A dust concentration monitoring device and method based on physical and data dual drive

Abstract

A dust concentration monitoring device and method based on physical and data dual drive, equipment: the oscillation measurement unit includes a detachable collision separator, an oscillation measurement mechanism, a mass flow controller and a flow pump; the oscillation measurement mechanism is mainly composed of an oscillation tube, a filter membrane holder, a magnetic steel, an electromagnet, an oscillation drive module and a high-precision frequency measurement module; the particle measurement unit includes a particle flight chamber, a high-efficiency filter, a flow pump, a current limiting component, a particle measurement chamber, a light trap, a laser, a photoelectric sensing unit, a temperature and humidity sensor and a mass flow controller; two lasers, two light traps, two filters and two photoelectric sensing units are arranged in the particle measurement chamber. Method: The parameter physical model is combined with the machine learning model, and the outputs of the two are comprehensively calculated through the neural network fusion layer, and the final predicted value of the dust concentration is finally output. The device and method have a high degree of automation, and can realize high-precision monitoring of dust concentration in the environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of dust concentration detection, and in particular relates to a dust concentration monitoring device and method based on physical and data dual drive. Background Art

[0002] Dust monitoring is of great significance in the fields of environmental protection and industrial hygiene. Dust particles in the air not only have an adverse impact on environmental quality, but also pose a serious threat to human health, especially inhalable particles (such as PM2.5 and PM10), whose high concentration exposure is closely related to health problems such as respiratory diseases, cardiovascular diseases and lung cancer. Therefore, accurate and real-time monitoring of dust concentration is crucial to protecting public health and environmental safety.

[0003] At present, dust monitoring mainly relies on two technical means: one is the traditional method based on physical principles, and the other is the data-driven machine learning method. Traditional dust monitoring methods include gravity method, beta ray method and light scattering method. The gravity method determines the dust concentration by collecting particles in the air and measuring their mass. It has high accuracy and reliability, but lacks real-time performance. The beta ray method estimates the concentration by measuring the absorption of beta rays by particles. Although it has high accuracy, the instrument cost is high and the maintenance process is complicated. The light scattering method uses the scattering characteristics of particles to measure light and can provide real-time data, but it is easily affected by environmental conditions and changes in particle characteristics, resulting in inaccurate measurement results. In order to overcome the shortcomings of these traditional methods, in recent years, data-driven machine learning methods have gradually been applied to the field of dust monitoring. These methods use a large amount of historical monitoring data to predict dust concentration by training models, and have good adaptability and flexibility. However, a single machine learning model may have insufficient generalization ability when facing different environments and complex practical application scenarios, resulting in unstable prediction accuracy.

[0004] Therefore, it is urgent to provide a hybrid model that combines physical models and machine learning models to fully improve the prediction accuracy of dust. Summary of the invention

[0005] In view of the problems existing in the above-mentioned prior art, the present invention provides a dust concentration monitoring device and method based on dual physical and data drive. The device has good reliability, strong versatility, and a high degree of automation, and can achieve high-precision monitoring of dust concentration in the environment; the method has simple steps and a high degree of automation, and can automatically monitor the dust concentration in the environment and effectively ensure the prediction accuracy.

[0006] In order to achieve the above-mentioned object, the present invention provides a dust concentration monitoring device based on physical and data dual drive, including a shell, an oscillation measurement unit, a particle measurement unit and a data processing unit, wherein the shell is respectively provided with a first air inlet, a second air inlet, a third air inlet, a fourth air inlet, a first air outlet and a second air outlet which are connected to its inner cavity;

[0007] The oscillation measurement unit includes a detachable collision separator, an oscillation measurement mechanism, a second mass flow controller and a fourth flow pump; the detachable collision separator is installed at the fourth air inlet; the oscillation measurement mechanism is mainly composed of an oscillation tube, a filter membrane holder, a magnetic steel, an electromagnet, an oscillation drive module and a high-precision frequency measurement module; the oscillation tube is a frustum-shaped quartz oscillation tube, whose cross-section is elliptical, with the small diameter end at its upper end serving as an air inlet, and the large diameter end at its lower end serving as an air outlet, and connected to the second air outlet through a connecting pipe C; the filter membrane holder is a thin-walled frustum structure, which is located above the oscillation tube, the large diameter end at the upper end of the filter membrane holder is installed with a filter membrane, and is connected to the fourth air inlet through a connecting pipe D, and the small diameter end at its lower end is connected to the fourth air inlet through a connecting pipe D. The diameter end is connected with the small diameter end of the oscillating tube; a pair of magnets are symmetrically distributed on opposite sides of the outside of the long axis direction of the oscillating tube, and are tightly adhered to the side walls of the upper part of the oscillating tube; a pair of electromagnets are symmetrically supported on the outside of the pair of magnets and are on the same horizontal line with the center of the magnets, and are used to drive the magnets to vibrate; the oscillation drive module is connected to the pair of electromagnets, and is used to output dynamic drive signals to the electromagnets; the high-precision frequency measurement module is installed on the outside of the oscillating tube and is at the same height position as the electromagnets, and is used to sense the change of magnetic field strength to achieve high-precision measurement of the oscillation frequency; the second mass flow controller and the fourth flow pump are connected in series to the connecting pipeline C from the air inlet side to the air outlet side in sequence;

[0008] The particle measurement unit includes a particle flight chamber, a first high efficiency filter, a first flow pump, a first current limiting component, a second high efficiency filter, a second flow pump, a second current limiting component, a particle measurement chamber, a first light trap, a first laser, a second light trap, a second laser, a first photoelectric sensing unit, a first filter, a second photoelectric sensing unit, a second filter, a temperature and humidity sensor, a first mass flow controller and a third flow pump;

[0009] The particle flight chamber is located in the shell, and is divided into an air intake section, a mixing section and an acceleration section from top to bottom. The air intake section is a conical structure, the small diameter end of which is located at the upper part, and a chamber inlet A is opened. The mixing section is a cylindrical structure, and a chamber inlet B and a chamber inlet C are opened on opposite sides of the upper part. The acceleration section is a conical structure, the small diameter end of which is located at the upper part, and a chamber outlet A is opened; the chamber inlet A is connected to the first air inlet through a connecting pipe A, the chamber inlet B is connected to the second air inlet through an air flow branch 1, and the chamber inlet C is connected to the third air inlet through an air flow branch 2; the air outlet ends of the air flow branch 1 and the air flow branch 2 are arranged at a relative angle to the central axis of the particle flight chamber, and are not perpendicular to the particle flight chamber;

[0010] The first high-efficiency filter, the first flow pump, and the first flow limiting component are sequentially connected in series on the air flow branch 1 from the air inlet side to the air outlet side; the second high-efficiency filter, the second flow pump, and the second flow limiting component are sequentially connected in series on the air flow branch 2 from the air inlet side to the air outlet side;

[0011] The particle measurement chamber is located in the shell and below the particle flight chamber. The top and bottom of the particle measurement chamber are relatively provided with a chamber outlet D and a chamber outlet B communicating with the inner cavity thereof, wherein the chamber outlet D is communicated with the chamber outlet A, and the chamber outlet B is connected with the first air outlet through a communication pipe B;

[0012] The first optical trap and the first laser are mounted on the side walls on the left and right sides of the upper part of the inner cavity of the particle measurement chamber in a relative manner;

[0013] The second optical trap and the second laser are mounted on the side walls on the left and right sides of the lower part of the inner cavity of the particle measurement chamber opposite to each other;

[0014] The first photoelectric sensing unit is mounted on the rear side wall of the particle measurement chamber and is arranged at 90° to the first laser. At the same time, it is at the same height as the center of the first laser.

[0015] The first filter is coaxially supported on the front side of the first photoelectric sensing unit;

[0016] The second photoelectric sensing unit is mounted on the front side wall of the particle measurement chamber and is arranged at 90 degrees to the second laser and at the same height as the center of the second laser;

[0017] The second filter is coaxially supported on the front side of the second photoelectric sensing unit;

[0018] The temperature and humidity sensor is arranged in the particle measurement chamber;

[0019] The first mass flow controller and the third flow pump are connected in series to the connecting pipe B from the air inlet side to the air outlet side;

[0020] The data processing unit includes a data acquisition and control module and a data processing system;

[0021] The data acquisition and control module is respectively connected to the first mass flow controller, the second mass flow controller, the first flow pump, the second flow pump, the third flow pump, the fourth flow pump, the first laser, the second laser, the first photoelectric sensing unit, the second photoelectric sensing unit, the temperature and humidity sensor, the oscillation drive module and the high-precision frequency measurement module;

[0022] The data processing system is connected to the data acquisition and control module, and has a particle size distribution measurement model and a dual-drive dust concentration calibration model based on physical drive and data drive.

[0023] Furthermore, in order to facilitate relevant personnel to observe the monitored dust concentration data in real time, a display screen is also included, which is embedded on the outer surface of the shell and connected to the data acquisition and control module.

[0024] The fourth air inlet and the first air inlet are both opened on the top plate of the shell, the second air outlet and the first air outlet are both opened on the bottom plate of the shell, and the first air inlet and the first air outlet are relatively distributed, and the fourth air inlet and the second air outlet are relatively distributed; the second air inlet is opened on the left side plate of the shell, and the third air inlet is opened on the right side plate of the shell, and the second air inlet and the third air inlet are relatively distributed.

[0025] Furthermore, in order to enable the clean airflow entering from the two airflow branches to better form a protective airflow outside the dusty airflow, the length of the mixing section accounts for a greater proportion than the lengths of the air intake section and the acceleration section.

[0026] Furthermore, in order to fully absorb the light beam transmitted or scattered onto the side wall of the particle measurement chamber to avoid adversely affecting the receiving signal of the photoelectric sensing unit, the particle measurement chamber is made of black matte or black light-absorbing material, and a non-adsorbent material with anti-adhesion properties is coated on the inner wall.

[0027] Further, in order to ensure measurement accuracy, the wavelengths of the first laser and the second laser are different, and the center wavelengths of the first filter and the second filter correspond to the wavelengths of the first laser and the second laser, respectively.

[0028] In the present invention, by installing a detachable collision separator at the fourth air inlet, the detachable collision separator can be easily assembled and detached from the shell, and at the same time, it is also easy to replace collision separators of different models. The first high-efficiency filter and the second high-efficiency filter are respectively arranged in the first gas branch and the second gas branch, so that the dust in the dust-containing airflow can be effectively filtered by the first high-efficiency filter and the second high-efficiency filter during the operation of the first flow pump and the second flow pump, thereby forming two clean airflows. A chamber inlet A is provided at the air inlet section at the top of the particle flight chamber, and chamber inlets B and C are provided at the mixing section in the middle of the particle flight chamber. At the same time, the chamber inlet A is connected to the first air inlet through the connecting pipe A, and the outlet ends of the two airflow branches are connected to the chamber inlets B and C at relative angles. In this way, a sheath flow structure can be formed, and then the two clean airflows entering from both sides can be used to form a protective airflow for the dusty airflow introduced from the top, and the dusty airflow can flow in a straight line, and then when passing through the particle measurement chamber, it can be convenient for the laser and the photoelectric sensing unit to better act on the dusty airflow, so as to obtain more accurate measurement data. By installing a light trap on the side opposite to the laser, it is convenient to effectively absorb the laser that passes through the dusty airflow, and then it can be avoided that the light beam transmitted to the left wall of the particle measurement chamber is scattered to the photoelectric sensing unit to cause measurement errors. By installing the first laser and the second laser on the upper and lower parts of the right side wall of the particle measurement chamber, respectively, and installing the first light trap and the second light trap on the upper and lower parts of the left side wall of the particle measurement chamber, respectively, and installing the first photoelectric sensing unit and the second photoelectric sensing unit on the rear side wall of the particle measurement chamber, respectively, it is convenient to use the two lasers to emit laser beams of different wavelengths, and at the same time, the two light traps can be used to absorb the light beams transmitted to the left side wall of the particle measurement chamber, and the two photoelectric sensors can be used to receive the light signals scattered by the particles in the dust-containing airflow, and convert them into electrical signals respectively. By setting the temperature and humidity sensor, it is convenient to collect the temperature and humidity signals in the particle measurement chamber in real time, and then it is convenient to obtain the temperature and humidity data in the particle measurement chamber. The first mass flow controller is set on the connecting pipe B, which can facilitate the precise control of the flow of the gas path inside the particle measurement unit, and then it can ensure the signal collection accuracy during the detection process, which is conducive to obtaining more accurate test data.The magnet and the electromagnet are sequentially arranged on the outside of the oscillation tube, and the electromagnet is connected to the oscillation driving module, so that the oscillation driving module can be used to drive the magnet to vibrate through the electromagnet, and then the oscillation tube can be driven to oscillate with equal amplitude; through the setting of the high-precision frequency measurement module, it is convenient to realize the high-precision measurement of the oscillation frequency by inducing the change of the magnetic field strength during the equal amplitude oscillation of the oscillation tube, and then the dust mass deposited on the filter membrane can be calculated by the change of the oscillation frequency; by connecting the second mass flow controller in series on the connecting pipeline C, it is convenient to accurately control the flow of the gas path inside the oscillation measurement unit, and then the dust concentration prediction value A can be calculated according to the flow of the gas path inside the oscillation measurement unit and the dust mass deposited on the filter membrane. Through the setting of the data acquisition and control module, the collected signals can be received, and each component can be reliably controlled. Through the setting of the data processing system, it is convenient to obtain the dust concentration data measured by the oscillation measurement unit based on the signal collected by the photoelectric sensing unit, the temperature and humidity signal collected by the temperature and humidity sensor, and the flow data fed back by the first mass flow controller. Furthermore, the data processing system is provided with a particle size distribution measurement model and a dual-drive dust concentration calibration model based on physical drive and data drive. On the one hand, the particle size distribution data of the particles can be obtained based on the two-way electrical signals sent by the first and second photoelectric sensing units and the flow data fed back by the first mass flow controller using the particle size distribution measurement model. On the other hand, the dual-drive dust concentration calibration model can be used to predict the dust concentration data based on the particle size distribution data and the temperature and humidity data, and can be compared with the dust concentration data measured by the oscillation measurement unit, and further obtain the dust concentration prediction value by comprehensive calculation. The device has good reliability, strong versatility, and high degree of automation, and can realize high-precision monitoring of dust concentration in the environment.

[0029] The present invention also provides a dust concentration monitoring method based on physical and data dual drive, using a dust concentration monitoring device based on physical and data dual drive, comprising the following steps:

[0030] S1: establishing communication between the data acquisition and control module and the first mass flow controller, the second mass flow controller, the first flow pump, the second flow pump, the third flow pump, the fourth flow pump, the first laser, the second laser, the first photoelectric sensing unit, the second photoelectric sensing unit, the oscillation drive module, the high-precision frequency measurement module and the data processing system;

[0031] S2: In the non-intake state, the oscillation driving module is controlled to output a dynamic driving signal to the electromagnet, and the dynamic driving signal is used to drive the electromagnet excitation coil to energize so that the magnetic flux increases and acts on the magnetic steel on the oscillation tube, thereby driving the oscillation tube to oscillate with equal amplitude, and the high-precision frequency measurement module is used to sense the change in magnetic field intensity to achieve high-precision measurement of the initial state oscillation frequency, thereby obtaining the initial state oscillation frequency of the oscillation tube;

[0032] S3: Control the first flow pump, the second flow pump, the third flow pump, the first mass flow controller, the first laser, the second laser, the first photoelectric sensing unit and the second photoelectric sensing unit to start working, and ensure that the gas path inside the particle measurement unit is in a normal working flow state; synchronously, control the fourth flow pump, the second mass flow controller and the oscillation driving module to start working, and ensure that the gas path inside the oscillation measurement unit is in a normal working flow state;

[0033] A third flow pump is used to provide negative pressure to the particle flight chamber through the particle measurement chamber, and the air flow rate through the connecting pipe B is adjusted by the first mass flow controller, so that the dust-laden air flow enters the air intake section of the particle flight chamber through the connecting pipe A, and then flows to the acceleration section through the mixing section;

[0034] A first flow pump and a second flow pump are used to provide negative pressure to airflow branch one and airflow branch two respectively, and the dust-containing airflow entering airflow branch one and airflow branch two is filtered by a first high-efficiency filter and a second high-efficiency filter respectively to form two clean airflows. At the same time, the airflow flow in airflow branch one and airflow branch two is limited by a first flow limiting component and a second flow limiting component respectively, so that the two clean airflows after flow limiting enter the particle flight chamber relatively and meet in the mixing section of the particle flight chamber, and a layer of sheath flow with higher flow rate is formed around the dust-containing airflow, so as to ensure that the particles in the dust-containing airflow do not diverge and can flow to the acceleration section along a path that is approximately straight; under the acceleration effect of the acceleration section, the dust-containing airflow forms a sample flow with better linearity, and the sample flow flows through the particle measurement chamber under the action of negative pressure. , and discharged to the outside through the first air outlet; in this process, two laser beams emitted by the first laser and the second laser arranged on the right side of the particle measurement chamber are focused on the mixed airflow passing through the center of the particle measurement chamber, and the first light trap and the second light trap arranged on the left side of the particle measurement chamber are used to absorb the two transmitted light beams irradiated on the left side plate of the particle measurement chamber, and the first filter and the second filter are used to filter the two scattered light beams refracted to the rear side plate of the particle measurement chamber, and then the first photoelectric sensing unit and the second photoelectric sensing unit are used to receive the two scattered light signals generated by the dust particles respectively, and further convert the two scattered light signals into two electrical signals; at the same time, the temperature and humidity signals in the particle measurement chamber are collected in real time by the temperature and humidity sensor, and the temperature and humidity data are obtained based on the temperature and humidity signals;

[0035] A fourth flow pump is used to provide negative pressure to the connecting pipe D through the connecting pipe C, the oscillating pipe and the filter membrane support. At the same time, the air flow rate through the connecting pipe C is adjusted by the second mass flow controller, so that the dust-laden air flow passes through the filter membrane, the filter membrane support and the oscillating pipe in sequence and is discharged to the outside through the second air outlet; at the same time, the frequency change of the oscillating pipe is monitored in real time by a high-precision frequency measurement module to obtain the oscillation frequency of the oscillating pipe in the detection state;

[0036] S4: The data processing system calculates the dust mass deposited on the filter membrane according to the change of the oscillation frequency, and calculates the dust concentration prediction value A in the test environment in combination with the flow rate of the second mass flow controller. At the same time, the dust concentration prediction value A is sent to the display for real-time display;

[0037] At the same time, the data processing system filters, amplifies and digitizes the two electrical signals based on the built-in particle size distribution measurement model, extracts features from the processed signals, and then obtains the particle size distribution data of the particles through the particle size distribution algorithm; further, the data processing system uses the built-in dual-drive dust concentration calibration model to process the particle size distribution data and the temperature and humidity data. Specifically, the particle size distribution data and the temperature and humidity data are first input into the model layer containing the parameterized physical model and the machine learning model. The parameterized physical model uses the determined model parameters and the temperature and humidity data to correct the particle size distribution data, and performs Secondary statistical classification is performed to obtain different particle size channel data, and then the total scattered light intensity value is obtained by using the Mie scattered light intensity calculation value of the particles in different channels. The dust concentration prediction value B is calculated using the total scattered light intensity value, the channel scattered light intensity calculation value and the flow rate fed back by the first mass flow controller. After receiving the corrected particle size distribution data and temperature and humidity data, the machine learning model uses the model parameters to predict the dust concentration and obtains the dust concentration prediction value C. The dust concentration prediction value B and the dust concentration prediction value C are then input into the fusion layer, and the difference between the dust concentration prediction value B and the dust concentration prediction value C exceeds the difference When the dust concentration prediction value B is greater than the threshold, the weight is adjusted to give a higher weight to the parameterized physical model, and the dust concentration prediction value B is added to the historical database of the machine learning model, so that the machine learning model updates and recalculates the model based on the updated historical database, and outputs the new dust concentration prediction value B and the new dust concentration prediction value C to the fusion layer until the difference between the received dust concentration prediction value B and the dust concentration prediction value C is within the difference threshold. The fusion layer then uses the neural network fusion algorithm to process the input data. In the process of processing the input data through multiple neural network layers, each layer converts its input through the activation function to capture The neural network captures and extracts the key feature information of the input data, and effectively combines the information of different models by adjusting the weights and biases of each layer, and outputs the dust concentration prediction value D. Furthermore, the data processing system compares the dust concentration prediction value A with the dust concentration prediction value D, and calculates the partial derivative of the loss function with respect to the output of the fusion layer, and then calculates the model parameter gradient using the chain rule for the physical model parameters, the machine learning model parameters, and the fusion layer parameters, and uses the optimization algorithm to update the model parameters according to the calculated gradient. Finally, the fusion layer performs a comprehensive calculation on the outputs of the parameterized physical model and the machine learning model, and obtains the dust concentration prediction value.

[0038] S5: Control the fourth flow pump, the second mass flow controller and the oscillation drive module to stop working, and continue to maintain the normal working state of the first flow pump, the second flow pump, the third flow pump, the first mass flow controller, the first laser, the second laser, the first photoelectric sensing unit and the second photoelectric sensing unit, so that the particle measurement unit works alone for a period of time, and outputs the obtained dust concentration prediction value to the display screen for real-time display;

[0039] S6: Control the fourth flow pump, the second mass flow controller and the oscillation drive module to restart working, and update the model parameters according to the method in S4;

[0040] S7: Repeat S3 to S6, and use the periodic dust concentration measurement data of the oscillation measurement unit to continuously optimize the model parameters. Finally, the fusion layer outputs the final predicted value of the dust concentration and outputs it to the display screen for real-time display.

[0041] In the present invention, in the non-intake state, the initial state oscillation frequency of the oscillating tube is obtained, which can be used as the reference frequency data, and then in the intake state, the dust mass deposited on the filter membrane can be obtained by only monitoring the frequency change of the oscillating tube, thereby, in combination with the flow data fed back by the second mass flow controller, the dust concentration prediction value A monitored by the oscillation measurement unit can be accurately obtained. In the measurement process, negative pressure is provided to the two air flow branches, and two high-efficiency filters are used for filtering, so that two clean air flows can be formed, so that the two clean air flows meet at the mixing section in the particle flight chamber at relative angles, and at the same time, the dust-containing air flow is introduced into the air intake section of the particle flight chamber, so that the two clean air flows can form a protective air flow at the periphery of the dust-containing air flow, and then the dust-containing air flow can flow through the particle measurement chamber in an approximately straight line, which is conducive to the light speed emitted by the laser to act more accurately on the particles in the dust-containing air flow, and then the scattering phenomenon can be better generated, and at the same time, it can ensure that the photoelectric sensing unit receives the scattered light intensity signal more accurately, and further obtain a more accurate electrical signal. The particle size distribution data of the particles can be accurately obtained by using the particle size distribution measurement model through the particle size distribution algorithm. When the particle size distribution data and temperature and humidity data are processed by the dual-driven dust concentration calibration model, the particle size distribution data can be corrected based on the temperature and humidity data using the parameterized physical model in the model layer, and the corrected particle size distribution data can be obtained, and then the scattered light intensity value can be accurately obtained. Combined with the flow rate fed back by the first mass flow controller, the dust concentration prediction value B measured by the parameterized physical model can be obtained more accurately. Therefore, the system behavior can be basically described by the parameterized physical model to ensure the credibility of the prediction. The machine learning model can obtain the dust concentration prediction value C based on the corrected particle size distribution data and temperature and humidity data through machine learning, which can give full play to the advantages of the data-driven characteristics of the machine learning model, learn complex nonlinear relationships through a large amount of data, and enhance the flexibility and generalization ability of the model. The fusion layer is used to compare the difference between the dust concentration prediction values ​​B and C, and the weights are adjusted according to the difference. This can make the dust concentration prediction value C output by the machine learning model closer to the true value. The prediction results of the two models are further optimized through the neural network, realizing the end-to-end optimization process of the overall system.

[0042] The present invention combines the parameter physical model with the machine learning model and performs comprehensive calculation of the outputs of the two through the neural network fusion layer, making comprehensive use of the respective advantages of the physical model and the machine learning model. The physical model provides a reliable basic description of the system with strong explanatory power and credibility; while the machine learning model can automatically extract features from a large amount of data and capture complex nonlinear relationships, thereby improving the flexibility and accuracy of the prediction. This combination significantly improves the accuracy of the prediction, and can simultaneously utilize the nonlinear characteristics of physical laws and data-driven, and finally output more accurate prediction results through the optimization of the fusion layer. Through the back-propagation mechanism, each part (physical model, machine learning model and fusion layer) is updated collaboratively during the training process, ensuring the optimal performance and consistency of the overall system. The robustness of the system is enhanced. The physical model can still guarantee the basic prediction accuracy when the data is limited or the data quality is poor, while the machine learning model learns experience from historical data, enhancing the model's adaptability to different environments and conditions, thereby providing a more comprehensive system performance.

[0043] The present invention makes full use of the theoretical basis of the physical model and the data-driven characteristics of the machine learning model, thereby fully utilizing the advantages of both and significantly improving the accuracy and stability of the prediction. The method has simple steps and a high degree of automation. It can automatically monitor the dust concentration in the environment and effectively ensure the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a structural schematic diagram of the monitoring device in the present invention;

[0045] Figure 2 It is a schematic diagram of the arrangement of the photoelectric sensing unit and the filter in the particle measurement chamber of the present invention;

[0046] Figure 3 It is a structural schematic diagram of the dual-drive dust concentration calibration model in the present invention;

[0047] Figure 4 It is a flow chart of the monitoring method part in the present invention.

[0048] In the figure: 1, housing, 2, first current limiting component, 3, second current limiting component, 4, second high efficiency filter, 5, first high efficiency filter, 6, first mass flow controller, 7, second mass flow controller, 8, third flow pump, 9, first flow pump, 10, second flow pump, 11, fourth flow pump, 12, particle flight chamber, 13, particle measurement chamber, 14, first laser, 15, second laser, 16, first photoelectric sensing unit, 17, second photoelectric sensing unit, 18, first filter, 19, second filter, 20, first light trap, 21. Second light trap, 22. Oscillating tube, 23. Magnetic steel, 24. Oscillation drive module, 25. High-precision frequency measurement module, 26. Filter membrane holder, 27. Removable collision separator, 28. Temperature and humidity sensor, 29. Data acquisition and control module, 30. Data processing system, 31. Display screen, 32. First air inlet, 33. Second air inlet, 34. Third air inlet, 35. Fourth air inlet, 36. First air outlet, 37. Second air outlet, 38. Air flow branch one, 39. Air flow branch two, 40. Connecting pipeline A, 41. Connecting pipeline B, 42. Connecting pipeline C, 43. Connecting pipeline D. DETAILED DESCRIPTION

[0049] The present invention will be further described below.

[0050] like Figure 1 and Figure 2 As shown, the present invention provides a dust concentration monitoring device based on physical and data dual drive, including a shell 1, an oscillation measurement unit, a particle measurement unit and a data processing unit, and the shell 1 is respectively provided with a first air inlet 32, a second air inlet 33, a third air inlet 34, a fourth air inlet 35, a first air outlet 36 and a second air outlet 37 connected to its inner cavity;

[0051] The oscillation measurement unit includes a detachable collision separator 27, an oscillation measurement mechanism, a second mass flow controller 7 and a fourth flow pump 11; the detachable collision separator 27 is installed at the fourth air inlet 35; the oscillation measurement mechanism is mainly composed of an oscillation tube 22, a filter membrane holder 26, a magnetic steel 23, an electromagnet, an oscillation drive module 24 and a high-precision frequency measurement module 25; the oscillation tube 22 is a frustum-shaped quartz oscillation tube with an elliptical cross-section, the small diameter end of the upper end serves as the air inlet, the large diameter end of the lower end serves as the air outlet, and is connected to the second air outlet 37 through a connecting pipe C42; the filter membrane holder 26 is a thin-walled frustum structure, which is located at the oscillation tube 22. Above the oscillating tube 22, a filter membrane is installed at the large diameter end of the upper end of the filter membrane holder 26, and is connected to the fourth air inlet 35 through a connecting pipe D43, and the small diameter end of the lower end is connected to the small diameter end of the oscillating tube 22; a pair of magnets 23 are axially symmetrically distributed on opposite sides of the outside of the long axis direction of the oscillating tube 22, and are tightly adhered to the side walls of the upper part of the oscillating tube 22; a pair of electromagnets are symmetrically supported on the outside of the pair of magnets 23 and are on the same horizontal line with the center of the magnets 23, and are used to drive the magnets 23 to vibrate; the oscillation drive module 24 is connected to the pair of electromagnets, and is used to output a dynamic drive signal to the electromagnets; the high-precision frequency measurement module 25 is installed on the outside of the oscillating tube 22 and is at the same height as the electromagnets, and is used to sense the change in magnetic field strength to achieve high-precision measurement of the oscillation frequency; the second mass flow controller 7 and the fourth flow pump 11 are connected in series to the connecting pipe C42 from the air inlet side to the air outlet side;

[0052] The particle measurement unit includes a particle flight chamber 12, a first high efficiency filter 5, a first flow pump 9, a first current limiting component 2, a second high efficiency filter 4, a second flow pump 10, a second current limiting component 3, a particle measurement chamber 13, a first light trap 20, a first laser 14, a second light trap 21, a second laser 15, a first photoelectric sensing unit 16, a first filter 18, a second photoelectric sensing unit 17, a second filter 19, a temperature and humidity sensor 28, a first mass flow controller 6 and a third flow pump 8;

[0053] The particle flight chamber 12 is located in the shell 1, and is divided into an air intake section, a mixing section and an acceleration section from top to bottom. The air intake section is a conical structure, the small diameter end of which is located at the upper part, and a chamber inlet A is opened. The mixing section is a cylindrical structure, and the upper part is provided with a chamber inlet B and a chamber inlet C on two opposite sides. The acceleration section is a conical structure, the small diameter end of which is located at the upper part, and a chamber outlet A is opened; the chamber inlet A is connected to the first air inlet 32 ​​through a connecting pipe A40, the chamber inlet B is connected to the second air inlet 33 through an air flow branch 1 38, and the chamber inlet C is connected to the third air inlet 34 through an air flow branch 2 39; the air outlet ends of the air flow branch 1 38 and the air flow branch 2 39 are arranged at a relative angle to the central axis of the particle flight chamber 12, and are not perpendicular to the particle flight chamber 12, wherein the air outlet ends of the air flow branch 1 38 and the air outlet ends of the air flow branch 2 39 are both oriented toward the acceleration section;

[0054] The first high-efficiency filter 5, the first flow pump 9, and the first flow limiting component 2 are connected in series to the first air flow branch 38 from the air inlet side to the air outlet side; the second high-efficiency filter 4, the second flow pump 10, and the second flow limiting component 3 are connected in series to the second air flow branch 39 from the air inlet side to the air outlet side;

[0055] The particle measurement chamber 13 is located in the housing 1 and below the particle flight chamber 12. The top and bottom of the particle measurement chamber 13 are relatively provided with a chamber outlet D and a chamber outlet B communicating with the inner cavity thereof, wherein the chamber outlet D is communicated with the chamber outlet A, and the chamber outlet B is connected to the first air outlet 36 via a communication pipe B41.

[0056] The first optical trap 20 and the first laser 14 are mounted on the side walls on the left and right sides of the upper inner cavity of the particle measurement chamber 13 relative to each other;

[0057] The second optical trap 21 and the second laser 15 are installed oppositely on the left and right side walls of the lower part of the particle measurement chamber 13;

[0058] The first photoelectric sensing unit 16 is mounted on the rear side wall of the particle measurement chamber 13 and is arranged at 90° to the first laser 14. At the same time, it is at the same height as the center of the first laser 14.

[0059] The first filter 18 is coaxially supported on the front side of the first photoelectric sensing unit 16;

[0060] The second photoelectric sensing unit 17 is mounted on the front side wall of the particle measurement chamber 13 and is arranged at 90 degrees to the second laser 15 and at the same height as the center of the second laser 15;

[0061] The second filter 19 is coaxially supported on the front side of the second photoelectric sensing unit 17;

[0062] The temperature and humidity sensor 28 is disposed in the particle measurement chamber 13;

[0063] The first mass flow controller 6 and the third flow pump 8 are connected in series to the connecting pipe B41 from the air inlet side to the air outlet side;

[0064] The data processing unit includes a data acquisition and control module 29 and a data processing system 30;

[0065] The data acquisition and control module 29 is respectively connected to the first mass flow controller 6, the second mass flow controller 7, the first flow pump 9, the second flow pump 10, the third flow pump 8, the fourth flow pump 11, the first laser 14, the second laser 15, the first photoelectric sensing unit 16, the second photoelectric sensing unit 17, the temperature and humidity sensor 28, the oscillation driving module 24 and the high-precision frequency measurement module 25;

[0066] The data processing system 30 is connected to the data acquisition and control module 29 .

[0067] The data processing system 30 includes a particle size distribution measurement model and a dual-drive dust concentration calibration model based on physical drive and data drive; the particle size distribution measurement model uses the data acquisition and control module 29 to receive two electrical signals emitted by the first photoelectric sensing unit 16 and the second photoelectric sensing unit 17, and filters, amplifies and digitizes the two electrical signals, and extracts features from the processed signals, and uses the particle size distribution algorithm to convert these feature information into particle size distribution data;

[0068] The dual-drive dust concentration calibration model receives the particle size distribution data output by the particle size distribution measurement model and the temperature and humidity data transmitted by the temperature and humidity sensor 28, and inputs these data distributions into the model layer, which includes a parameterized physical model and a machine learning model respectively;

[0069] The parameterized physical model uses the determined physical model parameters and temperature and humidity data to correct the particle size distribution data, performs secondary statistical classification on the corrected particle size distribution data to divide and count different particle size channel data, uses the Mie scattered light intensity calculation value of particles in different channels to obtain the total scattered light intensity value, and uses the total scattered light intensity value and the channel scattered light intensity calculation value to calculate the dust concentration prediction value; the corrected particle size distribution data calculated by the parameterized physical model and the temperature and humidity sensor data are jointly input into the machine learning model, and the machine learning model predicts the dust concentration using the constructed model parameters; the dust concentration prediction value of the parameterized physical model and the dust concentration prediction value of the machine learning model are input into the fusion layer, and the fusion layer uses a neural network fusion algorithm The input data is processed by multiple neural network layers, and each layer transforms its input through an activation function to capture and extract key feature information. The neural network effectively combines the information of different models by adjusting the weights and biases of each layer, and outputs the final predicted value of dust concentration. The fusion layer sets a difference threshold. When the dust concentration predicted value of the parameterized physical model exceeds the dust concentration predicted value of the machine learning model, the fusion layer adjusts the weight and introduces a weight factor to control the influence of the output of the physical model and the machine learning model, giving the physical model a higher weight. At the same time, the dust concentration predicted value of the physical model will be added to the historical database of the machine learning, and the machine learning model will perform model updates and recalculations based on the updated historical database.

[0070] In order to facilitate relevant personnel to observe the monitored dust concentration data in real time, a display screen 31 is also included. The display screen 31 is embedded on the outer surface of the shell 1 and is connected to the data acquisition and control module 29.

[0071] The fourth air inlet 35 and the first air inlet 32 ​​are both opened on the top plate of the shell 1, the second air outlet 37 and the first air outlet 36 are both opened on the bottom plate of the shell 1, and the first air inlet 32 ​​and the first air outlet 36 are relatively distributed, and the fourth air inlet 35 and the second air outlet 37 are relatively distributed; the second air inlet 33 is opened on the left plate of the shell 1, and the third air inlet 34 is opened on the right plate of the shell 1, and the second air inlet 33 and the third air inlet 34 are relatively distributed.

[0072] In order to enable the clean airflow entering from the two airflow branches to better form a protective airflow outside the dusty airflow, the length of the mixing section accounts for a greater proportion than the length of the air intake section and the acceleration section.

[0073] In order to fully absorb the light beam transmitted or scattered onto the side wall of the particle measurement chamber to avoid adversely affecting the receiving signal of the photoelectric sensing unit, the particle measurement chamber 13 is made of black matte or black light-absorbing material, and a non-adsorbent material with anti-adhesion properties is coated on the inner wall.

[0074] In order to ensure measurement accuracy, the wavelengths of the first laser 14 and the second laser 15 are different, and the central wavelengths of the first filter 18 and the second filter 19 correspond to the wavelengths of the first laser 14 and the second laser 15 respectively.

[0075] In the present invention, by installing a detachable collision separator at the fourth air inlet, the detachable collision separator can be easily assembled and detached from the shell, and at the same time, it is also easy to replace collision separators of different models. The first high-efficiency filter and the second high-efficiency filter are respectively arranged in the first gas branch and the second gas branch, so that the dust in the dust-containing airflow can be effectively filtered by the first high-efficiency filter and the second high-efficiency filter during the operation of the first flow pump and the second flow pump, thereby forming two clean airflows. A chamber inlet A is provided at the air inlet section at the top of the particle flight chamber, and chamber inlets B and C are provided at the mixing section in the middle of the particle flight chamber. At the same time, the chamber inlet A is connected to the first air inlet through the connecting pipe A, and the outlet ends of the two airflow branches are connected to the chamber inlets B and C at relative angles. In this way, a sheath flow structure can be formed, and then the two clean airflows entering from both sides can be used to form a protective airflow for the dusty airflow introduced from the top, and the dusty airflow can flow in a straight line, and then when passing through the particle measurement chamber, it can be convenient for the laser and the photoelectric sensing unit to better act on the dusty airflow, so as to obtain more accurate measurement data. By installing a light trap on the side opposite to the laser, it is convenient to effectively absorb the laser that passes through the dusty airflow, and then it can be avoided that the light beam transmitted to the left wall of the particle measurement chamber is scattered to the photoelectric sensing unit to cause measurement errors. By installing the first laser and the second laser on the upper and lower parts of the right side wall of the particle measurement chamber, respectively, and installing the first light trap and the second light trap on the upper and lower parts of the left side wall of the particle measurement chamber, respectively, and installing the first photoelectric sensing unit and the second photoelectric sensing unit on the rear side wall of the particle measurement chamber, respectively, it is convenient to use the two lasers to emit laser beams of different wavelengths, and at the same time, the two light traps can be used to absorb the light beams transmitted to the left side wall of the particle measurement chamber, and the two photoelectric sensors can be used to receive the light signals scattered by the particles in the dust-containing airflow, and convert them into electrical signals respectively. By setting the temperature and humidity sensor, it is convenient to collect the temperature and humidity signals in the particle measurement chamber in real time, and then it is convenient to obtain the temperature and humidity data in the particle measurement chamber. The first mass flow controller is set on the connecting pipe B, which can facilitate the precise control of the flow of the gas path inside the particle measurement unit, and then it can ensure the signal collection accuracy during the detection process, which is conducive to obtaining more accurate test data.The magnet and the electromagnet are sequentially arranged on the outside of the oscillation tube, and the electromagnet is connected to the oscillation driving module, so that the oscillation driving module can be used to drive the magnet to vibrate through the electromagnet, and then the oscillation tube can be driven to oscillate with equal amplitude; through the setting of the high-precision frequency measurement module, it is convenient to realize the high-precision measurement of the oscillation frequency by inducing the change of the magnetic field strength during the equal amplitude oscillation of the oscillation tube, and then the dust mass deposited on the filter membrane can be calculated by the change of the oscillation frequency; by connecting the second mass flow controller in series on the connecting pipeline C, it is convenient to accurately control the flow of the gas path inside the oscillation measurement unit, and then the dust concentration prediction value A can be calculated according to the flow of the gas path inside the oscillation measurement unit and the dust mass deposited on the filter membrane. Through the setting of the data acquisition and control module, the collected signals can be received, and each component can be reliably controlled. Through the setting of the data processing system, it is convenient to obtain the dust concentration data measured by the oscillation measurement unit based on the signal collected by the photoelectric sensing unit, the temperature and humidity signal collected by the temperature and humidity sensor, and the flow data fed back by the first mass flow controller. Furthermore, the data processing system is provided with a particle size distribution measurement model and a dual-drive dust concentration calibration model based on physical drive and data drive. On the one hand, the particle size distribution data of the particles can be obtained based on the two-way electrical signals sent by the first and second photoelectric sensing units and the flow data fed back by the first mass flow controller using the particle size distribution measurement model. On the other hand, the dual-drive dust concentration calibration model can be used to predict the dust concentration data based on the particle size distribution data and the temperature and humidity data, and can be compared with the dust concentration data measured by the oscillation measurement unit, and further obtain the dust concentration prediction value by comprehensive calculation. The device has good reliability, strong versatility, and high degree of automation, and can realize high-precision monitoring of dust concentration in the environment.

[0076] like Figure 3 and Figure 4 As shown, the present invention also provides a dust concentration monitoring method based on physical and data dual drive, using a dust concentration monitoring device based on physical and data dual drive, including the following steps:

[0077] S1: establishing communication between the data acquisition and control module 29 and the first mass flow controller 6, the second mass flow controller 7, the first flow pump 9, the second flow pump 10, the third flow pump 8, the fourth flow pump 11, the first laser 14, the second laser 15, the first photoelectric sensing unit 16, the second photoelectric sensing unit 17, the oscillation driving module 24, the high-precision frequency measurement module 25 and the data processing system 30;

[0078] S2: In the non-intake state, the oscillation driving module 24 is controlled to output a dynamic driving signal to the electromagnet, and the dynamic driving signal is used to drive the electromagnet excitation coil to energize so that the magnetic flux increases and acts on the magnetic steel 23 on the oscillation tube 22, thereby driving the oscillation tube 22 to oscillate with equal amplitude, and the high-precision frequency measurement module 25 is used to sense the change in magnetic field intensity to achieve high-precision measurement of the initial state oscillation frequency, and obtain the initial state oscillation frequency of the oscillation tube 22;

[0079] S3: Control the first flow pump 9, the second flow pump 10, the third flow pump 8, the first mass flow controller 6, the first laser 14, the second laser 15, the first photoelectric sensing unit 16 and the second photoelectric sensing unit 17 to start working, and ensure that the gas path inside the particle measurement unit is in a normal working flow state; synchronously, control the fourth flow pump 11, the second mass flow controller 7 and the oscillation driving module 24 to start working, and ensure that the gas path inside the oscillation measurement unit is in a normal working flow state;

[0080] The third flow pump 8 is used to provide negative pressure to the particle flight chamber 12 through the particle measurement chamber 13, and the air flow rate through the connecting pipe B41 is adjusted by the first mass flow controller 6, so that the dust-laden air flow enters the air intake section of the particle flight chamber 12 through the connecting pipe A40, and then flows to the acceleration section through the mixing section;

[0081] A first flow pump 9 and a second flow pump 10 are used to provide negative pressure to air flow branch 1 38 and air flow branch 2 39 respectively, and the dust-containing airflow entering air flow branch 1 38 and air flow branch 2 39 is filtered by a first high-efficiency filter 5 and a second high-efficiency filter 4 respectively to form two clean airflows. At the same time, the airflow flow in air flow branch 1 38 and air flow branch 2 39 is limited by a first flow limiting component 2 and a second flow limiting component 3 respectively, so that the two clean airflows after flow limiting enter the particle flight chamber 12 relatively and meet in the mixing section of the particle flight chamber 12, and a layer of sheath flow with higher flow rate is formed around the dust-containing airflow, so as to ensure that the particles in the dust-containing airflow do not diverge and can flow to the acceleration section along an approximately straight path; under the acceleration effect of the acceleration section, the dust-containing airflow forms a sample flow with better linearity, and the sample flow flows through the particle measurement chamber 13 under the action of negative pressure, and is The first air outlet 36 discharges to the outside; in this process, the two laser beams emitted by the first laser 14 and the second laser 15 arranged on the right side of the particle measurement chamber 13 are focused on the mixed airflow passing through the center of the particle measurement chamber 13, and the first light trap 20 and the second light trap 21 arranged on the left side of the particle measurement chamber 13 are used to absorb the two transmitted light beams irradiated on the left side plate of the particle measurement chamber 13, and the first filter 18 and the second filter 19 are used to filter the two scattered light beams refracted to the rear side plate of the particle measurement chamber 13, and then the first photoelectric sensing unit 16 and the second photoelectric sensing unit 17 are used to receive the two scattered light signals generated by the dust particles, and further convert the two scattered light signals into two electrical signals; at the same time, the temperature and humidity signals in the particle measurement chamber 13 are collected in real time by the temperature and humidity sensor 28, and the temperature and humidity data are obtained based on the temperature and humidity signals;

[0082] The fourth flow pump 11 is used to provide negative pressure to the connecting pipeline D43 through the connecting pipeline C42, the oscillating tube 22 and the filter membrane holder 26. At the same time, the second mass flow controller 7 is used to adjust the air flow through the connecting pipeline C42, so that the dust-containing airflow passes through the filter membrane, the filter membrane holder 26 and the oscillating tube 22 in sequence and is discharged to the outside through the second air outlet 37; at the same time, the high-precision frequency measurement module 25 is used to monitor the frequency change of the oscillating tube in real time to obtain the oscillation frequency of the oscillating tube 22 in the detection state;

[0083] S4: The data processing system 30 calculates the dust mass deposited on the filter membrane according to the change of the oscillation frequency, and calculates the dust concentration prediction value A in the test environment in combination with the flow rate of the second mass flow controller 7, and at the same time, sends the dust concentration prediction value A to the display 31 for real-time display;

[0084] At the same time, the data processing system 30 filters, amplifies and digitizes the two electrical signals based on the built-in particle size distribution measurement model, extracts features from the processed signals, and then obtains particle size distribution data of the particles through a particle size distribution algorithm. The characteristic signals proposed by the particle size distribution measurement model include the flight time of the particles, the scattered light intensity received by the first photoelectric sensing unit 16, and the scattered light intensity received by the second photoelectric sensing unit 17;

[0085] As a preferred embodiment, the particle size distribution algorithm is constructed through massive experiments and related physical models, by passing particles of standard particle size through the particle measurement chamber 13 under a calibrated flow rate in a standard test environment, and by collecting the scattered light intensity received by the first photoelectric sensing unit 16 and the second photoelectric sensing unit signal 17, thereby determining the classification algorithm of particles of different particle sizes;

[0086] Furthermore, the data processing system 30 uses the built-in dual-drive dust concentration calibration model to process the particle size distribution data and the temperature and humidity data. Specifically, the particle size distribution data and the temperature and humidity data are first input into the model layer including the parameterized physical model and the machine learning model; the parameterized physical model first parameterizes the adjustable parameters in the physical model, and the physical model mainly uses the κ-Köhler theoretical model in combination with the temperature and humidity data to process the wet particle size distribution of the particles to obtain the dry particle size distribution characteristics of the particles. Thus, the parameterized physical model can use the determined model parameters and temperature and humidity data to correct the particle size distribution data, and further perform secondary statistical classification on the corrected particle size distribution data to obtain different particle size channel data, and then use the Mie scattering principle to calculate the scattered light intensity calculated value of the particle size channel representing a single particle, and further calculate the scattered light intensity calculated value of all particles in the particle size channel, and then use the Mie scattering intensity calculated value of particles in different channels to calculate. The total scattered light intensity calculation value of particles in all channels is used, and the relationship between the scattered light intensity calculation value of particles in each channel, the total scattered light intensity calculation value of particles in all channels and the dust concentration is established by using a heuristic dynamic optimization algorithm. Then, combined with the flow rate fed back by the first mass flow controller 6, the dust concentration prediction value B output by the physical model can be obtained; after receiving the corrected particle size distribution data and temperature and humidity data, the machine learning model uses the model parameters to predict the dust concentration and obtains the dust concentration prediction value C; the dust concentration prediction value B and the dust concentration prediction value C are then input into the fusion layer, and the fusion layer obtains the dust concentration prediction value B and When the difference between the dust concentration prediction values ​​C exceeds the difference threshold, the weight is adjusted to assign a higher weight to the parameterized physical model, and the dust concentration prediction value B is added to the historical database of the machine learning model, so that the machine learning model updates and recalculates the model based on the updated historical database, and outputs the new dust concentration prediction value B and the new dust concentration prediction value C to the fusion layer until the difference between the received dust concentration prediction value B and the dust concentration prediction value C is within the difference threshold. The fusion layer then uses a neural network fusion algorithm to process the input data. During the processing of the input data through multiple neural network layers, each layer converts its input through an activation function to capture and extract key feature information of the input data. The neural network adjusts the weights and biases of each layer to effectively combine the information of different models and output the dust concentration prediction value D. Further, the data processing system 30 compares the dust concentration prediction value A with the dust concentration prediction value D, and calculates the partial derivative of the loss function with respect to the output of the fusion layer. Then, the model parameter gradient is calculated using the chain rule for the physical model parameters, the machine learning model parameters, and the fusion layer parameters, and the optimization algorithm is used to update the model parameters according to the calculated gradient.Finally, the fusion layer performs comprehensive calculations on the outputs of the parameterized physical model and the machine learning model, and obtains the dust concentration prediction value to ensure the accuracy of the output;

[0087] S5: Control the fourth flow pump 11, the second mass flow controller 7 and the oscillation driving module 24 to stop working, and continue to maintain the normal working state of the first flow pump 9, the second flow pump 10, the third flow pump 8, the first mass flow controller 6, the first laser 14, the second laser 15, the first photoelectric sensing unit 16 and the second photoelectric sensing unit 17, so that the particle measurement unit works alone for a period of time, and outputs the obtained dust concentration prediction value to the display screen 31 for real-time display;

[0088] S6: Control the fourth flow pump 11, the second mass flow controller 7 and the oscillation driving module 24 to restart working, and update the model parameters according to the method in S4;

[0089] S7: Repeat S3 to S6, and use the periodic dust concentration measurement data of the oscillation measurement unit to continuously optimize the model parameters to achieve end-to-end optimization of the overall system. Finally, the fusion layer outputs the final predicted value of the dust concentration and outputs it to the display screen 31 for real-time display.

[0090] In the present invention, in the non-intake state, the initial state oscillation frequency of the oscillating tube is obtained, which can be used as the reference frequency data, and then in the intake state, the dust mass deposited on the filter membrane can be obtained by only monitoring the frequency change of the oscillating tube, thereby, in combination with the flow data fed back by the second mass flow controller, the dust concentration prediction value A monitored by the oscillation measurement unit can be accurately obtained. In the measurement process, negative pressure is provided to the two air flow branches, and two high-efficiency filters are used for filtering, so that two clean air flows can be formed, so that the two clean air flows meet at the mixing section in the particle flight chamber at relative angles, and at the same time, the dust-containing air flow is introduced into the air intake section of the particle flight chamber, so that the two clean air flows can form a protective air flow at the periphery of the dust-containing air flow, and then the dust-containing air flow can flow through the particle measurement chamber in an approximately straight line, which is conducive to the light speed emitted by the laser to act more accurately on the particles in the dust-containing air flow, and then the scattering phenomenon can be better generated, and at the same time, it can ensure that the photoelectric sensing unit receives the scattered light intensity signal more accurately, and further obtain a more accurate electrical signal. The particle size distribution data of the particles can be accurately obtained by using the particle size distribution measurement model through the particle size distribution algorithm. When the particle size distribution data and temperature and humidity data are processed by the dual-driven dust concentration calibration model, the particle size distribution data can be corrected based on the temperature and humidity data using the parameterized physical model in the model layer, and the corrected particle size distribution data can be obtained, and then the scattered light intensity value can be accurately obtained. Combined with the flow rate fed back by the first mass flow controller, the dust concentration prediction value B measured by the parameterized physical model can be obtained more accurately. Therefore, the system behavior can be basically described by the parameterized physical model to ensure the credibility of the prediction. The machine learning model can obtain the dust concentration prediction value C based on the corrected particle size distribution data and temperature and humidity data through machine learning, which can give full play to the advantages of the data-driven characteristics of the machine learning model, learn complex nonlinear relationships through a large amount of data, and enhance the flexibility and generalization ability of the model. The fusion layer is used to compare the difference between the dust concentration prediction values ​​B and C, and the weights are adjusted according to the difference. This can make the dust concentration prediction value C output by the machine learning model closer to the true value. The prediction results of the two models are further optimized through the neural network, realizing the end-to-end optimization process of the overall system.

[0091] The present invention combines the parameter physical model with the machine learning model and performs comprehensive calculation of the outputs of the two through the neural network fusion layer, making comprehensive use of the respective advantages of the physical model and the machine learning model. The physical model provides a reliable basic description of the system with strong explanatory power and credibility; while the machine learning model can automatically extract features from a large amount of data and capture complex nonlinear relationships, thereby improving the flexibility and accuracy of the prediction. This combination significantly improves the accuracy of the prediction, and can simultaneously utilize the nonlinear characteristics of physical laws and data-driven, and finally output more accurate prediction results through the optimization of the fusion layer. Through the back-propagation mechanism, each part (physical model, machine learning model and fusion layer) is updated collaboratively during the training process, ensuring the optimal performance and consistency of the overall system. The robustness of the system is enhanced. The physical model can still guarantee the basic prediction accuracy when the data is limited or the data quality is poor, while the machine learning model learns experience from historical data, enhancing the model's adaptability to different environments and conditions, thereby providing a more comprehensive system performance.

[0092] The present invention makes full use of the theoretical basis of the physical model and the data-driven characteristics of the machine learning model, thereby fully utilizing the advantages of both and significantly improving the accuracy and stability of the prediction. The method has simple steps and a high degree of automation. It can automatically monitor the dust concentration in the environment and effectively ensure the prediction accuracy.

Claims

1. A dust concentration monitoring method based on physical and data dual drive, using a dust concentration monitoring device based on physical and data dual drive, the dust concentration monitoring device based on physical and data dual drive comprising a housing (1), an oscillation measurement unit, a particle measurement unit, a data processing unit and a display screen (31); the housing (1) is provided with a first air inlet (32), a second air inlet (33), a third air inlet (34), a fourth air inlet (35), a first air outlet (36) and a second air outlet (37) which are connected to its inner cavity; the oscillation measurement unit comprises a detachable collision separator (27), an oscillation measurement mechanism, a second mass flow controller (7) and a fourth flow pump (11); the detachable collision separator (27) is installed at the fourth air inlet (35); the oscillation measurement mechanism is mainly composed of an oscillation tube (22), a filter membrane holder (26), a magnetic steel (23), an electromagnet, an oscillation drive module (24) and a high-precision frequency measurement module (25); the oscillation The tube (22) is a frustum-shaped quartz oscillating tube, and its cross section is elliptical. The small diameter end of its upper end serves as an air inlet, and the large diameter end of its lower end serves as an air outlet, and is connected to the second air outlet (37) through a connecting pipe C (42); the filter membrane holder (26) is a thin-walled frustum-shaped structure, which is located above the oscillating tube (22). The large diameter end of the upper end of the filter membrane holder (26) is equipped with a filter membrane, and is connected to the fourth air inlet (35) through a connecting pipe D (43). The small diameter end of its lower end is connected to the small diameter end of the oscillating tube (22). The oscillating tube (22) is connected to the oscillating tube (22); a pair of magnetic steels (23) are axially symmetrically distributed on opposite sides of the outside of the oscillating tube (22) in the long axis direction, and are tightly adhered to the side walls of the upper part of the oscillating tube (22); a pair of electromagnets are symmetrically supported on the outside of the pair of magnetic steels (23) and are on the same horizontal line as the center of the magnetic steels (23), and are used to drive the magnetic steels (23) to vibrate; the oscillation driving module (24) is connected to the pair of electromagnets, and is used to output a dynamic driving signal to the electromagnets; the high-precision frequency measurement module (25) The particle measuring unit comprises a particle flight chamber (12), a first high efficiency filter (5), a first flow pump (9), a first current limiting component (2), a second high efficiency filter (4), a second flow pump (10), a second current limiting component (3), a particle measuring chamber (13), a first light trap (20), a first laser (14), a second light trap (21), a second laser (15), a first photoelectric sensing unit (16), a first filter (18), a second photoelectric sensing unit (17), a second filter (19), a temperature and humidity sensor (28), a first mass flow controller (6) and a third flow pump (8);The particle flight chamber (12) is located in the shell (1), and is divided into an air intake section, a mixing section and an acceleration section from top to bottom. The air intake section is a conical structure, the small diameter end of which is located at the upper part, and a chamber inlet A is opened. The mixing section is a cylindrical structure, and a chamber inlet B and a chamber inlet C are opened on two opposite sides of the upper part. The acceleration section is a conical structure, the small diameter end of which is located at the upper part, and a chamber outlet A is opened. The chamber inlet A is connected to the first air inlet (32) through a connecting pipe A (40), the chamber inlet B is connected to the second air inlet (33) through an air flow branch (38), and the chamber inlet C is connected to the second air inlet (33) through an air flow branch (38). The particle measurement chamber (13) is connected to the third air inlet (34) through the air flow branch 2 (39); the air outlet ends of the air flow branch 1 (38) and the air flow branch 2 (39) are arranged at a relative angle to the central axis of the particle flight chamber (12) and are not perpendicular to the particle flight chamber (12); the first high-efficiency filter (5), the first flow pump (9), and the first flow limiting component (2) are connected in series to the air flow branch 1 (38) from the air inlet side to the air outlet side; the second high-efficiency filter (4), the second flow pump (10), and the second flow limiting component (3) are connected in series to the air flow branch 2 (39) from the air inlet side to the air outlet side; the particle measurement chamber (13) is located in the housing (1) and below the particle flight chamber (12); the top and bottom of the particle measurement chamber (13) are relatively provided with a chamber outlet D and a chamber outlet B which are connected to the inner cavity thereof, wherein; The chamber outlet D is connected to the chamber outlet A, and the chamber outlet B is connected to the first gas outlet (36) through a connecting pipe B (41); the first light trap (20) and the first laser (14) are mounted on the side walls on the left and right sides of the upper inner cavity of the particle measurement chamber (13); the second light trap (21) and the second laser (15) are mounted on the side walls on the left and right sides of the lower inner cavity of the particle measurement chamber (13); the first photoelectric sensing unit (16) is mounted on the rear side wall of the particle measurement chamber (13) and is arranged at 90 degrees to the first laser (14), and at the same time, it has the same center height as the first laser (14); the first filter (18) is coaxially supported on the front side of the first photoelectric sensing unit (16); the second photoelectric sensing unit (17) is mounted on the front side wall of the particle measurement chamber (13) and is arranged at 90 degrees to the second laser (15), and at the same time, it has the same center height as the first laser (14); the center heights of the first and second filters (19) are the same; the second filter (19) is coaxially supported on the front side of the second photoelectric sensing unit (17); the temperature and humidity sensor (28) is arranged in the particle measurement chamber (13); the first mass flow controller (6) and the third flow pump (8) are sequentially connected in series to the connecting pipe B (41) from the air inlet side to the air outlet side; the data processing unit comprises a data acquisition and control module (29) and a data processing system (30); the data acquisition and control module (29) is respectively connected to the first mass flow controller (6), the second mass flow controller (7), the first flow pump (9), the second flow pump (10), the third flow pump (8), the fourth flow pump (11), the first laser (14), the second laser (15), the first photoelectric sensing unit (16), the second photoelectric sensing unit (17), the temperature and humidity sensor (28), the oscillation drive module (24) and the high-precision frequency measurement module (25); The data processing system (30) is connected to the data acquisition and control module (29), and has a particle size distribution measurement model and a dual-drive dust concentration calibration model based on physical drive and data drive; the display screen (31) is embedded on the outer surface of the housing (1) and connected to the data acquisition and control module (29); It is characterized by comprising the following steps: S1: establishing communication between the data acquisition and control module (29) and the first mass flow controller (6), the second mass flow controller (7), the first flow pump (9), the second flow pump (10), the third flow pump (8), the fourth flow pump (11), the first laser (14), the second laser (15), the first photoelectric sensing unit (16), the second photoelectric sensing unit (17), the oscillation driving module (24), the high-precision frequency measurement module (25) and the data processing system (30); S2: In a non-intake state, the oscillation driving module (24) is controlled to output a dynamic driving signal to the electromagnet, and the dynamic driving signal is used to drive the electromagnet excitation coil to energize so that the magnetic flux increases and acts on the magnetic steel (23) on the oscillation tube (22), thereby driving the oscillation tube (22) to oscillate with equal amplitude, and the high-precision frequency measurement module (25) is used to sense the change in magnetic field intensity to achieve high-precision measurement of the initial state oscillation frequency, thereby obtaining the initial state oscillation frequency of the oscillation tube (22); S3: Controlling the first flow pump (9), the second flow pump (10), the third flow pump (8), the first mass flow controller (6), the first laser (14), the second laser (15), the first photoelectric sensing unit (16) and the second photoelectric sensing unit (17) to start working, and ensuring that the gas path inside the particle measurement unit is in a normal working flow state; synchronously, controlling the fourth flow pump (11), the second mass flow controller (7) and the oscillation driving module (24) to start working, and ensuring that the gas path inside the oscillation measurement unit is in a normal working flow state; A third flow pump (8) is used to provide negative pressure to the particle flight chamber (12) via the particle measurement chamber (13), and a first mass flow controller (6) is used to adjust the flow rate of the airflow passing through the connecting pipe B (41), so that the dust-laden airflow enters the air intake section of the particle flight chamber (12) through the connecting pipe A (40), and then flows to the acceleration section through the mixing section; A first flow pump (9) and a second flow pump (10) are used to provide negative pressure to the airflow branch one (38) and the airflow branch two (39), respectively, and the dust-containing airflow entering the airflow branch one (38) and the airflow branch two (39) is filtered by the first high-efficiency filter (5) and the second high-efficiency filter (4) to form two clean airflows. At the same time, the airflows in the airflow branch one (38) and the airflow branch two (39) are limited by the first flow limiting component (2) and the second flow limiting component (3), respectively, so that the two clean airflows after the flow limiting are relatively entered into the particle flight chamber (12) and meet in the mixing section of the particle flight chamber (12), and a layer of sheath flow with a higher flow rate is formed around the dust-containing airflow, so that the particles in the dust-containing airflow are not dispersed and can flow to the acceleration section along a path that is approximately straight. Under the acceleration effect of the acceleration section, the dust-containing airflow forms a sample flow with better linearity. Under the action of the negative pressure, the sample flow flows through the particle measurement chamber (13) and is filtered by the first flow limiting component (2). An air outlet (36) is used to discharge to the outside; in this process, two laser beams emitted by a first laser (14) and a second laser (15) arranged on the right side of the particle measurement chamber (13) are used to focus on the mixed airflow passing through the center of the particle measurement chamber (13), and the first light trap (20) and the second light trap (21) arranged on the left side of the particle measurement chamber (13) are used to absorb the two transmitted light beams irradiated on the left side plate of the particle measurement chamber (13), and the first filter (18) and the second filter (19) are used to filter the two scattered light beams refracted to the rear side plate of the particle measurement chamber (13), and then the first photoelectric sensing unit (16) and the second photoelectric sensing unit (17) are used to receive the two scattered light signals generated by the dust particles, and further convert the two scattered light signals into two electrical signals; at the same time, the temperature and humidity signals in the particle measurement chamber (13) are collected in real time by the temperature and humidity sensor (28), and temperature and humidity data are obtained based on the temperature and humidity signals; A fourth flow pump (11) is used to provide negative pressure to the connecting pipe D (43) via the connecting pipe C (42), the oscillating pipe (22) and the filter membrane holder (26). At the same time, the air flow rate passing through the connecting pipe C (42) is adjusted by a second mass flow controller (7) so that the dust-laden air flow passes through the filter membrane, the filter membrane holder (26) and the oscillating pipe (22) in sequence and is discharged to the outside through the second air outlet (37); at the same time, a high-precision frequency measurement module (25) is used to monitor the frequency change of the oscillating pipe in real time to obtain the oscillation frequency of the oscillating pipe (22) in the detection state; S4: The data processing system (30) calculates the mass of dust deposited on the filter membrane according to the change of the oscillation frequency, and calculates the dust concentration prediction value A in the test environment in combination with the flow rate of the second mass flow controller (7), and at the same time, sends the dust concentration prediction value A to the display (31) for real-time display; At the same time, the data processing system (30) filters, amplifies and digitizes the two electrical signals based on the built-in particle size distribution measurement model, extracts features from the processed signals, and then obtains particle size distribution data of the particles through a particle size distribution algorithm; further, the data processing system (30) uses the built-in dual-drive dust concentration calibration model to process the particle size distribution data and the temperature and humidity data. Specifically, the particle size distribution data and the temperature and humidity data are first input into a model layer including a parameterized physical model and a machine learning model. The parameterized physical model uses the determined model parameters and the temperature and humidity data to correct the particle size distribution data, and performs secondary statistical classification on the corrected particle size distribution data to obtain different particle size channel data, and then uses the Mie scattered light intensity calculation value of the particles in different channels to obtain the total scattered light intensity value, and uses the total scattered light intensity value, the divided light intensity value and the calculated light intensity value to obtain the total scattered light intensity value. The dust concentration prediction value B is calculated based on the calculated value of the scattered light intensity and the flow rate fed back by the first mass flow controller (6); after receiving the corrected particle size distribution data and the temperature and humidity data, the machine learning model uses the model parameters to predict the dust concentration and obtains the dust concentration prediction value C; the dust concentration prediction value B and the dust concentration prediction value C are then input into the fusion layer. When the difference between the dust concentration prediction value B and the dust concentration prediction value C exceeds the difference threshold, the fusion layer adjusts the weight, assigns a higher weight to the parameterized physical model, and adds the dust concentration prediction value B to the machine learning In the historical database of the model, the machine learning model is updated and recalculated according to the updated historical database, and a new dust concentration prediction value B and a new dust concentration prediction value C are output to the fusion layer until the difference between the received dust concentration prediction value B and the dust concentration prediction value C is within the difference threshold. The fusion layer then uses a neural network fusion algorithm to process the input data. During the processing of the input data through multiple neural network layers, each layer converts its input through an activation function to capture and extract key feature information of the input data. The neural network effectively combines information from different models by adjusting the weights and biases of each layer to output a dust concentration prediction value D. Further, the data processing system (30) compares the dust concentration prediction value A with the dust concentration prediction value D, and calculates the partial derivative of the loss function with respect to the output of the fusion layer. Then, the chain rule is used to calculate the model parameter gradient for the physical model parameters, the machine learning model parameters and the fusion layer parameters, and the optimization algorithm is used to update the model parameters according to the calculated gradient. Finally, the fusion layer performs a comprehensive calculation on the outputs of the parameterized physical model and the machine learning model to obtain the dust concentration prediction value. S5: Control the fourth flow pump (11), the second mass flow controller (7) and the oscillation drive module (24) to stop working, and continue to maintain the normal working state of the first flow pump (9), the second flow pump (10), the third flow pump (8), the first mass flow controller (6), the first laser (14), the second laser (15), the first photoelectric sensing unit (16) and the second photoelectric sensing unit (17), so that the particle measurement unit works alone for a period of time, and outputs the obtained dust concentration prediction value to the display screen (31) for real-time display; S6: Control the fourth flow pump (11), the second mass flow controller (7) and the oscillation drive module (24) to restart operation, and update the model parameters according to the method in S4; S7: Repeat S3 to S6, and use the periodic dust concentration measurement data of the oscillation measurement unit to continuously optimize the model parameters. Finally, the fusion layer outputs the final predicted value of the dust concentration and outputs it to the display screen (31) for real-time display.

2. According to claim 1, a dust concentration monitoring method based on physical and data dual drive is characterized in that: The fourth air inlet (35) and the first air inlet (32) are both opened on the top plate of the shell (1); the second air outlet (37) and the first air outlet (36) are both opened on the bottom plate of the shell (1); the first air inlet (32) and the first air outlet (36) are relatively distributed; the fourth air inlet (35) and the second air outlet (37) are relatively distributed; the second air inlet (33) is opened on the left plate of the shell (1); the third air inlet (34) is opened on the right plate of the shell (1); the second air inlet (33) and the third air inlet (34) are relatively distributed.

3. The dust concentration monitoring method based on physical and data dual drive according to claim 1 is characterized in that: The mixing section length accounts for a larger proportion than the intake section and the acceleration section length.

4. The dust concentration monitoring method based on physical and data dual drive according to claim 1 is characterized in that: The particle measurement chamber (13) is made of matte black or light-absorbing black material, and a non-adsorbent material with anti-adhesion properties is coated on the inner wall.

5. The dust concentration monitoring method based on physical and data dual drive according to claim 1 is characterized in that: The first laser (14) and the second laser (15) have different wavelengths, and the central wavelengths of the first filter (18) and the second filter (19) correspond to the wavelengths of the first laser (14) and the second laser (15), respectively.

Images