Waste gas treatment efficiency online monitoring method and system based on artificial intelligence

Abstract

The invention discloses a waste gas treatment efficiency online monitoring method and system based on artificial intelligence, and relates to the technical field of waste gas treatment monitoring, and the method comprises the steps: deploying a sensor in a waste gas treatment device, collecting waste gas treatment data, obtaining process parameters, and constructing a sliding window for normalization treatment to output waste gas treatment characteristics after time alignment; constructing a waste gas treatment neural network by using a three-branch time sequence encoder, defining a sheath strength agent to calculate gating weight fusion branch output, defining a multiple objective function as a training objective, mapping a potential state through an energy constraint potential state space, and outputting predicted waste gas treatment efficiency; defining an optimization objective function and initializing optimization variables based on the waste gas treatment efficiency, updating and outputting the optimization variables through combination of random disturbance and SPSA, adjusting waste gas treatment process parameters based on the optimization variables, and forming an adjustment record for storage. The stability and the treatment efficiency of the waste gas treatment system are remarkably improved, and energy efficiency maximization is achieved.

Description

Technical Field

[0001] This invention relates to the field of waste gas treatment monitoring technology, and in particular to an online monitoring method and system for waste gas treatment efficiency based on artificial intelligence. Background Technology

[0002] With the increasing integration of industrial production and environmental protection, waste gas treatment has become one of the important environmental protection technologies, especially in industries such as power and chemicals. While using SF6 (sulfur hexafluoride) gas as an insulating material can effectively improve equipment performance, the environmental impact of the waste gas generated during its use cannot be ignored. SF6 waste gas has a strong greenhouse effect and is one of the gases with the highest global warming potential index. Therefore, how to efficiently and environmentally treat SF6 waste gas has become a pressing technical problem for industry. Traditional SF6 waste gas treatment methods, such as physical adsorption, chemical adsorption, and wet treatment, can reduce SF6 concentration to some extent, but they have low treatment efficiency, complex processes, and high energy consumption. In addition, these traditional methods may generate harmful byproducts during treatment, requiring further treatment and control. Existing technologies for SF6 waste gas treatment typically rely on conventional gas adsorption technologies and high-temperature treatment, but they still have many shortcomings in terms of efficiency, energy consumption, and environmental impact.

[0003] In recent years, microwave plasma technology, as an emerging waste gas treatment technology, has gradually become an important direction for solving this problem by achieving efficient decomposition of harmful gases in waste gas through high-energy electron excitation. The advantages of microwave plasma technology include fast reaction rate, high energy utilization rate, stable operation under normal pressure, and good degradation effect on most gases. Some research results at home and abroad show that microwave plasma technology can effectively decompose SF6 waste gas and convert SF6 into relatively harmless substances. However, the existing technology for microwave plasma treatment still has shortcomings. It cannot accurately obtain the SF6 waste gas treatment effect in real time under different operating conditions, which leads to the inability to accurately control the microwave plasma state to obtain the best waste gas treatment efficiency. Summary of the Invention

[0004] In view of the above-mentioned existing problems, the present invention provides an online monitoring method and system for exhaust gas treatment efficiency based on artificial intelligence, which solves the problem that the existing technology cannot accurately obtain the SF6 exhaust gas treatment effect in real time under different operating conditions, resulting in the inability to accurately control the microwave plasma state to obtain the best exhaust gas treatment efficiency.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an online monitoring method for waste gas treatment efficiency based on artificial intelligence, which includes the following steps: Sensors are deployed in the waste gas treatment device to collect waste gas treatment data and obtain process parameters. After aligning the time, a sliding window is constructed for normalization processing to output waste gas treatment characteristics. A three-branch temporal encoder is used to construct a neural network for exhaust gas treatment. The sheath strength proxy computation gate weight fusion branch output is defined, and multiple objective functions are defined as training objectives. The output predicts the exhaust gas treatment efficiency by mapping the latent state space through energy constraints. The objective function for waste gas treatment efficiency is defined and the optimization variables are initialized. The optimization variables are updated by combining random perturbation with SPSA. The waste gas treatment process parameters are adjusted based on the optimization variables, and the adjustment records are stored.

[0006] As a further preferred embodiment, deploying sensors in the waste gas treatment device to collect waste gas treatment data and obtain process parameters refers to deploying a reflectometer, an OES optical fiber, and an FTIR sensor on the waste gas treatment device to collect the reflection power ratio, respectively. The emission spectrum and original spectral band intensity were obtained simultaneously, along with process parameters including waste gas flow rate Q and gas ratio. and microwave power While maintaining exhaust gas flow without ignition, frequency scanning is performed, and the resonant frequency is recorded. and half-power bandwidth .

[0007] As a further preferred embodiment, the normalization process after aligning the time and constructing a sliding window to output exhaust gas treatment features involves aligning the collected exhaust gas treatment data and process parameters over time and constructing a sliding window of length L and step size S to obtain the reflection power ratio of each sampling point within the sliding window. and at the resonant frequency point A narrowband fast scan is performed, and the ratio of the resonant frequency to the half-power width is calculated as the quality factor. ; Based on the preset OES fiber optic transmission lines, each transmission line is configured with a fixed bandwidth within a sliding window. Calculate the OES trapezoidal integral ; Synchronization is achieved by using a preset set of infrared bands (acquired via an FTIR sensor) within a sliding window at a fixed bandwidth. Calculate the FTIR trapezoidal integral ; Baseline-robust scaling normalization is performed on all feature channels in a sliding window, and a proxy metric DPI is constructed. All feature channels are spliced ​​together to form the exhaust gas treatment feature. Output.

[0008] As a further preferred embodiment, the exhaust gas treatment neural network is constructed using a three-branch time-series encoder, and the sheath strength proxy calculation gate weight fusion branch output is defined as the exhaust gas treatment neural network constructed using a three-branch time-series encoder, including an electromagnetic branch, a spectroscopic branch, and a process branch. The electromagnetic branch employs a causal one-dimensional dilated convolution combined with lightweight self-attention pairs on the electromagnetic sequence. Process it; The spectroscopic branch uses a learnable bandpass filter group combined with channel-by-channel SE gating to process the OES trapezoidal integral and the FTIR trapezoidal integral. The process branch uses a controlled GRU to manage process parameters. Process it; The sheath strength of the waste gas treatment device is defined based on the quality factor and absorption power. ; Gating weights for three-branch calculation using sheath strength proxy ; The three-branch outputs are fused and the output time step is represented based on the gating weights. ; Calculate energy based on waste gas treatment characteristics And further calculate the gating time attention ; The time step representation is fused and output as a window context based on gated temporal attention. .

[0009] As a further preferred embodiment, the definition of a multiple objective function as the training objective, and the output of the predicted exhaust gas treatment efficiency after mapping the latent states through energy constraints in the latent state space, are based on the window context. Mapping latent states And initialized to processing efficiency ; Based on electromagnetic sequence The transpose is used to calculate the energy flow E; For each step k=(t, …, t+S) within the time window, based on process parameters Energy constraints combined with time step characterization Update potential states and processing efficiency; Multiple objective functions are defined based on the neural network for exhaust gas treatment, including mean squared error (MSE), reachability gating regularization, and fusion entropy regularization. A weighted combination of mean squared error (MSE), reachability gating regularization, and fusion entropy regularization is formed into a multi-objective function. The neural network for exhaust gas treatment is iteratively trained and optimized using the objective function. The iteration stops when convergence and outputs the predicted exhaust gas treatment efficiency. .

[0010] As a further preferred embodiment, the step of defining an optimization objective function based on exhaust gas treatment efficiency and initializing optimization variables, and updating the output optimization variables by combining random perturbation with SPSA to initialize the optimization variables, is described. The predicted waste gas treatment efficiency is combined with process parameters to define an optimization objective function J; For optimization variables Apply random perturbations and recalculate the objective function value after each perturbation. SPSA is used to update and optimize the optimization variables; Set a target convergence threshold, and stop optimization when the optimization meets the convergence condition, then output the final optimization variables.

[0011] As a further preferred option, adjusting the waste gas treatment process parameters based on the optimized variables and storing the adjustment record means adjusting the process parameters of the waste gas treatment device according to the final optimized variables, and waiting for a step size S after the parameters are issued to re-predict the waste gas treatment efficiency. If the waste gas treatment efficiency is greater than the set threshold, the optimized variables and the adjusted waste gas treatment process parameters are adjusted and stored as an adjustment record through adjustment testing.

[0012] Secondly, the present invention provides an online monitoring system for waste gas treatment efficiency based on artificial intelligence, comprising: The feature processing module is used to deploy sensors in the waste gas treatment device to collect waste gas treatment data and obtain process parameters. After aligning the time, a sliding window is constructed for normalization processing to output waste gas treatment features. The prediction module is used to construct a neural network for exhaust gas treatment using a three-branch temporal encoder, define the sheath strength proxy calculation gate weight fusion branch output, and define multiple objective functions as training objectives. After mapping the latent state through energy constraints, the output predicts the exhaust gas treatment efficiency. The adjustment record module is used to define an optimization objective function based on the waste gas treatment efficiency and initialize optimization variables. It updates the output optimization variables through random perturbation combined with SPSA, adjusts the waste gas treatment process parameters based on the optimization variables, verifies the adjustment effect, and stores the adjustment record.

[0013] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence as described in the first aspect of the present invention.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence as described in the first aspect of the present invention.

[0015] The beneficial effects of this invention are as follows: This invention collects waste gas treatment parameters and process parameters for feature mapping, and uses a three-branch time-series encoder neural network to predict and optimize the waste gas treatment process in real time. It also dynamically adjusts the waste gas treatment process parameters by combining multiple objective optimization functions of waste gas treatment efficiency and energy efficiency, which significantly improves the stability and treatment efficiency of the waste gas treatment system and maximizes energy efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0017] Figure 1 This is a flowchart of the online monitoring method for waste gas treatment efficiency based on artificial intelligence in Example 1; Figure 2 This is a structural diagram of the online monitoring system for waste gas treatment efficiency based on artificial intelligence in Example 1; Figure 3 This is a structural diagram of the waste gas treatment device in Example 1. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0021] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides an online monitoring method for waste gas treatment efficiency based on artificial intelligence, including the following steps: S1. Deploy sensors in the waste gas treatment device to collect waste gas treatment data and obtain process parameters. After aligning the time, construct a sliding window for normalization processing and output waste gas treatment characteristics. Specifically, deploying sensors in the waste gas treatment unit to collect waste gas treatment data and obtain process parameters refers to deploying reflectometers, OES optical fibers, and FTIR sensors on the waste gas treatment unit to collect reflected power ratios, respectively. (Ratio of reflected power to incident power), emission spectrum and original spectral band intensity, and simultaneously acquire process parameters including waste gas flow rate Q and gas ratio. and microwave power While keeping the exhaust gas flowing in without ignition, perform frequency scanning on the resonant cavity and record the resonant frequency points. and half-power bandwidth The exhaust gas treatment device includes a microwave source, a magnetron, a circulator, a tuner, a gradient waveguide, a resonant cavity, and a short-circuit piston. A double-layer quartz discharge tube is coaxially embedded in the resonant cavity window of the compressed rectangular waveguide. The outer tube is heat-insulated, and the inner tube is a plasma reaction zone. A conical multi-nozzle nozzle is fixed at the inlet end of the inner tube, with the nozzle tip facing the center of the cavity.

[0022] Furthermore, after aligning the time, a sliding window is constructed for normalization processing to output the exhaust gas treatment characteristics. The collected exhaust gas treatment data and process parameters are time-aligned, and a sliding window of length L and step size S is constructed to obtain the reflection power ratio of each sampling point within the sliding window. and at the resonant frequency point A narrowband fast scan is performed, and the ratio of the resonant frequency to the half-power width is calculated as the quality factor. ; Based on the preset OES fiber optic transmission lines (obtained via OES fiber), each transmission line is assigned a fixed bandwidth within the sliding window. Calculate the OES trapezoidal integral :

[0023] in Let be the frequency of the i-th transmission line. For integration variables, For the cumulative spectrum within the window, To establish the baseline dark level / background spectrum, near the resonant frequency, the microwave source power was fixed, and the short-circuit piston and tuner were adjusted to minimize the reflected power ratio. The corresponding resonant frequency, half-power width, gas flow rate, reflected power ratio, and the positions of the short-circuit piston and tuner were recorded as the baseline. Synchronization is achieved by using a preset set of infrared bands (acquired via an FTIR sensor) within a sliding window at a fixed bandwidth. Calculate the FTIR trapezoidal integral :

[0024] in Let the intensity of the j-th infrared band be... The cumulative spectrum within the window of the original spectral band intensity. The baseline dark level / background spectrum of the original band intensity. For integration variables; For all feature channels ( , , , Baseline-robust scaling normalization is performed within a sliding window, and a proxy metric DPI is constructed:

[0025] in ~ For fixed weights, , , as well as These are the features after normalization. It is the minimum quality factor; All feature channels are spliced ​​together to form the exhaust gas treatment feature. Output.

[0026] By unifying timestamps and establishing sliding windows, signals from different sources are sampled synchronously on the same time scale. This not only avoids time delay errors but also ensures causal consistency in subsequent feature correlations, thereby improving the coupling degree between features and prediction stability. The fixed bandwidth integration method integrates the energy distribution of the emission and absorption regions, reduces the influence of random noise, and retains comprehensive information of multiple reaction intermediates. The result makes the correspondence between spectral features and microwave energy states more robust, realizing the joint mapping of chemical reactions and energy states. DPI integrates two types of information: energy absorption (OES, FTIR) and energy matching (RΓ, Qinst), expressing plasma activity and coupling efficiency on a unified scale. Its physical meaning can be understood as "the degree of synergy between energy input and reaction response." This index is directly observable when predicting waste gas treatment efficiency and can be used as a feedback quantity for online control. By splicing the above normalized features in the time dimension to generate a unified vector, not only is the temporal correlation preserved but also information fusion of different physical channels is achieved. This feature vector provides a high-dimensional input basis for subsequent neural networks or energy constraint models, with strong physical correlation and information completeness, providing structural support for intelligent optimization.

[0027] S2. A three-branch temporal encoder is used to construct a neural network for exhaust gas treatment. The sheath strength proxy calculation gate weight fusion branch output is defined, and multiple objective functions are defined as training objectives. The exhaust gas treatment efficiency is predicted by mapping the latent state to the latent state space through energy constraints. Specifically, a three-branch time-series encoder is used to construct a neural network for exhaust gas treatment. The sheath strength proxy calculation gate weight fusion branch output is defined. The three-branch time-series encoder is used to construct the neural network for exhaust gas treatment, including electromagnetic branch, spectroscopic branch and process branch. The electromagnetic branch employs causal one-dimensional dilated convolution combined with lightweight self-attention pairs for electromagnetic sequences. To process, From the characteristics of waste gas treatment Obtain from, ,in This represents the dimensionless normalized absorption power. Incident power:

[0028]

[0029]

[0030] Where K is the kernel length and d is the hole ratio. For convolution weights, For convolution bias, For the attention weights within the branch, For the hidden representation of electromagnetic branches, For electromagnetic branch output, The convolution stride is... W and W are the attention parameters within the branch. For transpose; The spectroscopic branch employs a learnable bandpass filter group combined with channel-wise SE gating to process the OES trapezoidal integral and the FTIR trapezoidal integral:

[0031]

[0032]

[0033] Where SE stands for squeeze-and-excitation channel recalibration. The spectral eigenvectors are composed of OES trapezoidal integrals and FTIR trapezoidal integrals. For one-dimensional convolution, To hide the branch of spectroscopy, Output for the spectroscopy branch; The process branch uses a controlled GRU to manage process parameters. To process, From the characteristics of waste gas treatment Obtain from:

[0034]

[0035]

[0036]

[0037]

[0038] in For the sigmoid function, , , , , as well as For GRU weights and biases, This represents a hidden state for process branches. , , for The recursion, This is element-wise multiplication; The sheath strength of the waste gas treatment device is defined based on the quality factor and absorption power. :

[0039] in The dimensionless normalization coefficient; Gating weights for three-branch calculation using sheath strength proxy :

[0040]

[0041] in and These are learnable parameters; The three-branch outputs are fused and the output time step is represented based on the gating weights. :

[0042] in , as well as These are the gating weights for the three-branch outputs; Calculate energy based on waste gas treatment characteristics And further calculate the gating time attention :

[0043]

[0044] in and For gating time attention parameters, This is the energy bias coefficient; The time step representation is fused and output as a window context based on gated temporal attention. :

[0045] Where L is the length of the time window.

[0046] The electromagnetic branch expands the receptive field through dilated convolution, enabling it to capture periodic reflections and transient mismatches in electromagnetic signals. Simultaneously, the causal structure ensures the temporal unidirectionality of predictions, preventing information leakage. Combined with lightweight self-attention, the network can automatically identify critical moments in changes in reflection power and quality factor, achieving precise tracking of energy coupling states. The spectral branch addresses the high noise and high correlation characteristics of OES and FTIR signals, adaptively selecting active emission and absorption bands through learnable filters to achieve precise focusing on spectral energy distribution. SE gating further performs channel recalibration, ensuring the model still highlights relevant spectral bands under complex conditions such as high temperature and different gas ratios. The process branch introduces explicit external control inputs (such as flow rate, microwave power, and ratio) to form a temporal recursive description of process dynamics. Gating mechanisms adjust the coupling ratio between historical states and current control variables, giving the network good generalization ability under different operating conditions. By using sheath strength as a proxy for energy state driving quantities, the electromagnetic, spectral, and process outputs are weighted and fused, achieving physical consistency of information across different modes. The gating mechanism automatically increases spectral weights to capture enhanced reaction features when energy coupling is enhanced; when mismatch occurs, electromagnetic branch weights increase to reflect the mismatch state, thus forming a dynamic closed loop of "energy state - signal weights - feature output". This solves the problems of lack of physical reference and unstable weight allocation with data fluctuations in existing methods for multimodal feature fusion. The introduction of the energy function makes time attention no longer dependent on pure statistical features, but guided by energy balance, selecting locations with key physical events (such as discharge mutations and energy peaks) for feature weighting. This energy-constrained attention can suppress the interference of redundant time slices, allowing the model to focus on key periods reflecting energy transfer and reaction conversion efficiency, thereby improving the sensitivity and interpretability of predictions. The context fusion output generated by gating time attention enables simultaneous modeling of short-term dynamics and long-term trends. This design plays a special role in this invention: the exhaust gas treatment process has nonlinear and time-delay characteristics, and the window context can capture multi-scale energy change patterns without increasing model complexity, providing robust high-dimensional input for subsequent efficiency prediction and energy consumption optimization.

[0047] Furthermore, a multi-objective function is defined as the training objective. After mapping the latent states through energy constraints in the latent state space, the output predicts the exhaust gas treatment efficiency based on the window context. Mapping latent states And initialized to processing efficiency :

[0048]

[0049] in For ReLU function, , , as well as For learnable parameters, m is the dimension; Based on electromagnetic sequence The transpose of the energy flow E is calculated as follows:

[0050]

[0051]

[0052] in For linear terms, It is a positive definite diagonal. , , as well as These are learnable parameters; For each step k=(t, …, t+S) within the time window, based on process parameters Energy constraints combined with time step characterization Update potential states and processing efficiency:

[0053]

[0054] in For a stable linear matrix, For control mapping, For small MLP networks, For the energy descent step size, For energy flow gradient, The parameters for the linearly stable subsystem can be obtained through learning; B corresponds to the process parameters. It can be acquired through learning. It can be acquired through pre-training or later learning; Multiple objective functions are defined based on the neural network for exhaust gas treatment, including mean squared error (MSE), reachability gating regularization, and fusion entropy regularization. The reachability gating regularity for:

[0055]

[0056] in and The sheath mapping coefficient, To predict the upper limit of a single step change; The fusion entropy regularization for:

[0057] Where i represents the electromagnetic branch, the spectroscopic branch, and the technological branch; A weighted combination of mean squared error (MSE), reachability gating regularization, and fusion entropy regularization is formed into a multi-objective function. The neural network for exhaust gas treatment is iteratively trained and optimized using the objective function. The iteration stops when convergence and outputs the predicted exhaust gas treatment efficiency. .

[0058] Dynamic modeling in energy space is achieved through energy constraints. This design ensures that changes in the latent state are consistent with actual energy input and output, solving the problem of prediction instability caused by the neglect of physical laws in traditional deep models. Energy flow correction effectively avoids overfitting and state drift, enabling the network to maintain energy conservation in long-term predictions. This mechanism physically corresponds to the energy dissipation path of plasma discharge, allowing the network to output smooth and reasonable efficiency change curves even when faced with sudden energy fluctuations. The diagonalization design of the stability matrix A ensures the independent decay characteristics of each latent dimension, giving the model asymptotic convergence during training. The control map B allows external process inputs (such as power or gas flow rate) to act on the latent state in a learnable manner, solving the problem that traditional models cannot simultaneously model the relationship between "control quantity - energy state - reaction result", thus realizing the joint operation of energy and process parameters. Optimization is achieved by establishing dynamic constraints through reachability gating regularization, ensuring that efficiency changes align with the system's energy release rate. This not only improves the model's safety and reliability but also prevents misleading the control system during actual operation. Theoretically, it ensures that the predicted results are consistent with the temporal reachability of the energy transfer rate. During the three-branch feature fusion process, if the model over-relies on a certain feature channel (such as electromagnetic signals), it can lead to a decrease in the utilization rate of spectral and process information. Fusion entropy regularization suppresses the state of excessively low weight entropy, prompting the model to maintain a dynamic balance among different information sources. By weighting and combining mean square error, reachability gating regularization, and fusion entropy regularization, the model training simultaneously optimizes the three major indicators of accuracy, stability, and information utilization. This joint objective solves the common problem that a single loss function cannot adequately consider multi-dimensional performance, making the model training results more in line with the requirements of physical interpretability and industrial deployability.

[0059] S3. Define the optimization objective function based on the waste gas treatment efficiency and initialize the optimization variables. Update the output optimization variables by combining random perturbation with SPSA. Adjust the waste gas treatment process parameters based on the optimization variables and store the adjustment records. Specifically, an optimization objective function is defined based on exhaust gas treatment efficiency, and optimization variables are initialized. The output optimization variables are then updated using random perturbation combined with SPSA (Special Persistent Stress Analysis). ,in For exhaust gas flow rate adjustment, This is the amount for adjusting the exhaust gas ratio. The microwave power adjustment term is used, and the objective function J is defined by combining the predicted waste gas treatment efficiency with process parameters:

[0060] in , as well as For weight parameters, This is the preset minimum quality factor. This is a penalty term for the quality factor, ensuring that the quality factor of the resonant cavity is not lower than the preset minimum value; For optimization variables Apply random perturbations and recalculate the objective function value after each perturbation. SPSA is used to update and optimize the variables:

[0061]

[0062] in To update the gradient, for vector, and The step size sequence is obtained by fixing a constant beforehand. This is a projection operation on the actuator boundary and slope constraint set; Set a target convergence threshold, and stop optimization when the optimization meets the convergence condition, then output the final optimization variables.

[0063] By using random perturbations instead of traditional multidimensional partial derivative calculations, global gradient approximation can be achieved with only two function evaluations. Due to the time-varying and strong nonlinearity of the parameters in the exhaust gas treatment system, conventional gradient calculations are easily affected by noise amplification. However, SPSA's symmetric random perturbations can statistically eliminate measurement noise, making the optimization path more robust. The introduction of the projection operator Πc ensures that the parameters remain within the allowable range of the equipment after each update, such as the upper and lower limits of flow rate, maximum power input, and safety boundary of mixing ratio. Unlike traditional unconstrained optimization, this design embeds equipment safety mechanisms at the algorithm level, avoiding the situation of "numerical optimal but physically unrealizable". By applying nonlinear penalties for cases where the quality factor is below a threshold, the optimization algorithm can maintain high energy coupling of the resonant cavity while improving efficiency. The introduction of power and reflection terms in the optimization objective allows the algorithm to not only focus on degradation efficiency but also actively suppress energy waste and electromagnetic backflow. Especially under high load conditions, the algorithm can automatically reduce power output to maintain high absorption rate, realizing the implicit constraint of "maximizing plasma energy utilization".

[0064] Furthermore, adjusting the waste gas treatment process parameters based on the optimized variables and forming an adjustment record for storage means adjusting the process parameters of the waste gas treatment device according to the final optimized variables, and waiting for a step size S after the parameters are issued to re-predict the waste gas treatment efficiency. If the waste gas treatment efficiency is greater than the set threshold, the optimized variables and the adjusted waste gas treatment process parameters are adjusted and stored as an adjustment record through adjustment testing.

[0065] This embodiment also provides an online monitoring system for waste gas treatment efficiency based on artificial intelligence, including: The feature processing module is used to deploy sensors in the waste gas treatment device to collect waste gas treatment data and obtain process parameters. After aligning the time, a sliding window is constructed for normalization processing to output waste gas treatment features. The prediction module is used to construct a neural network for exhaust gas treatment using a three-branch temporal encoder, define the sheath strength proxy calculation gate weight fusion branch output, and define multiple objective functions as training objectives. After mapping the latent state through energy constraints, the output predicts the exhaust gas treatment efficiency. The adjustment record module is used to define an optimization objective function based on the waste gas treatment efficiency and initialize optimization variables. It updates the output optimization variables through random perturbation combined with SPSA, adjusts the waste gas treatment process parameters based on the optimization variables, verifies the adjustment effect, and stores the adjustment record.

[0066] This embodiment also provides a computer device applicable to the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence as proposed in the above embodiment. The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0067] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0068] In summary, this invention collects waste gas treatment parameters and process parameters for feature mapping, and uses a three-branch time-series encoder neural network for real-time prediction and optimization of the waste gas treatment process. By combining multiple objective optimization functions for waste gas treatment efficiency and energy efficiency, the waste gas treatment process parameters are dynamically adjusted, which significantly improves the stability and treatment efficiency of the waste gas treatment system and maximizes energy efficiency.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for online monitoring of waste gas treatment efficiency based on artificial intelligence, characterized in that, Includes the following steps: Sensors are deployed in the waste gas treatment device to collect waste gas treatment data and obtain process parameters. After aligning the time, a sliding window is constructed for normalization processing to output waste gas treatment characteristics. A three-branch temporal encoder is used to construct a neural network for exhaust gas treatment. The sheath strength proxy computation gate weight fusion branch output is defined, and multiple objective functions are defined as training objectives. The output predicts the exhaust gas treatment efficiency by mapping the latent state space through energy constraints. The objective function for waste gas treatment efficiency is defined and the optimization variables are initialized. The optimization variables are updated by combining random perturbation with SPSA. The waste gas treatment process parameters are adjusted based on the optimization variables, and the adjustment records are stored.

2. The online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in claim 1, characterized in that, The step of deploying sensors in the waste gas treatment device to collect waste gas treatment data and obtain process parameters refers to deploying reflectometers, OES optical fibers, and FTIR sensors on the waste gas treatment device to collect reflection power ratios, respectively. The emission spectrum and original spectral band intensity were obtained simultaneously, along with process parameters including waste gas flow rate Q and gas ratio. and microwave power While maintaining exhaust gas flow without ignition, frequency scanning is performed, and the resonant frequency is recorded. and half-power bandwidth .

3. The online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in claim 2, characterized in that, The process of aligning the time and constructing a sliding window for normalization to output exhaust gas treatment characteristics refers to aligning the collected exhaust gas treatment data and process parameters in time and constructing a sliding window of length L and step size S to obtain the reflection power ratio of each sampling point within the sliding window. and at the resonant frequency point A narrowband fast scan is performed, and the ratio of the resonant frequency to the half-power width is calculated as the quality factor. ; Based on the preset OES fiber optic transmission lines, each transmission line is configured with a fixed bandwidth within a sliding window. Calculate the OES trapezoidal integral ; Synchronous transmission via a preset set of infrared bands within a sliding window at a fixed bandwidth Calculate the FTIR trapezoidal integral ; Baseline-robust scaling normalization is performed on all feature channels in a sliding window, and a proxy metric DPI is constructed. All feature channels are spliced ​​together to form the exhaust gas treatment feature. Output.

4. The online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in claim 1, characterized in that, The method of constructing a waste gas treatment neural network using a three-branch time-series encoder defines the sheath strength proxy calculation gate weight fusion branch output as the method of constructing a waste gas treatment neural network using a three-branch time-series encoder, including electromagnetic branch, spectral branch and process branch; The electromagnetic branch employs a causal one-dimensional dilated convolution combined with lightweight self-attention pairs on the electromagnetic sequence. Process it; The spectroscopic branch uses a learnable bandpass filter group combined with channel-by-channel SE gating to process the OES trapezoidal integral and the FTIR trapezoidal integral. The process branch uses a controlled GRU to manage process parameters. Process it; The sheath strength of the waste gas treatment device is defined based on the quality factor and absorption power. ; Gating weights for three-branch calculation using sheath strength proxy ; The three-branch outputs are fused and the output time step is represented based on the gating weights. ; Calculate energy based on waste gas treatment characteristics And further calculate the gating time attention ; The time step representation is fused and output as a window context based on gated temporal attention. .

5. The online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in claim 1, characterized in that, The defined multiple objective function is used as the training objective. After mapping the latent states through energy constraints in the latent state space, the output predicts the exhaust gas treatment efficiency, which is determined based on the window context. Mapping latent states And initialized to processing efficiency ; Based on electromagnetic sequence The transpose is used to calculate the energy flow E; For each step k=(t, …, t+S) within the time window, based on process parameters Energy constraints combined with time step characterization Update potential states and processing efficiency; Multiple objective functions are defined based on the neural network for exhaust gas treatment, including mean squared error (MSE), reachability gating regularization, and fusion entropy regularization. A weighted combination of mean squared error (MSE), reachability gating regularization, and fusion entropy regularization is formed into a multi-objective function. The neural network for exhaust gas treatment is iteratively trained and optimized using the objective function. The iteration stops when convergence and the predicted exhaust gas treatment efficiency is output. .

6. The online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in claim 1, characterized in that, The objective function is defined based on exhaust gas treatment efficiency, and optimization variables are initialized. The output optimization variables are then updated using random perturbation combined with SPSA (Speed ​​Automated Analysis) to indicate the initialized optimization variables. The predicted waste gas treatment efficiency is combined with process parameters to define an optimization objective function J; For optimization variables Apply random perturbations and recalculate the objective function value after each perturbation. SPSA is used to update and optimize the optimization variables; Set a target convergence threshold, and stop optimization when the optimization meets the convergence condition, then output the final optimization variables.

7. The online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in claim 1, characterized in that, The process of adjusting the waste gas treatment process parameters based on the optimization variables and storing the adjustment record refers to adjusting the process parameters of the waste gas treatment device according to the final optimization variables, and waiting for a step size S after the parameters are issued to re-predict the waste gas treatment efficiency. If the waste gas treatment efficiency is greater than the set threshold, the optimization variables and the adjusted waste gas treatment process parameters are adjusted and stored as an adjustment record through adjustment testing.

8. An online monitoring system for waste gas treatment efficiency based on artificial intelligence, based on the online monitoring method for waste gas treatment efficiency based on artificial intelligence as described in any one of claims 1 to 7, characterized in that, include: The feature processing module is used to deploy sensors in the waste gas treatment device to collect waste gas treatment data and obtain process parameters. After aligning the time, a sliding window is constructed for normalization processing to output waste gas treatment features. The prediction module is used to construct a neural network for exhaust gas treatment using a three-branch temporal encoder, define the sheath strength proxy calculation gate weight fusion branch output, and define multiple objective functions as training objectives. After mapping the latent state through energy constraints, the output predicts the exhaust gas treatment efficiency. The adjustment record module is used to define an optimization objective function based on the waste gas treatment efficiency and initialize optimization variables. It updates the output optimization variables through random perturbation combined with SPSA, adjusts the waste gas treatment process parameters based on the optimization variables, verifies the adjustment effect, and stores the adjustment record.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the online monitoring method for exhaust gas treatment efficiency based on artificial intelligence as described in any one of claims 1 to 7.

Images