CN104535465A - PM2.5 concentration detection method and device based on neural network - Google Patents

Abstract

The invention provides a PM2.5 concentration detection method and device based on a neural network. A laser detection system is firstly built based on the Fraunhofer diffraction theory, then a corresponding function relation between a light strength digital signal and a PM2.5 concentration value is verified according to the light strength digital signal collected by the laser detection system based on the Fraunhofer diffraction theory, the light strength digital signal is adopted as an input quantity, a regularized neural network model is built, and the PM2.5 concentration value is output. The method overcomes the defect that a PM2.5 concentration detection method in the prior art is low in automation degree, repeated detection can be achieved, detection precision is high, and calculation is easy and convenient.

Description

PM2.5 concentration detection method and device based on neural network

The technical field is as follows:

the invention relates to the field of concentration detection, in particular to a PM2.5 concentration detection method and device based on a neural network.

Background art:

in recent years, severe haze and polluted air appear in the middle and east of China successively, and relevant researches show that PM2.5 is the chief culprit of haze weather. PM2.5 refers to particles with aerodynamic diameter less than or equal to 2.5 μm in the atmosphere, also called as particles capable of entering lung, compared with the thicker atmospheric particles, PM2.5 has small particle size, is rich in a large amount of toxic and harmful substances, has long retention time in the atmosphere and long conveying distance, pollutes the atmospheric environment and poses serious threat to human health.

The existing PM2.5 concentration detection method mostly adopts a manual gravimetric method, a beta-ray attenuation method and a micro-oscillation balance method. However, the manual gravimetric method for detecting PM2.5 has the problems of low automation degree, poor detection repeatability, easy generation of accumulation errors, medium consumption and the like, and the lower limit of particle detection of the beta-ray attenuation method and the micro-oscillation balance method is difficult to reach an ideal level.

The invention content is as follows:

aiming at the defects of the prior art, the invention provides a PM2.5 concentration detection method and device based on a neural network, which overcome the defect of low automation degree of the PM2.5 concentration detection method in the prior art, can realize repeated detection, and has high detection precision and simple and convenient calculation.

In one aspect, the invention provides a PM2.5 concentration detection method based on a neural network, including:

establishing a laser detection system based on the Fraunhofer diffraction principle;

verifying that the light intensity digital signal has a corresponding function relation with a PM2.5 concentration value based on a Fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system;

and establishing a regularization neural network model by taking the light intensity digital signal as an input quantity, and outputting a PM2.5 concentration value.

Optionally, the laser detection system comprises: the device comprises a power supply unit, a He-Ne laser, a filter lens, a beam expanding lens, an air pump, an air chamber, a signal receiving unit, a detection unit and a calculation unit;

wherein the power supply unit is used for providing power supply for the He-Ne laser, the air pump, the detection unit and the calculation unit;

the signal receiving unit comprises a Fourier lens and 70 paths of photodetectors and is used for receiving the optical signals diffracted by the air particles and converting the optical signals into light intensity analog signals;

the detection unit is used for converting the light intensity analog signal into a light intensity digital signal after amplification processing;

the calculation unit is used for modeling according to the light intensity digital signal and calculating to obtain a PM2.5 concentration value;

the power supply unit is respectively connected with the He-Ne laser, the air pump, the detection unit and the calculation unit, the He-Ne laser is connected with the filter lens, the air chamber is respectively connected with the beam expanding lens, the air pump and the signal receiving unit, the signal receiving unit is connected with the detection unit, and the detection unit is connected with the calculation unit.

Optionally, verifying, based on a fraunhofer diffraction principle, that the light intensity digital signal has a corresponding functional relationship with a PM2.5 concentration value according to the light intensity digital signal acquired by the laser detection system includes:

and calculating and processing the light energy distribution vector measured by the photoelectric detector to obtain the particle size distribution of the particle group to be measured, adding the particle size distribution percentages with the diameters smaller than 2.5 mu m to obtain a PM2.5 concentration value, and verifying that the light intensity digital signal and the PM2.5 concentration value have a corresponding relation.

Optionally, the particle size distribution of the particle group to be detected is calculated by the following formula,

E＝TW

wherein E is the light energy distribution vector measured by the photoelectric detector, $T=\left[\begin{array}{ccc}{t}_{11}& ...& t_{1M}\\ ...& & \\ t_{N1}& ...& t_{\mathrm{NM}}\end{array}\right]$ w is a particle size distribution vector of the particle group to be detected;

wherein, $t_{\mathrm{NM}}=C\frac{1}{D_M}[J_0^2\left(X_{M,N}\right)+J_1^2\left(X_{M,N}\right)-J_0^2\left(X_{M,N-1}\right)-J_1^2\left(X_{M,N+1}\right)]$ is of diameter D_MThe light energy of the nth detection ring of the detector,of particles of different diameter M, being a fixed constantClass I₀Is the intensity of incident light, J₀(X_M,N) Is a zero order Bessel function, J₁(X_M,N) Is a first order Bessel function.

Alternatively, the PM2.5 concentration value, calculated by the following equation,

wherein, beta_{Density of air}The value of 1.205kg/m at 20 ℃ under standard atmospheric pressure³And alpha is the particle size distribution with a diameter of less than 2.5 μm.

Optionally, the step of establishing a regularization neural network model by using the light intensity digital signal as an input quantity and outputting a value of the concentration of PM2.5 includes:

initializing neural network parameters;

and training the weight of the neural network model by adopting a least mean square algorithm, and adjusting the weight.

Optionally, the initializing neural network parameters specifically include:

selecting the number of hidden layer nodes of the neural network model to be equal to the number of training samples, selecting a Gaussian function as an excitation function, selecting the center of the excitation function as a sample data center, and initializing the weight and excitation function expansion constant of the neural network.

In another aspect, the present invention provides a PM2.5 concentration detection apparatus based on a neural network, including:

the laser detection system establishing unit is used for establishing a laser detection system based on the Fraunhofer diffraction principle;

the functional relationship verification unit is used for verifying that the light intensity digital signal and the PM2.5 concentration value have a corresponding functional relationship based on the Fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system;

and the neural network model calculation unit is used for establishing a regularization neural network model by taking the light intensity digital signal as an input quantity and outputting a PM2.5 concentration value.

Optionally, the neural network model calculating unit includes:

the parameter initialization module is used for initializing neural network parameters;

and the weight training unit is used for training the weight of the neural network model by adopting a least mean square algorithm and adjusting the size of the weight.

According to the technical scheme, the PM2.5 concentration detection method and device based on the neural network firstly establish a laser detection system based on the Fraunhofer diffraction principle, then verify that the light intensity digital signal and the PM2.5 concentration value have a corresponding function relation based on the Fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system, establish a regularized neural network model by taking the light intensity digital signal as an input quantity, and output the PM2.5 concentration value, and the method overcomes the defect that the PM2.5 concentration detection method in the prior art is low in automation degree, can realize repeated detection, and is high in detection precision and simple and convenient to calculate.

Description of the drawings:

fig. 1 is a schematic flow chart of a PM2.5 concentration detection method based on a neural network according to a first embodiment of the present invention;

FIG. 2 is a schematic diagram of a neural network according to a first embodiment of the present invention;

fig. 3 is a schematic flow chart of a PM2.5 concentration detection method based on a neural network according to a second embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a laser inspection system according to a second embodiment of the present invention;

FIG. 5 is a diagram illustrating a neural network model fitting 5 sets of data according to a second embodiment of the present invention;

fig. 6 is a schematic structural diagram of a PM2.5 concentration detection apparatus based on a neural network according to a third embodiment of the present invention;

fig. 7 provides comparison results of concentration values of PM2.5 according to the third embodiment of the present invention.

The specific implementation mode is as follows:

the following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.

Fig. 1 shows a schematic flow chart of a method for detecting a PM2.5 concentration based on a neural network according to a first embodiment of the present invention, as shown in fig. 1, the method of this embodiment is described as follows.

101. A laser detection system is established based on the Fraunhofer diffraction principle.

In this step, the laser detection system includes: the device comprises a power supply unit, a He-Ne laser, a filter lens, a beam expanding lens, an air pump, an air chamber, a signal receiving unit, a detection unit and a calculation unit;

wherein the power supply unit is used for providing power supply for the He-Ne laser, the air pump, the detection unit and the calculation unit;

the signal receiving unit comprises a Fourier lens and 70 paths of photodetectors and is used for receiving the optical signals diffracted by the air particles and converting the optical signals into light intensity analog signals;

the detection unit is used for converting the light intensity analog signal into a light intensity digital signal after amplification processing;

the calculation unit is used for modeling according to the light intensity digital signal and calculating to obtain a PM2.5 concentration value;

the power supply unit is respectively connected with the He-Ne laser, the air pump, the detection unit and the calculation unit, the He-Ne laser is connected with the filter lens, the air chamber is respectively connected with the beam expanding lens, the air pump and the signal receiving unit, the signal receiving unit is connected with the detection unit, and the detection unit is connected with the calculation unit.

102. And verifying the corresponding functional relation between the light intensity digital signal and the PM2.5 concentration value based on the Fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system.

In this step, it should be noted that before the neural network model is established, performance verification needs to be performed on the neural network model, and further, the effectiveness of the method is strictly described. Therefore, in order to obtain such a complex function mapping relationship through the neural network, a unique PM2.5 concentration value can be obtained by calculating based on the fraunhofer diffraction principle according to the light intensity digital signal collected by the laser detection system, so that the functional correspondence relationship between the light intensity digital signal and the PM2.5 concentration value is verified.

103. And establishing a regularization neural network model by taking the light intensity digital signal as an input quantity, and outputting a PM2.5 concentration value.

In this step, fig. 2 shows a schematic diagram of a neural network structure provided in the first embodiment of the present invention, and as shown in fig. 2, the number of hidden layer nodes is selected as the sample number, and the light intensity digital signal X is processed_iN＝(x_i1,x_i2,...,x_iN) The method is used for solving the problem that the input quantity of the neural network is p, wherein i is 1,2, and N is the dimension of an input node, in the embodiment, N is 70, and each training sample corresponds to one hidden layer node, namely, p hidden layer neurons are taken. A Gaussian function is adopted as an excitation function of a hidden layer and an output layer, the nonlinear mapping capability of the regularized neural network model is embodied, and the expression is as follows:

<math> <mrow> <msub> <mi>G</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>X</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>exp</mi> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <mrow> <mo>|</mo> <mo>|</mo> <mi>X</mi> <mo>-</mo> <msub> <mi>C</mi> <mi>iN</mi> </msub> <mo>|</mo> <mo>|</mo> </mrow> <mrow> <mn>2</mn> <msubsup> <mi>&delta;</mi> <mi>i</mi> <mn>2</mn> </msubsup> </mrow> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> <mi>i</mi> <mo>=</mo> <mn>1,2</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <mi>p</mi> </mrow> </math>

wherein, $\left|\right|X-C_{\mathrm{iN}}\left|\right|=\sqrt{{(x_1-c_{i1})}^2+{(x_2-c_{i2})}^2+...+{(x_N-c_{\mathrm{iN}})}^2}$ is a two-dimensional norm of the number,_ia spreading constant that is a radial odd function;

let the weight matrix be W ═ W₁,w₂,...,w_p)^TAnd if the output value is PM2.5 concentration, each node of the p-dimensional hidden layer outputs, and the expression is as follows:

<math> <mrow> <mi>F</mi> <mrow> <mo>(</mo> <mi>X</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>p</mi> </munderover> <msub> <mi>w</mi> <mi>i</mi> </msub> <msub> <mi>G</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>X</mi> <mo>)</mo> </mrow> </mrow> </math>

it should be noted that, in this embodiment, to simplify the operation, facilitate the transplantation of the single chip microcomputer, and miniaturize the system, each hidden layer basis function center is selected as C_iN＝X_iNThat is, the data center of each basis function corresponds to the corresponding sample itself, and the radial expansion function is taken asWherein d is_maxIs the maximum distance between samples.

Therefore, the invention only needs to train the weight matrix, adjust the weight to fit the mapping relation, determine the parameters of the neural network model, perform online or offline fitting and estimate the system performance.

According to the PM2.5 concentration detection method and device based on the neural network, firstly, a laser detection system is established based on a Fraunhofer diffraction principle, then, according to light intensity digital signals collected by the laser detection system, the corresponding function relationship between the light intensity digital signals and PM2.5 concentration values is verified based on the Fraunhofer diffraction principle, the light intensity digital signals are used as input quantities, a regularized neural network model is established, and the PM2.5 concentration values are output.

Fig. 3 is a schematic flow chart of a PM2.5 concentration detection method based on a neural network according to a second embodiment of the present invention, and as shown in fig. 3, the method of this embodiment is as follows.

301. According to a laser detection system established based on the Fraunhofer diffraction principle, 70 paths of photoelectric signal data are collected.

In this step, specifically, the power supply unit converts alternating current into direct current through the switching power supply, outputs stabilized voltage 3V direct current through the three-terminal voltage stabilization integrated circuit LM317 in cooperation with the potentiometer for use by the He-Ne laser, the He-Ne laser generates red monochromatic laser with a wavelength λ of 0.6328 μm, the laser emitted from the He-Ne laser is processed by the filter lens and the beam expanding lens to form a bundle of parallel monochromatic light as incident light for laser granularity measurement, the air pump is used to sufficiently and uniformly disperse air in the sample cell in the air chamber, fraunhofer diffraction occurs when the incident light passes through the air chamber, the signal receiving unit receives the diffracted light by the fourier lens with a focal length of 180mm and converges the light onto the photodetector, the detector is of a fan-shaped structure and consists of 70 fan-shaped rings corresponding to the same central angle, each ring detects scattered light energy within a corresponding scattering angle range, a small hole is formed in the center of a sector and used for centering a light path, a light intensity analog signal received by a detector is converted into a light intensity digital signal through a detection unit, a calculation unit adopts a calculation singlechip to perform modeling and calculation to obtain air granularity and a PM2.5 concentration value, and as shown in fig. 4, fig. 4 shows a structural schematic diagram of a laser detection system provided by a second embodiment of the invention.

302. According to the Fraunhofer diffraction principle, the corresponding functional relation between the 70 paths of photoelectric signal data and the PM2.5 concentration value is verified.

In this step, according to the Fraunhofer diffraction principle, when light passes through a particle with a diameter D, the distribution of diffraction intensity at any angle is calculated by the following formula,

<math> <mrow> <mi>I</mi> <mrow> <mo>(</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>I</mi> <mn>0</mn> </msub> <mfrac> <mrow> <msup> <mi>&pi;</mi> <mn>2</mn> </msup> <msup> <mi>D</mi> <mn>4</mn> </msup> </mrow> <mrow> <mn>16</mn> <msup> <mi>f</mi> <mn>2</mn> </msup> <msup> <mi>&lambda;</mi> <mn>2</mn> </msup> </mrow> </mfrac> <msup> <mrow> <mo>[</mo> <mfrac> <mrow> <msub> <mrow> <mn>2</mn> <mi>J</mi> </mrow> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>X</mi> <mo>)</mo> </mrow> </mrow> <mi>X</mi> </mfrac> <mo>]</mo> </mrow> <mn>2</mn> </msup> </mrow> </math>

wherein, I (theta) is the diffraction light intensity distribution of the light intensity digital signal under any angle, I₀Is the intensity of the incident light, f is the focal length of the fourier lens,d is the diameter of the air particle, λ is the wavelength of the incident light, J₁(X) is a first order Bessel function, thenThe emitted light is distributed in the n-th detection ring (the ring radius is from S)_nTo S_n+1Corresponding angle from theta_nTo theta_n+1) The light energy above, calculated by the following formula,

<math> <mrow> <msub> <mi>e</mi> <mi>n</mi> </msub> <mo>=</mo> <msubsup> <mo>&Integral;</mo> <msub> <mi>&theta;</mi> <mi>n</mi> </msub> <msub> <mi>&theta;</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> </msubsup> <mi>I</mi> <mrow> <mo>(</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mi>&pi;</mi> <mi>sin</mi> <mrow> <mo>(</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mi>d&theta;</mi> <mo>=</mo> <msubsup> <mo>&Integral;</mo> <msub> <mi>S</mi> <mi>n</mi> </msub> <msub> <mi>S</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> </msubsup> <mi>I</mi> <mrow> <mo>(</mo> <mi>S</mi> <mo>)</mo> </mrow> <mi>&pi;SdS</mi> <mo>,</mo> <mrow> <mo>(</mo> <mi>n</mi> <mo>=</mo> <mn>1,2</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mi>N</mi> <mo>)</mo> </mrow> </mrow> </math>

since theta is very small, thenThe expression is obtained by substituting the expression into the above formula and carrying out recursion on the formula by a Bessel function,

<math> <mrow> <msub> <mi>e</mi> <mi>n</mi> </msub> <mo>=</mo> <mfrac> <msup> <mi>&pi;D</mi> <mn>2</mn> </msup> <mn>8</mn> </mfrac> <msub> <mi>I</mi> <mn>0</mn> </msub> <mo>[</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mi>n</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mi>n</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>]</mo> </mrow> </math>

wherein,J₀representing a zero order Bessel function.

Assuming that the collected air consists of M particles of diameter, let D be_iThe particles of (A) have Q_iThe total diffraction light energy of the particle group in the nth ring is calculated by the following formula,

<math> <mrow> <msub> <mi>e</mi> <mi>n</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>&pi;I</mi> <mn>0</mn> </msub> <mn>8</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <msub> <mi>Q</mi> <mi>i</mi> </msub> <msubsup> <mi>D</mi> <mi>i</mi> <mn>2</mn> </msubsup> <mo>[</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>+</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>]</mo> </mrow> </math>

the usual particle distribution is expressed in parts by weight, the weight as a function of number, calculated by the following formula,

<math> <mrow> <msub> <mi>Q</mi> <mi>i</mi> </msub> <mo>=</mo> <mfrac> <msub> <mrow> <mn>6</mn> <mi>W</mi> </mrow> <mi>i</mi> </msub> <mrow> <mi>&pi;&rho;</mi> <msubsup> <mi>D</mi> <mi>i</mi> <mn>3</mn> </msubsup> </mrow> </mfrac> </mrow> </math>

wherein ρ is the density of the particulate matter, and further, the total diffracted light energy of the particle group in the nth ring is calculated by the following formula,

<math> <mrow> <msub> <mi>e</mi> <mi>n</mi> </msub> <mo>=</mo> <mfrac> <msub> <mrow> <mn>3</mn> <mi>I</mi> </mrow> <mn>0</mn> </msub> <mrow> <mn>4</mn> <mi>&rho;</mi> </mrow> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <mfrac> <msub> <mi>W</mi> <mi>i</mi> </msub> <msub> <mi>D</mi> <mi>i</mi> </msub> </mfrac> <mo>[</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>+</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>]</mo> </mrow> </math>

assuming that the detector consists of N rings, a linear system of equations consisting of N systems of equations can be established

<math> <mfenced open='' close=''> <mtable> <mtr> <mtd> <msub> <mi>e</mi> <mn>1</mn> </msub> <mo>=</mo> <mfrac> <mrow> <mn>3</mn> <msub> <mi>I</mi> <mn>0</mn> </msub> </mrow> <mrow> <mn>4</mn> <mi>&rho;</mi> </mrow> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <mfrac> <msub> <mi>W</mi> <mi>i</mi> </msub> <msub> <mi>D</mi> <mi>i</mi> </msub> </mfrac> <mo>[</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>+</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>]</mo> </mtd> </mtr> <mtr> <mtd> <msub> <mi>e</mi> <mn>2</mn> </msub> <mo>=</mo> <mfrac> <msub> <mrow> <mn>3</mn> <mi>I</mi> </mrow> <mn>0</mn> </msub> <mrow> <mn>4</mn> <mi>&rho;</mi> </mrow> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <mfrac> <msub> <mi>W</mi> <mi>i</mi> </msub> <msub> <mi>D</mi> <mi>i</mi> </msub> </mfrac> <mo>[</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>+</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>3</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>3</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>]</mo> </mtd> </mtr> <mtr> <mtd> <mo>.</mo> <mo>.</mo> <mo>.</mo> </mtd> </mtr> <mtr> <mtd> <msub> <mi>e</mi> <mi>N</mi> </msub> <mo>=</mo> <mfrac> <msub> <mrow> <mn>3</mn> <mi>I</mi> </mrow> <mn>0</mn> </msub> <mrow> <mn>4</mn> <mi>&rho;</mi> </mrow> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <mfrac> <msub> <mi>W</mi> <mi>i</mi> </msub> <msub> <mi>D</mi> <mi>i</mi> </msub> </mfrac> <mo>[</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>N</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>+</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>N</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>0</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>N</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>J</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <msub> <mi>X</mi> <mrow> <mi>i</mi> <mo>,</mo> <mi>N</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mo>]</mo> </mtd> </mtr> </mtable> </mfenced> </math>

The above equation can be written in matrix form, as shown below,

E＝TW

wherein E is the light energy distribution vector measured by the photoelectric detector, $T=\left[\begin{array}{ccc}{t}_{11}& ...& t_{1M}\\ ...& & \\ t_{N1}& ...& t_{\mathrm{NM}}\end{array}\right]$ w is a particle size distribution vector of the particle group to be detected;

wherein, $t_{\mathrm{NM}}=C\frac{1}{D_M}[J_0^2\left(X_{M,N}\right)+J_1^2\left(X_{M,N}\right)-J_0^2\left(X_{M,N-1}\right)-J_1^2\left(X_{M,N+1}\right)]$ is of diameter D_MThe light energy of the nth detection ring of the detector,for a fixed constant, M is the type of particles of different diameter, I₀Is the intensity of incident light, J₀(X_M,N) Is a zero order Bessel function, J₁(X_M,N) Is a first order Bessel function.

According to the above linear equation set, the particle size distribution of the particle group to be measured can be obtained, the particle sizes with the diameter less than 2.5 μm are added to obtain the particle size alpha of PM2.5, and then the concentration of PM2.5 can be calculated by the following formula,

wherein, beta_{Density of air}The value of 1.205kg/m at 20 ℃ under standard atmospheric pressure³And alpha is the particle size distribution with a diameter of less than 2.5 μm.

Therefore, the particle size distribution of the particle group to be detected is obtained according to the light energy distribution vector, the particle size distribution percentages with the diameters smaller than 2.5 mu m are added and multiplied by the air density to obtain a PM2.5 concentration value, and the corresponding function relation between the light intensity digital signal and the PM2.5 concentration value is verified.

303. And establishing a regularization neural network model by taking the light intensity digital signal as an input quantity.

304. Neural network parameters are initialized.

In this step, specifically: selecting the number of hidden layer nodes of the neural network model to be equal to the number of training samples, selecting a Gaussian function as an excitation function, selecting the center of the excitation function as a sample data center, and initializing the weight and excitation function expansion constant of the neural network.

305. And taking the concentration value of the standard PM2.5 as an output quantity, training the weight of the neural network model by adopting a least mean square algorithm, and adjusting the weight.

In this step, it should be noted that the photoelectric data of the laser detection system and the real-time corresponding PM2.5 concentration value need to be collected, data can be collected once every hour continuously, meanwhile, the corresponding real-time PM2.5 data is collected based on the central station of the chinese environment monitoring, 200 groups of data are collected altogether, in order to prevent the contingency and inaccuracy of the detection result, the data is enhanced to be representative, 100 groups of data with a certain interval and containing a large dynamic range are selected as a training set to establish a regularized neural network model, 30 data are selected as a test set from the remaining 100 groups of data, and the established regularized neural network model is verified.

In this embodiment, the weight value can be initialized to an arbitrary value, and training is performed by using the least mean square algorithm learning rule using the sample data and the known standard PM2.5 value, the learning signal expression is as follows,

e_j＝d_j-W^TG_j(x) j＝1,2,...p

the weight vector adjustment expression is as follows,

△W＝ηe_jG_j(x)

the weighted vector adjustment expression is as follows,

W(k+1)＝W(k)+△W

wherein η is a learning rate, η is initialized to any positive number of 0-1, as shown in fig. 5, and fig. 5 shows a schematic diagram of fitting the neural network model provided by the second embodiment of the present invention to 5 sets of data.

In order to guide modeling and calculation in the calculation unit, data needs to be learned and trained as efficiently as possible under the condition of ensuring convergence, so that the performance of the neural network is evaluated by using the iteration number j and the fitting error. The school rate eta is properly increased, the iteration times j can be reduced, the data fitting is accelerated, but the neural network fitting divergence can be caused by overlarge eta, the system is unstable, namely the iteration times represent the calculation complexity of the system, and the fitting error is the allowable deviation amount when each group of data is fitted. The larger the fitting error is, the lower the fitting accuracy is, but the corresponding iteration times are reduced, and the fitting error represents the calculation accuracy of the neural network algorithm. In the present invention, since the PM2.5 data precision given by the official is an integer number of bits, it is sufficient to set the fitting error to 1 or less in the training termination condition.

306. And inputting the acquired light intensity digital signal, and obtaining and outputting a PM2.5 concentration value according to the neural network model.

The embodiment is based on the Fraunhofer diffraction principle, establishes a laser detection system, verifies that the collected photoelectric data and the PM2.5 concentration really have an actual mapping relation, thereby constructing a regularized neural network calculation model, and acquires the mapping relation to be solved through learning and training the neural network, so that the PM2.5 concentration is efficiently and accurately detected, the whole monitoring system has self-learning capability, and has higher practical value.

Fig. 6 is a schematic structural diagram of a PM2.5 concentration detection apparatus based on a neural network according to a third embodiment of the present invention, and as shown in fig. 6, the PM2.5 concentration detection apparatus based on a neural network in this embodiment includes: the system comprises a laser detection system establishing unit 61, a functional relation verifying unit 62 and a neural network model calculating unit 63.

The laser detection system establishing unit 61 is used for establishing a laser detection system based on the Fraunhofer diffraction principle;

the functional relationship verification unit 62 is configured to verify that a corresponding functional relationship exists between the light intensity digital signal and the PM2.5 concentration value based on the fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system;

the neural network model calculation unit 63 is configured to establish a regularization neural network model using the light intensity digital signal as an input quantity, and output a PM2.5 concentration value.

Wherein, the neural network model calculation unit further comprises:

the parameter initialization module is used for initializing neural network parameters;

and the weight training module is used for training the weight of the neural network model by adopting a least mean square algorithm and adjusting the size of the weight.

Finally, comparing the PM2.5 concentration value in a month in Shenyang city with the standard PM2.5 concentration value of the national environment monitoring official website, the PM2.5 concentration value calculated based on the fraunhofer diffraction model calculation and the PM2.5 concentration value calculated based on the neural network calculation according to the present invention, wherein the comparison result of the PM2.5 concentration values is shown in fig. 7, it can be seen that the deviation between the PM2.5 concentration value obtained by using the neural network algorithm and the result of the national official website is small, which indicates that the method and the apparatus for detecting the PM2.5 concentration based on the neural network according to the present invention are feasible, efficient in calculation and high in detection accuracy.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions and scope of the present invention as defined in the appended claims.

Claims (9)

1. A PM2.5 concentration detection method based on a neural network is characterized by comprising the following steps:

establishing a laser detection system based on the Fraunhofer diffraction principle;

verifying that the light intensity digital signal has a corresponding function relation with a PM2.5 concentration value based on a Fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system;

and establishing a regularization neural network model by taking the light intensity digital signal as an input quantity, and outputting a PM2.5 concentration value.

2. The method of claim 1, wherein the laser detection system comprises: the device comprises a power supply unit, a He-Ne laser, a filter lens, a beam expanding lens, an air pump, an air chamber, a signal receiving unit, a detection unit and a calculation unit;

wherein the power supply unit is used for providing power supply for the He-Ne laser, the air pump, the detection unit and the calculation unit;

the signal receiving unit comprises a Fourier lens and 70 paths of photodetectors and is used for receiving the optical signals diffracted by the air particles and converting the optical signals into light intensity analog signals;

the detection unit is used for converting the light intensity analog signal into a light intensity digital signal after amplification processing;

the calculation unit is used for modeling according to the light intensity digital signal and calculating to obtain a PM2.5 concentration value;

the power supply unit is respectively connected with the He-Ne laser, the air pump, the detection unit and the calculation unit, the He-Ne laser is connected with the filter lens, the air chamber is respectively connected with the beam expanding lens, the air pump and the signal receiving unit, the signal receiving unit is connected with the detection unit, and the detection unit is connected with the calculation unit.

3. The method of claim 1, wherein the verifying the digital light intensity signal according to the collected digital light intensity signal of the laser detection system based on Fraunhofer diffraction principle has a corresponding function relationship with the concentration value of PM2.5 comprises:

and calculating and processing the light energy distribution vector measured by the photoelectric detector to obtain the particle size distribution of the particle group to be measured, adding the particle size distribution percentages with the diameters smaller than 2.5 mu m to obtain a PM2.5 concentration value, and verifying that the light intensity digital signal and the PM2.5 concentration value have a corresponding relation.

4. The method according to claim 3, wherein the particle size distribution of the particle group to be measured is calculated by the following formula,

E＝TW

wherein E is the light energy distribution vector measured by the photoelectric detector, $T=\left[\begin{array}{ccc}{t}_{11}& ...& t_{1M}\\ ...& & \\ t_{N1}& ...& t_{\mathrm{NM}}\end{array}\right]$ w is a particle size distribution vector of the particle group to be detected;

wherein, $t_{\mathrm{NM}}=C\frac{1}{D_M}[J_0^2\left(X_{M,N}\right)+J_1^2\left(X_{M,N}\right)-J_0^2\left(X_{M,N+1}\right)-J_1^2\left(X_{M,N+1}\right)]$ is of diameter D_MThe light energy of the nth detection ring of the detector,for a fixed constant, M is the type of particles of different diameter, I₀Is the intensity of incident light, J₀(X_M,N) Is a zero order Bessel function, J₁(X_M,N) Is a first order Bessel function.

5. The method of claim 1, wherein the PM2.5 concentration value is calculated by the following formula,

wherein, beta_{Density of air}At 20 deg.C under normal atmospheric pressureThe value is 1.205kg/m³And alpha is the particle size distribution with a diameter of less than 2.5 μm.

6. The method according to claim 1, wherein the step of establishing a regularized neural network model by using the light intensity digital signal as an input quantity and outputting a value of the concentration of PM2.5 comprises:

initializing neural network parameters;

and training the weight of the neural network model by adopting a least mean square algorithm, and adjusting the weight.

7. The method according to claim 6, wherein the initializing neural network parameters are specifically:

selecting the number of hidden layer nodes of the neural network model to be equal to the number of training samples, selecting a Gaussian function as an excitation function, selecting the center of the excitation function as a sample data center, and initializing the weight and excitation function expansion constant of the neural network.

8. A PM2.5 concentration detection device based on a neural network is characterized by comprising:

the laser detection system establishing unit is used for establishing a laser detection system based on the Fraunhofer diffraction principle;

the functional relationship verification unit is used for verifying that the light intensity digital signal and the PM2.5 concentration value have a corresponding functional relationship based on the Fraunhofer diffraction principle according to the light intensity digital signal acquired by the laser detection system;

and the neural network model calculation unit is used for establishing a regularization neural network model by taking the light intensity digital signal as an input quantity and outputting a PM2.5 concentration value.

9. The apparatus of claim 8, wherein the neural network model computation unit comprises:

the parameter initialization module is used for initializing neural network parameters;

and the weight training module is used for training the weight of the neural network model by adopting a least mean square algorithm and adjusting the size of the weight.