Particle parameter inversion method and device, electronic equipment and medium

Abstract

The application discloses a kind of fire detection technical field's particulate matter parameter inversion method, device, electronic equipment and medium, the method includes: setting initial weight coefficient for each discrete value in physical property parameter, form four groups respectively corresponding to the weight coefficient vector of physical property parameter;Measurement vector measured by fire smoke detector based on multi-channel spectral analysis is obtained;In each iteration, using compressed sensing method, any one of four groups of weight coefficient vectors is used as target weight coefficient vector, the remaining three groups of weight coefficient vectors are fixed, and the optimization solution of target weight coefficient vector is obtained based on measurement vector, to obtain the optimization result of target weight coefficient vector;The difference between the optimization result of four groups of weight coefficient vectors obtained after iteration and four groups of weight coefficient vectors is calculated, and whether to continue updating iteration to four groups of weight coefficient vectors is judged.Using the method can improve the identification accuracy of fire smoke particles and interference source particles in complex environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire detection, in particular to a particulate matter parameter inversion method and device, electronic equipment and medium. BACKGROUND

[0002] In the field of fire detection, photoelectric smoke fire detectors usually use a single wavelength light source to detect the change in scattering intensity of smoke particles on incident light to determine the occurrence of fire. This detection method based on single wavelength scattering intensity change has high sensitivity and can effectively identify fire smoke in ideal environments, but is easily affected by various interference factors in actual complex environments, especially dust, water mist and other non-fire particles, which can easily cause false alarms and reduce the reliability and practicality of the system.

[0003] In related technologies, some technical solutions use two different wavelength light sources to measure the multi-angle extinction and scattering characteristics of fire smoke through a fire smoke scattering and extinction characteristic measuring device, and then invert the particulate matter parameters. Some technologies use a single laser light source to measure the extinction or scattering response of smoke particles to analyze smoke characteristics and measure the extinction or scattering of fire smoke. Although these technical solutions have high discrimination ability in laboratory conditions, they often rely on large optical platforms and complex modulation systems, resulting in large overall device size and high cost, which is not suitable for widespread deployment in engineering practice.

[0004] To address the above problems, related technologies attempt to introduce miniaturized or low-cost devices to meet actual application needs, but such solutions still have significant limitations in the accuracy and discrimination ability of particulate matter parameter inversion. How to ensure the miniaturization and low cost of the device while obtaining sufficient dimensional optical signals to improve the accuracy of particulate matter parameter inversion remains a key technical challenge in the field of fire detection. SUMMARY

[0005] The present application aims to at least partially address one of the technical problems in the related art. To this end, the purpose of the present application is to provide a particulate matter parameter inversion method, device, electronic equipment and medium to improve the accuracy of identifying fire smoke particles and interference source particles in complex environments.

[0006] To achieve the above purpose, the first aspect of the present application provides a particulate matter parameter inversion method, comprising:

[0007] Discrete value ranges of physical parameters of particulate matters are constructed, and an initial weight coefficient is set for each discrete value of the physical parameters, forming four weight coefficient vectors corresponding to the physical parameters respectively; the physical parameters include a particle size logarithmic distribution average value, a particle size logarithmic distribution standard deviation, a refractive index real part, and a refractive index imaginary part; the weight coefficient vectors represent distribution of different discrete values in the corresponding physical parameters;

[0008] A measurement vector measured by a fire smoke detector based on multi-channel spectral analysis is acquired;

[0009] In each iteration, a compressed sensing method is used, any one of the four weight coefficient vectors is taken as a target weight coefficient vector, the other three weight coefficient vectors are fixed, and the target weight coefficient vector is optimized and solved based on the measurement vector, to obtain an optimization result of the target weight coefficient vector; in each iteration, the four weight coefficient vectors are sequentially optimized and solved respectively;

[0010] Differences between the optimization results of the four weight coefficient vectors obtained after iteration and the four weight coefficient vectors are calculated, and whether to continue updating iteration of the four weight coefficient vectors is determined according to the difference results.

[0011] In addition, the particulate matter parameter inversion method of the above-mentioned embodiments of the present application can also have the following additional technical features:

[0012] According to an embodiment of the present application, whether to continue updating iteration of the four weight coefficient vectors according to the difference results includes:

[0013] If the difference result is greater than a preset threshold, the four weight coefficient vectors are updated based on the optimization result, and the next iteration is performed on the updated four weight coefficient vectors.

[0014] According to an embodiment of the present application, whether to continue updating iteration of the four weight coefficient vectors according to the difference results includes:

[0015] If the difference result is less than or equal to a preset threshold, the optimization result of the four weight coefficient vectors obtained in the current iteration is taken as a final inversion result.

[0016] According to an embodiment of the present application, the differences between the optimization results of the four weight coefficient vectors obtained after iteration and the four weight coefficient vectors include

[0017] The relative errors between the optimization results of the four weight coefficient vectors and the corresponding weight coefficient vectors are calculated respectively, and the average value of the four relative errors is calculated.

[0018] According to one embodiment of the present application, the compressed sensing method is used to take any one of the four groups of weight coefficient vectors as a target weight coefficient vector, fix the other three groups of weight coefficient vectors, and solve the target weight coefficient vector based on the measurement vector to obtain an optimization result of the target weight coefficient vector, including:

[0019] In the case that the other three groups of weight coefficient vectors remain unchanged, an association matrix is constructed according to the Mie scattering theory, which associates the measurement vector with the target weight coefficient vector;

[0020] The optimization result of the target weight coefficient vector is solved based on a non-negative sparse regression method, so that the product of the optimization result of the target weight coefficient vector and the association matrix is closest to the measurement vector.

[0021] According to one embodiment of the present application, the measurement vector includes scattering signals of 8 central wavelength channels, and the central wavelengths corresponding to the central wavelength channels include 415 nm, 445 nm, 480 nm, 515 nm, 555 nm, 590 nm, 630 nm and 680 nm.

[0022] According to one embodiment of the present application, the initial weight values are all set to , where n is the number of discretized values of the physical property parameters.

[0023] To achieve the above purpose, a particle parameter inversion device is provided in a second embodiment of the present application, which includes:

[0024] A construction module is configured to construct a discretized value range of a physical property parameter of a particle, set an initial weight coefficient for each discrete value of the physical property parameter, and form four groups of weight coefficient vectors corresponding to the physical property parameter respectively; the physical property parameter includes a particle size logarithmic distribution average value, a particle size logarithmic distribution standard deviation, a real part of a refractive index and an imaginary part of a refractive index; the weight coefficient vector represents the distribution of different discrete values in the corresponding physical property parameter;

[0025] An acquisition module is configured to acquire a measurement vector measured by a fire smoke detector based on multi-channel spectral analysis;

[0026] An inversion module is configured to, in each iteration, take any one of the four groups of weight coefficient vectors as a target weight coefficient vector, fix the other three groups of weight coefficient vectors, and solve the target weight coefficient vector based on the measurement vector to obtain an optimization result of the target weight coefficient vector; in each iteration, the four groups of weight coefficient vectors are sequentially subjected to optimization and solving respectively.

[0027] The iteration judgment module is configured to calculate the difference between the optimization result of the four sets of weight coefficient vectors obtained after iteration and the four sets of weight coefficient vectors, and judge whether to continue updating the four sets of weight coefficient vectors according to the difference result.

[0028] To achieve the above object, the third aspect of the present application provides an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above particle parameter inversion method when executing the computer program.

[0029] To achieve the above object, the fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the above particle parameter inversion method when being executed.

[0030] The particle parameter inversion method, device, electronic device and medium of the embodiments of the present application can effectively represent the multi-dimensional physical properties of particles in a limited parameter space with low computational complexity by constructing the discretized value range of particle parameters and forming four sets of weight coefficient vectors corresponding to the logarithmic distribution average value of particle size, the logarithmic distribution standard deviation, the real part of refractive index and the imaginary part of refractive index based on the initial weight coefficients. On this basis, combined with the multi-channel spectral detection results, the four sets of weight coefficient vectors are respectively optimized in the coordinate descent manner by using the compressed sensing method, the non-target parameters are fixed in each iteration, and only the target parameters are solved, which can effectively avoid the instability of high-dimensional parameter space. At the same time, by judging the difference between the optimization result and the original weight coefficient vector, it is dynamically controlled whether to perform the next iteration, thereby improving the convergence efficiency and stability of the parameter inversion process. Based on this, the accurate inversion of the multi-physical parameters of particles in fire smoke can be realized while ensuring the calculation efficiency, and then a reliable basis for effectively distinguishing fire smoke from interference sources is provided. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 A flowchart of the particle parameter inversion method in an embodiment;

[0032] Figure 2 A flowchart of the optimization processing of the target weight coefficient vector in an embodiment;

[0033] Figure 3 A flowchart of the particle parameter inversion of fire smoke particles and interference sources based on multi-channel spectral analysis in an embodiment;

[0034] Figure 4 A structural block diagram of the particle parameter inversion device in an embodiment. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0036] The implementation details of the technical solutions of the embodiments of the present application are described in detail below.

[0037] In one embodiment, as shown in Figure 1 , a flowchart of a particle parameter inversion method is provided, which can include the following steps:

[0038] Step S101, construct the discretized value range of the physical parameters of the particles, and set an initial weight coefficient for each discrete value of the physical parameters to form four weight coefficient vectors corresponding to the physical parameters respectively.

[0039] For the physical and optical properties of the particles, the discretized value range of four physical parameters of the particles is constructed. The four physical parameters include the logarithmic distribution average of particle size, the logarithmic distribution standard deviation of particle size, the real part of refractive index, and the imaginary part of refractive index. Among them, the logarithmic distribution average of particle size represents the mean value of particle size in logarithmic scale, reflecting the size of typical particles; the logarithmic distribution standard deviation of particle size represents the dispersion degree of particle size in logarithmic scale, and the larger the value is, the wider the distribution is; the real part of refractive index represents the refractive property of particles in the visible light band, describing the refractive ability of particles to light; the imaginary part of refractive index represents the absorption property of particles in the visible light band, describing the absorption ability of particles to light.

[0040] For each physical parameter, a physically reasonable value range is preset, and the continuous values in the range are discretized into a limited number of value points. For example, the logarithmic distribution average of particle size is divided into to , a total of n discrete values; the logarithmic distribution standard deviation of particle size is divided into to ; the real part of refractive index is divided into to ; and the imaginary part of refractive index is divided into to .

[0041] An initial weight coefficient is assigned to each discrete value, which is used to represent the estimation of the proportion of the current value in the actual physical parameter. Since no measurement data has been obtained at the initial stage, the initial weight coefficients of all discrete values are preset values, which will be iteratively optimized by measurement data in the subsequent.

[0042] The weight coefficients corresponding to the n discrete values of each group of physical property parameters are sequentially arranged to form a weight coefficient vector, respectively. It should be noted that the weight coefficient vector here is an initial weight coefficient vector. Among them, the weight coefficient vector of the logarithmic distribution average of particle size is denoted as , , which corresponds to the logarithmic distribution average of particle size to , respectively. The weight coefficient vector of the standard deviation of the logarithmic distribution of particle size is denoted as , , which corresponds to the standard deviation of the logarithmic distribution of particle size to ; the weight coefficient vector of the real part of the refractive index is denoted as , , which corresponds to the real part of the refractive index to ; the weight coefficient vector of the imaginary part of the refractive index is denoted as , , which corresponds to the imaginary part of the refractive index to .

[0043] It should be noted that the logarithmic distribution average of particle size, the standard deviation of the logarithmic distribution of particle size, the real part of the refractive index, and the imaginary part of the refractive index are taken as the physical property parameters of the particulate matter, which helps to deeply analyze the composition of particulate matter in the environment from the perspective of the physical properties of particulate matter, and then used as identification variables to distinguish fire smoke from non-fire interference sources.

[0044] In one embodiment, for the four groups of physical property parameter discrete value sets constructed, it is assumed that the number of discrete values of each parameter is n, i.e. each group of parameters has n different possible values. Before obtaining specific measurement data, to avoid introducing prior bias, the initial weight coefficient corresponding to each discrete value is set to the same value . For example, for the n discrete values of the logarithmic distribution average of particle size to , the weight coefficient vector is constructed, where each takes , and the weight coefficient vectors corresponding to other physical property parameters are sequentially set.

[0045] Based on this, the setting of the initial weight coefficient follows the principle of uniform distribution, which can make each parameter dimension remain neutral in the initial stage of the inversion algorithm, avoiding that some discrete values occupy an unreasonable proportion before optimization.

[0046] In step S102, a measurement vector measured by a fire smoke detector based on multi-channel spectral analysis is obtained.

[0047] The fire smoke detector comprises a plurality of wavelength channels, each corresponding to a different incident light wavelength, and can monitor the scattering or transmission characteristics of particulate matter at different wavelengths. The detector is deployed in an environment to be monitored to sense the effect of fire smoke particles in the air on incident light in real time and record the response values of each channel.

[0048] In practical applications, when fire smoke particles appear in the environment, particles of different sizes, refractive indices and concentration distributions have different degrees of scattering and absorption effects on the light signals of each wavelength channel, thereby exhibiting different light intensity attenuation characteristics on each channel. By reading the measurement values of all channels of the detector, a set of original spectral measurement vectors are obtained, where represents the output signal intensity of the i-th wavelength channel.

[0049] In one embodiment, the fire smoke detector based on multi-channel spectral analysis comprises a plurality of preset central wavelength channels for simultaneously measuring the optical scattering characteristics of particulate matter in the air at different wavelength ranges. The plurality of preset central wavelength channels can be 8 independent optical channels corresponding to the following 8 central wavelengths: 415 nm, 445 nm, 480 nm, 515 nm, 555 nm, 590 nm, 630 nm and 680 nm. These wavelengths cover different color light segments in the visible light range, which helps to more comprehensively characterize the spectral response characteristics of particulate matter. Based on this, the measurement vector measured by the fire smoke detector based on multi-channel spectral analysis includes the scattering signals of the 8 central wavelength channels.

[0050] In step S103, in each iteration, the compressed sensing method is used to take any one of the four sets of weight coefficient vectors as the target weight coefficient vector, fix the remaining three sets of weight coefficient vectors, and optimize and solve the target weight coefficient vector based on the measurement vector to obtain the optimization result of the target weight coefficient vector.

[0051] In order to improve the accuracy and stability of the particulate matter parameter inversion, an alternating optimization strategy based on compressed sensing theory is used to iteratively solve the four sets of weight coefficient vectors involved in the multi-channel scattering model.

[0052] The compressed sensing method is a mathematical technique for reconstructing sparse signals under under-sampling conditions. Let the measurement vector be , which is composed of the scattering intensities of the plurality of central wavelength channels obtained by the fire smoke detector based on multi-channel spectral analysis. It is assumed that the inversion model includes four response characteristic subspaces corresponding to the four sets of weight coefficient vectors to be optimized, denoted as , , and Each weight coefficient vector reflects the contribution weight of the particulate matter to the scattering signal of each wavelength channel under a certain type of feature response mode.

[0053] In each iteration process, one of the weight coefficient vectors is selected as the target weight coefficient vector of the current round by using alternating optimization, and the remaining three weight coefficient vectors are kept unchanged. Based on the measurement vector and the fixed three weight coefficient vectors, the target weight coefficient vector is optimized and solved by using common compressed sensing reconstruction algorithms such as Orthogonal Matching Pursuit (OMP) and Basis Pursuit (BP).

[0054] After the optimization of the weight coefficients of the current group is completed, the next group of weight coefficient vectors is selected as the optimization target, and the above process is repeated. When the optimization structure of the four weight coefficient vectors is obtained, it indicates that one iteration is completed.

[0055] For example, the weight coefficient vector corresponding to the average value of the logarithmic distribution of the particle size is selected as the target weight coefficient vector, and the other weight coefficient vectors 、 and are fixed. After data processing, the optimization result of the weight coefficient vector is calculated . Similarly, the other weight coefficient vectors are processed accordingly, and the optimization results of the other weight coefficient vectors 、 and are finally calculated.

[0056] It should be noted that the iterative processing can reduce the dimension of the compressed data while retaining the key feature information of the measurement signal, improving the inversion efficiency and accuracy, and realizing effective conversion from multi-dimensional measurement information to high-dimensional physical parameters, breaking through the limitations in complex particulate matter identification.

[0057] In one embodiment, Figure 2 a schematic diagram of the optimization processing flow of the target weight coefficient vector is shown, which can include the following steps:

[0058] Step S201, under the condition that the remaining three weight coefficient vectors are kept unchanged, an association matrix relating the target weight coefficient vector to the measurement vector is constructed according to the Mie scattering theory.

[0059] Optionally, one of the four sets of weight coefficient vectors is taken as the target weight coefficient vector, and the other three sets of weight coefficient vectors are fixed. Based on the Mie scattering theory, the mathematical mapping relationship between the scattering response of each channel and the contribution of different physical parameters is established by considering the scattering characteristics of light of different wavelengths on different particle size components, thereby obtaining the correlation matrix that links the target weight coefficient vector and the measurement vector.

[0060] In step S202, the optimization result of the target weight coefficient vector is solved based on the non-negative sparse regression method, so that the product of the optimization result of the target weight coefficient vector and the correlation matrix is closest to the measurement vector.

[0061] Based on the constructed correlation matrix, the non-negative sparse regression method (such as non-negative least angle regression NN-LARS or non-negative sparse Bayesian learning) is used to optimize and solve the target weight coefficient vector. The goal of this optimization process is to minimize the reconstruction error between the target weight coefficient vector and the measurement vector, while satisfying the non-negative constraint condition to ensure that the weight coefficient conforms to the physical meaning of the scattering intensity. The final optimization result of the target weight coefficient vector is the optimal solution that makes the linear combination closest to the current measurement vector under the condition that the current three sets of coefficients are fixed.

[0062] For example, the target weight coefficient vector is the weight coefficient vector corresponding to the logarithmic distribution average of the particle size The measurement vector is denoted as The optimization result can be represented by the following linear model:

[0063]

[0064] wherein, is the correlation matrix that links the target weight coefficient vector and the measurement vector , and each element of the correlation matrix is composed of the unit response value of each particle size particle at different wavelengths calculated based on the Mie scattering theory; is the target weight coefficient vector after optimization processing.

[0065] In step S104, the difference between the optimization result of the four sets of weight coefficient vectors obtained after iteration and the four sets of weight coefficient vectors is calculated, and whether to continue updating and iterating the four sets of weight coefficient vectors is determined according to the difference result.

[0066] After each round of iteration is completed, the difference between the optimization result of the four sets of weight coefficient vectors obtained in the current round of iteration and the corresponding four sets of weight coefficient vectors in the previous round is calculated. The difference result is used to measure the change amplitude of the weight coefficient in the continuous iteration process, so as to determine whether the optimization process has tended to be stable.

[0067] Based on the difference result, it is judged whether the four groups of weight coefficient vectors need to be continuously updated and iterated, that is, whether the current four groups of weight coefficient vectors have reached a convergence state, and if not, the updating and iteration is continuously performed.

[0068] It should be noted that the weight coefficient vector reflects the contribution degree of different particle parameters (including particle size, refractive index, etc.) to the scattering measurement vector, and the optimization result actually corresponds to the fitting degree of each particle parameter combination under the current measurement condition. By continuously adjusting these weight coefficients through optimization iteration, the scattering response of the weight coefficient combination is closest to the actual measurement data. The finally converged weight coefficient vector indicates which combination of discretized parameter values can most accurately reflect the physical characteristics of the real particulate matter under the current environmental conditions. Therefore, the convergence of the weight coefficient not only characterizes whether the model inversion process is stable, but also indirectly maps the composition of the actual particulate matter in the environment, enabling the identification and differentiation of different particulate compositions, thereby enhancing the reliability and anti-interference ability of fire detection.

[0069] In one embodiment, to achieve effective optimization and convergence control of the four groups of weight coefficient vectors, a preset threshold for breaking out of the loop is set here In practical applications, the threshold Generally, 0.01 can be selected.

[0070] After each iteration is completed, if the calculated difference result is greater than the preset threshold , it indicates that the update of the current four groups of weight coefficient vectors has not converged, that is, the change amplitude between the vectors is still large, and the optimization process needs to be continuously performed. Therefore, based on the optimization result of the current round, the respective corresponding weight coefficient vectors are updated, that is, the optimization result replaces the weight coefficient vectors of the previous round, and the next iteration operation is entered based on the updated four groups of weight coefficient vectors.

[0071] In one embodiment, if the difference result is less than or equal to the preset threshold, it indicates that the change amplitude of the four groups of weight coefficient vectors has been in a stable interval in consecutive rounds, and the optimization process has reached the expected convergence standard, at which time the iteration process is terminated, and the optimization result of the four groups of weight coefficient vectors obtained in the current round is output as the final particulate matter parameter inversion result.

[0072] In practical applications, the inversion result, that is, the final output weight coefficient vector, is subjected to weighted summation or statistical feature extraction, etc., which can directly calculate the estimated values of parameters such as the mean of the logarithmic distribution of the particle size, the standard deviation of the logarithmic distribution, the real part and the imaginary part of the refractive index, etc., thereby realizing effective mapping and output of the actual particulate matter parameters from the weight coefficients.

[0073] In one embodiment, in order to accurately evaluate the convergence degree of the four groups of weight coefficient vectors in the iteration process, the relative error is used as a measurement standard when calculating the difference between the optimization results of the four groups of weight coefficient vectors and the current weight coefficient vectors.

[0074] Firstly, the relative error between the optimization result of each group of weight coefficient vectors obtained in the current iteration and the corresponding weight coefficient vector in the last iteration is calculated. Taking the weight coefficient vector and the optimization result of the weight coefficient vector as an example, the calculation formula of the relative error of this group of weight coefficient vectors is:

[0075]

[0076] wherein, ||.|| represents the norm of a vector, which can be the two-norm. In this way, the relative errors of the remaining three groups are calculated respectively, and finally the four groups of relative errors are obtained. The calculation results of the four groups of relative errors are averaged to obtain the difference result for iteration control.

[0077] In practical applications, Figure 3 Fig. 1 shows a flowchart of the parameter inversion of fire smoke particles and interference sources based on multi-channel spectral analysis.

[0078] Step 1: For the physical properties of fire smoke, the discrete value range of the physical parameters of particulate matter is constructed. The physical parameters include the logarithmic distribution average of particle size, the logarithmic distribution standard deviation of particle size, the real part of refractive index, and the imaginary part of refractive index.

[0079] Step 2: Set an initial weight coefficient for each discrete value in the physical parameters to form four groups of weight coefficient vectors of the physical parameters, denoted as , , and . In practical applications, the initial weight coefficient can be taken as , wherein, n represents the number of discrete values in a group of physical parameters.

[0080] Step 3: In order to realize the convergence criterion, a set threshold is set.

[0081] Step 4: Obtain the measurement vector by measuring the fire smoke detector.

[0082] Step 5: Use the compressed sensing method to optimize , , and based on the measurement vector, and obtain ,​​​ , and . Wherein, the optimization process herein can refer to the content of step S103 and embodiments related to step S103.

[0083] Step 6, after a round of iteration, calculate the difference results between , , and and , , and .

[0084] Step 7, judge whether the difference results are greater than a preset threshold .

[0085] Step 8, if the difference results are greater than the preset threshold , update , , and based on , , and , and return to step 5 for the next round of iterative update.

[0086] Step 9, if the difference results are less than or equal to the preset threshold , output , , and as the parameters of particulate matter.

[0087] In the above embodiments, based on the difference in physical nature between fire smoke and interference sources, the average value of the logarithmic distribution of particle size, the standard deviation of the logarithmic distribution, the real part of the refractive index, and the imaginary part of the refractive index are selected as the inversion target. Directly from the physical and optical characteristics of particulate matter, the recognition and judgment are made, which significantly improves the recognition accuracy of fire and non-fire interference sources. By constructing the discrete value space of the physical parameters and introducing the corresponding weight coefficient vector, the high-dimensional parameter inversion problem is converted into an optimization problem of sparse coefficients, and further using the advantage of compressed sensing method in processing high-dimensional sparse signals, an efficient cyclic solution algorithm is constructed. With the support of multi-channel spectral measurement data, accurate inversion of multi-parameter, multi-dimensional physical information can be efficiently completed. At the same time, by judging the difference between the optimization results and the original weight coefficient vector, it is dynamically controlled whether to perform the next round of iteration, thereby improving the convergence efficiency and stability of the parameter inversion process. Based on the above design, not only the accuracy and efficiency of the particulate matter parameter inversion are improved, but also the intelligent discrimination of fire smoke and interference sources based on physical parameters is realized.

[0088] In one embodiment, a particulate matter parameter inversion device is provided, referring to Figure 4 As shown in the figure, the particulate matter parameter inversion device 300 can include a construction module 301, an acquisition module 302, an inversion module 303 and an iteration judgment module 304. Among them, the construction module 301 is used to construct the discretization value range of the physical property parameters of the particulate matter, and set an initial weight coefficient for each discrete value in the physical property parameters to form four groups of weight coefficient vectors corresponding to the physical property parameters respectively; the physical property parameters include the logarithmic distribution average of particle size, the logarithmic distribution standard deviation of particle size, the real part of refractive index and the imaginary part of refractive index; the weight coefficient vector represents the distribution of different discrete values in the corresponding physical property parameters;

[0089] The acquisition module 302 is used to acquire the measurement vector measured by the fire smoke detector based on multi-channel spectral analysis;

[0090] The inversion module 303 is used to, in each iteration, adopt the compressed sensing method, take any one of the four groups of weight coefficient vectors as the target weight coefficient vector, fix the remaining three groups of weight coefficient vectors, and optimize and solve the target weight coefficient vector based on the measurement vector to obtain the optimization result of the target weight coefficient vector; wherein the optimization and solving of the four groups of weight coefficient vectors are performed in turn in each iteration;

[0091] The iteration judgment module 304 is used to calculate the difference between the optimization result of the four groups of weight coefficient vectors obtained after iteration and the four groups of weight coefficient vectors, and judge whether to continue updating iteration of the four groups of weight coefficient vectors according to the difference result.

[0092] In one embodiment, the iteration judgment module 304 is specifically used to, if the difference result is greater than a preset threshold, update the four groups of weight coefficient vectors based on the optimization result, and perform the next round of iteration on the updated four groups of weight coefficient vectors.

[0093] In one embodiment, the iteration judgment module 304 is specifically used to, if the difference result is less than or equal to the preset threshold, take the optimization result of the four groups of weight coefficient vectors obtained in the current iteration as the final inversion result.

[0094] In one embodiment, the iteration judgment module 304 is specifically used to calculate the relative error between the optimization result of the four groups of weight coefficient vectors and the corresponding weight coefficient vector respectively, and calculate the average value of the four groups of relative errors.

[0095] In one embodiment, the inversion module 303 is specifically used to, under the condition that the remaining three groups of weight coefficient vectors remain unchanged, construct the correlation matrix that links the measurement vector and the target weight coefficient vector according to the Mie scattering theory;

[0096] The optimization result of the target weight coefficient vector is solved based on a non-negative sparse regression method, so that the product of the optimization result of the target weight coefficient vector and the correlation matrix is closest to the measurement vector.

[0097] In one embodiment, the fire smoke detector based on multi-channel spectral analysis acquires scattering signals of 8 central wavelength channels, and the central wavelengths corresponding to the central wavelength channels include 415 nm, 445 nm, 480 nm, 515 nm, 555 nm, 590 nm, 630 nm and 680 nm.

[0098] In one embodiment, the initial weight values are all set to Wherein n is the number of discretized values in the physical property parameters.

[0099] The specific limitations of the particulate matter parameter inversion device 300 can refer to the limitations of the particulate matter parameter inversion method described above, and will not be repeated here. Each module in the above particulate matter parameter inversion device 300 can be realized by software, hardware and combinations thereof, in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above modules.

[0100] In one embodiment, an electronic device is provided, including a memory and a processor, the memory stores a computer program, and the processor implements a particulate matter parameter inversion method when executing the computer program.

[0101] In one embodiment, a computer storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement a particulate matter parameter inversion method.

[0102] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0103] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise explicitly and specifically limited.

[0104] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and that variations, modifications, substitutions and changes can be made by those skilled in the art without departing from the scope of the present application.

Claims

1. A particle parameter inversion method, characterized in that: include: Constructing a discrete range of values ​​for the physical property parameters of the particles and setting an initial weight coefficient for each discrete value in the physical property parameters to form four sets of weight coefficient vectors corresponding to the physical property parameters; the physical property parameters include the mean value of the logarithmic distribution of the particle size, the standard deviation of the logarithmic distribution of the particle size, the real part of the refractive index, and the imaginary part of the refractive index; the weight coefficient vectors represent the distribution of different discrete values ​​in the corresponding physical property parameters; Obtaining a measurement vector measured by a fire smoke detector based on multi-channel spectral analysis; In each round of iteration, a compressed sensing method is used, any one of the four groups of weight coefficient vectors is used as the target weight coefficient vector, the remaining three groups of weight coefficient vectors are fixed, and the target weight coefficient vector is optimized and solved based on the measurement vector to obtain an optimization result of the target weight coefficient vector, including: constructing a correlation matrix that links the measurement vector and the target weight coefficient vector according to Mie scattering theory when the remaining three groups of weight coefficient vectors remain unchanged; solving the optimization result of the target weight coefficient vector based on a non-negative sparse regression method so that the product of the optimization result of the target weight coefficient vector and the correlation matrix is ​​closest to the measurement vector; wherein, in each round of iteration, the optimization solution is performed on each of the four groups of weight coefficient vectors in turn; After the iterative calculation, the optimization results of the four sets of weight coefficient vectors and the differences between the four sets of weight coefficient vectors are obtained, and it is determined whether to continue updating and iterating the four sets of weight coefficient vectors based on the difference results.

2. The particle parameter inversion method according to claim 1, characterized in that: The step of determining whether to continue iterating the four sets of weight coefficient vectors based on the difference results includes: If the difference result is greater than a preset threshold, the four groups of weight coefficient vectors are updated based on the optimization result, and the next round of iteration is performed on the updated four groups of weight coefficient vectors.

3. The particle parameter inversion method according to claim 1, characterized in that: The step of determining whether to continue iterating the four sets of weight coefficient vectors based on the difference results includes: If the difference result is less than or equal to the preset threshold, the optimization results of the four groups of weight coefficient vectors obtained in the current round of iteration are used as the final inversion results.

4. The particle parameter inversion method according to claim 1, characterized in that: The optimization results of the four sets of weight coefficient vectors obtained after the iterative calculation and the differences between the four sets of weight coefficient vectors include: The relative errors between the optimization results of the four groups of weight coefficient vectors and the corresponding weight coefficient vectors are calculated respectively, and the average value of the four groups of relative errors is calculated.

5. The particle parameter inversion method according to claim 1, characterized in that: The measurement vector includes scattered signals of 8 central wavelength channels, and the central wavelengths corresponding to the central wavelength channels include 415 nm, 445 nm, 480 nm, 515 nm, 555 nm, 590 nm, 630 nm and 680 nm.

6. The particle parameter inversion method according to claim 1, characterized in that: The initial weight values ​​are all set to, where n is the number of discretized values ​​in the physical property parameters.

7. A particle parameter inversion device, characterized in that: The particle parameter inversion method according to any one of claims 1 to 6, wherein the device comprises: A construction module is used to construct a discrete value range of the physical property parameters of the particulate matter and set an initial weight coefficient for each discrete value of the physical property parameters to form four sets of weight coefficient vectors corresponding to the physical property parameters; the physical property parameters include the mean value of the logarithmic distribution of the particle size, the standard deviation of the logarithmic distribution of the particle size, the real part of the refractive index, and the imaginary part of the refractive index; the weight coefficient vectors represent the distribution of different discrete values ​​of the corresponding physical property parameters; an acquisition module, for acquiring a measurement vector measured by a fire smoke detector based on multi-channel spectral analysis; an inversion module, configured to, in each round of iteration, employ a compressed sensing method, use any one of the four groups of weight coefficient vectors as a target weight coefficient vector, fix the remaining three groups of weight coefficient vectors, and optimize and solve the target weight coefficient vector based on the measurement vector to obtain an optimized result of the target weight coefficient vector; wherein, in each round of iteration, the optimization and solving are performed on each of the four groups of weight coefficient vectors in sequence; The iterative judgment module is used to calculate the difference between the optimization results of the four sets of weight coefficient vectors obtained after iteration and the four sets of weight coefficient vectors, and judge whether to continue updating and iterating the four sets of weight coefficient vectors based on the difference results.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the particle parameter inversion method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the particle parameter inversion method according to any one of claims 1 to 6 is implemented.

Images