A method and device for killing microorganism aerosol by overflow

Abstract

The application discloses a kind of overcurrent microbial aerosol disinfecting method and device, it is related to air sterilization field, rely on air duct shell, sensor, pulse strong light sterilization component, excimer ultraviolet sterilization component, light board and control chip, after collecting inlet fluorescence spectrum and flow rate, obtain inlet bacteria component table by spectral analysis quantitative method, according to the pulse sensitive coefficient of bacteria, excimer sensitive coefficient and inlet content clustering obtain pulse cluster and excimer cluster, determine killing light intensity in combination with weibull model and light intensity setting algorithm, construct including reflection enhancement light intensity distribution, solve the optimal xenon lamp, ultraviolet power group by swarm intelligence algorithm, update the scale parameter of bacteria after outlet detection, realize accurate targeted disinfection, improve sterilization efficiency and adaptability, significantly reduce energy consumption.

Description

Technical Field

[0001] This invention relates to the field of air sterilization, specifically to a flow-through microbial aerosol sterilization method and apparatus. Background Technology

[0002] As a significant carrier of airborne pathogens, bacterial aerosols are widely present in various settings, including medical environments, public transportation, and workplaces. The bacteria, fungi, and other microorganisms they carry can easily cause cross-infection, posing a serious threat to human health and public health security. With increasing demands for air quality and upgraded epidemic prevention requirements in various enclosed spaces, developing efficient, precise, and low-energy-consumption air sterilization technologies to achieve rapid and effective elimination of bacterial aerosols has become a critical requirement for ensuring public health security, possessing significant practical importance and application value.

[0003] While current air sterilization technology has made some progress, it still has significant shortcomings in practical applications. On the one hand, existing sterilization technologies mostly employ a single disinfection method, lacking the ability to accurately identify and classify airborne bacterial aerosols. This makes it difficult to accurately obtain the specific types, contents, and sensitivity characteristics of bacteria, leading to a blind disinfection process. This results in the inability to target bacteria with different sensitivity characteristics, affecting disinfection efficiency and thoroughness, and may also lead to resource waste due to over-disinfection. On the other hand, existing technologies lack a scientific and systematic modeling method for disinfection light intensity design. They do not fully consider the impact of light propagation loss and reflection enhancement on light intensity distribution, and disinfection parameters are mostly fixed settings. They cannot dynamically adjust power configuration and related parameters according to the actual disinfection effect, resulting in uneven light intensity distribution, excessive energy consumption, and difficulty in adapting to the dynamic changes of bacterial aerosols in different scenarios. This leads to poor adaptability and economy in long-term operation. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by proposing a flow-through microbial aerosol disinfection method and device. By accurately identifying and clustering inlet bacteria components, combining light intensity model construction and swarm intelligence algorithm optimization of power groups, and dynamically updating bacterial parameters through outlet detection, it achieves precise targeted disinfection of bacteria with different characteristics. While ensuring the disinfection effect meets the standards for the entire area, it significantly reduces energy consumption and greatly improves the targeting, effectiveness, economic efficiency, and long-term adaptability of the device.

[0005] The technical solution to achieve the purpose of this invention is as follows: A method for disinfection of microbial aerosols via flow-through includes the following steps: In the current cycle, inlet fluorescence spectra and flow rates are collected. The inlet fluorescence spectrum was smoothed and its features extracted using spectral analysis and quantitative methods. The endmember matrix was obtained through kernel nonnegative matrix decomposition. abundance matrix The ingress bacterial composition table is obtained by combining a multi-task model mapping, based on flow rate. Calculate and determine the delayed acquisition time; Using a clustering algorithm, iterative clustering is performed based on the sensitivity coefficient of each bacterium and the corresponding ingress content in the ingress bacterial composition table to determine pulse clusters and excimer clusters. The sensitivity coefficients include pulse sensitivity coefficients and excimer sensitivity coefficients. Initialize the annihilation light intensity using a light intensity setting algorithm. Based on the shape and scale parameters of each bacterium in the pulsed or excimer cluster, the theoretical kill rate of the pulsed or excimer cluster is calculated using the Weibull model. The kill light intensity is then iteratively adjusted based on the deviation between the theoretical kill rate and the ideal kill rate. The intensity of the pulse annihilation light was obtained. and excimer laser annihilation intensity ; Based on the lamp layout and reflector reflection, the pulse intensity model of each pulse xenon lamp and the excimer light intensity model of each ultraviolet lamp are superimposed to obtain the pulse segment light intensity distribution. and excimer segment light intensity distribution To achieve pulse annihilation light intensity respectively and excimer laser annihilation intensity With the goal of minimizing total energy consumption, the optimal xenon lamp power group is determined iteratively using a swarm intelligence algorithm. and optimal ultraviolet power group And it performs regulation, among which the pulsed light intensity model and the excimer light intensity model are constructed based on the principles of light emission and light propagation loss; At the delayed acquisition time, the exit fluorescence spectrum was acquired and the exit bacterial composition table was obtained by using the spectral analysis quantitative method. The actual kill rate of each bacterial was calculated by comparing the exit bacterial composition table with the inlet bacterial composition table. The scale parameters of each bacterial were updated by combining the Weibull model and waiting for the next cycle.

[0006] Furthermore, the quantitative method based on spectral analysis includes the following steps: A Gaussian weighted moving average is performed on each wavelength point of the fluorescence spectrum. The wavelength distance between each non-center wavelength point and the center wavelength point is obtained from a window with a fixed width centered on each wavelength point and converted into a corresponding Gaussian weight through a Gaussian function. The fluorescence intensity of each non-center wavelength point is weighted and summed according to the corresponding normalized Gaussian weight to obtain the smoothed fluorescence intensity of the center wavelength point. The smoothed fluorescence spectrum is obtained by iterating through the spectrum. The fluorescence spectrum includes the inlet fluorescence spectrum and the outlet fluorescence spectrum. The characteristic peaks in the smoothed fluorescence spectrum are extracted and adjacent characteristic peaks with a peak position difference greater than or equal to the peak position difference threshold are integrated into a characteristic peak group. The ratio of the fluorescence intensity peak of the latter characteristic peak to the former characteristic peak in each characteristic peak group is divided by the standard ratio corresponding to the characteristic peak group in the pre-constructed fingerprint library and the mean value is taken to obtain the correction coefficient. The correction coefficient is multiplied by the smoothed fluorescence spectrum to obtain the unbiased fluorescence spectrum. The global skewness and global kurtosis of the unbiased fluorescence spectrum are calculated, and the peak position, peak height, full width at half maximum (FWHM), and peak area of ​​all characteristic peaks in the unbiased fluorescence spectrum are extracted. The unbiased fluorescence spectrum is decomposed using continuous wavelet transform, and the mean and variance of the wavelet coefficients at five scales are extracted. All coefficients are then concatenated to obtain the multimodal vector. ; Kernel nonnegative matrix factorization is employed, and the multimodal vectors are decomposed using radial basis kernels. Mapping to a high-dimensional kernel space and decomposing it to obtain the optimal endmember matrix and the optimal abundance matrix ; Based on a multi-task model, a convolutional neural network is used to extract the normalized optimal endmember matrix. The deep features of each endmember are extracted and flattened, and then mapped using a fully connected layer to obtain the category label corresponding to each endmember. The normalized optimal abundance matrix is ​​then processed using a fully connected layer with an attention mechanism. Each abundance in the table is mapped to the corresponding entry content, and the entry bacterial composition table is obtained by integrating all species tags and corresponding entry contents.

[0007] Furthermore, kernel nonnegative matrix factorization includes the following steps: Multimodal vectors By mapping radial basis kernels to a high-dimensional kernel space, a hybrid feature matrix is ​​obtained. ; Mixing feature matrices within the kernel space Approximate endmember matrix With abundance matrix The product of these terms yields the factorization; Based on the representation theorem of radial basis kernels, the endmember matrix is... Represented as a nonnegative coefficient matrix Transpose and hybrid feature matrix The product of the products is then substituted back into the decomposition to obtain the first decomposition, which is then multiplied by the mixture feature matrix on both sides. The transpose of the first expression yields the second decomposition. Taking the transpose of both sides simultaneously, the hybrid characteristic matrix is ​​defined. With the mixed feature matrix The product of transposes is the kernel matrix. Simplifying, we obtain the equivalent approximation; Using the Frobenius norm of the difference between the left and right sides of the equivalent approximation as the core objective function, the abundance matrix is... The L1 norm is determined by a preset regularization coefficient. Introducing the core objective function, we obtain the objective function. Minimizing this objective function is the optimization objective, using a non-negative coefficient matrix. abundance matrix The optimization problem is constructed with nonnegativity as a constraint, and the optimal nonnegative coefficient matrix is ​​obtained by iteratively solving the problem using the alternating least squares method. and the optimal abundance matrix And further calculate the optimal endmember matrix .

[0008] Furthermore, the sensitivity coefficients of each bacterium in the inlet bacterial composition table are obtained through the bacterial feature library. The bias coefficient of each bacterium is obtained by subtracting its excimer sensitivity coefficient from its pulse sensitivity coefficient. This bias coefficient is then concatenated with the corresponding inlet content and normalized to obtain the standard clustering features of each bacterium. The KMeans++ algorithm is used to cluster bacteria based on these standard clustering features, which include the standard bias coefficient and the standard inlet content. The bacteria with the largest and smallest standard bias coefficients are designated as pulse cluster centers and excimer cluster centers, respectively. Each bacterium is assigned to the cluster corresponding to the cluster center with the closest feature distance. The feature distance is equal to the difference in standard bias coefficients multiplied by a first weight plus the difference in standard inlet content multiplied by a second weight. A bacterium is selected through a process of iteration. The difference between the minimum feature distance between the selected bacterium and the bacteria in another cluster and the average feature distance between the selected bacterium and the other bacteria in the same cluster is divided by the larger of the minimum feature distance between the selected bacterium and the bacteria in another cluster and the average feature distance between the selected bacterium and the other bacteria in the same cluster. This yields the silhouette coefficient of the selected bacterium. The average silhouette coefficient of all bacteria is then taken as the global silhouette coefficient. The average standard clustering features of all bacteria in the two clusters are calculated separately and used as the new pulse cluster center and excimer cluster center, respectively. The next round of iterative clustering is then executed again until the number of iterations equals the maximum number of iterations or the change in the global silhouette coefficient between two adjacent iterations is less than or equal to the change threshold. The iteration is then stopped, and the pulse cluster and excimer cluster are obtained.

[0009] Furthermore, the light intensity setting algorithm includes the following steps: Obtain the sensitivity coefficient, ingress content, and shape and scale parameters involved in the Weibull model for each bacterium in the cluster. Divide the ingress content of each bacterium in the cluster by the sum of the ingress contents of all bacteria in the cluster to obtain the ingress content weight of each bacterium. Initialize the killing light intensity. In this context, "cluster" refers to either a pulse cluster or an excimer cluster. In the current iteration, the scale parameter of each bacterium in the cluster is divided by the sensitivity coefficient to obtain the specific killing light intensity of each bacterium. The specific killing light intensity is either the specific pulse killing light intensity or the specific excimer killing light intensity. The theoretical killing rate of each bacterium is calculated by combining the Weibull model and then weighted by the inlet content weight to obtain the cluster killing rate corresponding to the cluster. The inverse of the ratio of the cluster annihilation rate to the annihilation rate threshold is added to obtain the annihilation rate deviation rate. If the absolute value of the annihilation rate deviation rate is less than the deviation rate threshold, the annihilation light intensity of the current iteration is adjusted. Cluster annihilation light intensity refers to the intensity of pulsed annihilation light. Or excimer laser annihilation intensity If the absolute value of the kill rate deviation rate is greater than or equal to the deviation rate threshold, the kill rate deviation rate is multiplied by the adjustment step size to obtain the undetermined adjustment amount. If the undetermined adjustment amount is within the feasible range, it is directly used as the actual adjustment amount. If the undetermined adjustment amount is less than the lower limit of the feasible range or greater than the upper limit of the feasible range, the lower limit or upper limit of the feasible range is used as the actual adjustment amount, and the actual adjustment amount is superimposed on the kill light intensity of the current iteration. And then start the next iteration.

[0010] Furthermore, the radiant flux of the pulsed xenon lamp Equal to pulse luminescence efficiency Xenon lamp power Pulse width The product of the radiant flux and the solid angle of the hemisphere, the emission of a pulsed xenon lamp follows the Lamborghian radiation characteristics, and the initial radiant intensity of the Lamborghian source is equal to the radiant flux multiplied by the solid angle of the hemisphere. The ratio of the initial radiant intensity of a pulsed xenon lamp to the light intensity, when diffused in the form of light, undergoes inverse square attenuation of spherical waves and air absorption attenuation. The inverse square attenuation of spherical waves indicates that the energy flux density is inversely proportional to the square of the propagation distance, while air absorption attenuation follows an exponential decay law. A pulsed light intensity model of the pulsed xenon lamp is constructed, and the radiant flux of the ultraviolet lamp tube is... Equal to ultraviolet radiation efficiency and ultraviolet power The product of these two factors also follows the principles of Lambertian radiation characteristics and light propagation loss. An excimer light intensity model for ultraviolet lamps is constructed. The pulse light intensity model or the excimer light intensity model is used to calculate the light intensity at any coordinate in space when the pulsed xenon lamp or ultraviolet lamp is running at a specific power at a specific coordinate.

[0011] Furthermore, the radiant flux of the pulsed xenon lamp Equal to pulse luminescence efficiency Xenon lamp power Pulse width The product of these terms allows a pulsed xenon lamp to approximate a linear light source. The longitudinal direction of the air duct is defined as... In the axial direction, for a line light source, it can be divided into countless infinitesimal segments using the method of infinitesimal elements. Each micro-segment The radiant flux is equal to the radiant flux of the pulsed xenon lamp. With micro segment The product divided by the length of the pulsed xenon lamp Each micro-segment A Lamborgh source can be approximated as a point-like source. The initial radiative intensity of the Lamborgh source is equal to the radiative flux and the solid angle of the hemisphere. The ratio, due to each infinitesimal segment Acting on any coordinate within the pulse sterilization segment Only in the direction perpendicular to the pulsed xenon lamp, each micro-segment The initial radiation intensity multiplied by the cosine of the deviation angle and measured over the xenon lamp length Integrating, we obtain the initial radiant intensity of the pulsed xenon lamp, which is equal to the radiant flux of the pulsed xenon lamp. Divide by twice the length of the pulsed xenon lamp Introducing light propagation loss, light propagates from the pulsed xenon lamp to any coordinate. At that time, the energy flux density of the line light source undergoes both cylindrical wave attenuation and air absorption attenuation. For cylindrical wave attenuation, the energy flux density of the line light source varies with the vertical Euclidean distance. The decay is inversely proportional to the light intensity. For light absorbed by air, the decay follows an exponential law as light propagates through air. This leads to the construction of a pulsed xenon lamp pulse intensity model. Similarly, ultraviolet lamps can be considered as line light sources. The excimer light intensity model for ultraviolet lamps is based on the characteristics of ultraviolet energy conversion and the principle of light propagation loss. The radiant flux of the ultraviolet lamp... Equal to ultraviolet radiation efficiency and ultraviolet power The product of these factors, following the principles of Lambertian radiation characteristics and light propagation loss, is used to construct an excimer light intensity model for the ultraviolet lamp. The pulsed light intensity model or excimer light intensity model is used to calculate arbitrary coordinates in space when the pulsed xenon lamp or ultraviolet lamp is running at a specific power. The light intensity at that location.

[0012] Furthermore, by pre-measuring the light intensity at typical points inside the air duct with and without reflectors, and fitting the results to obtain the pulse enhancement coefficient, the following methods were employed. and UV enhancement coefficient Determine the coordinate ranges of the pulse sterilization segment and the excimer laser sterilization segment, sum the pulse intensity models of all pulsed xenon lamps, and multiply by the pulse enhancement coefficient. The intensity distribution of the pulse segment was obtained. Sum the excimer light intensity models of all UV lamps and multiply by the UV enhancement factor. The excimer segment light intensity distribution was obtained. .

[0013] Furthermore, define the fitness function. , to extinguish the intensity of the pulsed light Subtract pulse segment light intensity distribution The distribution of pulse light intensity difference was obtained. Design coordinates Impulse penalty function at the location Distribution of pulse light intensity differences median coordinate The larger of the pulse intensity difference at a given point and 0 will determine the excimer laser annihilation intensity. Subtract excimer segment light intensity distribution The excimer light intensity difference distribution was obtained. Design coordinates excimer penalty function at the location Distribution of excimer light intensity difference median coordinate Design an energy penalty function based on the larger of the excimer light intensity difference at the location and 0. It equals the sum of the xenon lamp power of all pulse xenon lamps and the ultraviolet power of all ultraviolet lamps. and The pulse penalty function is used to distinguish between the xenon lamp power group and the ultraviolet power group. In the pulse sterilization section All coordinates within The integral of the Sigmoid function value superimposed with the excimer penalty function In the excimer sterilization zone All coordinates within The integral of the Sigmoid function value minus the energy penalty function. The Sigmoid function value is used to obtain the fitness function. To maximize the fitness function To achieve the objective, the search space is divided based on the feasible ranges of xenon lamp power and ultraviolet power. A swarm intelligence algorithm is then used to iteratively search within the search space to obtain the fitness function. Optimal Xenon Lamp Power Group at Maximum and optimal ultraviolet power group .

[0014] A flow-through microbial aerosol disinfection device includes an air duct shell, a sensor, a pulsed light sterilization component, an excimer ultraviolet sterilization component, a reflector, and a control chip; The duct housing provides mounting support for the sensor, pulsed light sterilization component, excimer ultraviolet sterilization component, reflector and control chip; The sensors include an inlet fluorescence sensor, a flow rate sensor, and an outlet fluorescence sensor. The inlet fluorescence sensor and the flow rate sensor receive activation signals and acquire inlet fluorescence spectra and flow rates. The output fluorescence sensor receives the delayed activation signal, collects the output fluorescence spectrum, and feeds it back to the control chip. The pulsed light sterilization component includes 12 xenon lamps arranged in a 4×3 array. The xenon lamps are parallel to the airflow direction and are fixed by acrylic brackets arranged longitudinally along the air duct. The excimer ultraviolet sterilization component includes nine ultraviolet lamps arranged in a 3×3 array, with the ultraviolet lamps parallel to the airflow direction; Reflectors cover the pulse sterilization and excimer sterilization sections inside the duct casing to enhance irradiation intensity; The control chip records system time and sends an activation signal to the sensor at the beginning of each cycle, while also receiving inlet fluorescence spectra and flow rates. The composition of the inlet bacteria was obtained using spectral analysis and quantitative analysis based on flow rate. The delayed acquisition time was determined, the sensitivity coefficient of each bacterium in the ingress bacterial composition table was obtained and compared with the corresponding ingress content, the pulse cluster and excimer cluster were determined using a clustering algorithm, and the pulse sterilization light intensity was obtained using a light intensity setting algorithm. and excimer laser annihilation intensity Based on the lamp layout and reflector reflectivity, pulse intensity distributions were constructed using pulsed xenon lamp pulse intensity models and excimer light intensity models for ultraviolet lamps, respectively. and excimer segment light intensity distribution To meet the pulse annihilation light intensity and excimer laser annihilation intensity With the goal of minimizing total energy consumption, the optimal xenon lamp power group is determined iteratively using a swarm intelligence algorithm. and optimal ultraviolet power group The pulsed light sterilization component and the excimer ultraviolet sterilization component are adjusted, and the delayed acquisition time is waited for. A delayed activation signal is sent to the sensor, the exit fluorescence spectrum is received, and the exit bacterial composition table is obtained by spectral analysis quantification method. The actual killing rate of each bacterial is calculated by comparing the exit bacterial composition table and the inlet bacterial composition table, and the scale parameters of each bacterial are updated.

[0015] Compared with the prior art, the present invention has the following advantages: Data is collected using inlet fluorescence and flow rate sensors. Fluorescence spectra are smoothed, corrected, and feature extracted using spectral analysis and quantitative methods, and multimodal vectors are constructed. Kernel nonnegative matrix factorization and a multi-task model are used to accurately determine the types and corresponding concentrations of inlet bacteria. By combining the pulse sensitivity coefficient and excimer laser sensitivity coefficient of each bacterium in the bacterial feature library with its inlet concentration, clustering features are constructed. Iterative clustering algorithms are used to identify bacterial clusters suitable for pulsed light sterilization and excimer ultraviolet sterilization, achieving precise targeted classification and sterilization of bacteria with different characteristics. This significantly improves the specificity and effectiveness of sterilization, avoiding the indiscriminate nature of traditional sterilization methods. Based on the principles of light emission and light propagation loss, light intensity models for pulsed xenon lamps and ultraviolet lamps are constructed respectively. Combined with the reflective enhancement effect of reflectors, a uniform and stable light intensity distribution is formed. With the goal of meeting the required light intensity for sterilization while minimizing total energy consumption, an iterative search using a swarm intelligence algorithm determines the optimal xenon lamp power group and ultraviolet power group. Simultaneously, data is collected using an outlet sensor, and an outlet bacterial composition table is obtained through spectral analysis and quantitative methods. After comparing and calculating the actual sterilization rate, the bacterial scale parameters are updated, achieving dynamic adaptation and adjustment of sterilization parameters. This design ensures that the sterilization effect across the entire area meets the standards while significantly reducing energy consumption, improving the operational economy and long-term adaptability of the device. Attached Figure Description

[0016] Figure 1 Flowchart of the flow-through microbial aerosol disinfection method; Figure 2 Flowchart of kernel nonnegative matrix decomposition; Figure 3 This is a diagram of the internal structure of a flow-through microbial aerosol disinfection device. Figure 4 This is a diagram of the external structure of a flow-through microbial aerosol disinfection device.

[0017] Explanation of reference numerals in the attached drawings: 100, air duct housing; 101, sensor; 102, pulsed light sterilization component; 103, excimer ultraviolet sterilization component; 104, reflector; 105, control chip. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0019] Example 1 like Figure 1 As shown, this invention discloses a method for disinfection of flow-through microbial aerosols, comprising the following steps: In the current cycle, inlet fluorescence spectrum and flow rate are acquired using inlet fluorescence sensor and flow rate sensor. The inlet fluorescence spectrum was smoothed and corrected using spectral analysis quantification, and its features were extracted. The endmember matrix was obtained through kernel nonnegative matrix decomposition. abundance matrix The inlet bacterial composition table is obtained based on the multi-task model mapping, and the total length of the air duct is used as the basis for the calculation. and flow rate Calculate the delay duration and determine the delayed acquisition time for the current cycle; The sensitivity coefficient of each bacterium in the inlet bacterial composition table is obtained and combined with the corresponding inlet content to form the clustering feature of each bacterium. Using a clustering algorithm, the bacteria are iteratively clustered according to the silhouette coefficient to obtain pulse clusters and excimer clusters. The sensitivity coefficient includes pulse sensitivity coefficient and excimer sensitivity coefficient. The sensitivity coefficient reflects the effectiveness of the bacteria for pulsed light sterilization and excimer ultraviolet sterilization. The sensitivity coefficient is determined by inactivation experiment. Using a light intensity setting algorithm, the shape and scale parameters of each bacterium in the pulsed or excimer cluster are obtained, and the killing light intensity is adjusted based on the scale parameters and sensitivity coefficient. The specific killing light intensity for each type of bacteria is obtained, and the theoretical killing rate of pulsed clusters or excimer clusters is calculated using the Weibull model. The killing light intensity is then iteratively adjusted based on the deviation between the theoretical and ideal killing rates. The pulse annihilation light intensity was obtained respectively. and excimer laser annihilation intensity In this model, both shape and scale parameters are built-in parameters. The shape parameter is determined based on the type of bacteria, and different bacteria have different shape parameters. The shape parameter is determined by fitting the measurement through inactivation experiments. The Weibull model is used to evaluate the lifespan distribution of bacteria. Based on the lamp layout of the pulsed high-intensity light sterilization component 102 and the excimer ultraviolet sterilization component 103, and considering the reflectivity of the reflector 104, the pulse light intensity distribution of each pulsed xenon lamp is obtained by superimposing the pulse light intensity model. The excimer light intensity distribution of the excimer segment is obtained by superimposing the excimer light intensity model of each ultraviolet lamp. With pulse segment light intensity distribution and excimer segment light intensity distribution Each satisfies the pulse annihilation light intensity and excimer laser annihilation intensity With the goal of minimizing total energy consumption, the optimal xenon lamp power group is determined iteratively using a swarm intelligence algorithm. and optimal ultraviolet power group And it performs regulation, among which the pulsed light intensity model is constructed based on the principle of pulsed emission and the principle of light propagation loss, and the excimer light intensity model is constructed based on the ultraviolet energy conversion characteristics and the principle of light propagation loss; Waiting until the delayed acquisition time, the exit fluorescence spectrum is acquired through the exit fluorescence sensor. The exit bacterial composition table is obtained by using the spectral analysis quantitative method. The actual kill rate of each bacterial is calculated by comparing the exit bacterial composition table with the inlet bacterial composition table. The inlet content, shape parameters and sensitivity coefficient of each bacterial are combined to substitute back into the Weibull model. The scale parameters of each bacterial are updated in the current cycle and wait until the next cycle.

[0020] Furthermore, in the inactivation experiment, pure cultures of each bacterium were prepared at concentrations ranging from 1000 to 100000 CFU / m³. 3Standardized aerosol samples were used to ensure the purity of a single bacterial species. The experimental environment, with an airflow velocity controlled between 0.5 and 1.5 m / s, was kept consistent with the actual operating conditions of the device's duct. Five to eight different pulse light intensities and an equal number of excimer light intensities were designed. The actual effect value of each light intensity was calibrated using a photometer to correct errors caused by light propagation loss. The standardized aerosols were uniformly introduced, and the corresponding sterilization components were activated for disinfection. After disinfection, outlet aerosol samples were collected using a sterile sampler. The number of surviving bacteria was counted using the plate count method. Each light intensity was repeated three times, with a no-light blank group included to eliminate interference from natural aerosol settling. Basic data such as the set light intensity, actual light intensity, initial bacterial concentration, surviving bacterial concentration, and surviving bacterial concentration in the blank group were recorded. The corrected kill rate was calculated, and the average of the three experiments for each light intensity was taken. The data sets corresponding to the actual light intensity values ​​and the average kill rate are formed. The data sets are substituted into the Weibull model, and the least squares method is used for iterative optimization to obtain the shape parameters and specific kill light intensity of each bacterium. The actual light intensities corresponding to the pulsed light sterilization component 102 and the excimer ultraviolet sterilization component 103 when the kill rate reaches 63.2% are extracted from the actual light intensity values ​​and the average kill rate data, respectively, as scale parameters. According to the definition that the specific kill light intensity is equal to the scale parameter divided by the sensitivity coefficient, the pulse sensitivity coefficient and excimer sensitivity coefficient of each bacterium are back-derived. The obtained shape parameters, pulse sensitivity coefficient, and excimer sensitivity coefficient are substituted into the Weibull model to verify whether the deviation between the predicted kill rate and the measured kill rate is less than or equal to 5% to ensure reliability. After verification, these parameters are associated with the bacterial species labels and stored to construct a bacterial feature library.

[0021] Furthermore, the quantitative method based on spectral analysis includes the following steps: A Gaussian weighted moving average is performed on each wavelength point of the fluorescence spectrum. A window with a width of 5 is defined with each wavelength point as the center. The fluorescence intensity of each wavelength point within the window and the wavelength distance between each non-center wavelength point and the center wavelength point are obtained. The non-center wavelength points are the wavelength points other than the center wavelength point within the window. The Gaussian weight of each non-center wavelength point is calculated using a Gaussian function. The Gaussian weight of each non-center wavelength point within the window is normalized by L1. The fluorescence intensity of each non-center wavelength point is weighted and summed according to the normalized Gaussian weights to obtain the smoothed fluorescence intensity of the center wavelength point. The fluorescence spectrum is traversed point by point to obtain the smoothed fluorescence spectrum. The fluorescence spectrum includes the inlet fluorescence spectrum and the outlet fluorescence spectrum. Based on the conventional condition that the wavelength resolution of the fluorescence spectrum is 1 nm, the standard deviation of the Gaussian function is finely adjusted between 1 and 2 according to the actual spectral resolution. In this embodiment, it is set to 1.5. Characteristic peaks with fluorescence intensity peaks greater than or equal to three times the average fluorescence intensity of the smoothed fluorescence spectrum are extracted. Adjacent characteristic peaks with peak position differences greater than or equal to the peak position difference threshold are integrated into characteristic peak groups. The fluorescence intensity peak of the next characteristic peak in each characteristic peak group is divided by the fluorescence intensity peak of the previous characteristic peak to obtain the measured ratio of each characteristic peak group. The standard ratio of the corresponding characteristic peak group in the fingerprint database is then used. The average of the ratios of the measured ratio and the standard ratio of each characteristic peak group is taken as a correction coefficient and multiplied by the fluorescence intensity values ​​of all wavelength points in the smoothed fluorescence spectrum to offset the systematic intensity shift caused by baseline drift, thus correcting and obtaining the unbiased fluorescence spectrum. The fingerprint database is pre-constructed under conditions of excitation wavelength 365 nm, ambient temperature 25 °C, relative humidity 50%, and aerosol concentration of 5000 to 10000 CFU / m³. 3 Under standard experimental conditions, each bacterium was measured three times. The intensity ratio of each characteristic peak group was calculated for the three measurements of each bacterium. The average of the three intensity ratios was taken as the standard ratio of the corresponding characteristic peak group for each bacterium. In this embodiment, the peak position difference threshold was set to 2nm. When the peak position difference between adjacent characteristic peaks, i.e., the peak position of the next peak minus the peak position of the previous peak, is greater than or equal to 2nm, they are determined to be the same characteristic peak group; otherwise, they are regarded as independent characteristic peaks. Characteristic peaks in the unbiased fluorescence spectrum whose fluorescence intensity peaks are greater than or equal to three times the average fluorescence intensity of the unbiased fluorescence spectrum are identified, and their peak positions, peak heights, half-widths (FWHMs), and peak areas are extracted to constitute macroscopic features. The sym8 wavelet basis is used as the basis function for continuous wavelet transform, with scales set to 1, 2, 3, 4, and 5. The unbiased fluorescence spectrum is decomposed at these five scales using continuous wavelet transform, and the mean and variance of the wavelet coefficients at each scale are extracted to constitute microscopic features. The global skewness and global kurtosis of the unbiased fluorescence spectrum are calculated to constitute statistical features. The macroscopic, microscopic, and statistical features are then concatenated to obtain a multimodal vector. Among them, macroscopic features reflect the content of bacteria and inlet, microscopic features distinguish closely related bacteria from the perspective of subtle fluctuations in peak shape, and statistical features supplement the description of peak shape differences. Continuous wavelet transform, global skewness calculation, and global kurtosis calculation are all existing mature algorithms or methods, and are not the focus of the technical solution described in this application, so they will not be elaborated on here. Since the fluorescence spectra of bacteria are not simply linear superpositions—for example, the interaction of metabolites from two bacteria can cause nonlinear changes in spectral peak positions or intensities—kernel nonnegative matrix factorization is employed, using radial basis kernels to transform the multimodal vectors. Mapping to a high-dimensional kernel space to capture nonlinear relationships and decomposing to obtain the optimal endmember matrix. and the optimal abundance matrix Each endmember in the endmember matrix corresponds to a bacterium in the bacterial feature library, and the abundance of the endmember is associated with the corresponding entry content. The kernel width of the radial basis kernel determines the similarity measure of samples in the high-dimensional space, and the value ranges from 0.1 to 10. The optimal kernel width is determined by minimizing the decomposition error. In this embodiment, the kernel width is set to 1. Based on the multi-task model, for the optimal endmember matrix Perform a classification task and use a convolutional neural network to extract the normalized optimal endmember matrix. The deep features of each endmember are extracted and flattened before being input into a fully connected layer. The dimensionality is reduced to 1×1 using cubic linear modulation, and the species label corresponding to each endmember is obtained by mapping using the Sigmoid function. The species label is used to define the bacteria. The optimal abundance matrix after normalization is then used. To perform the regression task, a fully connected layer with an attention mechanism is used. For each abundance, linear modulation, attention mechanism, and ReLU function are sequentially applied to achieve regression mapping, yielding the corresponding entry content. All species labels and their corresponding entry contents are then concatenated into an entry bacterial component table. The multi-task model requires pre-training. Pure cultures and mixed cultures of common bacteria are collected as aerosol samples. Fluorescence spectra are acquired under standard experimental conditions: excitation wavelength 365 nm, temperature 25℃, and relative humidity 50%. Multimodal vectors are obtained through preliminary steps of spectral analysis and quantification. The optimal endmember matrix is ​​then obtained through kernel nonnegative matrix factorization. and the optimal abundance matrix The true species labels and ingestion content of each bacterium in each sample were determined using the plate counting method. A dataset of 1000 samples was constructed, and the training and validation sets were divided in a 4:1 ratio. The optimal endmember matrix for each sample was then determined. and the optimal abundance matrix The input content was normalized, and the model parameters of the multi-task model were initialized. Batch training mode was adopted, with a batch size of 32. The Adam optimizer was selected, with an initial learning rate of 0.001. Every 50 rounds, the learning rate decayed to 0.8 times the original value, and the number of training rounds was set to 200. During forward propagation, classification and regression tasks were executed in parallel. The cross-entropy loss between the predicted and true class labels was calculated for the classification task, and the mean squared error loss between the predicted and true input content was calculated for the regression task. The total loss was the average of the two losses. The model parameters were updated through the backpropagation algorithm. The total loss of the validation set was monitored in real time during training, and an early stopping strategy was adopted. If the validation set loss did not decrease for 10 consecutive rounds, training was stopped. The accuracy of the classification task was evaluated by the confusion matrix, and the accuracy of the regression task was evaluated by the mean absolute error. Training was considered successful if and only if the accuracy was greater than or equal to 95% and the mean absolute error was less than or equal to 10%. The model parameters were saved. If the target was not met, the number of network layers, the number of neurons, or the learning rate decay coefficient were adjusted, and retraining was performed.

[0022] Specifically, in the multi-task model, the optimal endmember matrix Dimensions , Represents the dimension of hidden features in the kernel space. The number of known bacterial species in the representative bacterial feature library, and the optimal endmember matrix. After normalization, the input is fed into a convolutional neural network, which consists of three convolutional layers and three fully connected layers. The first convolutional layer uses 3×1 kernels with 32 kernels and a stride of 1. After ReLU activation, max pooling is performed. The second convolutional layer uses 3×1 kernels with 64 kernels and a stride of 1. After ReLU activation, max pooling is performed. The third convolutional layer uses 3×1 kernels with 128 kernels and a stride of 1. After ReLU activation, max pooling and flattening are performed to obtain a dimension of ( A vector of -8)×128 is passed through three fully connected layers, with the number of neurons in the three fully connected layers being 128, 64, and 128 respectively. The first two fully connected layers use the ReLU activation function, and the third fully connected layer uses the Sigmoid activation function. The output... A dimensional class label vector, which indicates the class label corresponding to each endmember.

[0023] Specifically, in the multi-task model, the optimal abundance matrix The dimension is Each row corresponds to the relative abundance of a known bacterium. The data is processed sequentially through a first hidden layer, a scaled dot product attention mechanism, a second hidden layer, and an output layer. The first hidden layer contains 64 neurons, which, after linear modulation and ReLU activation, produce 64-dimensional intermediate features. The scaled dot product attention mechanism, with a head count of 4, splits the 64-dimensional intermediate features according to the head count and calculates attention weights, outputting a weighted 64-dimensional intermediate feature. The second hidden layer contains 32 neurons, which, after linear modulation and ReLU activation, output 32-dimensional transitional output features. Finally, the output layer... The output layer of each neuron undergoes dimensionality adjustment, outputting... An ingress content vector of dimension, which indicates the ingress content corresponding to each endmember.

[0024] like Figure 2 As shown, further, kernel nonnegative matrix factorization includes the following steps: Multimodal vectors By mapping radial basis function kernels to a high-dimensional kernel space, the nonlinear relationships in multimodal vectors are approximated as linear relationships within the kernel space, resulting in a hybrid feature matrix. ; Within the kernel space, the objective is to mix the feature matrices. Approximate endmember matrix With abundance matrix The product of these terms yields the factorization formula. Among them, the endmember matrix Each column in the matrix represents a fingerprint vector of an endmember within the kernel space, and the abundance matrix... Each row corresponds to an endmember, and each column indirectly reflects the relative abundance of a bacterium, while satisfying sparsity, i.e., abundance matrix. Most of the elements in each column are 0; Due to the mixed feature matrix Without explicit representation, according to the representation theorem of radial basis kernels, any vector in the kernel space can be represented by a mixture of characteristic matrices. The row vectors of each row are linearly represented, and the endmember matrix is... Represented as a nonnegative coefficient matrix Transpose and hybrid feature matrix The product and back-substitution of the mixed feature matrix The decomposition formula is used to obtain the first decomposition formula. Multiply both sides of the first decomposition by the mixed characteristic matrix. The transpose of the expression yields the second factorization. Take the transpose of both sides of the second decomposition equation and combine the mixed characteristic matrix. With the mixed feature matrix The product of transposes is defined as the kernel matrix. Since the transpose of the kernel matrix remains unchanged, simplification yields the equivalent approximation. Therefore, the original mixed feature matrix The decomposition problem is transformed into an approximation of both sides of the equation, avoiding the mixing of characteristic matrices. Explicit computation; The core objective function is the Frobenius norm of the difference between the left and right sides of the equivalent approximation. This core objective function is used to measure the kernel matrix. The fitting error will affect the abundance matrix. The L1 norm and the preset regularization coefficient Multiply and superimpose with the core objective function to obtain the objective function, thereby synergistically introducing the abundance matrix. The requirement of sparsity, with minimizing the objective function as the optimization objective, and using a non-negative coefficient matrix. abundance matrix An optimization problem is constructed with nonnegativity as a constraint. The optimal nonnegative coefficient matrix is ​​obtained by iteratively solving the optimization problem using the alternating least squares method. and the optimal abundance matrix The optimal nonnegative coefficient matrix The transpose multiplied by the hybrid characteristic matrix Obtain the optimal endmember matrix Wherein, regularization coefficient The value is set to 0.01. Alternating least squares is an existing mature algorithm and is not the focus of this application's technical solution, so it will not be elaborated on here.

[0025] Furthermore, the sensitivity coefficients of each bacterium in the ingress bacterial composition table are obtained through a bacterial feature library, which stores the sensitivity coefficients of each bacterium. The bias coefficient of each bacterium is obtained by subtracting its excimer sensitivity coefficient from its pulse sensitivity coefficient. The ingress content and bias coefficient of each bacterium are concatenated to form its clustering feature. The KMeans++ algorithm is used for clustering. Normalization is performed on the clustering features of each bacterium to obtain its standard clustering feature. The standard clustering feature includes the standard bias coefficient and the standard ingress content. The number of clusters, the maximum number of iterations, and the change threshold are set to 2, 50, and 0.01, respectively. The bacteria with the largest and smallest standard bias coefficients are used as the pulse cluster center and the excimer cluster center, respectively. Each bacterium is assigned to the cluster corresponding to the cluster center with the closest feature distance. The feature distance is equal to the difference in standard bias coefficients multiplied by a first weight plus the difference in standard ingress content multiplied by a second weight. The first and second weights are 0.7 and 0.3, respectively. The clustering is then iterated through... Select any bacterium, determine the minimum feature distance between the selected bacterium and the bacterium in another cluster, and the average feature distance between the selected bacterium and the remaining bacterium in the same cluster, and calculate the difference. Divide the difference by the larger of the minimum feature distance between the selected bacterium and the bacterium in another cluster and the average feature distance between the selected bacterium and the remaining bacterium in the same cluster to obtain the silhouette coefficient of the selected bacterium. Take the average of the silhouette coefficients of all bacteria to obtain the global silhouette coefficient. The closer the silhouette coefficient is to 1, the more reasonable it is for the selected bacterium to belong to the current cluster. Calculate the average standard clustering features of all bacteria in the two clusters respectively, and use them as the new pulse cluster center and excimer cluster center in turn, and re-execute the next round of iterative clustering until the number of iterations is equal to the maximum number of iterations or the change in the global silhouette coefficient between two adjacent iterations is less than or equal to the change threshold. Stop the iteration and obtain the pulse cluster and the excimer cluster. Among them, the bacteria in the pulse cluster are more suitable for killing with the pulsed light sterilization component 102, while the bacteria in the excimer cluster are more suitable for killing with the excimer ultraviolet sterilization component 103.

[0026] Furthermore, the light intensity setting algorithm includes the following steps: The sensitivity coefficient and ingestion content of each bacterium in the cluster are obtained, and the shape and scale parameters of each bacterium in the cluster are synchronously retrieved from the bacterium feature library. The shape parameter reflects the steepness of the bacterium inactivation curve, and the scale parameter is used to calibrate the bacterium's specific inactivation light intensity. The ingestion content weight of each bacterium in the cluster is calculated, which is equal to the bacterium's ingestion content divided by the sum of the ingestion contents of all bacteria in the cluster. The inactivation light intensity is then initialized. Set the kill rate threshold and the deviation rate threshold, where cluster refers to pulse cluster or excimer cluster; In the current iteration, the scale parameter of each bacterium within the cluster is divided by the sensitivity coefficient to obtain the specific killing light intensity for each bacterium. The theoretical killing rate of each bacterium is then calculated using the Weibull model, which is as follows: , When the cluster is a pulsed cluster or an excimer cluster, the sensitivity coefficient is the pulse sensitivity coefficient or the excimer sensitivity coefficient, and the specific killing light intensity is the specific pulse killing light intensity or the specific excimer killing light intensity. For the first in the cluster Individual bacteria are killed by light intensity Theoretical kill rate , and They are respectively the first in the cluster The specific killing light intensity and shape parameters of each bacterium are used to calculate the cluster killing rate by multiplying the theoretical killing rate of each bacterium within the cluster by the ingestion content weight and summing the results. ; The inverse of the ratio of the cluster annihilation rate to the annihilation rate threshold is added to obtain the annihilation rate deviation rate. If the absolute value of the annihilation rate deviation rate is less than the deviation rate threshold, then the annihilation light intensity of the current iteration is... Cluster annihilation light intensity refers to the intensity of pulsed annihilation light. Or excimer laser annihilation intensity If the absolute value of the kill rate deviation rate is greater than or equal to the deviation rate threshold, the kill rate deviation rate is multiplied by the adjustment step size to obtain the undetermined adjustment amount. The feasible interval is used to truncate the undetermined adjustment amount. If the undetermined adjustment amount is greater than or equal to the lower limit of the feasible interval and less than or equal to the upper limit of the feasible interval, then the undetermined adjustment amount is used as the actual adjustment amount. If the undetermined adjustment amount is less than the lower limit of the feasible interval or greater than the upper limit of the feasible interval, then the lower limit or the upper limit of the feasible interval is used as the actual adjustment amount, based on the kill light intensity in the current iteration. The cumulative and actual adjustment amounts will be used to obtain the annihilation light intensity for the next iteration. The next iteration is then initiated, with the following thresholds set: kill rate threshold, deviation rate threshold, initial kill light intensity of the pulsed xenon lamp, initial kill light intensity of the ultraviolet lamp, adjustment step size, feasible range of the pulsed xenon lamp, and feasible range of the ultraviolet lamp: 95%, 2%, and 50 mW / cm², respectively. 2 30mW / cm 2 5mW / cm 2 20mW / cm 2 Up to 100mW / cm 2 and 10mW / cm 2 Up to 60mW / cm 2 .

[0027] Furthermore, the pulsed light intensity model of the pulsed xenon lamp is constructed based on the principles of pulsed emission and light propagation loss, and the radiant flux of the pulsed xenon lamp... It is the total light energy radiated outward per unit time, determined by the pulse luminescence efficiency. Xenon lamp power The radiant flux of a pulsed xenon lamp is determined by its pulsed operating characteristics and based on the principle of light energy conversion. Equal to pulse luminescence efficiency Xenon lamp power Pulse width The product of , since the pulsed xenon lamps are arranged longitudinally along the air duct, they can be approximated as parallel, equal-length line light sources relative to the pulse sterilization section. The longitudinal direction of the air duct is defined as . In the axial direction, for a line light source, it can be divided into countless infinitesimal segments using the method of infinitesimal elements. Each micro-segment The radiant flux is equal to the radiant flux of the pulsed xenon lamp. With micro segment The product divided by the length of the pulsed xenon lamp Each micro-segment A Lamborgh source can be approximated as a point-like source. The initial radiative intensity of the Lamborgh source is equal to the radiative flux and the solid angle of the hemisphere. The ratio, when each infinitesimal segment When diffused outward in the form of light, it exhibits a spherical diffusion pattern, but it can actually effectively act on any coordinate within the pulse sterilization zone. The only direction perpendicular to the pulsed xenon lamp is, therefore, when all micro-segments... xenon lamp length When performing integration, each infinitesimal segment The initial radiation intensity needs to be multiplied by a cosine of the deviation angle, where the deviation angle is for each infinitesimal element. to any coordinate The angle between the direction of the xenon lamp and the direction perpendicular to the pulsed xenon lamp, due to the length of the xenon lamp. Much larger than the maximum vertical length between the upper and lower surfaces within the pulse sterilization zone, an engineering approximation can be made during integration, such that the cosine of the deviation angle is... The initial radiant intensity of the pulsed xenon lamp is calculated by replacing the spherical radiant characteristics with cylindrical radiant characteristics. This initial radiant intensity is equal to the radiant flux of the pulsed xenon lamp. Divide by twice the length of the pulsed xenon lamp Introducing light propagation loss, light propagates from the pulsed xenon lamp to any coordinate. At that time, it undergoes cylindrical wave attenuation and air absorption attenuation. For cylindrical wave attenuation, the cylindrical wave attenuation and arbitrary coordinates Vertical Euclidean distance to the pulsed xenon lamp Directly related, the energy flux density of a line light source varies with the vertical Euclidean distance. The decay is inversely proportional to the attenuation rate. For air absorption attenuation, light propagating in air follows an exponential decay law; therefore, the first... A pulsed xenon lamp at coordinates Pulse annihilation light intensity at the location Specifically as follows: , in, For the first The radiant flux of a pulsed xenon lamp coordinates With the The vertical Euclidean distance of each pulse xenon lamp This refers to the pulse absorption coefficient, and all 12 pulse xenon lamps are of the same model and size. The length of each pulse xenon lamp is uniformly referred to. Similarly, ultraviolet lamps can also be considered as linear light sources. The excimer light intensity model of ultraviolet lamps is constructed based on the ultraviolet energy conversion characteristics and the principle of light propagation loss. The radiant flux of ultraviolet lamps... Equal to ultraviolet radiation efficiency and ultraviolet power The product of also follows the Lamborghian radiation characteristics and the principle of light propagation loss; therefore, the first A UV lamp at coordinates Excimer laser intensity at the site Specifically as follows: , in, For the first The radiant flux of a single ultraviolet lamp, coordinates With the The vertical Euclidean distance between each UV lamp tube The value represents the ultraviolet absorption coefficient, and all nine ultraviolet lamps are of the same model and size. The length of each UV lamp tube is used to refer to the UV lamp itself. It is worth noting that pulsed xenon lamps and ultraviolet lamps follow the same propagation loss principle. The difference lies in the principle by which pulsed xenon lamps and ultraviolet lamps generate radiation intensity.

[0028] Specifically, in this embodiment, the pulse luminescence efficiency... The settings should refer to the luminous efficiency range of mainstream pulsed xenon lamps in the industry, and combine the energy conversion characteristics of pulsed xenon lamps to ensure that the energy proportion of the effective sterilization band 200-400nm in the pulsed light meets the luminous efficiency standard in QB / T 5587-2021 "Pulsed High-Intensity Light Sterilizer for Food Processing Machinery". This standard is a light industry standard issued by the Ministry of Industry and Information Technology and can be publicly searched through the National Standardization Management Committee's National Standard Information Public Service Platform; pulsed luminous efficiency The pulsed xenon lamp length is set to 0.7 lm / W. and pulse width Pulse absorption coefficients at 1000 mm and 100 μs, respectively. The settings are derived from the air absorption characteristics data of 200-400nm pulsed light in the publicly published monograph "Modern Atmospheric Optics" under standard atmospheric conditions, with the pulse absorption coefficient set at 0.01m. -1 Ultraviolet radiation efficiency The design is based on the energy conversion characteristics of 222nm excimer ultraviolet lamps, ensuring that the proportion of 222nm ultraviolet light effectively used for sterilization in the energy generated by excimer discharge meets the requirements of GB 28235-2020 "Hygienic Requirements for Ultraviolet Sterilizers". This standard is a mandatory national standard issued by the State Administration for Market Regulation and the Standardization Administration of China, and can be publicly searched through the National Standards Information Public Service Platform of the Standardization Administration of China. (Ultraviolet radiation efficiency...) Set to 0.4, UV absorption coefficient The settings refer to the technical parameters related to atmospheric attenuation of ultraviolet light recorded in GB / T 32092-2015 "Technical Terminology for Ultraviolet Disinfection". These parameters can be publicly retrieved through the National Standardization Management Committee's National Standards Information Public Service Platform. The ultraviolet absorption coefficient... Set to 0.02m -1 UV lamp length With pulse xenon lamp length The same, also 1000mm.

[0029] Furthermore, since reflectors 104 are installed in the pulse sterilization section and excimer sterilization section inside the duct housing 100, typical points are selected inside the duct, and the light intensity at these typical points is measured under the same conditions with and without reflectors 104. Through multiple sets of pre-measurements, the pulse enhancement coefficient of the reflector 104 for the pulsed xenon lamp is determined by fitting. The UV enhancement coefficient of reflector 104 for UV lamp tube Since the pulsed light sterilization component 102 is arranged in a 4×3 array and the coordinate ranges of the pulse sterilization section and the excimer sterilization section are determined according to the size of the flow-through microbial aerosol disinfection device, the pulse light intensity models of the 12 pulsed xenon lamps of the pulsed light sterilization component 102 are summed and multiplied by the pulse enhancement coefficient. The intensity distribution of the pulse segment was obtained. ,in, , Representing the pulsed sterilization segment, similarly, the excimer light intensity models of the nine UV lamps in the excimer UV sterilization component 103 are summed and multiplied by the UV enhancement coefficient. The excimer segment light intensity distribution was obtained. ,in, , Representing the excimer laser sterilization segment, in this embodiment, the pulse enhancement coefficient... and UV enhancement coefficient They are 1.5 and 1.4 respectively.

[0030] Furthermore, a fitness function is uniformly set for both the pulse sterilization segment and the excimer sterilization segment. The pulse segment light intensity distribution is required. In the pulse sterilization section arbitrary coordinates within The intensity of the light at that location is greater than or equal to the intensity of the pulse annihilation light. , to extinguish the intensity of the pulsed light Subtract pulse segment light intensity distribution The distribution of pulse light intensity difference was obtained. Design coordinates Impulse penalty function at the location Distribution of pulse light intensity differences median coordinate The larger of the pulse intensity difference and 0 at the point is the pulse intensity distribution. In the pulse sterilization section arbitrary coordinates within The intensity of the light at that location is greater than or equal to the intensity of the pulse annihilation light. Time, coordinates Impulse penalty function at the location The value is 0 when the light intensity distribution of the pulse segment is... In the pulse sterilization section arbitrary coordinates within The light intensity at that location is less than the intensity of the pulse annihilation light. Time, coordinates Impulse penalty function at the location Similarly, when the value is negative, the light intensity distribution in the excimer segment is required. In the excimer sterilization zone arbitrary coordinates within The light intensity at the point of application is greater than or equal to the excimer laser annihilation light intensity. The intensity of the excimer laser beam will be reduced. Subtract excimer segment light intensity distribution The excimer light intensity difference distribution was obtained. Design coordinates excimer penalty function at the location Distribution of excimer light intensity difference median coordinate The excimer light intensity difference at that point is a relatively large value compared to 0. It is worth noting that this is due to the intensity distribution in the pulse segment. and excimer segment light intensity distribution The energy penalty function is designed based on the direct determination of the xenon lamp power of the 12 pulse xenon lamps and the ultraviolet power of the 9 ultraviolet lamps. It equals the sum of the xenon lamp power of 12 pulse xenon lamps and the ultraviolet power of 9 ultraviolet lamps. and These are the xenon lamp power group and the ultraviolet power group, used to record the xenon lamp power of 12 pulsed xenon lamps and the ultraviolet lamp power of 9 ultraviolet lamps, respectively, and the pulse penalty function is applied. In the pulse sterilization section All coordinates within The integral of the Sigmoid function value superimposed with the excimer penalty function In the excimer sterilization zone All coordinates within The integral of the Sigmoid function value minus the energy penalty function. The Sigmoid function value is used to obtain the fitness function. When the light intensity distribution of the pulse segment In the pulse sterilization section arbitrary coordinates within The light intensity at that location is less than the intensity of the pulse annihilation light. or excimer segment light intensity distribution In the excimer sterilization zone arbitrary coordinates within The light intensity at that location is less than the light intensity of the excimer laser annihilation. Or when the sum of the xenon lamp power and ultraviolet power is too high, the fitness function The value should be small to maximize the fitness function. To achieve the objective, the search space is divided based on the feasible ranges of xenon lamp power and ultraviolet power. A swarm intelligence algorithm is then used to iteratively search within the search space to obtain the fitness function. The maximum xenon lamp power of the 12 pulse xenon lamps is combined with the ultraviolet power of the 9 ultraviolet lamps to form the optimal xenon lamp power group. and optimal ultraviolet power group The feasible range for xenon lamp power is 3W to 10W, and the feasible range for ultraviolet power is 10W to 30W. The swarm intelligence algorithm is an existing algorithm, including but not limited to the gray wolf optimization algorithm, particle swarm algorithm, whale foraging algorithm and mouse swarm algorithm, which is not the focus of the technical solution of this application and will not be described in detail here.

[0031] Example 2 like Figures 3-4 As shown, a flow-through microbial aerosol disinfection device includes an air duct shell 100, a sensor 101, a pulsed light sterilization component 102, an excimer ultraviolet sterilization component 103, a reflector 104, and a control chip 105. The duct housing 100 serves as the basic load-bearing structure of the device, providing mounting support for the sensor 101, pulsed light sterilization component 102, excimer ultraviolet sterilization component 103, reflector 104, and control chip 105. It is manufactured based on corrosion-resistant and lightweight materials, including stainless steel and engineering plastics, and is designed as a rectangular sealed structure with a cross-sectional area of ​​700 mm × 700 mm to form a sealed disinfection duct, ensuring that the air containing bacterial aerosols flows along a fixed path to avoid leakage. The duct housing 100 includes a 100 mm inlet detection section, a 1000 mm pulse sterilization section, a 1000 mm excimer sterilization section, and a 100 mm outlet detection section. The inlet detection section has two preset sensor mounting holes, and the inner wall of the outlet detection section has one preset sensor mounting hole. The inner walls of the pulse sterilization section and the excimer sterilization section have preset sterilization component fixing grooves and reflector mounting holes. The outer wall of the duct housing 100 has a preset chip mounting groove in the center. There are three sensors 101 in total: an inlet fluorescence sensor, a flow rate sensor, and an outlet fluorescence sensor. The inlet fluorescence sensor and the flow rate sensor are fixed in two sensor mounting holes in the inlet detection section, respectively. The outlet fluorescence sensor is fixed in the sensor mounting hole in the outlet detection section. The detection ends of the sensors 101 all face the inside of the air duct and are perpendicular to the airflow direction. The mounting gaps are sealed with silicone rubber. Data interaction is achieved with the control chip 105 through the I2C / SPI bus. The sensor receives activation signals. The inlet fluorescence sensor emits excitation light of a specific wavelength into the air duct to capture the excitation fluorescence signal of bacteria, converts it into an inlet fluorescence spectrum, and feeds it back to the control chip 105. The flow rate sensor synchronously emits laser light into the inlet detection section. The flow rate is measured based on the Doppler frequency shift of the laser and fed back to the control chip 105. The sensor receives delayed activation signals. The outlet fluorescence sensor emits excitation light of a specific wavelength into the air duct to capture the excitation fluorescence signal of bacteria, converts it into an outlet fluorescence spectrum, and feeds it back to the control chip 105. The pulsed high-intensity light sterilization component 102 is deployed in the sterilization component fixing slot of the pulse sterilization section in the air duct housing 100 and is connected to the control chip 105 via a cable. It includes 4 sets of pulsed xenon lamp arrays, each set of pulsed xenon lamp arrays includes 3 xenon lamp tubes. The diameter and preset xenon lamp power of a single xenon lamp tube are 8 mm and 6 watts, respectively. They are arranged in a 4×3 array, with 4 sets along the width of the air duct and 3 lamps along the height of the air duct. The xenon lamp tubes are parallel to the airflow direction and are fixed by acrylic brackets. The spacing between the xenon lamp tubes is set to be less than or equal to 200 mm. The spacing between the xenon lamp tubes and the side wall of the air duct housing 100 is set to be less than or equal to 140 mm. The cross-sectional area of ​​the acrylic bracket strips is 50 mm × 50 mm. They are arranged longitudinally along the air duct, and the spacing between the bracket strips is set to be between 150 mm and 200 mm to form a turbulent environment and prolong the bacterial irradiation time. The excimer ultraviolet sterilization component 103 is deployed in the sterilization component fixing slot of the excimer sterilization section in the air duct housing 100 and is connected to the control chip 105 via a cable. It includes 9 ultraviolet lamps, each with a preset ultraviolet power of 20 watts, arranged in a 3×3 array. There are 3 ultraviolet lamps along the width and height of the air duct. The ultraviolet lamps are parallel to the airflow direction. The spacing between the ultraviolet lamps and the distance between the ultraviolet lamps and the side wall of the air duct housing 100 are both less than or equal to 200 mm. The reflector 104 covers the pulse sterilization section and excimer sterilization section inside the air duct shell 100. It is made of a rectangular aluminum plate with a thickness of 1 mm and a reflectivity greater than or equal to 0.9. The outer edge of the reflector 104 is fixed to the reflector mounting hole with bolts, and the reflective surface faces the center of the air duct to enhance the irradiation intensity. The control chip 105 executes the flow-through microbial aerosol disinfection method, records the system time, sends an activation signal to the sensor 101 at the beginning of each cycle, and receives the inlet fluorescence spectrum and flow rate. The composition table of inlet bacteria was obtained using spectral analysis and quantitative method, based on the total length of the air duct. and flow rate The delayed acquisition time is calculated and determined. The sensitivity coefficient of each bacterium in the ingress bacterial composition table is obtained and combined with the corresponding ingress content to form the clustering feature of each bacterium. The clustering algorithm is used to determine the pulse clusters and excimer clusters, and the pulse sterilization light intensity is obtained through the light intensity setting algorithm. and excimer laser annihilation intensity Based on the lamp layout of the pulsed high-intensity light sterilization component 102 and the excimer ultraviolet sterilization component 103, and considering the reflectivity of the reflector 104, the pulse light intensity distribution of each pulsed xenon lamp is obtained by superimposing the pulse light intensity model. The excimer light intensity distribution of the excimer segment is obtained by superimposing the excimer light intensity model of each ultraviolet lamp. The pulse segment light intensity distribution and the excimer segment light intensity distribution respectively satisfy the pulse annihilation light intensity and excimer laser annihilation intensity With the goal of minimizing total energy consumption, the optimal xenon lamp power group is determined iteratively using a swarm intelligence algorithm. and optimal ultraviolet power group The pulsed light sterilization component 102 and the excimer ultraviolet sterilization component 103 are adjusted, and the delayed acquisition time is waited for. A delayed activation signal is sent to the sensor 101, the exit fluorescence spectrum is received, and the exit bacterial composition table is obtained by spectral analysis quantification method. The actual killing rate of each bacterial is calculated by comparing the exit bacterial composition table and the inlet bacterial composition table, and the scale parameters of each bacterial are updated.

[0032] This invention discloses a flow-through microbial aerosol disinfection method and device, relating to the field of air sterilization. It constructs a closed-loop disinfection system based on a duct shell 100, a sensor 101, a pulsed light sterilization component 102, an excimer ultraviolet sterilization component 103, a reflector 104, and a control chip 105. In each cycle, inlet fluorescence spectra and flow rates are collected by an inlet fluorescence sensor and a flow rate sensor. The fluorescence spectra are smoothed, corrected, and feature-extracted using spectral analysis and quantitative methods, and multimodal vectors are constructed. Kernel nonnegative matrix factorization and a multi-task model are used to accurately obtain the types and corresponding contents of inlet bacteria, providing precise data support for targeted disinfection and avoiding the blindness of traditional disinfection methods. Clustering features are constructed based on the pulse sensitivity coefficient, excimer sensitivity coefficient, and inlet content of each bacterium in a bacterial feature library. A clustering algorithm iteratively divides pulse clusters and excimer clusters, allowing bacteria with different sensitivity characteristics to be matched with either pulsed light sterilization or excimer ultraviolet sterilization, significantly improving the targeting and effectiveness of disinfection. By utilizing a light intensity setting algorithm, combined with the Weibull model and the shape and scale parameters of bacteria, the pulsed and excimer laser light intensities are iteratively adjusted and determined. Pulsed and excimer laser light intensity models are constructed based on the principles of luminescence and light propagation loss. Combined with the reflective enhancement effect of reflector 104, a uniform and stable light intensity distribution is formed, ensuring balanced disinfection across the entire area. With the goal of meeting the disinfection light intensity requirements while minimizing total energy consumption, a swarm intelligence algorithm iteratively searches for the optimal xenon lamp power group and the optimal ultraviolet power group, achieving an optimal balance between energy consumption and disinfection effect. During delayed acquisition, the exit fluorescence spectrum is collected by an exit fluorescence sensor. The exit bacterial composition table is obtained through spectral analysis and quantitative analysis. After comparing and calculating the actual kill rate, the bacterial scale parameters are updated, enabling dynamic adaptation and adjustment of disinfection parameters. This allows the device to continuously adapt to the dynamic changes in bacterial aerosols, significantly improving long-term operational adaptability and economy.

[0033] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for disinfection of microbial aerosols via flow-through, characterized in that, Includes the following steps: In the current cycle, inlet fluorescence spectrum and flow rate are collected, and inlet bacterial composition table is obtained using spectral analysis quantification method. Delayed collection time is determined based on flow rate. Clustering was performed based on the pulse sensitivity coefficient, excimer sensitivity coefficient, and inlet concentration of bacteria to identify pulse clusters and excimer clusters; Using a light intensity setting algorithm, the theoretical killing rate of each bacterium in the pulse cluster or excimer cluster is calculated based on the shape and scale parameters of each bacterium, combined with the Weibull model. The pulse killing light intensity and excimer killing light intensity are determined iteratively based on the deviation between the theoretical killing rate and the ideal killing rate. Based on the principles of light emission and light propagation loss, the pulse intensity model of the pulsed xenon lamp and the excimer light intensity model of the ultraviolet lamp are determined. Combining the lamp layout and considering the reflectivity of the reflector, the pulse intensity distribution and the excimer light intensity distribution are obtained. With the goal of achieving the pulse annihilation light intensity and the excimer annihilation light intensity respectively while minimizing the total energy consumption, the optimal xenon lamp power group and the optimal ultraviolet power group are determined iteratively using a swarm intelligence algorithm and the regulation is executed. At the delayed acquisition time, the exit fluorescence spectrum was acquired, and the exit bacterial composition table was obtained by spectral analysis quantification method. The actual kill rate of each bacterial was calculated by comparing it with the inlet bacterial composition table. The scale parameters of each bacterial were updated by combining the Weibull model and waiting for the next cycle.

2. The method for disinfection of flow-through microbial aerosols as described in claim 1, characterized in that, The quantitative method of spectral analysis includes the following operations: smoothing and correcting the fluorescence spectrum, extracting characteristic peaks and integrating characteristic peak groups with the required peak position difference, calculating correction coefficients to obtain unbiased fluorescence spectrum, extracting multimodal vectors, obtaining endmember matrix and abundance matrix through kernel nonnegative matrix decomposition, and using a multi-task model to map and obtain the entry bacterial component table.

3. The method for disinfection of flow-through microbial aerosols as described in claim 1, characterized in that, The clustering process includes: obtaining the sensitivity coefficient and ingestion content of each type of bacteria, calculating the bias coefficient, constructing and normalizing clustering features, using clustering algorithms to iteratively determine the pulse clustering center and the quasi-molecular clustering center, and dividing the bacteria into corresponding clusters to obtain pulse clusters and quasi-molecular clusters.

4. The method for disinfection of flow-through microbial aerosols as described in claim 1, characterized in that, The light intensity setting algorithm includes: obtaining the shape and scale parameters of each bacterium in the cluster, calculating the inlet content weight, initializing the killing light intensity, calculating the theoretical killing rate of each bacterium using the Weibull model and combining it with the inlet content weight to obtain the cluster killing rate, iteratively adjusting the killing light intensity based on the deviation between the cluster killing rate and the killing rate threshold, and obtaining the pulse killing light intensity and the excimer killing light intensity.

5. The method for disinfection of flow-through microbial aerosols as described in claim 1, characterized in that, The construction of the pulsed light intensity distribution and the excimer light intensity distribution includes: constructing a pulsed light intensity model for a pulsed xenon lamp and an excimer light intensity model for an ultraviolet lamp based on the principles of light emission and light propagation loss; combining the reflection effect of the reflector; superimposing all pulsed light intensity models to obtain the pulsed light intensity distribution; and superimposing all excimer light intensity models to obtain the excimer light intensity distribution.

6. The method for disinfection of flow-through microbial aerosols as described in claim 1, characterized in that, Determining the optimal xenon lamp power set and the optimal ultraviolet power set involves: setting a fitness function, dividing the search space according to the feasible intervals of xenon lamp power and ultraviolet power, aiming to satisfy the annihilation light intensity and minimize total energy consumption, and iteratively searching through a swarm intelligence algorithm to obtain the optimal xenon lamp power set and the optimal ultraviolet power set.

7. The method for disinfection of flow-through microbial aerosols as described in claim 3, characterized in that, The construction of cluster features includes: calculating the difference between the pulse sensitivity coefficient and the excimer sensitivity coefficient of each bacterium to obtain the bias coefficient, and splicing the bias coefficient and the corresponding inlet content to obtain the cluster features.

8. The method for disinfection of flow-through microbial aerosols as described in claim 2, characterized in that, The kernel nonnegative matrix decomposition includes: mapping multimodal vectors to a high-dimensional kernel space through a radial basis kernel, constructing an equivalent approximation containing the kernel matrix, introducing regularization coefficients to construct an objective function, using nonnegativity as a constraint, iteratively solving to obtain the optimal nonnegative coefficient matrix and the optimal abundance matrix, and calculating the optimal endmember matrix.

9. A flow-through microbial aerosol disinfection device, characterized in that, The system includes a control chip (105), which records system time, sends an activation signal at the beginning of each cycle, receives inlet fluorescence spectrum and flow rate, obtains inlet bacterial composition table using spectral analysis quantification method and determines delayed acquisition time based on flow rate, obtains sensitivity coefficient of each bacterial inlet bacterial composition table and corresponding inlet content, determines pulse cluster and excimer cluster using clustering algorithm and obtains pulse killing light intensity and excimer killing light intensity through light intensity setting algorithm, constructs pulse segment light intensity distribution and excimer segment light intensity distribution based on pulsed xenon lamp pulse light intensity model and ultraviolet lamp excimer light intensity model respectively, with the goal of satisfying pulse killing light intensity and excimer killing light intensity and minimizing total energy consumption, uses swarm intelligence algorithm to iteratively determine optimal xenon lamp power group and optimal ultraviolet power group and executes regulation, waits until delayed acquisition time to send delayed activation signal, receives outlet fluorescence spectrum and obtains outlet bacterial composition table using spectral analysis quantification method, calculates actual killing rate of each bacterial based on comparison of outlet bacterial composition table and inlet bacterial composition table and updates scale parameters of each bacterial.

10. The flow-through microbial aerosol disinfection device as described in claim 9, characterized in that, It also includes a duct housing (100), a sensor (101), a pulsed light sterilization component (102), an excimer ultraviolet sterilization component (103), and a reflector (104); the duct housing (100) provides mounting support for the sensor (101), the pulsed light sterilization component (102), the excimer ultraviolet sterilization component (103), the reflector (104), and the control chip (105); the sensor (101) includes an inlet fluorescence sensor, a flow rate sensor, and an outlet fluorescence sensor, the inlet fluorescence sensor and the flow rate sensor receiving activation signals. The system collects the inlet fluorescence spectrum and flow rate and feeds them back to the control chip (105). The outlet fluorescence sensor receives the delayed activation signal, collects the outlet fluorescence spectrum and feeds it back to the control chip (105). The pulsed light sterilization component (102) includes 12 xenon lamps arranged in a 4×3 array. The xenon lamps are parallel to the airflow direction and are fixed by a bracket arranged longitudinally along the air duct. The excimer ultraviolet sterilization component (103) includes 9 ultraviolet lamps arranged in a 3×3 array. The ultraviolet lamps are parallel to the airflow direction. The reflector (104) is used to enhance the irradiation intensity.

Images