Methods, systems, and terminals for estimating environmental particulate matter concentration based on sensor data

Abstract

This invention relates to a method, system, and terminal for estimating environmental particulate matter concentration based on sensor data. The method includes three stages: modeling, evaluation, and application. These three stages, deployed on hardware devices, can be updated and invoked periodically. The modeling stage takes a training time series set and modeling configuration parameters as input and outputs candidate estimation models. The evaluation stage takes a test time series set as input, and after processing the candidate estimation models obtained in the modeling stage, outputs the test error. The optimal estimation model is then output based on the candidate estimation model with the smallest test error. The application stage takes the original time series as input, and after calculation by the optimal estimation model, outputs the estimated time series. By using the method of this application to process sensor data at the software level, the measurement capability of sensors for environmental particulate matter concentration can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of environmental data detection technology, and more specifically, to a method, system, and terminal for estimating environmental particulate matter concentration based on sensor data. Background Technology

[0002] The concentration level of air pollutants is an important indicator for environmental governance, among which particulate matter is mainly represented by PM2.5. 2.5 (fine particulate matter) and PM 10 Mainly inhalable particulate matter, which is closely related to environmental quality and human health;

[0003] Current status regarding PM 2.5 and PM 10 In addition to measurement methods based on precision instruments, such as the oscillating balance method and the beta-ray method commonly used by official monitoring stations, sensors based on the principle of light scattering have also been widely used for concentration monitoring. Although sensors still have a significant gap in accuracy and stability compared to precision instruments, their low cost and small size still provide favorable hardware support for deploying sensing systems with high spatiotemporal resolution.

[0004] Currently, mainstream sensors based on the principle of light scattering can achieve a certain correlation in response trend when measuring the concentration of environmental particulate matter compared with standardized precision instruments. However, under certain conditions, especially high humidity and large short-term humidity changes, the correlation between sensor measurement results and standardized precision instrument measurement results is poor.

[0005] One important reason is that current mainstream sensors based on the principle of light scattering typically calculate the intensity of the light signal scattered by suspended particles in the detection chamber, ignoring the particle size. Some sensor models, at the factory, simply use standard instruments to analyze the obtained scattered signal against the PM2.5 concentration. 2.5 The concentration was calibrated, but for PM... 10 Concentration is directly measured by PM 2.5 The concentration is calculated using a simple linear mapping (multiplied by a coefficient and then added to a numerical value). In this case, the PM2.5 concentration output by this type of particulate sensor is... 10 Concentration is just a relation to PM 2.5 Concentration-highly correlated sequences (as per the instruction manual) Figure 1 As shown, it does not truly reflect environmental PM levels under many conditions. 10 The true concentration.

[0006] In actual working conditions, changes in factors such as humidity can affect PM levels. 2.5 and PM 10The correlation between the trend and the actual environmental concentration weakens (as per the instruction manual). Figure 2 (as shown), which leads to measurement deviations in such particulate matter sensors.

[0007] Hardware improvements require particle size separation within the sensor, specifically for PM2.5. 2.5 and PM 10 The scattered and received signals are statistically analyzed separately. On the one hand, this will greatly increase the complexity of the structural design and the cost of the equipment; on the other hand, it is difficult to improve the performance of already deployed particulate matter sensors. Therefore, software improvements are needed to estimate and approximate environmental PM levels based on existing information. 10 The true concentration became a practical and feasible method;

[0008] In traditional calibration problems, this issue is considered a classic regression problem, where regression modeling establishes a functional mapping between the input (multiple) parameters and the output; methods such as multiple (non-)linear regression can be used. However, this problem actually presents several additional challenges, requiring optimization of the mechanism design to effectively improve the performance of regression modeling. These challenges include:

[0009] (1) Hardware varies from person to person, which affects the regression model and its parameters.

[0010] (2) Lack of standard metrology equipment to provide calibration data support.

[0011] (3) The sample representativeness is insufficient in the regression modeling process within a limited time.

[0012] To address the aforementioned challenges, an environmental PM based on sensor data is needed. 2.5 / PM 10 Concentration estimation methods effectively improve the sensor's ability to measure environmental particulate matter concentration at the software level. Summary of the Invention

[0013] The technical problem to be solved by the present invention is to provide a method for estimating environmental particulate matter concentration based on sensor data, an environmental particulate matter concentration estimation system based on sensor data, an environmental particulate matter concentration estimation terminal based on sensor data, and a computer-readable storage medium, in order to address the above-mentioned deficiencies of the prior art.

[0014] The technical solution adopted by this invention to solve its technical problem is:

[0015] A method for estimating environmental particulate matter concentration based on sensor data is constructed, comprising the following steps:

[0016] Step 1: Select the training and testing time periods;

[0017] Step 2: Extract the raw environmental data output by the sensor to generate training time series and test time series sets with hourly intervals; the training time series samples are from the training period, and the test time series samples are from the test period.

[0018] Step 3: Set the modeling configuration parameters and determine the structure of the regression model:

[0019] Multiple different fitting itemsets are selected, and the training time series data is input into the composition formula corresponding to the selected fitting itemsets to obtain multiple different fitting itemsets. Based on the different fitting itemsets and their corresponding fitting coefficient vectors, the structures of multiple candidate estimation models are obtained.

[0020] Step 4: Construct multiple matrices based on different sets of fitted items and their corresponding fitted coefficient vectors;

[0021] Step 5: Solve for the fitting coefficient vectors corresponding to the matrices based on the obtained multiple matrices and vector calculation formulas. Substitute the obtained multiple fitting coefficient vectors into the structure of the candidate estimation model to obtain multiple candidate estimation models.

[0022] Step 6: Input the test time series set into each candidate estimation model, test the error value of each candidate estimation model, and select the candidate estimation model with the smallest error value as the preferred estimation model;

[0023] Step 7: Input the second-level data from the sensor into the optimal estimation model to obtain the sensor's real-time estimated output value of the environmental particulate matter concentration.

[0024] The environmental particulate matter concentration estimation method based on sensor data described in this invention, wherein:

[0025] In the first step, the training time period is denoted as [T1, T2], and the testing time period is denoted as [T3, T4].

[0026] In the second step, the original PM output of the sensor 2.5 or PM 10 The time series is denoted as X(t); the humidity time series output by the humidity sensing module is denoted as Hum(t); the temperature time series output by the temperature sensing module is denoted as Tem(t); t represents a second-level time.

[0027] Training time series C train ={X1(h),Hum1(h),Tem1(h),R1(h)}, where:

[0028] X1(h): X(t) is the sequence X(h) after all the original samples are merged according to the hourly mean. The terms that satisfy h∈[T1,T2] are taken out to form X1(h);

[0029] Hum1(h): Hum(t) is the sequence Hum(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T1,T2] are taken out to form Hum1(h).

[0030] Tem1(h): Tem(t) is the sequence Tem(h) after all original samples are merged according to the hourly mean. The items that satisfy h∈[T1,T2] are taken out to form Tem1(h);

[0031] R1(h): h∈[T1,T2] is the reference calibration sequence of the training time series set;

[0032] Test timing set C test ={X2(h),Hum2(h),Tem2(h),R2(h)}, where:

[0033] X2(h): X(t) is the sequence X(h) after all the original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken out to form X2(h);

[0034] Hum2(h): Hum(t) is the sequence Hum(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken to form Hum2(h);

[0035] Tem2(h): Tem(t) is the sequence Tem(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken out to form Tem2(h);

[0036] R2(h): h∈[T3,T4] is the reference calibration sequence of the test timing set.

[0037] The environmental particulate matter concentration estimation method based on sensor data described in this invention, wherein in the third step, the fitting itemset FI includes the first-order fitting itemset FI0 and the delayed fitting itemset FI0. d and supplementary fitting itemset FI c Multiple in;

[0038] A minimum set of fitted terms: The minimum number of fitted terms that the model needs to include, defined as:

[0039] FI0={X(h),Hum(h),Tem(h),1};

[0040] The first-order fitted itemset corresponds to the most basic multiple linear regression modeling:

[0041] E(h)=a*X(h)+b*Hum(h)+c*Tem(h)+d;

[0042] Delayed fitting itemset FI dBased on the fitted itemset FI0, the previous historical data is added to the fitted itemset, defined as:

[0043] FI d ={X(h),X(h-1)…Hum(h),Hum(h-1)…Tem(h),Tem(h-1)…1};

[0044] The regression modeling corresponding to the time-delayed fitting itemset is as follows:

[0045] E(h)=a0*X(h)+b0*Hum(h)+b1*Hum(h-1)+c0*Tem(h)+c1*Tem(h-1)+d;

[0046] Supplementing the fitted itemset FI c : In the time-delay fitting itemset FI d Based on this, the fitting effect in detail is improved by adding a quadratic term, which includes the square of any first-order term itself and / or the product of two different first-order terms;

[0047] The alternative model M is used with its fitted itemset FI and fitted coefficient vector B = [b1, b2, b3, ..., bb2]. L ] T Let L = |FI| represent the number of terms in FI, denoted as M. FI,B ; Let FI be denoted as FI = {a i (h)|1≤i≤L,L=|FI|},Alternative model M FI,B The structure is as follows:

[0048]

[0049] The environmental particulate matter concentration estimation method based on sensor data described in this invention, wherein in the fourth step:

[0050] Parameter estimation is performed under the condition of least squares fitting. A matrix is ​​constructed based on the fitted itemset FI and the fitted coefficient vector B, resulting in an overdetermined system of equations:

[0051] AB = R;

[0052] in:

[0053]

[0054] B = [b1, b2, ..., bb] L ] T ;

[0055] R = [R1(h), R1(h-1)...R1(hN)] T ;

[0056] The values ​​in matrix A and vector R can both be derived from the self-trained time series set C. train The corresponding sequence in the sequence can be obtained directly or by calculation.

[0057] The environmental particulate matter concentration estimation method based on sensor data described in this invention, wherein in the fifth step:

[0058] The formula for calculating vector B is:

[0059] B = (A T A) -1 A T R;

[0060] Substitute vector B into the candidate model M FI,B From the structural formula, we obtain the alternative estimation model M under the condition of the fitting itemset FI. FI,B .

[0061] The environmental particulate matter concentration estimation method based on sensor data described in this invention, wherein in the sixth step:

[0062] Under the condition of applying the fitted itemset FI, the model MFI,B is obtained on the training time set Ctrain, and the training time set C is calculated. train The estimated output E(h) is denoted as:

[0063] E1(h)=M FI,B (X1(h),Hum1(h),Tem1(h));

[0064] Model M FI,B In test timing set C test The estimated output is:

[0065] E2(h)=M FI,B (X2(h),Hum2(h),Tem2(h));

[0066] Then model M can be defined. FI,B In test timing set C test The error on is Err(M) FI,B C test The calculation formula is as follows:

[0067]

[0068] Where P is the current model M FI,B and the maximum number of test time series that Ctest can support; w(·) is the weighting function, defined as:

[0069] Represents the probability density function of the R2(h) sequence;

[0070] If there are k candidate estimation models, the preferred estimation model is the one that minimizes the error on the test time series set Ctest:

[0071] arg min Err(M FI(k),B(k) C test ).

[0072] The environmental particulate matter concentration estimation method based on sensor data described in this invention, wherein in the seventh step:

[0073] By replacing the hourly data during training and testing in the optimal estimation model with second-level data from the sensor, we can obtain the sensor's real-time estimate of environmental particulate matter concentration.

[0074] E PM (t)=M FI,B (X(t),Hum(t),Tem(t)).

[0075] An environmental particulate matter concentration estimation system based on sensor data is applied to the aforementioned environmental particulate matter concentration estimation method based on sensor data, comprising a data sampling unit, a data processing unit, and a data output unit.

[0076] The data sampling unit is used to select the training time period and the testing time period, and to sample the raw environmental data output by the sensor.

[0077] The data processing unit is used to perform the calculations as described in steps one through seven.

[0078] The data output unit is used to output the calculated data from the data processing unit.

[0079] An environmental particulate matter concentration estimation terminal based on sensor data includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0080] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0081] The beneficial effects of this invention are as follows: The method of this application is divided into three stages: modeling, evaluation, and application. After these three stages are deployed on the hardware device, they can be updated and called periodically. In the modeling stage, the input is the training time series set and modeling configuration parameters, and the output is the candidate estimation model. In the evaluation stage, the input is the test time series set. After the test time series set is processed by the candidate estimation model obtained in the modeling stage, the output is the test error. The optimal estimation model can be output based on the candidate estimation model with the smallest test error. In the application stage, the input is the original time series. After calculation by the optimal estimation model, the output is the estimated time series. By using the method of this application to process the sensor data at the software level, the sensor's ability to measure the concentration of environmental particulate matter can be effectively improved. Attached Figure Description

[0082] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0083] Figure 1 It is a certain type of light scattering particulate matter sensor PM 2.5 and PM 10 Trend chart (top line represents PM) 10 The bottom line is PM 2.5 );

[0084] Figure 2 PM in real environment 10 A diagram illustrating the difference between official site readings and sensor readings (thicker lines represent sensor data, thinner lines represent standard data from the official monitoring station);

[0085] Figure 3 This is a flowchart of a preferred embodiment of the environmental particulate matter concentration estimation method based on sensor data.

[0086] Figure 4 This is a schematic diagram of the logic principle of the environmental particulate matter concentration estimation method based on sensor data according to a preferred embodiment of the present invention.

[0087] Figure 5 This is a case study of the environmental particulate matter concentration estimation method based on sensor data, a preferred embodiment of the present invention.

[0088] Figure 6 This is a block diagram illustrating the principle of an environmental particulate matter concentration estimation system based on sensor data, according to a preferred embodiment of the present invention. Detailed Implementation

[0089] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, a clear and complete description will be provided below in conjunction with the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0090] The preferred embodiment of the present invention is a method for estimating environmental particulate matter concentration based on sensor data, such as... Figure 3 As shown, see also Figure 4 and 5 This includes the following steps:

[0091] S01: Select the training and testing time periods;

[0092] The training time period is denoted as [T1, T2], and the testing time period is denoted as [T3, T4]. The specified training time period and testing time period can be one before the other or partially overlap.

[0093] S02: Extract the raw environmental data output by the sensor to generate a training time series set and a test time series set with hourly intervals; the training time series set samples are from the training period, and the test time series set samples are from the test period.

[0094] PM raw output from sensor 2.5 or PM 10 The time series is denoted as X(t); the humidity time series output by the humidity sensing module is denoted as Hum(t); the temperature time series output by the temperature sensing module is denoted as Tem(t); t represents a second-level time.

[0095] Training time series C train ={X1(h),Hum1(h),Tem1(h),R1(h)}, where:

[0096] X1(h): X(t) is the sequence X(h) after all the original samples are merged according to the hourly mean. The terms that satisfy h∈[T1,T2] are taken out to form X1(h);

[0097] Hum1(h): Hum(t) is the sequence Hum(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T1,T2] are taken out to form Hum1(h).

[0098] Tem1(h): Tem(t) is the sequence Tem(h) after all original samples are merged according to the hourly mean. The items that satisfy h∈[T1,T2] are taken out to form Tem1(h);

[0099] R1(h): h∈[T1,T2] is the reference calibration sequence of the training time series set, which can be regarded as the real PM in the environment. 10 Concentration time series data can typically be obtained from hourly data released by official environmental monitoring stations in the area where the sensing devices are deployed.

[0100] Test timing set C test ={X2(h),Hum2(h),Tem2(h),R2(h)}, where:

[0101] X2(h): X(t) is the sequence X(h) after all the original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken out to form X2(h);

[0102] Hum2(h): Hum(t) is the sequence Hum(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken to form Hum2(h);

[0103] Tem2(h): Tem(t) is the sequence Tem(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken out to form Tem2(h);

[0104] R2(h): h∈[T3,T4] is the reference calibration sequence of the test time series set, which can be regarded as the real PM in the environment. 10 Concentration time series data can typically be obtained from hourly data released by official environmental monitoring stations in the area where the sensing devices are deployed.

[0105] S03: Set modeling configuration parameters and determine the structure of regression modeling:

[0106] Multiple different fitting itemsets are selected, and the training time series data is input into the composition formula corresponding to the selected fitting itemsets to obtain multiple different fitting itemsets. Based on the different fitting itemsets and their corresponding fitting coefficient vectors, the structures of multiple candidate estimation models are obtained.

[0107] The fitted itemset FI includes the first-order fitted itemset FI0 and the delayed fitted itemset FI0. d and supplementary fitting itemset FI c Multiple itemsets; it should be noted that there can be other forms besides these three types of fitting itemsets;

[0108] A minimum set of fitted terms: The minimum number of fitted terms that the model needs to include, defined as:

[0109] FI0={X(h),Hum(h),Tem(h),1};

[0110] The first-order fitted itemset corresponds to the most basic multiple linear regression modeling:

[0111] E(h)=a*X(h)+b*Hum(h)+c*Tem(h)+d;

[0112] Delayed fitting itemset FI d Based on the fitted itemset FI0, historical data from previous terms are added to the fitted itemset. This is mainly to reflect the gain of the estimation performance from the "change process," and is defined as follows:

[0113] FI d ={X(h),X(h-1)…Hum(h),Hum(h-1)…Tem(h),Tem(h-1)…1};

[0114] The regression modeling corresponding to the time-delayed fitting itemset is as follows:

[0115] E(h)=a0*X(h)+b0*Hum(h)+b1*Hum(h-1)+c0*Tem(h)+c1*Tem(h-1)+d;

[0116] Supplementing the fitted itemset FI c : In the time-delay fitting itemset FI d Based on this, the fitting effect in detail is improved by adding a quadratic term, which includes the square of any first-order term itself and / or the product of two different first-order terms;

[0117] For example:

[0118] When Fic = {X(h), Hum(h), Hum(h-1)Tem(h), Hum2(h-1), 1}, it means that the modeling requires the concentration and humidity at the current time, the humidity at the previous time and its quadratic term, and the product of the current temperature and the humidity at the previous time. The regression modeling corresponding to this time-delayed fitting itemset is as follows:

[0119] E(h)=a0*X(h)+b0*Hum(h)+c0*Hum(h-1)*Tem(h)+d0*Hum2(h-1)+e0;

[0120] From the above definition, we can know that F I0 Included in FI d FI d Included in FI c Here, these three are collectively referred to as the fitting itemset FI; within the framework of multiple linear regression, an alternative model M is defined by its fitting itemset FI and the fitting coefficient vector B = [b1, b2, b3…b ... L ] T Let L = |FI| represent the number of terms in FI, denoted as M. FI,B ; Let FI be denoted as FI = {a i(h)|1≤i≤L,L=|FI|},Alternative model M FI,B The structure is as follows:

[0121]

[0122] S04: Construct multiple matrices based on different sets of fitted items and their corresponding fitted coefficient vectors;

[0123] Parameter estimation is performed under the condition of least squares fitting. A matrix is ​​constructed based on the fitted itemset FI and the fitted coefficient vector B, resulting in an overdetermined system of equations:

[0124] AB = R;

[0125] in:

[0126]

[0127] B = [b1, b2, ..., bb] L ] T ;

[0128] R = [R1(h), R1(h-1)...R1(hN)] T ;

[0129] The values ​​in matrix A and vector R can both be derived from the self-trained time series set C. train The corresponding sequences in the dataset are obtained directly or by computation; the size of matrix A is (N+1)×L, generally N>>L, and N should be chosen as large as possible, but it must be ensured that the required data in A and R can be found in the training time set C. train Found;

[0130] S05: Based on the obtained multiple matrices and vector calculation formulas, solve the fitting coefficient vectors corresponding to the matrices, and substitute the obtained multiple fitting coefficient vectors into the structure of the candidate estimation model to obtain multiple candidate estimation models.

[0131] The formula for calculating vector B is:

[0132] B = (A T A) -1 A T R;

[0133] Substitute vector B into the candidate model M FI,B From the structural formula, we obtain the alternative estimation model M under the condition of the fitting itemset FI. FI,B ;

[0134] S06: Input the test time series set into each candidate estimation model, test the error value of each candidate estimation model, and select the candidate estimation model with the smallest error value as the preferred estimation model;

[0135] Under the condition of applying the fitting itemset FI, for the training time series set, all elements in matrix A can be calculated from X1(h), Hum1(h), and Tem1(h). Therefore, the fitting itemset FI can be applied to the training time series set C. train Model M is obtained above FI,B Calculate the training time series set C train The estimated output E(h) is denoted as:

[0136] E1(h)=M FI,B (X1(h),Hum1(h),Tem1(h));

[0137] Model M FI,B In test timing set C test The estimated output is:

[0138] E2(h)=M FI,B (X2(h),Hum2(h),Tem2(h));

[0139] Then model M can be defined. FI,B In test timing set C test The error on is Err(M) FI,B C test The calculation formula is as follows:

[0140]

[0141] Where P is the current model M FI,B and test timing set C test The maximum number that can be supported; w(·) is a weighting function, which is related to the probability distribution of the R2(h) sequence. The more reference calibration values ​​there are, the smaller the error should be, defined as:

[0142] Represents the probability density function of the R2(h) sequence;

[0143] If there are k candidate estimation models, the preferred estimation model is the one that minimizes the error on the test time series set Ctest:

[0144] arg min Err(M FI(k),B(k) C test );

[0145] S07: Input the second-level data from the sensor into the optimal estimation model to obtain the real-time estimated output value of the sensor for the concentration of particulate matter in the environment;

[0146] By replacing the hourly data during training and testing in the optimal estimation model with second-level data from the sensor, we can obtain the sensor's real-time estimate of environmental particulate matter concentration.

[0147] E PM (t)=M FI,B (X(t),Hum(t),Tem(t));

[0148] The method of this application consists of three stages: modeling, evaluation, and application. These three stages, once deployed on hardware, can be updated and invoked periodically. The modeling stage takes a training time series set and modeling configuration parameters as input and outputs a candidate estimation model. The evaluation stage takes a test time series set as input, and after processing the candidate estimation models obtained in the modeling stage, outputs the test error. The optimal estimation model is then output based on the candidate estimation model with the smallest test error. The application stage takes the original time series as input, and after calculation by the optimal estimation model, outputs the estimated time series. By using the method of this application to process sensor data at the software level, the sensor's ability to measure environmental particulate matter concentration can be effectively improved.

[0149] [Design Concept]

[0150] (1) In view of the individual differences in hardware, our method is based on providing an independent regression model and parameters for each set of sensing terminals, and can realize regression calibration function on both the terminal device and the software platform server.

[0151] (2) In view of the lack of standard metrology equipment to provide calibration data, our designed method will try to obtain local coarse-grained reference data under the actual working conditions of the hardware equipment as the data required for calibration modeling.

[0152] (3) In view of the insufficient representativeness of the sample during the regression modeling process within a limited time, our designed method can be updated and called, and the regression model and its parameters can be updated periodically, so as to make effective use of the limited sample data and improve the adaptability under different working conditions.

[0153] [Equipment Requirements]

[0154] Hardware requirements: A terminal device equipped with a particulate matter sensor capable of measuring PM2.5. 2.5 and PM 10 The device integrates a concentration sensor, a temperature sensor, and a humidity sensor; the terminal device supports a wireless data transmission module; and the terminal device supports a local data entry module.

[0155] Software requirements: This method can be implemented on both the device and server sides. When implemented on the device side, the terminal device equipped with the particulate matter sensor must have embedded programming capabilities and support remote program updates. When implemented on the server side, general software deployment requirements are sufficient.

[0156]

Example

[0157] Step 1: Specify the training time period [2021-09-01 00:00:00, 2021-09-30 23:59:59] and the testing time period [2021-10-01 00:00:00, 2021-10-10 23:59:59];

[0158] Step 2: Extract and generate training time series and test time series sets. After organizing them into hourly averages, each time series in the training time series set has 720 samples, and each time series in the test time series set has 240 samples.

[0159] C train ={X1(1)…X1(720); Hum1(1)…Hum1(720); Tem1(1)…Tem1(720); R1(1)…R1(720)}

[0160] C test ={X2(1)…X2(240); Hum2(1)…Hum2(240); Tem2(1)…Tem2(240); R2(1)…R2(240)}

[0161] Step 3: Set the modeling configuration parameters and determine the model structure.

[0162] Here, two sets of fitting terms are set up, corresponding to two models.

[0163] FI1={X(h),Hum(h),Hum(h-1),Tem(h),1}

[0164] FI2={X(h),Hum(h),Hum(h-1)Tem(h),1}

[0165] The corresponding model structure is

[0166] FI1:E(h)=b 11 *X(h)+b 12 *Hum(h)+b 13 *Hum(h-1)+b 14 *Tem(h)+b 15

[0167] FI2:E(h)=b 21 *X(h)+b 22 *Hum(h)+b 23 *Hum(h-1)*Tem(h)+b 24 *Tem(h)+b 25

[0168] Step 4: Construct the matrix

[0169] for

[0170]

[0171] for

[0172]

[0173] Step 5: Obtain alternative estimation models

[0174] The formula for calculation is as follows:

[0175] B1 = (A1) T A1) -1 A1 T R1 B2=(A2 T A2) -1 A2 T R2

[0176] Step Six: Model Evaluation

[0177] Using the test set

[0178] for

[0179]

[0180] for

[0181]

[0182] Calculation error

[0183]

[0184]

[0185] like Select Model Choose the preferred estimation model; otherwise, select the model. As the preferred estimation model

[0186] Step 7: Apply the model

[0187] If the model As a preferred estimation model, the sensor measures environmental PM2.5. 10 The output of the concentration estimate is:

[0188] E PM (t)=b 11 *X(t)+b 12 *Hum(t)+b 13 *Hum(t-1)+b 14 *Tem(t)+b15 ;

[0189] [Case Study Results] (See attached) Figure 5 As shown. Especially when relative humidity changes abruptly or is high, the proposed method enables the sensor to estimate ambient PM2.5 levels more effectively. 10 The concentration is closer to the measurement trend of local standard monitoring stations;

[0190] An environmental particulate matter concentration estimation system based on sensor data is applied to the aforementioned sensor data-based environmental particulate matter concentration estimation method, such as... Figure 6 As shown, it includes a data sampling unit 1, a data processing unit 2, and a data output unit 3;

[0191] Data sampling unit 1 is used to select the training time period and the testing time period, and to sample the raw environmental data output by the sensor.

[0192] Data processing unit 2 is used to perform calculations as described in steps one through seven;

[0193] Data output unit 3 is used to output the calculated data from the data processing unit;

[0194] By processing the sensor data through the methods described in this application, the sensor's ability to measure the concentration of environmental particulate matter can be effectively improved.

[0195] An environmental particulate matter concentration estimation terminal based on sensor data includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0196] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0197] The computer-readable storage medium in embodiments of the present invention may include any entity or device capable of carrying computer program code, a recording medium, such as ROM / RAM, disk, optical disk, flash memory, etc.

[0198] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for estimating environmental particulate matter concentration based on sensor data, characterized in that, Includes the following steps: Step 1: Select the training and testing time periods; Step 2: Extract the raw environmental data output by the sensor to generate training time series and test time series sets with hourly intervals; the training time series samples are from the training period, and the test time series samples are from the test period. Step 3: Set the modeling configuration parameters and determine the structure of the regression model: Multiple different fitting itemsets are selected, and the training time series data is input into the composition formula corresponding to the selected fitting itemsets to obtain multiple different fitting itemsets. Based on the different fitting itemsets and their corresponding fitting coefficient vectors, the structures of multiple candidate estimation models are obtained. The fitted itemset includes multiple itemsets from the first-order fitted itemset, the delayed fitted itemset, and the supplementary fitted itemset; Step 4: Construct multiple matrices based on different sets of fitted items and their corresponding fitted coefficient vectors; Step 5: Solve for the fitting coefficient vectors corresponding to the matrices based on the obtained multiple matrices and vector calculation formulas. Substitute the obtained multiple fitting coefficient vectors into the structure of the candidate estimation model to obtain multiple candidate estimation models. Step 6: Input the test time series set into each candidate estimation model, test the error value of each candidate estimation model, and select the candidate estimation model with the smallest error value as the preferred estimation model; Step 7: Input the second-level data from the sensor into the optimal estimation model to obtain the sensor's real-time estimated output value of the environmental particulate matter concentration.

2. The method for estimating environmental particulate matter concentration based on sensor data according to claim 1, characterized in that: In the first step, the training time period is denoted as [T1, T2], and the testing time period is denoted as [T3, T4]. In the second step, the original PM output of the sensor 2.5 or PM 10 The time series is denoted as X(t); the humidity time series output by the humidity sensing module is denoted as Hum(t); the temperature time series output by the temperature sensing module is denoted as Tem(t); t represents a second-level time. Training time series C train ={X1(h),Hum1(h),Tem1(h),R1(h)}, where: X1(h): X(t) is the sequence X(h) after all the original samples are merged according to the hourly mean. The terms that satisfy h∈[T1,T2] are taken out to form X1(h); Hum1(h): Hum(t) is the sequence Hum(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T1,T2] are taken out to form Hum1(h); Tem1(h): Tem(t) is the sequence Tem(h) after all original samples are merged according to the hourly mean. The items that satisfy h∈[T1,T2] are taken out to form Tem1(h); R1(h): h∈[T1,T2] is the reference calibration sequence of the training time series set; Test timing set C test ={X2(h),Hum2(h),Tem2(h),R2(h)}, where: X2(h): X(t) is the sequence X(h) after all the original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken out to form X2(h); Hum2(h): Hum(t) is the sequence Hum(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken to form Hum2(h); Tem2(h): Tem(t) is the sequence Tem(h) after all original samples are merged according to the hourly mean. The terms that satisfy h∈[T3,T4] are taken out to form Tem2(h); R2(h): h∈[T3,T4] is the reference calibration sequence of the test timing set.

3. The method for estimating environmental particulate matter concentration based on sensor data according to claim 2, characterized in that, FI is the fitted itemset, FI0 is the first-order fitted itemset, and FI d For the time-delay fitting itemset, FI c To supplement the fitted itemset; A minimum set of fitted terms: The minimum number of fitted terms that the model needs to include, defined as: FI0={X(h),Hum(h),Tem(h),1}; The first-order fitted itemset corresponds to the most basic multiple linear regression modeling: E(h)=a*X(h)+b* Hum(h)+c*Tem(h)+d; Delayed fitting itemset FI d Based on the fitted itemset FI0, the previous historical data is added to the fitted itemset, defined as: FI d ={X(h),X(h-1)…Hum(h),Hum(h-1)…Tem(h),Tem(h-1)…1}; The regression modeling corresponding to the time-delayed fitting itemset is as follows: E(h)=a0*X(h)+b0*Hum(h)+b1*Hum(h-1)+c0*Tem(h)+c1*Tem(h-1)+d; Supplement the fitted itemset FI c : In the time-delay fitting itemset FI d Based on this, the fitting effect in detail is improved by adding a quadratic term, which includes the square of any first-order term itself and / or the product of two different first-order terms; The alternative model M is used with its fitted itemset FI and fitted coefficient vector B=[b1,b2,b3…b L ] T Let L = |FI| represent the number of terms in FI, denoted as M. FI,B Let FI be denoted as FI = {a} i (h)|1≤i≤L,L=|FI|},Alternative model M FI,B The structure is as follows: 。 4. The method for estimating environmental particulate matter concentration based on sensor data according to claim 3, characterized in that, In the fourth step: Parameter estimation is performed under the condition of least squares fitting. A matrix is ​​constructed based on the fitted itemset FI and the fitted coefficient vector B, resulting in an overdetermined system of equations: AB=R; in: ； ； ； The values ​​in matrix A and vector R can both be derived from the self-trained time series set C. train The corresponding sequence in the sequence can be obtained directly or by calculation.

5. The method for estimating environmental particulate matter concentration based on sensor data according to claim 4, characterized in that, In the fifth step: The formula for calculating vector B is: ； Substitute vector B into the candidate model M FI,B From the structural formula, we obtain the alternative estimation model M under the condition of the fitting itemset FI. FI,B .

6. The method for estimating environmental particulate matter concentration based on sensor data according to claim 5, characterized in that, In the sixth step: Under the condition of applying the fitted itemset FI, the model MFI,B is obtained on the training time set Ctrain, and the training time set C is calculated. train The estimated output E(h) is denoted as: ； Model M FI,B In test timing set C test The estimated output is: ； Then model M can be defined. FI,B In test timing set C test The error on is Err(M) FI,B C test The calculation formula is as follows: ； Where P is the current model M FI,B and the maximum number of test time series that Ctest can support; w(·) is the weighting function, defined as: , Represents the probability density function of the R2(h) sequence; If there are k candidate estimation models, the preferred estimation model is the one that minimizes the error on the test time series set Ctest: 。 7. The method for estimating environmental particulate matter concentration based on sensor data according to claim 6, characterized in that, In the seventh step: By replacing the hourly data during training and testing in the optimal estimation model with second-level data from the sensor, we can obtain the sensor's real-time estimate of environmental particulate matter concentration. 。 8. An environmental particulate matter concentration estimation system based on sensor data, applied to the environmental particulate matter concentration estimation method based on sensor data as described in any one of claims 1-7, characterized in that, It includes a data sampling unit, a data processing unit, and a data output unit; The data sampling unit is used to select the training time period and the testing time period, and to sample the raw environmental data output by the sensor. The data processing unit is used to perform the calculations as described in steps one through seven. The data output unit is used to output the calculated data from the data processing unit.

9. A terminal for estimating environmental particulate matter concentration based on sensor data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images