Methods, systems and sensors for measuring atmospheric aerosol concentration and particle optical scattering

Abstract

This invention relates to a method, system, and sensor for measuring atmospheric aerosol concentration and particle spectrum optical scattering, belonging to the field of aerosol optical measurement technology. The measurement method based on the principle of optical scattering offers advantages such as high measurement speed and non-contact measurement, allowing for rapid measurement without damaging the aerosol being measured. The deep learning model of this invention is trained using a large amount of data, exhibiting high generalization ability and accuracy. Using deep learning methods for regression tasks yields higher accuracy than traditional methods, resulting in more accurate measurements of aerosol concentration and particle spectrum. The optical scattering measurement device for the three-dimensional distribution field of atmospheric aerosol concentration and particle spectrum of this invention enables high-precision and wide-range optical scattering measurements of the three-dimensional distribution field of atmospheric aerosol concentration and particle spectrum.

Description

Technical Field

[0001] This invention belongs to the field of aerosol optical measurement technology, and more specifically, relates to methods, systems and sensors for measuring atmospheric aerosol concentration and particle optical scattering. Background Technology

[0002] With the acceleration of global industrialization and the expansion of urbanization, the accurate measurement of aerosol concentration and particle size distribution, as a core indicator of atmospheric environmental monitoring, has become a key technological requirement in environmental science, meteorological forecasting, public health, and the low-altitude economy. Aerosols directly affect the energy balance of the Earth-atmosphere system through the scattering and absorption of solar radiation, and their particle size distribution characteristics are significantly correlated with pollutant diffusion, cloud microphysical processes, and the spread of respiratory diseases. Given the current context of frequent extreme weather events and increasingly prominent air quality problems, the development of high-precision, high spatiotemporal resolution aerosol monitoring technologies is of great significance.

[0003] Currently, aerosol concentration and particle size distribution measurement techniques are mainly divided into two categories: contact sampling methods and non-contact sampling methods. While contact sampling methods such as optical particle counting or electromobility methods can provide accurate particle size distribution data, they require pumping to disrupt the in-situ state of the aerosols, and their sampling efficiency is limited by flow control and particulate deposition effects. In dynamic measurement environments, the response delay of these methods (typically >1 minute) makes it difficult to capture transient fluctuations in aerosol concentration. Furthermore, these sampling methods can only acquire local point data and cannot achieve spatial integration measurement across the entire optical path, leading to statistical biases in the concentration and particle size distribution within the region. Light scattering measurement, as a non-contact measurement method, has advantages such as strong real-time performance and wide measurement range, and is gradually becoming the mainstream method in the field of aerosol measurement. The multi-angle static light scattering (MAELS) technology, developed in recent years, receives scattered light intensity signals at different scattering angles (e.g., 30°, 90°, 150°) using multiple fixed detectors, and then uses Mie scattering theory to invert the particle size distribution. Because measurement systems typically only have 3-5 detectors, their angular resolution is insufficient, making it difficult to accurately reconstruct the aerosol distribution field of the measurement area. Furthermore, the Mie theory assumes that particles are uniformly spherical with a known refractive index, while actual aerosols (such as soot aggregates and mineral dust) often exhibit non-spherical, porous, or mixed hygroscopic characteristics, leading to significant errors between theoretical calculations and measured scattered light intensity. In summary, existing technologies suffer from limited measurement range and overly strong model assumptions. Therefore, there is an urgent need to develop an optical scattering measurement method and sensor for atmospheric aerosol concentration and particle spectrum three-dimensional distribution field with high accuracy and a wide measurement range. Summary of the Invention

[0004] This invention proposes an optical scattering measurement sensor and method for atmospheric aerosol concentration and three-dimensional particle distribution field. The measurement method based on the principle of optical scattering has the advantages of fast measurement speed and non-contact measurement, allowing for rapid measurement without damaging the aerosol being measured. The optical scattering measurement device for atmospheric aerosol concentration and three-dimensional particle distribution field of this invention can achieve high-precision and wide-range optical scattering measurements of atmospheric aerosol concentration and three-dimensional particle distribution field.

[0005] According to a first aspect of the present invention, an optical scattering measurement method for atmospheric aerosol concentration and three-dimensional particle spectrum distribution field is provided, comprising the following steps:

[0006] (1) Determine the size of the measurement space and the number of grids after three-dimensional spatial discretization, and construct a multi-wavelength, multi-scattering angle aerosol scattered light intensity dataset;

[0007] (2) Build a regression model based on aerosol concentration and particle size distribution. Train the regression model on the aerosol scattered light intensity dataset obtained in step (1) and save the model with the minimum loss function as the optimal model.

[0008] (3) The laser emits a first wavelength beam perpendicular to the cube grid. N photodetectors collect the scattered light intensity signal respectively and transmit the collected signal to the main controller. Then the laser emits a second wavelength beam, and the N photodetectors continue to collect the scattered light intensity signal until all M wavelengths of the laser are emitted and collected, and M×N scattered light intensity data are obtained.

[0009] (4) After the M×N scattered light intensity data obtained in step (3) are expanded in one dimension, they are input into the optimal model saved in step (2) for regression to obtain the aerosol concentration and particle spectrum of the grid through which the laser beam passes.

[0010] (5) After the laser moves to the next measurement position, the operation is repeated according to steps (3) and (4) until the concentration and particle size distribution of all grids in the space are measured, and the three-dimensional distribution field of aerosol concentration and particle size distribution is obtained.

[0011] Preferably, in step (1), the aerosol particle spectrum in each grid is described by a log-normal model, denoted as . :

[0012]

[0013] in and These represent the median particle size and standard deviation, respectively, with d ranging from 10 to 1000 nm; particle size Set to 250~750 nm; set the standard deviation Set it to 1.2~2.0.

[0014] Preferably, in step (2), a regression model based on aerosol concentration and particle size distribution is constructed; the input of the deep learning model is scattered light intensity data, where the first layer is a hidden layer, and the fully connected layer of the hidden layer expands the dimension of the input data; then the data is standardized, and the feature vector of the expanded data is processed; then the ReLU activation function is added to introduce nonlinearity. The formula of the ReLU activation function is as follows, where x is the output of the previous neural network layer:

[0015]

[0016] Next, neurons are randomly deactivated, and deactivated neurons are set to zero to prevent overfitting. The second layer is an expansion layer, a fully connected layer that further expands the dimensions of the input data, then standardizes the data, introduces the ReLU activation function, and finally maintains the same deactivation rate. The third layer is a compression layer, a fully connected layer that compresses the data dimensions to prevent overparameterization, then performs standardization and activation. The fourth layer is a feature extraction layer, a fully connected layer that further compresses the data dimensions, extracts high-order features, and finally introduces an activation function. The last layer is the output layer, which maps the compressed data to the output data, which represents the aerosol concentration for each grid cell. Median particle size and standard deviation The regression model was trained on the entire scattered light intensity dataset, and the model with the minimum loss function was saved as the optimal model. The smooth mean squared error loss was used as the loss function during training, and the calculation formula is as follows:

[0017]

[0018] in, Let i be the true value of the i-th sample. Let be the predicted value for the i-th sample. To control the threshold of the smoothing range, L This is the loss value.

[0019] Preferably, in step (3), the M×N scattered light intensity data are obtained as a scattered light intensity data matrix as shown below; where Is represents the scattered light intensity, λ represents the wavelength, and D represents the detector;

[0020] .

[0021] According to another aspect of the present invention, an optical scattering measurement system for atmospheric aerosol concentration and three-dimensional particle spectrum distribution field is provided, comprising:

[0022] Dataset Construction Module: Used to determine the size of the measurement space and the number of grids after discretization of the three-dimensional space, and to construct a multi-wavelength, multi-scattering angle aerosol scattered light intensity dataset;

[0023] Dataset training module: used to build a regression model based on aerosol concentration and particle size distribution, train the regression model on the aerosol scattered light intensity dataset, and save the model with the minimum loss function as the optimal model;

[0024] Laser emission and acquisition module: used to emit a first wavelength beam of laser perpendicular to the cube grid, N photodetectors collect the scattered light intensity signal respectively, and transmit the collected signal to the main controller. Then the laser emits a second wavelength beam, and the N photodetectors continue to collect the scattered light intensity signal until all M wavelengths of the laser are emitted and collected, and M×N scattered light intensity data are obtained.

[0025] Data regression module: used to expand the acquired M×N scattered light intensity data in one dimension and input it into the saved optimal model for regression, to obtain the aerosol concentration and particle size distribution of the grid through which the laser beam passes.

[0026] The laser emission acquisition module after the measurement position changes: It is used to move the laser to the next measurement position and then operate again according to the laser emission acquisition module and the data regression module until the concentration and particle spectrum of all grids in the space are measured, so as to obtain the three-dimensional distribution field of aerosol concentration and particle spectrum.

[0027] According to another aspect of the present invention, an optical scattering measurement sensor for atmospheric aerosol concentration and particle size distribution three-dimensional field is provided, comprising a main controller, a laser, and N photodetectors; the main controller is used to control the movement of the sensor, to control whether the laser emits light, and to control the acquisition process of the photodetectors; the laser is used to emit beams of M different wavelengths; the photodetectors are used to receive scattered light intensity signals.

[0028] N is a positive integer greater than or equal to 2; M is a positive integer greater than or equal to 1.

[0029] Preferably, it also includes a power management module, which is used for sensor charging and discharging and power management.

[0030] Preferably, 3≤N≤5, 1≤M≤3.

[0031] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0032] (1) The measurement method based on the principle of optical scattering of the present invention has the advantages of fast measurement speed and non-contact measurement, and can quickly complete the measurement without damaging the aerosol being measured.

[0033] (2) The deep learning model of this invention is trained with a large amount of data. The model has high generalization ability and accuracy. The deep learning method has higher accuracy than the traditional method for regression tasks, making the measurement of aerosol concentration and particle size more accurate.

[0034] (3) The laser and photodetector of the sensor of the present invention are integrated into one design and the sensor can be moved arbitrarily. Therefore, the present invention can provide a portable measurement solution. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the sensor measurement according to the present invention.

[0036] Figure 2 This is a schematic diagram of the three-dimensional spatial discretization of the present invention.

[0037] Figure 3 This is a structural diagram of the deep learning model of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0039] This invention proposes an optical scattering measurement method and sensor for the three-dimensional distribution field of atmospheric aerosol concentration and particle size distribution. The sensor includes a main controller, a laser capable of emitting M wavelengths, N photodetectors, and a power management module. The main controller controls the movement of the sensor, the emission of the laser, and the acquisition process of the photodetectors. The laser emits a beam of a specific wavelength, the photodetectors receive the scattered light intensity signal, and the power management module manages the charging, discharging, and power of the entire system. The sensor can move arbitrarily in space, measuring the aerosol concentration and particle size distribution of the entire space through scanning. To facilitate measurement, the three-dimensional space is discretized into several uniform cubic grids. The scattered light intensity received by the photodetectors is the sum of the scattered light intensities of the laser beam passing through all grids. The aerosol concentration and particle size distribution in each grid are the measurement objects. To measure these parameters, a laser capable of emitting M wavelengths and N photodetectors are arranged on the sensor. The photodetectors receive backscattered light signals (scattering angle greater than 90°).

[0040] This invention achieves the measurement of aerosol concentration and particle size distribution in space through the following steps:

[0041] Step 1: Determine the size of the measurement space and the number of grids after discretization of the three-dimensional space, and construct a multi-wavelength, multi-scattering angle aerosol scattered light intensity dataset;

[0042] Step 2: Build a regression model based on aerosol concentration and particle size distribution, train the regression model on the entire scattered light intensity dataset, and save the optimal model;

[0043] Step 3: After the sensor reaches the measurement position, the laser emits a first wavelength beam perpendicular to the cube grid. The photodetector collects the scattered light intensity signal and transmits the collected signal to the main controller. The laser then emits a second wavelength beam, and the photodetector continues to collect the scattered light intensity signal until all M wavelengths are emitted and collected. The sensor acquires a total of M×N scattered light intensity data; where I represents the scattered light intensity, λ represents the wavelength, and D represents the detector.

[0044] ;

[0045] Step 4: Expand the M×N data obtained in Step 3 into one dimension and input it into the optimal model saved in Step 2 for regression to obtain the concentration and particle spectrum of the grid through which the laser beam passes.

[0046] Step 5: After the sensor moves to the next measurement position, repeat steps 3 and 4 until the concentration and particle size distribution of all grids in the space are measured, and finally obtain the three-dimensional distribution field of aerosol concentration and particle size distribution.

[0047] Figure 1 This is a schematic diagram of the sensor measurement of the present invention. The sensor includes a main controller, a power management module, a laser capable of emitting three different wavelengths of laser light, and five photodetectors. The main controller runs various algorithms, controls the movement of the sensor, determines whether the laser emits a beam, and manages the acquisition process of the photodetectors. The power management module is responsible for the charging, discharging, and power management of the entire system, prompting for charging when the battery is low. The laser can emit laser light with wavelengths of 450 nm, 650 nm, and 950 nm. The photodetectors receive the scattered light intensity signal and return the collected scattered light intensity data to the main controller. The sensor can be fixed to a drone, which carries the sensor and moves it within the measurement space to complete the measurement. Figure 2 This is a schematic diagram of the three-dimensional spatial discretization of the present invention. The space to be measured is a cuboid region with a size of 9 m × 9 m × 15 m, as shown below. Figure 2 The diagram shows the device uniformly divided into 45 identical cubes, each measuring 3 m × 3 m × 3 m. A drone carrying sensors moves within the xoy plane, with the laser beam perpendicular to the xoy plane (parallel to the z-axis).

[0048] Example 1

[0049] This invention achieves the measurement of aerosol concentration and particle size distribution in space through the following steps:

[0050] Step 1: The measurement area is a 9 m × 9 m × 15 m cuboid region. After discretization in three-dimensional space, the number of grids is 45 (3 × 3 × 5). The sensor moves in the xoy plane, and the laser beam can pass through 5 grids along the z-axis. Therefore, the intensity of the scattered light received by each photodetector is the sum of the scattered light intensities of the 5 grids. Based on Mie scattering theory, a multi-wavelength, multi-scattering-angle aerosol scattered light intensity dataset is constructed. The aerosol particle spectrum in each grid is described by a log-normal model, denoted as [data missing]. :

[0051] (1)

[0052] in and denoted as median particle size and standard deviation, respectively, with d ranging from 10 to 1000 nm.

[0053] Therefore, solving for the particle size distribution can be transformed into solving for the median particle size in a normal distribution model. and standard deviation To ensure that the simulation dataset can cover aerosols of various concentrations and particle sizes, the median particle size is used when calculating the simulated scattered light intensity. Set the range to 250~750 nm, with an interval of 50 nm, and set the standard deviation. Set to 1.2~2.0, with an interval of 0.2, concentration The values ​​are expressed in normalized form, ranging from 0.2 to 1.0, with intervals of 0.2. After setting the simulation parameters, a dataset of scattered light intensity is constructed.

[0054] Step 2: Build a regression model based on aerosol concentration and particle size distribution, such as... Figure 3 As shown. The input to the deep learning model is scattered light intensity data, with a dimension of 1×15. The first layer is a hidden layer. First, a fully connected layer expands the input data from 15 dimensions to 256 dimensions; then the data is standardized to process the 256-dimensional feature vector; finally, the ReLU activation function is added to introduce non-linearity. The formula for the ReLU activation function is as follows:

[0055] (2)

[0056] Finally, 30% of neurons are randomly deactivated to prevent overfitting. The second layer is an expansion layer. First, the fully connected layer expands the input data from 256 dimensions to 512 dimensions to capture more complex features. Then, the data is standardized, and the ReLU activation function is introduced. Finally, the same deactivation rate is maintained. The third layer is a compression layer. First, the fully connected layer compresses the data dimension from 512 dimensions to 256 dimensions, mainly to prevent overparameterization. Then, standardization and activation are performed. The fourth layer is a feature extraction layer, which further compresses the data dimension to 128 dimensions to extract higher-order features. Finally, an activation function is introduced. The last layer is the output layer, which maps the 128-dimensional data to a 15-dimensional output. The 15 data points are the aerosol concentrations of 5 grids. Median particle size and standard deviation The regression model was trained on the entire scattered light intensity dataset, and the optimal model was saved. During training, the smooth mean squared error loss (SSL Loss) was used as the loss function, calculated as follows:

[0057] (3)

[0058] in, Let i be the true value of the i-th sample. Let be the predicted value for the i-th sample. To control the smoothing range, the threshold is typically set to 1.

[0059] Step 3: The UAV, carrying sensors, moves to the (1,3) coordinate position in the xoy plane. First, it emits a 450 nm wavelength laser along the z-axis, and five photodetectors receive the scattered light intensity. Then, it emits a 650 nm wavelength laser at the (1,3) coordinate position, and the five photodetectors continue to receive the scattered light intensity. Finally, it emits a 950 nm wavelength laser, and the five photodetectors receive the scattered light intensity. The resulting scattered light intensity data matrix is ​​shown below:

[0060]

[0061] Step 4: Expand the scattered light intensity data matrix obtained in Step 3 into one dimension and input it into the optimal model saved in Step 2 for regression analysis. This will yield the concentration of the laser beam passing through the grid. Particle size distribution (median particle size) and standard deviation ).

[0062] Step 5: After the sensor moves to the measurement position (2,3), repeat steps 3 and 4 until the concentration and particle size distribution of all grids in the space are measured, and finally obtain the three-dimensional distribution field of aerosol concentration and particle size distribution.

[0063] This invention enables the measurement of concentration and particle size distribution in 45 discrete grids of a space without damaging the aerosol being measured.

[0064] The present invention allows for the measurement of atmospheric aerosol concentration and particle size distribution in three-dimensional space simply by moving the sensor in a plane. The measurement speed is fast and the distance moved during the measurement process is small.

[0065] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of measuring atmospheric aerosol concentration and size distribution optical scattering, characterized by, Comprising the following steps: (1) determining the size of the measurement space and the number of grids after three-dimensional space discretization processing, constructing a multi-wavelength, multi-scattering angle aerosol scattering light intensity dataset; (2) building a regression model based on aerosol concentration and particle size, training the regression model on the aerosol scattering light intensity dataset obtained in step (1), and saving the model with the smallest loss function as the optimal model; Specifically, the input of the deep learning model is the scattered light intensity data, wherein the first layer is a hidden layer, and a full connection layer of the hidden layer expands the dimension of the input data; the data is standardized, and a feature vector of the expanded data is processed; then, a ReLU activation function is added to introduce nonlinearity, and the formula of the ReLU activation function is as follows, wherein x is the output of the previous layer of neural network The neurons are randomly inactivated again, and the inactivated neurons are set to zero to prevent overfitting; the full connection layer of the expansion layer expands the dimension of the input data again, then the data is standardized, a ReLU activation function is introduced, and finally the same inactivation rate is maintained; the full connection layer of the compression layer of the third layer compresses the data dimension to prevent over-parameterization, then the data is standardized and activated; the full connection layer of the feature extraction layer of the fourth layer further compresses the data dimension to extract high-order features, and finally an activation function is introduced; the last layer is the output layer, which finally maps the compressed data to the output data, and the output data is the aerosol concentration of each grid , median particle size , and standard deviation ; the regression model is trained on the entire scattered light intensity dataset, and the model with the smallest loss function is saved as the optimal model. During training, the smooth mean square error loss is used as the loss function, and the calculation formula is as follows: wherein, is the true value for the i th sample, is the predicted value for the i th sample, is a threshold value for the control smoothing range, L is a loss value; (3) the laser emits a first wavelength beam perpendicular to the cubic grid, N photodetectors respectively collect scattering light intensity signals, and transmit the collected signals to the main controller, then the laser emits a second wavelength beam, the N photodetectors continue to collect scattering light intensity signals, until the M wavelengths of the laser are all emitted and collected, M×N scattering light intensity data are obtained; (4) one-dimensional expansion of the M×N scattering light intensity data obtained in step (3) is input into the optimal model saved in step (2) for regression, and the aerosol concentration and particle size of the grid through which the laser beam passes are obtained; (5) the laser moves to the next measurement position and performs steps (3) and (4) again, until the concentration and particle size of all grids in the space are measured, and the three-dimensional distribution field of the aerosol concentration and particle size is obtained.

2. The method of atmospheric aerosol concentration and size distribution optical scattering measurements of claim 1, wherein, In step (1), the aerosol size distribution in each grid is described by a lognormal model, denoted as : wherein and respectively median particle size and standard deviation, d The value range of D50 is 10~1000 nm; the particle size The value range of D50 is set to 250~750 nm; the standard deviation The value range of D50 is set to 1.2~2.

0.

3. The method of atmospheric aerosol concentration and size distribution optical scattering measurements of claim 1, wherein, In step (3), the MxN scattered light intensity data is obtained as a scattered light intensity data matrix as shown below; wherein Is wherein I denotes the scattered light intensity, λ denotes the wavelength, and D denotes the detector. 。 4. An atmospheric aerosol concentration and size distribution optical scattering measurement system, characterized by, Comprise: Dataset construction module: for determining the size of the measurement space and the number of grids after three-dimensional space discretization processing, constructing a multi-wavelength, multi-scattering angle aerosol scattering light intensity dataset; Dataset training module: for building a regression model based on aerosol concentration and particle size, training the regression model on the aerosol scattering light intensity dataset, and saving the model with the smallest loss function as the optimal model; Laser emission and collection module: for emitting a first wavelength beam perpendicular to the cubic grid by the laser, N photodetectors respectively collect scattering light intensity signals, and transmit the collected signals to the main controller, then the laser emits a second wavelength beam, the N photodetectors continue to collect scattering light intensity signals, until the M wavelengths of the laser are all emitted and collected, M×N scattering light intensity data are obtained; The data regression module is used for inputting the obtained M*N scattered light intensity data after one-dimensional expansion into the saved optimal model for regression, so as to obtain the aerosol concentration and particle spectrum through which the laser beam passes; the input of the deep learning model is the scattered light intensity data, wherein the first layer is a hidden layer, the full connection layer of the hidden layer expands the dimension of the input data; then the data is standardized, and the feature vector of the expanded data is processed; then the ReLU activation function is added, and the nonlinearity is introduced, and the formula of the ReLU activation function is as follows, wherein x is the output of the previous layer of neural network The neurons are randomly inactivated again, and the inactivated neurons are set to zero to prevent overfitting; the full connection layer of the expansion layer expands the dimension of the input data again, then the data is standardized, a ReLU activation function is introduced, and finally the same inactivation rate is maintained; the full connection layer of the compression layer of the third layer compresses the data dimension to prevent over-parameterization, then the data is standardized and activated; the full connection layer of the feature extraction layer of the fourth layer further compresses the data dimension to extract high-order features, and finally an activation function is introduced; the last layer is the output layer, which finally maps the compressed data to the output data, and the output data is the aerosol concentration of each grid median particle size and standard deviation ; the regression model is trained on the entire scattered light intensity dataset, and the model with the smallest loss function is saved as the optimal model. During training, the smooth mean square error loss is used as the loss function, and the calculation formula is as follows: wherein, is the true value for the i th sample, is the predicted value for the i th sample, is a threshold value for the control smoothing range, L is a loss value; Laser emission and collection module after measurement position change: for moving the laser to the next measurement position and performing the laser emission and collection module and data regression module again, until the concentration and particle size of all grids in the space are measured, and the three-dimensional distribution field of the aerosol concentration and particle size is obtained.

Images