System for predicting domestic sewage and garbage based on urban color

Abstract

The present application relates to the field of environmental protection, and particularly relates to a system for predicting domestic sewage and garbage based on urban color. The present application aims at the problem of low accuracy and timeliness of output results, and is difficult to accurately and timely grasp the change trend of domestic pollution load. By using the fusion and transformation of high-resolution urban street scene images and satellite remote sensing images, the color, texture and other landscape features in the images are extracted, and based on the deep learning framework, an innovative prediction model framework of domestic pollution load based on image features is constructed, realizing the analysis of the change characteristics of domestic pollution in time and space, fitting the current AI+GIS application scene, and solving the problem of lag and roughness of macro data characteristics such as social development and population statistics. Compared with the traditional model based on social panel statistical data, the present application has the advantages of short time step and strong spatial details.

Description

Technical Field

[0001] This invention relates to the field of environmental protection technology, specifically to a system for predicting domestic sewage and garbage based on urban color. Background Technology

[0002] Population growth and urbanization have made domestic pollution a significant source of pollution affecting urban environmental quality, drawing widespread attention. In recent years, my country's annual domestic waste collection volume has reached 250 million tons, and urban sewage discharge has reached 50.2 billion cubic meters. To understand the changing trends of domestic pollution loads and achieve effective management, it is necessary to predict pollution loads. Socioeconomic data and demographic data profoundly influence the generation of domestic sewage and waste, thus the accuracy of pollution load predictions often depends on these macro-level data. However, macro-level data such as social development and demographic data suffer from lag and crudeness due to long data collection and processing times and low update frequencies. This results in prediction models based on traditional methods having low accuracy and timeliness, making it difficult to accurately and promptly grasp the changing trends of domestic pollution loads. Summary of the Invention

[0003] This invention addresses the problem that prediction models built based on traditional methods suffer from low accuracy and timeliness in their output results, making it difficult to accurately and promptly grasp the changing trends of pollution loads from residential sources.

[0004] Data acquisition module, intelligent analysis module, and result output module;

[0005] The data acquisition module is used to collect street view, remote sensing images, urban color features, nighttime lights, and population data of the application area.

[0006] The intelligent analysis module is used to construct a fusion model based on the Stacking method; and to obtain a trained fusion model based on the Stacking method.

[0007] The trained fusion model based on the Stacking method is used to predict the quality of domestic sewage and domestic waste generated based on street view, remote sensing images, urban color features, nighttime lights and population of the application area.

[0008] The result output module is used to output the prediction results of the intelligent analysis module.

[0009] The specific process of constructing a fusion model based on the Stacking method and obtaining a trained fusion model based on the Stacking method is as follows:

[0010] Step 1: Obtain the training set and input features of the fusion model;

[0011] Step 2: Construct a fusion model based on the Stacking method to obtain a trained fusion model using the Stacking method.

[0012] The specific process for obtaining the training set of the fusion model in step one is as follows:

[0013] A1: Obtain the regression relationship dataset; the regression relationship dataset includes a regression relationship training set and a regression relationship test set;

[0014] A2: Construct four basic models, and train the four basic models using a regression relationship training set to obtain four well-trained basic models.

[0015] The analysis results of the four basic models on the regression relationship training set are used as input features for the fusion model;

[0016] The analysis results of the four trained basic models on the regression relationship test set were used as the training set for the fusion model.

[0017] In step two, a fusion model based on the Stacking method is constructed to obtain a trained Stacking method fusion model. The specific process is as follows:

[0018] Linear regression was chosen as the fusion model based on the Stacking method. The fusion model was trained using the input features and training set of the fusion model. The hyperparameters of the fusion model, such as regularization parameters, learning rate, and number of iterations, were adjusted to obtain the trained fusion model based on the Stacking method.

[0019] Beneficial Effects: This method addresses the shortcomings of traditional prediction methods for domestic sewage and waste pollution loads, which rely on outdated and coarse macro-level data such as social development and demographics. It proposes a deep learning framework that transforms and integrates high-resolution urban street view images and satellite remote sensing imagery, extracting color and texture features to construct an innovative prediction model architecture for domestic pollution loads based on these image features. Compared to traditional models built on social panel statistical data, this approach offers advantages such as shorter time steps and greater spatial detail.

[0020] This invention, for the first time, utilizes the fusion and transformation of high-resolution urban street view images and satellite remote sensing images to extract landscape features such as color and texture from the images. Based on a deep learning framework, it constructs an innovative predictive model framework for domestic pollution load based on image features, enabling the analysis of the temporal and spatial characteristics of domestic pollution changes. Aligned with current AI+GIS application scenarios, it addresses the issues of lagging and coarseness in macro-level data features such as social development and demographics. Compared to traditional models built based on social panel statistical data, it offers advantages such as shorter time steps and stronger spatial detail. The patented method's application of GIS+AI provides powerful macro-level planning and intelligent decision-making capabilities, with broad application prospects in the field of intelligent management of domestic pollution in different urban and rural scenarios. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a system process for predicting domestic sewage and garbage based on urban color, according to the present invention. Detailed Implementation

[0022] This invention proposes a system for predicting domestic sewage and garbage based on urban color.

[0023] Specific implementation method one: Combining Figure 1 This embodiment describes a system for predicting domestic sewage and waste based on urban color analysis, comprising:

[0024] Data acquisition module, intelligent analysis module, and result output module;

[0025] The data acquisition module is used to collect street view, remote sensing images, urban color features, nighttime lights, and population data of the application area.

[0026] The intelligent analysis module is used to construct a fusion model based on the Stacking method; and to obtain a trained fusion model based on the Stacking method.

[0027] The trained fusion model based on the Stacking method is used to predict the quality of domestic sewage and domestic waste generated based on street view, remote sensing images, urban color features, nighttime lights and population of the application area.

[0028] The output module is used to output the prediction results of the intelligent analysis module. This method addresses the problem that traditional predictions of environmental pollution loads from domestic sources, such as sewage and waste, rely on macro-level data features like social development and demographics, which are often outdated and coarse. It proposes an innovative prediction model architecture for domestic pollution loads based on a deep learning framework. This model transforms and integrates high-resolution urban street view images and satellite remote sensing imagery, extracting color and texture features from the images to construct an innovative prediction model architecture for domestic pollution loads based on these image features. Compared to traditional models built on social panel statistical data, this model has advantages such as shorter time steps and stronger spatial detail.

[0029] This invention, for the first time, utilizes the fusion and transformation of high-resolution urban street view images and satellite remote sensing images to extract landscape features such as color and texture from the images. Based on a deep learning framework, it constructs an innovative predictive model framework for domestic pollution load based on image features, enabling the analysis of the temporal and spatial characteristics of domestic pollution changes. Aligned with current AI+GIS application scenarios, it addresses the issues of lagging and coarseness in macro-level data features such as social development and demographics. Compared to traditional models built based on social panel statistical data, it offers advantages such as shorter time steps and stronger spatial detail. The patented method's application of GIS+AI provides powerful macro-level planning and intelligent decision-making capabilities, with broad application prospects in the field of intelligent management of domestic pollution in different urban and rural scenarios.

[0030] Specific Implementation Method Two: The difference between this implementation method and Specific Implementation Method One is that...

[0031] The specific process of constructing the fusion model based on the Stacking method in the first step and obtaining the trained fusion model based on the Stacking method is as follows:

[0032] Step 1: Obtain the training set and input features of the fusion model;

[0033] Step 2: Construct a fusion model based on the Stacking method to obtain a trained fusion model using the Stacking method.

[0034] The other steps and parameters are the same as in Specific Implementation Method 1.

[0035] Specific Implementation Method Three: The difference between this implementation method and Specific Implementation Method Two is that...

[0036] The specific process for obtaining the training set and input features of the fusion model in step one is as follows:

[0037] A1: Obtain the regression relation dataset; the regression relation dataset includes a regression relation training set and a regression relation test set;

[0038] A2: Construct four basic models, input the regression relationship training set into the four basic models respectively for training, and use all the output results of the four basic models during the training process as the input features of the fusion model until convergence is obtained to obtain four trained basic models.

[0039] A3: Input the regression relationship test set into the four pre-trained basic models respectively, and use the output results of the four pre-trained basic models as the training set of the fusion model. Other steps and parameters are the same as in Implementation Method 2.

[0040] Specific Implementation Method Four: The difference between this implementation method and Specific Implementation Method Three is that...

[0041] The regression relationship dataset is obtained from A1; the specific process is as follows:

[0042] S1: Collect road network data for the application area.

[0043] S2: Acquire Landsat satellite remote sensing images of the application area of ​​the invention, and then perform radiometric correction, geometric correction and noise reduction operations on the Landsat satellite remote sensing images in sequence to obtain preprocessed satellite remote sensing data; ensure the image quality of the satellite remote sensing data;

[0044] S3: Establish the correlation between the road network data collected in S1 and the satellite remote sensing data collected in S2;

[0045] S4: Obtain the city color dataset based on the correlation between the road network data collected by S1 and the satellite remote sensing data collected by S2 obtained by S3;

[0046] S5: Obtain the regression relationship dataset based on the city color dataset obtained in S4.

[0047] The other steps and parameters are the same as in Specific Implementation Method Three.

[0048] Specific Implementation Method Five: The difference between this implementation method and Specific Implementation Method Four is that...

[0049] The specific process of collecting road network data of the application area in S1 is as follows:

[0050] S1.1: Generating sampling points: The specific process is as follows:

[0051] Based on the distance point generation tool in ArcMap software, sampling points at equal intervals of 2000 meters are generated in the application area, and each sampling point is numbered to obtain the sampling point sequence number;

[0052] S1.2: Obtain the current street view image and historical street view image at the sampling point location in S1.1; the specific process is as follows:

[0053] A Python web crawler based on the Baidu Street View API is constructed to extract current and historical street view images of sampling points, while recording the acquisition time of each street view image; that is, the road network data of the application area. Other steps and parameters are the same as in Implementation Method Four.

[0054] Specific Implementation Method Six: The difference between this implementation method and Specific Implementation Method Five is that...

[0055] The process of establishing the correlation between the road network data collected in S1 and the satellite remote sensing data collected in S2 in step S3 is as follows:

[0056] S3.1: Divide the application area of ​​the invention into multiple Thiessen polygon segmentation spatial units; the specific process is as follows:

[0057] Based on the sampling points generated in step 1, the application area of ​​the invention is divided into N Thiessen polygon segmentation spatial units using the Thiessen polygon algorithm. Each segmented spatial unit has one and only one sampling point; N is a positive integer.

[0058] S3.2: Use the GDAL module toolkit to segment the preprocessed satellite remote sensing image from S3.1 into spatial units;

[0059] S3.3: Establish the correlation between sampling points and segmented satellite remote sensing images.

[0060] Other steps and parameters are the same as in Specific Implementation Method 5.

[0061] Specific Implementation Method Seven: The difference between this implementation method and Specific Implementation Method Six is ​​that...

[0062] In step S4, the city color dataset is obtained based on the correlation between the road network data collected in S1 (obtained in step S3) and the satellite remote sensing data collected in S2; the specific process is as follows:

[0063] S4.1: Obtain the "Image Source-Color-Percentage" feature. The specific process is as follows:

[0064] S4.1: The specific process for obtaining the "Image Source-Color-Percentage" feature is as follows:

[0065] S4.1.1: The colors in the street view image obtained in S1 and the remote sensing image obtained in S2 are clustered into nine color categories using a convolutional neural network algorithm; the nine color categories are: red, orange, yellow, green, cyan, blue, purple, black, and white;

[0066] S4.1.2: Calculate the proportion of each color class in each image to all colors in the corresponding image at each sampling point;

[0067] S4.1.3: For each sampling point, the proportion of each color class in each image to all colors in the corresponding image and the corresponding image source are used as the "image source-color-percentage" feature, such as "street view-red-30%" and "remote sensing-green-20%".

[0068] Based on the convolutional neural network algorithm, a recognition model is constructed for the main colors contained in the street view image obtained in step 2 and the remote sensing image obtained in step 4. The colors in the images are clustered into red, orange, yellow, green, cyan, blue, purple, black, and white, and the color proportion of the above nine categories in the street view image and remote sensing image corresponding to a certain sampling point is calculated. The proportion of each color is used as a feature and stored as "image source-color-percentage", for example, "street view-red-30%".

[0069] S4.2: Obtain “population” characteristics; the specific process is as follows:

[0070] Based on the Thiessen polygons obtained by S3.1, the UN population density grid data for each year is segmented to obtain the number of residents within each sampling point. This feature reflects the resident population within the Thiessen polygon corresponding to the sampling point, and the total number of residents within the Thiessen polygon area is taken as the "population" feature.

[0071] S4.3: Obtain the "nighttime light" feature; the specific process is as follows:

[0072] Based on the Thiessen polygons obtained in S3.1, the NPP-VIIRS nighttime light datasets from different years are segmented to obtain the nighttime light intensity within each sampling point. This feature reflects the intensity of human activity within the Thiessen polygon corresponding to the sampling point. The sum of the nighttime light intensity within the Thiessen polygon region is taken as the "nighttime light" feature.

[0073] S4.4: Constructing a city color dataset, the specific process is as follows:

[0074] The data obtained in S1.1 ("Sampling Point Number"), S1.2 ("Image Sampling Time"), S4.1 ("Image Source-Color-Percentage"), S4.2 ("Population"), and S4.3 ("Nighttime Lights") are integrated according to their temporal relationship. Data from the same time point ("Sampling Point Number", "Image Sampling Time", "Image Source-Color-Percentage", "Population", and "Nighttime Lights") are grouped together to form a data set. All data sets are then combined to form a city color dataset. Other steps and parameters are the same as in Specific Implementation Method Six.

[0075] Specific Implementation Method Eight: The difference between this implementation method and Specific Implementation Method Seven is that...

[0076] S5.1: Obtain data on the amount of domestic waste and domestic sewage generated. The specific process is as follows:

[0077] The statistical spatial scope is based on cities / counties / districts, and the statistical time scope is based on years, and the data includes the amount of domestic waste generated and the amount of domestic sewage generated.

[0078] The statistical space to which the statistical data belongs is used as the "administrative region" feature, and the data sampling time of the statistical data is used as the "time" feature; the amount of domestic waste generated is used as the "waste amount" feature, and the amount of domestic sewage generated is used as the "sewage amount" feature.

[0079] S5.2: Determine all sampling points within the spatial range corresponding to the data on domestic waste generation and domestic sewage generation obtained in S5.1;

[0080] S5.3: Add the data corresponding to all sampling points determined in S5.2 in the city color dataset with the data on domestic waste generation and domestic sewage generation obtained in S5.1 to construct a dataset containing "administrative region", "time", "image source-color-percentage", "population", "night lights", "sewage volume" and "waste volume", i.e., a regression relationship dataset; other steps and parameters are the same as in specific implementation method seven.

[0081] Specific Implementation Method Nine: The difference between this implementation method and Specific Implementation Method Eight is that...

[0082] In A2, four basic models are constructed. These four basic models are trained using a regression relationship training set to obtain four well-trained basic models. The specific process is as follows:

[0083] Linear regression, support vector machine, random forest, and recurrent neural network are used as four basic models to analyze the regression relationship between various features and "sewage volume" and "garbage volume".

[0084] The training set of the regression relation dataset is divided into a first subset, a second subset, a third subset, and a fourth subset;

[0085] For linear regression, the first subset is used for training, and parameters such as regularization parameters, learning rate, and number of iterations of the linear regression model are adjusted to obtain a trained linear regression model.

[0086] For support vector machines, the second subset is used for training, and hyperparameters such as the kernel function, C parameter, and class function of the vector machine model are adjusted to obtain a trained vector machine model.

[0087] For random forests, a third subset is used for training, and parameters such as the number of trees, tree depth, minimum split sample number, and minimum leaf node sample number are adjusted to obtain a trained random forest model.

[0088] For recurrent neural networks, the fourth subset is used for training, and hyperparameters such as the number of hidden layer units, learning rate, and number of iterations of the recurrent neural network model are adjusted to obtain a trained recurrent neural network model.

[0089] The regression relationship training set is input into four basic models for training. The output of the four basic models during the training process is the regression relationship between various features in the regression relationship training set and "sewage volume" and "garbage volume".

[0090] In A3, the regression relationship test set is input into four trained basic models, and the output of the four trained basic models are: the regression relationships between various features in the regression relationship test set and "sewage volume" and "garbage volume".

[0091] The other steps and parameters are the same as in Specific Implementation Method Eight.

[0092] Specific Implementation Method Ten: The difference between this implementation method and Specific Implementation Method Nine is that...

[0093] In step two, a fusion model based on the Stacking method is constructed to obtain a trained Stacking method fusion model. The specific process is as follows:

[0094] Linear regression was chosen as the fusion model based on the Stacking method. The fusion model was trained using its input features and training set. Hyperparameters such as regularization, learning rate, and number of iterations were adjusted to obtain the trained Stacking-based fusion model.

[0095] The other steps and parameters are the same as in Specific Implementation Method Nine.

[0096] The above description is merely of preferred embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention, and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A system for predicting domestic sewage and waste based on urban color, characterized in that, include: Data acquisition module, intelligent analysis module, prediction module, and result output module; The data acquisition module is used to collect street view images, remote sensing images, urban color features, nighttime lights of the area, and population data of the application area to be tested. The intelligent analysis module is used to construct a fusion model based on the Stacking method, and obtain a trained fusion model based on the Stacking method; the specific process is as follows: Step 1: Obtain the training set and input features of the fusion model; The training set and input features of the fusion model are obtained from the regression relation dataset; the regression relation dataset includes: a regression relation training set and a regression relation test set; the process of obtaining the regression relation dataset is as follows: S1: Collect road network data of sampling points in the application area; S2: Collect Landsat satellite remote sensing imagery of the application area, and then perform radiometric correction, geometric correction and noise reduction operations on the Landsat satellite remote sensing imagery in sequence to obtain preprocessed satellite remote sensing data; S3: Construct the correlation between the road network data collected in S1 and the satellite remote sensing data collected in S2. S4: Obtain the city color dataset based on the correlation between the road network data collected in S1 (obtained through S3) and the satellite remote sensing data collected in S2; the specific process is as follows: S4.1: The specific process for obtaining the "Image Source-Color-Percentage" feature is as follows: S4.1.1: Input the street view image from the road network data obtained in S1 into a convolutional neural network for processing, and cluster the colors in the image into nine color categories; input the remote sensing image obtained in S2 into a convolutional neural network for processing, and cluster the colors in the image into nine color categories; the nine color categories are: red, orange, yellow, green, cyan, blue, purple, black, and white; S4.1.2: Calculate the proportion of each color category in each image to all colors in the corresponding image at each sampling point; S4.1.3: Use the proportion of each color category in each image to all colors in the corresponding image at each sampling point and the corresponding image source as the "image source-color-percentage" feature; the image source includes: street view images and remote sensing images; S4.2: Obtain the "population" feature; S4.3: Obtain the "nighttime light" feature; S4.4: Construct a city color dataset. The specific process is as follows: Integrate the data of "sampling point number", "image sampling time", "image source-color-percentage" obtained in S4.1, "population" obtained in S4.2, and "night lights" obtained in S4.3 according to the time relationship. Group the data of "sampling point number", "image sampling time", "image source-color-percentage", "population" and "night lights" at the same time point into data groups, and combine all data groups to form a city color dataset. S5: Obtain the regression relationship dataset based on the city color dataset obtained in S4; the specific process is as follows: S5.1: Obtain data on the amount of domestic waste generated and the amount of domestic sewage generated. The specific process is as follows: the statistical spatial scope is based on the city / county / district, and the statistical time scope is based on the year. Collect data on the amount of domestic waste generated and the amount of domestic sewage generated. Use the statistical space to which the statistical data belongs as the "administrative region" feature, and the data sampling time of the statistical data as the "time" feature. Use the obtained statistical amount of domestic waste generated as the "waste amount" feature, and the statistical amount of domestic sewage generated as the "sewage amount" feature. S5.2: Determine all sampling points within the spatial range corresponding to the data on domestic waste generation and domestic sewage generation obtained in S5.1; S5.3: Add the data corresponding to all sampling points determined in S5.2 in the city color dataset with the data on domestic waste generation and domestic sewage generation obtained in S5.1 to construct a dataset containing features such as "administrative region", "time", "image source-color-percentage", "population", "night lights", "sewage amount" and "waste amount", i.e., regression relation dataset; Step 2: Construct a fusion model based on the Stacking method to obtain a trained fusion model using the Stacking method; The prediction module is used to input the test data collected by the data acquisition module into the trained fusion model based on the Stacking method, and output the predicted amount of domestic sewage and domestic waste. The result output module is used to output the prediction results of the prediction module.

2. The system for predicting domestic sewage and garbage based on urban color as described in claim 1, characterized in that, The specific process of obtaining the training set and input features of the fusion model from the regression relation dataset in step one is as follows: A1: Construct four basic models, input the regression relationship training set into the four basic models respectively for training, and use all the output results of the four basic models during the training process as the input features of the fusion model. Train until convergence to obtain four well-trained basic models. A2: Input the regression relationship test set into the four trained basic models respectively, and use the output results of the four trained basic models as the training set of the fusion model.

3. The system for predicting domestic sewage and garbage based on urban color as described in claim 2, characterized in that, The specific process of collecting road network data from sampling points in the application area in S1 is as follows: S1.1: Generating sampling points: The specific process is as follows: Based on the distance point generation tool in ArcMap software, sampling points at equal intervals of 2000 meters are generated in the application area, and each sampling point is numbered to obtain the sampling point sequence number; S1.2: Obtain the current street view image and historical street view image at the sampling point location in S1.1; the specific process is as follows: A Python web crawler based on the Baidu Street View API extracts current and historical street view images of sampling points, while recording the acquisition time of each street view image; that is, the road network data of the application area.

4. The system for predicting domestic sewage and garbage based on urban color as described in claim 3, characterized in that, The process of establishing the correlation between the road network data collected in S1 and the satellite remote sensing data collected in S2 in step S3 is as follows: S3.1: Divide the application area into N Thiessen polygon segmentation spatial units; the specific process is as follows: Based on the sampling points generated in S1.1, the application area is divided into N Thiessen polygon segmentation spatial units using the Thiessen polygon algorithm. Each segmented spatial unit contains exactly one sampling point; N is a positive integer. S3.2: Use the GDAL module toolkit to segment the preprocessed satellite remote sensing image from S3.1 into spatial units; S3.3: Establish the correlation between sampling points and segmented satellite remote sensing images.

5. A system for predicting domestic sewage and garbage based on urban color as described in claim 4, characterized in that, The "population" feature is obtained in S4.2; The specific process is as follows: Based on the Thiessen polygons obtained in S3.1, the population density grid data for each year is divided to obtain the number of residents within each sampling point. The total number of residents within the Thiessen polygon area is used as the "population" feature. The "nighttime light" feature is obtained in S4.3; The specific process is as follows: Based on the Thiessen polygons obtained in S3.1, the NPP-VIIRS nighttime light dataset for each year is segmented to obtain the nighttime light intensity within each sampling point. The sum of the nighttime light intensity within the Thiessen polygon region is used as the "nighttime light" feature.

6. A system for predicting domestic sewage and garbage based on urban color as described in claim 5, characterized in that, In A1, four basic models are constructed. These four basic models are trained using a regression relationship training set to obtain four well-trained basic models. The specific process is as follows: The four basic models are linear regression, support vector machine, random forest, and recurrent neural network; The regression relation training set is divided into a first subset, a second subset, a third subset, and a fourth subset; For linear regression, the first subset is used for training, and the parameters of the linear regression model are adjusted, including: regularization parameter, learning rate, and number of iterations; thus, a trained linear regression model is obtained. For the support vector machine, a second subset is used for training, and the hyperparameters of the vector machine model are adjusted. The hyperparameters of the vector machine model include: kernel function, C parameter, and class function; thus, a trained vector machine model is obtained. For random forests, a third subset is used for training, and the parameters of the random forest model are adjusted. The parameters of the random forest model include: the number of trees, the depth of the trees, the minimum number of split samples, and the minimum number of leaf node samples; thus, a trained random forest model is obtained. For recurrent neural networks, the fourth subset is used for training, and the hyperparameters of the recurrent neural network model are adjusted. The hyperparameters of the recurrent neural network model include: the number of hidden layer units, the learning rate, and the number of iterations; thus, a trained recurrent neural network model is obtained. The output of the four basic models during training is as follows: regression relationships between various features in the training set and the features of "sewage volume" and "garbage volume", respectively; In A2, the regression relationship test set is input into four trained basic models. The output of the four trained basic models is the regression relationship between each feature in the regression relationship test set and the features of "sewage volume" and "garbage volume".

7. A system for predicting domestic sewage and garbage based on urban color as described in claim 6, characterized in that, In step two, a fusion model based on the Stacking method is constructed to obtain a trained Stacking method fusion model. The specific process is as follows: Linear regression algorithm is selected as the fusion model based on the Stacking method. The fusion model based on the Stacking method is trained using the input features and training set of the fusion model. The hyperparameters of the fusion model based on the Stacking method are adjusted, including regularization parameter, learning rate, and number of iterations, to obtain the trained fusion model based on the Stacking method.

Images