Sintering production index identification method, model and system based on multi-source information

Abstract

The present application relates to the technical field of sintering technology, and provides a sintering production index identification method, model and system based on multi-source information. The sintering production index identification method based on multi-source information obtains a visible light image and an infrared image of a sintering machine tail section at the same time, as well as the production parameters of the sintering machine at that time, and then extracts features from the visible light image and the infrared image to obtain a first feature vector of the visible light image and a second feature vector of the infrared image. After eigenvalue dimension reduction is performed on the first feature vector and the second feature vector, feature-level image fusion is performed to obtain dual-source image fusion features, and multi-modal data fusion is performed on the fused dual-source image fusion features and the production parameters to obtain a multi-modal feature vector containing more influencing factors on the key parameters of the sintered ore. Finally, the multi-modal feature vector is used as input to obtain the key parameters of the sintered ore using a pre-established multi-layer perception model.

Description

Technical Field

[0001] The present application relates to the technical field of sintering technology, and in particular to a sintering production index identification method, model and system based on multi-source information. Background Art

[0002] The sintering process is a basic link in the steel production process. It mainly involves placing the iron-containing raw materials that have been batched, mixed, and granulated on a sintering machine trolley, so that the iron-containing raw materials are roasted at high temperature, undergo a series of physical and chemical changes, and then are crushed to form porous, blocky sintered finished ore with a certain strength.

[0003] The tail section of the sintering machine is a key state image that connects the two complex physical and chemical processes of sintering and cooling. It has rich implicit information on the thermal state and the quality and output index characteristics of the sintered ore. In view of the current measurement methods, it takes a long time to obtain the sintered ore indicators after the sintering process parameters are adjusted, and the stability of the sintered ore indicators cannot be guaranteed.

[0004] In the prior art, artificial intelligence methods are also used to intelligently obtain the sintering process conditions from the tail section of the machine and predict the sintered ore indicators, thereby solving the important method of time delay in sintered ore inspection and testing. However, the image processing stage in the current prediction method uses a single image, or less data on the image, resulting in the extracted eigenvalues ​​losing a lot of useful information, which ultimately affects the accuracy of the prediction results. Summary of the invention

[0005] In order to timely and accurately predict the corresponding key parameters of sintered ore based on the tail section information, the embodiment of the present application provides a sintering production index identification method, model and system based on multi-source information.

[0006] The first aspect of the present application provides a sintering production index identification method based on multi-source information, characterized by comprising:

[0007] Obtain visible light images and infrared images of the tail section of the sintering machine at any moment, as well as the production parameters of the sintering machine at that moment;

[0008] Extracting features from the visible light image and the infrared image respectively to obtain a first feature vector of the visible light image and a second feature vector of the infrared image;

[0009] Performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector respectively to obtain a first representative feature and a second representative feature, wherein the first representative feature is the eigenvalue with the highest correlation among the N groups of the first eigenvector, and the second representative feature is the eigenvalue with the highest correlation among the N groups of the second eigenvector;

[0010] The first representative feature and the second representative feature are fused to obtain a dual-source image fusion feature;

[0011] Performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector;

[0012] Inputting the multimodal feature vector into a pre-established multi-layer perception model, the multi-layer perception model comprises an input layer, a hidden layer and an output layer which are fully connected in sequence; the input layer is used to input the multimodal feature vector; the hidden layer is used to perform inference operation according to the multimodal feature vector to obtain the key parameters of the sintered ore; the output layer is used to output the key parameters of the sintered ore;

[0013] The key parameters of the sintered ore output by the multi-layer perception model are obtained.

[0014] Optionally, the key parameters of the sintered ore include FeO content, solid energy consumption, drum strength or yield.

[0015] Optionally, the production parameters of the sintering machine include moisture content of the mixture, material layer thickness, sintering machine speed, sintering end point position, sintering end point temperature, air volume, negative pressure, ignition temperature, feed amount, fuel ratio, fuel particle size and quicklime ratio.

[0016] Optionally, in the step of performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector to obtain the first representative feature and the second representative feature, a specific method adopted is a principal component analysis dimensionality reduction method.

[0017] Optionally, the eigenvalue correlation in the first eigenvector and the eigenvalue correlation in the second eigenvector are both calculated using a grey correlation analysis method.

[0018] Optionally, the multi-layer perception model is pre-built through the following steps:

[0019] Obtain multiple sets of historically collected images and the production parameters of the sintering machine at the time of collecting the multiple sets of images, as well as the corresponding key parameters of the sintered ore; any set of image sets contains visible light images and infrared images of the tail section of the sintering machine at the same time;

[0020] Inputting the multiple sets of image sets and the production parameters of the sintering machine at the time of collecting the multiple sets of image sets, as well as the corresponding key parameters of the sintered ore into a pre-built multi-layer perception model for training and testing;

[0021] Based on the training and testing results, the multi-layer perception model is obtained.

[0022] The second aspect of the present application provides a sintering production index identification model based on multi-source information, and the sintering production index identification model based on multi-source information is used for a sintering production index identification method based on multi-source information provided in the first aspect of the present application, for details not disclosed in the second aspect of the present application.

[0023] The sintering production index identification model based on multi-source information includes:

[0024] An acquisition module is used to acquire the visible light image and infrared image of the tail section of the sintering machine at any moment, as well as the production parameters of the sintering machine at that moment;

[0025] An image feature extraction module, used to extract a first feature vector of a visible light image and a second feature vector of an infrared image;

[0026] A dimension reduction module, used for selecting N groups of eigenvalues ​​with the highest correlation in the first eigenvector to obtain a first representative feature, and for selecting N groups of eigenvalues ​​with the highest correlation in the second eigenvector to obtain a second representative feature;

[0027] A dual-source image feature fusion module, used for fusing the first representative feature and the second representative feature to obtain a dual-source image fusion feature;

[0028] A multimodal data fusion module, used for performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector;

[0029] A multi-layer perception model, the multi-layer perception model includes an input layer, a hidden layer and an output layer which are fully connected in sequence, the input layer is used to input a multimodal feature vector, the hidden layer is used to perform inference operations based on the multimodal feature vector, and the output layer is used to output key parameters of sintered ore.

[0030] Optionally, the input layer includes M dimensions, where M is the number of vectors of the multimodal feature vector, and is used to input the multimodal feature vector;

[0031] The output layer includes P dimensions, where P is the number of categories of the key parameters of sintered ore.

[0032] Optionally, the input layer includes 28 dimensions, which are used to input 8-dimensional first representative features, 8-dimensional second representative features and 12-dimensional production parameters of the multimodal feature vector;

[0033] The number of hidden layers is 2. The input layer is mapped to the first hidden layer after full connection. The first hidden layer includes 56 dimensions. The first hidden layer is mapped to the second hidden layer after full connection. The second hidden layer includes 28 dimensions. The second hidden layer is mapped to the output layer after full connection. The output layer has a total of 4 dimensions.

[0034] The third aspect of the present application also provides a sintering production index identification system based on multi-source information, comprising an infrared thermal imager and a visible light camera, and a controller connected to the infrared thermal imager and the visible light camera, wherein the infrared thermal imager is used to take an infrared image of a cross section of a sintering machine, and the visible light camera is used to take a visible light image of a cross section of a sintering machine tail; the controller is also connected to a sintering machine control terminal, and is used to obtain production parameters of the sintering machine;

[0035] The controller stores a recognition model for performing the following steps:

[0036] Obtain visible light images and infrared images of the tail section of the sintering machine at any moment, as well as the production parameters of the sintering machine at that moment;

[0037] Extracting features from the visible light image and the infrared image respectively to obtain a first feature vector of the visible light image and a second feature vector of the infrared image;

[0038] Performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector respectively to obtain a first representative feature and a second representative feature, wherein the first representative feature is the eigenvalue with the highest correlation among the N groups of the first eigenvector, and the second representative feature is the eigenvalue with the highest correlation among the N groups of the second eigenvector;

[0039] The first representative feature and the second representative feature are fused to obtain a dual-source image fusion feature;

[0040] Performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector;

[0041] Inputting the multimodal feature vector into a pre-established multi-layer perception model, wherein the multi-layer perception model comprises an input layer, a hidden layer and an output layer which are fully connected in sequence, wherein the input layer is used to input the multimodal feature vector, the hidden layer is used to perform reasoning operations according to the multimodal feature vector, and the output layer is used to output key parameters of the sintered ore;

[0042] The key parameters of the sintered ore output by the multi-layer perception model are obtained.

[0043] It can be seen from the above technical scheme that the present application provides a sintering production index identification method, model and system based on multi-source information. The sintering production index identification method based on multi-source information obtains the visible light image and infrared image of the tail section of the sintering machine at the same time, and the production parameters of the sintering machine at that time, and then extracts features from the visible light image and the infrared image to obtain the first feature vector of the visible light image and the second feature vector of the infrared image. After eigenvalue dimensionality reduction is performed on the first feature vector and the second feature vector, feature-level image fusion is performed to obtain dual-source image fusion features, and the fused dual-source image fusion features are subjected to multimodal data fusion with the production parameters to obtain a multimodal feature vector containing more influencing factors on the key parameters of sintered ore. Finally, the multimodal feature vector is used as input to obtain the key parameters of sintered ore using a pre-established multi-layer perception model. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0045] Figure 1 A schematic flow chart of a sintering production index identification method based on multi-source information provided in an embodiment of the present application:

[0046] Figure 2 A schematic diagram of the VGG network structure provided in the embodiment of the present application;

[0047] Figure 3 A schematic diagram of dual-source image feature fusion provided in an embodiment of the present application;

[0048] Figure 4 A schematic diagram of the structure of a multi-layer perceptron provided in an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the multi-layer perception model construction process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to timely and accurately predict the corresponding key parameters of sintered ore based on the tail section information, the embodiment of the present application provides a sintering production index identification method, model and system based on multi-source information.

[0051] like Figure 1 As shown, the first aspect of the embodiment of the present application discloses a sintering production index identification method based on multi-source information, including steps S101 to S107.

[0052] Step S101, obtaining a visible light image and an infrared image of the tail section of the sintering machine at any moment, as well as the production parameters of the sintering machine at that moment.

[0053] The visible light image is acquired by a visible light camera arranged at the end of the sintering machine, and the infrared image is acquired by an infrared thermal imager arranged at the end of the sintering machine. The visible light camera and the infrared thermal imager are controlled by the same host and adopt binocular vision principle to simultaneously acquire the visible light image and the infrared image.

[0054] The production parameters of the sintering machine include the moisture content of the mixture, the thickness of the material layer, the sintering machine speed, the sintering end point position, the sintering end point temperature, the air volume, the negative pressure, the ignition temperature, the feeding amount, the fuel ratio, the fuel particle size and the quicklime ratio. The production parameters of the sintering machine are provided by the control system of the sintering machine.

[0055] The sintering production index identification method based on multi-source information provided in the embodiment of the present application needs to fuse the image parameters of the tail section of the sintering machine with the production parameters of the sintering machine. Before fusing the image parameters with the production parameters, it is necessary to first perform data fusion on the visible light image and the infrared image, that is, feature-level image fusion. The feature-level image fusion is to extract feature information from the source image (visible light image and infrared image). These feature information source images are target or area areas, such as the sintering end face edge area, fire layer area, pore area or target distribution information. These feature information are then analyzed, processed and integrated to obtain the fused image features. This algorithm uses a deep learning VGG network to extract image features. Figure 2 As shown, it is a schematic diagram of the VGG network structure provided in an embodiment of the present application. After the VGG16 network extracts the features of the infrared image, the output of each convolutional layer visualizes the image. The features extracted by the VGG16 network at each layer have different meanings. The network can automatically adjust the parameters to obtain more features that cannot be extracted manually. At the same time, after the pooling operation, the size of the feature map becomes smaller and smaller, which can automatically eliminate redundant information and background interference and retain the target information of the image as much as possible. Specifically, it includes steps S102 to S104.

[0056] Step S102 , performing feature extraction on the visible light image and the infrared image respectively to obtain a first feature vector of the visible light image and a second feature vector of the infrared image.

[0057] The infrared image and the visible light image are respectively input into the VGG16 network. After feature extraction by the VGG16 network, the input images (infrared image and visible light image) are converted into feature vectors with certain meanings. Then, the output of the second fully connected layer of the VGG16 network is intercepted to obtain the first feature vector of the visible light image and the second feature vector of the infrared image. In the embodiment of the present application, the feature vector sizes of the first feature vector and the second feature vector are 1×1×4096.

[0058] Step S103, performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector respectively to obtain a first representative feature and a second representative feature, wherein the first representative feature is the eigenvalue with the highest correlation in the N groups of the first eigenvector, and the second representative feature is the eigenvalue with the highest correlation in the N groups of the second eigenvector.

[0059] Specifically, the VGG16 network uses the PCA dimensionality reduction method (principal component analysis dimensionality reduction method) to reduce the dimensionality of the first eigenvector and the second eigenvector, selects the N groups of eigenvalues ​​with the highest correlation in the first eigenvector as the first representative feature, and selects the N groups of eigenvalues ​​with the highest correlation in the second eigenvector as the second representative feature, wherein the eigenvalue correlation in the first eigenvector and the eigenvalue correlation in the second eigenvector are both calculated using the grey correlation analysis method. Specifically, in the embodiment of the present application, N=8 is selected.

[0060] Step S104: fuse the first representative feature and the second representative feature to obtain a dual-source image fusion feature.

[0061] like Figure 3 As shown, it is a schematic diagram of a dual-source image feature fusion provided in an embodiment of the present application. In the embodiment of the present application, after the first representative feature and the second representative feature are fused using the VGG16 network, the two-dimensional image data represented by the first representative feature and the second representative feature are also converted into one-dimensional feature data.

[0062] Step S105, performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector.

[0063] Although the dual-source image fusion features generated after feature-level image fusion and the production parameters of the sintering machine belong to the same one-dimensional data, the semantic structure modalities of the two are different. Therefore, it is necessary to combine the semantics of the converted single-modal feature vectors to achieve multi-modal fusion (multi-source heterogeneous data fusion), as shown in the following model:

[0064] F=f(W1V1+W2V2+…+W N V N );

[0065] Among them, F is the output result in the shared semantic subspace; V is the input of the feature vector (production parameters and dual-source image fusion features); W is the weight coefficient; the subscripts represent different modes, that is, the one-dimensional features and production parameter data after the feature-level image fusion; all submodal semantics are converted to the shared subspace by mapping f. Since it is temporarily impossible to know in advance the specific impact of each mode on the sintering process and the final prediction result relationship, the weight coefficient W represents the importance of each mode, and its specific value can be set according to experience, or it can be obtained in advance by fusion training. The final F is the multimodal feature vector after the fusion of multimodal data, and the multimodal feature vector represents the dual-source image (visible light image and infrared image) and production parameters.

[0066] Step S106, inputting the multimodal feature vector into a pre-established multi-layer perception model, wherein the multi-layer perception model includes an input layer, a hidden layer, and an output layer that are fully connected in sequence; the input layer is used to input the multimodal feature vector; the hidden layer is used to perform inference operations based on the multimodal feature vector to obtain key parameters of the sintered ore; and the output layer is used to output the key parameters of the sintered ore.

[0067] For the prediction of key parameters of sintered ore, the embodiment of the present application uses multi-layer perceptron nonlinear regression. Figure 4 As shown, it is a schematic diagram of the multi-layer perceptron structure provided in an embodiment of the present application. The input layer is determined by the number of input vectors, that is, the output F feature vector of the fusion algorithm; the output layer is determined by the number of categories that need to be predicted, that is, the number of categories of key parameters of sintered ore that need to be predicted. In the embodiment of the present application, the key parameters of sintered ore include FeO content, solid energy consumption, drum strength or yield; the hidden layer can be customized according to the actual prediction object.

[0068] In the multi-layer perceptron prediction and analysis structure, an input layer, two hidden layers and an output layer are set, and the entire network is connected in a fully connected manner. The input layer has a total of 28 dimensions of input data, including 8-dimensional infrared thermal image data, 8-dimensional visible light image data and 12-dimensional production parameters; the input layer is mapped to the first hidden layer after full connection, with a total of 56 dimensions; the first hidden layer is mapped to the second hidden layer after full connection again, with a total of 28 dimensions; the second hidden layer is finally mapped to the output layer after full connection, with a total of 4 dimensions, which respectively predicts FeO content, solid energy consumption, drum strength or finished product rate.

[0069] In the embodiment of the present application, the multi-layer perception model is pre-constructed by the following steps: Figure 5 As shown, it includes steps S501 to S503.

[0070] Step S501, obtaining multiple sets of image sets collected historically and production parameters of the sintering machine at the time of collecting the multiple sets of image sets, as well as corresponding key parameters of the sintered ore; any set of image sets contains visible light images and infrared images of the tail section of the sintering machine at the same time;

[0071] Step S502: input the multiple image sets, the sintering machine production parameters at the time of collecting the multiple image sets, and the corresponding sintered ore key parameters into a pre-built multi-layer perception model for training and testing.

[0072] Step S503, based on the training and testing results, obtain the multi-layer perception model.

[0073] In the process of establishing a multi-layer perception model in the embodiment of the present application, the nonlinear regression uses the mean square error as the evaluation index, the number of outputs of the output layer is set to 4, and the output O is directly provided to the square loss function used in the linear regression for loss calculation, that is, the minimum mean square error (MMSE-Loss).

[0074]

[0075] Among them, L is the loss value, O i Represents the i-th output result.

[0076] For each output loss value, the calculation method is as follows:

[0077]

[0078] Among them, O i ' indicates the i-th prediction result.

[0079] In the process of training the model, the goal is to make the defined square loss function Loss as small as possible.

[0080] Step S107, obtaining the key parameters of the sintered ore output by the multi-layer perception model.

[0081] The embodiment of the present application provides a sintering production index identification method based on multi-source information, which obtains a visible light image and an infrared image of the tail section of a sintering machine at the same moment, as well as the production parameters of the sintering machine at that moment, and then performs feature extraction on the visible light image and the infrared image to obtain a first feature vector of the visible light image and a second feature vector of the infrared image. After eigenvalue dimensionality reduction is performed on the first feature vector and the second feature vector, feature-level image fusion is performed to obtain dual-source image fusion features, and the fused dual-source image fusion features are subjected to multimodal data fusion with the production parameters to obtain a multimodal feature vector containing more influencing factors on the key parameters of sintered ore. Finally, the multimodal feature vector is used as input, and a pre-established multi-layer perception model is used to obtain the key parameters of sintered ore output by the multi-layer perception model.

[0082] The second aspect of the embodiment of the present application provides a sintering production index identification model based on multi-source information. The sintering production index identification model based on multi-source information is used for a sintering production index identification method based on multi-source information provided in the first aspect of the embodiment of the present application. For details not disclosed in the second aspect of the embodiment of the present application, please refer to the contents disclosed in the first aspect of the embodiment of the present application.

[0083] The sintering production index identification model based on multi-source information includes:

[0084] The acquisition module is used to acquire the visible light image and infrared image of the tail section of the sintering machine at any moment, as well as the production parameters of the sintering machine at that moment.

[0085] The image feature extraction module is used to extract the first feature vector of the visible light image and the second feature vector of the infrared image.

[0086] The dimension reduction module is used to select the N groups of eigenvalues ​​with the highest correlation in the first eigenvector to obtain the first representative feature, and to select the N groups of eigenvalues ​​with the highest correlation in the second eigenvector to obtain the second representative feature.

[0087] The dual-source image feature fusion module is used to fuse the first representative feature and the second representative feature to obtain the dual-source image fusion feature.

[0088] The multimodal data fusion module is used to perform multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector.

[0089] A multi-layer perception model, the multi-layer perception model includes an input layer, a hidden layer and an output layer which are fully connected in sequence, the input layer is used to input a multimodal feature vector, the hidden layer is used to perform inference operations based on the multimodal feature vector, and the output layer is used to output key parameters of sintered ore.

[0090] Furthermore, the input layer includes M dimensions, where M is the number of vectors of the multimodal feature vector, and is used to input the multimodal feature vector.

[0091] The output layer includes P dimensions, where P is the number of categories of the key parameters of sintered ore.

[0092] Furthermore, the input layer includes 28 dimensions, which are used to input 8 dimensions of first representative features, 8 dimensions of second representative features and 12 dimensions of production parameters of the multimodal feature vector.

[0093] The number of hidden layers is 2. The input layer is mapped to the first hidden layer after full connection. The first hidden layer includes 56 dimensions. The first hidden layer is mapped to the second hidden layer after full connection. The second hidden layer includes 28 dimensions. The second hidden layer is mapped to the output layer after full connection. The output layer has a total of 4 dimensions.

[0094] The third aspect of the embodiment of the present application also provides a sintering production index identification system based on multi-source information, including an infrared thermal imager and a visible light camera, and a controller connected to the infrared thermal imager and the visible light camera, wherein the infrared thermal imager is used to take infrared images of the cross-section of the sintering machine, and the visible light camera is used to take visible light images of the cross-section of the tail of the sintering machine; the controller is also connected to the sintering machine control terminal to obtain the production parameters of the sintering machine.

[0095] The controller stores a recognition model for performing the following steps:

[0096] Obtain visible light images and infrared images of the tail section of the sintering machine at any moment, as well as the production parameters of the sintering machine at that moment.

[0097] Feature extraction is performed on the visible light image and the infrared image respectively to obtain a first feature vector of the visible light image and a second feature vector of the infrared image.

[0098] Perform eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector respectively to obtain a first representative feature and a second representative feature, wherein the first representative feature is the eigenvalue with the highest correlation in N groups of the first eigenvector, and the second representative feature is the eigenvalue with the highest correlation in N groups of the second eigenvector.

[0099] The first representative feature and the second representative feature are fused to obtain a dual-source image fusion feature.

[0100] The production parameters are fused with the dual-source image fusion features to perform multimodal data fusion to obtain a multimodal feature vector.

[0101] The multimodal feature vector is input into a pre-established multi-layer perception model, which includes an input layer, a hidden layer and an output layer that are fully connected in sequence. The input layer is used to input the multimodal feature vector, the hidden layer is used to perform inference operations based on the multimodal feature vector, and the output layer is used to output key parameters of the sintered ore.

[0102] The key parameters of the sintered ore output by the multi-layer perception model are obtained.

[0103] It can be seen from the above technical scheme that the embodiment of the present application provides a sintering production index identification method and system based on multi-source information. The sintering production index identification method based on multi-source information obtains the visible light image and infrared image of the tail section of the sintering machine at the same time, and the production parameters of the sintering machine at that time, and then extracts features from the visible light image and the infrared image to obtain the first feature vector of the visible light image and the second feature vector of the infrared image. After eigenvalue dimensionality reduction is performed on the first feature vector and the second feature vector, feature-level image fusion is performed to obtain dual-source image fusion features, and the fused dual-source image fusion features are subjected to multimodal data fusion with the production parameters to obtain a multimodal feature vector containing more influencing factors on the key parameters of sintered ore. Finally, the multimodal feature vector is used as input to obtain the key parameters of sintered ore using a pre-established multi-layer perception model.

[0104] The present application is described in detail above in conjunction with specific implementation methods and exemplary examples, but these descriptions cannot be understood as limiting the present application. Those skilled in the art understand that, without departing from the spirit and scope of the present application, a variety of equivalent replacements, modifications or improvements can be made to the technical solution of the present application and its implementation methods, all of which fall within the scope of the present application. The scope of protection of the present application shall be subject to the attached claims.

Claims

1. A sintering production index identification method based on multi-source information, characterized in that: include: Obtaining visible light images and infrared images of the tail section of the sintering machine at any time, as well as the production parameters of the sintering machine at that time, the production parameters of the sintering machine including the moisture content of the mixture, the thickness of the material layer, the sintering machine speed, the sintering end position, the sintering end temperature, the air volume, the negative pressure, the ignition temperature, the feeding amount, the fuel ratio, the fuel particle size and the quicklime ratio; Extracting features from the visible light image and the infrared image respectively to obtain a first feature vector of the visible light image and a second feature vector of the infrared image; Performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector respectively to obtain a first representative feature and a second representative feature, wherein the first representative feature is the eigenvalue with the highest correlation among the N groups of the first eigenvector, and the second representative feature is the eigenvalue with the highest correlation among the N groups of the second eigenvector; The first representative feature and the second representative feature are fused to obtain a dual-source image fusion feature; Performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector; Inputting the multimodal feature vector into a pre-established multi-layer perception model, the multi-layer perception model comprises an input layer, a hidden layer and an output layer which are fully connected in sequence; the input layer is used to input the multimodal feature vector; the hidden layer is used to perform reasoning operations according to the multimodal feature vector to obtain key parameters of the sintered ore; the output layer is used to output the key parameters of the sintered ore, and the key parameters of the sintered ore include FeO content, solid energy consumption, drum strength or yield; The key parameters of the sintered ore output by the multi-layer perception model are obtained.

2. The sintering production index identification method based on multi-source information according to claim 1 is characterized in that: In the step of performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector to obtain the first representative feature and the second representative feature, a specific method used is the principal component analysis dimensionality reduction method.

3. The sintering production index identification method based on multi-source information according to claim 1 is characterized in that: The eigenvalue correlation in the first eigenvector and the eigenvalue correlation in the second eigenvector are both calculated using a grey correlation analysis method.

4. The sintering production index identification method based on multi-source information according to claim 1 is characterized in that: The multi-layer perception model is pre-built through the following steps: Obtain multiple sets of historically collected images and the production parameters of the sintering machine at the time of collecting the multiple sets of images, as well as the corresponding key parameters of the sintered ore; any set of image sets contains visible light images and infrared images of the tail section of the sintering machine at the same time; Inputting the multiple sets of image sets and the production parameters of the sintering machine at the time of collecting the multiple sets of image sets, as well as the corresponding key parameters of the sintered ore into a pre-built multi-layer perception model for training and testing; Based on the training and testing results, the multi-layer perception model is obtained.

5. A sintering production index identification model based on multi-source information, characterized in that: The sintering production index identification model based on multi-source information is used to execute the sintering production index identification method based on multi-source information according to any one of claims 1 to 4, and the sintering production index identification model based on multi-source information includes: An acquisition module is used to acquire a visible light image and an infrared image of the tail section of the sintering machine at any time, as well as the production parameters of the sintering machine at that time, wherein the production parameters of the sintering machine include the moisture content of the mixture, the thickness of the material layer, the sintering machine speed, the sintering end position, the sintering end temperature, the air volume, the negative pressure, the ignition temperature, the feeding amount, the fuel ratio, the fuel particle size and the quicklime ratio; An image feature extraction module, used to extract a first feature vector of a visible light image and a second feature vector of an infrared image; A dimension reduction module, used for selecting N groups of eigenvalues ​​with the highest correlation in the first eigenvector to obtain a first representative feature, and for selecting N groups of eigenvalues ​​with the highest correlation in the second eigenvector to obtain a second representative feature; A dual-source image feature fusion module, used for fusing the first representative feature and the second representative feature to obtain a dual-source image fusion feature; A multimodal data fusion module, used for performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector; A multi-layer perception model, the multi-layer perception model includes an input layer, a hidden layer and an output layer which are fully connected in sequence, the input layer is used to input a multimodal feature vector, the hidden layer is used to perform inference operations based on the multimodal feature vector, and the output layer is used to output key parameters of sintered ore, the key parameters of sintered ore include FeO content, solid energy consumption, drum strength or yield.

6. The sintering production index identification model based on multi-source information according to claim 5 is characterized in that: The input layer includes M dimensions, where M is the number of vectors of the multimodal feature vector, and is used to input the multimodal feature vector; The output layer includes P dimensions, where P is the number of categories of the key parameters of sintered ore.

7. The sintering production index identification model based on multi-source information according to claim 5 or 6, characterized in that: include: The input layer includes 28 dimensions, which are used to input 8-dimensional first representative features, 8-dimensional second representative features and 12-dimensional production parameters of the multimodal feature vector; The number of hidden layers is 2. The input layer is mapped to the first hidden layer after full connection. The first hidden layer includes 56 dimensions. The first hidden layer is mapped to the second hidden layer after full connection. The second hidden layer includes 28 dimensions. The second hidden layer is mapped to the output layer after full connection. The output layer has a total of 4 dimensions.

8. A sintering production index identification system based on multi-source information, characterized in that: It includes an infrared thermal imager and a visible light camera, and a controller connected to the infrared thermal imager and the visible light camera, wherein the infrared thermal imager is used to take an infrared image of the cross section of the sintering machine, and the visible light camera is used to take a visible light image of the cross section of the tail of the sintering machine; the controller is also connected to the sintering machine control terminal to obtain the production parameters of the sintering machine; The controller stores a recognition model for performing the following steps: Obtaining visible light images and infrared images of the tail section of the sintering machine at any time, as well as the production parameters of the sintering machine at that time, the production parameters of the sintering machine including the moisture content of the mixture, the thickness of the material layer, the sintering machine speed, the sintering end position, the sintering end temperature, the air volume, the negative pressure, the ignition temperature, the feeding amount, the fuel ratio, the fuel particle size and the quicklime ratio; Extracting features from the visible light image and the infrared image respectively to obtain a first feature vector of the visible light image and a second feature vector of the infrared image; Performing eigenvalue dimensionality reduction on the first eigenvector and the second eigenvector respectively to obtain a first representative feature and a second representative feature, wherein the first representative feature is the eigenvalue with the highest correlation among the N groups of the first eigenvector, and the second representative feature is the eigenvalue with the highest correlation among the N groups of the second eigenvector; The first representative feature and the second representative feature are fused to obtain a dual-source image fusion feature; Performing multimodal data fusion on the production parameters and the dual-source image fusion features to obtain a multimodal feature vector; Inputting the multimodal feature vector into a pre-established multi-layer perception model, the multi-layer perception model comprises an input layer, a hidden layer and an output layer which are fully connected in sequence, the input layer is used to input the multimodal feature vector, the hidden layer is used to perform reasoning operations according to the multimodal feature vector, and the output layer is used to output key parameters of the sintered ore, the key parameters of the sintered ore include FeO content, solid energy consumption, drum strength or yield rate; The key parameters of the sintered ore output by the multi-layer perception model are obtained.

Images