Metallurgy intelligent factory full-process dynamic optimization simulation and data interaction control method

Abstract

The invention relates to the technical field of metallurgical plant pollutant monitoring and processing, and particularly discloses a metallurgical intelligent plant full-process dynamic optimization simulation and data interaction control method, which comprises the following steps: building a pollutant monitoring module and a carbon footprint tracking module on a management platform, and establishing the association between the pollutant monitoring module and the carbon footprint tracking module; the pollutant monitoring module is used for monitoring the pollutant emission condition of a factory through video images and data acquired by a sensor; the work factory regularly intercepts videos through an installed fixed-point camera, a comparison image and a collected image are uploaded to a pollutant monitoring module for comparison, and the particulate matter emission condition is determined; a smoke monitoring sensor is arranged in the smoke exhaust chimney and is used as a prediction data support; the carbon footprint tracking module performs data processing and analysis according to the acquired information, and completes automatic accounting and data tracing of carbon emission; the real-time prediction of the pollutant concentration is realized, the dust removal control efficiency is improved, the energy consumption is reduced, and the automatic and accurate accounting of the carbon emission is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metallurgical plant pollutant monitoring and processing, and particularly relates to a metallurgical intelligent plant full-process dynamic optimization simulation and data interaction control method. BACKGROUND

[0002] At present, the steel industry has insufficient multi-source heterogeneous data fusion and sharing capability, and the pollutant emission prediction accuracy and over-standard traceability efficiency are insufficient. In the process of intelligent prediction and control of pollutants, the multi-source heterogeneous data of the whole process of industrial production has format barriers and semantic differences, and the data island problem is prominent, which restricts the construction of intelligent models driven by data, and it is necessary to establish a cross-platform data governance framework to realize full-process data penetration and efficient collaborative analysis. Traditional statistical models are difficult to handle the complex nonlinear and time-dependent relationship of pollutant emission data, and manual investigation of over-standard reasons is time-consuming and lacks quantitative means, so it is necessary to build a prediction model and an automatic traceback mechanism to improve the timeliness and accuracy of environmental protection supervision. The intelligent sensing and collaborative control capability of the dust removal system is insufficient, resulting in unstable smoke dust capture effect and difficulty in improving the control of unorganized emissions. Traditional dust removal relies on manual experience control, lacks real-time sensing of dust production scenes and quantitative methods for air volume, resulting in uneven dust collection, high energy consumption, and pipeline safety risks, and it is necessary to integrate machine vision, mechanism model and reinforcement learning technology to realize adaptive control of dust removal equipment and balance of energy saving goals.

[0003] Carbon emission accounting and data traceability are difficult. Enterprise carbon emission data is scattered in EMS, MES, ERP and other heterogeneous systems, and the data format and time granularity are not unified, resulting in difficulty in data integration, and the material flow data between processes have fuzzy boundaries, so it is necessary to build a dynamic correlation model of cross-system heterogeneous data to realize the spatio-temporal alignment of multi-source data (such as matching of EMS energy consumption data and MES process time sequence); domestic and foreign carbon accounting standards are updated frequently, enterprises need to meet multiple accounting rules at the same time, manual adjustment of accounting logic is inefficient, and carbon verification requires data traceability. Therefore, a metallurgical intelligent plant full-process dynamic optimization simulation and data interaction control method is needed to solve the problems of uneven dust collection, high energy consumption and low efficiency of manual adjustment of accounting logic of existing systems. SUMMARY

[0004] In view of the problems in the prior art, the present application aims to provide a metallurgical intelligent plant full-process dynamic optimization simulation and data interaction control method, and research on carbon emission automation and data traceability based on multi-source heterogeneous data collaboration. Specifically, by developing a unified protocol analysis module, the problem of data format diversification caused by the heterogeneity of carbon emission data sources is solved. A built-in configurable accounting rule engine is used to develop a dynamic calculation module and a visual carbon flow diagram to realize dynamic update of emission, simulate future carbon emission trends and optimize emission reduction paths.

[0005] The technical scheme adopted by the present application to solve its technical problems is: a metallurgical intelligent factory full-process dynamic optimization simulation and data interaction control method, comprising the following steps:

[0006] S1, a pollution monitoring module and a carbon footprint tracking module are built on the management platform, and the association between the pollution monitoring module and the carbon footprint tracking module is established.

[0007] S2, the pollution monitoring module monitors the pollution emission of the factory through video image and sensor data collection, including the emission inside the working factory area and the emission of the smoke stack;

[0008] S3, the working factory area regularly intercepts video through the installed fixed-point camera, and the picture from the non-production period video is taken as a comparison image, and the picture from the production period video is taken as a collection image, and the comparison image and the collection image are uploaded to the pollution monitoring module for comparison to determine the particulate matter emission;

[0009] S4, a flue gas monitoring sensor is installed in the smoke stack, the flue gas monitoring sensor collects the emission of pollution gas and particulate matter in the flue gas, and the data collected by the flue gas monitoring sensor is uploaded to the pollution monitoring module, and the pollution emission is predicted by the established CNN-LSTM model, which provides data support for automatic carbon emission accounting and data traceability;

[0010] S5, the pollution monitoring module and the carbon footprint tracking module share data in the same database, the carbon footprint tracking module processes and analyzes the collected information, and completes the automatic carbon emission accounting and data traceability.

[0011] Specifically, the pollution monitoring module and the carbon footprint tracking module in step S1 are associated through the shared database of the management platform.

[0012] Specifically, the video image in step S2 is obtained through multiple angle monitoring cameras distributed in the production factory area, and the monitoring cameras are numbered on the management platform, and the image number corresponding to the monitoring camera is numbered according to the monitoring camera number, then the time number of the month and day is added, and then a random number is added.

[0013] The sensor uses MH-Z40A infrared gas sensor integrated smoke density sensor.

[0014] Specifically, the non-production time and production time in the step S3 are calculated according to the scheduling time on the production scheduling system connected with the management platform, and during the non-production time, the image processing software in the pollutant monitoring module intercepts three 30-second videos at intervals, the image analysis software in the pollutant monitoring module performs screenshot processing on the middle frame of the video, intercepts the contrast images and stores them in the storage, and the image analysis software first performs contrast analysis on the three contrast images, and removes the unclear images and retains one contrast image;

[0015] The acquisition during the production time is performed through the same operation process as the contrast image, and then the acquisition images at different time nodes during the production time are respectively compared and analyzed through image matching with the contrast image of the last non-production time, so as to judge the diffusion of the dust pollutant, and when the similarity between the acquisition image and the contrast image is lower than 80%, a pre-alarm is given to indicate that the dust pollutant in the production area exceeds the standard, the workers perform protective work, the standby dust removal equipment is opened, and the power of the dust removal equipment is increased.

[0016] Specifically, the process of the image analysis software for image matching and contrast analysis is as follows:

[0017] 1) Feature detection: automatic feature detection is adopted, and closed boundaries, edges, contours, line intersection points, corner points and representative points such as barycenter or line end are taken as special;

[0018] 2) Feature matching: the corresponding relationship between the acquisition image and the detected features in the contrast image is established, and different feature descriptors and similarity measures are used to determine the accuracy of registration;

[0019] 3) Evaluation of image transformation model: in order to register the acquisition image and the contrast image, the parameters of the mapping function need to be estimated, and these parameters are calculated using the corresponding features obtained from step 2);

[0020] 4) Image transformation: the mapping is used to perform image transformation on the acquisition image for registration, and the effect is measured by correlation.

[0021] Specifically, the CNN-LSTM model in the step S4 includes:

[0022] 1) Input layer: first, the data is input into the CNN, and the data includes the data collected by the flue gas monitoring sensor and the comparison result data of the contrast image and the acquisition image;

[0023] 2) Convolution layer: the CNN extracts local features of the data through convolution and pooling operations to generate feature maps;

[0024] 3) Flatten layer: the feature maps output by the CNN are flattened into one-dimensional vectors;

[0025] 4) LSTM layer: The flattened vector is input into the LSTM layer. The LSTM layer processes the sequence data through the input gate, forget gate, and output gate mechanism to capture long-term dependencies.

[0026] 5) Output layer: Through the fully connected layer and activation function, the predicted results of pollutant emissions are output.

[0027] Specifically, the convolution formula of the CNN is expressed as a two-dimensional discrete convolution operation, which slides the convolution kernel on the input data and calculates the weighted sum of the local area. The mathematical expression is:

[0028] Z(x,y)=i∑j∑I(x-1,y-1)·K(i,j);

[0029] Among them, I is the input data, K is the convolution kernel, and Z is the output feature map.

[0030] Specifically, the calculation formula of the forget gate of the LSTM is:

[0031] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0032] The calculation formula of the input gate is:

[0033] The calculation formula of the output gate is:

[0034] Specifically, the carbon footprint tracking module in step S5 forms a data set based on the data collected in the database. The data set is divided into four parts: X and Y. X is a plurality of groups of factors most relevant to energy consumption obtained through preliminary screening, and Y is the output of carbon energy consumption to be predicted. The change value dX of X is normalized, and the value of dX is distributed between -1 and 1. The same process is performed on Y. DTW clusters X and uses the cluster label as a feature. The original value of Y is an increasing quantity, which increases like a linear function, while dY is a jumping value. A sparse feature extracts key features through a time-equivalent pulse event sequence, converting discrete data into continuous data. After completing the preliminary preparations, prediction is performed, and GRU and a large language model are selected.

[0035] The present invention has the following beneficial effects:

[0036] The metallurgical intelligent factory full-process dynamic optimization simulation and data interaction control method designed by the application realizes the deep integration and efficient sharing of structured, semi-structured and unstructured data by establishing a unified data standard and semantic mapping framework. The time series data prediction technology based on machine learning and deep learning is proposed, which realizes real-time prediction of pollutant concentration by combining the spatial feature extraction and time series dependence modeling ability of the CNN-LSTM / GRU hybrid model. The particulate matter capture and efficiency improvement control technology based on machine vision and mechanism model combines reinforcement learning and model dynamic calibration to improve dust control efficiency and reduce energy consumption. The environmental protection full-process intelligentization is realized, and the replicable scheme is formed to promote other high energy-consuming industries. The environmental protection efficiency is improved, and the pollutant online monitoring is stable and up to standard. The enterprise environmental protection illegal risk and cost are reduced, and the competitiveness is improved. The automatic and accurate calculation of carbon emissions is realized, which helps enterprises avoid excessive purchase of quotas or sell surplus quotas for profit in carbon trading, while obtaining green financial support. The full-process data real-time notarization and abnormal early warning are realized, which replaces the inefficient manual statistics, and the cross-departmental collaboration efficiency is greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a module flowchart of the metallurgical intelligent factory full-process dynamic optimization simulation and data interaction control method.

[0038] Figure 2 is a workflow diagram of the CNN-LSTM model.

[0039] Figure 3 is a data processing and transformation display diagram in the carbon footprint tracking module. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the application will be further clearly, completely and specifically explained in combination with the drawings in the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0041] As shown in Figure 1 , a metallurgical intelligent factory full-process dynamic optimization simulation and data interaction control method comprises the following steps:

[0042] 1. Build a pollutant monitoring module and a carbon footprint tracking module on the management platform, and establish the association between the pollutant monitoring module and the carbon footprint tracking module.

[0043] 2、Pollutant monitoring module monitors the factory's pollutant emissions through video images and sensor data collection, including the emissions inside the factory and the exhaust chimney. Video images are obtained through multiple angle monitoring cameras distributed inside the production factory, which are numbered on the management platform, such as LT001. When the image number of the monitoring camera is added, the time number is added to the back of the monitoring camera number according to the monitoring camera number, and a random number is added, such as LT0010711001.

[0044] The sensor uses MH-Z40A infrared gas sensor integrated smoke concentration sensor, which can monitor smoke concentration to determine particulate matter emissions, and can monitor gas types including but not limited to carbon monoxide, carbon dioxide, sulfur dioxide, nitrogen oxides.

[0045] 3、The working factory area is intercepted by the installed fixed-point camera at regular intervals, and the picture is intercepted from the video during the non-production period as a comparison image, and the picture is intercepted from the video during the production period as a collection image. The comparison image and the collection image are uploaded to the pollutant monitoring module for comparison to determine the particulate matter emission situation.

[0046] The non-production time and production time are calculated according to the shift time on the production scheduling system connected by the management platform. During the non-production period, the image processing software set in the pollutant monitoring module intercepts three 30-second videos at intervals, and the image analysis software set in the pollutant monitoring module processes the middle frame of the video to intercept comparison images and store them in the memory. The image analysis software first compares and analyzes the three comparison images, and discards the unclear images and keeps one comparison image; the unclear images are deleted directly.

[0047] The collection during the production time is processed by the same operation process as the comparison image, and then the collection images obtained at different time nodes during the production time are respectively compared and analyzed with the comparison image of the last non-production time. Images exceeding one month are deleted to maintain the speed of memory and software processing. The diffusion of dust pollutants is determined, and when the similarity between the collection image and the comparison image is less than 80%, a pre-alarm is given to indicate that the dust pollutants in the production area exceed the standard, and the workers perform protective work, turn on the standby dust removal equipment and increase the power of the dust removal equipment.

[0048] The process of image matching and comparison analysis by the image analysis software is as follows:

[0049] 1) Feature detection: automatic feature detection is used, closed boundary, edge, contour, line intersection, corner and representative point such as barycenter or line end as special.

[0050] 2) Feature matching: Establishes correspondence between features detected in the acquired image and in the contrast image, using different feature descriptors and similarity measures to determine the accuracy of the registration.

[0051] 3) Evaluation of the image transformation model: To register the acquired image with the contrast image, the parameters of the mapping function need to be estimated, using the corresponding features obtained from step 2).

[0052] 4) Image transformation: The acquired image is transformed using the mapping to register, measuring the effectiveness by correlation.

[0053] Correlation is essentially a similarity measure to understand how similar the acquired image and the contrast image are, if the two images are identical, the correlation is equal to 1; if the two images are completely unrelated, the correlation value is equal to 0; if the correlation value is equal to -1, it means that the images are completely inversely related, one is the negative of the other. By using correlation as an evaluation criterion, monomodal registration obtains satisfactory results. By optimizing the similarity criterion between the two images, the transformation parameters are estimated, mainly the translation and rotation of the rigid body. Correlation is mainly limited to single-mode image registration, especially for comparing a series of images, from which small changes caused by changes in dust concentration are found.

[0054] The correlation formula is represented as:

[0055] where x i , y i are the intensities of the i-th pixel of the acquired image and the contrast image, respectively.

[0056] x m , y m are the average intensities of the acquired image and the contrast image.

[0057] 4. A flue gas monitoring sensor is installed in the flue, and the flue gas monitoring sensor collects the emission of pollution gases and particulate matters in the flue gas. The data collected by the flue gas monitoring sensor is uploaded to the pollutant monitoring module, and the emission of pollutants is predicted by the CNN-LSTM model, which provides data support for automatic calculation of carbon emissions and data traceability.

[0058] As shown in Figure 2 , the CNN-LSTM model includes:

[0059] 1) Input layer: First, the data is input into the CNN, including the data collected by the flue gas monitoring sensor and the comparison result data of the contrast image and the acquired image.

[0060] 2) Convolution layer: CNN extracts local features of data through convolution and pooling operations to generate feature maps.

[0061] 3) Flatten layer: flatten the feature map output by CNN into a one-dimensional vector.

[0062] 4) LSTM layer: input the flattened vector into LSTM, which processes sequence data through input gate, forget gate and output gate mechanism, capturing long-time dependencies.

[0063] 5) Output layer: output the prediction result of pollutant emission through fully connected layer and activation function.

[0064] The specific operation steps of CNN-LSTM model building: import the necessary libraries (image contrast of dust particle emission in factory area, collector collected smoke emission in chimney), load and prepare data, define data processing function, data set division and normalization processing, define and train the built CNN-LSTM model, draw training and validation loss curve, model prediction and inverse normalization processing, evaluate prediction performance, draw prediction results, form contrast analysis legend.

[0065] The convolution formula of CNN is represented as two-dimensional discrete convolution operation, which slides the convolution kernel on the input data and calculates the weighted sum of local area, the mathematical expression is:

[0066] Z(x,y) = i∑j∑I(x-1,y-1)·K(i,j);

[0067] Where I is the input data, K is the convolution kernel, and Z is the output feature map.

[0068] The calculation formula of forget gate of LSTM is: t = σ(W f ·[h t-1 ,x t ]+b f ); Where σ is the range function, W f is the weight matrix, b f is the bias term, and [h t-1 ,x t ] represents the concatenation of h t-l and x t .

[0069] The forget gate determines the discarded information, and the level of the forget gate takes the current input x t and the hidden state h t-1 of the last time as input, and outputs a vector f t with a value range between (0, 1), each element in the vector represents the degree of retention or discard of the corresponding data, 0 means completely discard, and 1 means completely retain.

[0070] The calculation formula of input gate is: Among them, ⊙ represents element-by-element multiplication, C t-1 The status at the previous moment.

[0071] The input gate is responsible for deciding what new information to add to the state. Part of this is done by using the range layer to decide which values ​​need to be updated, which is called input gating. t , and the other part is to generate possible new information through the tanh layer.

[0072] The calculation formula of the output gate is:

[0073] The output gate controls the output of the state. The output gate is generated by the range layer. t Determines the degree of output; then, the state is transformed by the tanh layer and multiplied by the output gate to obtain the hidden state at the current moment.

[0074] 5. The pollutant monitoring module and the carbon footprint tracking module share data in the same database. The carbon footprint tracking module processes and analyzes data based on the collected information to complete automatic carbon emission accounting and data traceability.

[0075] The carbon footprint tracking module forms a dataset based on data collected from the database. The dataset is divided into four parts: X and Y. X is a set of factors most relevant to energy consumption, obtained through preliminary screening, and Y is the output of carbon energy consumption to be predicted. The change value dX of X is normalized, and the value of dX is distributed between -1 and 1. The same process is performed on Y. DTW clusters X and uses the cluster label as a feature. The original value of Y is an increasing quantity, which increases like a linear function, while dY is a jumping value. A sparse feature extracts key features through a sequence of time-equivalent pulse events, converting discrete data into continuous data.

[0076] After completing the preliminary preparations, make predictions and select GRU and large language models. Figure 3As shown, the specific operation consists of three steps. First, the time intervals within each hour are extracted. The red line in the upper left graph represents the time intervals where Y remains constant. Second, the hourly time intervals are averaged to produce the graph in the upper right. The averaged value is inversely proportional to the original discrete value. Third, the inverse of the average is directly taken to produce data that fits the original data, as shown in the figure below. The blue figure represents the original discrete data. The yellow figure represents the data after the second averaging operation, and the red figure represents the data after the inverse ratio is taken and aligned. Compared with the blue figure, the red and blue figures not only fit the trend but also convert the discrete data into continuous data. The processed red Y is used as the new output for prediction. The next step is feature selection, selecting the features most relevant to the three selected Y factors (for example, the three energy consumption features: Y1 particulate matter, Y2 carbon dioxide, and Y3 carbon monoxide). One method is correlation analysis, and the other is XGBoost. First, the correlation analysis is performed, using Y2, 1.0 MPa carbon dioxide. By calculating the correlation between each X and Y2, the 40% most relevant features are selected. Specifically, we first calculate a score function between each X and Y, as shown in the figure in the lower left corner. We then perform a statistical analysis, selecting the top 40% of factors, and then performing a final screening. In addition to calculating the correlation between X and Y, we also calculate the correlation between Y and Y. Y also has a strong autocorrelation with itself. Therefore, we also use past Y as input to predict future Y. Using XGBoost, we also screened for the same number of factors. The overlap between the two methods was approximately 50%. Finally, we used the previous year's data for prediction and found a relatively good fit for carbon traceability.

[0077] The present invention forms a pollutant precise prediction and low-carbon coordinated control technology, realizes the online prediction of environmental emission data, and traces the reasons for exceeding the standard in minutes; forms a dust removal intelligent control technology, realizes the improvement of particulate matter capture efficiency and on-demand intelligent regulation of the dust removal system; forms an accounting rule engine based on multi-source heterogeneous data, realizes the enterprise carbon emission accounting, online analysis management, dynamic tracking of carbon flow and precise data tracing, and effectively assists post personnel in formulating practical emission reduction plans.

[0078] Direct Economic Benefits: Reduced Carbon Compliance Costs: This will improve companies' real-time tracking and analysis of production emissions, combined with carbon pricing consulting to assist in the deployment of targeted carbon reduction pathways, effectively improving corporate carbon management and reducing their carbon trading compliance costs. Indirect Economic Benefits: The development of standardized, intelligent environmental management and control technologies will provide technical services to the industry, enhance corporate competitiveness, and potentially help companies achieve "Environmental Performance Grade A" and "Ultra-Low Emissions" ratings.

[0079] Social benefits: Through the construction of this project, the optimization of pollutant emission control is realized, and the regional environmental quality is effectively improved; it also provides a replicable intelligent environmental protection solution for the steel industry, promotes industrial transformation and upgrading through intelligent transformation, sets up a new benchmark for green development in the industry, and through dynamic matching of energy consumption double control indicators, builds a green and low-carbon development pattern in advance, seizes the opportunity in the field of carbon assets and carbon finance, and helps the steel industry to realize sustainable development under the "double carbon" target, and plays an important demonstration and leading role in the aspects of environment, industry, policy and low-carbon development.

[0080] The present application is not limited to the above-mentioned embodiments, and anyone should know that any structural changes made under the inspiration of the present application fall within the scope of protection of the present application.

[0081] The technology, shape and structure parts not described in detail in the present application are well-known technology.

Claims

1. A method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory, characterized in that: The following steps are involved: S1. Build a pollutant monitoring module and a carbon footprint tracking module on the management platform, and establish a link between the pollutant monitoring module and the carbon footprint tracking module; S2, the pollutant monitoring module uses video images and sensor data to monitor the factory's pollutant emissions, including emissions from the work area and exhaust chimneys; S3. The work area uses fixed-point cameras installed to regularly capture video. Screenshots from videos during non-production periods are used as comparison images, and screenshots from videos during production periods are used as collected images. The comparison images and collected images are uploaded to the pollutant monitoring module for comparison and determination of particulate matter emissions. S4. Install a flue gas monitoring sensor in the exhaust chimney to collect comprehensive information on pollutant gases and particulate matter emissions in the flue gas. The data collected by the flue gas monitoring sensor is uploaded to the pollutant monitoring module. The established CNN-LSTM model is used to analyze pollutant emissions and provide predictive data support for automatic carbon emission accounting and data traceability. S5. The pollutant monitoring module and the carbon footprint tracking module share data under the same database. The carbon footprint tracking module processes and analyzes data based on the collected information to complete automatic carbon emission accounting and data traceability.

2. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 1 is characterized in that: The pollutant monitoring module and the carbon footprint tracking module in step S1 are associated via a shared database of the management platform.

3. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 1 is characterized in that: The video images in step S2 are obtained by surveillance cameras distributed at multiple angles within the production plant. The surveillance cameras are numbered on the management platform. The images captured by the surveillance cameras are numbered by adding the month and day to the end of the surveillance camera number and then adding a random number. The sensor uses MH-Z40A infrared gas sensor integrated with smoke concentration sensor.

4. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 1 is characterized in that: The non-production time and production time in step S3 are calculated based on the shift schedule on the production scheduling system connected to the management platform. During the non-production period, three 30-second videos are captured by the image processing software in the pollutant monitoring module. The image analysis software in the pollutant monitoring module captures the middle frame of the video, captures a comparison image and stores it in a memory. The image analysis software first compares and analyzes the three comparison images, excludes unclear images, and retains one comparison image. The acquisition during the production time follows the same operation process as the above-mentioned comparison images. Then, the acquisition images obtained at different time nodes during the production time are matched and compared with the comparison images during the last non-production time to determine the diffusion of dust pollutants. When the similarity between the acquisition image and the comparison image is less than 80%, a pre-alarm is issued to remind that the dust pollutants in the production area exceed the standard. Workers perform protective operations, turn on the standby dust removal equipment, and increase the power of the dust removal equipment.

5. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 4 is characterized in that: The process of image matching and comparative analysis performed by the image analysis software is as follows: 1) Feature detection: Automatic feature detection is used to detect closed boundaries, edges, contours, line intersections, corners, and representative points such as centroids or line ends as special features; 2) Feature matching: establishing the correspondence between the features detected in the acquired image and the compared image, using different feature descriptors and similarity metrics to determine the accuracy of the registration; 3) Evaluation of the image transformation model: To register the acquired image with the comparison image, it is necessary to estimate the parameters of the mapping function. These parameters are calculated using the corresponding features obtained from step 2); 4) Image transformation: Use mapping to perform image transformation on the acquired images for registration, and measure the effect by correlation.

6. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 1 is characterized in that: The CNN-LSTM model in step S4 includes: 1) Input layer: First, input data into CNN, including data collected by the smoke monitoring sensor and the comparison result data between the comparison image and the collected image; 2) Convolutional layer: CNN extracts local features of the data through convolution and pooling operations to generate feature maps; 3) Flattening layer: Flattens the feature map output by CNN into a one-dimensional vector; 4) LSTM layer: The flattened vector is input into the LSTM layer. The LSTM layer processes the sequence data through the input gate, forget gate, and output gate mechanism to capture long-term dependencies. 5) Output layer: Through the fully connected layer and activation function, the predicted results of pollutant emissions are output.

7. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 6 is characterized in that: The convolution formula of the CNN is expressed as a two-dimensional discrete convolution operation, which slides the convolution kernel on the input data and calculates the weighted sum of the local area. The mathematical expression is: Z(x,y)=i∑j∑I(x-1,y-1)·K(i,j); Among them, I is the input data, K is the convolution kernel, and Z is the output feature map.

8. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 6 is characterized in that: The calculation formula of the forget gate of the LSTM is: f t =σ(W f ·[h t-1 ,x t ]+b f ); The calculation formula of the input gate is: The calculation formula of the output gate is:

9. The method for dynamic optimization simulation and data interactive control of the entire process of a metallurgical intelligent factory according to claim 1 is characterized in that: The carbon footprint tracking module in step S5 forms a data set based on the data collected in the database. The data set is divided into four parts: X and Y. X represents multiple groups of factors most relevant to energy consumption obtained through preliminary screening, and Y represents the output of carbon energy consumption to be predicted. The change value dX of X is normalized, and the value of dX is distributed between -1 and 1. The same process is performed on Y. DTW clusters X and uses the cluster label as a feature. The original value of Y is an increasing quantity, which increases like a linear function, while dY is a jumping value. A sparse feature extracts key features through a time-equivalent pulse event sequence, converting discrete data into continuous data. After completing the preliminary preparations, prediction is performed, and GRU and a large language model are selected.

Images