Methods for Spectral Aliasing Decoupling and Concentration Inversion under the Interaction of Multiple Environmental Factors

Abstract

This invention discloses a method for spectral aliasing decoupling and concentration inversion under the combined influence of multiple environmental factors, belonging to the field of industrial process control and environmental monitoring. It constructs an environment-spectral collaborative fusion decoupling model for concentration prediction, specifically including the following steps: acquiring absorption spectral signals of a specified mixed gas at different temperatures, pressures, and known concentrations, while simultaneously acquiring environmental parameter data to construct a multi-source dataset; and performing denoising, dimensionality reduction, and preprocessing on the multi-source dataset; constructing a self-supervised feature extraction network with adaptive modulation of environmental parameters to achieve deep fusion of spectral and environmental information; utilizing a self-supervised learning mechanism to improve the feature representation ability and generalization performance of the self-supervised feature extraction network; and constructing a BPBO-GRNN adaptive concentration inversion optimization model to achieve the inversion of mixed gas concentration and adaptive optimization of model parameters. This invention can achieve high-precision concentration inversion and stable detection of mixed gases.

Description

Technical Field

[0001] This invention belongs to the field of industrial process control and environmental monitoring, and specifically relates to a method for spectral aliasing decoupling and concentration inversion under the cross-influence of multiple environmental factors. Background Technology

[0002] In the fields of industrial process control and environmental monitoring, particularly in complex and harsh applications such as continuous monitoring of flue gas emissions from thermal power plants, extremely high technical requirements are placed on the accurate, real-time online measurement of gas concentrations. To address this challenge, tunable diode laser absorption spectroscopy (TDLAS) technology, with its inherent advantages of high sensitivity, high selectivity, and rapid online response, is widely regarded as a key technology in this field. However, when applied to the aforementioned complex industrial environments, its measurement accuracy and reliability still face severe technical challenges, limiting its full potential.

[0003] In the flue gas denitrification process of thermal power plants, ammonia is injected into the high-temperature flue gas as a reducing agent to reduce nitrogen oxides to harmless nitrogen. To ensure denitrification efficiency, the amount of ammonia injected is usually slightly excessive, resulting in some unreacted NH3 being discharged with the flue gas, forming "ammonia escape." Simultaneously, incomplete combustion of fuel in the boiler inevitably produces carbon monoxide. Therefore, CO and NH3 must coexist and mix thoroughly in the final exhaust flue. These two gases exhibit partial overlap in their absorption spectra in the commonly used near-infrared monitoring band, resulting in severe spectral aliasing.

[0004] The dynamic nature of industrial processes leads to dramatic and frequent fluctuations in temperature and pressure within flue gas ducts. These environmental parameters are not simply background noise, but profoundly alter the absorption spectral morphology of gas molecules in a coupled and nonlinear manner. Studies show that pressure alters spectral linewidth through collisional broadening, and temperature through Doppler broadening; more importantly, temperature directly modulates the integrated absorption intensity of the spectral lines by changing the molecular energy level population. Crucially, these effects exhibit strong molecular specificity; the spectral morphology of NH3 is extremely sensitive to pressure and temperature, while CO is relatively insensitive. Therefore, the ultimately observed spectral morphology is a complex result of the nonlinear coupling of concentration, pressure, and temperature. This difference and nonlinearity in the responses of different components to environmental parameters further exacerbates the complexity of spectral aliasing and decoupling.

[0005] High-precision spectral analysis of mixed gases under multi-source environmental interference has always been a research hotspot and challenge in the field of industrial sensing. Early research mainly relied on the Lambert-Beer law, using absorption peaks at specific wavelengths to invert concentration. However, in scenarios such as the coexistence of CO and NH3, the absorption spectra of these gases severely overlap in the near-infrared band, with almost no undisturbed independent absorption peaks, causing this type of method to fundamentally fail. To solve the spectral aliasing problem, scholars at home and abroad have introduced chemometric algorithms such as multiple linear regression, principal component regression, and partial least squares. These methods treat the mixed spectrum as a linear superposition of the spectra of each pure component, which can separate overlapping signals to a certain extent. However, the core linear assumption is severely shaken in dynamic industrial environments. Drastic fluctuations in pressure and temperature can cause significant nonlinear broadening and deformation of the spectral lines of molecules such as NH3. Linear models such as PLS cannot effectively model such nonlinear spectral changes driven by environmental parameters. When the operating conditions deviate from their calibration conditions, the model accuracy drops sharply, resulting in poor robustness. Although subsequent research has shifted to nonlinear models such as neural networks, the current mainstream strategy of simply splicing high-dimensional spectral data with low-dimensional environmental data has failed to effectively learn key interference mechanisms due to deep-seated defects such as information overload and feature attribute mismatch.

[0006] In summary, existing spectral inversion methods generally suffer from problems such as sensitivity to temperature and pressure fluctuations, failure of linearity assumptions, and insufficient fusion of multi-source features under complex industrial conditions, making it difficult to achieve high-precision concentration inversion and stable detection of mixed gases. Summary of the Invention

[0007] To address the molecular-specific nonlinear spectral deformation caused by drastic fluctuations in environmental parameters such as temperature and pressure when handling complex industrial scenarios like continuous emission monitoring systems for flue gas from thermal power plants, this invention proposes a method for spectral aliasing decoupling and concentration inversion under the combined influence of multiple environmental factors. This method achieves cross-modal fusion from static environmental input to dynamic control of spectral features through deep characterization and adaptive modulation of multimodal environmental parameters. Furthermore, it incorporates a generalized regression neural network (GRNN) adaptive optimization model based on the Raptor Optimization Algorithm (BPBO), overcoming the limitations of traditional algorithms that are highly dependent on parameters and prone to getting trapped in local optima. Compared to existing technologies, this invention achieves innovative breakthroughs in adaptive compensation and dynamic feature modeling of spectral nonlinear deformation, realizing adaptive compensation and dynamic feature modeling of spectral nonlinear deformation under temperature and pressure coupling interference, providing an original technical path for high-precision detection of multi-component gases in complex industrial environments.

[0008] The technical solution of the present invention is as follows:

[0009] A method for spectral aliasing decoupling and concentration inversion under the combined influence of multiple environmental factors is proposed. An environment-spectral collaborative fusion decoupling model is constructed for concentration prediction. This model includes a multi-source data processing module, a self-supervised feature extraction network, a self-supervised auxiliary decoding module, and a BPBO-GRNN adaptive concentration inversion optimization model. The method specifically includes the following steps:

[0010] Step 1: Collect the absorption spectral signals of the specified mixed gas at different temperatures, pressures, and known concentrations, and simultaneously collect environmental parameter data to construct a multi-source dataset;

[0011] Step 2: Denoise, reduce dimensionality, and preprocess the multi-source dataset using the multi-source data processing module;

[0012] Step 3: Construct a self-supervised feature extraction network with adaptive modulation of environmental parameters to achieve deep fusion of spectral and environmental information and obtain fused feature vectors;

[0013] Step 4: Construct a self-supervised auxiliary decoding module. This module utilizes a self-supervised learning mechanism to improve the feature representation ability and generalization performance of the self-supervised feature extraction network, and fully explore unlabeled data.

[0014] Step 5: Construct a BPBO-GRNN adaptive concentration inversion optimization model to realize the inversion of mixed gas concentration and adaptive optimization of model parameters;

[0015] Step 6: Collect the absorption spectrum signal of the unknown concentration of the gas to be measured under the actual operating conditions of the thermal power plant, and at the same time collect the actual environmental parameter data to construct a multi-source dataset; import the multi-source dataset into the environmental-spectral synergistic fusion decoupling model trained by the laboratory calibration data for concentration prediction.

[0016] Further, the specific process of step 1 is as follows: First, under different temperature and pressure conditions, the absorption spectrum signals of the CO and NH3 mixture with known concentration gradients are collected, and the corresponding environmental parameter data are recorded to construct a multi-source dataset; the multi-source dataset is used as the training set of the model; then, the aliased spectrum signals under unknown concentration conditions are collected within the same temperature and pressure control range and used as the model validation set.

[0017] Furthermore, in step 2, the data processing module includes an intelligent noise reduction module, a preprocessing module, and a dimensionality reduction module; the specific process is as follows:

[0018] Step 2.1: Construct an intelligent denoising module. The intelligent denoising module adopts a pre-trained one-dimensional U-Net convolutional neural network structure. The multi-source dataset is input into the one-dimensional U-Net convolutional neural network for denoising, and the mapping relationship is as follows:

[0019] ;

[0020] in, It is the first High-fidelity spectra of individual samples; It is the first The original spectra of each sample; These are the trainable parameters of a one-dimensional U-Net convolutional neural network; This represents the function mapping corresponding to the one-dimensional U-Net convolutional neural network structure;

[0021] Step 2.2: Introduce a non-parametric distribution alignment mechanism for quantile mapping and construct a preprocessing module. The preprocessing module includes a distribution feature analysis unit and a non-parametric distribution alignment unit. The specific working process is as follows:

[0022] Step 2.2.1: Based on the distribution characteristic analysis unit, perform empirical distribution estimation, and calculate the empirical cumulative distribution function for each spectral feature and environmental parameter:

[0023] ;

[0024] in, The empirical cumulative distribution function; Represents the total number of samples; The feature value threshold; Indicates the first Feature values ​​of each sample; For indicator functions, when Less than or equal to The value is 1 if the condition is met, otherwise it is 0.

[0025] Step 2.2.2: Perform quantile mapping based on non-parametric distribution alignment units to obtain high-fidelity spectra. The quantiles are mapped through the inverse cumulative distribution function of the target distribution to obtain the standardized result. :

[0026] ;

[0027] in, This is the inverse cumulative distribution function of the standard normal distribution;

[0028] Step 2.3: Construct the dimensionality reduction module.

[0029] Furthermore, in step 2.3, the dimensionality reduction module includes an encoder submodule, a decoder submodule, and a feature reconstruction and verification unit. The specific working process is as follows:

[0030] Step 2.3.1: The encoder submodule uses the autoencoder mapping function to process the normalized result. Mapping to a low-dimensional latent space yields low-dimensional feature vectors. :

[0031] ;

[0032] in, Represents the autoencoder mapping function; It is a set of trainable parameters; The compressed spectral dimensions;

[0033] Step 2.3.2: The decoder submodule reconstructs the low-dimensional feature vectors back to the original space using the decoder's inverse mapping function.

[0034] ;

[0035] in, This represents the inverse mapping function of the decoder; This is the set of trainable parameters for the decoder; The result of the reconstruction;

[0036] Step 2.3.3: The goal of the feature reconstruction verification unit is to minimize the mean square error between the input and the reconstructed spectrum. By minimizing the loss function, the network can learn the nonlinear feature structure of the spectral data; the loss function... for:

[0037] ;

[0038] in, and They represent the first Standardization and reconstruction results for each sample; This represents the L2 norm.

[0039] Furthermore, in step 3, the self-supervised feature extraction network includes an environment modulation branch, a spectral backbone network, and a feature fusion module; the specific process is as follows:

[0040] Step 3.1: Construct an environment modulation branch to obtain high-level environment feature vectors. Furthermore, a multi-head self-attention mechanism and a gating control structure are introduced;

[0041] Step 3.2: Construct a spectral backbone network consisting of multiple environmentally adaptive residual blocks stacked in series, and extract environmentally adaptive features from the spectral data step by step to obtain the environmentally adaptive spectral feature vector. ;

[0042] Step 3.3: Construct a feature fusion module to perform feature fusion, specifically by fusing the environmental adaptive spectral feature vector. High-level environmental feature vector The features are concatenated to obtain a fused feature vector. :

[0043] ;

[0044] in, This refers to the vector concatenation operation along the feature dimension.

[0045] Furthermore, in step 3.1, the environment modulation branch includes a multimodal environment parameter self-attention representation submodule, an environment feature deep reconstruction submodule, and an adaptive control unit. The specific working process is as follows:

[0046] Step 3.1.1: Construct a multimodal environment parameter self-attention representation submodule based on a multi-head self-attention mechanism to obtain environmental feature representations. :

[0047] ;

[0048] in, For the Softmax function; , , These are the query vector, key vector, and value vector corresponding to each attention head; For transpose; The dimension of the key vector;

[0049] Step 3.1.2: Construct a deep environmental feature reconstruction submodule using a multi-layer residual network structure; As the input to the deep reconstruction submodule of environmental features, let the initial feature vector be denoted as . This submodule consists of multiple stacked residual blocks; for the first... The residual blocks are calculated as follows:

[0050] ;

[0051] in, For the first The feature vector output by the residual block is used as the first... Input of each residual block; For the first The feature vectors output by each residual block; It is the ReLU activation function; For the first One residual block; The total number of residual blocks; after The hierarchical feature extraction and superposition of layer residual blocks yields the high-level environmental feature vector. :

[0052] ;

[0053] in, For the first The feature vectors output by each residual block;

[0054] Step 3.1.3: Construct an adaptive control unit composed of multiple fully connected layers. For the first layer in the spectral backbone network... An environment adaptive residual block that needs to be modulated; the adaptive control unit uses high-level environmental feature vectors. Generate the corresponding environmental modulation parameters; specifically as follows:

[0055] First, the high-level environmental feature vector The inputs are fed into three independent fully connected layers to generate preliminary environmental modulation parameters and a gating signal. The preliminary environmental modulation parameters include a preliminary scaling factor and a preliminary offset factor.

[0056] Then, based on the gated signal, the preliminary environmental modulation parameters are weighted and fused to obtain the final environmental modulation parameters:

[0057] ;

[0058] ;

[0059] in, For the first Gating signals for an environment-adaptive residual block; , The first Preliminary scaling factor and preliminary offset factor for an environment-adaptive residual block; , The first The final scaling factor and final offset factor of each environment-adaptive residual block; This indicates element-wise multiplication; It is a unit vector;

[0060] Finally, the obtained environmental modulation parameters are applied to the feature mapping of the convolutional block to achieve adaptive affine transformation.

[0061] Furthermore, in step 3.2, each environment-adaptive residual block adopts a residual structure, consisting of a main path and a shortcut path; the calculation formula for the main path is:

[0062] ;

[0063] ;

[0064] in, This is the output feature map for the first stage; This is the first batch of standardized operations; This represents a one-dimensional convolution operation; This is the output of the previous environment adaptive residual block; The final output of the main path; This is the second batch of standardized operations; This represents a one-dimensional convolution operation;

[0065] The formula for calculating the shortcut path is:

[0066] ;

[0067] in, Output the shortcut path; This indicates a one-dimensional convolution operation using a 1×1 convolution kernel; The dimension of the feature map;

[0068] After adding the outputs of the main path and the shortcut path element by element, the spectral feature map is obtained. :

[0069] ;

[0070] It is the output of a single environment-adaptive residual block, for the . An environment-adaptive residual block:

[0071] ;

[0072] in, For the first Spectral feature map of the output of an environment-adaptive residual block; For the first An environment-adaptive residual block; For the first Spectral feature map of the output of an environment-adaptive residual block;

[0073] via After sequential processing of each environment-adaptive residual block, the final spectral feature map is obtained. ;

[0074] A global average pooling operation is used for transformation to obtain the final environment-adaptive spectral feature vector. :

[0075] ;

[0076] in, This is a global average pooling operation.

[0077] Furthermore, in step 4, the self-supervised auxiliary decoding module includes an auxiliary decoder, a self-supervised constraint unit, and a feature optimization and verification unit. The specific working process is as follows:

[0078] Step 4.1: Fuse the feature vectors The input auxiliary decoder extracts physically consistent deep feature representations. The auxiliary decoder reconstructs the spectral data through a multi-layer fully connected structure, remapping the highly compressed features back to the original spectral space, and finally generating the reconstructed spectrum.

[0079] Step 4.2: Define the reconstruction loss function in the self-supervised constraint unit. The difference between the original spectrum and the reconstructed spectrum is measured:

[0080] ;

[0081] in, For the first The original spectra of each sample; For the first The reconstructed spectra of each sample; the number of original spectra corresponds to the number of reconstructed spectra.

[0082] Step 4.3: Construct a feature optimization and verification unit to continuously monitor reconstruction performance and adjust the structural parameters of the self-supervised feature extraction network during the training phase, ensuring that the model obtains stable feature representation under unlabeled data conditions;

[0083] During training, the loss function is reconstructed. As a self-supervised signal, all trainable parameters, including the auxiliary decoder, feature fusion module, environmental modulation branch, and spectral backbone network, are jointly optimized through an end-to-end backpropagation algorithm. During the training phase, an early stopping strategy is adopted to determine the optimal stopping point based on the reconstruction error of the validation set to prevent overfitting. After training, the auxiliary decoder is removed, and only the optimized self-supervised feature extraction network is retained.

[0084] Furthermore, in step 5, the BPBO-GRNN adaptive concentration inversion optimization model includes a GRNN concentration prediction submodule and a BPBO adaptive optimization submodule, the specific process of which is as follows:

[0085] Step 5.1: Construct a GRNN concentration prediction submodule based on the generalized regression neural network structure. Use a Gaussian kernel function to weight the feature similarity model and a smoothing factor to control the kernel bandwidth, thereby obtaining concentration prediction results that are sensitive to changes in both spectral and environmental characteristics; define the... The fused feature vector of each training sample Given a fused feature vector of any input, for With the training samples The similarity is measured by Euclidean distance. For the first The fused feature vectors of the training samples are transformed into kernel weights using a Gaussian kernel function:

[0086] ;

[0087] in, for and The kernel weights between them; It is an exponential function; It is the Euclidean distance function; It is a smoothing factor;

[0088] Concentration prediction value for:

[0089] ;

[0090] in, The total number of training samples; For the first The true concentration labels of each training sample;

[0091] Step 5.2: Construct the BPBO global adaptive optimization submodule. This submodule proposes a generalized regression neural network adaptive optimization mechanism based on the raptor optimization algorithm. By simulating the global exploration and local exploitation behavior of raptors during the predation process, it dynamically searches for the smoothing factor to find the optimal smoothing factor.

[0092] Furthermore, the specific process of step 5.2 is as follows:

[0093] First, the smoothing factor The optimization process is modeled as a minimization problem, with a smoothing factor. The individual positions are updated using the Raptor Optimization algorithm, which then searches for the optimal smoothing factor. For any given... Value, its fitness The calculation is as follows:

[0094] ;

[0095] in, The number of samples in the validation set; For the first One verification sample; For the first The true concentration label of the validation sample; Indicates the use of the current The constructed GRNN concentration prediction submodule predicts the concentration values ​​of the validation samples;

[0096] Set population size With maximum number of iterations Set the smoothing factor The search space is a preset range ,in , They are respectively The minimum and maximum values; random initialization within the search space. The positions of individual raptors form the initial population; the fitness value of each individual is calculated, and the position of the individual with the lowest fitness value is recorded as the current global optimum. ;

[0097] In each iteration of the Raptor optimization algorithm, a random number within the interval [0, 1] is generated. Preset a threshold As a switching threshold between the exploration and development phases; if the updated individual position exceeds the search space, it is truncated and mapped back to the search space range; the specific iterative process of the Raptor optimization algorithm is as follows:

[0098] like The algorithm will then enter the global exploration phase; the individual position update formula is as follows:

[0099] ;

[0100] in, For the first The individual in the first The position of the next iteration corresponds to the [number]th iteration. The smoothing factor value at the next iteration; The updated position; It is a random number in the range [-1, 1];

[0101] like The algorithm then enters the local development phase; this phase contains two sub-phases, determined by the number of iterations. and The relationship determines; when At that time, execute the dive-and-pursue strategy:

[0102] ;

[0103] in, For the first The average of all individual positions in the next iteration; It is a random number within the range [0, 1].

[0104] when At that time, execute a precise attack strategy:

[0105] ;

[0106] in, For the Lévy flight function;

[0107] After each location update, the fitness value is recalculated. If so, update the individual's position;

[0108] When the number of iterations reaches Finally, output the final globally optimal position. As the optimal smoothing factor;

[0109] Obtaining the optimal smoothing factor Subsequently, the BPBO-GRNN adaptive concentration inversion optimization model was based on Predict the concentration of any input fused feature vector to obtain the optimal concentration prediction value. :

[0110] .

[0111] The beneficial technical effects brought about by this invention are as follows.

[0112] 1. This invention constructs a dual-branch deep fusion structure of spectral and environmental parameters, and innovatively designs an environment adaptive residual block (EAR-Block). The dynamic modulation factor generated by the environmental parameters is embedded into the spectral feature extraction process to achieve dynamic matching and weight balance between high-dimensional spectra and low-dimensional environmental parameters, thereby improving the robustness of feature expression and effectively solving the problem of spectral information overshadowing environmental features in traditional methods.

[0113] 2. A multi-modal environmental parameter self-attention representation submodule is constructed based on a multi-head self-attention mechanism. By combining the nonlinear mapping and reconstruction of a multi-layer residual network, a high-dimensional feature embedding with global dependence is generated, realizing the deep fusion of spectral and environmental information.

[0114] 3. A self-supervised auxiliary decoding module is introduced to learn the physical laws of the spectrum and the environmental influence patterns using unlabeled spectral data (i.e., to achieve feature learning of unlabeled samples using spectral reconstruction constraints). This improves the model's adaptability and generalization ability to complex temperature and pressure coupling interference without requiring a large number of labeled samples.

[0115] 4. The Raptor Optimization (BPBO) algorithm is used to globally and adaptively optimize the smoothing factor of the generalized regression neural network (GRNN), avoiding the problems of low efficiency and local optima in manual parameter tuning, realizing intelligent optimization of model parameters, and significantly improving the accuracy and convergence performance of the concentration inversion model.

[0116] 5. Construct an end-to-end trainable multi-level fusion inversion framework, which significantly improves the real-time performance and engineering deployability of the model. It enables rapid feature extraction and real-time concentration prediction under complex working conditions, making it easy to embed into industrial monitoring systems for online applications. This significantly improves the overall response speed and engineering practical value of the system. Attached Figure Description

[0117] Figure 1 This is a flowchart of the spectral aliasing decoupling and concentration inversion method under the cross-influence of multiple environmental factors according to the present invention.

[0118] Figure 2 The diagram shows the denoising results and residual analysis of this invention; where (a) is the original spectrum, (b) is the denoised spectrum, and (c) is the residual analysis diagram.

[0119] Figure 3 This is a gradient comparison diagram of the actual and predicted concentrations of NH3 after applying the method of this invention.

[0120] Figure 4 This is a gradient comparison diagram of the actual CO concentration and the predicted CO concentration after applying the method of the present invention.

[0121] Figure 5 The residual distribution diagrams of NH3 and CO after applying the method of the present invention are shown; where (a) is the residual distribution diagram of NH3 and (b) is the residual distribution diagram of CO.

[0122] Figure 6 This is a gradient comparison between the actual and predicted NH3 concentrations after using a compensation model without a self-supervised feature extraction network.

[0123] Figure 7 This is a gradient comparison between the actual and predicted CO concentrations after using a compensation model without a self-supervised feature extraction network.

[0124] Figure 8 The residual distributions of NH3 and CO are shown in the diagrams after using the compensation model without self-supervised feature extraction network; where (a) is the residual distribution of NH3 and (b) is the residual distribution of CO. Detailed Implementation

[0125] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0126] This invention innovatively proposes a deep collaborative fusion technology based on self-supervised feature extraction, deep characterization of environmental parameters, and joint optimization of BPBO-GRNN. This scheme realizes for the first time a dynamic integrated modeling mechanism for environmental information, spectral features, and inversion models, breaking through the technical bottlenecks of traditional methods such as the separate modeling of spectral features and environmental parameters, the failure of linear assumptions, and reliance on manual parameter tuning. It aims to solve the problem of spectral nonlinear deformation and aliasing caused by drastic fluctuations in environmental parameters such as temperature and pressure, and overcome the problem of information saturation of low-dimensional environmental parameters by high-dimensional spectra, thereby achieving highly robust inversion of target gas concentrations in severely aliased spectra.

[0127] To address the challenges of spectral aliasing decoupling and concentration inversion of CO and NH3 under temperature-pressure coupling interference, this invention innovatively constructs a multi-module collaborative intelligent spectral processing and concentration inversion method. The core innovations of this method are mainly concentrated in the following modules: a multi-source data preprocessing module, which introduces a non-parametric distribution alignment mechanism based on quantile mapping, overcoming the limitations of traditional Z-score normalization which relies on the Gaussian assumption; a self-supervised feature extraction network, which constructs a deep fusion mechanism between environmental parameters and spectral features, introducing self-attention and gating control structures to achieve self-learning and dynamic adjustment of environmental features, while simultaneously employing an environmental adaptive residual block to embed an environmental adaptive modulation mechanism in the spectral backbone network to achieve dynamic compensation and fusion of spectral and environmental features; and a BPBO-GRNN adaptive concentration inversion optimization model, based on the Raptor optimization algorithm, to achieve intelligent optimization and globally optimal configuration of GRNN model parameters.

[0128] This invention achieves high-precision gas concentration inversion and adaptive compensation for environmental disturbances through a deep learning model integrating multimodal environmental parameters and spectral data. The spectral data is processed by an intelligent denoising module, which extracts deep features using a one-dimensional U-Net convolutional neural network while preserving spectral details and absorption peak structures. A dimensionality reduction module compresses the spectral data into low-redundancy features for easier subsequent processing. In the self-supervised feature extraction network block, environmental information undergoes high-dimensional feature embedding and nonlinear mapping through a self-attention mechanism and residual network to generate a dynamically modulated signal, which is then deeply fused with the spectral features. The environmental adaptive residual block further calibrates the spectral features to achieve dynamic compensation. The self-supervised auxiliary decoding module optimizes the network feature representation through self-supervised learning and pre-trains using unlabeled data to improve the model's generalization ability. Subsequently, the BPBO-GRNN adaptive concentration inversion optimization model combines GRNN and Raptor optimization algorithm to achieve high-precision concentration prediction by adaptively adjusting the smoothing factor. This invention effectively improves the accuracy and robustness of concentration inversion through multi-level feature fusion and optimization, exhibiting good adaptability and generalization ability in complex environments.

[0129] This invention addresses the complex operating conditions of flue gas emissions from thermal power plants, where temperature and pressure are uncertain. It selects typical mixed gases CO and NH3 as target objects and constructs an experimental system for spectral aliasing decoupling and concentration inversion under the influence of multiple environmental factors. The method is not only applicable to emission monitoring during the denitrification process of thermal power plants but can also be extended to other multi-component gas aliasing spectral measurement scenarios affected by environmental disturbances.

[0130] Experimental Conditions and Testing Environment: The experiment employed a self-built controlled environment spectral acquisition system. The system mainly includes a temperature control unit, a pressure regulation unit, a gas mixing module, and a spectral detection module. The temperature control unit is continuously adjustable within the range of 0–300℃; the pressure regulation unit can stably control the pressure within the range of 0.5–3.0 atm. Real-time monitoring and closed-loop feedback of operating parameters are achieved through high-precision temperature controllers and pressure sensors, ensuring the accuracy, controllability, and repeatability of the experimental environment parameters.

[0131] This invention constructs an environment-spectral synergistic fusion decoupled model for concentration prediction, mainly including a multi-source data processing module, a self-supervised feature extraction network, a self-supervised auxiliary decoding module, and a BPBO-GRNN adaptive concentration inversion optimization model. Figure 1 As shown, the method of the present invention mainly includes the following steps:

[0132] Step 1: Collect the absorption spectral signals of the specified mixed gas at different temperatures, pressures, and known concentrations, and simultaneously collect environmental parameter data to construct a multi-source dataset;

[0133] First, under different temperature (25–250℃) and pressure (0.8–2.5 atm) conditions, absorption spectra of CO and NH3 mixtures with known concentration gradients were collected, along with corresponding environmental parameter data, to construct a multi-source dataset. This dataset served as the training set for the model, used for feature learning and parameter optimization. Subsequently, aliased spectra under unknown concentration conditions were collected within the same temperature and pressure control range, serving as the model validation set to evaluate inversion accuracy and generalization performance.

[0134] During the experiment, the volume fractions of CO and NH3 were controlled using a standard gas mixing device, with CO concentration gradients ranging from 320 ppm to 1200 ppm and NH3 concentration gradients ranging from 320 ppm to 2000 ppm. The temperature and pressure settings covered the typical variation range of flue gas emission pipelines in thermal power plants, effectively simulating the gas mixing and spectral interference characteristics under real and complex operating conditions.

[0135] The collected spectra exhibit a significant absorption overlap region in the near-infrared band, manifested as partial aliasing of CO and NH3 spectral lines. Temperature and pressure variations leading to spectral line broadening and nonlinear changes in absorption intensity further exacerbated spectral interference, providing typical validation conditions for subsequent data denoising, dimensionality reduction, and concentration inversion modeling.

[0136] Step 2: Denoising, dimensionality reduction, and preprocessing of the multi-source dataset using the multi-source data processing module; the data processing module includes an intelligent denoising module, a preprocessing module, and a dimensionality reduction module. The specific process is as follows:

[0137] Step 2.1: Construct an intelligent denoising module for intelligent denoising of multi-source datasets;

[0138] The intelligent denoising module adopts a pre-trained one-dimensional U-Net convolutional neural network structure, which mainly includes an encoder, a decoder, and skip connection units. The encoder extracts deep feature information of the spectrum through multi-layer one-dimensional convolution and downsampling operations. The decoder reconstructs the features layer by layer through upsampling and convolution operations. Skip connection units are used to transfer and fuse shallow and deep features between corresponding layers to maintain the integrity of spectral details and absorption peak structure.

[0139] A multi-source dataset is input into a one-dimensional U-Net convolutional neural network for denoising. This module autonomously learns the mapping relationship from the noisy spectral domain to the clean spectral manifold from the data through a highly nonlinear mapping function, transforming any input raw spectrum into a denoised high-fidelity spectrum. The mapping relationship is as follows:

[0140] ;

[0141] in, It is the first High-fidelity spectra of individual samples; It is the first The original spectra of each sample; These are the trainable parameters of a one-dimensional U-Net convolutional neural network; This represents the function mapping corresponding to the one-dimensional U-Net convolutional neural network structure; specifically, the internal structure of this convolutional neural network is optimized for the input data.

[0142] The encoder progressively compresses the input spectrum through multiple layers of one-dimensional convolution and downsampling operations to extract deep feature information. Assume the encoder of this invention comprises... One-dimensional convolution and downsampling operations in the first layer, with the input feature map of the first layer. By the The original spectra of each sample Direct import, i.e. ;No. The input feature map of layer 1 is the first layer. Output feature map of the layer , No. The output feature map of the layer is The specific calculation process is as follows:

[0143] ;

[0144] ;

[0145] in, For the first Layers are used for feature maps of skip connections; This represents a one-dimensional convolution operation; It is a non-linear activation function; This is a max pooling operation used to compress and extract features.

[0146] The decoder functions in the opposite way to the encoder; it is responsible for accurately reconstructing the pure spectral signal from abstract feature representations. The decoder is configured accordingly. Layer upsampling operation. The decoder starts from the network layer... Output feature map of the layer Begin, through Layer-by-layer upsampling recovers the original dimensions of the spectrum. Let the decoder... The input feature map of the layer is Upsampling is performed through a transposed convolutional layer, and the calculation is as follows:

[0147] ;

[0148] in, For the decoder Upsampled feature map of the layer; This is a transposed convolutional layer.

[0149] Based on the upsampled feature maps, the skip connection unit first concatenates the high-resolution feature maps of the corresponding layers of the encoder along the channel dimension through long-distance skip connections; then, it directly supplies the shallow features of the encoder containing precise location information to the decoder, thereby achieving effective fusion of deep feature information and shallow detail information to generate the final output of the current layer of the decoder:

[0150] ;

[0151] ;

[0152] in, Indicates the encoder's first... The feature map output by the layer, the encoder's first layer Layer and Decoding Layer The layers are relatively symmetrical; For the decoder Feature map after layer fusion; This refers to the vector concatenation operation along the feature dimension; For the decoder The final output of the layer; This represents a one-dimensional convolution operation, used for preliminary feature extraction and channel integration of the feature map concatenated by skip connections; This represents a one-dimensional convolution operation, used in... Further feature reconstruction is performed based on the extracted features;

[0153] go through The upsampling and skip fusion of the layers, and finally the output convolutional layer, maps the multi-channel feature map back to a single channel, resulting in a denoised spectrum that is completely consistent with the input dimension and has a significantly improved signal-to-noise ratio:

[0154] ;

[0155] in, This indicates a one-dimensional convolution operation using a 1×1 kernel, used to compress multi-channel feature maps and map them to a single-channel denoised spectrum. For the decoder The final output of the layer;

[0156] The U-Net convolutional neural network employs a structure where the encoder downsamples to extract spectral features, the decoder upsamples to reconstruct the spectrum, and skip connections fuse multiple layers of features. The denoising results and residual analysis are as follows: Figure 2 As shown, the overall trend of the denoised spectrum remains highly consistent with the original spectrum, the peak structure is well preserved, and high-frequency noise components are significantly reduced. The residual plot exhibits an approximately zero-mean random distribution, indicating that the model effectively removes random noise without introducing significant systematic bias, thus verifying the superiority of the proposed method in terms of spectral detail preservation and noise suppression.

[0157] Step 2.2: Construct a preprocessing module to maintain the intrinsic rank structure of the data under non-Gaussian, skewed, and multimodal distribution conditions. This module overcomes the limitation of traditional Z-score standardization relying on the Gaussian assumption by introducing a non-parametric distribution alignment mechanism of quantile mapping (i.e., the limitation of traditional Z-score standardization that only aligns the mean and variance of the data and assumes that the data follows a Gaussian distribution).

[0158] This module mainly includes a distribution feature analysis unit and a non-parametric distribution alignment unit. The distribution feature analysis unit is used to identify and characterize the statistical distribution characteristics of different data sources, including skewness, multimodality, and differences in distribution shape. The non-parametric distribution alignment unit adopts a distribution alignment strategy based on quantile mapping. By constructing an empirical cumulative distribution function, it maps different modal data to a unified target distribution space, thereby achieving standardization and consistency at the overall distribution level.

[0159] The specific working process of the preprocessing module is as follows:

[0160] Step 2.2.1: Based on the distribution feature analysis unit, perform empirical distribution estimation. Calculate the empirical cumulative distribution function for each spectral feature and environmental parameter to represent the quantile position of the input feature value in the sample distribution. The specific definition is as follows:

[0161] ;

[0162] in, The empirical cumulative distribution function; Represents the total number of samples; The feature value threshold; Indicates the first Feature values ​​of each sample; For indicator functions, when Less than or equal to If the value is 1, then the value is 0; otherwise, the value is 0.

[0163] Step 2.2.2: Perform quantile mapping based on non-parametric distribution alignment units to obtain high-fidelity spectra. The quantiles are mapped through the inverse cumulative distribution function of the target distribution to obtain the standardized result. :

[0164] ;

[0165] in, This is the inverse cumulative distribution function of the standard normal distribution. This method makes no distributional assumptions about the original data distribution; instead, it directly maps the quantile information of each feature in the training set to a unified target distribution through the empirical cumulative distribution function, providing a reliable input foundation for subsequent autoencoder dimensionality reduction and bi-branch fusion modeling.

[0166] Step 2.3: To facilitate subsequent feature extraction and fusion, this invention constructs a dimensionality reduction module. This module, based on an autoencoder structure, performs nonlinear compression representation on high-dimensional spectral data, overcoming the limitation of traditional principal component analysis, which can only extract linear features. The compressed spectral features generated by this module not only have low redundancy but also retain the key structural characteristics of the spectral sequence. This module includes an encoder submodule, a decoder submodule, and a feature reconstruction and verification unit.

[0167] The encoder submodule, consisting of multiple fully connected layers, is used to perform feature compression on the standardized high-dimensional spectral data, autonomously learn the main information features in the spectral signal, and generate representative low-dimensional feature vectors.

[0168] The decoder submodule, which is symmetrical to the encoder structure, is used to reconstruct spectral data with the same spectral dimension as the original spectral data based on low-dimensional feature vectors.

[0169] The feature reconstruction and verification unit achieves automatic updates of the parameters of each layer of the encoder and decoder by minimizing the reconstruction error, thereby obtaining a stable and compact low-dimensional representation while maintaining the main physical features of the spectrum.

[0170] The specific working process of the dimensionality reduction module is as follows:

[0171] Step 2.3.1: The encoder submodule uses the autoencoder mapping function to process the normalized result. Mapping to a low-dimensional latent space yields low-dimensional feature vectors. The mapping relationship is as follows:

[0172] ;

[0173] in, Represents the autoencoder mapping function; It is a set of trainable parameters; This represents the compressed spectral dimensions.

[0174] Step 2.3.2: The decoder submodule then reconstructs the low-dimensional feature vectors back to the original space using the decoder's inverse mapping function to maintain data consistency.

[0175] ;

[0176] in, Represents the inverse mapping function of the decoder, ensuring The features retained in the original spectrum are sufficient to reflect key physical information such as the position, shape, and full width at half maximum of the absorption peaks; This is the set of trainable parameters for the decoder; This is the result of the reconstruction.

[0177] Step 2.3.3: The goal of the feature reconstruction verification unit is to minimize the mean square error (MSE) between the input and the reconstructed spectrum. By minimizing the loss function, the network can learn the nonlinear feature structure of the spectral data. Loss function It can be represented as:

[0178] ;

[0179] in, The total number of samples; and They represent the first Standardization and reconstruction results for each sample; Let L2 norm represent the difference between the input data and the reconstructed data. Minimizing the reconstruction loss helps the encoder preserve the local correlations and global structure of the spectral signal during dimensionality reduction, resulting in a lower-dimensional feature vector. It retains the topological properties of the spectral sequence, thus enabling it to serve as the convolutional input for subsequent environment-adaptive residual blocks (EAR-Block).

[0180] Step 3: Construct a self-supervised feature extraction network with adaptive modulation of environmental parameters to achieve deep fusion of spectral and environmental information and obtain fused feature vectors;

[0181] The self-supervised feature extraction network includes an environment modulation branch, a spectral backbone network, and a feature fusion module. The environment modulation branch further includes a multimodal environment parameter self-attention representation submodule, an environment feature deep reconstruction submodule, and an adaptive control unit. The specific working process of the self-supervised feature extraction network is as follows:

[0182] Step 3.1: The environment modulation branch constructs a deep fusion mechanism between multimodal environmental parameters and spectral features. By introducing a multi-head self-attention mechanism and a gating control structure, a unified framework for self-learning and dynamic allocation of environmental features is formed, which is used to realize the dynamic modulation and adaptive compensation of environmental parameters in the spectral feature extraction process. The specific working process of the environment modulation branch is as follows:

[0183] Step 3.1.1: Construct a multi-modal environmental parameter self-attention representation submodule based on the multi-head self-attention mechanism. By calculating the correlation and interaction between environmental parameters, generate a high-dimensional feature embedding that can reflect the importance distribution of each parameter under different working conditions, providing an accurate and robust environmental representation basis for subsequent modulation.

[0184] This invention introduces a multi-head self-attention mechanism to achieve dynamic correlation modeling between environmental parameters. This mechanism learns the correlations between parameters and assigns context-dependent weights to each environmental parameter, thereby obtaining an embedded representation with globally dependent features. The query matrix, key matrix, and value matrix are calculated for the quantile-mapped environmental parameter data. Similarity between parameters is calculated using scaled dot product attention, and the multi-head attention results are concatenated and linearly transformed to obtain the environmental feature representation. This environmental feature representation explicitly encodes the interactions and relative importance between various environmental parameters, thereby providing a more accurate representation of environmental information for subsequent modulation of spectral signals and concentration inversion.

[0185] ;

[0186] in, For the Softmax function; , , For each attention head, there are query vectors, key vectors, and value vectors; For transpose; The dimension of the key vector;

[0187] Step 3.1.2: A multi-layer residual network structure is used to construct a deep reconstruction submodule of environmental features. Nonlinear mapping and high-order reconstruction of environmental features are performed, which enhances the expressive power and stability of the features and realizes the structured transformation from static parameter representation to dynamic modulation signal, laying the foundation for the adaptive control of the model in the environment.

[0188] Will As the input to the deep reconstruction submodule of environmental features, let the initial feature vector be denoted as . This submodule consists of multiple stacked residual blocks. For the first... Each residual block, while preserving input feature information, learns the residual information between the input and target features through nonlinear mapping, thereby achieving feature enhancement and maintaining gradient stability in the deep network. Its calculation is as follows:

[0189] ;

[0190] in, For the first The feature vector output by the residual block is used as the first... Input of each residual block; For the first The feature vectors output by each residual block; It is the ReLU activation function; For the first Each residual block consists of a nonlinear transform subnetwork composed of two fully connected layers and a ReLU activation function, used to nonlinearly reconstruct environmental features; This represents the total number of residual blocks. After... The hierarchical feature extraction and superposition of layer residual blocks yields the high-level environmental feature vector. :

[0191] ;

[0192] in, For the first The feature vectors output by each residual block;

[0193] This high-level environmental feature vector further characterizes the deep dependencies between different environmental parameters, providing structured environmental characterization inputs for subsequent spectral signal modulation and concentration inversion modules.

[0194] Step 3.1.3: Construct an adaptive modulation unit consisting of multiple fully connected layers to generate modulation parameters such as scaling factors, offset factors, and gating signals. This unit achieves dynamic adjustment of the spectral feature extraction process through a continuously differentiable gating mechanism, ensuring the flexibility of feature fusion and the effectiveness of environmental compensation.

[0195] For the first in the spectral backbone network An environment adaptive residual block that needs to be modulated; the adaptive control unit uses high-level environmental feature vectors. Generate the corresponding environmental modulation parameters. This process includes the following steps:

[0196] Step 3.1.3.1: Transfer the high-level environmental feature vector The inputs are fed into three independent fully connected layers to generate preliminary environmental modulation parameters and a gating signal. The preliminary environmental modulation parameters include a preliminary scaling factor and a preliminary offset factor.

[0197] Step 3.1.3.2: Based on the gated signal, the preliminary environmental modulation parameters are weighted and fused to obtain the final environmental modulation parameters, which are calculated as follows:

[0198] ;

[0199] ;

[0200] in, For the first Gating signals for an environment-adaptive residual block; , The first Preliminary scaling factor and preliminary offset factor for an environment-adaptive residual block; , The first The final scaling factor and final offset factor of each environment-adaptive residual block; This indicates element-wise multiplication; It is a unit vector.

[0201] Step 3.1.3.3: Apply the obtained environmental modulation parameters to the feature mapping of the convolutional block to achieve adaptive affine transformation. When When it approaches 0, Approaching 1 When the value approaches 0, modulation is turned off, and the feature map retains its original state; when... When it approaches 1, and They approach their initial values ​​respectively and That is, modulation is fully turned on.

[0202] Step 3.2: Construct a spectral backbone network consisting of multiple cascaded and stacked environment-adaptive residual blocks (EAR-Blocks). Perform step-by-step environment-adaptive feature extraction on the spectral data to obtain environment-adaptive spectral feature vectors. The spectral backbone network incorporates an environment-adaptive modulation mechanism, achieving deep fusion and dynamic compensation of spectral and environmental information. This is a key technological innovation of this invention in improving model robustness and inversion accuracy. In this invention, [the network is constructed using...] EAR-Blocks are stacked in series to form a deep spectral backbone network. Each EAR-Block adopts a residual structure, consisting of a main path and a shortcut path. As the basic unit of the spectral backbone network, the EAR-Block achieves dynamic calibration of spectral features by embedding environmental modulation mechanisms into the deep residual structure. The main path extracts spectral features step-by-step through two one-dimensional convolutions and batch normalization. After the first batch normalization output, a modulation factor generated by environmental parameters is introduced to perform a channel-by-channel affine transformation on the features, achieving environmentally adaptive adjustment of spectral features. The main path first performs the first convolution and batch normalization, outputting a feature map modulated by the environment in the first stage. Then, a second convolution transformation is performed on the already modulated feature map to learn more complex feature combination relationships. The calculation process of each EAR-Block is as follows:

[0203] First, calculate the main path:

[0204] ;

[0205] ;

[0206] in, This is the output feature map for the first stage; This is the first batch of standardized operations; This represents a one-dimensional convolution operation used to extract the first-stage features of the main path; This is the output of the previous environment-adaptive residual block, where the input of the first environment-adaptive residual block is... Low-dimensional feature vectors ,Right now . The final output of the main path; This is the second batch of standardized operations; This represents a one-dimensional convolution operation used for further feature mapping of the first-stage features.

[0207] The shortcut path is responsible for ensuring stable information transmission and is integrated with the output of the main path. The calculation is as follows:

[0208] ;

[0209] in, Output the shortcut path; This indicates a one-dimensional convolution operation using a 1×1 kernel, used to adjust the number of channels or sequence dimensions in the shortcut path to align with the final output of the main path. The dimensions remain consistent; The dimension of the feature map;

[0210] After element-wise addition of the outputs from the main path and the shortcut path, the resulting ReLU activation yields the environmentally compensated robust spectral feature map. This enables direct information connection and gradient stabilization.

[0211] ;

[0212] This is the output of a single EAR-Block, representing the intermediate state of the features after one environment-adaptive transformation in the deep network. The first EAR-Block receives the low-dimensional feature vector after dimensionality reduction by the autoencoder. As its initial input The first set of parameters generated by the environmental modulation branch. Modulation was performed to obtain the spectral feature map of the first environment-adaptive residual block output. ;in, , These are the final scaling factor and final offset factor of the first environment-adaptive residual block, respectively. The second EAR-Block receives the output of the previous environment-adaptive residual block as its input and is determined by the second set of parameters. Modulation was performed to obtain the spectral feature map of the second environment-adaptive residual block output. And so on, information propagates level by level along the network. For the first... There are EAR-Blocks, including:

[0213] ;

[0214] in, For the first Spectral feature map of the output of an environment-adaptive residual block; For the first An environment-adaptive residual block; For the first Spectral feature map of the output of an environment-adaptive residual block;

[0215] via After sequential processing of each EAR-Block, the final spectral feature map is obtained. This feature map contains the deepest level of environmental adaptive spectral information.

[0216] To convert the aforementioned one-dimensional feature map into a fixed-length one-dimensional feature vector for easier subsequent feature fusion, a global average pooling operation is used to achieve the conversion, resulting in the final environment-adaptive spectral feature vector:

[0217] ;

[0218] in, This is an environmentally adaptive spectral feature vector; This is a global average pooling operation. This structure ensures stable gradient propagation while enabling dynamic control and enhancement of spectral features through environmental modulation.

[0219] Step 3.3: Construct a feature fusion module to perform feature fusion, specifically by fusing the environmental adaptive spectral feature vector. High-level environmental feature vector The features are concatenated to obtain a fused feature vector. :

[0220] ;

[0221] in, This refers to the vector concatenation operation along the feature dimension;

[0222] This dual-fusion architecture of implicit modulation and explicit splicing ensures that the output features are highly distinguishable from concentration information and environmental interference, providing a more robust feature foundation for subsequent prediction tasks.

[0223] Step 4: To enable the deep feature extraction network to learn the intrinsic physical laws and environmental influences of the spectrum without manually labeled data, this invention constructs a self-supervised auxiliary decoding module. This module utilizes a self-supervised learning mechanism to improve the feature representation ability and generalization performance of the self-supervised feature extraction network, and can be used to fully mine unlabeled data. This module mainly consists of an auxiliary decoder, a self-supervised constraint unit, and a feature optimization and verification unit. The specific working process is as follows:

[0224] Step 4.1: Fuse the feature vectors The input auxiliary decoder extracts physically consistent deep feature representations. The auxiliary decoder reconstructs the spectral data through a multi-layer fully connected structure, remapping the highly compressed features back to the original spectral space, and finally generating the reconstructed spectrum.

[0225] Step 4.2: To drive the parameter learning of the network, define the reconstruction loss function in the self-supervised constraint unit. The difference between the original spectrum and the reconstructed spectrum is measured to guide the parameter optimization of the feature extraction network; the calculation is as follows:

[0226] ;

[0227] in, For the first The original spectra of each sample; For the first The reconstructed spectra of each sample; the number of original spectra corresponds to the number of reconstructed spectra. This represents the total number of samples, i.e., the total number of spectra.

[0228] Step 4.3: Construct a feature optimization and verification unit to continuously monitor reconstruction performance and adjust the structural parameters of the self-supervised feature extraction network during the training phase, ensuring that the model obtains stable feature representation under unlabeled data conditions.

[0229] During training, the loss function is reconstructed. As a self-supervised signal, all trainable parameters, including the auxiliary decoder, feature fusion module, environmental modulation branch, and spectral backbone network, are jointly optimized through an end-to-end backpropagation algorithm. An early stopping strategy is employed during training, determining the optimal stopping point based on the reconstruction error of the validation set to prevent overfitting. After training, the auxiliary decoder is removed, retaining only the optimized self-supervised feature extraction network. This network exhibits stronger robustness and generalization ability, serving as the pre-training foundation for the concentration prediction model.

[0230] Step 5: To achieve high-precision concentration inversion under small sample conditions, a BPBO-GRNN adaptive concentration inversion optimization model is constructed regarding the relationship between the fused feature vector and the mixed gas concentration. This model is a joint decoupled model, which deeply integrates the BPBO algorithm and the GRNN model to construct a concentration inversion network with self-learning and self-adjustment capabilities. This model is used to achieve high-precision inversion of mixed gas concentration and adaptive optimization of model parameters, and is the core unit for completing the final concentration prediction in this invention. The model mainly consists of two parts: a GRNN concentration prediction submodule and a BPBO adaptive optimization submodule. BPBO global optimization is used to improve the robustness and prediction accuracy of the GRNN. The specific process is as follows:

[0231] Step 5.1: Construct a GRNN concentration prediction submodule based on a generalized regression neural network structure. This submodule achieves a nonlinear mapping between the input fused feature vector and concentration values ​​through kernel density estimation. It utilizes a Gaussian kernel function to weighted model feature similarity and uses a smoothing factor to control the kernel bandwidth, thereby obtaining concentration prediction results that are sensitive to changes in both spectral and environmental characteristics. GRNN is a nonparametric regression model based on kernel density estimation, and its output is obtained by a weighted average of the concentration labels of all training samples. Define the... The fused feature vector of each training sample Given a fused feature vector of any input, for With the training samples The similarity is measured by Euclidean distance. For the first The fused feature vectors of the training samples are transformed into kernel weights using a Gaussian kernel function:

[0232] ;

[0233] in, for and The kernel weights between them; It is an exponential function; It is the Euclidean distance function; This is a smoothing factor that controls the rate at which the similarity between samples decays. Samples with higher similarity have a greater weight in the prediction result.

[0234] Concentration prediction values ​​from the GRNN concentration prediction submodule Given by the following formula:

[0235] ;

[0236] in, The total number of training samples; For the first The true concentration labels of the training samples. As the formula shows, the performance of GRNN mainly depends on the smoothing factor. The value of . When If the value is too small, the kernel function bandwidth tends to be narrow, which can easily lead to overfitting; when... If the value is too large, the kernel function bandwidth will be too wide, which can easily lead to underfitting problems.

[0237] Step 5.2: Constructing the BPBO Global Adaptive Optimization Submodule. This submodule proposes a generalized regression neural network adaptive optimization mechanism based on the Raptor Optimization Algorithm, realizing intelligent optimization and globally optimal configuration of hyperparameters for the concentration inversion model, which is one of the key innovations of this invention. Based on the swarm intelligent search mechanism of the Raptor Optimization Algorithm, this submodule dynamically adjusts the GRNN smoothing factor value through a strategy combining global exploration and local exploitation, avoiding the inefficiency and local optima problems caused by traditional manual parameter tuning or grid search, thus achieving adaptive optimal configuration of model performance.

[0238] against For problems that rely on experience and are difficult to optimize globally, this invention introduces the BPBO algorithm for adaptive optimization. This algorithm simulates the global exploration and local exploitation behavior of birds of prey during hunting, dynamically searching for a smoothing factor to verify that minimizing the root mean square error is the optimization objective. First, the smoothing factor is... The optimization process is modeled as a minimization problem, with a smoothing factor. The individual positions are updated using the Raptor optimization algorithm. The concentration-labeled data used in this invention was collected by a self-built controllable gas mixing system. This system can precisely control and record the true concentration of each group of samples during the gas mixing process; therefore, each collected spectrum corresponds to a defined concentration label. The concentration-labeled dataset is divided into a training set and a validation set. The fitness function is defined as the root mean square error (RMSE) of the GRNN model on the validation set. For any given... Value, its fitness The calculation is as follows:

[0239] ;

[0240] in, The number of samples in the validation set; For the first One verification sample; For the first The true concentration label of the validation sample; Indicates the use of the current The constructed GRNN concentration prediction submodule predicts the concentration values ​​of the validation samples.

[0241] The BPBO algorithm finds the optimal smoothing factor by simulating the exploration and exploitation behavior of birds of prey during hunting. Population size is set. With maximum number of iterations Set the smoothing factor The search space is a preset range ,in , They are respectively The minimum and maximum values; random initialization within the search space. The positions of individual raptors form the initial population. The fitness value of each individual is calculated, and the position of the individual with the lowest fitness value is recorded as the current global optimum. In each iteration of the Raptor optimization algorithm, a random number within the interval [0, 1] is generated. Preset a threshold The BPBO algorithm serves as a threshold for switching between the exploration and development phases; if the updated individual position exceeds the search space, it is truncated and mapped back to the search space. The iterative update process of the BPBO algorithm is an adaptive, phased search strategy, its core being the dynamic balancing of global exploration and local development. In each iteration, the algorithm first generates a random number. With threshold The comparison will determine whether to proceed with the global exploration phase or the local development phase. If... The algorithm will then enter a global exploration phase. This phase simulates a raptor searching for prey over a wide area at high altitudes, aiming to escape local optima and explore a broader search space. The individual position update formula is as follows:

[0242] ;

[0243] in, For the first The individual in the first The position of the next iteration corresponds to the [number]th iteration. The smoothing factor value at the next iteration; The updated position; It is a random number in the range [-1, 1].

[0244] like The algorithm then enters a local development phase, simulating the raptor's dive and precise pursuit after spotting prey. This phase comprises two sub-phases, determined by the number of iterations. and The relationship determines this. When At this time, the algorithm executes a dive-pursuit strategy, which guides individuals to move towards the center of the population to quickly converge to a promising solution region. The formula for the dive-pursuit strategy is as follows:

[0245] ;

[0246] in, For the first The average of all individual positions in the next iteration; It is a random number within the range [0, 1]. This strategy guides the population to gather towards the central region of the current solution. When At that time, the algorithm executes a precision attack strategy, simulating a raptor locking onto a single prey for a precise attack, and utilizing Levi's flight to increase the randomness and diversity of the search. The formula for the precision attack strategy is as follows:

[0247] ;

[0248] in, The Lévy flight function is used, and since there is only one smoothing factor among the optimization variables, the input dimension is 1. After each position update, the fitness value is recalculated. Then update the individual's location.

[0249] When the number of iterations reaches Finally, output the final globally optimal position. As the optimal smoothing factor.

[0250] Obtaining the optimal smoothing factor Subsequently, the BPBO-GRNN adaptive concentration inversion optimization model was based on Predict the concentration of any input fused feature vector to obtain the optimal concentration prediction value. :

[0251] ;

[0252] The globally optimal smoothing factor obtained by introducing BPBO The model can use the optimized Gaussian kernel function to obtain the concentration inversion results of the test sample, thereby achieving high-precision adaptive concentration prediction under small sample conditions.

[0253] Step 6: Collect the absorption spectrum signal of the unknown concentration of the gas to be measured under the actual operating conditions of the thermal power plant, and at the same time collect the actual environmental parameter data to construct a multi-source dataset; import the multi-source dataset into the environmental-spectral synergistic fusion decoupling model trained by the laboratory calibration data for concentration prediction.

[0254] To verify the concentration inversion performance of the method of this invention, typical gases such as NH3 and CO are used as examples. The gradient comparison results between their actual and predicted concentrations are as follows: Figure 3 and Figure 4 As shown, the residual distribution is as follows Figure 5As shown in the figure. The results show that the method of the present invention has high prediction accuracy under different concentration gradients. The determination coefficients R² of NH3 and CO reach 0.9994 and 0.9992, respectively, and the root mean square errors RMSE are 3.05 and 2.36, respectively, indicating that the method of the present invention can still maintain stable concentration inversion performance under the conditions of temperature and pressure coupling interference.

[0255] To verify the effectiveness of the self-supervised feature extraction network in environmental interference compensation and further illustrate its technical effect in improving the accuracy and robustness of concentration inversion, a comparative ablation experiment was conducted. In the experiment, the original spectral signal after data preprocessing in step 2 was directly input into the BPBO-GRNN adaptive concentration inversion optimization model in step 5 for prediction (i.e., steps 3 and 4 were omitted). Tests were conducted under the same temperature (100℃~250℃) and pressure (0.8~2.5 atm) conditions. The gradient comparison results of the actual and predicted concentrations of CO and NH3 without the self-supervised feature extraction network compensation model are shown below. Figure 6 and Figure 7 As shown, the residual distribution is as follows Figure 8 As shown.

[0256] The comparative results show that the model without self-supervised feature extraction network compensation exhibits significant bias in predicting the concentrations of NH3 and CO, with root mean square errors (RMSE) of 70.22 and 36.12, respectively. The predicted concentrations deviate significantly from the actual concentrations, making it difficult to accurately invert the concentrations of the gas mixture. In contrast, the model in this invention, which incorporates a self-supervised feature extraction network, under the same conditions, achieves significantly higher concentrations. Figure 5 As shown, the root mean square error (RMSE) significantly decreased to 3.05 and 2.36, respectively, and the prediction determination coefficient (R²) increased to over 0.999, indicating a substantial improvement in prediction accuracy. The results demonstrate that the concentration inversion accuracy is significantly improved in the model incorporating the self-supervised feature extraction network, with the RMSE reduced by approximately 91.6% (NH3) and 96.6% (CO), validating the network's effective compensation for spectral drift and feature aliasing caused by temperature-pressure coupling. The prediction stability and robustness are enhanced, the model's performance is more consistent under complex conditions, and the error fluctuation is controlled within ±0.5% under different experimental conditions.

[0257] In summary, the spectral aliasing decoupling and concentration inversion method proposed in this invention, based on self-supervised feature extraction and BPBO-GRNN joint optimization, achieves adaptive compensation for spectral nonlinear deformation and dynamic feature modeling in complex industrial environments, demonstrating significant innovation and practical value. This method overcomes the technical bottlenecks of traditional models, such as inaccuracy and insufficient anti-interference capability under temperature-pressure coupling conditions, through adaptive environmental parameter characterization and dynamic spectral compensation mechanisms. It achieves high-precision, robust inversion and stable detection of multi-component gases. This invention also exhibits good system compatibility and can be embedded in a continuous emission monitoring system (CEMS) to achieve high-precision concentration monitoring of aliased gases such as CO and NH3 during power plant emissions, providing original and widely applicable technical support for industrial emission control and environmental supervision under complex operating conditions.

[0258] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for spectral unmixing and concentration retrieval under the cross-influence of multiple environmental factors, characterized in that, A environment-spectrum collaborative fusion decoupling model is constructed for concentration prediction, and the environment-spectrum collaborative fusion decoupling model comprises a multi-source data processing module, a self-supervised feature extraction network, a self-supervised auxiliary decoding module and a BPBO-GRNN adaptive concentration inversion optimization model; the method specifically comprises the following steps: Step 1, respectively collecting absorption spectrum signals of specified mixed gas under different temperatures, pressures and known concentrations, and collecting environmental parameter data to construct a multi-source data set; Step 2, denoising, dimensionality reduction and preprocessing the multi-source data set through the multi-source data processing module; Step 3, constructing an environment parameter adaptive modulation self-supervised feature extraction network to realize deep fusion of spectrum and environmental information and obtain a fusion feature vector; the self-supervised feature extraction network comprises an environment modulation branch, a spectrum backbone network and a feature fusion module; The specific process is as follows: Step 3.1, constructing environment modulation branch to obtain high-level environment feature vector And introduce multi-head self-attention mechanism and gating control structure; The environment modulation branch further comprises a multi-modal environment parameter self-attention representation submodule, an environment feature deep reconstruction submodule and an adaptive regulation unit, and the specific working process is as follows: Step 3.1.

1. Constructing a multi-modal environmental parameter self-attention representation submodule based on a multi-head self-attention mechanism, for obtaining an environmental feature representation : ； wherein, is a Softmax function; , , are query vector, key vector and value vector corresponding to each attention head, respectively; is a transpose; is a dimension of the key vector; Step 3.1.2, constructing the environmental feature deep reconstruction sub-module using a multi-layer residual network structure; and inputting the initial feature vector into the environmental feature deep reconstruction sub-module , and recording the initial feature vector as ; the sub-module is composed of a plurality of residual blocks stacked; for the i-th residual block, the calculation is as follows: ​ ； wherein, is the feature vector output by the th residual block, which is input to the th residual block; is the feature vector output by the th residual block; is a ReLU activation function; is the th residual block; is the total number of residual blocks; through the step-by-step feature extraction and superposition of the layers of residual blocks, a high-level environmental feature vector is obtained. ； wherein, is the feature vector output for the i-th residual block; Step 3.1.3: Construct an adaptive control unit composed of multiple fully connected layers. For the first layer in the spectral backbone network... An environment adaptive residual block that needs to be modulated; the adaptive control unit uses high-level environmental feature vectors. Generate the corresponding environmental modulation parameters; specifically as follows: First, the high-level environment feature vector is input to three independent fully connected layers respectively to generate a preliminary environment modulation parameter and a gating signal, the preliminary environment modulation parameter including a preliminary scaling factor and a preliminary offset factor; Then, based on the gating signal, the preliminary environment modulation parameters are weighted and fused to obtain the final environment modulation parameters: ； ； wherein, is a gating signal for the th environment-adaptive residual block; , are a preliminary scaling factor and a preliminary offset factor for the th environment-adaptive residual block, respectively; , are a final scaling factor and a final offset factor for the th environment-adaptive residual block, respectively; denotes element-wise multiplication; is a unit vector; Finally, the obtained environment modulation parameters are applied to the feature mapping of the convolution block to realize adaptive affine transformation; Step 3.2, constructing a spectral backbone network composed of a plurality of environment-adaptive residual blocks stacked in series, performing environment-adaptive feature extraction on the spectral data level by level to obtain an environment-adaptive spectral feature vector ; Step 3.3, constructing a feature fusion module to perform feature fusion, specifically, performing splicing on the environment adaptive spectral feature vector and the high-layer environment feature vector to obtain a fused feature vector : ； wherein, is a vector concatenation operation in the feature dimension; Step 4, constructing a self-supervised auxiliary decoding module, which uses a self-supervised learning mechanism to improve the feature expression ability and generalization performance of the self-supervised feature extraction network and fully excavates the unlabeled data; Step 5, constructing a BPBO-GRNN adaptive concentration inversion optimization model for realizing inversion of mixed gas concentration and adaptive optimization of model parameters; the BPBO-GRNN adaptive concentration inversion optimization model comprises a GRNN concentration prediction submodule and a BPBO adaptive optimization submodule, and the specific process is as follows: Step 5.1, based on the generalized regression neural network structure, the GRNN concentration prediction submodule is constructed, the feature similarity is weighted and modeled by using the Gaussian kernel function, and the kernel bandwidth is controlled by the smoothing factor, so as to obtain the concentration prediction result sensitive to the changes of spectrum and environmental characteristics; the fusion feature vector of the first training sample is defined as For any input fusion feature vector, the similarity with the first training sample is measured by the Euclidean distance, The fusion feature vector of the first training sample is defined as The fusion feature vector of the first training sample is defined as​​​​ ； wherein, is with between the kernel weights; is an exponential function; is an Euclidean distance function; is a smoothing factor; Concentration prediction value Is: ； wherein, is the total number of training samples; is the true concentration label of the th training sample; Step 5.2, constructing a BPBO global adaptive optimization submodule, which proposes a generalized regression neural network adaptive optimization mechanism based on the hawk optimization algorithm, dynamically searches for the optimal smoothing factor by simulating the global exploration and local development behavior of hawks in the process of predation, and finds the optimal smoothing factor; Step 6, collecting absorption spectrum signals of the gas to be measured under unknown concentration in actual working conditions of the thermal power plant, collecting actual environmental parameter data to construct a multi-source data set, and inputting the multi-source data set into the environment-spectrum collaborative fusion decoupling model trained by the laboratory calibration data to predict the concentration.

2. The method of claim 1, wherein, The specific process of step 1 is as follows: first, under different temperature and pressure conditions, absorption spectrum signals of CO and NH3 mixed gas with known concentration gradient are collected, and corresponding environmental parameter data is recorded to construct a multi-source data set; the multi-source data set is used as the training set of the model; Subsequently, under the same temperature and pressure control range, the mixed spectrum signal under unknown concentration is collected as the model verification set.

3. The method of claim 1, wherein, In step 2, the data processing module comprises an intelligent denoising module, a preprocessing module and a dimensionality reduction module; the specific process is as follows: Step 2.1, construct an intelligent denoising module, the whole intelligent denoising module adopts a pre-trained one-dimensional U-Net convolutional neural network structure; input the multi-source data set into the one-dimensional U-Net convolutional neural network for denoising, the mapping relationship is: ； wherein, is a high-fidelity spectrum of the th sample; is a raw spectrum of the th sample; are trainable parameters of a one-dimensional U-Net convolutional neural network; denotes a function mapping corresponding to the one-dimensional U-Net convolutional neural network structure. Step 2.2, introduce a non-parametric distribution alignment mechanism of quantile mapping to construct a preprocessing module; the preprocessing module includes a distribution feature analysis unit and a non-parametric distribution alignment unit, and the specific working process is: Step 2.2.1, based on the distribution feature analysis unit to estimate the empirical distribution, calculate the empirical cumulative distribution function of each spectral feature and environmental parameter: ； in, The empirical cumulative distribution function; Represents the total number of samples; The threshold value is the feature value. Indicates the first Feature values ​​of each sample; For indicator functions, when Less than or equal to The value is 1 if the condition is met, otherwise it is 0. Step 2.2.

2. Quantile mapping of the high-fidelity spectra based on the non-parametric distribution alignment unit maps the quantile values of the high-fidelity spectra through the inverse cumulative distribution function of the target distribution to obtain the normalized results :​ ； wherein is the inverse cumulative distribution function of the standard normal distribution; Step 2.3, construct a dimension reduction module.

4. The method of claim 3, wherein, In the step 2.3, the dimension reduction module includes an encoder submodule, a decoder submodule and a feature reconstruction verification unit, and the specific working process is: Step 2.3.1, the encoder sub-module maps the standardized results to a low-dimensional latent space via an autoencoder mapping function, resulting in a low-dimensional feature vector : ； wherein, represents an autoencoder mapping function; is a set of trainable parameters; is the compressed spectral dimension; Step 2.3.2, the decoder submodule reconstructs the low-dimensional feature vector back to the original space through the inverse mapping function of the decoder: ； wherein, denotes an inverse mapping function of the decoder; is a set of trainable parameters of the decoder; is a reconstruction result; Step 2.3.3, the target of the feature reconstruction verification unit is to minimize the mean square error between the input and the reconstructed spectrum, and the network can learn the nonlinear feature structure of the spectral data by minimizing the loss function; the loss function is: ； wherein, and denote the standardized result and the reconstructed result of the th sample, respectively; denotes the L2 norm.

5. The method of claim 4, wherein, In the step 3.2, each environment adaptive residual block adopts a residual structure composed of a main path and a shortcut path; the calculation formula of the main path is: ； ； wherein, is the output feature map of the first stage; is the first batch normalization operation; denotes a one-dimensional convolution operation; is the output of the previous environment-adaptive residual block; is the final output of the main path; is the second batch normalization operation; denotes a one-dimensional convolution operation; The calculation formula of the shortcut path is: ； wherein, is an output of the shortcut path; represents a one-dimensional convolution operation with a 1x1 convolution kernel; is a dimension of the feature map; The output of the main path and the output of the shortcut path are added element by element to obtain a spectral feature map : ； is the output of the single environment-adaptive residual block for the th environment-adaptive residual block: ； wherein, is the spectral feature map output by the environment-adaptive residual block; is the environment-adaptive residual block; spectral feature map output by the After successive processing via the environment-adaptive residual block, the final spectral feature map is obtained ; The global average pooling operation is adopted for conversion to obtain a final environment adaptive spectral feature vector : ； wherein, is a global average pooling operation.

6. The method of claim 5, wherein, In the step 4, the self-supervised auxiliary decoding module includes an auxiliary decoder, a self-supervised constraint unit and a feature optimization verification unit, and the specific working process is: Step 4.1, fusing feature vectors The auxiliary decoder extracts deep feature representation with physical consistency, and reconstructs the spectral data through a multi-layer fully connected structure, remaps the highly compressed features to the original spectral space, and finally generates reconstructed spectra; Step 4.

2. Defining the reconstruction loss function in the self-supervised constraint unit Measuring the difference between the original and reconstructed spectra: ； wherein, is the original spectrum of the th sample; is the reconstructed spectrum of the th sample; the number of original spectra and reconstructed spectra corresponds. Step 4.3, construct a feature optimization verification unit to continuously monitor the reconstruction performance and adjust the structure parameters of the self-supervised feature extraction network during the training stage, so as to ensure that the model obtains stable feature expression under the condition of no labeled data; During the training process, the reconstruction loss function As a self-supervised signal, all trainable parameters including auxiliary decoder, feature fusion module, environment modulation branch and spectral backbone network are jointly optimized by end-to-end backpropagation algorithm; Early stopping strategy is adopted in the training stage to determine the optimal stopping point according to the reconstruction error of the validation set to prevent overfitting; After training, the auxiliary decoder is removed and only the optimized self-supervised feature extraction network is retained.

7. The method of claim 6, wherein, The specific process of step 5.2 is: First, the smoothing factor is modeled as a minimization problem, the smoothing factor is updated as the individual position of the raptor optimization algorithm, and the optimal smoothing factor is found by the raptor optimization algorithm; for any given value, the fitness is calculated as follows: ； in, The number of samples in the validation set; For the first One verification sample; For the first The true concentration label of the validation sample; Indicates the use of the current The constructed GRNN concentration prediction submodule predicts the concentration values ​​of the validation samples; Setting population size With the maximum number of iterations ; Set smoothing factor The search space is a preset range , wherein , The minimum value and the maximum value of , respectively; within the search space, the positions of raptors are randomly initialized to form an initial population; the fitness value of each individual is calculated, and the position of the individual with the minimum fitness value is recorded as the current global optimal position ; In each iteration of the algorithm, a random number in the interval [0, 1] is generated , a threshold value is preset as the switching threshold between the exploration and development stages; if the updated individual position is outside the search space, it is truncated and mapped back within the search space range; the specific iteration process of the algorithm is as follows: If , the algorithm will enter the global exploration stage; individual position update formula is as follows: ； wherein, is the position of the i-th individual at the j-th iteration, is the position of the i-th individual at the j-th iteration, is the position of the i-th individual at the j-th iteration, is the position of the i-th individual at the j-th iteration, is the updated position of the i-th individual, is a random number in the range [-1, 1]. If , the algorithm enters a local development phase; this phase contains two sub-phases, determined by the relationship between the iteration number and ; when , a dive-and-chase strategy is executed: ； wherein, is the first iteration the average of all individual positions; is a random number in [0, 1] When the exact attack strategy is executed: ； wherein is the Levy flight function; After each position update, the fitness value is recalculated, and if the individual position is updated; When the number of iterations reaches the final global optimal position as the optimal smoothing factor; After obtaining the optimal smoothing factor Then, the BPBO-GRNN adaptive concentration inversion optimization model is established based on The concentration of the fusion feature vector is predicted for any input, and the optimal concentration prediction value is obtained : 。

Images