An ultra-low emission intelligent control system for the steel industry based on the Internet of Things

Abstract

The present invention relates to the field of emission control technology, and discloses an ultra-low emission intelligent control system for the steel industry based on the Internet of Things. The control system comprises: a sequence construction module for constructing a raw material sequence, a process sequence and an environmental sequence; a prediction model training module for constructing training samples of the prediction model; an objective function construction module for constructing a desulfurization efficiency model and a denitrification efficiency model as objective functions; a code generation module for randomly generating individual codes through a periodic mapping function; and an adjustment list generation module for outputting the code of the individual with the largest fitness value in an initialized population as an adjustment list through an optimization algorithm. The present invention uses a prediction model to comprehensively analyze raw material parameters, process parameters and environmental parameters to predict sulfur dioxide emissions and nitrogen oxide emissions at multiple moments in the future, and combines the optimization algorithm to realize automatic adjustment of equipment parameters, so as to maximize the amount of pollutant removal, minimize equipment energy consumption and minimize equipment operating costs.

Description

Technical Field

[0001] The present invention relates to the field of emission control technology, and more specifically, to an ultra-low emission intelligent control system for the steel industry based on the Internet of Things. Background Art

[0002] The steel industry is a vital foundational sector of the national economy and one of the most concentrated industrial sectors in terms of energy consumption and air pollution emissions. The steel production process includes key steps such as sintering, ironmaking, steelmaking, and rolling. Each step releases various types of air pollutants during high-temperature smelting, fuel combustion, and material transportation. Sintering, a high-temperature, open, continuous combustion process, involves large amounts of air being drawn through the sinter bed by exhaust fans, making it the most pollutant-intensive step within steel production.

[0003] Currently, pollution control for flaring tail gas primarily relies on treatment equipment such as desulfurization (FGD), denitrification (SCR / SNCR), and dust removal. Pollutant emissions are indirectly controlled by adjusting parameters such as exhaust fan frequency, ammonia injection volume, and desulfurization slurry injection volume. However, in actual production, the adjustment of these parameters often relies on the operator's experience and judgment, leading to the following problems: 1. Parameter adjustments are often based on existing emission fluctuations, resulting in significant lag in adjustment response; 2. Workers' adjustments based on experience are difficult to accurately match current operating conditions (for example, factors such as raw material fluctuations and environmental changes), resulting in inefficient pollutant treatment; 3. While pursuing emission compliance, the operating energy efficiency of equipment such as fans and spray systems is often overlooked, resulting in unnecessary energy consumption. Summary of the Invention

[0004] The present invention provides an ultra-low emission intelligent control system for the steel industry based on the Internet of Things, which solves the technical problems in the above-mentioned background technology.

[0005] The present invention provides an ultra-low emission intelligent control system for the steel industry based on the Internet of Things, comprising:

[0006] A sequence construction module is used to simultaneously collect raw material parameters, process parameters and environmental parameters at preset time intervals within a preset time period, and normalize them to construct raw material sequences, process sequences and environmental sequences respectively;

[0007] The prediction model training module is used to use the raw material sequence, process sequence, and environmental sequence as sample data for the prediction model, and collect sulfur dioxide emissions and nitrogen oxide emissions at N moments in the future as sample labels for the prediction model;

[0008] An objective function construction module is used to construct a desulfurization efficiency model and a denitrification efficiency model as objective functions according to the sulfur dioxide emissions, nitrogen oxide emissions, the first adjustment parameter and the second adjustment parameter at the next N moments;

[0009] A code generation module, which is used to randomly generate codes of individuals in the initialized population that meet the constraints through a periodic mapping function;

[0010] The adjustment list generation module is used to calculate the fitness value of the individual according to the individual code through the objective function until the iterative termination condition of the optimization algorithm is met, and the code of the individual with the largest fitness value in the initialized population is output as the adjustment list;

[0011] The adjustment list is represented by the desulfurization slurry injection amount, ammonia injection amount, exhaust fan frequency, air duct damper opening and bag filter backblowing frequency at the next N moments.

[0012] Furthermore, the iteration termination condition is that the number of iterations reaches a preset maximum number of iterations or the difference between the maximum fitness values ​​of two iterations is less than a preset difference threshold, where the preset maximum number of iterations and the preset difference threshold are both custom parameters; the number of future moments N = future time period / future time interval, where the future time period, future time interval, preset time period, preset time interval and the number of individuals in the initialized population are all custom parameters; the lengths of the raw material sequence, process sequence and environment sequence are the same, all equal to the preset time period divided by the preset time interval.

[0013] Furthermore, the prediction model consists of a first branch, a second branch, a third branch, a feature fusion layer, N first fully connected layers, and N second fully connected layers;

[0014] The first branch is used to perform time series analysis on the raw material sequence to obtain a first vector;

[0015] The first branch is constructed based on a bidirectional gated recurrent unit network model;

[0016] The second branch is used to perform timing analysis on the process sequence to obtain a second vector;

[0017] The second branch is built based on the temporal convolutional network model;

[0018] The third branch is used to perform time series analysis on the environment sequence to obtain the third vector;

[0019] The third branch is built based on the gated recurrent unit network model;

[0020] The feature fusion layer is used to perform feature fusion on the first vector, the second vector and the third vector to obtain a fusion vector;

[0021] Each of the N first fully connected layers inputs the fusion vector and outputs the sulfur dioxide emissions at the next N moments.

[0022] Each of the N second fully connected layers inputs the fusion vector and outputs the nitrogen oxide emissions at the next N moments.

[0023] The dimensions of the first vector, the second vector, the third vector, and the fusion vector are all custom parameters, and the activation functions of the first fully connected layer and the second fully connected layer are both GELU activation functions.

[0024] Furthermore, the feature fusion layer uniformly converts the first vector, the second vector, and the third vector into vectors of the same dimension through matrix operations, then stacks the three into a matrix form, and finally converts them into a fused vector through matrix operations.

[0025] Furthermore, the first adjustment parameters include: desulfurization slurry injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency; the desulfurization efficiency model first constructs a weighted power sum function of the first adjustment parameter at a single moment, and then fits the mapping relationship between it and the sulfur dioxide removal amount through the least squares method. Finally, according to the first adjustment parameters at the next N moments, the sulfur dioxide removal amount at the next N moments is obtained through the fitted desulfurization efficiency model, and then the sulfur dioxide emissions at the next N moments are subtracted respectively and the sum is calculated.

[0026] Furthermore, the second adjustment parameters include: ammonia injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency; the denitrification efficiency model first constructs a weighted power sum function of the second adjustment parameters at a single moment, and then fits the mapping relationship between it and the nitrogen oxide removal amount through the least squares method. Finally, according to the second adjustment parameters at the next N moments, the nitrogen oxide removal amount at the next N moments is obtained through the fitted denitrification efficiency model, and then the nitrogen oxide emissions at the next N moments are subtracted respectively and the sum is calculated.

[0027] Furthermore, the encoding of the individuals in the initialized population is the same as the representation of the adjustment list, and the constraint condition is that the desulfurization slurry injection amount, ammonia injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency at the randomly generated future N moments cannot exceed the corresponding preset upper and lower limits.

[0028] Furthermore, the periodic mapping function periodically generates a disturbance coefficient with a value range between 0 and 1 according to the individual serial number, multiplies the disturbance coefficient with the difference between the upper limit value and the lower limit value preset for each parameter in the adjustment list, and then sums it with the corresponding lower limit value to generate the individual code, where the individual serial number is represented by an ascending serial number with a starting value of 1 and an ending value of G, and G represents the number of individuals in the initialized population.

[0029] Furthermore, the objective function is the weighted sum of the output value of the desulfurization efficiency model, the output value of the denitrification efficiency model, the inverse of the energy consumption value corresponding to each parameter in the adjustment list, and the inverse of the operating cost, where the energy consumption value and operating cost corresponding to each parameter in the adjustment list are custom parameters.

[0030] Furthermore, the optimization algorithm is a sparrow search algorithm.

[0031] The beneficial effects of the present invention are as follows: the present invention uses a prediction model to comprehensively analyze raw material parameters, process parameters and environmental parameters to predict sulfur dioxide emissions and nitrogen oxide emissions at multiple moments in the future, and combines it with an optimization algorithm to realize automatic adjustment of equipment parameters, thereby maximizing pollutant removal, minimizing equipment energy consumption, and minimizing equipment operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a schematic diagram of an ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to the present invention.

[0033] In the figure: sequence construction module 101, prediction model training module 102, objective function construction module 103, code generation module 104, adjustment list generation module 105. DETAILED DESCRIPTION

[0034] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0035] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words 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 object being described changes, the relative positional relationship may also change accordingly.

[0036] like Figure 1 As shown in the figure, an ultra-low emission intelligent control system for the steel industry based on the Internet of Things includes:

[0037] A sequence construction module 101 is used to simultaneously collect raw material parameters, process parameters, and environmental parameters at preset time intervals within a preset time period, and normalize and construct them into a raw material sequence, a process sequence, and an environmental sequence respectively;

[0038] The prediction model training module 102 is used to use the raw material sequence, process sequence, and environment sequence as sample data for the prediction model, and collect sulfur dioxide emissions and nitrogen oxide emissions at N moments in the future as sample labels for the prediction model;

[0039] An objective function construction module 103 is used to construct a desulfurization efficiency model and a denitrification efficiency model as objective functions according to the sulfur dioxide emissions, nitrogen oxide emissions, the first adjustment parameter and the second adjustment parameter at the next N moments;

[0040] A code generation module 104 is used to randomly generate codes of individuals in the initialization population that meet the constraints through a periodic mapping function;

[0041] The adjustment list generation module 105 is used to calculate the fitness value of the individual according to the individual code through the objective function until the iterative termination condition of the optimization algorithm is met, and output the code of the individual with the largest fitness value in the initialized population as the adjustment list;

[0042] The adjustment list is represented by the desulfurization slurry injection amount, ammonia injection amount, exhaust fan frequency, air duct damper opening and bag filter backblowing frequency at the next N moments.

[0043] In one embodiment of the present invention, the iteration termination condition is that the number of iterations reaches a preset maximum number of iterations or the difference between the maximum fitness values ​​of two iteration numbers is less than a preset difference threshold, wherein the preset maximum number of iterations and the preset difference threshold are both custom parameters. Preferably, the preset maximum number of iterations is set to 200, and the preset difference threshold is set to 1.

[0044] In one embodiment of the present invention, the number of future moments N = future time period / future time interval, where the future time period, future time interval, preset time period, preset time interval and the number of individuals in the initialized population are all custom parameters. Preferably, the preset time period and the future time period are both set to 3 hours, and the future time interval and the preset time interval are both set to 10 minutes, then the number of future moments N = 18, and the number of individuals in the initialized population is set to 50.

[0045] In one embodiment of the present invention, the lengths of the raw material sequence, process sequence and environmental sequence are the same, which are all equal to the preset time period divided by the preset time interval, wherein the raw material parameters include: iron content of iron ore powder, sulfur content of iron ore powder, nitrogen content of iron ore powder, solid fuel ratio, raw material particle size, return ore ratio, limestone ratio and feed moisture content, etc.; process parameters include: exhaust fan frequency, sintering temperature, sintering layer thickness, desulfurization slurry injection amount, ammonia injection amount, etc.; environmental parameters include: ambient temperature, rainfall, wind speed, etc.

[0046] It should be noted that the solid fuel ratio indicates the ratio of coke powder or coal powder in the raw materials, which affects the combustion heat intensity; the raw material particle size indicates the average particle size of the raw materials, which indirectly affects the flue gas flow field; the return ore ratio indicates the proportion of sintering return materials in the new ingredients, which affects the material layer structure and composition stability; the limestone ratio provides alkaline substances to participate in the desulfurization reaction; the feed moisture content affects the material adhesion; the desulfurization slurry injection amount determines the desulfurization reactant supply rate, which directly affects the sulfur dioxide removal rate; the ammonia injection amount determines the reduction reaction raw material supply, which directly affects the nitrogen oxide removal rate; in addition, the above parameters can be collected and obtained through relevant instruments and sensors, such as the iron content, sulfur content and nitrogen content of iron ore powder can be obtained through portable X-ray fluorescence analysis, the sintering temperature can be obtained by an infrared thermometer, and the sintering layer thickness can be obtained by a laser displacement sensor, which will not be elaborated here.

[0047] In one embodiment of the present invention, the raw material parameters, process parameters, and environmental parameters are normalized separately by the Min-Max normalization method, or by the Z-score method, thereby eliminating dimensional differences and improving the generalization ability of the prediction model. This will not be elaborated here.

[0048] In one embodiment of the present invention, the prediction model consists of a first branch, a second branch, a third branch, a feature fusion layer, N first fully connected layers, and N second fully connected layers;

[0049] The first branch is used to perform time series analysis on the raw material sequence to obtain a first vector;

[0050] The second branch is used to perform timing analysis on the process sequence to obtain a second vector;

[0051] The third branch is used to perform time series analysis on the environment sequence to obtain the third vector;

[0052] The feature fusion layer is used to perform feature fusion on the first vector, the second vector and the third vector to obtain a fusion vector;

[0053] Each of the N first fully connected layers inputs the fusion vector and outputs the sulfur dioxide emissions at the next N moments.

[0054] Each of the N second fully connected layers inputs the fusion vector and outputs the nitrogen oxide emissions at the next N moments.

[0055] The number of dimensions of the first vector, the second vector, the third vector, and the fused vector are all custom parameters. Preferably, the number of dimensions of the first vector, the second vector, the third vector, and the fused vector are set to 64, 32, 16, and 128, respectively. The activation functions of the first fully connected layer and the second fully connected layer are both GELU activation functions, and can also be ReLU activation functions, which are not described here.

[0056] In one embodiment of the present invention, the first branch is constructed based on the BiGRU (bidirectional gated recurrent unit network) model, the second branch is constructed based on the TCN (temporal convolutional network) model, and the third branch is constructed based on the GRU (gated recurrent unit network) model. They can also be uniformly constructed based on the Transformer (self-attention converter) model.

[0057] In one embodiment of the present invention, the feature fusion layer uniformly converts the first vector, the second vector, and the third vector into vectors of the same number of dimensions through matrix operations, then stacks the three into a matrix form, and finally converts them into a fused vector through matrix operations.

[0058] It should be noted that, for example, if the vectors are uniformly converted to 128 dimensions, then according to the above content, the first vector needs to be multiplied by a 64×128 matrix, the second vector needs to be multiplied by a 32×128 matrix, and the third vector needs to be multiplied by a 16×128 matrix. Then, after stacking the three, a 3×128 matrix is ​​obtained, and finally, the fused vector is converted through matrix operations. The calculation formula is as follows:

[0059] ,in represents the stacked matrix, and represent the first weight parameter and the second weight parameter respectively, and represent the first bias parameter and the second bias parameter respectively, represents a natural constant, express Activation function.

[0060] It should be noted that It needs to be designed as a 1×3 size vector, then Need to synchronize the design to a vector of size 1×3, It needs to be designed into a 128×128 size matrix, then The synchronization design needs to be a vector of size 1×128.

[0061] In one embodiment of the present invention, the first adjustment parameter includes: desulfurization slurry injection amount, exhaust fan frequency, duct baffle opening and bag filter backblowing frequency; the desulfurization efficiency model first constructs a weighted power sum function of the first adjustment parameter at a single moment, and then fits the mapping relationship between it and the sulfur dioxide removal amount through the least squares method. Finally, according to the first adjustment parameter at the next N moments, the sulfur dioxide removal amount at the next N moments is obtained through the fitted desulfurization efficiency model, and then the sulfur dioxide emissions at the next N moments are subtracted respectively and summed up.

[0062] Specifically, the calculation formula of the desulfurization efficiency model is as follows:

[0063] ,in represents the desulfurization efficiency coefficient, 、 、 and They represent the desulfurization slurry injection volume, exhaust fan frequency, air duct damper opening and bag filter back-blowing frequency at the future i-th moment, 、 、 and They represent the first weight coefficient, the second weight coefficient, the third weight coefficient and the fourth weight coefficient of the desulfurization efficiency model respectively, 、 、 and They represent the first power coefficient, second power coefficient, third power coefficient and fourth power coefficient of the desulfurization efficiency model respectively. represents the amount of sulfur dioxide removed at the i-th moment in the future.

[0064] In one embodiment of the present invention, the second adjustment parameter includes: ammonia injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency; the denitrification efficiency model first constructs a weighted power sum function of the second adjustment parameter at a single moment, and then fits the mapping relationship between it and the nitrogen oxide removal amount through the least squares method. Finally, according to the second adjustment parameter at the next N moments, the nitrogen oxide removal amount at the next N moments is obtained through the fitted denitrification efficiency model, and then the nitrogen oxide emissions at the next N moments are subtracted respectively and summed up.

[0065] Specifically, the calculation formula of the denitrification efficiency model is as follows:

[0066] ,in represents the denitrification efficiency coefficient, 、 、 and They represent the ammonia injection amount, exhaust fan frequency, air duct damper opening and bag filter back-blowing frequency at the future i-th moment, 、 、 and represent the first weight coefficient, the second weight coefficient, the third weight coefficient and the fourth weight coefficient of the denitrification efficiency model respectively, 、 、 and They represent the first power coefficient, second power coefficient, third power coefficient and fourth power coefficient of the denitrification efficiency model respectively. represents the amount of nitrogen oxides removed at the i-th moment in the future.

[0067] In one embodiment of the present invention, the encoding of individuals in the initialized population is the same as the representation of the adjustment list, and the constraint condition is that the desulfurization slurry injection amount, ammonia injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency at the randomly generated future N moments cannot exceed the corresponding preset upper and lower limits.

[0068] For example, the upper and lower limits of the desulfurization slurry injection amount are set to 50-300L / min, the upper and lower limits of the ammonia injection amount are set to 20-150L / min, the exhaust fan frequency is set to 30-50Hz, the air duct baffle opening is set to 20-100%, and the bag dust collector backflush frequency is set to 0.5-6.0 times / min. The specific adjustments are made according to the equipment model and will not be elaborated here.

[0069] In one embodiment of the present invention, a periodic mapping function periodically generates a disturbance coefficient with a value range between 0 and 1 based on the serial number of the individual, multiplies the disturbance coefficient with the difference between the upper limit value and the lower limit value preset for each parameter in the adjustment list, and then sums it with the corresponding lower limit value to generate the individual code, where the serial number of the individual is represented by an ascending serial number with a starting value of 1 and an ending value of G, and G represents the number of individuals in the initialized population.

[0070] Specifically, the calculation formula of the periodic mapping function is as follows:

[0071] ,in , represents the perturbation coefficient, Indicates the individual's serial number, Indicates the number of future moments, Represents the modulo function.

[0072] For example, if the number of individuals in the initialized population is G=50 and the number of future moments is N=18, then the disturbance coefficients of the first and second moments of the individual with sequence number 1 are 0.239 and 0.273 respectively. Assuming that the upper and lower limits of the desulfurization slurry injection volume are set to 50~300L / min, the corresponding individual coding values ​​are (300-50)×0.239+50=109.75 and (300-50)×0.273+50=118.25 respectively.

[0073] It should be noted that compared with the encoding of purely randomly initialized individuals, the present invention generates perturbation coefficients through a periodic mapping function and then acts on the preset upper and lower limits of each parameter in the adjustment list, which can effectively avoid the problem of certain areas in the solution space being too dense or sparse, thereby facilitating global search.

[0074] In one embodiment of the present invention, the objective function is the weighted sum of the output value of the desulfurization efficiency model, the output value of the denitrification efficiency model, the inverse of the energy consumption value corresponding to each parameter in the adjustment list, and the inverse of the operating cost, wherein the energy consumption value and operating cost corresponding to each parameter in the adjustment list are custom parameters.

[0075] Specifically, the calculation formula of the objective function is as follows:

[0076] ,in represents the fitness value of the individual, It represents the desulfurization efficiency coefficient output by the desulfurization efficiency model. Represents the denitrification efficiency coefficient output by the denitrification efficiency model, and They represent the energy consumption value and operating cost of the jth parameter at the i-th moment in the future in the adjustment list, 、 、 and They represent the first, second, third and fourth weight coefficients of the objective function respectively, and are all custom parameters.

[0077] In one embodiment of the present invention, the optimization algorithm is a sparrow search algorithm, or a particle swarm optimization algorithm, which will not be described in detail here.

[0078] It should be noted that the intervals and thresholds are set for ease of comparison. The threshold size depends on the amount of sample data and the cardinality set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless numerical calculations. These formulas are derived from software simulations of the most recent real-world conditions using large amounts of data. The preset parameters in these formulas are set by those skilled in the art based on actual conditions.

[0079] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. An ultra-low emission intelligent control system for the steel industry based on the Internet of Things, characterized by: include: A sequence construction module is used to simultaneously collect raw material parameters, process parameters and environmental parameters at preset time intervals within a preset time period, and normalize them to construct raw material sequences, process sequences and environmental sequences respectively; The prediction model training module is used to use the raw material sequence, process sequence, and environmental sequence as sample data for the prediction model, and collect sulfur dioxide emissions and nitrogen oxide emissions at N moments in the future as sample labels for the prediction model; An objective function construction module is used to construct a desulfurization efficiency model and a denitrification efficiency model as objective functions according to the sulfur dioxide emissions, nitrogen oxide emissions, the first adjustment parameter and the second adjustment parameter at the next N moments; A code generation module, which is used to randomly generate codes of individuals in the initialized population that meet the constraints through a periodic mapping function; The adjustment list generation module is used to calculate the fitness value of the individual according to the individual code through the objective function until the iterative termination condition of the optimization algorithm is met, and the code of the individual with the largest fitness value in the initialized population is output as the adjustment list; The adjustment list is represented by the desulfurization slurry injection amount, ammonia injection amount, exhaust fan frequency, air duct damper opening and bag filter back-blowing frequency at the next N moments; The desulfurization efficiency model first constructs a weighted power sum function of the first adjustment parameter at a single moment, then uses the least squares method to fit the mapping relationship between it and the sulfur dioxide removal amount. Finally, based on the first adjustment parameter at the next N moments, the desulfurization efficiency model is used to obtain the sulfur dioxide removal amount at the next N moments after fitting, and then the sulfur dioxide emissions at the next N moments are subtracted and summed. The denitrification efficiency model first constructs a weighted power sum function of the second adjustment parameter at a single moment, then fits the mapping relationship between it and the nitrogen oxide removal amount through the least squares method, and finally obtains the nitrogen oxide removal amount at the next N moments through the fitted denitrification efficiency model based on the second adjustment parameter at the next N moments, and then subtracts the nitrogen oxide emissions at the next N moments and sums them up.

2. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The iteration termination condition is that the number of iterations reaches the preset maximum number of iterations or the difference between the maximum fitness values ​​of two iterations is less than the preset difference threshold, where the preset maximum number of iterations and the preset difference threshold are both custom parameters; the number of future moments N = future time period / future time interval, where the future time period, future time interval, preset time period, preset time interval and the number of individuals in the initialized population are all custom parameters; the lengths of the raw material sequence, process sequence and environment sequence are the same, all equal to the preset time period divided by the preset time interval.

3. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The prediction model consists of the first branch, the second branch, the third branch, the feature fusion layer, N first fully connected layers and N second fully connected layers; The first branch is used to perform time series analysis on the raw material sequence to obtain a first vector; The first branch is constructed based on a bidirectional gated recurrent unit network model; The second branch is used to perform timing analysis on the process sequence to obtain a second vector; The second branch is built based on the temporal convolutional network model; The third branch is used to perform time series analysis on the environment sequence to obtain the third vector; The third branch is built based on the gated recurrent unit network model; The feature fusion layer is used to perform feature fusion on the first vector, the second vector and the third vector to obtain a fusion vector; Each of the N first fully connected layers inputs the fusion vector and outputs the sulfur dioxide emissions at the next N moments. Each of the N second fully connected layers inputs the fusion vector and outputs the nitrogen oxide emissions at the next N moments. The dimensions of the first vector, the second vector, the third vector, and the fusion vector are all custom parameters, and the activation functions of the first fully connected layer and the second fully connected layer are both GELU activation functions.

4. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 3 is characterized in that: The feature fusion layer converts the first, second, and third vectors into vectors of the same dimension through matrix operations, stacks the three into a matrix form, and finally converts them into a fused vector through matrix operations.

5. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The first adjustment parameters include: desulfurization slurry injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency.

6. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The second adjustment parameters include: ammonia injection amount, exhaust fan frequency, air duct baffle opening and bag filter backblowing frequency.

7. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The encoding of individuals in the initialized population is the same as the representation of the adjustment list. The constraints are that the desulfurization slurry injection amount, ammonia injection amount, exhaust fan frequency, air duct baffle opening and bag filter backflushing frequency at the randomly generated N moments in the future cannot exceed the corresponding preset upper and lower limits.

8. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 7 is characterized in that: The periodic mapping function periodically generates a disturbance coefficient ranging from 0 to 1 based on the individual serial number. After multiplying the disturbance coefficient with the difference between the upper and lower limits preset for each parameter in the adjustment list, the sum is added with the corresponding lower limit value to generate the individual code, where the individual serial number is represented by an ascending serial number with a starting value of 1 and an ending value of G, where G represents the number of individuals in the initialized population.

9. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The objective function is the weighted sum of the output value of the desulfurization efficiency model, the output value of the denitrification efficiency model, the inverse of the energy consumption value corresponding to each parameter in the adjustment list, and the inverse of the operating cost, where the energy consumption value and operating cost corresponding to each parameter in the adjustment list are custom parameters.

10. The ultra-low emission intelligent control system for the steel industry based on the Internet of Things according to claim 1 is characterized in that: The optimization algorithm is the sparrow search algorithm.

Images