Urban and rural garbage production prediction method based on spatial scale extension and data fusion

Abstract

The present invention discloses a method for predicting urban and rural garbage production based on spatial scale extension and data fusion, which belongs to the field of environmental technology and comprises the following steps: S1: fusing scattered surface elements in the shp data of a map file of a measured area; S2: setting the garbage production data of all districts and towns in the measured area to a csv format, and respectively associating them with the corresponding district and town-level administrative regions; the present invention uses the fusion of spatial remote sensing data for the first time to perform machine learning regression prediction on the change of garbage production within a long period of time, and compares the application effects of different machine learning algorithms, realizing the analysis of the change characteristics of the production of different types of garbage in time and space, and being able to obtain the garbage generation status in remote areas where data statistics are difficult to perform, as well as the change of urban and rural garbage, especially for application in rural areas that are difficult to manage.

Description

Technical Field

[0001] The present invention belongs to the field of environmental technology, and in particular relates to a method for predicting urban and rural garbage production based on spatial scale extension and data fusion. Background Art

[0002] The macro-planning and management of garbage has developed into an important emerging industry at home and abroad, and has attracted widespread attention. In recent years, the amount of garbage generated in rural areas of my country has reached 200 million tons per year. In order to meet the needs of refined management and construction of sanitary environment in villages and towns, it is necessary to conduct a detailed analysis of the magnitude of garbage-related data on a temporal and spatial scale.

[0003] In the process of spatial analysis of social statistics, it is difficult to conduct effective data analysis and production forecasting due to the small amount of rural garbage generated, its dispersion and complex composition, the unclear spatial distribution characteristics of garbage, the lack of basic management data and the reality of broken data management. In addition, traditional garbage data statistical methods are often based on vehicle numbers and loads, which makes it difficult to trace the garbage generation situation to specific geographical units and villagers' settlements. Therefore, it is necessary to form relevant methods to serve the early layout of village and town garbage collection and transportation equipment. Summary of the invention

[0004] To solve the problems raised in the above background technology, the present invention provides a method for predicting urban and rural garbage production based on spatial scale extension and data fusion, which has the characteristics of solving the problem that garbage distribution characteristics in some areas are unclear, lack of basic management data and difficult to manage, which leads to difficulty in predicting production.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for predicting urban and rural garbage production based on spatial scale extension and data fusion, comprising the following steps:

[0006] S1: Obtain the map file shp data of the measured area from the open source network, and use the fusion function of GeoScene Pro2.1 software to fuse the scattered surface elements in the map file shp data of the measured area, so that each district and town in the measured area is composed of a separate surface element;

[0007] S2: The data of different types of garbage production in all districts and towns in the measured area are set in csv format and associated with the corresponding district and town-level administrative areas;

[0008] S3: The total number of towns and districts included in the measured area is recorded as n, and the n towns and districts are grouped into m towns and districts as a group, and multiple extended groups are obtained. The total garbage production of all towns and districts in each extended group is taken as the garbage production of each extended group, where m≤n;

[0009] S4: Collect training sample data at the same time scale as the garbage production of each extended group, and trim the part of the training sample data that is outside the measured area. The training sample data includes the following collected on the open source network: settlement data, population density data, VIIRS / DNB night light data, OSM road network data and water body contour data;

[0010] S5: Calculate the total and average values ​​of the settlement data, the total and average values ​​of the population density data, and the total and average values ​​of the VIIRS / DNB night light data for each extension group, break the OSM road network and water bodies in each extension group according to the boundaries of the district-level administrative districts, and calculate the total length of the OSM road network and the total area of ​​water bodies in each district and town in each extension group. Use the total and average values ​​of the settlement data, the total and average values ​​of the population density data, the total and average values ​​of the VIIRS / DNB night light data, the total length of the OSM road network, and the total area of ​​water bodies for each extension group as analysis features;

[0011] S6: Classify the garbage in the tested area according to the garbage classification policy of the administrative district to which the tested area belongs, calculate the high-value features of each type of garbage, and use them as the predicted sample data of this type of garbage;

[0012] S7: Use the LightGBM algorithm, CatBoost algorithm, XGBoost algorithm and model fusion algorithm to establish regression prediction models between the predicted sample data and the output of the same category of garbage, and use the predicted sample data to train the regression prediction models of the same category, and select the regression prediction model with the smallest MAE value in the same category of regression prediction models as the optimal prediction model for the category of garbage, so as to obtain the optimal prediction models for different categories of garbage;

[0013] S8: Use the spatial fishing net to grid the measured area and extract the test sample data of each grid. The test sample data includes the actual settlement data, population density data, VIIRS / DNB night light data, OSM road network data and water body contour data of the measured area;

[0014] S9: The test sample data of each grid is used as the input of the optimal prediction model of different categories of garbage to obtain the garbage production of different categories in each grid, thereby realizing the prediction of the production of different types of garbage in the entire tested area.

[0015] Furthermore, in the present invention, in step S3, the values ​​of m are: 2n, 0.25n, 0.5n, 0.75n, n-1 and n, and the current rounding method of m is rounding down.

[0016] In the present invention, further, in step S3, the method for selecting the extended group is:

[0017] When m≤n / 2 or m=n, m districts and towns are randomly selected from n districts and towns as a group, and the districts and towns in each extended group do not overlap. If the number of remaining districts and towns is less than m during the selection process, the selection of the extended group under the current value of m is completed;

[0018] When m>n / 2 and m≠n, m districts and towns are randomly selected from n districts and towns as a group. The districts and towns in each extended group overlap and are not completely the same.

[0019] Furthermore, in the step S5 of the present invention, the settlement data, population density data and VIIRS / DNB night light data are all expressed in the form of raster data, the total value of the raster data is the sum of the data in all the grids, the average value of the raster data is the quotient of the total value and the number of grids, and the total and average values ​​of the settlement data, population density data and VIIRS / DNB night light data are calculated using the tabular display zoning statistics function of GeoScene Pro2.1 software.

[0020] Furthermore, in the step S6 of the present invention, the method for calculating the high-value features of each type of garbage is the same, and the specific method is: grey correlation analysis is performed on the eight analysis features obtained in step S5 and the garbage output of the corresponding extended group respectively, to obtain the grey correlation analysis value of each analysis feature, and the obtained grey correlation analysis values ​​of each analysis feature are sorted from large to small, and the analysis features corresponding to the top five grey correlation analysis values ​​are selected as high-value features.

[0021] Furthermore, in the present invention, in step S7, the model fusion algorithm is to fuse the LightGBM algorithm, the CatBoost algorithm and the XGBoost algorithm.

[0022] In the present invention, in step S8, the unit size of the spatial fishing net is 500m×

[0023] 500m.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. The present invention uses the fusion of spatial remote sensing data for the first time to perform machine learning regression prediction on the changes in garbage production over a long period of time, and compares the application effects of different machine learning algorithms, realizing the analysis of the changing characteristics of different types of garbage production in time and space. It can obtain the garbage generation status in remote areas where data statistics are difficult to obtain, as well as the changes in urban and rural garbage, especially for application in rural areas that are difficult to manage.

[0026] 2. The prediction results of the present invention can serve the processes of generation, collection, transportation, transfer, treatment and recycling of rural garbage. This process is not only the operation of the garbage material flow and energy flow, but also serves the spatial linkage and matching coordination between various key links. It has broad application prospects in the field of intelligent solid waste management in different urban and rural scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the method for predicting urban and rural garbage production based on spatial scale extension and data fusion of the present invention;

[0028] Figure 2 This is the statistical chart of waste classification output in Hong Kong in 2019;

[0029] Figure 3 This is a diagram showing the effect of the prediction model in 2D space using the method of the present invention;

[0030] Figure 4 This is a rendering of the prediction model of construction waste in 3D space using the method of the present invention;

[0031] Figure 5 This is a prediction model effect diagram of urban garbage in 3D space using the method of the present invention;

[0032] Figure 6 This is a prediction model effect diagram of household garbage in 3D space using the method of the present invention;

[0033] Figure 7 This is a prediction model effect diagram of industrial and commercial waste in 3D space using the method of the present invention;

[0034] Figure 8 This is an effect diagram of the prediction model of the total amount of garbage in 3D space using the method of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] See also Figure 1 The present invention provides the following technical solution: a method for predicting urban and rural garbage production based on spatial scale extension and data fusion, comprising the following steps:

[0037] S1: Obtain the map file shp data of the measured area from the open source network, and use the fusion function of GeoScene Pro2.1 software to fuse the scattered surface elements in the map file shp data of the measured area, so that each district and town in the measured area is composed of a separate surface element;

[0038] S2: The data of different types of garbage production in all districts and towns in the measured area are set in csv format and associated with the corresponding district and town-level administrative areas;

[0039] S3: The total number of towns and districts included in the measured area is recorded as n, and the n towns and districts are grouped into m towns and districts as a group, and multiple extended groups are obtained by extension, wherein the values ​​of m are: 2n, 0.25n, 0.5n, 0.75n, n-1 and n, and the current rounding method of m is rounding down. The method for selecting the extended group is:

[0040] When m≤n / 2 or m=n, m districts and towns are randomly selected from n districts and towns as a group, and the districts and towns in each extended group do not overlap. If the number of remaining districts and towns is less than m during the selection process, the selection of the extended group under the current value of m is completed;

[0041] When m>n / 2 and m≠n, m districts and towns are randomly selected from n districts and towns as a group. The districts and towns in each extended group are overlapped and not completely the same.

[0042] The total garbage production of all districts and towns in each extended group is taken as the garbage production of each extended group;

[0043] S4: Collect training sample data at the same time scale as the garbage production of each extended group, and trim the part of the training sample data that is outside the measured area. The training sample data includes the following collected on the open source network: settlement data, population density data, VIIRS / DNB night light data, OSM road network data and water body contour data;

[0044] S5: The settlement data, population density data and VIIRS / DNB night light data are all expressed in the form of raster data. The total value of the raster data is the sum of the data in all the rasters, and the average value of the raster data is the quotient of the total value and the number of rasters. The total value and average value of the settlement data, the total value and average value of the population density data, and the total value and average value of the VIIRS / DNB night light data corresponding to each extended group are calculated using the table display zoning statistics function of GeoScene Pro2.1 software. The OSM road network and water bodies in each extended group are interrupted according to the boundaries of district-level administrative districts, and the total length of the OSM road network and the total area of ​​water bodies in each district and town in each extended group are calculated. The total value and average value of the settlement data, the total value and average value of the population density data, the total value and average value of the VIIRS / DNB night light data, the total length of the OSM road network and the total area of ​​water bodies corresponding to each extended group are used as analysis features.

[0045] S6: Classify the garbage in the tested area according to the garbage classification policy of the administrative district to which the tested area belongs, calculate the high-value features of each type of garbage respectively, and use them as the predicted sample data of this type of garbage. The method for calculating the high-value features of each type of garbage is the same, and the specific method is:

[0046] The eight analysis features obtained in step S5 are respectively subjected to grey correlation analysis with the garbage output of the corresponding extended group to obtain the grey correlation analysis value of each analysis feature, and the obtained grey correlation analysis values ​​of each analysis feature are sorted from large to small, and the analysis features corresponding to the top five grey correlation analysis values ​​are selected as high-value features;

[0047] S7: Use LightGBM algorithm, CatBoost algorithm, XGBoost algorithm and model fusion algorithm to establish regression prediction models between prediction sample data and the output of garbage of the same category, respectively. The model fusion algorithm is to fuse LightGBM algorithm, CatBoost algorithm and XGBoost algorithm, and use the prediction sample data to train the regression prediction model of the same category, and select the regression prediction model with the smallest MAE value in the same category of regression prediction models as the optimal prediction model of the category of garbage, so as to obtain the optimal prediction model of different categories of garbage;

[0048] S8: Use a 500m×500m spatial fishing net to grid the measured area and extract test sample data for each grid. The test sample data includes the actual settlement data, population density data, VIIRS / DNB night light data, OSM road network data and water body contour data of the measured area;

[0049] S9: The test sample data of each grid is used as the input of the optimal prediction model of different categories of garbage to obtain the garbage production of different categories in each grid, thereby realizing the prediction of the production of different types of garbage in the entire tested area. Specific embodiment:

[0051] Reference Figures 2 to 8 This embodiment is specifically described. This embodiment takes the spatiotemporal distribution of domestic waste in the Hong Kong Special Administrative Region and its surrounding towns in March 2021 as an example, takes four major areas including Hong Kong Island, Kowloon, New Territories, and outlying islands and 18 small areas as statistical units, and conducts an inversion study on waste classification production to meet the needs of exploring the spatiotemporal distribution of local waste production. The spatiotemporal distribution inversion of domestic waste in 2020 is obtained, which provides decision-making support for the layout of the future domestic waste collection, storage and transportation system. The specific implementation process is as follows:

[0052] Study area map: Obtain maps of 18 small areas in Hong Kong, such as Yuen Long and Tuen Mun, and merge their scattered surface elements to obtain complete surface elements of 18 areas. Merge 18 geographic units to obtain surface elements of the entire territory of Hong Kong. Extract relevant administrative districts in Hong Kong Island, Kowloon, New Territories and other areas and merge them to obtain surface elements of three large areas. The above 22 surface elements are merged into one surface element for the extraction of spatial geographic data;

[0053] Garbage production data collation: Extract the garbage production data of each year from the 2014-2019 solid waste monitoring report of the Hong Kong Environmental Protection Department and connect it to the corresponding administrative area elements, such as Figure 2 As shown;

[0054] Extend the spatial scale: Random sampling was conducted in the existing 18 administrative districts, and 9 geographical units formed by two random administrative areas, 4 geographical units formed by 4 random administrative areas, 2 geographical units formed by 9 random administrative areas, 4 geographical units formed by randomly selecting 4 administrative areas, and 18 administrative units formed by 17 random geographical units were constructed. Combined with the 22 administrative units formed in the past, the final geographical unit sample constructed included 59, and the statistical volume increased by nearly two times. The statistical data on population, garbage, etc. corresponding to each group of geographical units consisted of seven years of data from 2013 to 2019, that is, the sample data set for subsequent machine learning training contained 413 groups;

[0055] Processing of raster remote sensing data such as night lights and population: Download Worldpop settlements, population density (100m), and VIIRS / DNB night lights data (June) from 2014 to 2020 from the open source network, perform cropping and data extraction based on the 59 geographic units of Hong Kong described in step 3, and export the csv table;

[0056] Processing of vector data such as water bodies and road networks: Download water body contour data and OSM road network data from 2014 to 2020 from the open source network, perform cropping and data extraction based on the 59 geographic units of Hong Kong described in step 3, and export the csv table;

[0057] Machine learning model construction: Based on the above data, a machine learning model dataset containing regional names, spatial geographic features, and garbage production was constructed. First, a gray correlation analysis was performed to screen out low-correlation data, such as regional average population, average night lights, and average settlements. The removal of three types of mean features improved the accuracy of the model: the modeling effects of four methods, lightGBM, CatBoost, XGBoost, and model fusion, were compared, and XGBoost was the best.

[0058] Classify and predict garbage production: Save the four types of garbage prediction models trained in the previous step, divide the entire Hong Kong area into 500m spatial grids, generate grid surfaces and grid points, extract road network, population, settlement, and night light data according to the location of the fishing net, and perform feature engineering. In order to express that the garbage production in areas without lights, population, and settlements is theoretically 0, the data with both features and garbage production being 0 are added to the dataset and trained together. The features are exported into a table and input into the model to obtain the prediction results, which are then connected back to the fishing net grid. The model results need to be denoised. In areas without lights, population, and settlements, the machine learning fitting results are decimals close to 0 but cannot be exactly 0, so all similar points are returned to 0, resulting in the following: Figure 3 Results shown;

[0059] The three-dimensional display is finally obtained through the prediction method of village and town garbage production based on spatial scale extension and grid data fusion. Figure 4-8 The garbage classification results shown complete the spatial prediction results of local household garbage, industrial and commercial garbage, construction garbage, urban garbage, and total garbage in Hong Kong, which includes garbage generation data in rural areas within Hong Kong that are difficult to count.

[0060] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting urban and rural garbage production based on spatial scale extension and data fusion, characterized in that: The following steps are involved: S1: Obtain the map file shp data of the measured area from the open source network, and use the fusion function of GeoScene Pro2.1 software to fuse the scattered surface elements in the map file shp data of the measured area, so that each district and town in the measured area is composed of a separate surface element; S2: The data of different types of garbage production in all districts and towns in the measured area are set in csv format and associated with the corresponding district and town-level administrative areas; S3: The total number of towns and districts included in the measured area is recorded as n, and the n towns and districts are grouped into m towns and districts as a group, and multiple extended groups are obtained. The total garbage production of all towns and districts in each extended group is taken as the garbage production of each extended group, where m≤n; S4: Collect training sample data at the same time scale as the garbage production of each extended group, and trim the part of the training sample data that is outside the measured area. The training sample data includes the following collected on the open source network: settlement data, population density data, VIIRS / DNB night light data, OSM road network data and water body contour data; S5: Calculate the total and average values ​​of the settlement data, the total and average values ​​of the population density data, and the total and average values ​​of the VIIRS / DNB night light data for each extension group, break the OSM road network and water bodies in each extension group according to the boundaries of the district-level administrative districts, and calculate the total length of the OSM road network and the total area of ​​water bodies in each district and town in each extension group. Use the total and average values ​​of the settlement data, the total and average values ​​of the population density data, the total and average values ​​of the VIIRS / DNB night light data, the total length of the OSM road network, and the total area of ​​water bodies for each extension group as analysis features; S6: Classify the garbage in the tested area according to the garbage classification policy of the administrative district to which the tested area belongs, calculate the high-value features of each type of garbage, and use them as the predicted sample data of this type of garbage; The method for calculating the high-value features of each type of garbage is the same, and the specific method is: the eight analysis features obtained in step S5 are respectively subjected to grey correlation analysis with the garbage output of the corresponding extended group to obtain the grey correlation analysis value of each analysis feature, and the obtained grey correlation analysis values ​​of each analysis feature are sorted from large to small, and the analysis features corresponding to the top five grey correlation analysis values ​​are selected as high-value features; S7: Use the LightGBM algorithm, CatBoost algorithm, XGBoost algorithm and model fusion algorithm to establish regression prediction models between the predicted sample data and the output of the same category of garbage, and use the predicted sample data to train the regression prediction models of the same category, and select the regression prediction model with the smallest MAE value in the same category of regression prediction models as the optimal prediction model for the category of garbage, so as to obtain the optimal prediction models for different categories of garbage; S8: Use the spatial fishing net to grid the measured area and extract the test sample data of each grid. The test sample data includes the actual settlement data, population density data, VIIRS / DNB night light data, OSM road network data and water body contour data of the measured area; S9: The test sample data of each grid is used as the input of the optimal prediction model of different categories of garbage to obtain the garbage production of different categories in each grid, thereby realizing the prediction of the production of different types of garbage in the entire tested area.

2. The method for predicting urban and rural garbage production based on spatial scale extension and data fusion according to claim 1 is characterized by: In step S3, the values ​​of m are: 2n, 0.25n, 0.5n, 0.75n, n-1 and n, and the current rounding method of m is rounding down.

3. The method for predicting urban and rural garbage production based on spatial scale extension and data fusion according to claim 1 is characterized by: In step S3, the method for selecting the extended group is: When m≤n / 2 or m=n, m districts and towns are randomly selected from n districts and towns as a group, and the districts and towns in each extended group do not overlap. If the number of remaining districts and towns is less than m during the selection process, the selection of the extended group under the current value of m is completed; When m>n / 2 and m≠n, m districts and towns are randomly selected from n districts and towns as a group. The districts and towns in each extended group overlap and are not completely the same.

4. The method for predicting urban and rural garbage production based on spatial scale extension and data fusion according to claim 1 is characterized by: In step S5, the settlement data, population density data and VIIRS / DNB night light data are all expressed in the form of raster data. The total value of the raster data is the sum of the data in all grids, and the average value of the raster data is the quotient of the total value and the number of grids. The total value and average value of the settlement data, population density data and VIIRS / DNB night light data are calculated using the GeoScene Pro2.1 software's function of displaying zoning statistics in a table.

5. The method for predicting urban and rural garbage production based on spatial scale extension and data fusion according to claim 1 is characterized by: In step S7, the model fusion algorithm is to fuse the LightGBM algorithm, the CatBoost algorithm and the XGBoost algorithm.

6. The method for predicting urban and rural garbage production based on spatial scale extension and data fusion according to claim 1 is characterized by: In step S8, the unit size of the spatial fishing net is 500m×500m.

Images