Metallurgy waste gas purification system prediction and optimization method based on artificial intelligence

Abstract

The invention discloses a metallurgical waste gas purification system prediction and optimization method based on artificial intelligence. The method comprises the steps that waste gas component concentration, temperature, pressure, flow and reaction time data in the operation process of a waste gas purification system are collected in real time, and an initial multi-dimensional data sample set is constructed; performing adaptive preprocessing on the initial multi-dimensional data sample set to obtain a standardized input data set; inputting the standardized input data set into a neural operator model for real-time dynamic modeling to obtain state real-time prediction features; performing real-time collaborative optimization by using an improved FOX optimization algorithm based on the prediction features to obtain preliminary optimization control parameters; and implementing dynamic iterative optimization of a self-feedback enhanced optimization strategy to obtain an optimal control parameter set, and feeding back the optimal control parameter set in real time and performing closed-loop regulation. Efficient and accurate optimization control of the waste gas purification system is achieved, the waste gas purification efficiency and stability are remarkably improved, and the system dynamically adapts to complex production working conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of cross-integration of metallurgical waste gas purification and artificial intelligence, and in particular to a prediction and optimization method for a metallurgical waste gas purification system based on artificial intelligence. Background Art

[0002] With rapid industrial development and increasingly stringent environmental standards, the issue of waste gas pollution from the metallurgical industry has gradually become a focus of public attention. Metallurgical waste gas typically contains high concentrations of harmful gases such as sulfur dioxide, carbon monoxide, and nitrogen oxides. These substances not only pollute the environment but also pose a serious threat to human health. Therefore, efficient and stable waste gas purification technology has become a key research direction in this field.

[0003] Currently, optimization control technologies for metallurgical waste gas purification systems primarily utilize traditional feedback control methods and empirical modeling. Feedback control methods typically use PID controllers or classical control theory for parameter adjustment, while empirical modeling methods rely on manually designed rules and empirical data to adjust and control system parameters. These methods are relatively straightforward and easy to implement, and have been widely used within the industry with some success. However, with the increasing diversity and complexity of production conditions, particularly when metallurgical processes experience large fluctuations or process changes, traditional feedback control methods, due to their fixed parameters and hysteresis, struggle to effectively respond to and accurately control dynamic system changes in real time. This often leads to low system efficiency and even the risk of loss of control. Furthermore, empirical modeling methods rely heavily on historical data and manual experience, lacking versatility and adaptability. They struggle to effectively predict complex and changing waste gas treatment environments, suffer from significant hysteresis and large errors, and thus struggle to meet the current demand for precise control of waste gas purification.

[0004] In recent years, with the advancement of artificial intelligence (AI), particularly deep learning, various intelligent control strategies based on machine learning models and optimization algorithms have been gradually introduced into metallurgical waste gas purification systems, aiming to improve the system's dynamic response capabilities and precise control performance. Traditional neural network models (such as BP networks and CNN networks) and classical optimization algorithms (such as genetic algorithms and particle swarm optimization) have been initially applied in waste gas treatment. These solutions typically construct neural network models to predict waste gas treatment status and then combine them with intelligent optimization algorithms to adjust system control parameters. However, traditional neural network models suffer from poor generalization performance when predicting complex nonlinear systems and are susceptible to significant deviations due to disturbances in the actual production environment. Furthermore, classical optimization algorithms are prone to falling into local optimal solutions, resulting in low optimization efficiency and slow convergence. These algorithms struggle to meet the stringent real-time response requirements of industrial environments, making it difficult to achieve optimal waste gas treatment results.

[0005] In response to the shortcomings of the above-mentioned existing technologies, the neural operator technology that has emerged in recent years has gradually become an effective way to solve the problem of modeling and prediction of complex dynamic systems. Neural operators can efficiently simulate and predict nonlinear systems with multi-scale dynamic changes, have good generalization capabilities and real-time response characteristics, and have demonstrated significant performance advantages in fields such as fluid mechanics and power systems. In addition, new optimization algorithms such as the FOX optimization algorithm have also begun to attract widespread attention from industry and academia due to their good global search capabilities and efficient convergence characteristics. However, the current research on neural operators and FOX algorithms in the field of metallurgical waste gas purification systems is still in its infancy. How to effectively combine these cutting-edge artificial intelligence technologies to accurately predict the status of the waste gas system and achieve efficient real-time optimization and control of purification parameters still lacks a systematic technical solution.

[0006] Therefore, how to provide an artificial intelligence-based prediction and optimization method for metallurgical waste gas purification systems is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0007] One purpose of the present invention is to propose an artificial intelligence-based prediction and optimization method for a metallurgical waste gas purification system. The present invention adopts a technical route that synergizes neural operator technology with an improved FOX optimization algorithm. By collecting data such as waste gas component concentration, temperature, pressure, flow rate, and reaction time during the operation of the metallurgical waste gas purification system in real time, an initial multidimensional data sample set is constructed and adaptively preprocessed. A neural operator model is used to perform multi-scale dynamic modeling and prediction on the preprocessed data set in real time. The improved FOX optimization algorithm is used to perform multi-parameter collaborative optimization in real time to obtain preliminary optimized control parameters. Then, by designing and implementing a self-feedback reinforcement optimization strategy, the preliminary optimized control parameters are dynamically iteratively optimized in real time to obtain the optimal control parameter set, ultimately achieving closed-loop precise control of the waste gas purification system. The technical solution of the present invention effectively solves the prominent problems of insufficient waste gas prediction accuracy and difficulty in dynamically responding to complex production conditions in the prior art, as well as the traditional optimization algorithm's tendency to fall into local optimality and slow convergence. It has the significant advantages of high prediction accuracy, strong generalization performance, good real-time parameter optimization, fast convergence speed, and stable and reliable overall system control effect.

[0008] According to an embodiment of the present invention, a prediction and optimization method for a metallurgical waste gas purification system based on a neural operator combined with an improved FOX optimization algorithm includes the following steps: S1. Real-time collection of waste gas component concentration, temperature, pressure, flow rate and reaction time data during the operation of the metallurgical waste gas purification system to establish an initial multi-dimensional data sample set; S2. performing adaptive preprocessing on the initial multidimensional data sample set, including data anomaly removal, data denoising, data fusion and multi-scale feature extraction, to obtain a standardized input data set; S3. Inputting the standardized input data set into a neural operator model to perform real-time multi-scale dynamic modeling to obtain real-time prediction characteristics of the metallurgical waste gas system state; S4. Using the improved FOX optimization algorithm, taking the real-time prediction features as input, and combining it with an adaptive dynamic search strategy to perform multi-parameter collaborative optimization in real time, to obtain preliminary optimized control parameters of the exhaust gas purification system; S5. Implementing a self-feedback reinforcement optimization strategy on the preliminary optimized control parameters, performing real-time dynamic iterative optimization, and obtaining an optimal control parameter set; S6. Feedback the optimal control parameter set to the metallurgical waste gas purification system in real time, and control the metallurgical waste gas purification system to perform closed-loop regulation according to the optimal control parameter set.

[0009] Optionally, the S1 specifically includes: S11. Install a gas composition sensor at the exhaust gas inlet of the metallurgical exhaust gas purification system to collect concentration data of sulfur dioxide, carbon monoxide, and nitrogen oxides in real time, with a sampling frequency of 1 to 5 seconds; S12. Install high-precision temperature sensors at the exhaust gas inlet and reaction chamber of the metallurgical exhaust gas purification system to collect real-time data on the inlet gas temperature and the temperature inside the reaction chamber. The temperature sampling accuracy is ±0.1°C. S13. Install a pressure sensor inside the reaction chamber of the metallurgical waste gas purification system to collect real-time pressure data in the chamber. The pressure measurement range is 0.1~2.0MPa; S14. Install a flow meter at the gas inlet pipe of the metallurgical waste gas purification system to collect waste gas flow data in real time. The flow measurement range is 100~5000m³ / h; S15. Timestamp the start and end times of each purification process of the metallurgical waste gas purification system and record the reaction time data. The accuracy of the reaction time is ±0.5s. S16. Using the timestamp of the start of the reaction process as a reference, the real-time collected exhaust gas component concentration, temperature, pressure, and flow rate data are time-aligned according to the timestamp, and matched one by one to the corresponding reaction time period to form a multidimensional data sequence under a unified time reference; S17. Based on the multidimensional data sequence, data splicing and fusion are performed with timestamps as associated indexes to construct an initial multidimensional data sample set with unified timestamps and clear internal data structure.

[0010] Optionally, the S2 specifically includes: S21, taking the mean of each data sequence in the initial multidimensional data sample set as the benchmark, adopt the The abnormal data are identified and eliminated based on the criteria, is the standard deviation of each data series; S22. After the abnormal data are removed, the data sequence is subjected to data denoising using a wavelet threshold denoising method, wherein the wavelet basis function is Daubechies wavelet and the decomposition scale is set to 3 to 5 layers; S23, applying normalization processing to the denoised multidimensional data sequence, with the normalization processing range of the data being [0, 1]; S24. Perform data fusion on the normalized data series using principal component analysis, with the cumulative contribution rate of the principal component set at 90% to 95%; S25. Perform multi-scale feature extraction on the sequence after data fusion, and use the empirical mode decomposition method to decompose the fused data sequence into intrinsic mode functions step by step, with the decomposition order of each data sequence set to 4 to 6; S26. Concatenate the eigenmode functions of each order in descending order of scale to form a standardized input data set.

[0011] Optionally, the neural operator model specifically includes: The input mapping layer uses a fully connected neural network to map the normalized input data set to a high-dimensional feature space with a dimension of 128 to 256; The Fourier operator layer uses fast Fourier transform to perform frequency domain transformation on the mapped high-dimensional feature space data; The multi-scale frequency domain filter layer extracts multi-scale frequency domain features of low frequency 0sim0.1Hz, medium-low frequency 0.1~0.5Hz, medium frequency 0.5~1Hz, medium-high frequency 1~2Hz, and high frequency 2~5Hz respectively through frequency domain filters with a scale of 4~6 in the frequency domain space; The inverse Fourier transform layer performs inverse fast Fourier transform on the multi-scale frequency domain features after frequency domain filtering, maps the frequency domain features back to the time domain, and obtains the corresponding multi-scale time domain features; The feature fusion layer combines multi-scale temporal features and fuses them through a fully connected network. The output dimension after feature fusion is 64-128. The state prediction output layer outputs the real-time prediction features of the metallurgical waste gas purification system state based on the fused features through a fully connected network.

[0012] Optionally, the S3 specifically includes: S31. Use a fully connected neural network to map the standardized input data set to a high-dimensional feature space with a dimension of 128 to 256 to obtain high-dimensional feature data; S32. Perform fast Fourier transform on high-dimensional feature data: ; Where X(k) represents the frequency domain feature data corresponding to the kth frequency component, x(n) represents the data value of the nth time domain sampling point in the high-dimensional feature data, N represents the total length of the data sequence, k is the frequency index, ranging from 0, 1, 2, ..., N-1, n is the time domain index, ranging from 0, 1, 2, ..., N-1, and j represents the imaginary unit; S33, applying a frequency domain filter with a scale number of 4 to 6 to the frequency domain feature data to extract multi-scale frequency domain features with frequency ranges of 0 to 0.1 Hz, 0.1 to 0.5 Hz, 0.5 to 1 Hz, 1 to 2 Hz, and 2 to 5 Hz; S34. Perform inverse fast Fourier transform on the multi-scale frequency domain features: ; Where x(n) represents the time domain feature data value corresponding to the nth time domain sampling point, X(k) represents the data value of the kth frequency component in the frequency domain feature data, N represents the total length of the data sequence, n is the time domain index, ranging from 0, 1, 2, ..., N-1, k is the frequency index, ranging from 0, 1, 2, ..., N-1, and j represents the imaginary unit; S35, the obtained multi-scale time domain features are spliced in order of scale size and input into the fully connected network for fusion processing. The feature dimension after fusion is 64~128; S36. The fused features are output through a fully connected network to realize real-time prediction of the status of the metallurgical waste gas purification system.

[0013] Optionally, the improved FOX optimization algorithm specifically includes: In the algorithm initialization step, the population size is set to 20-50, and each individual corresponds to a set of exhaust gas purification control parameters, including a temperature range of 300-800°C, a pressure range of 0.1-2.0 MPa, a flow range of 100-5000 m³ / h, and a reaction time range of 10-120 s. Adaptive dynamic search step, by calculating the population fitness value, taking the comprehensive indicators of exhaust gas purification efficiency and economic cost as the target, and dynamically adjusting the search range and step size according to the current population fitness variance; The Lévy flight search step performs a random walk search, and the random flight step size follows the Lévy distribution with a parameter range of 0.5~1.5; Gaussian perturbation local search step, in which Gaussian perturbations with a standard deviation of 10% to 20% of the current parameter value are applied to the individual; In the global and local search switching steps, the search mode is switched through a probability mechanism. The initial switching probability is set to 0.3~0.5 and decreases linearly with the number of iterations. Elite retention and iterative update steps: after each iteration, the top 10% to 20% of individuals in fitness are retained and the population is updated to the next generation; Convergence judgment step: when the global optimal individual fitness value changes by less than 10 in 20 to 50 consecutive iterations, -5 When , the optimal control parameters are output.

[0014] Optionally, the S4 specifically includes: S41. Using the real-time prediction features as input, establish an initial population with a population size of 20 to 50 individuals. Each individual contains control parameters for temperature, pressure, flow rate, and reaction time. The temperature range is 300 to 800°C, the pressure range is 0.1 to 2.0 MPa, the flow rate range is 100 to 5000 m³ / h, and the reaction time range is 10 to 120 s. S42. Calculate the fitness value of each individual in the initial population, where the fitness value is defined as a comprehensive evaluation index of exhaust gas purification efficiency and economic cost; S43, adaptively adjusting the search range and step size based on the calculated population fitness value and its variance, with the search range adjustment ratio being 10% to 30% of the initial range, and the step size adjustment ratio being 5% of the initial step size; S44, using the Lévy flight random search strategy to update the position of the individual in the population, the Lévy flight step size obeys the Lévy distribution, and the characteristic parameter α of the Lévy distribution ranges from 0.5 to 1.5; S45. Perform a Gaussian perturbation local search on the individual parameters of the population obtained by the search, with the perturbation standard deviation being 10% to 20% of the individual's current parameter value; S46. Set the initial value of the switching probability between global search and local search to 0.3~0.5, and decrease it linearly with the number of iterations. Update the population and implement the elite retention strategy to retain the top 10 individuals in fitness after each iteration. S47, in the process of 20 to 50 consecutive iterations, the fitness value of the global optimal individual changes less than the convergence threshold 10 -5 The iteration is terminated when , and the preliminary optimized control parameters of the exhaust gas purification system are obtained.

[0015] Optionally, the self-feedback reinforcement optimization strategy specifically includes: Initial optimization control parameter acquisition: using the exhaust system state characteristics predicted in real time by the neural operator model as input, the improved FOX optimization algorithm is used to obtain the initial optimization control parameters; Real-time status data collection: input the preliminary optimized control parameters into the metallurgical waste gas purification system, and collect the actual status data of the metallurgical waste gas purification system in real time; State error calculation, calculation error function: ; Among them, E(t) represents the total state error at time t, Indicates the value of the ith actual measured state parameter of the metallurgical waste gas purification system, represents the value of the i-th predicted state parameter, represents the weight coefficient of the i-th state parameter, and n represents the total number of state parameters; Adaptive feedback gain calculation, dynamic calculation of feedback gain coefficient: ; in, represents the dynamic feedback gain coefficient, Indicates the initial feedback gain coefficient, alpha indicates the gain adjustment factor, and its value range is 0.1~0.5. Indicates the allowed error threshold; Control parameter correction: real-time correction of control parameters based on dynamic feedback gain coefficient: ; Among them, P(t+1) represents the optimized control parameters at the next moment, and P(t) represents the optimized control parameters at the current moment. Represents the gradient of the error function relative to the current parameter, defined as: ; in, represents the value of the i-th control parameter at the current moment, and m represents the total number of control parameters; The iterative optimization is terminated when the change of the global optimal individual fitness value in 20 to 50 consecutive iterations is less than the set convergence error threshold of 10 -4 ~10 -6 When , the iterative optimization is stopped and the final optimized control parameters are obtained.

[0016] Optionally, the S5 specifically includes: S51, inputting the preliminary optimized control parameters into the self-feedback reinforcement optimization strategy; S52, real-time collection of actual operating status data of the metallurgical waste gas purification system; S53, calculating a state error value according to the real-time state data; S54, dynamically updating the optimized control parameters according to the state error value; S55, performing multiple dynamic iterative optimizations on the updated optimized control parameters; S56, when the change range of the optimization control parameter during 20 to 50 consecutive iterations is less than the set convergence error threshold of 10 -4 ~10 -6 When , the optimal control parameter set is output.

[0017] Optionally, the S6 specifically includes: S61, transmitting the obtained optimal control parameter set to the control terminal of the metallurgical waste gas purification system in real time through the communication interface; S62, the control terminal adjusts the temperature of the reaction chamber in the metallurgical waste gas purification system in real time based on the centralized temperature parameters of the optimal control parameters, with a temperature control accuracy of ±0.5°C; S63, based on the optimal control parameter centralized pressure parameter, real-time control of the pressure in the reaction chamber, with a pressure control accuracy of ±0.01 MPa; S64, based on the optimal control parameter centralized flow parameters, adjust the exhaust gas inlet flow in real time, with a flow control accuracy of ±10m³ / h; S65, based on the optimal control parameter and the reaction time parameter, the exhaust gas residence time in the reaction chamber is controlled in real time, and the reaction time control accuracy is ±1s; S66: Real-time collection of exhaust gas state parameters after feedback control, and periodic transmission of the collected feedback data back to the control terminal, with a data transmission cycle of 1 to 2 seconds; S67. The control terminal performs a real-time comparison between the feedback state parameter and the optimal control parameter set. When the deviation of the feedback state parameter exceeds a preset threshold range, the optimization strategy is re-executed to update the optimal control parameter set.

[0018] The beneficial effects of the present invention are: (1) The present invention uses a neural operator model to perform real-time multi-scale dynamic prediction of the exhaust gas system state, which can achieve high-precision real-time prediction of exhaust gas component concentration, temperature, pressure, flow rate and reaction time and other state parameters, effectively improving the exhaust gas system state prediction accuracy and response speed, and significantly improving the adaptability and stability of the metallurgical exhaust gas purification system in complex production conditions and sudden changes in the environment.

[0019] (2) By designing and introducing an improved FOX optimization algorithm, the present invention can achieve rapid and accurate optimization of the optimization control parameters of the exhaust gas purification system, significantly improving the optimization efficiency and global optimization capability of the exhaust gas purification system, and showing better real-time responsiveness and stability in complex and dynamically changing metallurgical production conditions.

[0020] (3) In terms of real-time optimization and dynamic correction of the control parameters of the exhaust gas purification system, the present invention implements a self-feedback enhanced optimization strategy, effectively solving the problem of the existing technology that prediction and optimization are separated from each other and cannot achieve real-time linkage control. It breaks through the bottleneck of traditional exhaust gas purification control technology parameter optimization lag, slow convergence speed and easy to fall into local optimality, and realizes rapid dynamic iterative optimization of the control parameters of the exhaust gas purification system, thereby effectively improving the technical level and actual engineering application capabilities in the field of metallurgical exhaust gas purification. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is the overall technical flow chart of the artificial intelligence-based prediction and optimization method for metallurgical waste gas purification systems proposed by the present invention; Figure 2 This is a schematic diagram of the neural operator model structure of the artificial intelligence-based prediction and optimization method for metallurgical waste gas purification system proposed in the present invention; Figure 3 This is a flow chart of the improved FOX optimization algorithm for the prediction and optimization method of metallurgical waste gas purification system based on artificial intelligence proposed by the present invention. DETAILED DESCRIPTION

[0022] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0023] refer to Figure 1-Figure 3 A prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm includes the following steps: S1. Real-time collection of waste gas component concentration, temperature, pressure, flow rate and reaction time data during the operation of the metallurgical waste gas purification system to establish an initial multi-dimensional data sample set; S2. performing adaptive preprocessing on the initial multidimensional data sample set, including data anomaly removal, data denoising, data fusion and multi-scale feature extraction, to obtain a standardized input data set; S3. Inputting the standardized input data set into a neural operator model to perform real-time multi-scale dynamic modeling to obtain real-time prediction characteristics of the metallurgical waste gas system state; S4. Using the improved FOX optimization algorithm, taking the real-time prediction features as input, and combining it with an adaptive dynamic search strategy to perform multi-parameter collaborative optimization in real time, to obtain preliminary optimized control parameters of the exhaust gas purification system; S5. Implementing a self-feedback reinforcement optimization strategy on the preliminary optimized control parameters, performing real-time dynamic iterative optimization, and obtaining an optimal control parameter set; S6. Feedback the optimal control parameter set to the metallurgical waste gas purification system in real time, and control the metallurgical waste gas purification system to perform closed-loop regulation according to the optimal control parameter set.

[0024] The specific implementation method of real-time collection of waste gas component concentration, temperature, pressure, flow and reaction time data in the metallurgical waste gas purification system of the present invention is as follows: a gas composition sensor is installed at the waste gas inlet end, and the sampling frequency of the gas composition sensor is set to once per second; temperature data is collected by a high-precision thermocouple temperature sensor, and the installation position is set at the waste gas inlet end and inside the reaction chamber, and the temperature sensor sampling accuracy is ±0.1°C; pressure data is collected in real time by a pressure sensor installed inside the reaction chamber, and the measurement range is set to 0.1MPa to 2.0MPa; waste gas flow data is collected in real time by a flow meter installed in the gas inlet pipe, and the measurement range is set to 100m³ / h to 5000m³ / h; reaction time data is obtained by recording timestamps at the start and end of each waste gas purification process, and the timestamp recording accuracy is set to ±0.5 seconds. The data collected by the above-mentioned sensors are based on the timestamp of the reaction start time, and after strict time axis alignment processing, an initial multidimensional data sample set is formed, and further processed through 3 Criteria, wavelet threshold denoising, principal component analysis and empirical mode decomposition methods are used to realize data preprocessing and multi-scale feature extraction, and finally a standardized input data set is obtained.

[0025] The present invention proposes a prediction and optimization method for a metallurgical waste gas purification system based on a neural operator combined with an improved FOX optimization algorithm. Through real-time and accurate prediction of the waste gas system state, dynamic collaborative parameter optimization and closed-loop feedback control, the system parameters are always in a dynamic optimization state, thereby significantly improving the waste gas purification efficiency and the stability of the system operation, and reducing the overall operating cost. Compared with the existing technical solutions, it exhibits higher prediction accuracy and stronger adaptability under complex working conditions, and has obvious technical advantages.

[0026] In this embodiment, S1 specifically includes: S11. Install a gas composition sensor at the exhaust gas inlet of the metallurgical exhaust gas purification system to collect concentration data of sulfur dioxide, carbon monoxide, and nitrogen oxides in real time, with a sampling frequency of 1 to 5 seconds; S12. Install high-precision temperature sensors at the exhaust gas inlet and reaction chamber of the metallurgical exhaust gas purification system to collect real-time data on the inlet gas temperature and the reaction chamber temperature. The temperature sampling accuracy is ±0.1°C. S13. Install a pressure sensor inside the reaction chamber of the metallurgical waste gas purification system to collect real-time pressure data in the chamber. The pressure measurement range is 0.1~2.0MPa; S14. Install a flow meter at the gas inlet pipe of the metallurgical waste gas purification system to collect waste gas flow data in real time. The flow measurement range is 100~5000m³ / h; S15. Timestamp the start and end times of each purification process of the metallurgical waste gas purification system and record the reaction time data. The accuracy of the reaction time is ±0.5s. S16. Using the timestamp of the start of the reaction process as a reference, the real-time collected exhaust gas component concentration, temperature, pressure, and flow rate data are time-aligned according to the timestamp, and matched one by one to the corresponding reaction time period to form a multidimensional data sequence under a unified time reference; S17. Based on the multidimensional data sequence, data splicing and fusion are performed with timestamps as associated indexes to construct an initial multidimensional data sample set with unified timestamps and clear internal data structure.

[0027] During the implementation of the present invention, the real-time collection of the exhaust gas component concentration is specifically achieved through a gas composition sensor installed at the exhaust gas inlet end of the metallurgical exhaust gas purification system. The sensor type is an electrochemical sensor, wherein the measurement range of sulfur dioxide is set to 0 to 2000 ppm, the measurement range of carbon monoxide is set to 0 to 1000 ppm, and the measurement range of nitrogen oxides is set to 0 to 500 ppm; the high-precision temperature sensor used is a K-type thermocouple sensor with a temperature measurement range of 0 to 1200°C; the pressure sensor is a piezoresistive pressure sensor with an accuracy level of 0.1; the flow meter adopts a thermal mass flow meter with a flow measurement accuracy of ±1% of the full scale; the timestamp marking is achieved by the system's own high-precision synchronous clock with a clock accuracy of milliseconds; the collected multidimensional data is transmitted in real time to the central data processing unit via industrial Ethernet for unified time axis alignment processing and multidimensional data fusion.

[0028] The present invention ensures the basic data quality of subsequent data processing and optimization work by accurately collecting and constructing the initial multi-dimensional data sample set in the waste gas treatment process in real time, significantly improves the accuracy and stability of waste gas treatment status prediction and optimization, and provides reliable data support for the real-time dynamic optimization control of metallurgical waste gas purification systems. Compared with the existing technology, it significantly improves the accuracy and response efficiency of system operation.

[0029] In this embodiment, S2 specifically includes: S21. Taking the mean of each data sequence in the initial multidimensional data sample set as the benchmark, identify and eliminate abnormal data based on the 3σ criterion, where σ is the standard deviation of each data sequence; S22. After the abnormal data are removed, the data sequence is subjected to data denoising using a wavelet threshold denoising method, wherein the wavelet basis function is Daubechies wavelet and the decomposition scale is set to 3 to 5 layers; S23, applying normalization processing to the denoised multidimensional data sequence, with the normalization processing range of the data being [0, 1]; S24. Perform data fusion on the normalized data series using principal component analysis, with the cumulative contribution rate of the principal component set at 90% to 95%; S25. Perform multi-scale feature extraction on the sequence after data fusion, and use the empirical mode decomposition method to decompose the fused data sequence into intrinsic mode functions step by step, with the decomposition order of each data sequence set to 4 to 6; S26. Concatenate the eigenmode functions of each order in descending order of scale to form a standardized input data set.

[0030] The specific method for identifying and eliminating abnormal data during the implementation of the present invention is as follows: after calculating the mean and standard deviation of the initial data sequence one by one, abnormal data is determined based on the difference between each data point in the data sequence and the corresponding sequence mean. Any data point with a difference greater than 3 times the standard deviation is identified as abnormal data and directly eliminated; the wavelet threshold denoising process adopts a soft threshold denoising method, and the threshold size is calculated by multiplying the median of the absolute value of the decomposition coefficient of each layer by the logarithmic function; the data fusion of the principal component analysis adopts the covariance matrix calculation method, and the eigenvalue and the corresponding eigenvector are solved by the covariance matrix, and the principal components with a cumulative contribution rate of more than 90% are selected for fusion.

[0031] The present invention effectively removes abnormal noise in the data and extracts key features with high discrimination by comprehensively applying the 3σ criterion, wavelet threshold denoising and principal component analysis method, significantly improving the accuracy and robustness of the multi-scale dynamic modeling of the data input neural operator model, thereby effectively improving the reliability and stability of the real-time prediction of the status of the metallurgical waste gas purification system.

[0032] The neural operator model specifically includes: The input mapping layer uses a fully connected neural network to map the normalized input data set to a high-dimensional feature space with a dimension of 128 to 256; The Fourier operator layer uses fast Fourier transform to perform frequency domain transformation on the mapped high-dimensional feature space data; The multi-scale frequency domain filter layer extracts multi-scale frequency domain features of low frequency 0sim0.1Hz, medium-low frequency 0.1~0.5Hz, medium frequency 0.5~1Hz, medium-high frequency 1~2Hz, and high frequency 2~5Hz respectively through frequency domain filters with a scale of 4~6 in the frequency domain space; The inverse Fourier transform layer performs inverse fast Fourier transform on the multi-scale frequency domain features after frequency domain filtering, maps the frequency domain features back to the time domain, and obtains the corresponding multi-scale time domain features; The feature fusion layer combines multi-scale temporal features and fuses them through a fully connected network. The output dimension after feature fusion is 64-128. The state prediction output layer outputs the real-time prediction features of the metallurgical waste gas purification system state based on the fused features through a fully connected network.

[0033] In this embodiment, S3 specifically includes: S31. Use a fully connected neural network to map the standardized input data set to a high-dimensional feature space with a dimension of 128 to 256 to obtain high-dimensional feature data; S32. Perform fast Fourier transform on high-dimensional feature data: ; Where X(k) represents the frequency domain feature data corresponding to the kth frequency component, x(n) represents the data value of the nth time domain sampling point in the high-dimensional feature data, N represents the total length of the data sequence, k is the frequency index, ranging from 0, 1, 2, ..., N-1, n is the time domain index, ranging from 0, 1, 2, ..., N-1, and j represents the imaginary unit; S33, applying a frequency domain filter with a scale number of 4 to 6 to the frequency domain feature data to extract multi-scale frequency domain features with frequency ranges of 0 to 0.1 Hz, 0.1 to 0.5 Hz, 0.5 to 1 Hz, 1 to 2 Hz, and 2 to 5 Hz; S34. Perform inverse fast Fourier transform on the multi-scale frequency domain features: ; Where x(n) represents the time domain feature data value corresponding to the nth time domain sampling point, X(k) represents the data value of the kth frequency component in the frequency domain feature data, N represents the total length of the data sequence, n is the time domain index, ranging from 0, 1, 2, ..., N-1, k is the frequency index, ranging from 0, 1, 2, ..., N-1, and j represents the imaginary unit; S35, the obtained multi-scale time domain features are spliced in order of scale size and input into the fully connected network for fusion processing. The feature dimension after fusion is 64~128; S36. The fused features are output through a fully connected network to realize real-time prediction of the status of the metallurgical waste gas purification system.

[0034] During the implementation of the present invention, the specific method for obtaining real-time prediction features includes using a fully connected neural network to map a standardized input data set to a high-dimensional feature space with a dimension of 128256; extracting multi-scale frequency domain features with frequency ranges of 00.1Hz, 0.10.5Hz, 0.51Hz, 12Hz and 25Hz in the frequency domain space through a frequency domain filter with a scale number of 46; performing inverse fast Fourier transform on the multi-scale frequency domain features to obtain multi-scale time domain features; then splicing the obtained multi-scale time domain features in order of scale size and inputting them into a fully connected network for fusion processing, and finally obtaining real-time prediction features through a state prediction output layer. When implementing the improved FOX optimization algorithm, the initial population size is set to 2050 individuals, including four parameters: temperature (300800℃), pressure (0.12.0MPa), flow rate (1005000m³ / h) and reaction time (10120s). Each parameter corresponds to a clear numerical range. The search process adopts Lévy flight random search, Gaussian perturbation local search, adaptive dynamic search and elite retention mechanism to ensure reasonable switching between global and local searches. The condition for the parameter optimization iteration to converge is that the fitness value change is less than 10 for 2050 consecutive times. -5 During the implementation of the self-feedback reinforcement optimization strategy, the feedback gain coefficient is dynamically calculated and the control parameters are corrected in real time. The state error is calculated by the difference between the actual state data and the predicted state data. The total error value is quantified using the weighted absolute value error function. The control parameter correction is updated using the gradient of the error function. The termination condition of the iterative optimization is set to be less than 10 for 2050 consecutive fitness value changes. -4 ~10 -6 .

[0035] In the real-time feedback and closed-loop regulation of the optimal control parameters, the control parameters are transmitted to the control terminal of the metallurgical waste gas purification system in real time. The temperature, pressure, flow and reaction time in the reaction chamber are accurately adjusted in real time through the control terminal, and the feedback state parameters are collected in real time. When the deviation between the feedback state parameters and the optimal control parameters exceeds the preset threshold, the control terminal automatically re-executes the optimization strategy to update the parameters.

[0036] The present invention combines the neural operator with the improved FOX optimization algorithm to achieve accurate real-time prediction of the state parameters of the metallurgical waste gas purification system and multi-parameter collaborative optimization control, significantly improving the state prediction accuracy and optimization control response speed in the waste gas purification process, enhancing the stability and robustness of the dynamic adjustment of the waste gas purification process parameters, effectively improving the waste gas treatment efficiency and the accuracy of process parameter optimization, and significantly reducing the system operating costs.

[0037] The improved FOX optimization algorithm specifically includes: In the algorithm initialization step, the population size is set to 20-50, and each individual corresponds to a set of exhaust gas purification control parameters, including a temperature range of 300-800°C, a pressure range of 0.1-2.0 MPa, a flow range of 100-5000 m³ / h, and a reaction time range of 10-120 s. Adaptive dynamic search step, by calculating the population fitness value, taking the comprehensive indicators of exhaust gas purification efficiency and economic cost as the target, and dynamically adjusting the search range and step size according to the current population fitness variance; The Lévy flight search step performs a random walk search, and the random flight step size follows the Lévy distribution with a parameter range of 0.5~1.5; Gaussian perturbation local search step, in which Gaussian perturbations with a standard deviation of 10% to 20% of the current parameter value are applied to the individual; In the global and local search switching steps, the search mode is switched through a probability mechanism. The initial switching probability is set to 0.3~0.5 and decreases linearly with the number of iterations. Elite retention and iterative update steps: after each iteration, the top 10% to 20% of individuals in fitness are retained and the population is updated to the next generation; Convergence judgment step: when the global optimal individual fitness value changes by less than 10 in 20 to 50 consecutive iterations, -5 When , the optimal control parameters are output.

[0038] In this embodiment, the S4 specifically includes: S41. Using the real-time prediction features as input, establish an initial population with a population size of 20 to 50 individuals. Each individual contains control parameters for temperature, pressure, flow rate, and reaction time. The temperature range is 300 to 800°C, the pressure range is 0.1 to 2.0 MPa, the flow rate range is 100 to 5000 m³ / h, and the reaction time range is 10 to 120 s. S42. Calculate the fitness value of each individual in the initial population, where the fitness value is defined as a comprehensive evaluation index of exhaust gas purification efficiency and economic cost; S43, adaptively adjusting the search range and step size based on the calculated population fitness value and its variance, with the search range adjustment ratio being 10% to 30% of the initial range, and the step size adjustment ratio being 5% of the initial step size; S44, using the Lévy flight random search strategy to update the position of the individual in the population, the Lévy flight step size obeys the Lévy distribution, and the characteristic parameter α of the Lévy distribution ranges from 0.5 to 1.5; S45. Perform a Gaussian perturbation local search on the individual parameters of the population obtained by the search, with the perturbation standard deviation being 10% to 20% of the individual's current parameter value; S46. Set the initial value of the switching probability between global search and local search to 0.3~0.5, and decrease it linearly with the number of iterations. Update the population and implement the elite retention strategy to retain the top 10 individuals in fitness after each iteration. S47, in the process of 20 to 50 consecutive iterations, the fitness value of the global optimal individual changes less than the convergence threshold 10 -5 The iteration is terminated when , and the preliminary optimized control parameters of the exhaust gas purification system are obtained.

[0039] In this embodiment, in order to realize the prediction and optimization method of the metallurgical waste gas purification system combining the above-mentioned neural operator with the improved FOX optimization algorithm, the consistency and synchronization of real-time data acquisition and processing must be ensured during the actual deployment process. Specifically, the data acquisition end sensors should respectively use gas composition sensors, high-precision temperature sensors, pressure sensors and flow meters, and they should be fixedly installed at key locations such as the metallurgical waste gas inlet, reaction chamber and inlet pipe; the fully connected network structure in the neural operator model must be clearly defined as 3 to 5 layers, and the number of neurons in each layer must be set to 128, 256, and 128 respectively; the improved FOX optimization algorithm must clearly specify the selection principles and adjustment criteria of the initial parameters, such as the initial feedback gain coefficient is 0.5, the gain adjustment factor is 0.3, and the iterative convergence threshold is set to 10 -5 .

[0040] This invention combines a neural operator model with an improved FOX optimization algorithm to accurately and in real time predict the operating state of a metallurgical waste gas purification system and collaboratively optimize control parameters. Compared with traditional technologies, this invention effectively reduces state prediction errors, improves the accuracy and convergence speed of optimized control parameters, significantly enhances the operating efficiency and stability of the metallurgical waste gas purification process, and provides greater adaptability to complex operating conditions.

[0041] The self-feedback reinforcement optimization strategy specifically includes: Initial optimization control parameter acquisition: using the exhaust system state characteristics predicted in real time by the neural operator model as input, the improved FOX optimization algorithm is used to obtain the initial optimization control parameters; Real-time status data collection: input the preliminary optimized control parameters into the metallurgical waste gas purification system, and collect the actual status data of the metallurgical waste gas purification system in real time; State error calculation, calculation error function: ; Among them, E(t) represents the total state error at time t, Indicates the value of the ith actual measured state parameter of the metallurgical waste gas purification system, represents the value of the i-th predicted state parameter, represents the weight coefficient of the i-th state parameter, and n represents the total number of state parameters; Adaptive feedback gain calculation, dynamic calculation of feedback gain coefficient: ; in, represents the dynamic feedback gain coefficient, Indicates the initial feedback gain coefficient, alpha indicates the gain adjustment factor, and its value range is 0.1~0.5. Indicates the allowed error threshold; Control parameter correction: real-time correction of control parameters based on dynamic feedback gain coefficient: ; Among them, P(t+1) represents the optimized control parameters at the next moment, and P(t) represents the optimized control parameters at the current moment. Represents the gradient of the error function relative to the current parameter, defined as: ; in, represents the value of the i-th control parameter at the current moment, and m represents the total number of control parameters; The iterative optimization is terminated when the change of the global optimal individual fitness value in 20 to 50 consecutive iterations is less than the set convergence error threshold of 10 -4 ~10 -6 When , the iterative optimization is stopped and the final optimized control parameters are obtained.

[0042] In this embodiment, the S5 specifically includes: S51, inputting the preliminary optimized control parameters into the self-feedback reinforcement optimization strategy; S52, real-time collection of actual operating status data of the metallurgical waste gas purification system; S53, calculating a state error value according to the real-time state data; S54, dynamically updating the optimized control parameters according to the state error value; S55, performing multiple dynamic iterative optimizations on the updated optimized control parameters; S56, when the change range of the optimization control parameter during 20 to 50 consecutive iterations is less than the set convergence error threshold of 10 -4 ~10 -6 When , the optimal control parameter set is output.

[0043] In this implementation, the various sensors that collect real-time exhaust gas data are connected to the control terminal via an industrial-grade communication interface. The communication protocol uses Modbus RTU, and the transmission rate is set to 9600bps. A high-precision platinum resistance temperature sensor, a strain gauge pressure sensor, and a vortex flowmeter are used as the flowmeter. All sensors are strictly installed at the exhaust gas inlet or inside the reaction chamber to ensure data accuracy.

[0044] The present invention accurately collects waste gas composition, temperature, pressure, flow and other data in real time, uses neural operators and improved FOX optimization algorithm to construct a dynamic prediction and multi-parameter collaborative optimization mechanism, and introduces a self-feedback reinforcement optimization strategy to achieve accurate prediction and real-time optimization control of the operating status of the metallurgical waste gas purification system, effectively improving the waste gas purification efficiency, reducing the system operating cost, and significantly improving the control delay and poor parameter adaptability problems of traditional methods, thereby improving the intelligence level of waste gas purification in the metallurgical industry production process.

[0045] In this embodiment, S6 specifically includes: S61, transmitting the obtained optimal control parameter set to the control terminal of the metallurgical waste gas purification system in real time through the communication interface; S62, the control terminal adjusts the temperature of the reaction chamber in the metallurgical waste gas purification system in real time based on the centralized temperature parameters of the optimal control parameters, with a temperature control accuracy of ±0.5°C; S63, based on the optimal control parameter centralized pressure parameter, real-time control of the pressure in the reaction chamber, with a pressure control accuracy of ±0.01 MPa; S64, based on the optimal control parameter centralized flow parameters, adjust the exhaust gas inlet flow in real time, with a flow control accuracy of ±10m³ / h; S65, based on the optimal control parameter and the reaction time parameter, the exhaust gas residence time in the reaction chamber is controlled in real time, and the reaction time control accuracy is ±1s; S66: Real-time collection of exhaust gas state parameters after feedback control, and periodic transmission of the collected feedback data back to the control terminal, with a data transmission cycle of 1 to 2 seconds; S67. The control terminal performs a real-time comparison between the feedback state parameter and the optimal control parameter set. When the deviation of the feedback state parameter exceeds a preset threshold range, the optimization strategy is re-executed to update the optimal control parameter set.

[0046] In this implementation, the control terminal utilizes an industrial-grade embedded controller, which receives the optimal control parameter set in real time via an RS-485 communication interface. A built-in PID closed-loop control algorithm implements precise regulation of temperature, pressure, flow rate, and response time. The proportional parameter (P) of the PID control algorithm ranges from 2.0 to 5.0, the integral parameter (I) ranges from 0.01 to 0.05, and the differential parameter (D) ranges from 0.1 to 0.5, ensuring precise control of the system's operating state. Real-time feedback data is collected by an acquisition module within the control terminal, which includes a built-in 12-bit analog-to-digital converter (ADC) with a sampling frequency of ≥1Hz. This real-time data is transmitted to an industrial server via a communication protocol to ensure both real-time and accurate data transmission.

[0047] The present invention implements real-time closed-loop regulation of the optimal control parameter set to achieve precise control of the temperature, pressure, flow rate and reaction time of the metallurgical waste gas purification process. The system has a fast response speed and high control accuracy. The real-time feedback mechanism can quickly detect and correct system deviations, effectively ensuring the stability and optimization effect of the waste gas purification process, and significantly improving the real-time performance and reliability of the metallurgical waste gas purification system.

[0048] Example 1: In order to verify the feasibility of the present invention, the present invention was applied to the metallurgical waste gas purification system of Company B, a medium-sized steel smelting enterprise located in Tangshan, Hebei. Company B produces about 3 million tons of steel products annually. Its smelting process produces a large amount of high-temperature waste gas containing harmful substances such as sulfur dioxide, carbon monoxide, and nitrogen oxides, which need to undergo strict purification treatment before they can be discharged. Recently, the environmental monitoring department of Company B found that the existing waste gas purification system had insufficient control accuracy and slow parameter optimization and adjustment speed, resulting in occasional excessive waste gas emissions, seriously affecting the compliance of environmental protection indicators and the stability of corporate production. Therefore, Company B decided to apply the metallurgical waste gas purification system prediction and optimization method based on the neural operator combined with the improved FOX optimization algorithm proposed in the present invention to improve the efficiency and stability of waste gas purification.

[0049] In specific implementation, a gas composition sensor is installed at the exhaust gas inlet of the metallurgical exhaust gas purification system. High-precision temperature sensors, pressure sensors, and flow meters are installed in the exhaust gas inlet pipe and reaction chamber, respectively. These sensors collect real-time data on exhaust gas composition, temperature, pressure, and flow rate. Furthermore, a high-precision timer accurately records the start and end times of the exhaust gas purification reaction, forming a unified initial multi-dimensional data sample library.

[0050] Subsequently, taking the mean of each data sequence in the initial multidimensional data sample library as a benchmark, the 3σ criterion is used to identify and eliminate abnormal data. The Daubechies wavelet denoising method is used to denoise the data sequence after the abnormal data is eliminated, and the denoised data sequence is normalized. Finally, principal component analysis is used to fuse the data and extract the main features to obtain a standardized high-quality input data set.

[0051] Next, a neural operator model is applied to the standardized input dataset for real-time multi-scale dynamic modeling. The neural operator model first uses a fully connected neural network to map the data into a 128-dimensional high-dimensional feature space. It then extracts multi-scale frequency domain features using Fourier transforms and multi-scale frequency domain filters. These features are then converted into multi-scale time domain features using an inverse Fourier transform. After feature fusion through the fully connected network, the model outputs real-time prediction features for the exhaust gas purification system's state.

[0052] The real-time prediction features are then used as input to the improved FOX optimization algorithm, which performs real-time collaborative optimization of multiple parameters. Using a Lévy flight random search and Gaussian perturbation local search strategy, the initial optimized control parameters for the exhaust gas purification system are quickly obtained. Subsequently, a self-feedback reinforcement optimization strategy is used to dynamically adjust the initial optimized control parameters in real time. A state error function is established, and feedback gain coefficients are dynamically calculated. The parameters are iteratively optimized until convergence, ultimately outputting the optimal control parameter set.

[0053] Finally, the optimal control parameter set is transmitted in real time via a communication interface to the metallurgical waste gas purification system control terminal. An embedded controller then executes PID closed-loop control to precisely adjust temperature, pressure, flow rate, and response time. Simultaneously, status parameters are fed back in real time. If deviations exceed thresholds, further optimization adjustments are made to ensure continuous and stable system operation.

[0054] To verify the application effect, Company B conducted data monitoring and statistical analysis on the operation status of the exhaust gas purification system before and after implementation for a period of three months. The relevant data are summarized in Table 1 below: Table 1: Comparison of the operating conditions of the exhaust gas purification system before and after the implementation of the present invention ; It can be clearly seen from the above data that before the implementation of this invention, the control accuracy of Company B's exhaust gas purification system was obviously insufficient. The average deviations of temperature, pressure, flow and reaction time control were as high as ±3.5°C, ±0.08MPa, ±120m³ / h and ±9.5s respectively. The exhaust gas emission index compliance rate was only 87.3%, and the average number of failures per month was as high as 5 times. The system parameter adjustment took an average of about 45 minutes, and it was unable to quickly adapt to fluctuations in exhaust gas composition, seriously affecting production efficiency and environmental compliance.

[0055] After implementing the present invention, the control accuracy of system temperature, pressure, flow and reaction time was improved to within ±0.4°C, ±0.005MPa, ±8m³ / h and ±0.5s, respectively, significantly improving the control accuracy and response speed; the compliance rate of waste gas emission indicators was increased to 99.8%, the average monthly number of system failures was reduced to 0, and the average time for system parameter optimization and adjustment was greatly reduced from 45 minutes to 3 minutes, which greatly improved the waste gas treatment efficiency and significantly improved environmental compliance and system stability.

[0056] Furthermore, employees in Company B's production workshop reported that the implementation of this invention has resulted in a more stable exhaust gas treatment system, significantly improved on-site working conditions, and virtually eliminated exhaust gas leakage accidents, effectively enhancing production safety. Monitoring data from the company's environmental protection department also shows that since implementing the system, exhaust gas concentrations of sulfur dioxide, carbon monoxide, and nitrogen oxides have consistently remained well below national environmental emission standards. The company's environmental indicators have met the standards for three consecutive months, earning it high recognition and praise from the local government's environmental protection department.

[0057] Through the above specific embodiments and comparative analysis of relevant data, the prediction and optimization method of metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm proposed in the present invention not only has good applicability and high efficiency in actual industrial environment, but also provides a new and effective technical solution for steel and metallurgical enterprises to improve waste gas purification efficiency and enhance environmental protection level, and has broad industrial application value.

[0058] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm, characterized in that: The steps include: S1. Real-time collection of waste gas component concentration, temperature, pressure, flow rate and reaction time data during the operation of the metallurgical waste gas purification system to establish an initial multi-dimensional data sample set; S2. performing adaptive preprocessing on the initial multidimensional data sample set, including data anomaly removal, data denoising, data fusion and multi-scale feature extraction, to obtain a standardized input data set; S3. Inputting the standardized input data set into a neural operator model to perform real-time multi-scale dynamic modeling to obtain real-time prediction characteristics of the metallurgical waste gas system state; S4. Using the improved FOX optimization algorithm, taking the real-time prediction features as input, and combining it with an adaptive dynamic search strategy to perform multi-parameter collaborative optimization in real time, to obtain preliminary optimized control parameters of the exhaust gas purification system; S5. Implementing a self-feedback reinforcement optimization strategy on the preliminary optimized control parameters, performing real-time dynamic iterative optimization, and obtaining an optimal control parameter set; S6. Feedback the optimal control parameter set to the metallurgical waste gas purification system in real time, and control the metallurgical waste gas purification system to perform closed-loop regulation according to the optimal control parameter set.

2. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: Said S1 specifically includes: S11. Install a gas composition sensor at the exhaust gas inlet of the metallurgical exhaust gas purification system to collect concentration data of sulfur dioxide, carbon monoxide, and nitrogen oxides in real time, with a sampling frequency of 1 to 5 seconds; S12. Install high-precision temperature sensors at the exhaust gas inlet and reaction chamber of the metallurgical exhaust gas purification system to collect real-time data on the inlet gas temperature and the temperature inside the reaction chamber. The temperature sampling accuracy is ±0.1°C. S13. Install a pressure sensor inside the reaction chamber of the metallurgical waste gas purification system to collect real-time pressure data in the chamber. The pressure measurement range is 0.1~2.0MPa; S14. Install a flow meter at the gas inlet pipe of the metallurgical waste gas purification system to collect waste gas flow data in real time. The flow measurement range is 100~5000m³ / h; S15. Timestamp the start and end times of each purification process of the metallurgical waste gas purification system and record the reaction time data. The accuracy of the reaction time is ±0.5s. S16. Using the timestamp of the start of the reaction process as a reference, the real-time collected exhaust gas component concentration, temperature, pressure, and flow rate data are time-aligned according to the timestamp, and matched one by one to the corresponding reaction time period to form a multidimensional data sequence under a unified time reference; S17. Based on the multidimensional data sequence, data splicing and fusion are performed with timestamps as associated indexes to construct an initial multidimensional data sample set with unified timestamps and clear internal data structure.

3. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The S2 specifically includes: S21, taking the mean of each data sequence in the initial multidimensional data sample set as the benchmark, adopt the The abnormal data are identified and eliminated based on the criteria, is the standard deviation of each data series; S22. After the abnormal data are removed, the data sequence is subjected to data denoising using a wavelet threshold denoising method, wherein the wavelet basis function is Daubechies wavelet and the decomposition scale is set to 3 to 5 layers; S23, applying normalization processing to the denoised multidimensional data sequence, with the normalization processing range of the data being [0, 1]; S24. Perform data fusion on the normalized data series using principal component analysis, with the cumulative contribution rate of the principal component set at 90% to 95%; S25. Perform multi-scale feature extraction on the sequence after data fusion, and use the empirical mode decomposition method to decompose the fused data sequence into intrinsic mode functions step by step, with the decomposition order of each data sequence set to 4 to 6; S26. Concatenate the eigenmode functions of each order in descending order of scale to form a standardized input data set.

4. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The neural operator model specifically includes: The input mapping layer uses a fully connected neural network to map the normalized input data set to a high-dimensional feature space with a dimension of 128 to 256; The Fourier operator layer uses fast Fourier transform to perform frequency domain transformation on the mapped high-dimensional feature space data; The multi-scale frequency domain filter layer extracts multi-scale frequency domain features of low frequency 0sim0.1Hz, medium-low frequency 0.1~0.5Hz, medium frequency 0.5~1Hz, medium-high frequency 1~2Hz, and high frequency 2~5Hz respectively through frequency domain filters with a scale of 4~6 in the frequency domain space; The inverse Fourier transform layer performs inverse fast Fourier transform on the multi-scale frequency domain features after frequency domain filtering, maps the frequency domain features back to the time domain, and obtains the corresponding multi-scale time domain features; The feature fusion layer combines multi-scale temporal features and fuses them through a fully connected network. The output dimension after feature fusion is 64-128. The state prediction output layer outputs the real-time prediction features of the metallurgical waste gas purification system state based on the fused features through a fully connected network.

5. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The S3 specifically includes: S31. Use a fully connected neural network to map the standardized input data set to a high-dimensional feature space with a dimension of 128 to 256 to obtain high-dimensional feature data; S32. Perform fast Fourier transform on high-dimensional feature data: ； Where X(k) represents the frequency domain feature data corresponding to the kth frequency component, x(n) represents the data value of the nth time domain sampling point in the high-dimensional feature data, N represents the total length of the data sequence, k is the frequency index, ranging from 0, 1, 2, ..., N-1, n is the time domain index, ranging from 0, 1, 2, ..., N-1, and j represents the imaginary unit; S33, applying a frequency domain filter with a scale number of 4 to 6 to the frequency domain feature data to extract multi-scale frequency domain features with frequency ranges of 0 to 0.1 Hz, 0.1 to 0.5 Hz, 0.5 to 1 Hz, 1 to 2 Hz, and 2 to 5 Hz; S34. Perform inverse fast Fourier transform on the multi-scale frequency domain features: ； Where x(n) represents the time domain feature data value corresponding to the nth time domain sampling point, X(k) represents the data value of the kth frequency component in the frequency domain feature data, N represents the total length of the data sequence, n is the time domain index, ranging from 0, 1, 2, ..., N-1, k is the frequency index, ranging from 0, 1, 2, ..., N-1, and j represents the imaginary unit; S35, the obtained multi-scale time domain features are spliced in order of scale size and input into the fully connected network for fusion processing. The feature dimension after fusion is 64~128; S36. The fused features are output through a fully connected network to realize real-time prediction of the status of the metallurgical waste gas purification system.

6. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The improved FOX optimization algorithm specifically includes: In the algorithm initialization step, the population size is set to 20-50, and each individual corresponds to a set of exhaust gas purification control parameters, including a temperature range of 300-800°C, a pressure range of 0.1-2.0 MPa, a flow range of 100-5000 m³ / h, and a reaction time range of 10-120 s. Adaptive dynamic search step, by calculating the population fitness value, taking the comprehensive indicators of exhaust gas purification efficiency and economic cost as the target, and dynamically adjusting the search range and step size according to the current population fitness variance; The Lévy flight search step performs a random walk search, and the random flight step size follows the Lévy distribution with a parameter range of 0.5~1.5; Gaussian perturbation local search step, in which Gaussian perturbations with a standard deviation of 10% to 20% of the current parameter value are applied to the individual; In the global and local search switching steps, the search mode is switched through a probability mechanism. The initial switching probability is set to 0.3~0.5 and decreases linearly with the number of iterations. Elite retention and iterative update steps: after each iteration, the top 10% to 20% of individuals in fitness are retained and the population is updated to the next generation; Convergence judgment step: when the global optimal individual fitness value changes by less than 10 in 20 to 50 consecutive iterations, -5 When , the optimal control parameters are output.

7. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The S4 specifically includes: S41. Using the real-time prediction features as input, establish an initial population with a population size of 20 to 50 individuals. Each individual contains control parameters for temperature, pressure, flow rate, and reaction time. The temperature range is 300 to 800°C, the pressure range is 0.1 to 2.0 MPa, the flow rate range is 100 to 5000 m³ / h, and the reaction time range is 10 to 120 s. S42. Calculate the fitness value of each individual in the initial population, where the fitness value is defined as a comprehensive evaluation index of exhaust gas purification efficiency and economic cost; S43, adaptively adjusting the search range and step size based on the calculated population fitness value and its variance, with the search range adjustment ratio being 10% to 30% of the initial range, and the step size adjustment ratio being 5% of the initial step size; S44, using the Lévy flight random search strategy to update the position of the individual in the population, the Lévy flight step size obeys the Lévy distribution, and the characteristic parameter α of the Lévy distribution ranges from 0.5 to 1.5; S45. Perform a Gaussian perturbation local search on the individual parameters of the population obtained by the search, with the perturbation standard deviation being 10% to 20% of the individual's current parameter value; S46. Set the initial value of the switching probability between global search and local search to 0.3~0.5, and decrease it linearly with the number of iterations. Update the population and implement the elite retention strategy to retain the top 10 individuals in fitness after each iteration. S47, in the process of 20 to 50 consecutive iterations, the fitness value of the global optimal individual changes less than the convergence threshold 10 -5 The iteration is terminated when , and the preliminary optimized control parameters of the exhaust gas purification system are obtained.

8. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The self-feedback reinforcement optimization strategy specifically includes: Initial optimization control parameter acquisition: using the exhaust system state characteristics predicted in real time by the neural operator model as input, the improved FOX optimization algorithm is used to obtain the initial optimization control parameters; Real-time status data collection: input the preliminary optimized control parameters into the metallurgical waste gas purification system, and collect the actual status data of the metallurgical waste gas purification system in real time; State error calculation, calculation error function: ； Among them, E(t) represents the total state error at time t, Indicates the value of the ith actual measured state parameter of the metallurgical waste gas purification system, represents the value of the i-th predicted state parameter, represents the weight coefficient of the i-th state parameter, and n represents the total number of state parameters; Adaptive feedback gain calculation, dynamic calculation of feedback gain coefficient: ； in, represents the dynamic feedback gain coefficient, Indicates the initial feedback gain coefficient, alpha indicates the gain adjustment factor, and its value range is 0.1~0.

5. Indicates the allowed error threshold; Control parameter correction: real-time correction of control parameters based on dynamic feedback gain coefficient: ； Among them, P(t+1) represents the optimized control parameters at the next moment, and P(t) represents the optimized control parameters at the current moment. Represents the gradient of the error function relative to the current parameter, defined as: ； in, represents the value of the i-th control parameter at the current moment, and m represents the total number of control parameters; The iterative optimization is terminated when the change of the global optimal individual fitness value in 20 to 50 consecutive iterations is less than the set convergence error threshold of 10 -4 ~10 -6 When , the iterative optimization is stopped and the final optimized control parameters are obtained.

9. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1 is characterized in that: The S5 specifically includes: S51, inputting the preliminary optimized control parameters into the self-feedback reinforcement optimization strategy; S52, real-time collection of actual operating status data of the metallurgical waste gas purification system; S53, calculating a state error value according to the real-time state data; S54, dynamically updating the optimized control parameters according to the state error value; S55, performing multiple dynamic iterative optimizations on the updated optimized control parameters; S56, when the change range of the optimization control parameter during 20 to 50 consecutive iterations is less than the set convergence error threshold of 10 -4 ~10 -6 When , the optimal control parameter set is output.

10. The prediction and optimization method for metallurgical waste gas purification system based on neural operator combined with improved FOX optimization algorithm according to claim 1, characterized in that: The S6 specifically includes: S61, transmitting the obtained optimal control parameter set to the control terminal of the metallurgical waste gas purification system in real time through the communication interface; S62, the control terminal adjusts the temperature of the reaction chamber in the metallurgical waste gas purification system in real time based on the centralized temperature parameters of the optimal control parameters, with a temperature control accuracy of ±0.5°C; S63, based on the optimal control parameter centralized pressure parameter, real-time control of the pressure in the reaction chamber, with a pressure control accuracy of ±0.01 MPa; S64, based on the optimal control parameter centralized flow parameters, adjust the exhaust gas inlet flow in real time, with a flow control accuracy of ±10m³ / h; S65, based on the optimal control parameter and the reaction time parameter, the exhaust gas residence time in the reaction chamber is controlled in real time, and the reaction time control accuracy is ±1s; S66: Real-time collection of exhaust gas state parameters after feedback control, and periodic transmission of the collected feedback data back to the control terminal, with a data transmission cycle of 1 to 2 seconds; S67. The control terminal performs a real-time comparison between the feedback state parameter and the optimal control parameter set. When the deviation of the feedback state parameter exceeds a preset threshold range, the optimization strategy is re-executed to update the optimal control parameter set.

Images