An artificial intelligence-based metallurgical waste gas purification system prediction and optimization method

Abstract

The application discloses a metallurgical waste gas purification system prediction and optimization method based on artificial intelligence, which comprises the following steps: collecting waste gas component concentration, temperature, pressure, flow and reaction time data in the running process of the waste gas purification system in real time and constructing an initial multi-dimensional data sample set; performing self-adaptive pretreatment 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 to obtain state real-time prediction characteristics through real-time dynamic modeling; obtaining preliminary optimization control parameters through real-time collaborative optimization based on the prediction characteristics and using an improved FOX optimization algorithm; and implementing a self-feedback reinforcement optimization strategy to dynamically iteratively optimize to obtain an optimal control parameter set and realize real-time feedback closed-loop regulation and control. The application realizes efficient and accurate optimization control of the waste gas purification system, significantly improves waste gas purification efficiency and stability, and dynamically adapts to complex production conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metallurgical waste gas purification and artificial intelligence cross-fusion, and particularly relates to a metallurgical waste gas purification system prediction and optimization method based on artificial intelligence. BACKGROUND

[0002] With the rapid development of industry and the increasingly stringent environmental protection standards, the problem of waste gas pollution discharged in the production process of the metallurgical industry has gradually become the focus of social attention. Metallurgical waste gas usually contains high concentrations of harmful gases such as sulfur dioxide, carbon monoxide, and nitrogen oxides. These substances not only pollute the environment, but also seriously threaten human health. Therefore, efficient and stable waste gas purification treatment technology has become an important direction of research in this field.

[0003] Currently, the optimization control technology of the metallurgical waste gas purification system mainly adopts traditional feedback control methods and empirical model methods. The feedback control method generally adjusts parameters based on a PID controller or classical control theory, and the empirical model method adjusts and controls system parameters relying on artificial design rules and empirical data. These technical methods are relatively simple and direct in the implementation process, easy to implement in engineering, have been widely used in the industry, and have achieved certain results. However, with the diversification and complexity of production conditions, especially when large-scale fluctuations or process mutations occur in the metallurgical production process, the traditional feedback control method has difficulty in achieving effective real-time response and accurate regulation and control of dynamic changes in the system due to the fixedness of parameters and the hysteresis of response, often leading to low system operation efficiency or even the risk of loss of control. On the other hand, the empirical model method relies heavily on historical data and artificial experience, and has insufficient universality and adaptability, making it difficult to effectively predict complex and variable waste gas treatment environmental conditions, and there are obvious hysteresis and large errors, which makes it difficult to meet the current demand for precise control of waste gas purification.

[0004] In recent years, with the development of artificial intelligence technology, especially deep learning technology, various intelligent control strategies based on machine learning models and optimization algorithms have been introduced into the metallurgical waste gas purification system in order to improve the dynamic response capability and precise control effect of the system. Among them, traditional neural network models (such as BP network, CNN network) and classical optimization algorithms (such as genetic algorithm, particle swarm algorithm, etc.) have been preliminarily applied in the field of waste gas treatment. These technical solutions usually predict the waste gas treatment state by constructing a neural network model, and then adjust the system control parameters by combining intelligent optimization algorithms. However, the traditional neural network model has weak network generalization performance when predicting complex nonlinear systems, and is easily disturbed by actual production environment, resulting in significant deviation. At the same time, the classical optimization algorithm is prone to fall into local optimal solution, and has low optimization efficiency and slow convergence speed, which makes it difficult to meet the strict requirements of real-time response in industrial environment, causing the waste gas treatment process to be difficult to achieve ideal optimization effect.

[0005] In view of the deficiencies of the prior art, the neural operator technology emerging in recent years has gradually become an effective way to solve the problem of modeling and prediction of complex dynamic systems. The neural operator can efficiently simulate and predict nonlinear systems with multi-scale dynamic changes, has good generalization ability and real-time response characteristics, and has shown obvious performance advantages in the fields of fluid mechanics, power systems and the like. In addition, new optimization algorithms such as FOX optimization algorithm have also attracted widespread attention from the industry and academia due to their good global search ability and efficient convergence characteristics. However, the research on neural operators and FOX algorithms in the field of metallurgical waste gas purification systems is still in its infancy, and there is still a lack of systematic technical solutions on how to effectively combine these cutting-edge artificial intelligence technologies to accurately predict the state of the waste gas system and realize efficient and real-time optimization and regulation of the purification parameters.

[0006] Therefore, how to provide an artificial intelligence-based metallurgical waste gas purification system prediction and optimization method is a problem that those skilled in the art need to solve. SUMMARY

[0007] One object of the present application is to provide an artificial intelligence-based metallurgical waste gas purification system prediction and optimization method. The present application adopts a technical route in which a neural operator technology and an improved FOX optimization algorithm are used in cooperation. By real-time collection of data such as waste gas component concentration, temperature, pressure, flow rate and reaction time during the operation of the metallurgical waste gas purification system, an initial multi-dimensional data sample set is constructed and self-adaptive preprocessing is performed. A neural operator model is used to perform real-time multi-scale dynamic modeling and prediction on the preprocessed data set. An improved FOX optimization algorithm is used to perform real-time multi-parameter collaborative optimization to obtain preliminary optimized control parameters. Then, a self-feedback reinforcement optimization strategy is designed and implemented to perform real-time dynamic iterative optimization on the preliminary optimized control parameters to obtain an optimal control parameter set, and finally closed-loop accurate regulation of the waste gas purification system is realized. The technical solution of the present application effectively solves the problems of insufficient waste gas prediction accuracy, difficulty in dynamically responding to complex production conditions, and easy falling into local optimum and slow convergence of traditional optimization algorithms in the prior art, and 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 the metallurgical waste gas purification system prediction and optimization method based on the neural operator combined with the improved FOX optimization algorithm according to an embodiment of the present application, the following steps are included:

[0009] S1, real-time collection of data such as waste gas component concentration, temperature, pressure, flow rate and reaction time during the operation of the metallurgical waste gas purification system, establishment of an initial multi-dimensional data sample set;

[0010] S2, adaptive preprocessing of the initial multi-dimensional data sample set, including data outlier rejection, data denoising, data fusion and multi-scale feature extraction, to obtain a standardized input data set;

[0011] S3, inputting the standardized input data set into a neural operator model for real-time multi-scale dynamic modeling to obtain real-time prediction features of the metallurgical waste gas system state;

[0012] S4, using an improved FOX optimization algorithm, taking the real-time prediction features as input, combining an adaptive dynamic search strategy to perform real-time multi-parameter collaborative optimization to obtain preliminary optimization control parameters of the waste gas purification system;

[0013] S5, implementing a self-feedback reinforcement optimization strategy on the preliminary optimization control parameters for real-time dynamic iterative optimization to obtain an optimal control parameter set;

[0014] S6, feeding back the optimal control parameter set to the metallurgical waste gas purification system in real time, and controlling the metallurgical waste gas purification system to perform closed-loop regulation and control according to the optimal control parameter set.

[0015] Optionally, the S1 specifically comprises:

[0016] S11, installing a gas component sensor at the waste gas inlet end of the metallurgical waste gas purification system to collect concentration data of sulfur dioxide, carbon monoxide and nitrogen oxides in real time, with a sampling frequency of Once;

[0017] S12, installing high-precision temperature sensors at the waste gas inlet end and inside the reaction chamber of the metallurgical waste gas purification system to collect inlet gas temperature and reaction chamber temperature data in real time, with a temperature sampling accuracy of ;

[0018] S13, installing a pressure sensor inside the reaction chamber of the metallurgical waste gas purification system to collect chamber pressure data in real time, with a pressure measurement range of ;

[0019] S14, installing a flowmeter on the gas inlet pipeline of the metallurgical waste gas purification system to collect waste gas flow data in real time, with a flow measurement range of ;

[0020] S15, time stamping the start and end times of each purification process of the metallurgical waste gas purification system to record reaction time data, with a reaction time accuracy of ;

[0021] S16, taking the timestamp of the reaction process starting moment as the reference, the real-time collected exhaust gas component concentration, temperature, pressure and flow data are time axis aligned according to the timestamp, and matched one by one into the corresponding reaction time period, forming a multi-dimensional data sequence under the unified time reference;

[0022] S17, according to the multi-dimensional data sequence, the data is spliced and fused with the timestamp as the association index, and an initial multi-dimensional data sample set with unified timestamp and clear internal data structure is constructed.

[0023] Optionally, the S2 specifically comprises:

[0024] S21, taking the mean value of each data sequence in the initial multi-dimensional data sample set as the reference, adopting the outlier data identification and elimination method based on The criterion, wherein The standard deviation of each data sequence;

[0025] S22, for the data sequence after the elimination of abnormal data, the wavelet threshold denoising method is used for data denoising processing, wherein the wavelet basis function is Daubechies wavelet, and the decomposition scale is set to Layers;

[0026] S23, the normalized processing is adopted for the denoised multi-dimensional data sequence, and the data normalization processing range is ;

[0027] S24, data fusion is performed on the data sequence after the normalization processing, and the principal component analysis method is adopted for data fusion, and the principal component cumulative contribution rate is set to ;

[0028] S25, multi-scale feature extraction is performed on the sequence after data fusion, and the empirical mode decomposition method is adopted to decompose the fused data sequence into intrinsic mode functions step by step, and the decomposition order of each data sequence is set to Order;

[0029] S26, the intrinsic mode functions of each order are spliced in order from large to small scale, and a standardized input data set is constructed.

[0030] Optionally, the neural operator model specifically comprises:

[0031] An input mapping layer, which adopts a fully connected neural network to map the standardized input data set to a high-dimensional feature space with a dimension of ;

[0032] A Fourier operator layer, which adopts fast Fourier transform to perform frequency domain transformation on the mapped high-dimensional feature space data;

[0033] A multi-scale frequency domain filtering layer, which performs multi-scale frequency domain filtering on the frequency domain space data through a scale number of low frequency , middle-low frequency , middle frequency , middle-high frequency , high frequency ;

[0034] inverse Fourier transform layer, performing inverse fast Fourier transform on the multi-scale frequency domain features filtered in the frequency domain respectively, mapping the frequency domain features back to the time domain to obtain corresponding multi-scale time domain features;

[0035] feature fusion layer, splicing the multi-scale time domain features, and performing fusion processing through a fully connected network, wherein the output dimension of the fused features is ;

[0036] state prediction output layer, outputting real-time prediction features of the metallurgical waste gas purification system state through a fully connected network.

[0037] Optionally, the S3 specifically comprises:

[0038] S31, mapping the standardized input data set to a high-dimensional feature space with a dimension of using a fully connected neural network, to obtain high-dimensional feature data;

[0039] S32, performing fast Fourier transform on the high-dimensional feature data:

[0040] ;

[0041] wherein, denotes frequency domain feature data corresponding to the i-th frequency component, denotes the data value of the i-th time domain sampling point in the high-dimensional feature data, denotes the total length of the data sequence, is a frequency index, and the value range is , is a time domain index, and the value range is , denotes an imaginary unit; S33, using a frequency domain filter with a scale number of to extract multi-scale frequency domain features with frequency ranges of

[0042] , , , , and ;

[0043] ​S34, inverse fast Fourier transform is performed on the multi-scale frequency domain features respectively;

[0044] ;

[0045] wherein, represents the time domain feature data value corresponding to the i th time domain sampling point, represents the data value of the i th frequency component in the frequency domain feature data, represents the total length of the data sequence, is a time domain index, and the value range is , is a frequency index, and the value range is , represents an imaginary unit; S35, the obtained multi-scale time domain features are spliced in order of scale size and input to a fully connected network for fusion processing, and the feature dimension after fusion is ;

[0046] S36, the fused features are output by the fully connected network to obtain real-time prediction features of the metallurgical waste gas purification system state.

[0047]

[0048] Optionally, the improved FOX optimization algorithm specifically includes:

[0049] An algorithm initialization step, setting the population size to , and each individual corresponds to a set of waste gas purification control parameters, wherein the temperature range is , the pressure range is , the flow range is , and the reaction time range is ;

[0050] An adaptive dynamic search step, by calculating the population fitness value, taking the waste gas purification efficiency and economic cost comprehensive index as the target, and dynamically adjusting the search range and step length according to the current population fitness variance;

[0051] A Levy flight search step, performing random walk search, and the random flight step length obeys the Levy distribution with the parameter range of ;

[0052] A Gaussian disturbance local search step, in the local search stage, the individual is subjected to Gaussian disturbance with a standard deviation of the current parameter value ;

[0053] A global and local search switching step, the search mode is switched through a probability mechanism, the initial switching probability is set to , and decreases linearly with the number of iterations; ​​

[0054] elite reservation and iteration update step, after each iteration, the top individuals in fitness are reserved, and the population is updated to the next generation;

[0055] convergence determination step, when the fitness value of the global optimal individual changes less than in the continuous iteration process, the optimal control parameter is output.

[0056] Optionally, the S4 specifically comprises:

[0057] S41, taking the real-time prediction characteristics as input, an initial population is established, the population size is individuals, each individual contains the control parameters of temperature, pressure, flow and reaction time, the temperature value range is , the pressure value range is , the flow value range is , and the reaction time value range is ;

[0058] S42, the fitness value of each individual in the initial population is calculated respectively, and the fitness value is defined as the comprehensive evaluation index of exhaust gas purification efficiency and economic cost;

[0059] S43, according to the population fitness value and its variance obtained by calculation, the search range and step are adjusted adaptively, the search range adjustment ratio is of the initial range, and the step adjustment ratio is of the initial step;

[0060] S44, the population individual position is updated by using the Levy flight random search strategy, the Levy flight step is subject to Levy distribution, and the characteristic parameter of Levy distribution is ; ;

[0061] S45, the population individual parameters obtained by searching are subjected to Gaussian disturbance local search, and the disturbance standard deviation is of the current parameter value of the individual;

[0062] S46, the initial value of the switching probability between global search and local search is set to , and decreases linearly with the iteration number, the population is updated and the elite reservation strategy is executed, and the top individuals in fitness after each iteration are reserved;

[0063] S47, when the fitness value of the global optimal individual changes less than the convergence threshold in the continuous iteration process, the iteration is terminated, and the preliminary optimized control parameter of the exhaust gas purification system is obtained.

[0064] Optionally, the self-feedback reinforcement optimization strategy specifically comprises:

[0065] Preliminary optimization control parameter acquisition, taking the real-time predicted state characteristics of the exhaust system by the neural operator model as input, and using an improved FOX optimization algorithm to obtain the preliminary optimization control parameter;

[0066] Real-time state data acquisition, inputting the preliminary optimization control parameter into the metallurgical exhaust gas purification system, and collecting real-time actual state data of the metallurgical exhaust gas purification system;

[0067] State error calculation, calculating the error function:

[0068] ;

[0069] wherein, represents the total value of the state error at the th moment, represents the value of the th actual measured state parameter of the metallurgical exhaust gas purification system, represents the value of the th predicted state parameter, represents the weight coefficient of the th state parameter, represents the total number of state parameters;

[0070] Adaptive feedback gain calculation, dynamically calculating the feedback gain coefficient:

[0071] ;

[0072] wherein, represents the dynamic feedback gain coefficient, represents the initial feedback gain coefficient, represents the gain adjustment factor, and the value range is , represents the allowed error threshold;

[0073] Control parameter correction, real-time correction of the control parameter according to the dynamic feedback gain coefficient:

[0074] ;

[0075] wherein, represents the next moment optimization control parameter, represents the current moment optimization control parameter, represents the gradient of the error function relative to the current parameter, and is defined as:

[0076] ;

[0077] wherein, a number representing the current time, a number representing the current time, a number representing the total number of control parameters;

[0078] The iterative optimization is terminated when the fitness value of the global optimal individual changes by less than a set convergence error threshold times of iterations, the iterative optimization is stopped, and the final optimized control parameters are obtained.

[0079] Optionally, the S5 specifically includes:

[0080] S51, input the preliminary optimized control parameters to the self-feedback reinforcement optimization strategy;

[0081] S52, collect the actual running state data of the metallurgical off-gas purification system in real time;

[0082] S53, calculate the state error value according to the real-time state data;

[0083] S54, dynamically update the optimized control parameters according to the state error value;

[0084] S55, perform multiple dynamic iterative optimizations on the updated optimized control parameters;

[0085] S56, when the change range of the optimized control parameters is less than a set convergence error threshold times of continuous iterations, output the optimal control parameter set.

[0086] Optionally, the S6 specifically includes:

[0087] S61, transmit the obtained optimal control parameter set to the control terminal of the metallurgical off-gas purification system in real time through a communication interface;

[0088] S62, the control terminal adjusts the temperature of the reaction cavity in the metallurgical off-gas purification system in real time according to the temperature parameter in the optimal control parameter set, and the temperature control precision is ;

[0089] S63, control the pressure in the reaction cavity in real time according to the pressure parameter in the optimal control parameter set, and the pressure control precision is ;

[0090] S64, adjust the flow rate at the off-gas inlet end in real time according to the flow rate parameter in the optimal control parameter set, and the flow rate control precision is ;

[0091] S65, control the residence time of off-gas in the reaction cavity in real time according to the reaction time parameter in the optimal control parameter set, and the reaction time control precision is ; ​

[0092] S66, real-time collection of exhaust gas state parameters after feedback regulation, periodic transmission of collected feedback data back to the control terminal, data transmission period is one time;

[0093] S67, the control terminal compares the feedback state parameters with the optimal control parameter set in real time, and when the deviation of the feedback state parameters exceeds the preset threshold range, the optimization strategy is re-executed, and the optimal control parameter set is updated.

[0094] The beneficial effects of the present application are:

[0095] (1) The present application can realize high-precision real-time prediction of state parameters such as exhaust gas composition concentration, temperature, pressure, flow rate and reaction time by using a neural operator model for real-time multi-scale dynamic prediction of the state of the exhaust gas system, 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 under complex production conditions and sudden changes in the environment.

[0096] (2) The present application can realize fast and accurate optimization of the optimization control parameters of the exhaust gas purification system by designing and introducing an improved FOX optimization algorithm, significantly improving the optimization efficiency and global optimization ability of the exhaust gas purification system, and showing better real-time responsiveness and stability in complex and dynamic metallurgical production conditions.

[0097] (3) In the real-time optimization and dynamic correction of the control parameters of the exhaust gas purification system, the present application implements a self-feedback reinforcement optimization strategy, effectively solving the problem of mutual fragmentation of prediction and optimization in the prior art, which cannot realize real-time linkage control, breaking through the bottleneck of slow convergence speed and easy falling into local optimum in traditional exhaust gas purification control parameter optimization, realizing fast dynamic iterative optimization of the control parameters of the exhaust gas purification system, thereby effectively improving the technical level and practical engineering application ability in the field of metallurgical exhaust gas purification. BRIEF DESCRIPTION OF DRAWINGS

[0098] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0099] Fig. 1 A neural operator model structure diagram of a metallurgical exhaust gas purification system prediction and optimization method based on artificial intelligence proposed by the present application;

[0100] Fig. 2 A neural operator model structure diagram of a metallurgical exhaust gas purification system prediction and optimization method based on artificial intelligence proposed by the present application;

[0101] Fig. 3An improved FOX optimization algorithm flow chart of a metallurgical waste gas purification system prediction and optimization method based on artificial intelligence is proposed. DETAILED DESCRIPTION

[0102] The application will now be described in further detail with reference to the drawings. These drawings are simplified schematic diagrams which show the basic structure of the application in a schematic manner only, and thus only show the components relevant to the application.

[0103] REFERENCE Figs. 1-3 A metallurgical waste gas purification system prediction and optimization method based on a neural operator combined with an improved FOX optimization algorithm, comprising the following steps:

[0104] S1, real-time collection of waste gas composition concentration, temperature, pressure, flow rate and reaction time data during the running process of the metallurgical waste gas purification system, establishment of an initial multi-dimensional data sample set;

[0105] S2, adaptive preprocessing of the initial multi-dimensional data sample set, including data anomaly rejection, data denoising, data fusion and multi-scale feature extraction, to obtain a standardized input data set;

[0106] S3, input of the standardized input data set into a neural operator model for real-time multi-scale dynamic modeling, to obtain real-time prediction features of the metallurgical waste gas system state;

[0107] S4, use of the improved FOX optimization algorithm, with the real-time prediction features as input, combined with an adaptive dynamic search strategy to perform real-time multi-parameter collaborative optimization, to obtain preliminary optimization control parameters of the waste gas purification system;

[0108] S5, implementation of a self-feedback reinforcement optimization strategy on the preliminary optimization control parameters, real-time dynamic iterative optimization, to obtain an optimal control parameter set;

[0109] S6, real-time feedback of the optimal control parameter set to the metallurgical waste gas purification system, and control of the metallurgical waste gas purification system to perform closed-loop regulation and control according to the optimal control parameter set.

[0110] The specific implementation mode of real-time collection of waste gas component concentration, temperature, pressure, flow rate and reaction time data in the metallurgical waste gas purification system of the application is that a gas component sensor is installed at the waste gas inlet end, the sampling frequency of the gas component sensor is set to 1 second; the temperature data is collected by a high-precision thermocouple temperature sensor, the installation position is set at the waste gas inlet end and the inside of the reaction cavity, and the sampling accuracy of the temperature sensor is ±0.1℃; the pressure data is collected in real time by a pressure sensor installed in the inside of the reaction cavity, and the measurement range is set to 0.1MPa to 2.0MPa; the waste gas flow rate data is collected in real time by a flow meter installed on the gas inlet pipeline, and the measurement range is set to 100m³ / h to 5000m³ / h; the reaction time data is obtained by recording the time stamp at the start time and the end time of each waste gas purification process, and the time stamp recording accuracy is set to ±0.5 seconds. The data collected by the above sensors are all based on the time stamp at the start time of the reaction, and after strict time axis alignment processing, the initial multi-dimensional data sample set is formed, and further through the 3σ criterion, wavelet threshold denoising, principal component analysis and empirical mode decomposition method, the data preprocessing and multi-scale feature extraction are realized, and finally the standardized input data set is obtained.

[0111] The metallurgical waste gas purification system prediction and optimization method based on the neural operator combined with the improved FOX optimization algorithm provided by the application can realize real-time and accurate waste gas system state prediction, dynamic collaborative parameter optimization and closed-loop feedback regulation, so that the system parameters are always in a dynamic optimization state, thereby significantly improving the waste gas purification treatment efficiency and the stability of system operation, reducing the overall operation cost, and having higher prediction accuracy and stronger adaptability under complex working conditions compared with the prior art, and having obvious technical advantages.

[0112] In the embodiment, the S1 specifically includes:

[0113] S11, a gas component sensor is installed at the waste gas inlet end of the metallurgical waste gas purification system to collect the concentration data of sulfur dioxide, carbon monoxide and nitrogen oxides in real time, and the sampling frequency is 1 second;

[0114] S12, high-precision temperature sensors are installed at the waste gas inlet end and the inside of the reaction cavity of the metallurgical waste gas purification system to collect the inlet gas temperature and the temperature data in the reaction cavity in real time, and the temperature sampling accuracy is ;

[0115] S13, a pressure sensor is installed in the inside of the reaction cavity of the metallurgical waste gas purification system to collect the cavity pressure data in real time, and the pressure measurement range is ;

[0116] S14, a flow meter is installed on the gas inlet pipeline of the metallurgical waste gas purification system to collect the waste gas flow rate data in real time, and the flow rate measurement range is ;

[0117] S15, the start time and end time of each purification process of the metallurgical waste gas purification system are time stamped respectively, reaction time data is recorded, and the precision of the reaction time is ;

[0118] S16, based on the time stamp of the start time of the reaction process, the real-time collected waste gas component concentration, temperature, pressure and flow data are time axis aligned according to the time stamp, and are matched one by one into the corresponding reaction time period, forming a multi-dimensional data sequence under a unified time reference;

[0119] S17, according to the multi-dimensional data sequence, the data is spliced and fused based on the time stamp as the association index, and an initial multi-dimensional data sample set with unified time stamp and clear internal data structure is constructed.

[0120] In the implementation process of the present application, the real-time collection of waste gas component concentration is realized by installing a gas component sensor at the waste gas inlet end of the metallurgical waste gas purification system. The sensor type is selected as an electrochemical sensor, wherein the measurement range of sulfur dioxide is set to 0-2000 ppm, the measurement range of carbon monoxide is set to 0-1000 ppm, and the measurement range of nitrogen oxides is set to 0-500 ppm. The high-precision temperature sensor used is a K-type thermocouple sensor, and the temperature measurement range is 0-1200 DEG C. The pressure sensor is selected as a piezoresistive pressure sensor, and the accuracy level is 0.1 level. The flow meter adopts a thermal mass flow meter, and the flow measurement accuracy is ±1% of the full scale. The time stamp marking is realized by a high-precision synchronous clock provided by the system, and the clock accuracy is millisecond level. The collected multi-dimensional data is transmitted to the central data processing unit in real time through industrial Ethernet for unified time axis alignment processing and multi-dimensional data fusion.

[0121] The present application collects and constructs the initial multi-dimensional data sample set in the waste gas treatment process in real time and accurately, ensures the basic data quality of subsequent data processing and optimization work, significantly improves the accuracy and stability of waste gas treatment state prediction and optimization, provides reliable data support for real-time dynamic optimization control of the metallurgical waste gas purification system, and significantly improves the precision and response efficiency of system operation compared with the prior art.

[0122] In the present embodiment, the S2 specifically comprises:

[0123] S21, taking the mean value of each data sequence in the initial multi-dimensional data sample set as the reference, the abnormal data is identified and removed based on the criterion, wherein is the standard deviation of each data sequence;

[0124] S22, the data sequence after the abnormal data is eliminated, a wavelet threshold denoising method is used for data denoising processing, wherein the wavelet base function is Daubechies wavelet, and the decomposition scale is set to layer;

[0125] S23, the normalized processing is carried out on the multi-dimensional data sequence after denoising, and the data normalization processing range is ;

[0126] S24, data fusion is carried out on the data sequence after normalization processing, and the principal component analysis method is used for data fusion, and the principal component cumulative contribution rate is set to ;

[0127] S25, multi-scale feature extraction is carried out on the sequence after data fusion, and the empirical mode decomposition method is used to decompose the fused data sequence into intrinsic mode function step by step, and the decomposition order of each data sequence is set to order;

[0128] S26, the intrinsic mode functions of each order are spliced in order from large to small scale, and the standardized input data set is formed.

[0129] The specific method of abnormal data recognition and elimination in the implementation process of the application is as follows: after calculating the mean and standard deviation of the initial data sequence one by one, the difference between each data point in the data sequence and the corresponding sequence mean is used to determine the abnormal data, and the data point whose difference is greater than 3 times the standard deviation is identified as abnormal data and directly eliminated; the wavelet threshold denoising processing adopts soft threshold denoising mode, and the threshold size is the median of the absolute value of each layer decomposition coefficient multiplied by the logarithmic function calculation; the data fusion of principal component analysis adopts the covariance matrix calculation method, and the eigenvalue and corresponding eigenvector are solved through the covariance matrix, and the principal components with cumulative contribution rate above 90% are selected for fusion.

[0130] The application effectively removes abnormal noise in the data and extracts key features with high discrimination by comprehensively applying 3σ criterion, wavelet threshold denoising and principal component analysis method, significantly improves the precision and robustness of data input neural operator model for multi-scale dynamic modeling, and effectively improves the reliability and stability of real-time prediction of metallurgical waste gas purification system state.

[0131] The neural operator model specifically includes:

[0132] The input mapping layer maps the standardized input data set to a high-dimensional feature space with a dimension of by using a fully connected neural network;

[0133] The Fourier operator layer performs frequency domain transformation on the mapped high-dimensional feature space data by using fast Fourier transform;

[0134] Multi-scale frequency domain filtering layers pass through a number of scales in the frequency domain space. Frequency domain filters extract low frequencies respectively Mid-low frequency , intermediate frequency Medium and high frequency ,high frequency Multi-scale frequency domain characteristics;

[0135] The inverse Fourier transform layer performs inverse fast Fourier transform on the multi-scale frequency domain features after frequency domain filtering, mapping the frequency domain features back to the time domain to obtain the corresponding multi-scale time domain features.

[0136] The feature fusion layer concatenates multi-scale temporal features and fuses them through a fully connected network. The output dimension after feature fusion is [dimension not specified]. ;

[0137] The state prediction output layer outputs real-time predictive features of the state of the metallurgical waste gas purification system through a fully connected network.

[0138] In this embodiment, S3 specifically includes:

[0139] S31. Use a fully connected neural network to map the standardized input dataset to a dimension of . High-dimensional feature space, to obtain high-dimensional feature data;

[0140] S32. Perform Fast Fourier Transform on high-dimensional feature data:

[0141] ;

[0142] in, Indicates the first Frequency domain feature data corresponding to each frequency component Represents the first in high-dimensional feature data Data values ​​of each time-domain sampling point Indicates the total length of the data sequence. This is a frequency index, with a value range of [value range missing]. , This is a time-domain index, with a value range of [value range missing]. , Represents the imaginary unit;

[0143] S33. The number of scales used for frequency domain feature data is... The frequency domain filter extracts frequencies in the following ranges: , , , and Multi-scale frequency domain characteristics;

[0144] S34. Perform inverse fast Fourier transform on the multi-scale frequency domain features respectively:

[0145] ;

[0146] in, Indicates the first The temporal feature data values ​​corresponding to each temporal sampling point Represents the first in the frequency domain feature data Data values ​​of each frequency component Indicates the total length of the data sequence. This is a time-domain index, with a value range of [value range missing]. , This is a frequency index, with a value range of [value range missing]. , Represents the imaginary unit;

[0147] S35. The obtained multi-scale temporal features are concatenated in scale order and then input into a fully connected network for fusion processing. The fused feature dimension is: ;

[0148] S36. The fused features are output as real-time predictive features of the metallurgical waste gas purification system status through a fully connected network.

[0149] In the implementation of this invention, the specific method for obtaining real-time prediction features includes: using a fully connected neural network to map the standardized input dataset to a high-dimensional feature space of 128256 dimensions; extracting multi-scale frequency domain features with frequency ranges of 0.01Hz, 0.1-0.5Hz, 0.51Hz, 12Hz, and 25Hz respectively using a frequency domain filter with 46 scales; performing inverse fast Fourier transform on the multi-scale frequency domain features to obtain multi-scale time domain features; then concatenating the obtained multi-scale time domain features in order of scale size and inputting them into a fully connected network for fusion processing; finally obtaining real-time prediction features through the state prediction output layer. The improved FOX optimization algorithm, when implemented, sets the initial population size at 2050 individuals and includes four parameters: temperature (300-800℃), pressure (0.1-2.0 MPa), flow rate (100-5000 m³ / h), and reaction time (10-120 s). Each parameter corresponds to a specific numerical range. The search process alternates between Lévy fly-through random search, Gaussian perturbation local search, adaptive dynamic search, and elite retention mechanism to ensure reasonable switching between global and local searches. The convergence condition for parameter optimization iteration is that the fitness value change is less than 10⁻ for 2050 consecutive iterations. 5The feedback reinforcement optimization strategy dynamically calculates the feedback gain coefficient and corrects the control parameters in real time during the implementation process, the state error is calculated by the difference between the actual state data and the predicted state data, the error total value is quantified by using a weighted absolute error function, the control parameter correction is updated by using the gradient of the error function, and the iteration optimization termination condition is set as the change of the fitness value being less than 10⁻ 4 ~10⁻ 6 .

[0150] 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 cavity are accurately adjusted in real time through the control terminal, and the feedback state parameters are collected in real time; when the feedback state parameters deviate from the optimal control parameters by more than a preset threshold, the control terminal automatically re-executes the optimization strategy for parameter updating.

[0151] The present application realizes accurate real-time prediction and multi-parameter collaborative optimization control of the state parameters of the metallurgical waste gas purification system by combining the neural operator with the improved FOX optimization algorithm, significantly improves the state prediction accuracy and optimization control response speed in the waste gas purification process, enhances the stability and robustness of the dynamic adjustment of the waste gas purification process parameters, effectively improves the waste gas treatment efficiency and the accuracy of the process parameter optimization, and significantly reduces the system operation cost.

[0152] The improved FOX optimization algorithm specifically includes:

[0153] The algorithm initialization step sets the population size as Each individual corresponds to a set of waste gas purification control parameters, wherein the temperature range is , the pressure range is , the flow range is , and the reaction time range is ;

[0154] The adaptive dynamic search step calculates the population fitness value, takes the comprehensive index of waste gas purification efficiency and economic cost as the target, and dynamically adjusts the search range and step length according to the current population fitness variance;

[0155] The Levy flight search step performs random walk search, and the random flight step length obeys the Levy distribution with the parameter range of ;

[0156] The Gaussian disturbance local search step implements Gaussian disturbance with a standard deviation of the current parameter value in the local search stage;

[0157] The global and local search switching step switches the search mode through a probability mechanism, and the initial switching probability is set as , linearly decreases with iteration number;

[0158] an elitist strategy and an iteration update step, the top individuals in fitness are reserved after each iteration, and the population is updated to the next generation;

[0159] a convergence determination step, when the fitness value of the global optimal individual changes less than in the last iterations, the optimal control parameters are output.

[0160] In the embodiment, the S4 specifically comprises:

[0161] S41, an initial population is established with real-time prediction characteristics as input, the population size is individuals, each individual contains control parameters of temperature, pressure, flow rate and reaction time, the temperature value range is , the pressure value range is , the flow rate value range is , and the reaction time value range is ;

[0162] S42, the fitness value of each individual in the initial population is calculated respectively, and the fitness value is defined as a comprehensive evaluation index of exhaust gas purification efficiency and economic cost;

[0163] S43, the search range and the step size are adaptively adjusted according to the population fitness value and the variance size obtained by calculation, the search range adjustment ratio is of the initial range, and the step size adjustment ratio is of the initial step size;

[0164] S44, the population individual position is updated by using a Levy flight random search strategy, the Levy flight step size is subject to a Levy distribution, and the characteristic parameter of the Levy distribution is ; ;

[0165] S45, the population individual parameters obtained by searching are subjected to Gaussian disturbance local search, and the disturbance standard deviation is of the current parameter value of the individual;

[0166] S46, the initial value of the switching probability between global search and local search is set to , and linearly decreases with iteration number, the population is updated and the elitist strategy is executed, and the top individuals in fitness are reserved after each iteration;

[0167] S47, in the last iterations, the fitness value of the global optimal individual changes less than the convergence threshold The iteration is terminated, and the preliminary optimization control parameter of the exhaust gas purification system is obtained.

[0168] In the embodiment, to realize the metallurgical exhaust gas purification system prediction and optimization method of the improved FOX optimization algorithm combined with the neural operator, consistency and synchronism of real-time data acquisition and processing need to be ensured in actual deployment. The sensor of the data acquisition end should be a gas composition sensor, a high-precision temperature sensor, a pressure sensor and a flow meter, and be fixedly installed at key positions such as the metallurgical exhaust gas inlet, the reaction cavity and the inlet pipeline. The full connection network structure of the neural operator model needs to be determined as .

[0169] The neural operator model and the improved FOX optimization algorithm are combined to accurately and timely realize the operation state prediction and optimization control parameter collaborative optimization of the metallurgical exhaust gas purification system. Compared with the traditional technology, the present application can effectively reduce the state prediction error, improve the accuracy and convergence speed of the optimization control parameter, significantly improve the operation efficiency and stability of the metallurgical exhaust gas purification process, and has stronger adaptability to complex working conditions.

[0170] The self-feedback reinforcement optimization strategy specifically includes:

[0171] The preliminary optimization control parameter is obtained by using the improved FOX optimization algorithm with the real-time predicted state characteristics of the exhaust gas system as input;

[0172] Real-time state data acquisition, the preliminary optimization control parameter is input into the metallurgical exhaust gas purification system, and the actual state data of the metallurgical exhaust gas purification system is collected in real time;

[0173] State error calculation, the error function is calculated as:

[0174] ;

[0175] Wherein, represents the total state error value at the th moment, represents the value of the th actual measured state parameter of the metallurgical exhaust gas purification system, represents the value of the th predicted state parameter, represents the weight coefficient of the th state parameter, denotes the total number of state parameters;

[0176] Adaptive feedback gain calculation, dynamically calculating the feedback gain coefficient:

[0177] ;

[0178] wherein, denotes the dynamic feedback gain coefficient, denotes the initial feedback gain coefficient, denotes the gain adjustment factor, and the value range is , denotes the allowable error threshold;

[0179] Control parameter correction, real-time correction of the control parameter according to the dynamic feedback gain coefficient:

[0180] ;

[0181] wherein, denotes the next time optimization control parameter, denotes the current time optimization control parameter, denotes the gradient of the error function relative to the current parameter, and is defined as:

[0182] ;

[0183] wherein, denotes the value of the th control parameter at the current time, denotes the total number of control parameters;

[0184] Iteration optimization termination, when the global optimal individual fitness value changes less than the set convergence error threshold times in succession, stop iteration optimization, and obtain the final optimization control parameter.

[0185] In the embodiment, the S5 specifically comprises:

[0186] S51, input the preliminary optimization control parameter to the self-feedback reinforcement optimization strategy;

[0187] S52, real-time collection of metallurgical waste gas purification system actual operation state data;

[0188] S53, calculating the state error value according to the real-time state data;

[0189] S54, dynamically updating the optimization control parameter according to the state error value;

[0190] S55, multiple dynamic iteration optimization of the updated optimization control parameter;

[0191] ​S56, when the optimization control parameter changes less than the set convergence error threshold value in 20 to 50 iteration processes continuously, output the optimal control parameter set.

[0192] In the embodiment, various sensors for collecting waste gas data in real time are connected with the control terminal through an industrial communication interface, a Modbus RTU protocol is used as a communication protocol, and a transmission rate is set as 9600 bps. A high-precision temperature sensor is a platinum resistance sensor, a pressure sensor is a strain pressure sensor, and a flow meter is a vortex flow meter. All sensors are strictly installed at the waste gas inlet end or inside the reaction cavity to ensure data accuracy.

[0193] The present application realizes accurate prediction and real-time optimization control of the running state of the metallurgical waste gas purification system by collecting data such as waste gas composition, temperature, pressure, flow rate and the like in real time, using a neural operator and an improved FOX optimization algorithm to construct a dynamic prediction and multi-parameter collaborative optimization mechanism, and introducing a self-feedback reinforcement optimization strategy, effectively improving waste gas purification efficiency, reducing system operation cost, significantly improving the problems of control delay and poor parameter adaptability existing in the traditional method, and improving the intelligent level of waste gas purification in the metallurgical industrial production process.

[0194] In the embodiment, the S6 specifically comprises:

[0195] S61, the obtained optimal control parameter set is transmitted to the control terminal of the metallurgical waste gas purification system in real time through a communication interface;

[0196] S62, the control terminal adjusts the temperature of the reaction cavity in the metallurgical waste gas purification system in real time according to the temperature parameter in the optimal control parameter set, and the temperature control precision is ;

[0197] S63, the pressure in the reaction cavity is controlled in real time according to the pressure parameter in the optimal control parameter set, and the pressure control precision is ;

[0198] S64, the flow rate at the waste gas inlet end is adjusted in real time according to the flow rate parameter in the optimal control parameter set, and the flow rate control precision is ;

[0199] S65, the reaction time in the reaction cavity is controlled in real time according to the reaction time parameter in the optimal control parameter set, and the reaction time control precision is ;

[0200] S66, the waste gas state parameter after feedback regulation is collected in real time, and the collected feedback data is transmitted back to the control terminal periodically, and the data transmission period is once;

[0201] ​S67, the control terminal compares the feedback state parameter with the optimal control parameter set in real time, and when the deviation of the feedback state parameter exceeds the preset threshold range, the optimization strategy is re-executed, and the optimal control parameter set is updated.

[0202] In the embodiment, the control terminal adopts an industrial-grade embedded controller, receives the optimal control parameter set in real time through an RS-485 communication interface, and implements accurate regulation and control of temperature, pressure, flow rate and reaction time by built-in PID closed-loop control algorithm, wherein the proportional parameter (P) of the PID control algorithm is in the range of 2.0-5.0, the integral parameter (I) is in the range of 0.01-0.05, and the differential parameter (D) is in the range of 0.1-0.5, so as to ensure accurate and controllable system operation state. Real-time feedback data is collected by a collection module in the control terminal, the collection module is built-in 12-bit analog-to-digital converter (ADC), the sampling frequency is greater than or equal to 1 Hz, and real-time data is transmitted to an industrial server through a communication protocol, so as to ensure real-time and accuracy of data transmission.

[0203] The present application realizes accurate control of temperature, pressure, flow rate and reaction time in the metallurgical waste gas purification process by implementing real-time closed-loop regulation and control on the optimal control parameter set, has fast system response speed and high control precision, the real-time feedback mechanism can quickly detect and correct system deviation, effectively ensures stability and optimization effect of the waste gas purification process, and significantly improves real-time and reliability of the metallurgical waste gas purification system.

[0204] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and inventive concept of the present application within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. A metallurgical off-gas purification system prediction and optimization method based on a neural operator combined with an improved FOX optimization algorithm, characterized in that, The method comprises the following steps: S1, collecting in real time the concentration, temperature, pressure, flow and reaction time data of the metallurgical waste gas during the operation of the metallurgical waste gas purification system, and establishing an initial multi-dimensional data sample set; S2, adaptively preprocessing the initial multi-dimensional data sample set, including data anomaly elimination, 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 for real-time multi-scale dynamic modeling to obtain real-time prediction features of the metallurgical waste gas system state; S4, using an improved FOX optimization algorithm, taking the real-time prediction features as input, and combining an adaptive dynamic search strategy to perform real-time multi-parameter collaborative optimization to obtain preliminary optimization control parameters of the waste gas purification system; S5, implementing a self-feedback reinforcement optimization strategy on the preliminary optimization control parameters to perform real-time dynamic iterative optimization to obtain an optimal control parameter set; S6, feeding back the optimal control parameter set to the metallurgical waste gas purification system in real time, and controlling the metallurgical waste gas purification system to perform closed-loop regulation and control according to the optimal control parameter set; The improved FOX optimization algorithm specifically comprises: The algorithm initialization step sets the population size as Each individual corresponds to a set of exhaust purification control parameters, wherein the temperature range is , the pressure range is , the flow range is , and the reaction time range is ; An adaptive dynamic search step, which dynamically adjusts the search range and step size according to the current population fitness variance by calculating the population fitness value, taking the waste gas purification efficiency and economic cost comprehensive index as the target; The Levy flight search step performs a random walk search with random flight steps that follow a Levy distribution with parameter range 0 < a < 2. a Gaussian perturbation local search step, in which the individual is subjected to a Gaussian perturbation with a standard deviation of the current parameter value in the local search phase. The global and local search switching step switches the search mode through a probability mechanism, and the initial switching probability is set as , and linearly decreases with the iteration number. Elite retention and iterative update steps: retain the top fitness rankings after each iteration. Individuals, renewing the population to the next generation; Convergence determination step, when the fitness value of the global optimal individual in the continuous iteration process is less than the fitness value of the global optimal individual in the previous iteration process, output the optimal control parameter.

2. The method according to claim 1, wherein the method is characterized by, The S1 specifically comprises: S11, at the exhaust gas inlet end of the metallurgical waste gas purification system, install a gas component sensor to collect the concentration data of sulfur dioxide, carbon monoxide and nitrogen oxides in real time, the sampling frequency is one time; S12, high-precision temperature sensors are respectively installed at the exhaust gas inlet end of the metallurgical exhaust gas purification system and in the reaction cavity to collect inlet gas temperature and reaction cavity temperature data in real time, with a temperature sampling precision of ; S13, install a pressure sensor inside the reaction cavity of the metallurgical waste gas purification system to collect cavity pressure data in real time, with a pressure measurement range of ; S14, install a flow meter at the gas inlet pipeline of the metallurgical waste gas purification system to collect waste gas flow data in real time, and the flow measurement range is ; S15, the start time and the end time of each purification process of the metallurgical waste gas purification system are respectively time-stamped, reaction time data is recorded, and the precision of the reaction time is ; S16, taking the timestamp at the beginning of the reaction process as the reference, aligning the real-time collected waste gas composition concentration, temperature, pressure and flow data on the time axis according to the timestamp, and matching them one by one to the corresponding reaction time period to form a multi-dimensional data sequence under a unified time reference; S17, according to the multi-dimensional data sequence, the data is spliced and fused with the timestamp as the association index to build an initial multi-dimensional data sample set with unified timestamp and clear internal data structure.

3. The method according to claim 1, wherein the method is characterized by, The S2 specifically comprises: S21, based on the mean of each data sequence in the initial multi-dimensional data sample set, adopting the criterion based on to identify and eliminate abnormal data, wherein is the standard deviation of each data sequence. S22, the data sequence after eliminating abnormal data is processed by wavelet threshold denoising method, wherein the wavelet base function is Daubechies wavelet, and the decomposition scale is set as layer; S23, the normalized processing is applied to the de-noised multi-dimensional data sequence, and the normalized processing range of the data is ; S24, data fusion is performed on the normalized data sequence, the principal component analysis method is adopted for data fusion, and the principal component cumulative contribution rate is set to ; S25, multi-scale feature extraction is performed on the sequence after data fusion, and an empirical mode decomposition method is used to decompose the fused data sequence into intrinsic mode functions step by step, and the decomposition order of each data sequence is set to order; S26, splicing the intrinsic mode functions of each order in order of scale from large to small to form a standardized input data set.

4. The method according to claim 1, wherein the method is characterized by, The neural operator model specifically comprises: An input mapping layer maps the standardized input dataset to a high-dimensional feature space with dimensionality using a fully connected neural network. A Fourier operator layer that performs frequency domain transformation on the mapped high-dimensional feature space data using fast Fourier transform; a multi-scale frequency domain filtering layer extracts multi-scale frequency domain features of low frequency , middle-low frequency , middle frequency , middle-high frequency , and high frequency , respectively through a frequency domain filter with a scale number of ​ An inverse Fourier transform layer that performs inverse fast Fourier transform on the multi-scale frequency domain features filtered in the frequency domain to map the frequency domain features back to the time domain and obtain corresponding multi-scale time domain features; The feature fusion layer splices the multi-scale time domain features, performs fusion processing through a full connection network, and the output dimension after feature fusion is ; A state prediction output layer that outputs real-time prediction features of the metallurgical waste gas purification system state through a fully connected network.

5. The method for prediction and optimization of metallurgical off-gas cleaning system based on neural operator combined with improved FOX optimization algorithm as claimed in claim 1, wherein, The S3 specifically comprises: S31. Use a fully connected neural network to map the standardized input dataset to a dimension of . High-dimensional feature space, to obtain high-dimensional feature data; S32, performing fast Fourier transform on the high-dimensional feature data: ； wherein, represents the frequency-domain feature data corresponding to the th frequency component, represents the data value of the th time-domain sampling point in the high-dimensional feature data, represents the total length of the data sequence, is a frequency index, and takes a value in the range of , is a time-domain index, and takes a value in the range of , represents an imaginary unit; S33, a frequency domain filter with a scale number of is adopted on the frequency domain feature data, and multi-scale frequency domain features with frequency ranges of , , , and are extracted. S34, performing inverse fast Fourier transform on the multi-scale frequency domain features: ； wherein, denotes a time-domain feature data value corresponding to the th time-domain sample, denotes a data value of the th frequency component in the frequency-domain feature data, denotes the total length of the data sequence, is a time-domain index, and has a value range of , is a frequency index, and has a value range of , denotes an imaginary unit; S35, the obtained multi-scale time domain features are spliced in order of scale size and input to a full connection network for fusion processing, and the dimension of the fused features is ; S36, outputting the real-time prediction features of the metallurgical waste gas purification system state through a fully connected network.

6. The method for prediction and optimization of metallurgical off-gas cleaning system based on neural operator combined with improved FOX optimization algorithm as claimed in claim 1 wherein, The S4 specifically comprises: S41, inputting the real-time prediction features, establishing an initial population, the population size is individuals, each individual containing temperature, pressure, flow rate and reaction time control parameters, the temperature value range is , the pressure value range is , the flow rate value range is , and the reaction time value range is ; S42, calculating the fitness value of each individual in the initial population, and the fitness value is defined as the comprehensive evaluation index of waste gas purification efficiency and economic cost; S43, according to the population fitness value and the variance size obtained by calculation, adaptively adjusting the search range and the step, the search range adjustment ratio is initial range , and the step adjustment ratio is initial step S44, adopt the Levy flight random search strategy to update the population individual position, the Levy flight step length obeys the Levy distribution, the characteristic parameter of the Levy distribution is in the range of ; S45, Gaussian disturbance local search is performed on the population individual parameter obtained by the search, and the disturbance standard deviation is the current parameter value of the individual ; S46, set the switching probability initial value between global search and local search as , and linearly decrease with the iteration number, update the population and perform the elite reservation strategy, reserve the top individuals in fitness after each iteration. S47、in succession In the sub-iteration process, the fitness value of the global optimal individual changes less than the convergence threshold The iteration is terminated when the time is up, and the preliminary optimized control parameters of the exhaust purification system are obtained.

7. The method for prediction and optimization of metallurgical off-gas cleaning system based on neural operator combined with improved FOX optimization algorithm as claimed in claim 1 wherein, The self-feedback reinforcement optimization strategy specifically comprises: The preliminary optimization control parameter is obtained, and a neural operator model is used to predict the state characteristics of the exhaust system in real time. An improved FOX optimization algorithm is used to obtain the preliminary optimization control parameter; Real-time state data acquisition, inputting the preliminary optimization control parameter into the metallurgical exhaust gas purification system, and collecting the actual state data of the metallurgical exhaust gas purification system in real time; State error calculation, calculating the error function: ； wherein represents the number of the state parameter, represents the total value of the state error at the time t, represents the value of the actual measured state parameter of the metallurgical off-gas purification system, represents the value of the predicted state parameter, represents the weight coefficient of the state parameter, represents the total number of state parameters; Adaptive feedback gain calculation, dynamically calculating the feedback gain coefficient: ； wherein, represents a dynamic feedback gain coefficient, represents an initial feedback gain coefficient, represents a gain adjustment factor, which has a value range of , represents an allowable error threshold value; Control parameter correction, correcting the control parameter in real time according to the dynamic feedback gain coefficient: ； wherein, denotes the next time optimized control parameter, denotes the current time optimized control parameter, denotes the gradient of the error function with respect to the current parameter, defined as: ； in, Indicates the current time. The value of each control parameter, Indicates the total number of control parameters; The iterative optimization is terminated when the fitness value of the global optimal individual in the successive iteration is less than a set convergence error threshold the set convergence error threshold the final optimized control parameters are obtained.

8. The method according to claim 1, wherein the method is characterized by, The S5 specifically includes: S51, inputting the preliminary optimization control parameter into the self-feedback reinforcement optimization strategy; S52, collecting the actual operation state data of the metallurgical exhaust gas purification system in real time; S53, calculating the state error value according to the real-time state data; S54, dynamically updating the optimization control parameter according to the state error value; S55, performing multiple dynamic iterative optimizations on the updated optimization control parameter; S56, output the optimal control parameter set when the variation range of the optimization control parameter is less than the set convergence error threshold value in 20 to 50 iteration processes in succession times.

9. The method of claim 1, wherein the method is a method of predicting and optimizing a metallurgical off-gas cleaning system based on a neural operator combined with an improved FOX optimization algorithm. The S6 specifically includes: S61, transmitting the obtained optimal control parameter set to the control terminal of the metallurgical exhaust gas purification system in real time through a communication interface; S62, the terminal controls the temperature parameter in the optimal control parameter set to adjust the temperature of the reaction cavity in the metallurgical waste gas purification system in real time, and the temperature control precision is ; S63, according to the optimal control parameter set pressure parameter, real-time control of the pressure in the reaction cavity, the pressure control precision is ; S64, according to the optimal control parameter set flow parameter, real-time adjust the exhaust gas inlet end flow, flow control precision is ; S65, according to the optimal control parameter set, the reaction time parameter, the exhaust gas residence time in the reaction cavity is controlled in real time, and the reaction time control precision is ; S66, collecting the exhaust gas state parameters after feedback regulation in real time, periodically transmitting the collected feedback data back to the control terminal, and the data transmission period is one time; S67, the control terminal compares the feedback state parameter with the optimal control parameter set in real time, and when the deviation of the feedback state parameter exceeds the preset threshold range, the optimization strategy is re-executed, and the optimal control parameter set is updated.

Images