WO2021051609A1 - 细颗粒物污染等级的预测方法、装置及计算机设备 - Google Patents

细颗粒物污染等级的预测方法、装置及计算机设备 Download PDF

Info

Publication number
WO2021051609A1
WO2021051609A1 PCT/CN2019/118403 CN2019118403W WO2021051609A1 WO 2021051609 A1 WO2021051609 A1 WO 2021051609A1 CN 2019118403 W CN2019118403 W CN 2019118403W WO 2021051609 A1 WO2021051609 A1 WO 2021051609A1
Authority
WO
WIPO (PCT)
Prior art keywords
environmental
data
particulate matter
pollution level
fine particulate
Prior art date
Application number
PCT/CN2019/118403
Other languages
English (en)
French (fr)
Inventor
陈娴娴
阮晓雯
徐亮
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021051609A1 publication Critical patent/WO2021051609A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Definitions

  • Haze weather caused by fine particles is even more harmful to human health than sandstorms.
  • Particles with a particle size of 10 microns or more will be blocked from the outside of the human nose; particles with a particle size between 2.5 microns and 10 microns can enter the upper respiratory tract, but some can be excreted through sputum, etc., and can also be passed into the nasal cavity
  • the inner fluff is blocked, which is relatively less harmful to human health; while the fine particles with a diameter of less than 2.5 microns have a diameter equivalent to one-tenth the size of a human hair and are not easy to be blocked.
  • the present application provides a method, device and computer equipment for predicting the pollution level of fine particulate matter, which can perform targeted detection of the pollution level of fine particulate matter based on real-time environmental data, and can solve the problem of insufficient accuracy of fine particulate matter analysis results.
  • a method for predicting the pollution level of fine particulate matter including:
  • the trained convolutional neural network model is used to determine the pollution level of the fine particulate matter within a preset time period in the future.
  • a device for predicting the pollution level of fine particulate matter comprising:
  • the screening module is used to screen out target analysis data whose correlation with fine particles meets preset standards
  • a creating module which is used to create a fine particle spatial distribution map according to the concentration value of the fine particle in a preset historical time period
  • a training module for training a convolutional neural network model based on the target analysis data and the fine particle spatial distribution map
  • the judging module is configured to use the trained convolutional neural network model to judge the pollution level of the fine particulate matter within a preset time period in the future.
  • a non-volatile readable storage medium having computer readable instructions stored thereon, and the computer readable instructions are executed by a processor to realize the above-mentioned method for predicting the pollution level of fine particulate matter.
  • a computer device including a non-volatile readable storage medium, a processor, and a computer readable storage medium that is stored on the non-volatile readable storage medium and can run on the processor. Instructions, when the processor executes the computer-readable instructions, the method for predicting the pollution level of fine particulate matter is realized.
  • this application provides a method, device, and computer equipment for predicting the pollution level of fine particulate matter. Compared with the currently commonly used prediction methods, this application can obtain more comprehensive environmental data from multiple channels, and The target analysis data whose correlation with the fine particles meets the preset standard is selected from the environmental data as the data to be analyzed. In order to make the concentration change characteristics of the fine particles more obvious, it is also based on the concentration value of the fine particles in the preset historical time period. Create a fine particle spatial distribution map, then use the target analysis data and the fine particle spatial distribution map to train the convolutional neural network model, and use the trained convolutional neural network model to determine the pollution level of the fine particulate matter in the future preset time period.
  • the deep convolutional neural network model is applied to the prediction of the pollution level of fine particles, which improves work efficiency while also meeting the requirements of real-time prediction; in addition, it can also enhance the scientificity and accuracy of prediction , To make the determined pollution level of fine particles more real and reliable.
  • FIG. 1 shows a schematic flowchart of a method for predicting the pollution level of fine particulate matter provided by an embodiment of the present application
  • FIG. 2 shows a schematic flow chart of another method for predicting the pollution level of fine particulate matter provided by an embodiment of the present application
  • FIG. 3 shows a schematic structural diagram of a device for predicting the pollution level of fine particulate matter provided by an embodiment of the present application
  • Fig. 4 shows a schematic structural diagram of another device for predicting the pollution level of fine particulate matter provided by an embodiment of the present application.
  • an embodiment of the present application provides a method for predicting the pollution level of fine particulate matter. As shown in FIG. 1, the method includes :
  • the preset standard is that the correlation between the environmental index data to be screened and the fine particles is greater than the preset threshold, and the size of the preset threshold can be set according to actual application requirements. The larger the preset threshold is set, the greater the screening The correlation between the target analysis data and the fine particulate matter is stronger.
  • the purpose of setting up environmental data correlation detection is to initially filter out environmental index data that is not strongly related to fine particles, retain environmental data related to fine particles, and proceed to the next step of detection Analysis can eliminate the influence of irrelevant factors on the test data while reducing the workload.
  • the selected target analysis data should be a time series data set, such as the past week (or daily, this is refined according to specific demand scenarios) Different frequency can be adopted as required) sulfur dioxide, nitrogen dioxide, fine particulate matter (PM2.5), inhalable particulate matter (PM10), carbon monoxide, etc., as well as the number of medical records per week, pathogen detection results, and the coverage of the hospital Portrait data such as the epidemic situation within the scope. There are more weather features, such as whether there is precipitation, etc., as well as Weibo index and so on. Determine these structural feature data sets as target analysis data as an important part of training the model.
  • the preset historical time period can be set according to the actual scene, and the purpose of creating the fine particle spatial distribution map is to make the concentration change feature of the fine particle more obvious and facilitate the extraction of image features.
  • the concentration data of fine particles in the past time period can be obtained through various channels such as data collection, data procurement, and cooperation and sharing, to assist in the prediction of future fine particle concentration data.
  • the high-dimensional spatial distribution heat map can be used to visualize the changes in the spatial concentration of fine particles in a preset historical time period, which improves the visualization of image features and at the same time visualizes this high-dimensional spatial distribution map with regional characteristics It also serves as an important part of the training model to assist in building a neural network.
  • the high-dimensional spatial distribution heat map is an image
  • the convolutional neural network model mainly includes the following layers of network design.
  • the first is the image_in layer, which processes the image data into an image format that can be processed by the convolutional neural network; the second is the convolutional layer.
  • a total of 2 convolutional layers are designed.
  • One layer contains 64 convolution kernels, the second layer contains 128 convolution kernels, followed by a layer of pooling (Pooling), the method of choice is Max Pooling; in addition, it is also designed Two fully connected layers;
  • Softmax is used to calculate Loss based on cross entropy, and iteratively update the network parameters to output the pollution level of fine particulate matter corresponding to a city in the next day.
  • the above two parts of data information can be combined and input into the convolutional neural network model. training.
  • the convolutional neural network model After the convolutional neural network model reaches the training standard, it can be applied to the detection of the pollution level of fine particles in actual application scenarios, and to predict the pollution level of fine particles in a preset time period in the future.
  • the method for predicting the pollution level of fine particulate matter in this embodiment more comprehensive environmental data can be obtained from multiple channels, and the target analysis data whose correlation with the particulate matter meets the preset standard can be selected from the environmental data as the data to be analyzed.
  • the spatial distribution map of fine particulate matter is created based on the concentration value of fine particulate matter in the preset historical time period, and then the convolutional neural network model is trained using the target analysis data and the spatial distribution map of fine particulate matter. , Use the trained convolutional neural network model to determine the pollution level of fine particles in the future preset time period.
  • the convolutional neural network model is applied to the prediction of the pollution level of fine particulate matter, which improves work efficiency while also meeting the requirements of real-time prediction; in addition, it can also enhance the scientificity and accuracy of prediction. Make the determined pollution level of fine particles more real and reliable.
  • the method includes :
  • some new URLs can be obtained by parsing the link information in the webpage, adding these URLs to the download queue, and repeating the data capture operation until the entire network is traversed or certain conditions are met It will stop until later.
  • a preset number of mycoplasma pneumonia-related indexes can be reviewed by disease experts in sentinel hospitals, such as fog, haze, allergies, bronchiolitis, pneumonia and other keywords in designated areas Daily Baidu index value to obtain relevant public opinion feature data set.
  • the depth-first traversal strategy can also be used to climb the weather characteristics of each time dimension, such as temperature/hour, sunshine time/day, precipitation/hour, cloudy/rainy/thunderstorm/clear weather image data sets.
  • a weather feature portrait can be created.
  • the obtained environmental indicator data can include: sulfur dioxide (SO2), nitrogen dioxide (NO2), fine particulate matter PM2.5, inhalable particulate matter (PM10), carbon monoxide (CO), etc., and can also include weekly Portrait data such as the number of medical records, pathogen detection results, and the epidemic situation within the coverage of the hospital.
  • these environmental index data are preliminarily constructed into weather feature portraits, which contain data information of various dimensions.
  • the feature image when using the feature image to predict the pollution level of fine particles, the feature image can be continuously improved based on the collected or detected effective environmental data, which can be selectively refined or fully automated to perform well.
  • the other feature portraits of ” are merged into the portrait data set of the main model to increase the dimensionality of the portraits so as to be used for qualitative analysis of environmental data in the future.
  • each environmental indicator in the feature portrait can be regarded as independent, and the correlation with PM2.5 can be calculated separately, and the environmental indicator whose correlation is greater than or equal to the first preset threshold
  • the data is determined to be the data to be analyzed.
  • measuring the correlation between two variables can be achieved by calculating the correlation coefficient.
  • the calculation formula of the correlation coefficient is:
  • r is the calculated correlation coefficient between a certain environmental indicator and PM2.5
  • x i is the single indicator value of the environmental indicator within a preset time period
  • y i is the single PM2.5 value corresponding to the environmental indicator in the same preset time period
  • It is the average value of PM2.5 in the preset time period.
  • the corresponding correlation intensity can be determined based on the calculated correlation value.
  • the correlation intensity can be divided into different intervals by setting different intensity thresholds. , Among them, the division interval can be divided into: extremely strong correlation, strong correlation, medium degree correlation, weak correlation, extremely weak correlation or no correlation, etc.
  • the absolute value of r is determined to be extremely strong correlation when the absolute value of r is between 0.8 and 1; the absolute value of r is determined to be strong correlation when the absolute value of r is between 0.6 and 0.8; and the absolute value of r is determined to be moderate when the absolute value is between 0.3 and 0.6.
  • Degree correlation when the absolute value of r is between 0.1 and 0.3, it is determined as a weak correlation; when the absolute value of r is between 0.0 and 0.1, it is determined as a very weak correlation or no correlation.
  • the environmental indicators in the weakly correlated, extremely weakly correlated, or non-correlated intervals in the feature image can be filtered out, and the first preset threshold can be set to 0.3, and the environmental indicators with correlations above the medium degree can be further extracted , And determine its corresponding environmental indicator as the first environmental indicator.
  • the first environmental indicator with a smaller saturation needs to be filtered out.
  • the calculated saturation can be compared with the second preset threshold.
  • a clustering algorithm based on unsupervised learning fills in environmental data with abnormal data in the second environmental indicator, so as to obtain target analysis data with complete data.
  • the detection and elimination of outliers can be carried out through the method of unsupervised learning-clustering, and all kinds of outliers can be marked or assigned as null values, and then the marked outliers can be re-acquired.
  • the specific data can be aggregated The value corresponding to the center of the class fills in the missing data.
  • unsupervised learning-clustering refers to the clustering of two categories into a new category each time, until all the clusters are clustered into a single category.
  • the algorithm is as follows: define each observation (row or unit) as a category; calculate the distance between each category and other categories; merge the two categories with the shortest distance into one category, so that the number of categories is reduced by one; repeat the above steps, Until the classes containing all observations are merged into a single class.
  • step 206 of the embodiment may specifically include: obtaining the fine particle concentration value corresponding to each time point in the preset historical time period; drawing the fine particle concentration Scatter plot of the value; based on the Kriging interpolation method to create a spatial distribution map of fine particles on the basis of the scatter plot.
  • the steps of creating a spatial distribution map of fine particles based on the scatter plot based on the Kriging interpolation method can be: filter out the target data points falling within the preset search range from the scatter plot; determine the target data point correspondence Mathematical function of spatial variation; according to mathematical function, assign values to data points falling on regular grid cells in order to obtain the spatial distribution map of fine particles.
  • the regular grid is usually a square, but can also be a regular grid such as a rectangle or a triangle.
  • the regular grid divides the regional space into regular grid cells, and each grid cell corresponds to a value. Mathematically, it can be expressed as a matrix, and in computer implementation, it is a 2-dimensional array.
  • the first one is the grid grid view.
  • the value of a grid cell is considered to be the elevation value of all points in it, that is, the elevation within the ground area corresponding to the grid cell is uniform. the height of.
  • point grid view which considers that the value of a grid cell is the elevation of the center point of the grid or the average elevation value of the grid cell.
  • Kriging also known as spatial local estimation or spatial local interpolation, is a method of unbiased optimal estimation of the value of regionalized variables in a limited area based on the theory of variogram and structural analysis. kind of method. In this method, the estimated value of any point is obtained by the linear combination of n valid sample values Z(x i) within the influence range of the point.
  • a corresponding pollution level can be configured for the fine particulate matter spatial distribution map, where the pollution level can be classified as first-level pollution based on the fine particulate matter spatial distribution.
  • first-level pollution based on the fine particulate matter spatial distribution.
  • second-degree pollution good
  • third-degree pollution light pollution
  • fourth-degree pollution medium pollution
  • fifth-degree pollution severe pollution
  • sixth-degree pollution severe pollution
  • the initial convolutional neural network model is created in advance according to the design needs.
  • the difference from the convolutional neural network model is: the initial convolutional neural network model is only initially created, it has not passed the model training, and has not met the preset standards, and
  • the convolutional neural network model mentioned in this application refers to a model that has reached a preset standard through model training and can be applied to predict the pollution level of fine particulate matter.
  • each environmental indicator is independently analyzed with the pollution level to determine the independent determination rule for the corresponding pollution level.
  • the determination rules corresponding to different environmental indicators determine the pollution levels of fine particles that are independent of each other, and the final determination result is determined based on the pollution levels of fine particles corresponding to each environmental index.
  • step 210 of the embodiment may specifically include: obtaining environmental data corresponding to environmental indicators; calculating environmental data corresponding to each pollution level based on the detection classification probability The category probability of the corresponding category; the pollution level with the highest probability of the corresponding category is determined as the first fine particulate pollution level determined according to environmental indicators.
  • Softmax is used in the multi-classification process. It maps the output of multiple neurons to the (0,1) interval, which can be understood as a probability. According to the Softmax function, the probability of each classification is solved, and the probability is the largest The classification is used as the current air pollution situation, and then the air pollution degree is classified. Among them, the formula of Softmax is:
  • the input of Softmax is the feature vector obtained from the fully connected layer. Assuming that the spatial distribution map of fine particles used for model training is I, the first-level pollution (excellent), the second-level pollution (good), and the third-level pollution can be discussed in this application. Pollution (light pollution), grade 4 pollution (medium pollution), grade 5 pollution (severe pollution), grade 6 pollution (severe pollution) 6 classification problems (types are represented by 1, 2, 3, 4, 5, and 6) , Various environmental indicators get the feature vector constructed with each pollution level before reaching the Softmax layer, that is to say, the feature vector is a 6*1 vector, that is, aj represents the jth value in this 6*1 vector ; The ak in the denominator represents the 6 values in the 6*1 vector, T represents the number of categories, and the range of j is also 1 to T.
  • the numerator is always a positive number, and the denominator is the sum of multiple positive numbers, so the denominator must also be a positive number, so Sj is a positive number and the range is (0,1). If you are not training the model, but testing the model, when a sample passes through the Softmax layer and outputs a T*1 vector, the index of the largest value in the vector will be used as the predicted label of the sample.
  • category 4 is the first level of fine particulate pollution determined by the model based on this environmental indicator, which corresponds to the fourth level of pollution (medium pollution).
  • the target analysis data contains five environmental indicators a, b, c, d, and e.
  • the pollution level of the first fine particulate matter determined based on the environmental indicator a is the first level pollution
  • the first fine particulate matter determined based on the environmental indicator b is the pollution level of particulate matter
  • the pollution level of particulate matter is secondary pollution
  • the pollution level of the first fine particulate matter determined according to environmental index c is level 3 pollution
  • the pollution level of the first fine particulate matter determined according to environmental index d is first pollution, which is determined according to environmental index e
  • the pollution level of the first fine particulate matter is the first-level pollution. Since the first fine-particulate pollution level of the largest determined level is the third-level pollution, it can be determined that the estimated pollution level of the fine particulate matter is the third-level pollution, which corresponds to light pollution.
  • the detection accuracy rate of the model is determined by calculating the ratio of the amount of successfully recognized data to the total amount of recognition.
  • the third preset threshold is the accuracy value used to determine whether the model meets the training standard.
  • the specific value can be determined according to the actual needs of the application scenario. Make settings.
  • the feature can be extracted again Vector and determine the judgment rules, replace and update the original judgment rules, and retrain the initial target detection model until it reaches the training standard.
  • the initial convolutional neural network model when it is determined that the initial convolutional neural network model has been trained, it can be used to predict the pollution level of fine particulate matter.
  • real-time environmental data can be input into the model, and the convolutional neural network model can be output
  • the pollution level of fine particles is determined as the pollution level of fine particles within a preset time period in the future.
  • the second fine particle pollution level is the pollution level corresponding to different environmental index data determined by the trained convolutional neural network model, and each second fine particle pollution level corresponds to the result of a single pollution index matching identification. .
  • the imported real-time environmental data contains five environmental indicators a, b, c, and d.
  • the pollution level of the first fine particulate matter determined according to the environmental indicator a is the first level pollution
  • the first fine particulate matter determined according to the environmental indicator b is the first level of pollution.
  • the pollution level of particulate matter is second-level pollution
  • the first fine particulate matter pollution level determined according to environmental index c is third-level pollution
  • the first fine particulate matter pollution level determined according to environmental index d is fourth-level pollution.
  • the pollution level of fine particulate matter is level four pollution, so it can be determined that the pollution level of fine particulate matter in the future preset time period is level four pollution, which corresponds to moderate pollution.
  • the specifics can also include : Output the pollution level of fine particles; if it is determined that there is fine particle pollution based on the pollution level, an alarm message will be output.
  • the alarm information may include text alarm information, picture alarm information, audio alarm information, video alarm information, light alarm information, vibration alarm information, etc.
  • the information about the pollution of fine particles can be output through various forms such as audio, video, or text.
  • the structured data set is used as an important data block of the training model.
  • the data block of the training model is used to assist in building a neural network. Then use the above two parts of data and information to train the convolutional neural network model.
  • the convolutional neural network model After the convolutional neural network model meets the training standards, import the real-time environmental data into the convolutional neural network model to predict the pollution of fine particulate matter in the future preset time period The predicted content is output and displayed.
  • the pollution level is determined to be light pollution, moderate pollution, heavy pollution, and severe pollution, an alarm message will be output to serve as a warning.
  • the convolutional neural network model is applied to the prediction of the pollution level of fine particulate matter, which improves work efficiency while also meeting the requirements of real-time prediction; in addition, it can also enhance the scientificity and accuracy of prediction. Make the determined pollution level of fine particles more real and reliable.
  • an embodiment of the present application provides a device for predicting the pollution level of fine particulate matter.
  • the device includes: a screening module 31, a creation module 32, Training module 33, determination module 34.
  • the screening module 31 is used to screen out target analysis data whose correlation with fine particles meets preset standards;
  • the creating module 32 is used to create a fine particle spatial distribution map according to the concentration value of the fine particle in a preset historical time period;
  • the training module 33 is used to train the convolutional neural network model based on the target analysis data and the spatial distribution map of fine particles;
  • the judging module 34 is configured to use the trained convolutional neural network model to judge the pollution level of fine particulate matter within a preset time period in the future.
  • the screening module 31 is specifically used to collect all environmental indicator data within a preset historical time period and create feature portraits;
  • the first environmental indicator whose correlation with fine particulate matter is greater than or equal to the first preset threshold is selected from the portrait;
  • the saturation of the environmental data corresponding to each first environmental indicator is calculated;
  • the first environmental data whose saturation is greater than or equal to the second preset threshold is calculated.
  • the environmental indicator is determined as the second environmental indicator; the clustering algorithm based on unsupervised learning fills in the environmental data with abnormal data in the second environmental indicator, so as to obtain target analysis data with complete data.
  • the creation module 32 is specifically used to obtain the fine particle concentration value corresponding to each time point in the preset historical time period; draw a scatter diagram of the fine particle concentration value; based on Kriging
  • the interpolation method creates a spatial distribution map of fine particles on the basis of the scatter plot; among them, when creating a spatial distribution map of fine particles on the basis of the scatter plot using the Kriging interpolation method, the creation module 32 is specifically used to start from the scatter plot.
  • the target data points that fall within the preset search range are selected; the mathematical function corresponding to the spatial change of the target data point is determined; the data points falling on the regular grid cell are assigned values according to the mathematical function to obtain the spatial distribution of fine particles Figure.
  • the training module 33 is specifically used to determine each pollution level contained in the spatial distribution map of fine particulate matter, and configure the level label; analyze various environmental indicators in the target data
  • the data and the spatial distribution map of fine particulate matter are input into the initial convolutional neural network model to extract various environmental indicator data and the feature vector of each pollution level; determine the judgment rule of each environmental indicator corresponding to the pollution level according to the feature vector; obtain the basis for judgment of each environmental indicator
  • If the accuracy rate is greater than or equal to the third preset threshold, it is determined that the initial convolutional neural network model has passed the training; if it is determined that the matching accuracy rate of the pollution level is less than the third preset threshold, it is determined that the initial convolutional neural network model has not
  • the training module 33 is specifically used to obtain environmental data corresponding to the environmental indicator; based on the detection classification probability, the category probability of the environmental data corresponding to each pollution level is calculated; The pollution level with the highest category probability is determined as the first fine particulate pollution level determined according to environmental indicators.
  • the determination module 34 is specifically used to input real-time environmental data into the trained convolutional neural network In the model, each second fine particulate matter pollution level determined by different environmental index data is obtained; the second finest particulate matter pollution level with the largest level is determined as the pollution level of fine particulate matter within a preset time period in the future.
  • the device further includes: an output module 35.
  • the output module 35 can be used to output the pollution level of fine particles
  • the output module 35 can also be used to output alarm information if it is determined that there is fine particulate pollution according to the pollution level.
  • an embodiment of the present application also provides a storage medium on which computer-readable instructions are stored.
  • the computer-readable instructions are executed by a processor, the above-mentioned figure is realized.
  • 1 and Figure 2 show the method of predicting the pollution level of fine particulate matter.
  • the technical solution of this application can be embodied in the form of a software product.
  • the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk, etc.), including several
  • the instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute the methods in each implementation scenario of the present application.
  • an embodiment of the present application also provides a computer device, which may be a personal computer, Server, network device, etc.
  • the physical device includes a storage medium and a processor; the storage medium is used to store computer-readable instructions; the processor is used to execute the computer-readable instructions to implement the above-mentioned details as shown in FIGS. 1 and 2 Prediction method of particulate pollution level.
  • the computer device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and so on.
  • the user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like.
  • the network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), and the like.
  • the structure of the computer device provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine some components, or arrange different components.
  • the non-volatile readable storage medium may also include an operating system and a network communication module.
  • the operating system is a program of physical equipment hardware and software resources used to predict the pollution level of fine particulate matter, and supports the operation of information processing programs and other software and/or computer-readable instructions.
  • the network communication module is used to implement communication between various components in the non-volatile readable storage medium, and communication with other hardware and software in the physical device.
  • this application can be implemented by means of software plus necessary general hardware platforms, or can be selected through correlation analysis, saturation calculation, and clustering algorithm.
  • the data block of the training model is used to assist in building a neural network.
  • the above two parts of data are combined to train the convolutional neural network model.
  • the real-time environmental data is imported into the convolutional neural network model to predict the pollution of fine particulate matter in the future preset time period
  • the predicted content is output and displayed.
  • the pollution level is determined to be light pollution, moderate pollution, heavy pollution, and severe pollution
  • an alarm message will be output to serve as a warning.
  • the convolutional neural network model is applied to the prediction of the pollution level of fine particles, which improves work efficiency while also meeting the requirements of real-time prediction; in addition, it can also enhance the scientificity and accuracy of prediction. Make the determined pollution level of fine particles more real and reliable.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Tourism & Hospitality (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Educational Administration (AREA)
  • General Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Game Theory and Decision Science (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

一种细颗粒物污染等级的预测方法、装置及计算机设备，涉及大气监测领域，可以基于实时环境数据对细颗粒物污染等级进行针对性检测，并且能够解决细颗粒物分析结果不够准确的问题。其中方法包括：筛选出与细颗粒物相关性符合预设标准的目标分析数据(101)；根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图(102)；基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型(103)；利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级(104)。该方法适用于对细颗粒物污染等级的预测。

Description

细颗粒物污染等级的预测方法、装置及计算机设备 技术领域
本申请要求与2019年9月20日提交中国专利局、申请号为201910894165.0、申请名称为“细颗粒物污染等级的预测方法、装置及计算机设备”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
背景技术
由细颗粒物造成的灰霾天气对人体健康的危害甚至要比沙尘暴更大。粒径10微米以上的颗粒物，会被挡在人的鼻子外面；粒径在2.5微米至10微米之间的颗粒物，能够进入上呼吸道，但部分可通过痰液等排出体外，另外也会被鼻腔内部的绒毛阻挡，对人体健康危害相对较小；而粒径在2.5微米以下的细颗粒物，直径相当于人类头发的十分之一大小，不易被阻挡。被吸入人体后会直接进入支气管，干扰肺部的气体交换，引发包括哮喘、支气管炎和心血管病等方面的疾病。因此，在一定的时空范围内PM2.5的空间分布及其扩散规律研究对社会人类健康生活产生巨大的正面效应。
目前科学界已经对PM2.5的重视程度在逐年递增，但对于检测体系内人员来说，由于工种、人力资源、技术能力等限制，很难进行有效的侦测建模；而工业界人群由于企业性质、重要数据资源限制等层面的局限性，很难有足够的资源支持必要的团队投入环境层面的研究，因此目前来说较多的研究者还存在定性的叙述研究及伪定量的研究调研中。很少有研究团队能基于真实大数据，并根据现实地理条件针对性地进行PM2.5进行污染侦测与建模，从而使分析结果不够准确，无法对PM2.5流行的现状作出实质性贡献与建设性意见。
发明内容
有鉴于此，本申请提供了一种细颗粒物污染等级的预测方法、装置及计算机设备，能够基于实时环境数据对细颗粒物污染等级进行针对性检测，并且能够解决细颗粒物分析结果不够准确的问题。
根据本申请的一个方面，提供了一种细颗粒物污染等级的预测方法，该方法包括：
筛选出与细颗粒物相关性符合预设标准的目标分析数据；
根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图；
基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型；
利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级。
根据本申请的另一个方面，提供了一种细颗粒物污染等级的预测装置，该装置包括：
筛选模块，用于筛选出与细颗粒物相关性符合预设标准的目标分析数据；
创建模块，用于根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图；
训练模块，用于基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型；
判定模块，用于利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级。
根据本申请的又一个方面，提供了一种非易失性可读存储介质，其上存储有计算机可读指令，计算机可读指令被处理器执行时实现上述细颗粒物污染等级的预测方法。
根据本申请的再一个方面，提供了一种计算机设备，包括非易失性可读存储介质、处理器及存储在非易失性可读存储介质上并可在处理器上运行的计算机可读指令，处理器执行所述计算机可读指令时实现上述细颗粒物污染等级的预测方法。
借由上述技术方案，本申请提供的一种细颗粒物污染等级的预测方法、装置及计算机设备，与目前普遍采用的预测方式相比，本申请可从多个途径获取较为全面的环境数据，并从环境数据中筛选出与细颗粒物相关性符合预设标准的目标分析数据作为待分析数据，为了使细颗粒物的浓度变化特征更为明显，故还基于预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图，之后利用目标分析数据以及细颗粒物空间分布图训练卷积神经网络模型，利用训练好的卷积神经网络模型判定未来预设时间段内细颗粒物的污染等级。在本申请中，将深度卷积神经网络模型应用到对细颗粒物污染等级的预测中，提高了工作效率的同时，也能满足预测实时性的要求；另外还可增强预测的科学性、准确性，使判定出的细颗粒物污染等级更加真实可靠。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本地申请的不当限定。在附图中：
图1示出了本申请实施例提供的一种细颗粒物污染等级的预测方法的流程示意图；
图2示出了本申请实施例提供的另一种细颗粒物污染等级的预测方法的流程示意图；
图3示出了本申请实施例提供的一种细颗粒物污染等级的预测装置的结构示意图；
图4示出了本申请实施例提供的另一种细颗粒物污染等级的预测装置的结构示意图。
具体实施方式
下文将参考附图并结合实施例来详细说明本申请。需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互结合。
针对目前不能基于实时环境数据对细颗粒物污染等级进行针对性检测，使分析结果不够准确的问题，本申请实施例提供了一种细颗粒物污染等级的预测方法，如图1所示，该方法包括：
101、筛选出与细颗粒物相关性符合预设标准的目标分析数据。
其中，预设标准为待筛选的环境指标数据需要与细颗粒物的相关性大于预设阈值，而预设阈值的大小可根据实际应用需求进行设定，预设阈值设定的数值越大，筛选出的目标分析数据与细颗粒物的相关性也越强。
在具体的应用场景中，设置环境数据相关性检测的目的为，初步滤除掉与细颗粒物相关性不强的环境指标数据，保留与细颗粒物存在相关性的环境数据，并进行下一步的检测分析，能够在减少工作量的同时，排除不相关因素对检测数据的影响。由于本申请是用于预测未来预设时间段内的细颗粒物浓度数据，故筛选出的目标分析数据应该是一个时序数据集，譬如过去每周(或者每天，这个按具体的需求场景的细化需求，可以采取不同的频次)二氧化硫、二氧化氮、细颗粒物(PM2.5)、可吸入颗粒物(PM10)、一氧化碳等浓度情况，以及每周病案个数、病原体检测结果、以及医院所覆盖的范围内的疫情情况等画像数据。还有更多的天气特征，譬如是否存在降水等情况，还有微博指数等等。将这些结构性特征数据集确定为目标分析数据，作为用来训练模型的一块重要组成部分。
102、根据预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图。
其中，预设历史时间段可根据实际场景进行设定，创建细颗粒物空间分布图的目的是，可使细颗粒物的浓度变化特征更为明显、便于提取图像特征。
在具体的应用场景中，可以通过数据收集、数据采购、合作共享等多种途径获取过去时间段的细颗粒物浓度数据，以辅助对未来细颗粒物浓度数据的预测。本申请中，可通过高维空间分布热度图来可视化呈现预设历史时间段内细颗粒物的空间浓度变化状况，提高图像特征可视化的同时，还可将这个带有地区特征的高维空间分布图同样充当训练模型的一块重要组成部分，来辅助搭建神经网络网络。因为高维空间分布热图是图像，因此可通过设定卷积核、隐层数、神经元个数等，充分地提取高维空间分布热图的图像信息，由于最终希望预测的是每一个区域的未来一天(一周)的细颗粒物浓度，因此在考虑时序信息的同时，还需要考虑地域信息，而各个地点的浓度是有关联性的，这也是绘制空间分布图、并且要用卷积神经网络去提取信息的重要原因。通过对高维分布图的信息提取，可获取到另一块训练模型的数据块。
103、基于目标分析数据及细颗粒物空间分布图训练卷积神经网络模型。
其中，卷积神经网络模型主要包括以下几层网络设计，首先是image_in层，将图像数据处理成卷积神经网络能处理的图像格式；其次是卷积层，共设计2个卷积层，第一层包 含64个卷积核，第二层包含了128个卷积核，紧接着跟上了一层池化(Pooling)层，选择的方法是最大池化(Max Pooling)；另外还设计了两个完全连接层(fully connected layers)；最终通过softmax基于cross entropy计算Loss，并进行对网络参数的迭代更新，输出某市未来一天对应的细颗粒物污染等级。
相应的，基于实施例步骤101、102中的内容，在提取出目标分析数据及细颗粒物空间分布图对应的图像信息后，可将以上两部分数据信息进行合并，输入卷积神经网络模型中进行训练。
104、利用训练好的卷积神经网络模型判定未来预设时间段内细颗粒物的污染等级。
在具体的应用场景中，在卷积神经网络模型达到训练标准后，可将其应用于实际应用场景下细颗粒物的污染等级的检测，并预测未来预设时间段内细颗粒物的污染等级。
通过本实施例中细颗粒物污染等级的预测方法，可从多个途径获取较为全面的环境数据，并从环境数据中筛选出与细颗粒物相关性符合预设标准的目标分析数据作为待分析数据，为了使细颗粒物的浓度变化特征更为明显，故还基于预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图，之后利用目标分析数据以及细颗粒物空间分布图训练卷积神经网络模型，利用训练好的卷积神经网络模型判定未来预设时间段内细颗粒物的污染等级。在本申请中，将卷积神经网络模型应用到对细颗粒物污染等级的预测中，提高了工作效率的同时，也能满足预测实时性的要求；另外还可增强预测的科学性、准确性，使判定出的细颗粒物污染等级更加真实可靠。
进一步的，作为上述实施例具体实施方式的细化和扩展，为了完整说明本实施例中的具体实施过程，提供了另一种细颗粒物污染等级的预测方法，如图2所示，该方法包括：
201、收集预设历史时间段内的所有环境指标数据并创建特征画像。
在具体的应该用场景中，为了全面提取出对细颗粒物(PM2.5)具有影响力的环境数据，对于各个环境指标数据的获取，除了从智慧城市、城市卫健委、环保局获取记录数据的方式外，还可基于网络爬虫技术、深度优先遍历策略、微博指数爬虫等数据获取方式提取出更为深层次、广范围的环境数据。在基于网络爬虫技术获取环境指标数据时，网络爬虫系统会选择一些比较重要的、出度(网页中链出超链接数)较大的网站的URL作为种子URL集合，网络爬虫系统以这些种子集合作为初始URL，开始数据的抓取。在得到网页的内容将其存储后，再经过解析网页中的链接信息可以得到一些新的URL，将这些URL加入下载队列，如此反复进行数据抓取操作，直至遍历了整个网络或者满足某种条件后才会停止下来。在通过微博指数的爬虫时，可通过哨点医院的疾病专家审核得到的预设数量 个与支原体肺炎相关的指数，譬如雾、霾、过敏、细支气管炎、肺炎等关键词在指定地区的每天的百度指数数值，从而获取得到相关舆情特征数据集。另外，还可通过深度优先遍历策略爬取得到每个时间维度的天气特征，譬如温度/小时，日照时间/天，降水/小时，阴/雨/雷暴/晴等天气画像数据集。在获取到各个环境指标数据之后，可创建天气特征画像。例如，获取到的环境指标数据可包括：二氧化硫(SO2)、二氧化氮(NO2)、细颗粒物PM2.5、可吸入颗粒物(PM10)、一氧化碳(CO)等浓度情况，此外还可包括每周病案个数、病原体检测结果、以及医院所覆盖的范围内的疫情情况等画像数据。进而将这些环境指标数据初步构建成天气特征画像，天气特征画像中包含各个维度的数据信息。
相应的，在后续过程中，利用特征画像对细颗粒物污染等级进行预测时，还可基于收集到的或者检测出的有效环境数据不断完善特征画像，可选择性地提炼或完全自动化地将表现良好的其他特征画像合并入主模型的画像数据集来增大画像维度，以便日后用于环境数据的定性分析。
202、从特征画像中筛选出与细颗粒物相关性大于或等于第一预设阈值的第一环境指标。
对于本实施例，在进行相关性检测时，可将特征画像中各个环境指标视为独立，分别计算与PM2.5之间的相关性，将相关性大于或等于第一预设阈值的环境指标数据确定为待分析数据。
相应的，衡量两个变量之间的相关性可用相关系数的计算方法来实现。相关系数的计算公式为：
Figure PCTCN2019118403-appb-000001
其中，r为计算出的某一环境指标与PM2.5之间的相关系数，x i为该环境指标在预设时间段内的单个指标数值，
Figure PCTCN2019118403-appb-000002
为该环境指标在预设时间段内的指标平均值，y i为同一预设时间段内与环境指标对应的单个PM2.5数值，
Figure PCTCN2019118403-appb-000003
为PM2.5在该预设时间段内的平均值。上式用于衡量x与y的相关程度，显然r<＝1，r的值越大，说明x i对应的环境指标与PM2.5之间的相关性越高，反之，r的值越小，说明x i对应的环境指标与PM2.5相关性越低，当r＝0时，代表两者最不相关，即完全不相关。
在计算出各个环境指标与PM2.5之间的相关系数后，即可基于计算出的相关数值判定 对应的相关强度，具体可通过设定不同的强度阈值来将相关性强度划分为不同的区间，其中，划分区间可分为：极强相关、强相关、中等程度相关、弱相关、极弱相关或无相关等。例如，根据实际情况，将r的绝对值在0.8～1时确定为极强相关；将r的绝对值在0.6～0.8时确定为强相关；将r的绝对值在0.3～0.6时确定为中等程度相关；将r的绝对值在0.1～0.3时确定为弱相关；将r的绝对值在0.0～0.1时确定为极弱相关或无相关。之后可将特征画像中处于弱相关、极弱相关或无相关区间内的环境指标滤除，即可设定第一预设阈值为0.3，进一步提取得到相关性在中等程度相关以上的各个环境指标，并将其对应的环境指标确定为第一环境指标。
203、计算各个第一环境指标对应环境数据的饱和度。
其中，计算各个第一环境指标数据饱和度的方法可为：分别提取出各个第一环境指标对应的环境数据，将同一环境指标内完整无缺失的环境数据个数与环境数据总个数的比值确定为该环境指标对应的特征饱和度。例如，提取出二氧化硫这一环境指标的数据总个数为500，进一步确定其中完整无缺失的环境数据个数为450，则可计算出二氧化硫这一环境指标对应环境数据的饱和度为：450/500*100％＝90％。
204、将饱和度大于或等于第二预设阈值的第一环境指标确定为第二环境指标。
相应的，在计算出各个第一环境指标对应环境数据的饱和度后，需要将饱和度较小的第一环境指标滤除，具体可将计算出的饱和度与第二预设阈值进行比较，删除饱和度小于第二预设阈值的环境指标，保留饱和度较高的环境指标数据，以便进行进一步的分析。例如，设定第二预设阈值为0.7，则可将饱和度γ<0.7的环境指标剔除，只分析γ>＝0.7的环境指标。
205、基于非监督学习的聚类算法填充第二环境指标中数据异常的环境数据，以便获取得到数据完整的目标分析数据。
在具体的应用场景中，在获取得到第二环境指标后，还需要对第二环境指标对应的环境数据进行全方位的输出勘探和结构性补缺，进一步提取出完整的结构化数据集。具体可通过非监督学习-聚类的方法进行异常值的检测以及剔除，将各类异常值进行标记或赋值为空值，之后对标记的异常值进行重新获取，具体可根据数据所聚成的类中心对应的取值对缺失数据进行填充。
其中，非监督学习-聚类是指每一次把两类聚成新的一类，直到所有的类聚成单个类为止。算法如下：定义每个观测值(行或单元)为一类；计算每类和其他各类的距离；把距离最短的两类合并成一类，这样类的个数就减少一个；重复上述步骤，直至包含所有观测 值的类合并成单个的类为止。
206、根据预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图。
对于本实施例，在具体的应用场景中，为了创建出细颗粒物空间分布图，实施例步骤206具体可以包括：获取预设历史时间段内各个时间点对应的细颗粒物浓度值；绘制细颗粒物浓度值的散点图；基于克里格插值法在散点图的基础上创建细颗粒物空间分布图。其中，基于克里格插值法在散点图的基础上创建细颗粒物空间分布图的步骤可为：从散点图中筛选出落在预设搜索范围内的目标数据点；确定目标数据点对应空间变化的数学函数；依据数学函数为落在规则格网单元上的数据点赋值，以便获取得到细颗粒物空间分布图。
其中，规则格网，通常是正方形，也可以是矩形，三角形等规则网格。规则网格将区域空间切分为规则的格网单元，每个格网单元对应一个数值。数学上可以表示为矩阵，在计算机实现中则是一个2维数组。对于每个网格的数值有两种不同的解释，第一种：格网栅格观点，认为格网单元的数值是其中所有点的高程值，即格网单元对应的地面面积内高程是均一的高度。第二种：点栅格观点，认为网格单元的数值是网格中心点的高程或该网格单元的平均高程值。计算任何不是网格中心的数据点的高程值，使用周围4点的高程值，可使用克里格插值法。克里格法(Kriging)，也称空间局部估计或空间局部插值，是建立在变异函数理论及结构分析的基础上，在有限区域内对区域化变量的取值进行无偏最优估计的一种方法。该方法将任一个点的估计值通过该点影响范围内的n个有效样本值Z(x i)的线性组合得到。
207、确定细颗粒物空间分布图中包含的各个污染等级，并配置等级标签。
对于本实施例，在具体的应用场景中，在创建细颗粒物空间分布图后，可为细颗粒物空间分布图配置对应的污染等级，其中，污染等级可根据细颗粒物空间分布情况划分为一级污染(优)、二级污染(良)、三级污染(轻度污染)、四级污染(中度污染)、五级污染(重度污染)、六级污染(严重污染)。
208、将目标分析数据中的各类环境指标数据及细颗粒物空间分布图输入初始卷积神经网络模型中，提取各类环境指标数据与各个污染等级的特征向量。
其中，初始卷积神经网络模型为预先根据设计需要创建的，与卷积神经网络模型的区别是：初始卷积神经网络模型只是初步创建完成，未通过模型训练，且未满足预设标准，而本申请中提及的卷积神经网络模型是指通过模型训练，已达到预设标准、可应用于对细颗粒物污染等级预测的模型。
209、依据特征向量确定各个环境指标对应污染等级的判定规则。
对于本实施例，由于不同环境指标对细颗粒物的影响存在差异，故将各个环境指标分别与污染等级进行独立分析，确定对应污染等级的独立判定规则，在进行细颗粒物污染等级判定时，可依据不同环境指标对应的判定规则确定出彼此独立的细颗粒物污染等级，并基于各个环境指标对应细颗粒物污染等级，确定最终的判定结果。
210、获取各个环境指标依据判定规则确定出的第一细颗粒物污染等级。
对于本实施例，在具体的应用场景中，为了确定出的第一细颗粒物污染等级，实施例步骤210具体可以包括：获取环境指标对应的环境数据；基于探测分类概率计算环境数据对应各个污染等级的类别概率；将对应类别概率最大的污染等级确定为根据环境指标判定出的第一细颗粒物污染等级。
其中，Softmax用于多分类过程中，它将多个神经元的输出，映射到(0,1)区间内，可以看成概率来理解，根据Softmax函数进而求解各个分类的概率，将概率最大的分类作为当前空气污染状况，进而对空气污染程度进行分类。其中，Softmax的公式为：
Figure PCTCN2019118403-appb-000004
Softmax的输入是从全连接层获取到的特征向量，假设用于模型训练的细颗粒物空间分布图是I，在本申请中可讨论一级污染(优)、二级污染(良)、三级污染(轻度污染)、四级污染(中度污染)、五级污染(重度污染)、六级污染(严重污染)6分类问题(类别用1，2，3，4，5，6表示)，各类环境指标在到达Softmax层之前就得到了与各个污染等级构建的特征向量，也就是说特征向量是一个6*1的向量，即aj表示这个6*1的向量中的第j个值；而分母中的ak则表示6*1的向量中的6个值，T表示类别数，j的范围也是1到T。因为e^x恒大于0，所以分子永远是正数，分母又是多个正数的和，所以分母也肯定是正数，因此Sj是正数，而且范围是(0,1)。如果现在不是在训练模型，而是在测试模型，那么当一个样本经过Softmax层并输出一个T*1的向量时，就会取这个向量中值最大的那个数的index作为这个样本的预测标签。
例如：假设获取到某项环境指标与各个污染等级构建的特征向量为[1，2，3，4，5，6]，那么经过Softmax层后得到对应的预测概率为[0.09，0.24，0.67，0.85，0.43，0.38]，这四个数字表示这个样本属于第1，2，3，4，5，6类的概率分别是0.09，0.24，0.67，0.85，0.43，0.38，因第四类概率最高，故可说明类别4为模型根据该项环境指标判定出的第一细颗粒物污染等级，即对应四级污染(中度污染)。
211、将级别最大的第一细颗粒物污染等级确定为预估细颗粒物污染等级。
例如，目标分析数据中共包含五个环境指标a、b、c、d、e，其中依据环境指标a确定出的第一细颗粒物污染等级为一级污染、依据环境指标b确定出的第一细颗粒物污染等级为二级污染，依据环境指标c确定出的第一细颗粒物污染等级为三级污染，依据环境指标d确定出的第一细颗粒物污染等级为一级污染，依据环境指标e确定出的第一细颗粒物污染等级为一级污染，由于确定级别最大的第一细颗粒物污染等级为三级污染，故可确定预估细颗粒物污染等级为三级污染，即对应轻度污染。
212、匹配预估细颗粒物污染等级与实际配置的污染等级。
对于本实施例，在对初始卷积神经网络模型过程中，为了检测模型分析识别的精度，故需要将预估细颗粒物污染等级与实际配置的污染等级进行比较，获取识别成功的数据量，并通过计算识别成功的数据量与总识别量的比值来确定模型的检测正确率。
213、若确定污染等级的匹配正确率大于或等于第三预设阈值，则判定初始卷积神经网络模型通过训练。
其中，第三预设阈值为用于判定模型是否达到训练标准的精度值，第三预设阈值越大，代表卷积神经网络模型的训练精度越高，具体数值可根据应用场景的实际需求来进行设定。
214、若确定污染等级的匹配正确率小于第三预设阈值，则判定初始卷积神经网络模型未通过训练，则更新各个环境指标的判定规则，以使初始目标检测模型通过训练。
对于本实施例，在具体的应用场景中，当判定初始卷积神经网络模型未通过训练时，说明依据特征向量确定出的各个环境指标对应污染等级的判定规则可能存在误差，故可重新提取特征向量并确定判定规则，替换更新原有的判定规则后，重新训练初始目标检测模型，直至其达到训练标准。
215、将实时环境数据输入到训练好的卷积神经网络模型中。
在具体的应用场景中，当确定初始卷积神经网络模型通过训练后，即可投入到对细颗粒物污染等级的预测中，具体可将实时环境数据输入到模型中，将卷积神经网络模型输出的细颗粒物污染等级确定为未来预设时间段内细颗粒物的污染等级。
216、获取由不同环境指标数据确定出的各个第二细颗粒物污染等级。
其中，第二细颗粒物污染等级为利用训练好的卷积神经网络模型确定出的不同环境指标数据对应的各个污染等级，每一个第二细颗粒物污染等级对应单独一项污染指标匹配识别出的结果。
217、将级别最大的第二细颗粒物污染等级确定为未来预设时间段内细颗粒物的污染等级。
例如，导入的实时环境数据中共包含五个环境指标a、b、c、d，其中依据环境指标a确定出的第一细颗粒物污染等级为一级污染、依据环境指标b确定出的第一细颗粒物污染等级为二级污染，依据环境指标c确定出的第一细颗粒物污染等级为三级污染，依据环境指标d确定出的第一细颗粒物污染等级为四级污染，由于确定级别最大的第一细颗粒物污染等级为四级污染，故可确定未来预设时间段内细颗粒物的污染等级为四级污染，即对应中度污染。
在具体的应用场景中，为了将判定出的细颗粒物的污染等级及时通知给相关工作人员，作为一种优选方式，在判定出未来预设时间段内细颗粒物的污染等级后，具体还可以包括：输出细颗粒物的污染等级；若根据污染等级判定存在细颗粒物污染，则输出报警信息。
其中，报警信息可包括文字报警信息、图片报警信息、音频报警信息、视频报警信息、灯光报警信息、震动报警信息等。可通过音频、视频、或文字等多种形式，将细颗粒物存在污染的信息输出。
通过上述细颗粒物污染等级的预测方法，可通过相关性分析、饱和度计算以及聚类算法筛选出相关性较强的环境数据，并进行全方位的输出勘探和结构性补缺，进而获取得到较为完整的结构化数据集，将其作为训练模型的一个重要数据块。为了使细颗粒物的浓度变化特征更为明显，还需要基于预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图，并将这个带有地区特征的高维空间分布图同样充当另一块训练模型的数据块，用于辅助搭建神经网络网络。之后利用以上两部分数据信息合并训练卷积神经网络模型，在卷积神经网络模型符合训练标准后，将实时的环境数据导入卷积神经网络模型中，预测未来预设时间段内细颗粒物的污染等级，并将预测出的内容输出显示，当确定污染等级为轻度污染、中度污染、重度污染、严重污染中的任意一项时，将会输出报警信息，起到警示的作用。在本申请中，将卷积神经网络模型应用到对细颗粒物污染等级的预测中，提高了工作效率的同时，也能满足预测实时性的要求；另外还可增强预测的科学性、准确性，使判定出的细颗粒物污染等级更加真实可靠。
进一步的，作为图1和图2所示方法的具体体现，本申请实施例提供了一种细颗粒物污染等级的预测装置，如图3所示，该装置包括：筛选模块31、创建模块32、训练模块33、判定模块34。
筛选模块31，用于筛选出与细颗粒物相关性符合预设标准的目标分析数据；
创建模块32，用于根据预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图；
训练模块33，用于基于目标分析数据及细颗粒物空间分布图训练卷积神经网络模型；
判定模块34，用于利用训练好的卷积神经网络模型判定未来预设时间段内细颗粒物的污染等级。
在具体的应用场景中，为了筛选出与细颗粒物相关性符合预设标准的目标分析数据，筛选模块31，具体用于收集预设历史时间段内的所有环境指标数据并创建特征画像；从特征画像中筛选出与细颗粒物相关性大于或等于第一预设阈值的第一环境指标；计算各个第一环境指标对应环境数据的饱和度；将饱和度大于或等于第二预设阈值的第一环境指标确定为第二环境指标；基于非监督学习的聚类算法填充第二环境指标中数据异常的环境数据，以便获取得到数据完整的目标分析数据。
相应的，为了创建出细颗粒物空间分布图，创建模块32，具体用于获取预设历史时间段内各个时间点对应的细颗粒物浓度值；绘制细颗粒物浓度值的散点图；基于克里格插值法在散点图的基础上创建细颗粒物空间分布图；其中，在利用克里格插值法在散点图的基础上创建细颗粒物空间分布图时，创建模块32，具体用于从散点图中筛选出落在预设搜索范围内的目标数据点；确定目标数据点对应空间变化的数学函数；依据数学函数为落在规则格网单元上的数据点赋值，以便获取得到细颗粒物空间分布图。
在具体的应用场景中，为了训练卷积神经网络模型，训练模块33，具体用于确定细颗粒物空间分布图中包含的各个污染等级，并配置等级标签；将目标分析数据中的各类环境指标数据及细颗粒物空间分布图输入初始卷积神经网络模型中，提取各类环境指标数据与各个污染等级的特征向量；依据特征向量确定各个环境指标对应污染等级的判定规则；获取各个环境指标依据判定规则确定出的第一细颗粒物污染等级；将级别最大的第一细颗粒物污染等级确定为预估细颗粒物污染等级；匹配预估细颗粒物污染等级与实际配置的污染等级；若确定污染等级的匹配正确率大于或等于第三预设阈值，则判定初始卷积神经网络模型通过训练；若确定污染等级的匹配正确率小于第三预设阈值，则判定初始卷积神经网络模型未通过训练，则更新各个环境指标的判定规则，以使初始目标检测模型通过训练。
相应的，为了获取各个环境指标依据对应的第一细颗粒物污染等级，训练模块33，具体用于获取环境指标对应的环境数据；基于探测分类概率计算环境数据对应各个污染等级的类别概率；将对应类别概率最大的污染等级确定为根据环境指标判定出的第一细颗粒物污染等级。
在具体的应用场景中，为了利用训练好的卷积神经网络模型判定未来预设时间段内细颗粒物的污染等级，判定模块34，具体用于将实时环境数据输入到训练好的卷积神经网络模型中；获取由不同环境指标数据确定出的各个第二细颗粒物污染等级；将级别最大的第 二细颗粒物污染等级确定为未来预设时间段内细颗粒物的污染等级。
相应的，为了将判定出的细颗粒物的污染等级直观显示并设置相应提醒，如图4所示，本装置还包括：输出模块35。
输出模块35，可用于输出细颗粒物的污染等级；
相应的，输出模块35，还可用于若根据污染等级判定存在细颗粒物污染，则输出报警信息。
需要说明的是，本实施例提供的一种细颗粒物污染等级的预测装置所涉及各功能单元的其它相应描述，可以参考图1至图2中的对应描述，在此不再赘述。
基于上述如图1和图2所示方法，相应的，本申请实施例还提供了一种存储介质，其上存储有计算机可读指令，该计算机可读指令被处理器执行时实现上述如图1和图2所示的细颗粒物污染等级的预测方法。
基于这样的理解，本申请的技术方案可以以软件产品的形式体现出来，该软件产品可以存储在一个非易失性存储介质(可以是CD-ROM，U盘，移动硬盘等)中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施场景的方法。
基于上述如图1、图2所示的方法，以及图3、图4所示的虚拟装置实施例，为了实现上述目的，本申请实施例还提供了一种计算机设备，具体可以为个人计算机、服务器、网络设备等，该实体设备包括存储介质和处理器；存储介质，用于存储计算机可读指令；处理器，用于执行计算机可读指令以实现上述如图1和图2所示的细颗粒物污染等级的预测方法。
可选地，该计算机设备还可以包括用户接口、网络接口、摄像头、射频(Radio Frequency，RF)电路，传感器、音频电路、WI-FI模块等等。用户接口可以包括显示屏(Display)、输入单元比如键盘(Keyboard)等，可选用户接口还可以包括USB接口、读卡器接口等。网络接口可选的可以包括标准的有线接口、无线接口(如蓝牙接口、WI-FI接口)等。
本领域技术人员可以理解，本实施例提供的计算机设备结构并不构成对该实体设备的限定，可以包括更多或更少的部件，或者组合某些部件，或者不同的部件布置。
非易失性可读存储介质中还可以包括操作系统、网络通信模块。操作系统是用于细颗粒物污染等级预测的实体设备硬件和软件资源的程序，支持信息处理程序以及其它软件和/或计算机可读指令的运行。网络通信模块用于实现非易失性可读存储介质内部各组件之间的通信，以及与该实体设备中其它硬件和软件之间通信。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到本申请可以借助软件加必要的通用硬件平台的方式来实现，也可通过相关性分析、饱和度计算以及聚类算法筛选出相关性较强的环境数据，并进行全方位的输出勘探和结构性补缺，进而获取得到较为完整的结构化数据集，将其作为训练模型的一个重要数据块。为了使细颗粒物的浓度变化特征更为明显，还需要基于预设历史时间段内细颗粒物的浓度值创建细颗粒物空间分布图，并将这个带有地区特征的高维空间分布图同样充当另一块训练模型的数据块，用于辅助搭建神经网络网络。之后将以上两部分数据信息合并训练卷积神经网络模型，在卷积神经网络模型符合训练标准后，将实时的环境数据导入卷积神经网络模型中，预测未来预设时间段内细颗粒物的污染等级，并将预测出的内容输出显示，当确定污染等级为轻度污染、中度污染、重度污染、严重污染中的任意一项时，将会输出报警信息，起到警示的作用。在本申请中，将卷积神经网络模型应用到对细颗粒物污染等级的预测中，提高了工作效率的同时，也能满足预测实时性的要求；另外还可增强预测的科学性、准确性，使判定出的细颗粒物污染等级更加真实可靠。
本领域技术人员可以理解附图只是一个优选实施场景的示意图，附图中的模块或流程并不一定是实施本申请所必须的。本领域技术人员可以理解实施场景中的装置中的模块可以按照实施场景描述进行分布于实施场景的装置中，也可以进行相应变化位于不同于本实施场景的一个或多个装置中。上述实施场景的模块可以合并为一个模块，也可以进一步拆分成多个子模块。
上述本申请序号仅仅为了描述，不代表实施场景的优劣。以上公开的仅为本申请的几个具体实施场景，但是，本申请并非局限于此，任何本领域的技术人员能思之的变化都应落入本申请的保护范围。

Claims (20)

  1. 一种细颗粒物污染等级的预测方法，其特征在于，包括：
    筛选出与细颗粒物相关性符合预设标准的目标分析数据；
    根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图；
    基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型；
    利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级。
  2. 根据权利要求1的方法，其特征在于，所述筛选出与细颗粒物相关性符合预设标准的目标分析数据，具体包括：
    收集所述预设历史时间段内的所有环境指标数据并创建特征画像；
    从所述特征画像中筛选出与所述细颗粒物相关性大于或等于第一预设阈值的第一环境指标；
    计算各个所述第一环境指标对应环境数据的饱和度；
    将所述饱和度大于或等于第二预设阈值的所述第一环境指标确定为第二环境指标；
    基于非监督学习的聚类算法填充所述第二环境指标中数据异常的环境数据，以便获取得到数据完整的所述目标分析数据。
  3. 根据权利要求2的方法，其特征在于，所述细颗粒物相关性的计算公式为：
    Figure PCTCN2019118403-appb-100001
    其中，r为各个待测环境指标与PM2.5之间的相关系数，x i为待测环境指标在预设时间段内的单个指标数值，
    Figure PCTCN2019118403-appb-100002
    为待测环境指标在预设时间段内的指标平均值，y i为同一预设时间段内与待测环境指标对应的单个PM2.5数值，
    Figure PCTCN2019118403-appb-100003
    为PM2.5在该预设时间段内的平均值。
  4. 根据权利要求3的方法，其特征在于，所述根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图，具体包括：
    获取预设历史时间段内各个时间点对应的细颗粒物浓度值；
    绘制所述细颗粒物浓度值的散点图；
    基于克里格插值法在所述散点图的基础上创建细颗粒物空间分布图；
    所述基于克里格插值法在所述散点图的基础上创建细颗粒物空间分布图，具体包括：
    从所述散点图中筛选出落在预设搜索范围内的目标数据点；
    确定所述目标数据点对应空间变化的数学函数；
    依据所述数学函数为落在规则格网单元上的数据点赋值，以便获取得到细颗粒物空间分布图。
  5. 根据权利要求4的方法，其特征在于，所述基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型，具体包括：
    确定所述细颗粒物空间分布图中包含的各个污染等级，并配置等级标签；
    将所述目标分析数据中的各类环境指标数据及所述细颗粒物空间分布图输入初始卷积神经网络模型中，提取所述各类环境指标数据与所述各个污染等级的特征向量；
    依据所述特征向量确定各个环境指标对应污染等级的判定规则；
    获取各个所述环境指标依据所述判定规则确定出的第一细颗粒物污染等级；
    将级别最大的所述第一细颗粒物污染等级确定为预估细颗粒物污染等级；
    匹配所述预估细颗粒物污染等级与实际配置的污染等级；
    若确定所述污染等级的匹配正确率大于或等于第三预设阈值，则判定所述初始卷积神经网络模型通过训练；
    若确定所述污染等级的匹配正确率小于所述第三预设阈值，则判定所述初始卷积神经网络模型未通过训练，则更新各个所述环境指标的判定规则，以使所述初始目标检测模型通过训练。
  6. 根据权利要求5的方法，其特征在于，所述获取各个所述环境指标依据所述判定规则确定出的第一细颗粒物污染等级，具体包括：
    获取所述环境指标对应的环境数据；
    基于探测分类概率计算所述环境数据对应各个污染等级的类别概率；
    将对应所述类别概率最大的污染等级确定为根据所述环境指标判定出的第一细颗粒物污染等级。
  7. 根据权利要求6的方法，其特征在于，所述利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级，包括：
    将实时环境数据输入到训练好的卷积神经网络模型中；
    获取由不同环境指标数据确定出的各个第二细颗粒物污染等级；
    将级别最大的所述第二细颗粒物污染等级确定为未来预设时间段内细颗粒物的污染等级。
  8. 根据权利要求7的方法，其特征在于，在所述利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级之后，具体还包括：
    输出所述细颗粒物的污染等级；
    若根据所述污染等级判定存在细颗粒物污染，则输出报警信息。
  9. 一种细颗粒物污染等级的预测装置，其特征在于，包括：
    筛选模块，用于筛选出与细颗粒物相关性符合预设标准的目标分析数据；
    创建模块，用于根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图；
    训练模块，用于基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型；
    判定模块，用于利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级。
  10. 根据权利要求9所述的装置，其特征在于，所述筛选模块，具体用于收集所述预设历史时间段内的所有环境指标数据并创建特征画像；从所述特征画像中筛选出与所述细颗粒物相关性大于或等于第一预设阈值的第一环境指标；计算各个所述第一环境指标对应环境数据的饱和度；将所述饱和度大于或等于第二预设阈值的所述第一环境指标确定为第二环境指标；基于非监督学习的聚类算法填充所述第二环境指标中数据异常的环境数据，以便获取得到数据完整的所述目标分析数据。
  11. 根据权利要求10所述的装置，其特征在于，所述创建模块，具体用于获取预设历史时间段内各个时间点对应的细颗粒物浓度值；绘制所述细颗粒物浓度值的散点图；基于克里格插值法在所述散点图的基础上创建细颗粒物空间分布图；所述基于克里格插值法在所述散点图的基础上创建细颗粒物空间分布图，具体包括：从所述散点图中筛选出落在预设搜索范围内的目标数据点；确定所述目标数据点对应空间变化的数学函数；依据所述数学函数为落在规则格网单元上的数据点赋值，以便获取得到细颗粒物空间分布图。
  12. 根据权利要求11所述的装置，其特征在于，所述训练模块，具体用于确定所述细颗粒物空间分布图中包含的各个污染等级，并配置等级标签；将所述目标分析数据中的各类环境指标数据及所述细颗粒物空间分布图输入初始卷积神经网络模型中，提取所述各类环境指标数据与所述各个污染等级的特征向量；依据所述特征向量确定各个环境指标对应污染等级的判定规则；获取各个所述环境指标依据所述判定规则确定出的第一细颗粒物污染等级；将级别最大的所述第一细颗粒物污染等级确定为预估细颗粒物污染等级；匹配所述预估细颗粒物污染等级与实际配置的污染等级；若确定所述污染等级的匹配正确率大于或等于第三预设阈值，则判定所述初始卷积神经网络模型通过训练；若确定所述污染等级的匹配正确率小于所述第三预设阈值，则判定所述初始卷积神经网络模型未通过训练，则 更新各个所述环境指标的判定规则，以使所述初始目标检测模型通过训练。
  13. 根据权利要求12所述的装置，其特征在于，所述训练模块，具体还用于获取所述环境指标对应的环境数据；基于探测分类概率计算所述环境数据对应各个污染等级的类别概率；将对应所述类别概率最大的污染等级确定为根据所述环境指标判定出的第一细颗粒物污染等级。
  14. 根据权利要求13所述的装置，其特征在于，所述判定模块，具体用于将实时环境数据输入到训练好的卷积神经网络模型中；获取由不同环境指标数据确定出的各个第二细颗粒物污染等级；将级别最大的所述第二细颗粒物污染等级确定为未来预设时间段内细颗粒物的污染等级。
  15. 根据权利要求14所述的装置，其特征在于，所述装置还包括：输出模块；
    所述输出模块，用于输出所述细颗粒物的污染等级；若根据所述污染等级判定存在细颗粒物污染，则输出报警信息。
  16. 一种非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，计算机可读指令被处理器执行时实现细颗粒物污染等级的预测方法，包括：
    筛选出与细颗粒物相关性符合预设标准的目标分析数据；根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图；基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型；利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级。
  17. 根据权利要求16所述的计算机可读存储介质，其特征在于，所述计算机可读指令被处理器执行时实现所述筛选出与细颗粒物相关性符合预设标准的目标分析数据，包括：
    收集所述预设历史时间段内的所有环境指标数据并创建特征画像；从所述特征画像中筛选出与所述细颗粒物相关性大于或等于第一预设阈值的第一环境指标；计算各个所述第一环境指标对应环境数据的饱和度；将所述饱和度大于或等于第二预设阈值的所述第一环境指标确定为第二环境指标；基于非监督学习的聚类算法填充所述第二环境指标中数据异常的环境数据，以便获取得到数据完整的所述目标分析数据。
  18. 根据权利要求17所述的计算机可读存储介质，其特征在于，所述计算机可读指令被处理器执行时实现所述根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图，包括：
    获取预设历史时间段内各个时间点对应的细颗粒物浓度值；绘制所述细颗粒物浓度值的散点图；基于克里格插值法在所述散点图的基础上创建细颗粒物空间分布图；所述基于克里格插值法在所述散点图的基础上创建细颗粒物空间分布图，具体包括：从所述散点图 中筛选出落在预设搜索范围内的目标数据点；确定所述目标数据点对应空间变化的数学函数；依据所述数学函数为落在规则格网单元上的数据点赋值，以便获取得到细颗粒物空间分布图。
  19. 一种计算机设备，包括非易失性可读存储介质、处理器及存储在非易失性可读存储介质上并可在处理器上运行的计算机可读指令，其特征在于，处理器执行所述计算机可读指令时实现细颗粒物污染等级的预测方法，包括：
    筛选出与细颗粒物相关性符合预设标准的目标分析数据；根据预设历史时间段内所述细颗粒物的浓度值创建细颗粒物空间分布图；基于所述目标分析数据及所述细颗粒物空间分布图训练卷积神经网络模型；利用训练好的所述卷积神经网络模型判定未来预设时间段内所述细颗粒物的污染等级。
  20. 根据权利要求19所述的计算机设备，其特征在于，所述所述计算机可读指令被处理器执行时实现所述筛选出与细颗粒物相关性符合预设标准的目标分析数据，包括：
    收集所述预设历史时间段内的所有环境指标数据并创建特征画像；从所述特征画像中筛选出与所述细颗粒物相关性大于或等于第一预设阈值的第一环境指标；计算各个所述第一环境指标对应环境数据的饱和度；将所述饱和度大于或等于第二预设阈值的所述第一环境指标确定为第二环境指标；基于非监督学习的聚类算法填充所述第二环境指标中数据异常的环境数据，以便获取得到数据完整的所述目标分析数据。
PCT/CN2019/118403 2019-09-20 2019-11-14 细颗粒物污染等级的预测方法、装置及计算机设备 WO2021051609A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910894165.0 2019-09-20
CN201910894165.0A CN110796284B (zh) 2019-09-20 2019-09-20 细颗粒物污染等级的预测方法、装置及计算机设备

Publications (1)

Publication Number Publication Date
WO2021051609A1 true WO2021051609A1 (zh) 2021-03-25

Family

ID=69438735

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/118403 WO2021051609A1 (zh) 2019-09-20 2019-11-14 细颗粒物污染等级的预测方法、装置及计算机设备

Country Status (2)

Country Link
CN (1) CN110796284B (zh)
WO (1) WO2021051609A1 (zh)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113418841A (zh) * 2021-06-23 2021-09-21 四川省生态环境监测总站 一种空气质量颗粒物浓度预测数据的补全方法
CN115240060A (zh) * 2022-09-22 2022-10-25 海南浙江大学研究院 一种海星灾害预警方法及系统
CN115358572A (zh) * 2022-08-17 2022-11-18 生态环境部南京环境科学研究所 一种污染场地地表颗粒物释放暴露快速评估方法
CN115453064A (zh) * 2022-09-22 2022-12-09 山东大学 一种细颗粒物空气污染成因分析方法及系统
CN115774953A (zh) * 2022-11-09 2023-03-10 安徽新宇环保科技股份有限公司 一种基于数据处理的污染时空风险监管评估系统及方法
CN115860286A (zh) * 2023-03-01 2023-03-28 江苏省生态环境监测监控有限公司 一种基于时序门机制的空气质量预测方法及系统
CN116297062A (zh) * 2023-02-22 2023-06-23 中国测绘科学研究院 一种面向多维时空因子的pm2.5浓度精细化时空模拟与定量分析方法
CN116340768A (zh) * 2023-02-28 2023-06-27 江苏省环境工程技术有限公司 一种智能化道路积尘负荷监测方法及监测装置
CN116384754A (zh) * 2023-06-02 2023-07-04 北京建工环境修复股份有限公司 一种基于深度学习的在产化工园区环境污染风险评估方法
CN116403092A (zh) * 2023-06-02 2023-07-07 北京建工环境修复股份有限公司 一种基于图像学习的地下水napl污染程度判定方法及系统
CN116758342A (zh) * 2023-06-01 2023-09-15 中国地质科学院矿产资源研究所 一种基于稀土矿产区域的大气污染等级评估方法及装置
CN116883755A (zh) * 2023-07-20 2023-10-13 广州新城建筑设计院有限公司 一种乡村建设环境监测方法、系统、设备及存储介质
CN117171675A (zh) * 2023-11-02 2023-12-05 北京建工环境修复股份有限公司 基于多源数据的水环境微生物检测方法、系统及介质
CN117330710A (zh) * 2023-09-26 2024-01-02 桂林电子科技大学 一种大气污染检测装置及大气污染趋势预测方法
CN117557044A (zh) * 2023-11-22 2024-02-13 山东德才建设有限公司 一种基于物联网平台的建筑作业环境监控方法及设备

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111428405B (zh) * 2020-03-20 2023-07-07 北京百分点科技集团股份有限公司 细颗粒物浓度仿真方法、装置、存储介质及电子设备
CN114580572B (zh) * 2022-04-07 2023-04-07 中科三清科技有限公司 一种异常值的识别方法、装置、电子设备及存储介质
CN117538492B (zh) * 2024-01-10 2024-03-26 深圳市恒义建筑技术有限公司 建筑空间中污染物的在线检测方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160125307A1 (en) * 2013-06-05 2016-05-05 Yu Zheng Air quality inference using multiple data sources
CN106529081A (zh) * 2016-12-03 2017-03-22 安徽新华学院 一种基于神经网络的pm2.5实时等级预测方法和系统
CN107798425A (zh) * 2017-10-16 2018-03-13 中国科学院地理科学与资源研究所 一种基于大数据的时空混淆暴露度评估系统及方法
CN108009674A (zh) * 2017-11-27 2018-05-08 上海师范大学 基于cnn和lstm融合神经网络的空气pm2.5浓度预测方法
CN108108836A (zh) * 2017-12-15 2018-06-01 清华大学 一种基于时空深度学习的臭氧浓度分布预测方法和系统

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103325068B (zh) * 2013-05-29 2016-12-28 广东电网公司电力科学研究院 一种实时动态三维电网污区分布图的绘制方法
CN104809511B (zh) * 2014-01-28 2018-04-17 国际商业机器公司 绝缘子污染预测方法及设备
CN105740643B (zh) * 2016-03-15 2018-08-07 杭州电子科技大学 一种基于城市区域网格自适应的pm2.5浓度推测方法
CN106056210B (zh) * 2016-06-07 2018-06-01 浙江工业大学 一种基于混合神经网络的pm2.5浓度值预测方法
CN107300550A (zh) * 2017-06-21 2017-10-27 南京大学 一种基于bp神经网络模型预测大气重金属浓度的方法
CN109087277B (zh) * 2018-06-11 2021-02-26 北京工业大学 一种空气细颗粒物pm2.5测量方法
CN109523013B (zh) * 2018-10-15 2021-08-06 西北大学 基于浅层卷积神经网络的空气颗粒物污染程度估计方法
CN109816163A (zh) * 2019-01-15 2019-05-28 深圳中兴网信科技有限公司 污染度预测方法、预测系统及计算机可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160125307A1 (en) * 2013-06-05 2016-05-05 Yu Zheng Air quality inference using multiple data sources
CN106529081A (zh) * 2016-12-03 2017-03-22 安徽新华学院 一种基于神经网络的pm2.5实时等级预测方法和系统
CN107798425A (zh) * 2017-10-16 2018-03-13 中国科学院地理科学与资源研究所 一种基于大数据的时空混淆暴露度评估系统及方法
CN108009674A (zh) * 2017-11-27 2018-05-08 上海师范大学 基于cnn和lstm融合神经网络的空气pm2.5浓度预测方法
CN108108836A (zh) * 2017-12-15 2018-06-01 清华大学 一种基于时空深度学习的臭氧浓度分布预测方法和系统

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113418841A (zh) * 2021-06-23 2021-09-21 四川省生态环境监测总站 一种空气质量颗粒物浓度预测数据的补全方法
CN115358572B (zh) * 2022-08-17 2023-04-28 生态环境部南京环境科学研究所 一种污染场地地表颗粒物释放暴露快速评估方法
CN115358572A (zh) * 2022-08-17 2022-11-18 生态环境部南京环境科学研究所 一种污染场地地表颗粒物释放暴露快速评估方法
CN115453064A (zh) * 2022-09-22 2022-12-09 山东大学 一种细颗粒物空气污染成因分析方法及系统
CN115240060A (zh) * 2022-09-22 2022-10-25 海南浙江大学研究院 一种海星灾害预警方法及系统
CN115453064B (zh) * 2022-09-22 2023-09-05 山东大学 一种细颗粒物空气污染成因分析方法及系统
CN115774953A (zh) * 2022-11-09 2023-03-10 安徽新宇环保科技股份有限公司 一种基于数据处理的污染时空风险监管评估系统及方法
CN115774953B (zh) * 2022-11-09 2024-04-26 安徽新宇环保科技股份有限公司 一种基于数据处理的污染时空风险监管评估系统及方法
CN116297062A (zh) * 2023-02-22 2023-06-23 中国测绘科学研究院 一种面向多维时空因子的pm2.5浓度精细化时空模拟与定量分析方法
CN116297062B (zh) * 2023-02-22 2023-11-17 中国测绘科学研究院 一种面向多维时空因子的pm2.5浓度精细化时空模拟与定量分析方法
CN116340768B (zh) * 2023-02-28 2023-10-20 江苏省环境工程技术有限公司 一种智能化道路积尘负荷监测方法及监测装置
CN116340768A (zh) * 2023-02-28 2023-06-27 江苏省环境工程技术有限公司 一种智能化道路积尘负荷监测方法及监测装置
CN115860286A (zh) * 2023-03-01 2023-03-28 江苏省生态环境监测监控有限公司 一种基于时序门机制的空气质量预测方法及系统
CN116758342B (zh) * 2023-06-01 2023-12-15 中国地质科学院矿产资源研究所 一种基于稀土矿产区域的大气污染等级评估方法及装置
CN116758342A (zh) * 2023-06-01 2023-09-15 中国地质科学院矿产资源研究所 一种基于稀土矿产区域的大气污染等级评估方法及装置
CN116403092B (zh) * 2023-06-02 2023-08-18 北京建工环境修复股份有限公司 一种基于图像学习的地下水napl污染程度判定方法及系统
CN116384754B (zh) * 2023-06-02 2023-08-04 北京建工环境修复股份有限公司 一种基于深度学习的在产化工园区环境污染风险评估方法
CN116403092A (zh) * 2023-06-02 2023-07-07 北京建工环境修复股份有限公司 一种基于图像学习的地下水napl污染程度判定方法及系统
CN116384754A (zh) * 2023-06-02 2023-07-04 北京建工环境修复股份有限公司 一种基于深度学习的在产化工园区环境污染风险评估方法
CN116883755A (zh) * 2023-07-20 2023-10-13 广州新城建筑设计院有限公司 一种乡村建设环境监测方法、系统、设备及存储介质
CN116883755B (zh) * 2023-07-20 2024-03-26 广州新城建筑设计院有限公司 一种乡村建设环境监测方法、系统、设备及存储介质
CN117330710A (zh) * 2023-09-26 2024-01-02 桂林电子科技大学 一种大气污染检测装置及大气污染趋势预测方法
CN117171675A (zh) * 2023-11-02 2023-12-05 北京建工环境修复股份有限公司 基于多源数据的水环境微生物检测方法、系统及介质
CN117171675B (zh) * 2023-11-02 2024-01-12 北京建工环境修复股份有限公司 基于多源数据的水环境微生物检测方法、系统及介质
CN117557044A (zh) * 2023-11-22 2024-02-13 山东德才建设有限公司 一种基于物联网平台的建筑作业环境监控方法及设备
CN117557044B (zh) * 2023-11-22 2024-04-09 山东德才建设有限公司 一种基于物联网平台的建筑作业环境监控方法及设备

Also Published As

Publication number Publication date
CN110796284A (zh) 2020-02-14
CN110796284B (zh) 2022-05-17

Similar Documents

Publication Publication Date Title
WO2021051609A1 (zh) 细颗粒物污染等级的预测方法、装置及计算机设备
Hermosilla et al. Regional detection, characterization, and attribution of annual forest change from 1984 to 2012 using Landsat-derived time-series metrics
Lin et al. Real-time traffic accidents post-impact prediction: Based on crowdsourcing data
CN108446470B (zh) 基于车辆轨迹数据和人口分布的医疗设施可达性分析方法
CN110427533B (zh) 基于时序粒子跟踪的污染传播模式可视分析方法及系统
Goovaerts et al. Detection of temporal changes in the spatial distribution of cancer rates using local Moran’s I and geostatistically simulated spatial neutral models
Yao et al. Discovering the homogeneous geographic domain of human perceptions from street view images
CN109493119B (zh) 一种基于poi数据的城市商业中心识别方法及系统
CN112820415B (zh) 一种基于gis的慢性病时空演化特征分析及环境健康风险监测系统及方法
WO2021068781A1 (zh) 一种疲劳状态识别方法、装置和设备
WO2020108219A1 (zh) 基于交通安全风险的群体划分与差异性分析方法及系统
Chaudhary et al. Time series based LSTM model to predict air pollutant’s concentration for prominent cities in India
CN115409069A (zh) 村镇建筑的识别方法、分类方法、装置、电子设备和介质
Ahn et al. A digital twin city model for age-friendly communities: Capturing environmental distress from multimodal sensory data
WO2023143000A1 (zh) 一种基于多源大数据的老龄友好街道建成环境审计系统
CN111462919A (zh) 一种基于滑动窗口时序模型的虫媒疾病预测方法及系统
Su et al. New particle formation event detection with Mask R-CNN
CN111223567A (zh) 一种基于区域医学影像的疑似疾病风险范围的计算方法
CN113326472B (zh) 一种基于时序多变量数据的模式提取与演化可视分析方法
CN106952251B (zh) 一种基于吸附模型的图像显著性检测方法
CN113379334B (zh) 基于有噪声轨迹数据的路段自行车骑行质量鉴别方法
CN115879594A (zh) 一种基于地理探测器的城市定居人口分布趋势预测方法
CN114360735A (zh) 传染病时空聚集性探测分析方法、系统及电子设备
Kou Abnormal pattern recognition in spatial data
Awuah Effects of spatial resolution, land-cover heterogeneity and different classification methods on accuracy of land-cover mapping

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19945762

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19945762

Country of ref document: EP

Kind code of ref document: A1