Hazardous gas removal method and system based on special gas supply equipment

Abstract

This invention discloses a method and system for removing hazardous gases from a special gas supply equipment, relating to the field of hazardous gas removal technology. The method includes: numerically simulating the gas flow path inside the equipment to identify hazardous gas retention areas; applying a magnetoacoustic coupling field to the retention areas to desorb the hazardous gases and migrating the desorbed gases to a preset safe treatment area; extracting and fusing optical and spectral features using a learnable encoder to obtain the physicochemical characteristics of the residual hazardous gases; constructing a digital twin model to simulate the residual diffusion behavior and retention risk of the residual hazardous gases within the equipment; optimizing the parameters of the magnetoacoustic coupling field and iteratively optimizing it until the hazardous gases are completely removed from the equipment. This invention improves the efficiency and directionality of hazardous gas removal, reduces the safety risks caused by residues, and thus ensures the stability of the special gas supply equipment operation and the safety of the operating environment.

Description

Technical Field

[0001] This invention relates to the field of hazardous gas removal technology, and more specifically, to a method and system for removing hazardous gases based on a special gas supply device. Background Technology

[0002] In high-precision industrial fields such as semiconductor manufacturing, microelectronic device processing, and advanced material synthesis, specialty gases are widely used as key process media in core processes such as etching, deposition, and doping. Specialty gas supply equipment, as a core device ensuring process stability and safety, often experiences gas residue, adsorption, or localized concentration accumulation, leading to hazardous gas stagnation areas and subsequently causing cross-contamination, equipment corrosion, and even safety accidents.

[0003] Currently, traditional methods for removing hazardous gases mainly rely on continuous ventilation, inert gas flushing, or high-temperature desorption. These methods suffer from low emission efficiency, high energy consumption, and difficulty in precise control. With the increasing complexity of microscale structures and the diversification of gas types, traditional technologies can no longer meet the demands for efficient, safe, and intelligent gas removal. Furthermore, existing technologies generally lack the ability to dynamically sense and track the spatial distribution of residual gases, making it difficult to predict and actively control gas retention paths. This results in low removal efficiency, uncontrollable processes, and potential secondary pollution and safety hazards.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of the problems in related technologies, the present invention proposes a method and system for removing hazardous gases based on special gas supply equipment, so as to overcome the above-mentioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for removing hazardous gases based on a special gas supply device is provided, the method comprising: S1. Collect multi-dimensional status data during equipment operation, and use a pre-built intelligent prediction model to perform numerical simulation of the gas flow path inside the equipment to identify the stagnation area of ​​hazardous gases. S2. Apply a magnetoacoustic coupling field to the retention area to desorb the hazardous gas. Use micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, determine the gas transport path, and migrate the desorbed hazardous gas to the preset safe treatment area. S3. Collect optical response data and terahertz spectrum data from the equipment after migration processing, and use a learnable encoder to extract and fuse optical and spectral features to obtain the physicochemical characteristics of residual hazardous gases. S4. Based on the obtained physicochemical characteristics of residual hazardous gases, construct a digital twin model to simulate the residual diffusion behavior and retention risk of residual hazardous gases in the equipment; S5. Based on the simulation results of the digital twin, optimize the parameters of the magnetoacoustic coupling field, and optimize it iteratively until the hazardous gas is completely discharged from the equipment.

[0007] Furthermore, the step of applying a magnetoacoustic coupling field to desorb hazardous gases in the retention region, using micromagnetic simulation equations to dynamically evolve the skyrmion formation process, determining the gas transport path, and migrating the desorbed hazardous gases to a preset safe treatment area includes: S21. Apply a magnetoacoustic coupling field to the area where hazardous gases are retained, and desorb the hazardous gases inside the equipment through the synergistic effect of the magnetic field and ultrasound. S22. Construct a spatial distribution model of the magnetoacoustic coupling field, and use micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path. S23. Using acoustic pressure wave focusing technology, a directional acoustic pressure gradient is established within the determined gas transport channel, and the desorbed hazardous gas is driven to migrate along a preset direction and flow into a preset safe treatment area.

[0008] Furthermore, the construction of a spatial distribution model of the magnetoacoustic coupling field, and the application of micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path, includes: S221. A spatial distribution model of the magnetoacoustic coupling field inside the equipment is constructed using the finite element analysis algorithm, and a corresponding spatial distribution prediction map is generated. S222. Based on the generated spatial distribution prediction map, the formation process of skyrmions is dynamically simulated using micromagnetic simulation equations, and the topological invariants of skyrmions are extracted to generate a skyrmion distribution feature map. S223. Based on the generated skyrmion distribution feature map, extract the feature parameters of skyrmions using a convolutional neural network, and establish a skyrmion topology array based on the feature parameters. S224. The established skymin subtopology array is mapped to a weighted topology graph structure, and the connectivity of the weighted topology graph structure is analyzed using the shortest path algorithm to extract the transport paths used to guide gas flow, which serve as transport channels for gas migration.

[0009] Furthermore, based on the generated spatial distribution prediction map, the formation process of skyrmions is dynamically simulated using micromagnetic simulation equations, and the topological invariants of skyrmions are extracted to generate a skyrmion distribution feature map, including: S2221. Based on the spatial distribution prediction map of the magnetoacoustic coupling field, the nonlinear terms in the micromagnetic equation are solved using a quantum computing simulator, the dynamic evolution simulation of the skyrmion formation process is performed, and a magnetic configuration evolution map is generated. S2222. Based on the generated magnetic configuration evolution diagram, the spatiotemporal parameter combination of the magnetic field and ultrasound is adjusted in real time using a reinforcement learning algorithm. The stability of the skyrmion topology is used as a reward function for feedback training. Through continuous iterative optimization, the final magnetic configuration state is determined. S2223. Based on the determined final magnetic configuration state, construct a hybrid neural network model and use parallel computing to extract the topological invariants of skyrmions; S2224. Use a relocation optimization algorithm based on topological invariants to identify skyrmion centers and generate a skyrmion distribution feature map containing topological feature parameters.

[0010] Furthermore, the calculation formula for solving the nonlinear terms in the micromagnetic equation using a quantum computing simulator is as follows: ; In the formula, Represents the total Hamiltonian of micromagnetism; J Indicates the strength of exchange coupling; i Indicates the grid index value; j Indicates the grid index value; Indicates the first i Spin operator vectors at magnetic lattice points; Indicates the first j Spin operator vectors at magnetic lattice points; D This represents the strength of the Dzyaloshinskii-Moriya interaction; A unit vector representing the vertical direction; β ∠ represents the magnetoelastic coupling coefficient; ∠ represents the divergence. u Represents the acoustic displacement field; The first in the vertical direction i Spin operator vectors at magnetic lattice points; γ e Indicates the electron gyromagnetic ratio; h Represents the reduced Planck constant; B mag This represents the vector of static magnetic field strength.

[0011] Furthermore, the step of identifying skymin center points using a relocation optimization algorithm based on topological invariants and generating a skymin distribution feature map containing topological feature parameters includes: S22241. Based on the topological invariants of skyrmions, the candidate centroids of skyrmions are identified using the topological association analysis algorithm. Combined with the distributed path search strategy, the shortest topological association path between each candidate centroid and the preset centroid is calculated, and the corresponding topological feature matching degree is evaluated. S22242. Based on the topological feature matching degree of the candidate center points, map the candidate center points to the particle positions in the particle swarm optimization algorithm, initialize the position and velocity of each particle, and construct a fitness function to initialize the individual optimal position and the global optimal position. S22243. Based on the initialization results, evaluate the convergence index and structural aggregation degree of the particle swarm, and update the inertial weights of the particles using the topological adaptive formula. S22244. Based on the updated inertia weight, iteratively update the particle's velocity and position, recalculate the particle's fitness, and update the individual optimal position and the global optimal position. S22245. Determine whether the fitness of the current global best particle meets the preset convergence threshold. If it does, execute S22246; otherwise, return to S22243 to continue iterative optimization. S22246. Output the position corresponding to the globally optimal particle as the target center point of the skyrmion, and generate the skyrmion distribution feature map by combining the corresponding topological invariants.

[0012] Furthermore, the optical response data and terahertz spectrum data in the acquired and processed equipment are extracted and fused using a learnable encoder to obtain the physicochemical characteristics of the residual hazardous gases, including: S31. Based on the migration results of hazardous gases, the residual hazardous gases in the equipment are detected using an integrated optical metasurface sensing structure. Based on the detection results, the corresponding optical response data is obtained, and the terahertz spectrum data of the corresponding area is collected synchronously through a terahertz transmitter. S32. Based on the optical response data and terahertz spectrum data, use a learnable encoder to extract optical features and spectral features, and construct a multi-scale feature pyramid to generate a joint representation that reflects the physical correlation. S33. Based on the generated joint characterization, a multi-task deep learning network is constructed to identify the type, concentration and spatial distribution information of residual hazardous gases, and the physicochemical characteristics of residual hazardous gases are extracted by combining physicochemical property constraints and molecular prior knowledge graphs.

[0013] Furthermore, the step of extracting optical and spectral features using a learnable encoder based on optical response data and terahertz spectral data, and constructing a multi-scale feature pyramid to generate a joint representation reflecting physical correlations includes: S321. Using a learnable encoder based on physical inspiration, optical features and spectral features are extracted from optical response data and terahertz spectrum data, respectively. S322. Based on the correlation constraints between geometric structures, the extracted optical features and spectral features are aligned in spatial structure to construct a multi-scale feature pyramid with spatial consistency. S323. Based on the constructed multi-scale feature pyramid, optical features and spectral features at different scales are fused, and the fusion results of each scale are integrated by using a gating mechanism to generate multi-scale fused features. S324. Graph neural networks are used to map multi-scale fused features to molecular graph structure space to capture the topological relationships and chemical bond connection characteristics between molecules. Combined with a pre-set knowledge base of gas physicochemical properties, a joint characterization reflecting physical correlations is output.

[0014] Furthermore, based on the constructed multi-scale feature pyramid, optical and spectral features at different scales are fused, and a gating mechanism is used to integrate the fusion results at each scale to generate multi-scale fused features, including: S3231. Based on the constructed multi-scale feature pyramid, a cross-modal interactive attention mechanism is used to fuse optical features and spectral features layer by layer, and the absorption peak position information of gas molecules is introduced as a physical prior to optimize the fusion process. S3232. Based on the optimization results, a gating mechanism is used to dynamically weight and integrate the fusion outputs of each scale layer, and the contribution of features at different scales is adaptively adjusted through the gating parameters to generate multi-scale fusion features.

[0015] According to another aspect of the present invention, a hazardous gas removal system based on a special gas supply device is also provided, the system comprising: The stagnation area identification module is used to collect multi-dimensional status data during equipment operation, and use a pre-built intelligent prediction model to perform numerical simulation of the gas flow path inside the equipment to identify the stagnation area of ​​hazardous gases. The hazardous gas migration module is used to apply a magnetoacoustic coupling field to the retention area to desorb hazardous gases. It uses micromagnetic simulation equations to dynamically evolve the formation process of skyrmions, determine the gas transport path, and migrate the desorbed hazardous gases to a preset safe treatment area. The physicochemical feature acquisition module is used to collect optical response data and terahertz spectrum data in the equipment after migration treatment. It uses a learnable encoder to extract and fuse optical features and spectrum features to obtain the physicochemical features of residual hazardous gases. The residual hazardous gas simulation module is used to construct a digital twin model based on the acquired physicochemical characteristics of residual hazardous gases, and to simulate the residual diffusion behavior and retention risk of residual hazardous gases in the equipment. The iterative loop module is used to optimize the parameters of the magnetoacoustic coupling field based on the simulation results of the digital twin, and to optimize through iterative loops until the hazardous gases are completely discharged from the equipment.

[0016] The beneficial effects of this invention are as follows: 1. This invention identifies the retention area of ​​hazardous gases, applies a magnetoacoustic coupling field to achieve precise desorption, and combines micro-magnetic simulation to determine the gas transport path, thereby improving the efficiency and directionality of hazardous gas removal, reducing the safety risks caused by residues, and ensuring the stability of special gas supply equipment operation and the safety of the operating environment.

[0017] 2. This invention extracts the physicochemical characteristics of hazardous gases by collecting and fusing optical response data and terahertz spectrum data, and plans gas transport paths by combining micromagnetism and skyrmion mechanisms, thereby avoiding irregular diffusion and retention of hazardous gases inside the equipment and improving the reliability and controllability of the gas removal process.

[0018] 3. This invention simulates the diffusion behavior and retention risk of residual hazardous gases by constructing a digital twin, optimizes the magnetoacoustic coupling field parameters in real time, and adopts an iterative closed-loop control strategy to ensure efficient and stable removal process, improve the intelligence level of gas removal, and reduce energy consumption and operational risks. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a method for removing hazardous gases based on a special gas supply device according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a hazardous gas removal system based on a special gas supply device according to an embodiment of the present invention.

[0021] In the picture: 1. Retention Area Identification Module; 2. Hazardous Gas Migration Module; 3. Physicochemical Characteristic Acquisition Module; 4. Residual Hazardous Gas Simulation Module; 5. Iterative Loop Module. Detailed Implementation

[0022] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0023] According to embodiments of the present invention, a method and system for removing hazardous gases based on a special gas supply device are provided.

[0024] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for eliminating hazardous gases based on a special gas supply device includes: S1. Collect multi-dimensional status data during equipment operation, and use a pre-built intelligent prediction model to perform numerical simulation of the gas flow path inside the equipment to identify the stagnation area of ​​hazardous gases.

[0025] It should be further explained that by collecting multi-dimensional status data during equipment operation and using a pre-built intelligent prediction model, numerical simulations of the gas flow path inside the equipment are performed to identify areas where hazardous gases may accumulate, including: First, multi-dimensional state data involving pressure, temperature, flow rate, and component concentration are collected during equipment operation. Then, real-time data is input into a pre-trained intelligent prediction model (such as a fluid dynamics model based on CFD or machine learning) to accurately simulate the gas flow path and distribution state inside the equipment through numerical simulation. Finally, based on the simulation results, dead zones in airflow circulation, low-pressure zones, or abnormal stagnation areas of hazardous gases caused by adsorption effects are identified, providing a basis for subsequent intervention.

[0026] S2. Apply a magnetoacoustic coupling field to the retention area to desorb the hazardous gas. Use micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, determine the gas transport path, and migrate the desorbed hazardous gas to the preset safe treatment area.

[0027] In this optional embodiment, the step of applying a magnetoacoustic coupling field to desorb hazardous gases in the retention region, using micromagnetic simulation equations to dynamically evolve and simulate the formation process of skyrmions, determining the gas transport path, and migrating the desorbed hazardous gases to a preset safe treatment area includes: S21. Apply a magnetoacoustic coupling field to the area where hazardous gases are retained, and desorb the hazardous gases inside the equipment through the synergistic effect of the magnetic field and ultrasound.

[0028] It should be further explained that applying a magnetoacoustic coupling field to the retention area of ​​the hazardous gas, and using the synergistic effect of the magnetic field and ultrasound, desorbs the hazardous gas inside the equipment, including: A gradient magnetic field (such as using rare-earth permanent magnets or electromagnetic systems) is established in the retention area to attract paramagnetic hazardous gases (such as O2, NO, etc.) to the magnetic poles through magnetic force. Simultaneously, high-power ultrasound (frequency 20–200kHz, sound intensity 5–200W / cm²) is applied to break the adsorption bonds between gas molecules and the equipment surface by utilizing the acoustic cavitation effect and mechanical vibration. At the same time, the sound wave energy is converted into heat energy to further reduce the gas adhesion force. The magnetic field and ultrasound work together to enhance the desorption effect of gas molecules from the inner surface or structure of the equipment, and finally achieve effective desorption and removal of hazardous gases inside the equipment.

[0029] S22. Construct a spatial distribution model of the magnetoacoustic coupling field, and use micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path.

[0030] In this optional embodiment, the construction of a spatial distribution model of the magnetoacoustic coupling field and the application of micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path, includes: S221. A spatial distribution model of the magnetoacoustic coupling field inside the equipment is constructed using the finite element analysis algorithm, and a corresponding spatial distribution prediction map is generated.

[0031] S222. Based on the generated spatial distribution prediction map, the formation process of skyrmions is dynamically simulated using micromagnetic simulation equations, and the topological invariants of skyrmions are extracted to generate a skyrmion distribution characteristic map.

[0032] In this optional embodiment, the step of dynamically simulating the formation process of skyrmions using micromagnetic simulation equations based on the generated spatial distribution prediction map, and extracting the topological invariants of skyrmions to generate a skyrmion distribution feature map includes: S2221. Based on the spatial distribution prediction map of the magnetoacoustic coupling field, the nonlinear terms in the micromagnetic equation are solved using a quantum computing simulator, the dynamic evolution simulation of the skyrmion formation process is performed, and a magnetic configuration evolution map is generated.

[0033] In this optional embodiment, the calculation formula for solving the nonlinear terms in the micromagnetic equation using a quantum computing simulator is as follows: ; In the formula, Represents the total Hamiltonian of micromagnetism; J Indicates the strength of exchange coupling; i Indicates the grid index value; j Indicates the grid index value; Indicates the first i Spin operator vectors at magnetic lattice points; Indicates the first j Spin operator vectors at magnetic lattice points; D This represents the strength of the Dzyaloshinskii-Moriya interaction; A unit vector representing the vertical direction; β ∠ represents the magnetoelastic coupling coefficient; ∠ represents the divergence. u Represents the acoustic displacement field; The first in the vertical direction i Spin operator vectors at magnetic lattice points; γ e Indicates the electron gyromagnetic ratio; h Represents the reduced Planck constant; B mag This represents the vector of static magnetic field strength.

[0034] S2222. Based on the generated magnetic configuration evolution diagram, the spatiotemporal parameter combination of the magnetic field and ultrasound is adjusted in real time using a reinforcement learning algorithm. The stability of the skyrmion topology is used as a reward function for feedback training. Through continuous iterative optimization, the final magnetic configuration state is determined.

[0035] S2223. Based on the determined final magnetic configuration state, construct a hybrid neural network model and use parallel computing to extract the topological invariants of skyrmions.

[0036] S2224. Use a relocation optimization algorithm based on topological invariants to identify skyrmion centers and generate a skyrmion distribution feature map containing topological feature parameters.

[0037] It should be further explained that, based on the generated spatial distribution prediction map, the formation process of skyrmions is dynamically simulated using micromagnetic simulation equations, and the topological invariants of skyrmions are extracted to generate a skyrmion distribution characteristic map. A specific implementation example is as follows: Based on the spatial distribution prediction map of the magnetoacoustic coupling field (gradient magnetic field strength 0.5-1.2 T / m, ultrasonic frequency 20 kHz), the nonlinear terms in the micromagnetic LLG equations were solved using a quantum computing simulator (such as Qiskit or QuTip). Dynamic evolution showed that skyrmions formed Néel-type structures with a diameter of approximately 50 nm within 2 ns. Subsequently, a reinforcement learning algorithm was used, with topological stability (vortex number conservation) as the reward function, to adjust the magnetic field pulse width (10-100 ns) and acoustic intensity (5-200 W / cm²) in real time. After 500 iterations, the magnetic configuration energy decreased to Skyrmion migration speed was increased to 120 m / s; based on the optimized magnetic configuration, a CNN-GNN hybrid neural network was constructed to extract topological invariants in parallel (skyrmion number Q=±1, accounting for 98.7%, Hall angle deviation <0.02 rad); finally, the skyrmion center point was identified by the topological relocation algorithm (positioning accuracy ±2 nm), and a distribution map containing characteristic parameters such as vortex polarity and helicity was generated (feature dimension 128-dimensional).

[0038] In this optional embodiment, the step of identifying skyrmion centers using a relocation optimization algorithm based on topological invariants and generating a skyrmion distribution feature map containing topological feature parameters includes: S22241. Based on the topological invariants of skyrmions, the candidate centroids of skyrmions are identified using the topological association analysis algorithm. Combined with the distributed path search strategy, the shortest topological association path between each candidate centroid and the preset centroid is calculated, and the corresponding topological feature matching degree is evaluated. S22242. Based on the topological feature matching degree of the candidate center points, map the candidate center points to the particle positions in the particle swarm optimization algorithm, initialize the position and velocity of each particle, and construct a fitness function to initialize the individual optimal position and the global optimal position. S22243. Based on the initialization results, evaluate the convergence index and structural aggregation degree of the particle swarm, and update the inertial weights of the particles using the topological adaptive formula. S22244. Based on the updated inertia weight, iteratively update the particle's velocity and position, recalculate the particle's fitness, and update the individual optimal position and the global optimal position. S22245. Determine whether the fitness of the current global best particle meets the preset convergence threshold. If it does, execute S22246; otherwise, return to S22243 to continue iterative optimization. S22246. Output the position corresponding to the globally optimal particle as the target center point of the skyrmion, and generate the skyrmion distribution feature map by combining the corresponding topological invariants.

[0039] It should be further explained that by introducing topological correlation analysis and distributed path search algorithms based on skyrmion topological invariants, candidate skyrmion centers are accurately identified, and a particle swarm optimization model is constructed. Candidate centers are mapped to particles for multi-round iterative optimization, and particle inertia weights and position update strategies are dynamically adjusted to improve the convergence speed of the solution space and the search capability for the global optimum. Combined with a fitness function and a topological feature matching degree evaluation mechanism, high-precision extraction of skyrmion target centers is achieved, and finally, a skyrmion distribution feature map with clear topological physical meaning is generated. This enhances the accuracy and stability of topological structure identification and provides highly reliable support for topological control of gas transport paths.

[0040] S223. Based on the generated skyrmion distribution feature map, extract the feature parameters of the skyrmions using a convolutional neural network, and establish a skyrmion topology array based on the feature parameters.

[0041] S224. The established skymin subtopology array is mapped to a weighted topology graph structure, and the connectivity of the weighted topology graph structure is analyzed using the shortest path algorithm to extract the transport paths used to guide gas flow, which serve as transport channels for gas migration.

[0042] It should be further explained that the following is a specific implementation of constructing a spatial distribution model of the magnetoacoustic coupling field and using micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path: A spatial distribution model of the magnetoacoustic coupling field (gradient magnetic field 0.8-1.5 T / m, ultrasonic frequency 18 kHz) was constructed based on finite element analysis. The magneto-acoustic-electric coupling equations were solved using COMSOL multiphysics simulation software, generating a near-field electromagnetic distribution map (wavelength range λ=1.5 mm). Based on this distribution map, the micromagnetic simulation equations (LLG equations) were solved using the micromagnetic simulation software MuMax3. Dynamic evolution showed that skyrmions form Néel-type structures with diameters of 50-100 nm within 2.5 ns in [Ta / CoFeB / MgO]×15 multilayer films (vortex number Q=1 accounting for 92.3%, Hall angle deviation <0.03 rad). A convolutional neural network was used to extract 128-dimensional feature parameters such as skyrmion center coordinates, polarity, and helicity, constructing a topological array (array density...). Finally, the array is mapped to a weighted topology graph, and the connectivity is analyzed using the Dijkstra algorithm to output the gas transport path (the migration speed is improved by 35%, and the path curvature radius is optimized to 200nm).

[0043] S23. Using acoustic pressure wave focusing technology, a directional acoustic pressure gradient is established within the determined gas transport channel, and the desorbed hazardous gas is driven to migrate along a preset direction and flow into a preset safe treatment area.

[0044] It should be further explained that the use of acoustic pressure wave focusing technology to establish a directional acoustic pressure gradient within a defined gas transport channel, and to drive the desorbed hazardous gas to migrate along a preset direction and converge into a preset safe treatment area includes: First, based on sound field measurement data (such as sound pressure distribution detected by hydrophone scanning or laser deflection), a sound focusing device (such as a gradient spiral structure metamaterial or a sawtooth acoustic metasurface) is designed and deployed. By adjusting the sound wave frequency (such as 3.5–4.5 kHz) and phase difference, a standing wave field with alternating distributions of high-intensity sound pressure nodes (sound pressure peak region) and low-pressure antinodes (sound pressure minimum region) is formed within the channel. Then, according to the topological characteristics of the skyrmion transport path, the spatiotemporal parameters of the transducer array (such as sound intensity 5–200 W / cm², frequency 18–20 kHz) are adjusted to establish a sound pressure gradient (gradient intensity 0.8–1.5 T / m) pointing towards the safe region between nodes. Finally, the desorbed hazardous gas molecules are driven by acoustic radiation force to migrate directionally from the high-pressure region to the low-pressure region and flow into the safe treatment zone along a preset path (such as a low-resistance channel optimized by Dijkstra's algorithm). During the process, the migration efficiency is monitored in real time and the sound field parameters are dynamically adjusted to ensure complete gas discharge.

[0045] S3. Collect optical response data and terahertz spectrum data from the equipment after migration processing, and use a learnable encoder to extract and fuse optical and spectral features to obtain the physicochemical characteristics of residual hazardous gases.

[0046] In this optional embodiment, the optical response data and terahertz spectrum data in the acquired migration-processed device are used to extract and fuse optical and spectral features using a learnable encoder to obtain the physicochemical characteristics of the residual hazardous gas, including: S31. Based on the migration results of hazardous gases, the residual hazardous gases in the equipment are detected using an integrated optical metasurface sensing structure. Based on the detection results, the corresponding optical response data is obtained, and the terahertz spectrum data of the corresponding area is collected synchronously through a terahertz transmitter.

[0047] S32. Based on the optical response data and terahertz spectrum data, use a learnable encoder to extract optical features and spectral features, and construct a multi-scale feature pyramid to generate a joint representation that reflects the physical correlation.

[0048] In this optional embodiment, the step of extracting optical and spectral features using a learnable encoder based on optical response data and terahertz spectral data, and constructing a multi-scale feature pyramid to generate a joint representation reflecting physical correlation includes: S321. Using a learnable encoder based on physical inspiration, optical features and spectral features are extracted from optical response data and terahertz spectrum data, respectively.

[0049] S322. Based on the correlation constraints between geometric structures, the extracted optical features and spectral features are aligned in spatial structure to construct a multi-scale feature pyramid with spatial consistency.

[0050] S323. Based on the constructed multi-scale feature pyramid, optical features and spectral features at different scales are fused, and the fusion results of each scale are integrated using a gating mechanism to generate multi-scale fused features.

[0051] In this optional embodiment, the process of fusing optical and spectral features at different scales based on the constructed multi-scale feature pyramid, and integrating the fusion results at each scale using a gating mechanism to generate multi-scale fused features includes: S3231. Based on the constructed multi-scale feature pyramid, a cross-modal interactive attention mechanism is used to fuse optical features and spectral features layer by layer, and the absorption peak position information of gas molecules is introduced as a physical prior to optimize the fusion process. S3232. Based on the optimization results, a gating mechanism is used to dynamically weight and integrate the fusion outputs of each scale layer, and the contribution of features at different scales is adaptively adjusted through the gating parameters to generate multi-scale fusion features.

[0052] S324. Graph neural networks are used to map multi-scale fused features to molecular graph structure space to capture the topological relationships and chemical bond connection characteristics between molecules. Combined with a pre-set knowledge base of gas physicochemical properties, a joint characterization reflecting physical correlations is output.

[0053] It should be further explained that, through a physics-inspired learnable encoder, multi-source features are accurately extracted from optical response data and terahertz spectral data. By combining geometric structure correlation constraints to achieve feature space alignment, a multi-scale feature pyramid with spatial consistency is constructed, effectively improving the completeness and hierarchy of feature representation. Furthermore, a gating mechanism is used to fuse multi-scale features, enhancing the comprehensive utilization of information at different scales. A graph neural network is used to map the fused features to the molecular graph structure space, deeply mining the topological relationships and chemical bond characteristics of gas molecules. Combined with a physicochemical property knowledge base, a joint characterization with strong physical consistency and high analytical capability is achieved, significantly improving the recognition accuracy of gas physicochemical features and the generalization ability of the model, providing solid data support for subsequent gas recognition and path optimization.

[0054] S33. Based on the generated joint characterization, a multi-task deep learning network is constructed to identify the type, concentration and spatial distribution information of residual hazardous gases, and the physicochemical characteristics of residual hazardous gases are extracted by combining physicochemical property constraints and molecular prior knowledge graphs.

[0055] It should be noted that physicochemical characteristics include: molecular structure characteristics, electrical and polar properties, spectral response characteristics, thermodynamic and kinetic parameters, and topological and graphical structure indices.

[0056] S4. Based on the obtained physicochemical characteristics of residual hazardous gases, construct a digital twin model to simulate the residual diffusion behavior and retention risk of residual hazardous gases in the equipment.

[0057] It should be further explained that, based on the obtained physicochemical characteristics of the residual hazardous gases, a digital twin model is constructed to simulate the residual diffusion behavior and retention risks of the residual hazardous gases within the equipment, including: First, a digital twin model is established by dividing the data into three-dimensional meshes, and gas diffusion physics equations (such as mass conservation equations and turbulence models) and equipment material property parameters are integrated. Second, optical features (such as gas distribution thermograms) and spectral features (such as the absorption peak positions of NH3 at 10.4 THz and H2S at 8.7 THz) are mapped to the digital twin model. Combined with machine learning models (such as diffusion predictors based on AdaBoost ensemble learning), the multi-physics coupled diffusion behavior of residual hazardous gases is simulated, and their retention concentration and diffusion rate in dead zones are quantified. Finally, a risk heatmap is output by rendering a gas concentration gradient cloud map in real time (red >1000ppm is a high-risk area) and combining it with a retention time threshold (>300s triggers an alarm).

[0058] S5. Based on the simulation results of the digital twin, optimize the parameters of the magnetoacoustic coupling field, and optimize it iteratively until the hazardous gas is completely discharged from the equipment.

[0059] It should be further noted that, based on the simulation results of the digital twin, the parameters of the magnetoacoustic coupling field are optimized, and iterative optimization is performed until the hazardous gases are completely discharged. The equipment includes: First, based on the residual gas diffusion behavior and retention risk heat map simulated by the digital twin (e.g., high-risk areas with concentrations >1000ppm), intelligent optimization algorithms (e.g., whale optimization algorithm) are used to dynamically adjust the magnetoacoustic coupling field parameters: the magnetic field gradient intensity is optimized to 0.8-1.5T / m, the acoustic intensity to 50-200W / cm², and the ultrasonic frequency is calibrated to 18-20kHz. The acoustic pressure gradient of the gas transport path is directionally enhanced according to the acoustic pressure node distribution (controlled by metamaterial acoustic metasurface). Then, an iterative loop is executed: after applying the optimized parameter combination, the gas migration efficiency (e.g., migration speed, residual concentration) is monitored in real time. The acoustic signal feedback is processed by a convolutional neural network (U-Net model) to calculate the signal-to-noise ratio improvement (measured to be over 40%) and the change in residual gas concentration. If the preset safety threshold (e.g., <10ppm) is not reached, the closed-loop process of digital twin simulation-parameter optimization-migration monitoring is re-entered until the gas concentration in the entire equipment area is continuously lower than the safety threshold and the retention time is <300 seconds.

[0060] like Figure 2As shown, according to another embodiment of the present invention, a hazardous gas removal system based on a special gas supply device is also provided, the system comprising: The stagnation area identification module 1 is used to collect multi-dimensional status data during equipment operation, and use a pre-built intelligent prediction model to perform numerical simulation of the gas flow path inside the equipment to identify the stagnation area of ​​hazardous gases. Hazardous gas migration module 2 is used to apply a magnetoacoustic coupling field to the retention area to desorb hazardous gases. It uses micromagnetic simulation equations to dynamically evolve the formation process of skyrmions, determine the gas transport path, and migrate the desorbed hazardous gases to a preset safe treatment area. The physicochemical feature acquisition module 3 is used to collect optical response data and terahertz spectrum data in the equipment after migration treatment, and to extract and fuse optical features and spectrum features using a learnable encoder to obtain the physicochemical features of residual hazardous gases. The residual hazardous gas simulation module 4 is used to construct a digital twin model based on the acquired physicochemical characteristics of the residual hazardous gas to simulate the residual diffusion behavior and retention risk of the residual hazardous gas in the equipment. Iterative loop module 5 is used to optimize the parameters of the magnetoacoustic coupling field based on the simulation results of the digital twin, and to optimize through iterative loop until the hazardous gas is completely discharged from the equipment.

[0061] In summary, by utilizing the technical solutions described above, the present invention identifies the retention areas of hazardous gases, applies a magnetoacoustic coupling field to achieve precise desorption, and combines micromagnetic simulation to determine the gas transport path, thereby improving the efficiency and directionality of hazardous gas removal, reducing the safety risks caused by residues, and ensuring the stability of the special gas supply equipment and the safety of the operating environment. By collecting and fusing optical response data and terahertz spectrum data, the physicochemical characteristics of hazardous gases are extracted, and gas transport paths are planned using micromagnetism and skyrmion mechanisms, avoiding irregular diffusion and retention of hazardous gases within the equipment, and improving the reliability and controllability of the gas removal process. By constructing a digital twin to simulate the diffusion behavior and retention risks of residual hazardous gases, the magnetoacoustic coupling field parameters are optimized in real time, and an iterative closed-loop control strategy is adopted to ensure efficient and stable removal, improving the intelligence level of gas removal while reducing energy consumption and operational risks.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for eliminating hazardous gases based on a special gas supply equipment, characterized in that, The method includes: S1. Collect multi-dimensional status data during equipment operation, and use a pre-built intelligent prediction model to perform numerical simulation of the gas flow path inside the equipment to identify the stagnation area of ​​hazardous gases. S2. Apply a magnetoacoustic coupling field to the retention area to desorb the hazardous gas. Use micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, determine the gas transport path, and migrate the desorbed hazardous gas to the preset safe treatment area. S3. Collect optical response data and terahertz spectrum data from the equipment after migration processing, and use a learnable encoder to extract and fuse optical and spectral features to obtain the physicochemical characteristics of residual hazardous gases. S4. Based on the obtained physicochemical characteristics of residual hazardous gases, construct a digital twin model to simulate the residual diffusion behavior and retention risk of residual hazardous gases in the equipment; S5. Based on the simulation results of the digital twin, optimize the parameters of the magnetoacoustic coupling field, and iterate and optimize until the harmful gas is completely discharged from the equipment.

2. The method for eliminating hazardous gases based on a special gas supply equipment according to claim 1, characterized in that, The process of applying a magnetoacoustic coupling field to desorb hazardous gases in the retention area, using micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, determining the gas transport path, and migrating the desorbed hazardous gases to a preset safe treatment area includes: S21. Apply a magnetoacoustic coupling field to the area where hazardous gases are identified, and desorb the hazardous gases inside the equipment through the synergistic effect of the magnetic field and ultrasound. S22. Construct a spatial distribution model of the magnetoacoustic coupling field, and use micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path. S23. Using acoustic pressure wave focusing technology, a directional acoustic pressure gradient is established within the determined gas transport channel, and the desorbed hazardous gas is driven to migrate along a preset direction and flow into a preset safe treatment area.

3. The method for eliminating hazardous gases based on a special gas supply equipment according to claim 2, characterized in that, The construction of a spatial distribution model of the magnetoacoustic coupling field, and the application of micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, generating a skyrmion distribution characteristic map to determine the gas transport path, includes: S221. A spatial distribution model of the magnetoacoustic coupling field inside the equipment is constructed using the finite element analysis algorithm, and a corresponding spatial distribution prediction map is generated. S222. Based on the generated spatial distribution prediction map, the formation process of skyrmions is dynamically simulated using micromagnetic simulation equations, and the topological invariants of skyrmions are extracted to generate a skyrmion distribution feature map. S223. Based on the generated skyrmion distribution feature map, extract the feature parameters of skyrmions using a convolutional neural network, and establish a skyrmion topology array based on the feature parameters. S224. The established skymin subtopology array is mapped to a weighted topology graph structure, and the connectivity of the weighted topology graph structure is analyzed using the shortest path algorithm to extract the transport paths used to guide gas flow, which serve as transport channels for gas migration.

4. The method for eliminating hazardous gases based on a special gas supply equipment according to claim 3, characterized in that, The method based on the generated spatial distribution prediction map uses micromagnetic simulation equations to dynamically simulate the formation process of skyrmions, extracts the topological invariants of skyrmions, and generates a skyrmion distribution feature map, including: S2221. Based on the spatial distribution prediction map of the magnetoacoustic coupling field, the nonlinear terms in the micromagnetic equation are solved using a quantum computing simulator, the dynamic evolution simulation of the skyrmion formation process is performed, and a magnetic configuration evolution map is generated. S2222. Based on the generated magnetic configuration evolution diagram, the spatiotemporal parameter combination of the magnetic field and ultrasound is adjusted in real time using a reinforcement learning algorithm. The stability of the skyrmion topology is used as a reward function for feedback training. Through continuous iterative optimization, the final magnetic configuration state is determined. S2223. Based on the determined final magnetic configuration state, construct a hybrid neural network model and use parallel computing to extract the topological invariants of skyrmions; S2224. Use a relocation optimization algorithm based on topological invariants to identify skyrmion centers and generate a skyrmion distribution feature map containing topological feature parameters.

5. A method for eliminating hazardous gases based on a special gas supply device according to claim 4, characterized in that, The formula for solving the nonlinear terms in the micromagnetic equations using a quantum computing simulator is as follows: ； In the formula, Represents the total Hamiltonian of micromagnetism; J Indicates the strength of exchange coupling; i Indicates the grid index value; j Indicates the grid index value; Indicates the first i Spin operator vectors on magnetic lattice points; Indicates the first j Spin operator vectors on magnetic lattice points; D This represents the strength of the Dzyaloshinskii-Moriya interaction; A unit vector representing the vertical direction; β ∠ represents the magnetoelastic coupling coefficient; ∠ represents the divergence. u Represents the acoustic displacement field; The first in the vertical direction i Spin operator vectors on magnetic lattice points; γ e Indicates the electron gyromagnetic ratio; h Represents the reduced Planck constant; B mag This represents the vector of static magnetic field strength.

6. A method for eliminating hazardous gases based on a special gas supply device according to claim 5, characterized in that, The process of identifying skymin center points using a relocation optimization algorithm based on topological invariants and generating a skymin distribution feature map containing topological feature parameters includes: S22241. Based on the topological invariants of skyrmions, the candidate centroids of skyrmions are identified using the topological association analysis algorithm. Combined with the distributed path search strategy, the shortest topological association path between each candidate centroid and the preset centroid is calculated, and the corresponding topological feature matching degree is evaluated. S22242. Based on the topological feature matching degree of the candidate center points, map the candidate center points to the particle positions in the particle swarm optimization algorithm, initialize the position and velocity of each particle, and construct a fitness function to initialize the individual optimal position and the global optimal position. S22243. Based on the initialization results, evaluate the convergence index and structural aggregation degree of the particle swarm, and update the inertial weights of the particles using the topological adaptive formula. S22244. Based on the updated inertia weight, iteratively update the particle's velocity and position, recalculate the particle's fitness, and update the individual optimal position and the global optimal position. S22245. Determine whether the fitness of the current global best particle meets the preset convergence threshold. If it does, execute S22246; otherwise, return to S22243 to continue iterative optimization. S22246. Output the position corresponding to the globally optimal particle as the target center point of the skyrmion, and generate the skyrmion distribution feature map by combining the corresponding topological invariants.

7. A method for eliminating hazardous gases based on a special gas supply equipment according to claim 1, characterized in that, The optical response data and terahertz spectrum data in the acquired and processed equipment are used to extract and fuse optical and spectral features using a learnable encoder to obtain the physicochemical characteristics of residual hazardous gases, including: S31. Based on the migration results of hazardous gases, the residual hazardous gases in the equipment are detected using an integrated optical metasurface sensing structure. Based on the detection results, the corresponding optical response data is obtained, and the terahertz spectrum data of the corresponding area is collected synchronously through a terahertz transmitter. S32. Based on the optical response data and terahertz spectrum data, use a learnable encoder to extract optical features and spectral features, and construct a multi-scale feature pyramid to generate a joint representation that reflects the physical correlation. S33. Based on the generated joint characterization, a multi-task deep learning network is constructed to identify the type, concentration and spatial distribution information of residual hazardous gases, and the physicochemical characteristics of residual hazardous gases are extracted by combining physicochemical property constraints and molecular prior knowledge graphs.

8. A method for eliminating hazardous gases based on a special gas supply device according to claim 7, characterized in that, The process of extracting optical and spectral features using a learnable encoder based on optical response data and terahertz spectral data, and constructing a multi-scale feature pyramid to generate a joint representation reflecting physical correlations includes: S321. Using a learnable encoder based on physical inspiration, optical features and spectral features are extracted from optical response data and terahertz spectrum data, respectively. S322. Based on the correlation constraints between geometric structures, the extracted optical features and spectral features are aligned in spatial structure to construct a multi-scale feature pyramid with spatial consistency. S323. Based on the constructed multi-scale feature pyramid, optical features and spectral features at different scales are fused, and the fusion results of each scale are integrated by using a gating mechanism to generate multi-scale fused features. S324. Graph neural networks are used to map multi-scale fused features to molecular graph structure space to capture the topological relationships and chemical bond connection characteristics between molecules. Combined with a pre-set knowledge base of gas physicochemical properties, a joint characterization reflecting physical correlations is output.

9. A method for eliminating hazardous gases based on a special gas supply equipment according to claim 8, characterized in that, The constructed multi-scale feature pyramid fuses optical and spectral features at different scales, and uses a gating mechanism to integrate the fusion results at each scale, generating multi-scale fused features including: S3231. Based on the constructed multi-scale feature pyramid, a cross-modal interactive attention mechanism is used to fuse optical features and spectral features layer by layer, and the absorption peak position information of gas molecules is introduced as a physical prior to optimize the fusion process. S3232. Based on the optimization results, a gating mechanism is used to dynamically weight and integrate the fusion outputs of each scale layer, and the contribution of features at different scales is adaptively adjusted through the gating parameters to generate multi-scale fusion features.

10. A hazardous gas removal system based on a special gas supply device, used to implement the hazardous gas removal method based on a special gas supply device according to any one of claims 1-9, characterized in that, The system includes: The stagnation area identification module is used to collect multi-dimensional status data during equipment operation, and use a pre-built intelligent prediction model to perform numerical simulation of the gas flow path inside the equipment to identify the stagnation area of ​​hazardous gases. The hazardous gas migration module is used to apply a magnetoacoustic coupling field to the retention area to desorb hazardous gases. It uses micromagnetic simulation equations to dynamically evolve the formation process of skyrmions, determine the gas transport path, and migrate the desorbed hazardous gases to a preset safe treatment area. The physicochemical feature acquisition module is used to collect optical response data and terahertz spectrum data in the equipment after migration treatment. It uses a learnable encoder to extract and fuse optical features and spectrum features to obtain the physicochemical features of residual hazardous gases. The residual hazardous gas simulation module is used to construct a digital twin model based on the acquired physicochemical characteristics of residual hazardous gases, and to simulate the residual diffusion behavior and retention risk of residual hazardous gases in the equipment. The iterative loop module is used to optimize the parameters of the magnetoacoustic coupling field based on the simulation results of the digital twin, and to optimize through iterative loops until the hazardous gases are completely discharged from the equipment.

Images