Multi-source solid waste data supplementing method based on satellite-ground big data migration application

Abstract

The invention belongs to the field of environmental engineering, and particularly relates to a multi-source solid waste data supplementing method based on satellite-ground big data migration application. The objective of the invention is to solve the problem of low prediction result accuracy of a prediction model caused by lack of analysis modeling start solid waste data in a data scarce region in the prior art. The invention provides a multi-source solid waste data supplementing method based on satellite-ground big data migration application. A multi-source solid waste data supplementing model is constructed; an SARIMA model is constructed; acquiring explanatory variable data of a to-be-supplemented multi-source solid waste data area, and inputting the explanatory variable data of the to-be-supplemented multi-source solid waste data area into the trained multi-source solid waste data supplementing model to obtain the annual solid waste output of the to-be-supplemented multi-source solid waste data area; and inputting the annual solid waste output of the to-be-supplemented multi-source solid waste data area into the trained SARIMA model to obtain the monthly solid waste output of the to-be-supplemented multi-source solid waste data area. The problem that the prediction result of the prediction model is low in accuracy is solved.

Description

Technical Field

[0001] The present invention belongs to the field of environmental engineering, and in particular relates to a multi-source solid waste data supplementation method based on satellite-to-ground big data migration application. Background Art

[0002] Focusing on the theme of high-quality development, the requirements for improving the quality of urban environment are constantly increasing, and the demand for the reduction, reuse and resource utilization of domestic waste is becoming more urgent, which puts forward new and higher requirements for the classification and treatment of domestic waste in my country. In modern cities, the generation of various solid wastes such as industrial waste, hazardous waste and recyclable waste is increasing, and traditional waste treatment methods have been unable to cope with the complexity and diversity of these multi-source solid wastes. Scientific and effective classification and recycling of these wastes will not only help reduce environmental pollution and reduce resource waste, but also improve the efficiency of waste treatment and the recycling rate of resources. In particular, the proper management of hazardous wastes can prevent the potential threat of harmful substances to human health and ecosystems. The classification management of multi-source solid wastes will not only help improve the quality of urban environment, but also enhance the public's environmental awareness and promote the development of society in a green and low-carbon direction. Therefore, promoting the classification and recycling of garbage has important environmental, economic and social significance.

[0003] In this context, building a novel prediction method that can accurately predict the generation and distribution of different types of solid waste is crucial to developing more efficient waste management strategies. However, accurate prediction of the generation of multi-source solid waste in cities still faces challenges.

[0004] The current garbage prediction methods are mainly based on statistical data, which are often scarce and discontinuous, with a rough scale, and only disclose annual total information. There is a lack of detailed production data of different types of solid waste, making it difficult to build an effective prediction model. Therefore, the present invention is proposed to construct a multi-solid waste component collaborative migration prediction framework based on historical discontinuous data, fully release the data value based on historical waste production rules, and use machine learning combined with deep learning to achieve collaborative and accurate prediction of multiple solid wastes, and realize the extrapolation of prediction results, providing sufficient data for the precise management of multi-source solid waste. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the existing technology lacks solid waste data for analysis and modeling in data-scarce areas, resulting in low accuracy of prediction results of prediction models. A multi-source solid waste data supplementation method based on satellite-to-ground big data migration application is proposed, which includes the following steps:

[0006] 1. Construct a multi-source solid waste data supplement model; obtain a trained multi-source solid waste data supplement model; construct a SARIMA model and obtain a trained SARIMA model;

[0007] 2. Collect explanatory variable data (i.e., satellite-ground big data) of the multi-source solid waste data area to be supplemented, input the explanatory variable data of the multi-source solid waste data area to be supplemented into the trained multi-source solid waste data supplementation model, and obtain the annual solid waste garbage output of the multi-source solid waste data area to be supplemented;

[0008] 3. Input the annual solid waste production of the multi-source solid waste data area to be supplemented into the trained SARIMA model to obtain the monthly solid waste production of the multi-source solid waste data area to be supplemented

[0009] In the first step, a multi-source solid waste data supplement model is constructed; and a trained multi-source solid waste data supplement model is obtained;

[0010] Construct a SARIMA model and obtain a trained SARIMA model; the specific process is:

[0011] Step 1: Collect historical explanatory variable datasets;

[0012] Based on the open source satellite-earth big data on the Internet, collect the explanatory variables required for modeling in all the collectible time periods and spatial ranges;

[0013] The historical explanatory variable data include: annual explanatory variable data; monthly explanatory variable data;

[0014] Step 2: Collect historical target variable data sets;

[0015] The historical target variable data set includes: the annual solid waste production in the administrative area, the monthly solid waste production in the administrative area; the number of types of solid waste is Y, and Y is a positive integer;

[0016] The administrative regions include: city-level administrative regions, county-level administrative regions, and district-level administrative regions;

[0017] Generally speaking, the output of four types of solid waste is collected nationwide at the city, county, district, year and month scales. There are Y types of specific classifications based on actual conditions:

[0018] Step 3: Construct a training sample dataset based on the historical explanatory variable dataset collected in step 1 and the historical target variable dataset collected in step 2;

[0019] Performing data enhancement on the training sample data set to obtain an enhanced training sample data set;

[0020] Step 4: According to the number of garbage types in the target variable, construct Z base regression model groups, Z = Y;

[0021] Each base regression model group corresponds to a waste type;

[0022] Step 5: Construct logistic regression meta-model;

[0023] Step 6: construct a multi-source solid waste data supplement model based on the Z basis regression model groups obtained in step 4 and the logistic regression meta-model obtained in step 5; train the multi-source solid waste data supplement model based on the enhanced training sample data set obtained in step 3 to obtain a trained multi-source solid waste data supplement model;

[0024] Step 7: Construct a SARIMA model. According to the enhanced training sample data set obtained in step 3, train the SARIMA model to obtain a trained SARIMA model.

[0025] SARIMA (Seasonal Autoregressive Integrated Moving Average) model is an existing statistical model for time series analysis; the present invention adopts SARIMA model to analyze the temporal autoregressive characteristics of historical monthly solid waste, predict the generation of different types of solid waste in different months, and calculate the proportion of monthly garbage generation in the existing solid waste data to the existing annual solid waste.

[0026] That is, based on the trained SARIMA model, the proportion of different types of solid waste in each month in each year is obtained; the proportion of different types of solid waste in each month in each year that can be analyzed based on the SARIMA model is multiplied by the predicted annual output to obtain the amount of different types of solid waste generated each month in the data-missing area, thereby realizing transfer learning from data-existing areas to data-free areas in scenarios with limited data.

[0027] The beneficial effects of the present invention are:

[0028] The core innovation of this method is to use cross-regional transfer learning technology to fit multiple solid waste prediction models trained in areas with sufficient data to the satellite-ground big data, thus avoiding the limitations of single model prediction.

[0029] A variety of basic solid waste prediction models are integrated in the form of meta-model integration, and other data are migrated to data-scarce areas. Combined with feature fine-tuning and enhancement, the prediction capability of the solid waste production prediction model is improved.

[0030] This invention releases the value of data to the greatest extent, and has solved the problem of lack of analysis and modeling startup data for solid waste in data-scarce areas, as well as difficulties in prediction and management. While realizing the transfer of knowledge on solid waste generation characteristics between regions, it has significantly improved the application breadth of the prediction model, and provided assistance for the coordinated management of multi-source solid waste in urban scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of a multi-source solid waste data supplementation method based on satellite-to-ground big data migration application of the present invention;

[0032] Figure 2It is a schematic diagram of the spatiotemporal distribution of the amount of domestic waste generated predicted by the supplementary data of the present invention. DETAILED DESCRIPTION

[0033] Specific implementation method 1: Combination Figure 1 The invention is described, comprising:

[0034] 1. Construct a multi-source solid waste data supplement model; obtain a trained multi-source solid waste data supplement model; construct a SARIMA model and obtain a trained SARIMA model;

[0035] 2. Collecting the satellite-ground big data of the multi-source solid waste data area to be supplemented, inputting the satellite-ground big data of the multi-source solid waste data area to be supplemented into the trained multi-source solid waste data supplementation model to obtain the annual solid waste garbage output of the multi-source solid waste data area to be supplemented;

[0036] 3. Input the annual solid waste production of the area with multi-source solid waste data to be supplemented into the trained SARIMA model to obtain the monthly solid waste production of the area with multi-source solid waste data to be supplemented.

[0037] Specific implementation method 2: The difference between this implementation method and specific implementation method 1 is that:

[0038] In the first step, a multi-source solid waste data supplement model is constructed; and a trained multi-source solid waste data supplement model is obtained;

[0039] Construct a SARIMA model and obtain a trained SARIMA model; the specific process is:

[0040] Step 1: Collect historical explanatory variable datasets;

[0041] Based on the open source satellite-earth big data on the Internet, collect the explanatory variables required for modeling in all the collectible time periods and spatial ranges;

[0042] The historical explanatory variable data include: annual explanatory variable data; monthly explanatory variable data;

[0043] Step 2: Collect historical target variable data sets;

[0044] The historical target variable data set includes: the annual solid waste production in the administrative area, the monthly solid waste production in the administrative area; the number of types of solid waste is Y, and Y is a positive integer;

[0045] The administrative regions include: city-level administrative regions, county-level administrative regions, and district-level administrative regions;

[0046] Generally speaking, the output of four types of solid waste is collected nationwide at the city, county, district, year and month scales. There are Y types of specific classifications based on actual conditions.

[0047] Step 3: Construct a training sample dataset based on the historical explanatory variable dataset collected in step 1 and the historical target variable dataset collected in step 2;

[0048] Performing data enhancement on the training sample data set to obtain an enhanced training sample data set;

[0049] Step 4: According to the number of garbage types in the target variable, construct Z base regression model groups, Z = Y;

[0050] Each base regression model group corresponds to a waste type;

[0051] Step 5: Construct logistic regression meta-model;

[0052] Step 6: construct a multi-source solid waste data supplement model based on the Z basis regression model groups obtained in step 4 and the logistic regression meta-model obtained in step 5; train the multi-source solid waste data supplement model based on the enhanced training sample data set obtained in step 3 to obtain a trained multi-source solid waste data supplement model;

[0053] Step 7: Construct a SARIMA model. According to the enhanced training sample data set obtained in step 3, train the SARIMA model to obtain a trained SARIMA model.

[0054] SARIMA (Seasonal Autoregressive Integrated Moving Average) model is an existing statistical model for time series analysis; the present invention adopts SARIMA model to analyze the temporal autoregressive characteristics of historical monthly solid waste, predict the generation of different types of solid waste in different months, and calculate the proportion of monthly garbage generation in the existing solid waste data to the existing annual solid waste.

[0055] That is, based on the trained SARIMA model, the proportion of different types of solid waste in each month in each year is obtained; the proportion of different types of solid waste in each month in each year that can be analyzed based on the SARIMA model is multiplied by the predicted annual output to obtain the amount of different types of solid waste generated each month in the data-missing area, thereby realizing transfer learning from data-existing areas to data-free areas in scenarios with limited data.

[0056] The other steps and parameters are the same as those in the first specific implementation.

[0057] Specific implementation method three: This implementation method is different from the specific implementation method one in that:

[0058] The explanatory variable data in step 1 include: night light data, population distribution data, age distribution data, vegetation index data, built-up area data, and road network data;

[0059] The above data has a fine spatial scale and can be used to extract relevant data based on flexible administrative boundaries, which can be used to simultaneously serve solid waste resource data prediction models at multiple spatial scales.

[0060] Among them, the VIIRS satellite provides high-resolution night light data for monitoring urbanization and economic activities;

[0061] The LandScan Global provides accurate population distribution data, and combines it with the seventh national census to analyze the age structure characteristics and obtain age distribution data;

[0062] Providing NDVI vegetation index data through MODIS and Sentinel-2 satellites for monitoring ecosystem health around cities;

[0063] Sentinel-2 and Landsat satellites provide global built-up area data to show the characteristics of urban expansion;

[0064] OpenStreetMap updates global road network data in real time to show the traffic prosperity of different regions.

[0065] The above spatial data were divided into the same spatial scale as the solid waste data using administrative regions for modeling;

[0066] The other steps and parameters are the same as those in the first to second embodiments.

[0067] Specific implementation method 4: This implementation method is different from specific implementation methods 1 to 4 in that:

[0068] The number of solid waste types Y in step 2 is 20; the solid waste types include: catering waste, kitchen waste, other waste, hazardous waste, bulk waste; coal gangue, fly ash, smelting slag, chemical waste slag, construction waste; waste mineral oil, waste acid, waste alkali, waste organic solvent, medical waste; waste metal, waste glass, waste fabric, waste paper, waste plastic;

[0069] The solid waste is divided into four categories, each of which has five types; the four categories are: domestic waste y1, industrial solid waste y2, hazardous waste y3 and recyclable waste y4;

[0070] The domestic waste y1 includes: catering waste y11, kitchen waste y12, other waste y13, hazardous waste y14, and bulk waste y15;

[0071] The industrial solid waste y2 includes: coal gangue y21, fly ash y22, smelting slag y23, chemical waste slag y24, and construction waste y25;

[0072] The hazardous waste y3 includes: waste mineral oil y31, waste acid y32, waste alkali y33, waste organic solvent y34, and medical waste y35;

[0073] The recyclable garbage y4 includes: waste metal y41, waste glass y42, waste fabric y43, waste paper y44, and waste plastic y45;

[0074] The data on the production of the above 20 types of garbage are often only occasionally counted in municipal administrative areas, lacking detailed data by county and district, and the statistical calibers of different places are inconsistent, with only sporadic statistical reports and significant temporal and spatial discontinuities. For example, a certain county only disclosed the recyclable garbage for the whole year of 20xx, and the data for other years were not publicly reported; at the same time, the city where the county is located did not report the city's annual recyclable garbage data. The other steps and parameters are the same as those of specific implementation methods one to three.

[0075] Specific implementation method 5: This implementation method is different from specific implementation methods 1 to 4 in that:

[0076] In step 3, the training sample data set is formed according to the historical explanatory variable data set collected in step 1 and the historical target variable data set collected in step 2. The specific process is as follows:

[0077] The data of the same time and place in the historical explanatory variable data set and the historical target variable data set collected in step 1 are merged into a set of sample data, and finally X sets of sample data are obtained, where X is a positive integer, and the X sets of sample data are used as the training sample data set;

[0078] In step 3, data enhancement is performed on the training sample data set to obtain an enhanced training sample data set. The specific process is as follows:

[0079] Step 31: Perform time shift processing on each training sample data in the training sample data set to obtain a training sample data set after time shift processing; merge the training sample data set after time shift processing with the original training sample data set to obtain a merged training sample data set;

[0080] Step 32: Perform noise injection processing on the merged training sample data set to obtain a training sample data set after noise injection processing;

[0081] Step 33: Generate adversarial sample data using a generative adversarial network based on the training sample data set after noise injection to obtain an adversarial sample data set, and merge the adversarial sample data set with the training sample data set after noise injection to obtain an enhanced training sample data set;

[0082] The other steps and parameters are the same as those in Specific Embodiments 1 to 4.

[0083] Specific implementation method 6: This implementation method is different from specific implementation methods 1 to 5 in that:

[0084] In the step 31, the specific process of performing time shift processing on one of the training sample data to obtain a training sample data after time shift processing is as follows:

[0085] The monthly data in a training sample data is shifted by plus or minus 1 month to obtain the training sample after time shift processing; the training sample data set is expanded, which can improve the subsequent time prediction ability of multi-source solid waste production.

[0086] The specific process of performing noise injection processing on one of the merged training sample data in step 32 to obtain a training sample data after noise injection processing is as follows:

[0087] Adding noise data to the merged training sample data, wherein the noise data obeys a Gaussian distribution with a standard deviation of 0.01, and the noise data is used to simulate statistical errors;

[0088] The generated adversarial network in step 33 is a GAN generated adversarial network

[0089] ; Other steps and parameters are the same as those in Specific Implementation Methods 1 to 5.

[0090] Specific implementation method 7: This implementation method is different from specific implementation methods 1 to 6 in that:

[0091] In step 4, Z base regression model groups are constructed according to the types of garbage in the target variable. The specific process is as follows:

[0092] Each base regression model group includes a convolutional neural network base regression model (CNN), a long short-term memory network base regression model (LSTM), a random forest base regression model (RF), and a gradient boosted decision tree base regression model (GBDT);

[0093] The network structure of the convolutional neural network base regression model is 4 convolutional layers, the convolution kernel size of each convolutional layer is 3x3, and the number of channels of the 4 convolutional layers is 64, 128, 256, and 512 respectively;

[0094] The number of units of the long short-term memory network is 100, and the sequence length of the long short-term memory network is 12;

[0095] The number of decision trees of the random forest-based regression model is 100, and the maximum depth of the random forest-based regression model is 10;

[0096] The learning rate of the gradient boosting decision tree-based regression model is 0.01, the number of decision trees of the gradient boosting decision tree-based regression model is 500, and the maximum depth of the gradient boosting decision tree-based regression model is 5;

[0097] The other steps and parameters are the same as those in Specific Embodiments 1 to 6.

[0098] Specific implementation eight: This implementation differs from specific implementations one to seven in that:

[0099] The specific process of constructing the logistic regression meta-model in step 5 is as follows:

[0100] The regularization parameter (C) of the logistic regression meta-model is set to 1.0 to balance the model complexity and regularization strength;

[0101] The penalty term of the logistic regression meta-model is L2, L2 ridge regression, which adds a penalty term of the sum of squared weights to the loss function. This can help the model avoid over-reliance on certain features, thereby reducing the risk of overfitting.

[0102] The optimization algorithms of the logistic regression meta-model include: lbfgs optimization algorithm and liblinear optimization algorithm. The lbfgs optimization algorithm is an optimization algorithm suitable for large-scale data sets, which can effectively process large-scale feature data and has a fast convergence speed, and can optimize the training process of large-scale data sets.

[0103] If the data set is relatively small, the liblinear optimization algorithm is selected. For smaller data sets, the 'liblinear' optimization algorithm can provide faster training speed.

[0104] The maximum number of iterations of the logistic regression metamodel is set to 1000;

[0105] The other steps and parameters are the same as those in Specific Embodiments 1 to 7.

[0106] Specific implementation method 9: This implementation method is different from specific implementation methods 1 to 8 in that:

[0107] In step 6, a multi-source solid waste data supplement model is constructed based on the Z base regression model groups obtained in step 4 and the logistic regression meta-model obtained in step 5; the specific process is:

[0108] The multi-source solid waste data supplementation model includes the base regression model group and the logistic regression meta-model obtained in step 5 in sequence;

[0109] In step 6, the multi-source solid waste data supplement model is trained according to the enhanced training sample data set obtained in step 3 to obtain a trained multi-source solid waste data supplement model; the specific process is:

[0110] Step 61: Input the enhanced training sample data sets into the Z base regression model groups in the multi-source solid waste data supplementation model, and the Z base regression model groups output R prediction results of solid waste production; R = Z;

[0111] Step 62: Input the R solid waste prediction results into the logistic regression meta-model in the multi-source solid waste data supplementation model, and the logistic regression meta-model outputs the annual multi-source solid waste data;

[0112] Step 63: Train the trained multi-source solid waste data supplement model; the other steps and parameters of the trained multi-source solid waste data supplement model are the same as those of the specific implementation methods one to eight.

[0113] Specific implementation method 10: This implementation method is different from specific implementation methods 1 to 9 in that:

[0114] In the step 61, the yi-th type of solid waste data in the enhanced training sample data set is input into the zi-th base regression model group to obtain the prediction result of the ri-th type of solid waste output, yi∈Y,zi∈Z,ri∈R; the specific process is:

[0115] Step 6: Input the first type of solid waste data in the enhanced training sample data set into the convolutional neural network base regression model of the zith base regression model group to obtain the intermediate result ri1;

[0116] Step 612: Input the first type of solid waste data in the enhanced training sample data set into the long short-term memory network base regression model of the zith base regression model group to obtain the intermediate result ri2;

[0117] Step 613: Input the first type of solid waste data in the enhanced training sample data set into the random forest based regression model of the zith based regression model group to obtain the intermediate result ri3;

[0118] Step 614: input the first type of solid waste data in the enhanced training sample data set into the gradient boosting decision tree based regression model of the zith based regression model group to obtain the intermediate result ri4;

[0119] Step 615: The intermediate results ri1, intermediate results ri1, intermediate results ri1, and intermediate results ri1 constitute the predicted results of the ri-th type of solid waste production.

[0120] The other steps and parameters are the same as those in Specific Implementations 1 to 9.

[0121] Combined with the simulation analysis of specific implementation methods one to ten: Since the statistics on domestic waste generation are often only estimated based on urban treatment facilities, this method is used to fill the scarce data on domestic waste generation in villages and towns around the Bohai Sea from 2001 to 2020, and supports the local forecast of urban and rural domestic waste generation from 2025 to 2040.

[0122] (1) Based on this method, an ensemble learning framework based on multiple machine learning algorithms was constructed. A base regression model was constructed using satellite-ground big data and the amount of domestic waste produced in urban areas from 2001 to 2020. The ensemble learning method based on logistic regression was used to predict the amount of domestic waste in remote rural areas of 41 prefecture-level cities, thus filling the missing rural data.

[0123] (2) By using the existing satellite-ground big data and the linear regression method on a grid basis, we derived the future characteristics of the satellite-ground big data for each city in the future and achieved the coordinated prediction of the amount of domestic waste in urban areas and rural areas. It is estimated that the area will generate 83.1 million tons of domestic waste in 2040.

[0124] The above only describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the above specific implementation methods. Although the present invention has been disclosed as above in the preferred embodiments, it is not used to limit the present invention. Any technician familiar with the profession can make some changes or modify the technical contents disclosed above into equivalent embodiments with equivalent changes without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A multi-source solid waste data supplementation method based on satellite-to-ground big data migration application, characterized in that: The following steps are involved:

1. Construct a multi-source solid waste data supplement model; obtain a trained multi-source solid waste data supplement model; construct a SARIMA model and obtain a trained SARIMA model; 2. Collect explanatory variable data of the multi-source solid waste data area to be supplemented, input the explanatory variable data of the multi-source solid waste data area to be supplemented into the trained multi-source solid waste data supplementation model, and obtain the annual solid waste garbage output of the multi-source solid waste data area to be supplemented; 3. Input the annual solid waste production of the area with multi-source solid waste data to be supplemented into the trained SARIMA model to obtain the monthly solid waste production of the area with multi-source solid waste data to be supplemented.

2. According to claim 1, a multi-source solid waste data supplementation method based on satellite-to-ground big data migration application is characterized in that: In the first step, a multi-source solid waste data supplement model is constructed; and a trained multi-source solid waste data supplement model is obtained; Construct a SARIMA model and obtain a trained SARIMA model; the specific process is: Step 1: Collect historical explanatory variable datasets; The historical explanatory variable data include: annual explanatory variable data; monthly explanatory variable data; Step 2: Collect historical target variable data sets; The historical target variable data set includes: the annual solid waste production in the administrative area, the monthly solid waste production in the administrative area; the number of types of solid waste is Y, and Y is a positive integer; The administrative regions include: city-level administrative regions, county-level administrative regions, and district-level administrative regions; Step 3: Construct a training sample dataset based on the historical explanatory variable dataset collected in step 1 and the historical target variable dataset collected in step 2; Performing data enhancement on the training sample data set to obtain an enhanced training sample data set; Step 4: According to the number of garbage types in the target variable, construct Z base regression model groups, Z = Y; Each base regression model group corresponds to a waste type; Step 5: Construct logistic regression meta-model; Step 6: construct a multi-source solid waste data supplement model based on the Z basis regression model groups obtained in step 4 and the logistic regression meta-model obtained in step 5; train the multi-source solid waste data supplement model based on the enhanced training sample data set obtained in step 3 to obtain a trained multi-source solid waste data supplement model; Step 7: Construct a SARIMA model. According to the enhanced training sample data set obtained in step 3, train the SARIMA model to obtain a trained SARIMA model.

3. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 2 is characterized in that: The explanatory variable data in step 1 include: night light data, population distribution data, age distribution data, vegetation index data, built-up area data, and road network data.

4. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 3 is characterized in that: The number Y of solid waste types in step 2 is 20; the solid waste types include: catering waste, kitchen waste, other waste, hazardous waste, bulk waste; coal gangue, fly ash, smelting slag, chemical waste slag, construction waste; waste mineral oil, waste acid, waste alkali, waste organic solvent, medical waste; waste metal, waste glass, waste fabric, waste paper, waste plastic.

5. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 4 is characterized in that: In step 3, the training sample data set is formed according to the historical explanatory variable data set collected in step 1 and the historical target variable data set collected in step 2. The specific process is as follows: The data of the same year and the same location in the historical explanatory variable dataset and the historical target variable dataset collected in step 1 are merged into a set of annual sample data, and finally X1 sets of annual training sample data are obtained, where X1 is a positive integer. The data of the same month and the same location in the historical explanatory variable dataset and the historical target variable dataset collected in step 1 are merged into a set of monthly sample data, and finally X2 sets of monthly training sample data are obtained, where X2 is a positive integer. Combine the annual training sample data of group X1 and the monthly training sample data of group X2 to form a training sample data set; The historical explanatory variable data in the set of annual training sample data only include the annual solid waste output of one type of waste; The historical explanatory variable data in the set of monthly training sample data only include the monthly solid waste output of one type of waste; In step 3, data enhancement is performed on the training sample data set to obtain an enhanced training sample data set. The specific process is as follows: Step 31: Perform time shift processing on each monthly training sample data in the training sample data set to obtain a training sample data set after time shift processing; merge the training sample data set after time shift processing with the original training sample data set to obtain a merged training sample data set; Step 32: Perform noise injection processing on the merged training sample data set to obtain a training sample data set after noise injection processing; Step 33: Based on the training sample data set after noise injection processing, use the generative adversarial network to generate adversarial sample data to obtain an adversarial sample data set, merge the adversarial sample data set and the training sample data set after noise injection processing to obtain an enhanced training sample data set.

6. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 5 is characterized in that: In the step 31, the specific process of performing time shift processing on one of the training sample data to obtain a training sample data after time shift processing is as follows: Perform a time shift of positive 1 month and a time shift of negative 1 month on a monthly training sample data to obtain a training sample after time shift processing; The specific process of performing noise injection processing on one of the merged training sample data in step 32 to obtain a training sample data after noise injection processing is as follows: Noise data is added to the merged training sample data, and the noise data obeys a Gaussian distribution with a standard deviation of 0.

01. The generative adversarial network in step 33 is a GAN generative adversarial network.

7. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 6 is characterized in that: In step 4, Z base regression model groups are constructed according to the types of garbage in the target variable. The specific process is as follows: Each base regression model group includes a convolutional neural network base regression model, a long short-term memory network base regression model, a random forest base regression model, and a gradient boosting decision tree base regression model; The network structure of the convolutional neural network based regression model is 4 convolutional layers, the convolution kernel size of each convolutional layer is 3x3, and the number of channels of the 4 convolutional layers is 64, 128, 256, and 512 respectively; The number of units of the long short-term memory network is 100, and the sequence length of the long short-term memory network is 12; The number of decision trees of the random forest-based regression model is 100, and the maximum depth of the random forest-based regression model is 10; The learning rate of the gradient boosting decision tree based regression model is 0.01, the number of decision trees of the gradient boosting decision tree based regression model is 500, and the maximum depth of the gradient boosting decision tree based regression model is 5.

8. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 7 is characterized in that: The specific process of constructing the logistic regression meta-model in step 5 is as follows: The regularization parameter C of the logistic regression meta-model is set to 1.0, The penalty term of the logistic regression meta-model is selected as L2, The optimization algorithms of the logistic regression meta-model include: lbfgs optimization algorithm, liblinear optimization algorithm, The maximum number of iterations of the logistic regression meta-model is set to 1000.

9. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 8 is characterized in that: In step 6, a multi-source solid waste data supplement model is constructed based on the Z base regression model groups obtained in step 4 and the logistic regression meta-model obtained in step 5; the specific process is: The multi-source solid waste data supplementation model includes the base regression model group and the logistic regression meta-model obtained in step 5 in sequence; In step 6, the multi-source solid waste data supplement model is trained according to the enhanced training sample data set obtained in step 3 to obtain a trained multi-source solid waste data supplement model; the specific process is: Step 61: Input the enhanced training sample data sets into the Z base regression model groups in the multi-source solid waste data supplementation model, and the Z base regression model groups output R prediction results of solid waste production; R = Z; Step 62: Input the R solid waste prediction results into the logistic regression meta-model in the multi-source solid waste data supplementation model, and the logistic regression meta-model outputs the annual multi-source solid waste data; Step 63: Train the trained multi-source solid waste data supplement model; obtain the trained multi-source solid waste data supplement model.

10. The multi-source solid waste data supplementation method based on satellite-to-ground big data migration application according to claim 9 is characterized in that: In the step 61, the yi-th type of solid waste data in the enhanced training sample data set is input into the zi-th base regression model group to obtain the prediction result of the ri-th type of solid waste output, yi∈Y,zi∈Z,ri∈R; the specific process is: Step 6: Input the first type of solid waste data in the enhanced training sample data set into the convolutional neural network base regression model of the zith base regression model group to obtain the intermediate result ri1; Step 612: Input the first type of solid waste data in the enhanced training sample data set into the long short-term memory network base regression model of the zith base regression model group to obtain the intermediate result ri2; Step 613: Input the first type of solid waste data in the enhanced training sample data set into the random forest based regression model of the zith based regression model group to obtain the intermediate result ri3; Step 614: input the first type of solid waste data in the enhanced training sample data set into the gradient boosting decision tree based regression model of the zith based regression model group to obtain the intermediate result ri4; Step 615: The intermediate results ri1, intermediate results ri1, intermediate results ri1, and intermediate results ri1 constitute the predicted results of the ri-th type of solid waste production.

Images