：
Figure PCTCN2022117768-appb-000020
其中，
Figure PCTCN2022117768-appb-000021
表示所述差分特征向量的一范数，且||V 1 TV 2|| F表示所述全场景单应关联矩阵的Frobenius范数，⊙表示按位置点乘、
Figure PCTCN2022117768-appb-000022
表示按位置相加，
Figure PCTCN2022117768-appb-000023
表示按位置相减。最后，计算所述修正后温度特征向量和所述产物特征向量的按位置加权和以得到所述分类特征向量。应可以理解，该所述深度单应对齐通过根据向量差分表征的特征深度信息特性，来进行向量间的基于场景深度流的单应性对齐，并通过向量间的全场景单应关联矩阵进行深度感知，从而在向量之间存在特征分布的深度差异的基础上进行向量之间的稠密深度拟合，以获得与所述产物特征向量V 1在深度上单应对齐的所述修正的温度特征向量V 2'，进而提高后续分类的准确性。
图3图示了根据本申请实施例的用于氟化锂制备的能源管理控制系统中 特征融合模块的框图。如图3所示，所述特征融合模块270，包括：向量差分单元271，用于计算所述产物特征向量和所述温度特征向量之间的差分特征向量；对数运算单元272，用于计算所述差分特征向量中各个位置的特征值的对数函数值以得到对数差分特征向量；深度信息特性值计算单元273，用于计算所述差分特征向量的一范数；全场景单应关联矩阵构造单元274，用于计算所述产物特征向量与所述温度特征向量的转置向量之间的乘积以得到全场景单应关联矩阵；深度感知单元275，用于计算所述全场景单应关联矩阵的Frobenius范数；对齐单元276，用于以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项对所述对数差分特征向量进行处理以得到与所述产物特征向量在深度上单应对齐的修正后温度特征向量；融合单元277，用于计算所述修正后温度特征向量和所述产物特征向量的按位置加权和以得到所述分类特征向量。
具体地，在本申请实施例中，所述能源管理结果生成模块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当前时间点的煅烧器的温度应增大或应减小。也就是，在本申请的技术方案中，在融合所述修正的温度特征向量和所述产物特征向量后，进一步将所述分类特征向量通过分类器中进行分类处理，以获得用于表示当前时间点的煅烧器的温度应增大或应减小的分类结果。
更具体地，在本申请实施例中，所述能源管理结果生成模块，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。
综上，基于本申请实施例的所述用于氟化锂制备的能源管理控制系统200被阐明，其通过基于深度学习的卷积神经网络模型来对于煅烧器内部的温度关联特征进行实时动态地提取，以及对于煅烧产物的结构变化特征以及内部的热量分布特征进行深层次地挖掘，进而结合这三者在时序上的特征信息来智能地调整所述煅烧器的温度，以在优化能源的同时保证最终氟化锂的成品质量。
如上所述，根据本申请实施例的用于氟化锂制备的能源管理控制系统200可以实现在各种终端设备中，例如用于氟化锂制备的能源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氟化锂制备的能源管 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，该用于氟化锂制备的能源管理控制系统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，该用于氟化锂制备的能源管理控制系统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，该用于氟化锂制备的能源管理控制系统200与该终端设备也可以是分立的设备，并且该用于氟化锂制备的能源管理控制系统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4图示了用于氟化锂制备的能源管理控制系统的控制方法的流程图。如图4所示，根据本申请实施例的用于氟化锂制备的能源管理控制系统的控制方法，包括步骤：S110，通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体；S120，通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；S130，将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；S140，将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；S150，将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；S160，将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；S170，融合所述产物特征向量和所述温度特征向量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当前时间点的煅烧器的温度应增大或应减小。
图5图示了根据本申请实施例的用于氟化锂制备的能源管理控制系统的控制方法的架构示意图。如图5所示，在所述用于氟化锂制备的能源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煅烧产物的热红外图像(例如，如图5中所示意的P1)通过使用空间注意力 机制的第一卷积神经网络(例如，如图5中所示意的CNN1)以得到热红外特征图(例如，如图5中所示意的F1)；接着，将获得的所述多个预定时间点的煅烧产物的X-Ray扫描图(例如，如图5中所示意的P2)通过使用三维卷积核的第二卷积神经网络(例如，如图5中所示意的CNN2)以得到X-Ray特征图(例如，如图5中所示意的F2)；然后，将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络(例如，如图5中所示意的CNN3)以得到产物特征向量(例如，如图5中所示意的VF1)；接着，将获得的所述煅烧器在多个预定时间点的煅烧温度值(例如，如图5中所示意的Q)通过包含一维卷积层和全连接层的时序编码器(例如，如图5中所示意的E)以得到温度特征向量(例如，如图5中所示意的VF2)；然后，融合所述产物特征向量和所述温度特征向量以得到分类特征向量(例如，如图5中所示意的VF)；以及，最后，将所述分类特征向量通过分类器(例如，如图5中所示意的圈S)以得到分类结果，所述分类结果用于表示当前时间点的煅烧器的温度应增大或应减小。
更具体地，在步骤S110和S120中，通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体，并通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图。应可以理解，考虑到若想对于煅烧器的温度进行智能准确地控制，以提高氟化锂成品的质量，就需要对于所述煅烧器内部的温度进行实时动态地测量，并且还需要对于煅烧产物的结构特征以及热量分布进行分析。因此，在本申请的技术方案中，采用温度传感器来采集所述煅烧器中的热量信息，以及通过热红外相机和X-Ray扫描仪来采集所述煅烧产物的内部热量分布以及内部的结构信息，进一步再通过基于深度学习的卷积神经网络模型来对其隐含的关联特征分布进行深层挖掘，进而在分类时能够保证对于当前时间点的煅烧器的温度调节更适于氟化锂的制备。
也就是，具体地，在本申请的技术方案中，首先，通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体。并且通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图，这里，所述热红外相机能够采集到所述煅烧产物的内部热量 分布情况，以更准确地对于所述煅烧器内部的温度进行控制，且所述X-Ray扫描仪能够采集到所述煅烧产物的内部结构以及形状的变化信息，以对于氟化锂产物的形成制备进行更好地监控。
更具体地，在步骤S130中，将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图。也就是，在本申请的技术方案中，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煅烧产物的热红外图像进行特征挖掘，但是，考虑到在对于所述煅烧产物的热红外图像进行特征挖掘时，应该更加关注于所述煅烧产物的热量变化特征，而需要去除其周围的干扰热量特征。因此，在本申请的技术方案中，使用空间注意力机制的第一卷积神经网络来对所述多个预定时间点的煅烧产物的热红外图像进行特征提取，以提取出所述多个预定时间点的煅烧产物的热红外图像的局部隐含热量特征分布，从而得到热红外特征图。
更具体地，在步骤S140和步骤S150中，将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图，并将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量。应可以理解，考虑到在利用所述X-Ray扫描仪来挖掘所述煅烧产物的内部空间结构特征时，需要关注到所述煅烧产物的动态变化特征，以防止反应过快而导致氟化锂成品的质量达不到应有的要求。因此，在本申请的技术方案中，使用三维卷积核的第二卷积神经网络来对所述多个预定时间点的煅烧产物的X-Ray扫描图进行处理，以得到具有所述煅烧产物动态变化特征的X-Ray特征图。这样，进一步将所述热红外特征图和所述X-Ray特征图进行级联以融合所述煅烧产物中热量特征分布以及内部结构变化的特征信息以获得级联特征图，再将得到的级联特征图通过作为过滤器的第三卷积神经网络中进行特征提取，以得到产物特征向量。相应地，在一个具体示例中，使用作为过滤器的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产物特征向量，其中，所述第三卷积神经网络的第一层的输入为所述级联特征图。
更具体地，在步骤S160中，将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量。应 可以理解，对于所述煅烧器在多个预定时间点的煅烧温度值，考虑到所述各个时间点的煅烧温度值之间存在着关联性，因此，在本申请的技术方案中，使用包含嵌入层的上下文编码器对所述煅烧器在多个预定时间点的煅烧温度值进行编码以提取所述多个预定时间点的煅烧温度值之间的基于全局的高维语义特征以更适于表征所述煅烧器内部温度关联的本质特征，从而得到温度特征向量。
更具体地，在步骤S170中，融合所述产物特征向量和所述温度特征向量以得到分类特征向量。应可以理解，考虑到所述产物特征向量是通过级联的所述第一卷积神经网络和所述第三卷积神经网络以及所述三维卷积神经网络和所述第三卷积神经网络获得的，其在特征深度方面会大于通过所述时序编码器获得的所述温度特征向量，因此，在本申请的技术方案中，在将所述产物特征向量，例如记为V 1与所述温度特征向量，例如记为V 2融合之前，首先需要将所述温度特征向量进行深度单应对齐。
更具体地，在步骤S180中，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当前时间点的煅烧器的温度应增大或应减小。也就是，在本申请的技术方案中，在融合所述修正的温度特征向量和所述产物特征向量后，进一步将所述分类特征向量通过分类器中进行分类处理，以获得用于表示当前时间点的煅烧器的温度应增大或应减小的分类结果。
综上，基于本申请实施例的所述用于氟化锂制备的能源管理控制系统的控制方法被阐明，其通过基于深度学习的卷积神经网络模型来对于煅烧器内部的温度关联特征进行实时动态地提取，以及对于煅烧产物的结构变化特征以及内部的热量分布特征进行深层次地挖掘，进而结合这三者在时序上的特征信息来智能地调整所述煅烧器的温度，以在优化能源的同时保证最终氟化锂的成品质量。
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. 一种用于氟化锂制备的能源管理控制系统，其特征在于，包括：煅烧温度采集模块，用于通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体；产物数据采集模块，用于通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；热红外编码模块，用于将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；透视图编码模块，用于将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；关联编码模块，用于将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；能源数据编码模块，用于将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；特征融合模块，用于融合所述产物特征向量和所述温度特征向量以得到分类特征向量；以及能源管理结果生成模块，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当前时间点的煅烧器的温度应增大或应减小。
  2. 根据权利要求1所述的用于氟化锂制备的能源管理控制系统，其中，所述热红外编码模块，进一步用于使用所述第一卷积神经网络的各层在层的正向传递中对输入数据进行：对所述输入数据进行基于二维卷积核的卷积处理以生成卷积特征图；对所述卷积特征图进行池化处理以生成池化特征图；
    对所述池化特征图进行激活处理以生成激活特征图；对所述激活特征图进行沿通道维度的全局平均池化以获得空间特征矩阵；对所述空间特征矩阵进行卷积处理和激活处理以生成权重向量；以及以所述权重向量中各个位置的权重值分别对所述激活特征图的各个特征矩阵进行加权以获得生成特征图；其中，所述第一卷积神经网络模型的最后一层输出的所述生成特征图为所述热红外特征图。
  3. 根据权利要求2所述的用于氟化锂制备的能源管理控制系统，其中，所述透视图编码模块，进一步用于所述使用三维卷积核的第二卷积神经网络在层的正向传递中对输入数据分别进行：基于所述三维卷积核对所述输入数据进行三维卷积处理以得到卷积特征图；对所述卷积特征图进行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以得到激活特征图； 其中，所述第二卷积神经网络的最后一层的输出为所述X-Ray特征图，所述第二卷积神经网络的第一层的输入为所述多个预定时间点的煅烧产物的X-Ray扫描图。
  4. 根据权利要求3所述的用于氟化锂制备的能源管理控制系统，其中，所述关联编码模块，级联单元，用于将所述热红外特征图和所述X-Ray特征图进行级联以获得级联特征图；以及特征提取单元，用于使用作为过滤器的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产物特征向量，其中，所述第三卷积神经网络的第一层的输入为所述级联特征图。
  5. 根据权利要求4所述的用于氟化锂制备的能源管理控制系统，其中，所述能源数据编码模块，包括：输入向量构造单元，用于将所述煅烧器在多个预定时间点的煅烧温度值按照时间维度排列为一维的输入向量；全连接编码单元，用于使用所述时序编码器的全连接层以如下公式对所述输入向量进行全连接编码以提取出所述输入向量中各个位置的特征值的高维隐含特征，其中，所述公式为：
    Figure PCTCN2022117768-appb-100001
    其中X是所述输入向量，Y是输出向量，W是权重矩阵，B是偏置向量，
    Figure PCTCN2022117768-appb-100002
    表示矩阵乘；一维卷积编码单元，用于使用所述时序编码器的一维卷积层以如下公式对所述输入向量进行一维卷积编码以提取出所述输入向量中各个位置的特征值间的高维隐含关联特征，其中，所述公式为：
    Figure PCTCN2022117768-appb-100003
    其中，a为卷积核在x方向上的宽度、F为卷积核参数向量、G为与卷积核函数运算的局部向量矩阵，w为卷积核的尺寸。
  6. 根据权利要求5所述的用于氟化锂制备的能源管理控制系统，其中，所述特征融合模块，包括：向量差分单元，用于计算所述产物特征向量和所述温度特征向量之间的差分特征向量；对数运算单元，用于计算所述差分特征向量中各个位置的特征值的对数函数值以得到对数差分特征向量；深度信息特性值计算单元，用于计算所述差分特征向量的一范数；全场景单应关联矩阵构造单元，用于计算所述产物特征向量与所述温度特征向量的转置向量之间的乘积以得到全场景单应关联矩阵；深度感知单元，用于计算所述全场 景单应关联矩阵的Frobenius范数；对齐单元，用于以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项对所述对数差分特征向量进行处理以得到与所述产物特征向量在深度上单应对齐的修正后温度特征向量；融合单元，用于计算所述修正后温度特征向量和所述产物特征向量的按位置加权和以得到所述分类特征向量。
  7. 根据权利要求6所述的用于氟化锂制备的能源管理控制系统，其中，所述对齐单元，用于以所述差分特征向量的一范数作为加权系数以及以所述全场景单应关联矩阵的Frobenius范数作为偏置项以如下公式对所述对数差分特征向量进行处理以得到所述修正后温度特征向量；其中，所述公式为：
    Figure PCTCN2022117768-appb-100004
    其中，
    Figure PCTCN2022117768-appb-100005
    表示所述差分特征向量的一范数，且||V 1 TV 2|| F表示所述全场景单应关联矩阵的Frobenius范数，
    Figure PCTCN2022117768-appb-100006
    表示按位置点乘、
    Figure PCTCN2022117768-appb-100007
    表示按位置相加，
    Figure PCTCN2022117768-appb-100008
    表示按位置相减。
  8. 根据权利要求7所述的用于氟化锂制备的能源管理控制系统，其中，所述能源管理结果生成模块，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。
  9. 根据权利要求8所述的能源管理结果生成模块，还包括能源控制模块，用于基于所述分类结果，调整所述煅烧器的温度。
  10. 一种用于氟化锂制备的能源管理控制系统的控制方法，其特征在于，包括：通过部署于煅烧器内的温度传感器获取煅烧器在多个预定时间点的煅烧温度值，其中，所述煅烧器用于煅烧形成氟化锂成品的晶体；通过部署于所述煅烧器内的热红外相机和X-Ray扫描仪获取所述多个预定时间点的煅烧产物的热红外图像和X-Ray扫描图；将所述多个预定时间点的煅烧产物的热红外图像通过使用空间注意力机制的第一卷积神经网络以得到热红外特征图；将所述多个预定时间点的煅烧产物的X-Ray扫描图通过使用三维卷积核的第二卷积神经网络以得到X-Ray特征图；将所述热红外特征图和所述X-Ray特征图进行级联后通过作为过滤器的第三卷积神经网络以得到产物特征向量；将所述煅烧器在多个预定时间点的煅烧温度值通过包含一维卷积层和全连接层的时序编码器以得到温度特征向量；融合所述产物特征向量和所 述温度特征向量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当前时间点的煅烧器的温度应增大或应减小。
PCT/CN2022/117768 2022-07-08 2022-09-08 用于氟化锂制备的能源管理控制系统及其控制方法 WO2024007445A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210800838.3A CN115291646B (zh) 2022-07-08 2022-07-08 用于氟化锂制备的能源管理控制系统及其控制方法
CN202210800838.3 2022-07-08

Publications (1)

Publication Number Publication Date
WO2024007445A1 true WO2024007445A1 (zh) 2024-01-11

Family

ID=83822429

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/117768 WO2024007445A1 (zh) 2022-07-08 2022-09-08 用于氟化锂制备的能源管理控制系统及其控制方法

Country Status (2)

Country Link
CN (1) CN115291646B (zh)
WO (1) WO2024007445A1 (zh)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116000297A (zh) * 2023-01-03 2023-04-25 赣州市光华有色金属有限公司 高强度钨镧丝的制备装置及其方法
CN116003126B (zh) * 2023-03-22 2023-06-20 珠海和泽科技有限公司 静电吸盘表面陶瓷材料的制备方法及其系统
CN116048031B (zh) * 2023-03-31 2023-08-04 克拉玛依市蓝润环保科技有限责任公司 用于石油助剂生产的控制系统及其方法
CN116130721B (zh) * 2023-04-11 2023-06-16 杭州鄂达精密机电科技有限公司 氢燃料电池的状态诊断系统及方法
CN116907214B (zh) * 2023-05-09 2024-03-08 广东夏和瓷业有限公司 环保日用陶瓷的制备工艺及其系统
CN116652681A (zh) * 2023-05-24 2023-08-29 浙江沃沃阀门有限公司 锻钢阀阀盖自动化多工位机加工设备及方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180300906A1 (en) * 2015-10-09 2018-10-18 Zhejiang Dahua Technology Co., Ltd. Methods and systems for fusion display of thermal infrared and visible image
CN110544247A (zh) * 2019-09-03 2019-12-06 东莞德福得精密五金制品有限公司 人工智能云计算多途径设备系统运行缺陷检查监控方法
CN111782459A (zh) * 2020-06-03 2020-10-16 东南大学 一种基于管外温度图像与cnn的行波管内部温度预测方法
CN112313067A (zh) * 2018-04-23 2021-02-02 物化股份有限公司 激光烧结中的热控制
CN114001360A (zh) * 2021-12-31 2022-02-01 南京沃谱瑞环境研究院有限公司 一种基于物联网的焚烧回转窑智能控制方法及其系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180300906A1 (en) * 2015-10-09 2018-10-18 Zhejiang Dahua Technology Co., Ltd. Methods and systems for fusion display of thermal infrared and visible image
CN112313067A (zh) * 2018-04-23 2021-02-02 物化股份有限公司 激光烧结中的热控制
CN110544247A (zh) * 2019-09-03 2019-12-06 东莞德福得精密五金制品有限公司 人工智能云计算多途径设备系统运行缺陷检查监控方法
CN111782459A (zh) * 2020-06-03 2020-10-16 东南大学 一种基于管外温度图像与cnn的行波管内部温度预测方法
CN114001360A (zh) * 2021-12-31 2022-02-01 南京沃谱瑞环境研究院有限公司 一种基于物联网的焚烧回转窑智能控制方法及其系统

Also Published As

Publication number Publication date
CN115291646A (zh) 2022-11-04
CN115291646B (zh) 2022-12-23

Similar Documents

Publication Publication Date Title
WO2024007445A1 (zh) 用于氟化锂制备的能源管理控制系统及其控制方法
Eldesokey et al. Propagating confidences through cnns for sparse data regression
CN110570371B (zh) 一种基于多尺度残差学习的图像去雾方法
CN115079572B (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
CN107203745B (zh) 一种基于跨域学习的跨视角动作识别方法
CN114821450B (zh) 太阳能电池板加工用层压机及其控制方法
CN114871486B (zh) 龙门刨床及其加工控制方法
CN115330876B (zh) 基于孪生网络和中心位置估计的目标模板图匹配定位方法
Cooper et al. Information measures for object recognition accommodating signature variability
CN115294639A (zh) 色温可调灯带及其控制方法
CN113343966B (zh) 一种红外与可见光图像文本描述生成方法
CN107808391A (zh) 一种基于特征选择与光滑表示聚类的视频动态目标提取方法
Xu et al. Attention-guided polarization image fusion using salient information distribution
CN111881841B (zh) 一种基于双目视觉的人脸检测识别方法
WO2019090503A1 (zh) 一种智能终端的图像拍摄方法及图像拍摄系统
CN110633732B (zh) 一种基于低秩和联合稀疏的多模态图像识别方法
CN108830163B (zh) 一种基于局部判别cca的海关身份验证系统及方法
CN105323460B (zh) 图像处理设备及其控制方法
CN108965885B (zh) 一种基于帧压缩测量的视频在线重建与移动目标检测方法
CN113284127A (zh) 图像融合显示方法、装置、计算机设备和存储介质
CN114780917B (zh) 一种基于流判别模型的奶粉掺杂非定向检测方法
CN117876346A (zh) 一种绝缘子自主红外三维视觉检测方法和边缘智能装置
UrRehman et al. Inspection on infrared-based image processing
CN115439573B (zh) 一种基于温度信息编码的可见光转红外图像的生成方法
CN112489066B (zh) 一种配电设备红外热成像图像边缘的提取方法

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: 22950006

Country of ref document: EP

Kind code of ref document: A1