Blast furnace working condition identification method and system fusing process variable and video image

Abstract

The invention relates to the technical field of industrial working condition recognition, and discloses a blast furnace working condition recognition method and system fusing a process variable and a video image, and the method comprises the steps: carrying out the interpolation processing of the process variable, carrying out the averaging of a video image frame, and aligning the multi-source heterogeneous information to the same time step; thus, multi-source heterogeneous information with inconsistent sampling frequencies is aligned, a deep neural network model jointly driven by process variables and video images is constructed, representative working condition features in the multi-source heterogeneous information are extracted, the problem that working condition information is not completely utilized when only a single information source is used is solved, and the method is suitable for the multi-source heterogeneous information. A double-layer residual error connection module and a model stability evaluation index are designed, and accurate recognition of the working condition of the blast furnace is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of industrial operating condition identification, and in particular to a blast furnace operating condition identification method and system that integrates process variables and video images. Background Art

[0002] As the core process of the iron and steel metallurgical industry, the stable and smooth operation of blast furnace ironmaking is an important prerequisite for high-quality and efficient steel production. The blast furnace operating condition is a direct reflection of the blast furnace's operating status and can provide key feedback information for the closed-loop control of blast furnace ironmaking. Therefore, accurate identification of blast furnace operating conditions is a mandatory requirement in the blast furnace ironmaking process.

[0003] With the continuous advancement of key technologies such as sensor equipment and distributed control systems, a large amount of multi-source heterogeneous information has been accumulated in the blast furnace ironmaking process, including but not limited to process variables, real-time video streams and static images. According to the information source used for condition identification, the existing blast furnace condition identification methods can be summarized into two categories: condition identification methods based on process variables and condition identification methods based on video images. The condition identification method based on process variables uses the process variables collected by each sensor during the operation of the blast furnace for statistical analysis to establish a data-driven model, but this method requires a large number of samples for model training. In actual situations, the number of samples of abnormal conditions is often scarce, so it is difficult to train a data-driven model with good generalization performance. At present, the condition identification method based on video images often establishes an image feature extraction model through image enhancement, semantic segmentation and other technologies, but this method has high complexity, long training time and large computing resources required, which makes it face adaptability challenges in practical industrial applications. In addition, the existing technology often uses only a single information source, which makes the condition information incompletely utilized and the recognition accuracy low. Summary of the invention

[0004] The present application provides a blast furnace operating condition identification method and system integrating process variables and video images to solve the problems existing in the prior art.

[0005] In order to achieve the above objectives, the present application is implemented through the following technical solutions:

[0006] In a first aspect, the present application provides a method for identifying blast furnace operating conditions by integrating process variables and video images, comprising:

[0007] S1: Collect process variables of blast furnace working conditions and material surface video images to obtain multi-source heterogeneous information;

[0008] S2: performing interpolation processing on the process variables and averaging the video image frames to align the multi-source heterogeneous information to the same time step;

[0009] S3: Build a single-layer fusion network architecture;

[0010] S4: extracting a preliminary feature set of the aligned multi-source heterogeneous information based on the single-layer fusion network architecture;

[0011] S5: Based on the preliminary feature set and the pre-built two-layer fusion network architecture, training and optimization are performed to obtain a two-layer fusion network model;

[0012] S6: Identify the blast furnace operating condition type based on the double-layer fusion network model.

[0013] In a second aspect, the present application provides a blast furnace operating condition identification system that integrates process variables and video images, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0014] Beneficial effects:

[0015] The blast furnace operating condition recognition method that integrates process variables and video images provided in the present application interpolates the process variables and averages the video image frames, thereby aligning multi-source heterogeneous information to the same time step. In this way, multi-source heterogeneous information with inconsistent sampling frequencies is aligned, a deep neural network model jointly driven by process variables and video images is constructed, representative operating condition features are extracted from the multi-source heterogeneous information, and the problem of incomplete utilization of operating condition information when only a single information source is used is overcome. A double-layer residual connection module and a model stability evaluation index are designed to achieve accurate recognition of blast furnace operating conditions.

[0016] In a further technical solution, the present application proposes a deep neural network model jointly driven by process variables and video images, extracts process variables and video image features based on LSTM and CNN respectively, and then simplifies the features through CCA, and realizes the fusion of multi-source heterogeneous information;

[0017] In a further technical solution, the present application designs a two-layer residual connection module, which concatenates the preliminary features extracted by the single-layer network with the top-level input data after data alignment, thereby achieving effective extraction and full utilization of important features of multi-source heterogeneous data;

[0018] In a further technical solution, the present application designs a model training stability evaluation index, analyzes the time consumption of the training process, the accuracy of the training set, the loss of the training set, the accuracy of the test set and the correlation of the test set, and comprehensively evaluates the optimization speed, accuracy and fluctuation of the model; BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1This is one of the flow charts of a blast furnace operating condition identification method integrating process variables and video images according to a preferred embodiment of the present application;

[0020] Figure 2 This is a second flow chart of a blast furnace operating condition identification method that integrates process variables and video images according to a preferred embodiment of the present application;

[0021] Figure 3 (a) is a blast furnace top temperature variation curve of a preferred embodiment of the present application; Figure 3 (b) is a material surface image of each working condition of the preferred embodiment of the present application;

[0022] Figure 4 This is a schematic diagram of the indicator change curve of the model training process of the preferred embodiment of the present application;

[0023] Figure 5 This is a schematic diagram of the confusion matrix for working condition identification of a preferred embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solution of the present application is described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0025] Unless otherwise defined, the technical terms or scientific terms used in this application should be understood by people with ordinary skills in the field to which this application belongs. The "first", "second" and similar words used in this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, similar words such as "one" or "one" do not indicate quantity restrictions, but indicate the existence of at least one. Similar words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0026] See also Figure 1 The present application provides a method for identifying blast furnace operating conditions by integrating process variables and video images, comprising:

[0027] S1: Collect process variables of blast furnace working conditions and material surface video images to obtain multi-source heterogeneous information;

[0028] S2: performing interpolation processing on the process variables and averaging the video image frames to align the multi-source heterogeneous information to the same time step;

[0029] S3: Build a single-layer fusion network architecture;

[0030] S4: extracting a preliminary feature set of the aligned multi-source heterogeneous information based on the single-layer fusion network architecture;

[0031] S5: Based on the preliminary feature set and the pre-built two-layer fusion network architecture, training and optimization are performed to obtain a two-layer fusion network model;

[0032] S6: Identify the blast furnace operating condition type based on the double-layer fusion network model.

[0033] In this implementation, process variables are collected by a sensor cluster deployed inside and outside the blast furnace, and the material surface video image is obtained by taking a high-temperature industrial endoscope on the top of the blast furnace. Specifically, the process variables include parameters such as top temperature, hot blast pressure, oxygen-enriched flow, gas composition, and permeability index, which describe the state data of the blast furnace ironmaking process in detail. The video image is presented as a continuous information flow, intuitively showing the changes in the morphology of the materials in the blast furnace and their movement state.

[0034] Below, a specific implementation scheme of the blast furnace operating condition recognition method integrating process variables and video images is described in detail with a complete example as follows:

[0035] (1) Alignment of multi-source heterogeneous data

[0036] It is worth pointing out that in the blast furnace ironmaking process, the sampling methods of multi-source heterogeneous information are different, and the frequency of information collection shows significant differences. Specifically, the variables of the blast furnace ironmaking process exist in a discrete form, with a sampling interval of ΔT seconds, including parameters such as furnace top temperature, hot blast pressure, oxygen-enriched flow, gas composition, and permeability index, which describe the state data of the blast furnace ironmaking process in detail. The video image is presented as a continuous stream of information, which intuitively shows the changes in the morphology of materials and their motion state in the blast furnace. In order to synchronize the two in the time dimension, this application uses a data alignment method based on Gaussian functions to interpolate process variables and average video image frames, which specifically includes the following steps:

[0037] Step 1: Let Ψ represent the original process variable set, which contains p process variable types and m samples. c z represents the zth time window sequence of the cth process variable, where c = 1, 2, ..., p and z = 1, 2, ..., s. By setting a certain window width length, m samples are divided into s time window sequences, so the time window sequence The number of samples contained in is

[0038] Step 2: Interpolate the sample points in each time window sequence in turn, and let n be the value in the time window sequence. The number of new samples interpolated between any two original sample points in (t i ,x i ) represents the i-th original sample point, i=1,2,...,q, let point represents the jth interpolated new sample point between the i-th original sample point and its next adjacent original sample point, where Δt = 1 and j = 1,...,n, and n = ΔT-1. Every second, a new sample is added at each second. Step 3: Calculate the time window sequence The interpolation Gaussian function is modified to K(t), which satisfies the following relationship:

[0039]

[0040] Where σ is the bandwidth, t is the time, and μ is the mean;

[0041] Based on the Gaussian function K(t), the The size of the new sample point to be interpolated satisfies the following relationship:

[0042]

[0043] In the formula, x m is the true value of the mth original data point, t m is the time at which the mth original data point is located, and K represents the Gaussian function.

[0044] Step 4: As a continuous information stream, the video contains frame images per second. In order to align it with the process variable set with one sample per second after interpolation, this application will average all the frame images contained per second to obtain an image representing the time per second. Assume that the video contains num frame images per second, calculate the average value of num frame images at each pixel position per second, and obtain an image representing the time per second F η , the image F η Contains information about all video frame images within that second.

[0045] Through this data alignment method, process variables and video images with inconsistent sampling frequencies can be effectively aligned. The process variable after data alignment is set as X and the video image is set as Y, which provides a solid and reliable data foundation for subsequent modeling and blast furnace operating condition identification tasks.

[0046] (2) Designing a single-layer fusion network architecture

[0047] This application proposes a deep neural network model based on the joint driving of process variables and video images, in which a single-layer fusion network architecture is as follows: Figure 2 As shown in the “First Layer” in .

[0048] In the single-layer fusion network architecture, the process variable X and video image Y after data alignment are input, and the first deep neural network model LSTM and the second deep neural network model CNN are used to extract the features of X and Y respectively, and two feature sets F of heterogeneous information are obtained. 1 and F 2 .

[0049] For data sets X and Y, calculate the product of the covariance matrix:

[0050]

[0051] Among them, Σ XX ,Σ YY and Σ XY are the covariance matrices of X and Y and their cross-covariance matrices, γ is the product of the covariance matrices, and the singular value decomposition of γ satisfies the following relationship:

[0052] γ=ΓΣR T ; (4)

[0053]

[0054]

[0055] Among them, Γ, Σ and R are the matrix factors after singular value decomposition, J and L are weight parameters, are vector spaces with dimensions l×l, are vector spaces of dimensions m×m, and T represents transpose;

[0056] Calculate the feature set u after being simplified by the information simplifier 1 and v 1 , satisfying the following relationship:

[0057]

[0058]

[0059] Among them, the parameter J 1 , L 1 and Σ 1 are the weight parameters calculated by the first deep neural network model.

[0060] The role of the information simplifier is to reduce the typical correlation between two heterogeneous information feature sets, thereby reducing the common features of the heterogeneous information feature sets and retaining most of the unique features of the heterogeneous information feature sets.

[0061] (3) Propose a two-layer residual connection module

[0062] This application proposes a double-layer residual connection module to extract the feature set u through a single-layer fusion network architecture. 1 and v 1 , respectively concatenated with X and Y to obtain data sets X′ and Y′

[0063] X′=u 1 ⊕X (9)

[0064] Y′=v 1 ⊕Y (10)

[0065] Among them, ⊕ is the data concatenation operation.

[0066] Then, build a new fusion network architecture with the same structure but different parameters, such as Figure 2 As shown in the “second layer” in , the datasets X′ and Y′ are input into the new fusion network architecture to extract features, and two feature sets F of heterogeneous information are obtained. 3 and F 4 :

[0067]

[0068]

[0069] Among them, J 2 , L 2 and Σ 2 is the weight parameter calculated by the second deep neural network model;

[0070] Next, concatenate the feature set u 2 and v 2 , and input it into the fully connected layer and the Softmax output layer, and the blast furnace working condition recognition result is calculated. The model training and parameter optimization are performed through the cross entropy loss function of the following model

[0071]

[0072] Among them, θ 1 and θ 2 are the model parameter sets of LSTM and CNN in the first layer, θ 3 and θ 4 are the model parameter sets of LSTM and CNN in the second layer, θ 5 is the parameter set of the fully connected layer, is the parameter combination when the network achieves the optimal effect, y i represents the true label of the i-th sample, Represents the recognition result of the i-th sample.

[0073] The two-layer residual connection module enables the output of the previous layer to bypass the intermediate layer and connect directly to the subsequent layer. This architecture promotes the learning of residual mapping and ensures the effective propagation of input information within the network. It is worth noting that the network contains two independent feature extraction models, each trained on a different dataset, allowing for relatively independent training processes. Despite this independence, both feature extraction models use the same optimization strategy and loss function, thereby improving overall performance while maintaining model autonomy.

[0074] (4) Design model training stability evaluation indicators

[0075] It is worth pointing out that in order to evaluate the stability of the model during training, this application designs a model training stability evaluation index to evaluate the optimization speed, accuracy and fluctuation of the model. For the H training processes of the model, each training process is assumed to include h training stages, and the following statistical information of each training process is calculated:

[0076]

[0077]

[0078]

[0079]

[0080] in, The value is [t, d, l, e, g], which represents the five indicators of training process time, training set accuracy, training set loss, test set accuracy, and test set relevance. i is the data sequence of the corresponding indicator, max() and min() represent the solution sequence list i The maximum and minimum values ​​in . When it represents the accuracy of the training set or the accuracy of the test set, formula (14) takes max(); when it represents the loss of the training set or the correlation of the test set, formula (14) takes min(). represents the value of the corresponding element in the jth training stage of the i-th training process, and Respectively represent the mean and standard deviation of the corresponding elements in the i-th training process, where i = 1, 2, ..., H and j = 1, 2, ..., h. This application calculates and To evaluate the accuracy of the model during training.

[0081] In order to evaluate the fluctuation during model training, this application defines a fluctuation coefficient They represent the volatility coefficients of the five indicators t, d, l, e, and g, respectively, as shown in formula (17). The smaller the value, the smaller the model volatility relative to the mean.

[0082] In order to quantify the optimization speed of model training, this application defines the training stage sequence of each indicator when the four indicators d, l, e, and g reach the set threshold of the optimal value of each indicator, that is, For the training set accuracy and test set accuracy, the threshold settings are [85%, 90%, 95%], and for the training set loss and test set relevance, the threshold settings are [115%, 110%, 105%], respectively. The smaller the value, the faster the optimization speed during model training.

[0083] In summary, for the i-th training process, this application calculates the following statistical information for the five indicators t, d, l, e, and g respectively:

[0084]

[0085] in, Take values ​​from d, l, e, g, γ=2, 3, 4, 5, Respectively represent the training stage number of each indicator when it reaches the set threshold, and Respectively represent the extreme value, mean value, standard deviation and fluctuation coefficient of each indicator.

[0086] In order to evaluate the stability of the H training processes of the model, the above statistical information calculated in each training process is sorted and normalized, and the normalized result of the training process time t is obtained as follows: Then through the weight The normalized data of the four indicators d, l, e, and g are fused, and the fusion results are The calculation formula is:

[0087]

[0088] Among them, γ=2, 3, 4, 5, representing the four indicators d, l, e, and g respectively. Represents the normalized data vector of the four indicators d, l, e, and g, W 1 Represents the fusion weight vector.

[0089] Finally, the fusion results of the five indicators t, d, l, e, and g in the i-th training process can be obtained And use weights Fusion R i , and obtain the final model stability evaluation index SE.

[0090] SE i =W 2 ·R i (20)

[0091] Among them, W 2 is the fusion weight vector of the stability evaluation index calculation process, R i is the fusion result of the five indicators t, d, l, e, and g. SE i is the calculation result of the stability evaluation index of the ith training process. The value range of SE is [0,1]. The larger the value, the more stable the training process. In summary, SE analyzes the time consumption of the training process, the accuracy of the training set, the loss of the training set, the accuracy of the test set and the correlation of the test set, and comprehensively evaluates the optimization speed, accuracy and fluctuation during model training, objectively and accurately characterizing the stability of the model during training. In a preferred example, the training steps of the model can be optimized according to the evaluation index SE. For example, when the accuracy reflected by the evaluation index SE is low, the training process of the model is re-updated to obtain new parameters to improve the accuracy of the model. This is only an example and is not limited.

[0092] In summary, the present application proposes a blast furnace working condition recognition method that integrates process variables and video images. In view of the problem that process variables and video images are difficult to synchronize in time series due to different sampling frequencies, the present application designs a data alignment strategy based on Gaussian functions, which realizes the alignment of heterogeneous data with inconsistent sampling frequencies in the time dimension by interpolating process variables and averaging video frames. On this basis, the present invention establishes a deep neural network model that integrates multi-source heterogeneous information, uses LSTM and CNN to extract features of process variables and video images respectively, and simplifies information through CCA. In order to more effectively extract multi-source data features, the present application designs a two-layer residual connection module, combines preliminary features with top-level input data after data alignment, and improves the extraction and utilization of important features of multi-source heterogeneous data. In addition, the present application also proposes a model training stability evaluation index, analyzes the time consumption of the training process, the accuracy of the training set, the loss of the training set, the accuracy of the test set, and the correlation of the test set, and comprehensively evaluates the optimization speed, accuracy, and fluctuation of the model. The method proposed in the present invention can accurately identify 8 types of blast furnace operating conditions, solves the problem of incomplete utilization of operating condition information and inaccurate identification of blast furnace operating conditions caused by using only a single information source, and provides on-site workers with scientific and reliable operating condition identification results.

[0093] In one example, a 2650m 3Taking a blast furnace as an example, the blast furnace working condition recognition method provided by the present application that integrates process variables and video images is verified. Specifically, the data samples used include process variables and furnace top material surface videos within the same time period. The data examples are as follows: Figure 3 shown. Figure 3 (a) shows the variation curve of the blast furnace top temperature. The data size of the original process variable is (1,3899), and the size after data alignment is (1,38987). In order to fully capture the data features, the present invention adopts sliding window processing. The data size after sliding window processing is (1,11,38987). Similarly, 35 process variables are selected. After transposing the data set, the data size of X is (38987,11,35). Figure 3 (b) shows the material surface video images of 8 blast furnace working conditions. These images have been image enhanced and then aligned. The data size of the material surface video image Y is (38987, 256, 256). Finally, X and Y are input into the model to conduct experiments on the blast furnace working condition recognition task.

[0094] The index change curve of the model training process and the working condition recognition experimental results (recognition accuracy) are shown as follows: Figure 4 and Figure 5 As shown. It can be seen that the improved blast furnace operating condition identification method of the present invention has fast optimization speed during model training, good information extraction effect, high accuracy, and an average accuracy of 96.99%, and the recognition accuracy of each operating condition is above 95%. At the same time, for the 20 training processes of the model, the average value of the calculation results of the model stability evaluation index is 0.752, which shows that the model has good stability during training. In summary, the experimental results show that the proposed method has achieved a very high accuracy rate in the blast furnace operating condition identification task, can meet the needs of on-site work, and provides reliable decision-making support for on-site workers to monitor the operating status of blast furnaces in real time.

[0095] An embodiment of the present application also provides a blast furnace operating condition identification system that integrates process variables and video images, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0096] The blast furnace operating condition identification system integrating process variables and video images can implement various embodiments of the above-mentioned method and achieve the same beneficial effects, which will not be described in detail here.

[0097] The preferred specific embodiments of the present application are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present application without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art based on the concept of the present application through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for identifying blast furnace operating conditions by integrating process variables and video images, characterized in that: include: S1: Collect process variables of blast furnace working conditions and material surface video images to obtain multi-source heterogeneous information; S2: performing interpolation processing on the process variables and averaging the video image frames to align the multi-source heterogeneous information to the same time step; S3: Build a single-layer fusion network architecture; S4: extracting a preliminary feature set of the aligned multi-source heterogeneous information based on the single-layer fusion network architecture; S5: Based on the preliminary feature set and the pre-built two-layer fusion network architecture, training and optimization are performed to obtain a two-layer fusion network model; S6: Identify the blast furnace operating condition type based on the double-layer fusion network model.

2. The blast furnace operating condition identification method of integrating process variables and video images according to claim 1 is characterized in that: The interpolation processing of the process variable comprises: Let Ψ represent the original set of process variables, including p process variable types and m samples, Ψ c z represents the zth time window sequence of the cth process variable, where c = 1, 2, ..., p, and z = 1, 2, ..., s, and m samples are divided into s time window sequences by presetting the window width length; Interpolate the sample points in each time window sequence in turn, and let n be the c z The number of new samples interpolated between any two original sample points in (t i ,x i ) represents the i-th original sample point, i=1,2,...,q, let point represents the jth interpolated new sample point between the ith original sample point and its next adjacent original sample point, where Δt = 1 and j = 1, ..., n, and n = ΔT-1. Every second, a new sample is added at each second point; Calculate the time window sequence Ψ c z The interpolation Gaussian function is modified to K(t), which satisfies the following relationship: Where σ is the bandwidth of the interpolated Gaussian function K(t), t is time, and μ is the mean of the interpolated Gaussian function K(t); Based on the Gaussian function K(t), the The size of the new sample point to be interpolated satisfies the following relationship: In the formula, x m is the true value of the mth original data point, t m is the time at which the mth original data point is located, and K is the Gaussian function.

3. The blast furnace operating condition identification method of integrating process variables and video images according to claim 1 is characterized in that: Averaging the video image frames comprises: Assume that the video contains num frames per second, calculate the average value of num frames per second at each pixel position, and get an image F representing the time per second. η , the image F η Contains information about all video frame images within that second.

4. The blast furnace operating condition identification method of fusing process variables and video images according to claim 1 is characterized in that: The single-layer fusion network architecture includes a first deep neural network model, a second deep neural network model and an information simplifier, and the input end of the information simplifier is respectively connected to the output end of the first deep neural network model and the output end of the second deep neural network model.

5. The blast furnace operating condition identification method by integrating process variables and video images according to claim 4 is characterized in that: The S4 includes: The process variable X after data alignment is input into the first deep neural network model, and the video image Y after data alignment is input into the second deep neural network model, and the features of X and Y are extracted respectively to obtain two feature sets F1 and F2 of heterogeneous information; For data sets X and Y, calculate the product of the covariance matrix: Among them, Σ XX ,Σ YY and Σ XY are the covariance matrices of X and Y and their cross-covariance matrices, respectively, Y is the product of the covariance matrices, and the singular value decomposition of γ satisfies the following relationship: γ=GΣR T (4) Among them, Γ, Σ and R are the matrix factors after singular value decomposition, J and L are weight parameters, are vector spaces with dimensions l×l, are vector spaces of dimensions m×m, and T represents transpose; The feature sets u1 and v1 obtained by the information simplifier satisfy the following relationship: Among them, parameters J1, L1 and Σ1 are weight parameters calculated by the above-mentioned first deep neural network model.

6. The method for identifying blast furnace operating conditions by integrating process variables and video images according to claim 1, characterized in that: The S5 includes: The feature sets u1 and v1 extracted by the single-layer fusion network architecture are concatenated with the process variable X after data alignment and the video image Y after data alignment to obtain the data sets X′ and Y′, respectively, satisfying the following relationship: in, It is a data splicing operation; Building a new fusion network architecture, wherein the new fusion network architecture has the same structure as the single-layer fusion network architecture but different parameters, and forming a double-layer fusion network architecture based on the new fusion network architecture and the single-layer fusion network architecture; The datasets X′ and Y′ are input into the new fusion network architecture to extract features and obtain two feature sets F3 and F4 of heterogeneous information: Wherein, J2, L2 and Σ2 are weight parameters calculated by the second deep neural network model; Concatenate feature sets u2 and v2 and calculate the blast furnace operating condition recognition results The cross entropy loss function of the following model is used to perform model training and parameter optimization to obtain a two-layer fusion network model: Among them, θ1 and θ2 are the model parameter sets of the single-layer fusion network architecture, θ3 and θ4 are the model parameter sets of the new fusion network architecture, and θ5 is the parameter set of the fully connected layer. is the parameter combination when the network achieves the optimal effect, y i represents the true label of the i-th sample, Represents the recognition result of the i-th sample.

7. The method for identifying blast furnace operating conditions by integrating process variables and video images according to claim 1, characterized in that: The method further comprises: Constructing a model training stability evaluation index, and determining the optimization speed, accuracy and fluctuation of the double-layer fusion network model based on the model training stability evaluation index; Optimize the training model based on the evaluation results.

8. A blast furnace operating condition recognition system integrating process variables and video images, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 7 are implemented.

Images