Municipal solid waste output and component joint prediction method and system based on deep learning

Abstract

The invention discloses a deep learning-based municipal solid waste output and component combined prediction method and system. The method comprises the following steps of: constructing a comprehensive data set of a city; performing feature screening on the related influence factor data based on a Pearson correlation coefficient and grey correlation analysis to obtain key features; constructing a municipal solid waste output prediction model and a component prediction model based on a deep learning network, and training the models by using the comprehensive data set; based on a recursive feature self-updating mechanism, inputting the key features into the trained municipal solid waste output prediction model to obtain the predicted municipal solid waste output; and combining the predicted garbage generation amount with the key features, inputting the combined amount into the trained component prediction model to obtain the predicted household garbage component amount, and introducing a total amount consistency constraint condition to obtain the proportion of each component of the household garbage. The method can effectively predict the household garbage output and components of the target city.

Description

Technical Field

[0001] The present invention belongs to the technical field of municipal domestic waste management, and in particular relates to a method and system for jointly predicting the amount of municipal domestic waste generated and its components based on deep learning. Background Art

[0002] The changes in the amount of urban domestic waste generated are affected by the combined effects of multiple complex factors, and the degree of influence of each factor varies. Scientifically and accurately predicting the amount of domestic waste generated and the proportion of its components is of vital importance to urban management and environmental protection.

[0003] Existing methods for predicting municipal waste rely heavily on traditional statistical models or limited machine learning techniques, such as random forests and support vector machines. These methods are often based on small datasets or consider only a limited number of influencing factors, limiting the accuracy of predictions when faced with complex environmental changes across different years. Existing models have significant limitations in data usage and lack comprehensive integration of multidimensional data. Furthermore, some models rely too heavily on simple correlation analysis for feature selection, failing to accurately measure the complex impact of multidimensional features on the amount of municipal solid waste generated and the proportions of its components. Furthermore, there are also inconsistencies in the acquisition of future feature values.

[0004] This shows that, despite the outstanding prediction performance of deep learning models in other fields, their application in municipal solid waste management prediction is still relatively scarce. Therefore, to address the above issues, it is urgent to propose a deep learning-based joint prediction method and system for municipal solid waste generation and its composition, which has important practical significance for the effective management of municipal solid waste and the optimal allocation of resources. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a method and system for jointly predicting the amount of urban domestic waste generated and its components based on deep learning. This method and system can accurately predict the amount of urban domestic waste generated and the proportion of its components, provide data support for urban waste management and resource utilization, and solve the problems existing in the above-mentioned existing technologies.

[0006] To achieve the above objectives, the present invention provides a method for jointly predicting the amount of municipal solid waste generated and its components based on deep learning, comprising the following steps:

[0007] Constructing a comprehensive data set for the city, the comprehensive data set including data on the amount of municipal solid waste generated and related influencing factors, as well as data on waste composition;

[0008] Based on the Pearson correlation coefficient and grey relational analysis, the relevant influencing factor data are subjected to feature screening to obtain key features that affect the amount of municipal solid waste generated and the waste composition data;

[0009] Constructing a municipal solid waste generation amount prediction model and a component prediction model based on a deep learning network, and training the municipal solid waste generation amount prediction model and the component prediction model using the comprehensive dataset;

[0010] Based on the recursive feature self-update mechanism, the key features that affect the amount of municipal solid waste generated are input into the trained municipal solid waste generation prediction model to obtain the predicted amount of municipal solid waste generated.

[0011] The predicted amount of garbage generated is combined with the key features that affect the garbage component data and input into the trained component prediction model to obtain the predicted amount of domestic garbage components. At the same time, the total amount consistency constraint is introduced to obtain the proportion of each component of domestic garbage.

[0012] Optionally, after constructing the comprehensive data set of the city, the method further includes: performing data cleaning and normalization processing on the comprehensive data set.

[0013] Optionally, the process of performing feature screening on the relevant influencing factor data based on the Pearson correlation coefficient and grey relational analysis to obtain key features that affect the amount of municipal solid waste generated and the waste composition data includes:

[0014] The linear relationship between the relevant influencing factor data is quantified based on the Pearson correlation coefficient, and the nonlinear relationship between the relevant influencing factor data is captured based on grey correlation analysis. The linear and nonlinear relationships are weighted and integrated to obtain the key features that affect the generation of urban domestic waste and the waste composition data.

[0015] Optionally, the calculation formula of the Pearson correlation coefficient is:

[0016]

[0017] Among them, x i and y i are the data points of the two variables respectively; and is the mean of the two variables; n is the total number of data points.

[0018] Optionally, the calculation formula of the grey relational analysis is:

[0019]

[0020] Where m is the total number of data points, ρ is the resolution coefficient, is the reference sequence X0 and the comparison sequence X i The absolute difference between the data points.

[0021] Optionally, the architecture of the deep learning network includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network and a multi-head attention mechanism model.

[0022] Optionally, the one-dimensional convolutional neural network includes an input layer, a plurality of hidden layers and an output layer;

[0023] The hidden layer includes several repeatedly stacked convolution layers and pooling layers. In the convolution layer, a convolution kernel is used to perform a convolution operation on the input feature map to establish an input feature map; the input feature map and the output of the neuron are connected to obtain the output of the neuron; the sigmoid activation function is used as a bridge to connect the input feature map and the output of the neuron to provide input for the next layer of the network.

[0024] Optionally, the bidirectional long short-term memory network includes a forget gate, and the forget gate determines the degree of retention of input information through a sigmoid activation function.

[0025] Optionally, the method for training the municipal solid waste generation amount prediction model and the component prediction model includes: grid search and cross validation.

[0026] The present invention also provides a deep learning-based joint prediction system for municipal domestic waste generation and its components, which is used to implement the method described above and includes: a data acquisition module, a feature screening module, a model building module, a waste prediction module, and a component prediction module;

[0027] The data acquisition module is used to construct a comprehensive data set of the city, which includes data on the amount of urban domestic waste generated and related influencing factors, as well as data on waste composition;

[0028] The feature screening module is used to perform feature screening on the relevant influencing factor data based on the Pearson correlation coefficient and grey relational analysis to obtain key features that affect the amount of municipal solid waste generated and the waste composition data;

[0029] The model building module is used to build a municipal solid waste generation amount prediction model and a component prediction model based on a deep learning network, and use the comprehensive data set to train the municipal solid waste generation amount prediction model and the component prediction model;

[0030] The garbage prediction module is used to input key features that affect the amount of municipal solid waste generated into the trained municipal solid waste generation prediction model based on a recursive feature self-update mechanism to obtain the predicted amount of municipal solid waste generated;

[0031] The component prediction module is used to combine the predicted garbage generation amount with the key features that affect the garbage component data and input them into the trained component prediction model to obtain the predicted amount of domestic garbage components, while introducing the total amount consistency constraint condition to obtain the proportion of each component of domestic garbage.

[0032] Compared with the prior art, the present invention has the following advantages and technical effects:

[0033] Multi-source data fusion: This invention collects and integrates statistical yearbooks, municipal management, and environmental monitoring platform data of the target city or urban agglomeration, ensuring the comprehensiveness and accuracy of the data set, and providing rich feature information for subsequent predictions.

[0034] Accurate feature screening: This invention adopts a feature screening method that combines Pearson correlation coefficient and grey correlation analysis, which can simultaneously capture linear and nonlinear relationships in the data, effectively improving the comprehensiveness and accuracy of feature selection, and ensuring that the extracted features can best reflect the key factors of domestic waste generation and composition.

[0035] Efficient deep learning model: By integrating a one-dimensional convolutional neural network, a bidirectional long short-term memory network, and a multi-head attention mechanism model, the proposed deep learning model can fully utilize the local features, long-term dependencies, and attention mechanisms of important features in time series data, thereby improving the accuracy of the prediction of domestic waste generation and its component ratios.

[0036] Recursive feature self-update mechanism: Based on the recursive feature self-update mechanism and deep learning model, the present invention can reasonably predict future features, thereby providing more accurate prediction results.

[0037] Combined Forecasting and Calibration: This method combines the predicted results of household waste generation with the component prediction model and introduces a total consistency constraint to ensure the accuracy and consistency of each component. This calibration process makes the final prediction more reliable and accurately reflects the actual situation.

[0038] In summary, the present invention provides an efficient, accurate and reliable solution for the prediction and management of urban domestic waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0040] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0041] Figure 2This is a block diagram of the deep learning network structure of an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0043] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0044] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0045] Example 1

[0046] like Figure 1 As shown, this embodiment provides a method for jointly predicting the amount of municipal solid waste generated and its components based on deep learning, including the following steps:

[0047] Constructing a comprehensive data set for the city, the comprehensive data set including data on the amount of municipal solid waste generated and related influencing factors, as well as data on waste composition;

[0048] Based on the Pearson correlation coefficient and grey relational analysis, the relevant influencing factor data are subjected to feature screening to obtain key features that affect the amount of municipal solid waste generated and the waste composition data;

[0049] Constructing a municipal solid waste generation amount prediction model and a component prediction model based on a deep learning network, and training the municipal solid waste generation amount prediction model and the component prediction model using the comprehensive dataset;

[0050] Based on the recursive feature self-update mechanism, the key features that affect the amount of municipal solid waste generated are input into the trained municipal solid waste generation prediction model to obtain the predicted amount of municipal solid waste generated.

[0051] The predicted amount of garbage generated is combined with the key features that affect the garbage component data and input into the trained component prediction model to obtain the predicted amount of domestic garbage components. At the same time, the total amount consistency constraint is introduced to obtain the proportion of each component of domestic garbage.

[0052] The feasible process of multi-source heterogeneous data collection and fusion includes:

[0053] This example ensures the comprehensiveness and representativeness of the data used by collecting relevant data for the target city or urban cluster from city statistical yearbooks, municipal management data, and environmental monitoring platforms. By integrating these data sources, this example expands the time span and feature dimensions of data related to household waste generation and composition, significantly improving the stability and reliability of the model's predictions.

[0054] As a specific implementation, this example draws on data from sources including the Shanghai Statistical Yearbook, the Shanghai Greening Statistical Yearbook, and monitoring data from the Institute of Environmental Science and Engineering. Data on municipal solid waste generation and related influencing factors are extracted from these data, along with waste composition data from the Institute of Environmental Science and Engineering. This dataset covers multiple years and includes multi-dimensional features such as time series and socioeconomic characteristics, providing a solid foundation for building a predictive model.

[0055] The feasible process of data preprocessing stage includes:

[0056] First, data cleaning is performed. This process involves identifying and removing noisy data and filling missing values in the data to improve data quality. Subsequently, normalization is performed to convert data of different units and magnitudes to the same scale to ensure data consistency and improve the accuracy of model training. Furthermore, for some missing values in the data, this embodiment uses techniques such as mean filling or nearest neighbor interpolation to ensure the integrity and reliability of the dataset.

[0057] In order to enhance the generalization ability of the model, this embodiment also includes a data segmentation step, which divides the corresponding data into a training set, a validation set, and a test set in a ratio of 8:1:1, aiming to ensure the stability of the model during training and maintain the fairness of the evaluation in the testing phase, thereby improving the predictive performance and generalization ability of the model in practical applications.

[0058] The feature engineering adopted in this embodiment may include the following steps:

[0059] In terms of feature screening, the initial screening process was based on solid waste management domain knowledge. A feature screening method combining Pearson correlation coefficient and grey relational analysis was then used to comprehensively evaluate influencing factors. By weighting the Pearson and grey relational results, a secondary screening process was conducted to identify key features that significantly impacted the amount or composition of municipal solid waste. This approach ensured that the selected features accurately reflected the interactions of multiple factors, thereby improving the model's predictive accuracy and efficiency.

[0060] Furthermore, Pearson correlation analysis was used to evaluate the linear relationship, and grey relational analysis was used to evaluate the nonlinear relationship.

[0061] The calculation formula for the Pearson correlation coefficient is:

[0062]

[0063] Where: x i and y i are the data points of the two variables respectively; and is the mean of the two variables; n is the total number of data points.

[0064] This formula determines the degree of linear correlation between two variables by calculating the covariance between the two variables (the numerator) and dividing it by the product of the standard deviations of the two variables (the denominator).

[0065] The commonly used formula in grey relational analysis is the grey relational calculation formula, which is used to measure the similarity or correlation between the reference series and the comparison series. The following are the basic calculation steps and formulas in grey relational analysis:

[0066] First, the reference and comparison series are dimensionless to eliminate the effects of different units and orders of magnitude. A common method is to normalize the data to the interval [0,1].

[0067] For the reference sequence X0 and the comparison sequence X i (i=1, 2, ..., n), calculate their absolute differences at each data point:

[0068]

[0069] Where k is the serial number of the data point and n is the total number of comparison series.

[0070] Then calculate the correlation between the reference series and each comparison series:

[0071]

[0072] Where m is the total number of data points, and ρ is the resolution coefficient, which usually takes a value between [0, 1] and is used to adjust the influence of the maximum difference and the minimum difference in the correlation calculation.

[0073] According to the calculated correlation ξ i Sort the comparison series. The higher the correlation, the stronger the correlation between the factor and the reference series.

[0074] As a specific implementation method, the feature selection step of this example involves a comprehensive analysis of more than 40 potential influencing factors initially screened to identify key features closely related to municipal solid waste generation. Specifically, this example uses two methods, Pearson correlation analysis and grey relational analysis, to evaluate the linear and nonlinear relationships between each factor and municipal solid waste generation. Using these two analysis methods, this example screened out a total of 24 key features, as shown in Table 1:

[0075] Table 1

[0076]

[0077] In this example, Pearson correlation analysis is used to quantify linear relationships between factors, while grey relational analysis is used to capture nonlinear relationships between factors. To combine the results of these two analyses, this example employs a weighted calculation method that integrates the influence of linear and nonlinear features by optimizing the weighting ratio. The weighting ratio of 0.65 in this example is experimentally proven to be optimal, ensuring the scientific nature of the feature selection process and the high accuracy of model predictions.

[0078] Compared to traditional feature selection methods, the feature selection mechanism of this embodiment improves the scientific nature of feature selection and the model's prediction accuracy. By combining Pearson correlation analysis and grey relational analysis, this embodiment can more comprehensively identify and utilize key factors affecting household waste generation, providing a foundation for building an efficient and accurate prediction model.

[0079] Implementable, in terms of model selection:

[0080] like Figure 2 As shown, the model selection for the method for predicting the amount of municipal solid waste generated and its components in this embodiment is based on a deep learning model. The deep learning architecture involves a one-dimensional convolutional neural network (1DCNN), a bidirectional long short-term memory network (Bi-LSTM), a multi-head attention mechanism model, and a hybrid model (1D-C-BL-MA). By comprehensively comparing the performance of these models, this embodiment ultimately selects the best prediction model (1D-C-BL-MA).

[0081] The deep learning model used in this embodiment specifically includes: a one-dimensional convolutional neural network (1D CNN) that effectively extracts local features from time series data, and then captures long-term dependencies in the data through a bidirectional long short-term memory network (Bi-LSTM). The model integrates a multi-head attention mechanism (Multi-Attention), which can dynamically assign weights to the model, allowing the model to focus on key steps in the time series, thereby improving the model's understanding and prediction capabilities of time series data. This model structure, which combines local feature extraction, long-term dependency capture, and attention weighting, provides an efficient and accurate solution for predicting the amount of municipal solid waste generated.

[0082] Furthermore, the one-dimensional convolutional neural network (CNN) architecture of this embodiment includes an input layer, multiple hidden layers, and an output layer. The hidden layer is composed of multiple repeatedly stacked convolutional layers and pooling layers, and optionally includes one or two fully connected layers. In the convolutional layer, the input feature map is convolved using the corresponding convolution kernel to create the input feature map.

[0083] Connecting the input feature map and the output of the neuron, it can be calculated as:

[0084]

[0085] in, is the output of the first layer of JTH neurons, f(x) is a nonlinear function, is the scalar bias of the JTH neurons in layer 1, M j Indicates the selectivity of the input mapping.

[0086] The neurons in the convolutional layer are the input features of the max pooling layer, and are equipped with the function of reducing the dimensionality of the features extracted by the upper convolutional layer. The output of the max pooling neuron can be expressed as:

[0087]

[0088] Finally, the max pooling unit is flattened and input into the fully connected layer, and then the weighted sum value is used as the input of the sigmoid function. Subsequently, the sigmoid activation function is used as a bridge to connect the input feature map with the output of the neuron. The calculation of the sigmoid function can be expressed as:

[0089]

[0090] Among them, x represents the input of the convolutional layer, and σ(x) represents the output after activation by the sigmoid function.

[0091] In this way, the sigmoid function converts the signal of the input feature map into the activation output of the neuron, providing input for the next layer of the network. This structural design enables the convolutional neural network to effectively capture the local features of the input data and perform feature abstraction and learning at multiple levels.

[0092] Furthermore, this embodiment relates to an improved bidirectional long short-term memory (Bi-LSTM) unit that uses a multiplication gating mechanism to regulate information flow to optimize the transmission efficiency of error signals. Specifically, the bidirectional long short-term memory unit includes a forget gate that uses a sigmoid activation function layer to determine which information should be retained and which should be discarded. The output value of the sigmoid function ranges from [0, 1], indicating the degree of information retention, where 0 means completely discarding the information and 1 means completely retaining the information.

[0093] f t =σ(W f *[h t-1 ,x t ]+b f ),

[0094] In terms of input gate, it can store valuable information in the cell state with the help of sigmoid and tanh functions. The input gate function can be expressed as:

[0095] i t =σ(W i *[h t-1 ,x t ]+b i ),

[0096]

[0097] Long memory (Cell state) is the product of the previous Cell state and the new Cell state, which can be calculated as:

[0098]

[0099] The output gate also uses sigmoid and tanh functions to implement information filtering. The mathematical expression is as follows:

[0100] O t =σ(W0*[h t-1 ,x t ]+b0),

[0101] h t =O t *tan(C t ),

[0102] This design enables the forget gate to finely control the flow of information, effectively reducing the interference of useless information on model performance, thereby improving the efficiency and accuracy of the bidirectional long short-term memory network unit in processing sequence data.

[0103] Furthermore, in order to further improve the model's ability to capture temporal dependencies between features, this embodiment introduces a multi-head attention mechanism after the bidirectional long short-term memory network layer.

[0104] The attention mechanism can be described as mapping a query and a set of key-value pairs to an output, where the query, key, value, and output are all vectors.

[0105]

[0106] Among them, F i is the attention score calculated between a given query and a key. i is the attention weight after Softmax normalization. It represents the relative importance of the query and the i-th key. The larger the value, the stronger the correlation between the query and the key. x It represents the number of keys, that is, the total number of keys in a set of key-value pairs.

[0107] This embodiment constructs a prediction model for municipal solid waste generation and composition based on a deep learning framework. During the model training phase, this embodiment uses the aforementioned comprehensive dataset to train the models separately. The models are then trained through methods such as cross-validation and baseline model evaluation and comparison to optimize model parameters, thereby improving the model's generalization capabilities. The parameter optimization process includes two methods: grid search and cross-validation.

[0108] Grid search systematically searches for optimal parameter settings by iterating over multiple parameter combinations, while cross-validation evaluates the model's performance on different data subsets, ensuring its stability and generalization ability on unseen data. Through these methods, this example ensures that the model demonstrates stable and reliable prediction performance on the test set.

[0109] This embodiment employs a self-updating architecture with a recursive feature to predict future feature quantities. Compared to most current methods that use average growth rates to calculate future feature values, this method uses the interactions of various factors affecting the amount and composition of domestic waste as effective features, learning from historical data to recursively predict future feature quantities, further effectively predicting domestic waste production.

[0110] It is feasible that this embodiment adopts a step-by-step prediction logic, and performs the prediction of urban domestic waste generation and the prediction of component ratio as two independent steps, respectively predicting the domestic waste generation data within the target time range and the component data within the target time range. This strategy bases the component ratio prediction on the domestic waste volume data generated in the previous step.

[0111] Specifically, this embodiment predicts the amount of domestic waste generated based on future time characteristics; using the amount of domestic waste generated as a supplementary feature, together with the previously screened key feature data as new feature data, further predicts the component amounts of domestic waste. At the same time, a total amount consistency constraint is introduced to standardize and calibrate the prediction results to obtain the proportions of each component. This method logically ensures the scientific nature and rationality of the prediction process and improves the accuracy of component ratio prediction. Through this step-by-step prediction strategy, this embodiment can more accurately capture the dynamic relationship between domestic waste generation and component ratios, providing a more precise scientific basis for urban domestic waste management.

[0112] Furthermore, MSW components are categorized into kitchen waste, paper, rubber and plastic, textiles, wood and bamboo, bricks, ceramics, glass, metals, and the sum of other components. The proportions of each component are then predicted separately. This approach ensures that the predicted proportions are consistent with actual trends, providing a more accurate scientific basis for MSW management.

[0113] This embodiment prioritizes predicting the amount of domestic waste generated, and then continues to predict the components based on this prediction result to ensure the logical rationality and effectiveness of the prediction, thereby improving the accuracy of urban domestic waste classification and treatment.

[0114] As a specific implementation, this method uses municipal solid waste generation as one of the key input features for predicting component ratios. This is combined with the 24 previously selected key features as new feature values for prediction. By incorporating waste generation into the model's input feature set, this embodiment ensures that the prediction process fully considers its impact on composition, thereby improving the accuracy and practical application of component predictions.

[0115] In practice, this embodiment takes the coefficient of determination (R2), mean square error (MSE) and mean absolute error (MAE) as factors to consider for model evaluation.

[0116]

[0117] Among them, y i 、 are the actual value, predicted value and average value of the amount of municipal solid waste generated or the proportion of its components.2 This measure reflects the model's ability to explain data variation. Within the range [0, 1], values closer to 1 indicate a better model fit. MSE measures the average squared difference between the model's predicted and actual values. Within the range [0, +∞), smaller values indicate higher model accuracy. MAE measures the average absolute difference between the predicted and actual values. Within the range [0, +∞), smaller values indicate higher model prediction accuracy. Unlike MSE, MAE does not amplify outliers, making it more robust to these factors.

[0118] The prediction method described in this embodiment enhances the accuracy of predictions of municipal solid waste generation and its composition ratios, providing a decision-making aid for urban planning decision makers and solid waste management personnel.

[0119] like Figure 3 As shown, this embodiment also provides a deep learning-based joint prediction system for the amount of municipal domestic waste generated and its components, which is used to implement the method described above, including: a data acquisition module, a feature screening module, a model building module, a waste prediction module and a component prediction module;

[0120] The data acquisition module is used to construct a comprehensive data set of the city, which includes data on the amount of urban domestic waste generated and related influencing factors, as well as data on waste composition;

[0121] The feature screening module is used to perform feature screening on the relevant influencing factor data based on the Pearson correlation coefficient and grey relational analysis to obtain key features that affect the amount of municipal solid waste generated and the waste composition data;

[0122] The model building module is used to build a municipal solid waste generation amount prediction model and a component prediction model based on a deep learning network, and use the comprehensive data set to train the municipal solid waste generation amount prediction model and the component prediction model;

[0123] The garbage prediction module is used to input key features that affect the amount of municipal solid waste generated into the trained municipal solid waste generation prediction model based on a recursive feature self-update mechanism to obtain the predicted amount of municipal solid waste generated;

[0124] The component prediction module is used to combine the predicted garbage generation amount with the key features that affect the garbage component data and input them into the trained component prediction model to obtain the predicted amount of domestic garbage components, while introducing the total amount consistency constraint condition to obtain the proportion of each component of domestic garbage.

[0125] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for jointly predicting the amount of municipal solid waste generated and its components based on deep learning, characterized in that: The following steps are involved: Constructing a comprehensive data set for the city, the comprehensive data set including data on the amount of municipal solid waste generated and related influencing factors, as well as data on waste composition; Based on the Pearson correlation coefficient and grey relational analysis, the relevant influencing factor data are subjected to feature screening to obtain key features that affect the amount of municipal solid waste generated and the waste composition data; Constructing a municipal solid waste generation amount prediction model and a component prediction model based on a deep learning network, and training the municipal solid waste generation amount prediction model and the component prediction model using the comprehensive dataset; Based on the recursive feature self-update mechanism, the key features that affect the amount of municipal solid waste generated are input into the trained municipal solid waste generation prediction model to obtain the predicted amount of municipal solid waste generated. The predicted amount of garbage generated is combined with the key features that affect the garbage component data and input into the trained component prediction model to obtain the predicted amount of domestic garbage components. At the same time, the total amount consistency constraint is introduced to obtain the proportion of each component of domestic garbage.

2. The method according to claim 1, characterized in that After constructing the comprehensive data set of the city, the process also includes: performing data cleaning and normalization processing on the comprehensive data set.

3. The method according to claim 1, characterized in that The process of performing feature screening on the relevant influencing factor data based on the Pearson correlation coefficient and grey relational analysis to obtain key features that affect the amount of municipal solid waste generated and the waste composition data includes: The linear relationship between the relevant influencing factor data is quantified based on the Pearson correlation coefficient, and the nonlinear relationship between the relevant influencing factor data is captured based on grey correlation analysis. The linear and nonlinear relationships are weighted and integrated to obtain the key features that affect the generation of urban domestic waste and the waste composition data.

4. The method according to claim 3, characterized in that The calculation formula of the Pearson correlation coefficient is: Among them, x i and y i are the data points of the two variables respectively; and is the mean of the two variables; n is the total number of data points.

5. The method according to claim 3, characterized in that The calculation formula of the grey relational analysis is: Where m is the total number of data points, ρ is the resolution coefficient, is the reference sequence X0 and the comparison sequence X i The absolute difference between the data points.

6. The method according to claim 1, characterized in that The architecture of the deep learning network includes a one-dimensional convolutional neural network, a bidirectional long short-term memory network and a multi-head attention mechanism model.

7. The method according to claim 6, characterized in that The one-dimensional convolutional neural network includes an input layer, several hidden layers and an output layer; The hidden layer includes several repeatedly stacked convolution layers and pooling layers. In the convolution layer, a convolution kernel is used to perform a convolution operation on the input feature map to establish an input feature map; the input feature map and the output of the neuron are connected to obtain the output of the neuron; the sigmoid activation function is used as a bridge to connect the input feature map and the output of the neuron to provide input for the next layer of the network.

8. The method according to claim 6, characterized in that The bidirectional long short-term memory network includes a forget gate, which determines the retention degree of input information through a sigmoid activation function.

9. The method according to claim 1, characterized in that The method for training the municipal solid waste generation amount prediction model and the component prediction model includes: grid search and cross validation.

10. A deep learning-based joint prediction system for municipal solid waste generation and its components, characterized by: Used to implement the method according to any one of claims 1 to 9, comprising: a data acquisition module, a feature screening module, a model building module, a garbage prediction module and a component prediction module; The data acquisition module is used to construct a comprehensive data set of the city, which includes data on the amount of urban domestic waste generated and related influencing factors, as well as data on waste composition; The feature screening module is used to perform feature screening on the relevant influencing factor data based on the Pearson correlation coefficient and grey relational analysis to obtain key features that affect the amount of municipal solid waste generated and the waste composition data; The model building module is used to build a municipal solid waste generation amount prediction model and a component prediction model based on a deep learning network, and use the comprehensive data set to train the municipal solid waste generation amount prediction model and the component prediction model; The garbage prediction module is used to input key features that affect the amount of municipal solid waste generated into the trained municipal solid waste generation prediction model based on a recursive feature self-update mechanism to obtain the predicted amount of municipal solid waste generated; The component prediction module is used to combine the predicted garbage generation amount with the key features that affect the garbage component data and input them into the trained component prediction model to obtain the predicted amount of domestic garbage components, while introducing the total amount consistency constraint condition to obtain the proportion of each component of domestic garbage.

Images