Municipal waste output prediction method based on signaling remote sensing fusion

Abstract

The invention discloses an urban garbage generation quantity prediction method based on signaling remote sensing fusion, solves the problem that an existing garbage prediction method cannot perform real-time space-time refined prediction on household garbage, and belongs to the field of environmental engineering and artificial intelligence. The method comprises the steps that a daytime training set and a night training set are established, the daytime garbage generation amount is predicted through daytime population distribution data of the first N days, and the night garbage generation amount is predicted through night population distribution data and night light index characteristics of the first N days, the daytime population distribution data and the nighttime population distribution data are determined through mobile phone signaling data of residents in the region, and the nighttime light index is determined according to nighttime light data in the remote sensing image; according to the method, multiple data sources are fused, the prediction accuracy is improved, and particularly, the mobile phone signaling and the night light data are fused and used, the dynamic changes of the population density and the resident activity intensity in the city are fully considered, and the defect that a traditional method depends on static statistical data is overcome.

Description

Technical Field

[0001] The present invention relates to the detection and refined management of the generation amount of multi-source solid waste in cities, and belongs to the fields of environmental engineering and artificial intelligence. Background Art

[0002] Nowadays, 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. New and higher requirements are put forward for the classification and treatment of domestic waste, which requires scientific management planning of domestic waste and prediction of domestic waste volume. However, the accurate prediction of urban waste generation still faces challenges and it is difficult to meet the needs of urban waste classification and reduction. Existing waste prediction methods are mainly based on statistical data. Because statistical data has a time delay and ignores the dynamic population flow and activity intensity in the city, it often cannot provide sufficient support for the real-time spatiotemporal and detailed prediction of domestic waste. Summary of the invention

[0003] In view of the problem that the existing garbage prediction methods cannot make precise spatiotemporal predictions of real-time domestic garbage, the present invention provides a method for predicting the amount of urban garbage generated based on signaling remote sensing fusion.

[0004] The method for predicting the amount of urban garbage generated based on signaling remote sensing fusion of the present invention comprises: establishing a training set, including a training set 1 and a training set 2, wherein the input data in the training set 1 comprises the daytime population distribution data of the city for N days, and the output data is the amount of garbage generated during the day on the N+1th day; the input data in the training set 2 comprises the nighttime population distribution data and nighttime light index characteristics of N days, and the output data is the amount of garbage generated at night on the N+1th day;

[0005] Among them, the daytime population distribution data and nighttime population distribution data are determined by the mobile phone signaling data of residents in the area, and the night light index is determined based on the night light data in the remote sensing image;

[0006] Establish a daytime garbage prediction model and a nighttime garbage prediction model, using a neural network or deep learning model to implement them;

[0007] Use training set 1 and training set 2 to train the daytime garbage prediction model and nighttime garbage prediction model respectively;

[0008] The corresponding input data of the previous N days in the urban test area are input into the trained daytime garbage prediction model and nighttime garbage prediction model. The daytime garbage prediction model and the nighttime garbage prediction model output the daytime garbage generation and the nighttime garbage generation respectively.

[0009] Preferably, the method for establishing a training set includes:

[0010] Divide the urban space into grids, and determine the daytime and nighttime population distribution data in each grid according to the time and the mobile phone signaling data of the residents in the grid;

[0011] Obtain the night light data of each grid from remote sensing data, and process the night light data to obtain the light index characteristics;

[0012] According to the amount of garbage at each transfer station, the daytime and nighttime garbage generation in the transfer station service area is determined, and then the daytime and nighttime garbage generation in each grid in the city is determined.

[0013] Preferably, the nighttime light index determination method is:

[0014] Acquire nighttime light data of each grid from remote sensing data, wherein the nighttime light data is the NTL value of the grid, and correct the NTL value of each grid;

[0015] Calculate the EANTLI value for each grid:

[0016]

[0017] Where EANTLI is the lighting index, NTL is norm is the normalized night light intensity, EVI is the enhanced vegetation index, and NTL is the corrected night light intensity;

[0018] According to the EANTLI value of each grid, the EANTLI image of the city is obtained;

[0019] The EANTLI image is input into the convolutional neural network, and the features of EANTLI are extracted as the night light index features.

[0020] As a preferred method, the NTL value of each grid is corrected as follows:

[0021] Set the population density threshold and the night light intensity threshold, determine the population density based on the population distribution data of each grid, and update the night light intensity of the grid that is less than the population density threshold and greater than the night light intensity threshold to the average of the night light intensity in the surrounding grids.

[0022] As a preferred option, the daily garbage generation in the transfer station service area G day_total and nighttime garbage generation G night_total How to obtain:

[0023] G day_per_vehicle =H single ×G per_household ×T day

[0024] Gday_total =N day ×G day_per_vehicle

[0025] G night_per_vehicle =H single ×G per_household ×T night

[0026] G night_total =N night ×G night_per_vehiclr

[0027] In the formula, G day_per_vehicle , G night_per_vehicle represents the amount of garbage collected by each vehicle during the day and at night, respectively, single Indicates the number of households served by each vehicle in a single trip, G per_household represents the amount of garbage generated by each household, T day , T night Represent the number of trips per vehicle during the day and at night, respectively.

[0028] Preferably, both the daytime garbage prediction model and the nighttime garbage prediction model are implemented using the long short-term memory neural network LSTM. Both the daytime garbage prediction model and the nighttime garbage prediction model use the Sequential class to build a sequential model, add three LSTM layers, and set 128, 64, and 32 neurons in each LSTM layer respectively.

[0029] Preferably, a Dropout layer is added between LSTM layers.

[0030] The beneficial effects of the present invention are as follows: the present invention integrates multiple data sources to improve the accuracy of prediction, especially by integrating the use of mobile phone signaling and night light data, fully considering the dynamic changes in population density and residents' activity intensity in the city, and making up for the shortcomings of traditional methods that rely on static statistical data. The present invention is based on the CNN-LSTM model, and uses CNN to extract the features of the light index EANTLI and then inputs them into the LSTM, making full use of the local feature extraction ability of CNN, especially for the periodic change characteristics of residents' working days and rest days, which helps the LSTM model to better understand and utilize light data and improves the model's ability to reflect actual conditions. At the same time, in order to resist the correlation between variables, the Dropout layer is added as a regularization condition to prevent the occurrence of overfitting. Based on this method, the output of domestic garbage in the city has been refined in the time and space scale, and the high-frequency garbage volume distribution calculation within a single day has been realized, which can provide objective and reasonable suggestions for the development planning of urban environmental protection and sanitation, as well as the classification and treatment of domestic garbage and source reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1Schematic diagram of the prediction method of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0034] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0035] The method for predicting urban garbage generation based on signaling remote sensing fusion in this embodiment aims at the spatial distribution characteristics of urban garbage generation, integrates mobile phone signaling data, nighttime light data, vegetation index and garbage collection and transportation data, and constructs a training set of daytime and nighttime population distribution data, light index and garbage generation in the area to be tested. A daytime garbage prediction model and a nighttime garbage prediction model are established, and implemented using a neural network or a deep learning model; the daytime garbage prediction model and the nighttime garbage prediction model are trained respectively using the daytime and nighttime training sets; the corresponding input data of the previous N days in the area are input into the trained daytime garbage prediction model and the nighttime garbage prediction model, and the daytime garbage prediction model and the nighttime garbage prediction model respectively output the daytime garbage generation and the nighttime garbage generation of the day, thereby realizing an accurate downscaling prediction of garbage production. The method of this embodiment includes:

[0036] Step 1: Obtain different distribution information of residents during the day and at night through mobile phone signaling data. Mobile phone data can record user locations and is widely used in the study of human mobility. This implementation method can use the population distribution obtained by China Unicom through the mobile phone data collected, divide the data according to time (daytime, nighttime) and space (1×1km grid), determine the actual population number and population density in each grid, and obtain population distribution data.

[0037] Step 2: Obtain night light data through remote sensing images. Night light data is the night light intensity NTL value of the grid. Correct the night light data to improve the accuracy of the data and better reflect the actual population distribution and urban development. Perform point density analysis on the population distribution during the day and night to identify densely populated areas in the city during the day and at night. By setting point population density thresholds and night light intensity thresholds, mark some grids in the city with high night light intensity but sparse population. In subsequent predictions, the average night light intensity of the surrounding grids is used to replace the night light intensity value of this grid, thereby completing the preliminary correction of the night light data.

[0038] The NTL value gradually decreases from the city center to the suburbs, while the EVI index shows the opposite trend. The normalized NTL and the interpolated EVI index show a trend of gradually decreasing from the central area to the suburbs. This phenomenon can be used to alleviate the saturation of the city center in the potential saturated area and enhance the difference of NTL in the saturated area. A light index EANTLI that alleviates the saturation problem of night light data is calculated by combining the EVI index (Enhanced Vegetation Index) and the night light intensity. The calculation formula is as follows:

[0039]

[0040] Where, EANTLI is the improved lighting index; NTL norm is the normalized night light intensity; EVI is the enhanced vegetation index; NTL is the night light intensity.

[0041] According to the EANTLI value of each grid, the EANTLI image of the area is obtained.

[0042] Step 3: Input the EANTLI image into CNN to extract the features of EANTLI, so that the subsequent LSTM can better capture the features of night lights. Normalize the EANTLI image, divide and crop it according to the spatial grid (1×1km) in step 1. Select an appropriate CNN architecture for EANTLI feature extraction, set 4 convolutional layers and add a maximum pooling layer after every two convolutional layers to extract high-dimensional feature vectors, and finally convert it into a one-dimensional feature vector through a fully connected layer to facilitate the subsequent input of night light features into LSTM.

[0043] Step 4: By determining the specific parameters of the garbage collection and transportation vehicles, determine the amount of garbage at each transfer station, and then determine the amount of garbage generated in each grid, so as to facilitate the establishment of the subsequent model. By dividing the time of different shifts to distinguish between day and night shifts, the time of garbage generation is determined: garbage collection and transportation from 9 am to 9 pm is daytime generation, and the rest is nighttime generation. In this way, the number of vehicles for day and night shifts is determined, and then the number of households served by a garbage collection and transportation vehicle in a single trip and the number of daily trips of a vehicle are determined, so as to determine the daytime and nighttime garbage production in the service area of ​​each transfer station. The formula for calculating the amount of garbage is as follows:

[0044] G day_per_vehicle =H single ×G per_household ×T day

[0045] G day_total =N day ×G day_per_vehicle

[0046] G night_per_vehicle =H single ×G per_household ×T night

[0047] G night_total =N night ×G night_per_vehicle

[0048] In the formula, G day_per_vehicle , G night_per_vehicle represents the amount of garbage collected by each vehicle during the day and at night, respectively, single Indicates the number of households served by each vehicle in a single trip, G per_household represents the amount of garbage generated by each household, T day , T night Respectively represent the number of trips per vehicle during the day and at night; G day_total , G night_total Represents the amount of garbage generated during the day and at night respectively.

[0049] Step 5: After the above steps, the corresponding population, population density, EANTLI, and the amount of garbage generated during the day and night can be obtained in each service area. All data are divided into training set, test set, and validation set in a ratio of 7:2:1.

[0050] Step 6: Establish a daytime garbage prediction model and a nighttime garbage prediction model, which can be implemented by a neural network or a deep learning model. The neural network can be a long short-term memory neural network LSTM and a BP neural network. This embodiment adopts the LSTM model to implement it, and uses the training set data divided in step 4 to train the daytime and nighttime LSTM models respectively. The daytime garbage generation in the area on the same day is used as the dependent variable input of the LSTM (long short-term memory neural network) model, and the daytime population density and the number of mobile device signals in the area in the previous seven days are used as independent variables to complete the establishment of the daytime LSTM garbage prediction model; the nighttime garbage generation in the area on the same day is used as the dependent variable input of the LSTM (long short-term memory neural network) model, and the nighttime population density, the number of mobile device signals at night, and EANTLI in the area in the previous seven days are used as independent variables to complete the establishment of the nighttime LSTM garbage prediction model.

[0051] For both LSTM models, the Sequential class is used to build a sequential model, adding three LSTM layers, and setting 128, 64, and 32 neurons in each LSTM layer to reduce the computational cost and prevent overfitting. The work of LSTM neurons mainly relies on the forget gate, input gate, output gate, and a memory unit.

[0052] The forget gate is used to control the degree of retention of historical information. The forget coefficient calculation formula of the forget gate is as follows:

[0053] f t =σ(W f x t +U f h t-1 +b f )

[0054] In the formula, f t is the forgetting coefficient. A forgetting coefficient of 0 means that the state of the previous neuron is completely discarded, and a forgetting coefficient of 1 means that the state of the previous neuron is completely retained. σ is the sigmoid activation function, and the output value is between [0,1]. (W f ,U f ,b f ) is a set of weight coefficients of the forget gate.

[0055] After passing through the input gate, a new cell state is generated, where the input coefficient and the new internal cell state are calculated as follows:

[0056] i t =σ(W i x t +U i h t-1 +b i )

[0057] In the formula, i t is the input coefficient, the value range is [0,1], and its size represents the degree of retaining the new unit state; (W i ,U i ,b i ) is a set of weights for the input gate.

[0058] C′ t =tanh(W c x t +U c h t-1 +b c )

[0059] In the formula, C′ t is the new internal unit state, tanh is the activation function, and its value range is [-1,1], (W c ,U c ,b c ) is a set of weights used to compute the internal unit state sharing.

[0060] After calculating the forgetting coefficient and the input coefficient, the memory unit state is updated. The formula for updating the memory unit state is as follows:

[0061] C t =f t *C t-1 +i t *C′ t

[0062] In the formula, C t is the memory cell state at the current moment, C t-1 is the old memory cell state.

[0063] Finally, the result is output through the output gate. The output result is multiplied by the output coefficient and the memory cell state. The calculation process is as follows:

[0064] o t =σ(W o x t +U o h t-1 +b o )

[0065] h t =o t *tanh(C t )

[0066] In the formula, o t is the output coefficient, (W o ,U o ,b o ) is a set of weights for the output gate.

[0067] The model is compiled using the Adam optimizer and the mean square error loss function. The Adam optimizer is listed as follows:

[0068] m t =β1m t-1 +(1-β1)g t

[0069]

[0070]

[0071] In the formula, g t is the gradient at time t; m t and ν t are estimates of the first and second order moments of the gradient, respectively; β1 and β2 are exponential decay rates. The advantage of the Adam optimizer is that it can dynamically adjust the learning rate of each parameter. It does this by maintaining a moving average of past gradients and their square values, ensuring fast and stable learning.

[0072] Since the intensity of nighttime lights also reflects the density of population distribution to a certain extent, that is, there is a certain correlation between the independent variables, in order to ensure the fitting effect of the model, a Dropout layer is added between the layers of LSTM. The Dropout layer randomly discards some neurons, forcing other neurons in the network to be more robust in representing features, avoiding overfitting of the model to the training data, and helping to improve the performance of the model on unknown data.

[0073] Step 7: Use the validation set divided in step 5 to tune the daytime and nighttime LSTM garbage prediction model established in step 6. The mean square error is selected as an indicator to evaluate the prediction performance of the model, because the mean square error measures the average square difference between the predicted value and the actual value. The smaller the mean square error, the closer the model's prediction result is to the actual result, thereby ensuring that the model can accurately predict the amount of garbage generated. Use the validation set divided in step 5 to verify the LSTM model established in step 6. Repeat the training in step 6 and the verification in step 7 under different hyperparameters (learning rate, etc.) and model architectures (number of neurons, etc.), and select the hyperparameters and model architecture with the smallest loss function to complete the model establishment.

[0074] Step 8: Use the test set divided in step 4 to evaluate the effect of the final model in step 7 to ensure that it has good generalization ability and improve the practical application value of the model. Input the test set data divided in step 4 into the model of the final step 7 and check the mean square error of the final model prediction result. It should be close to the mean square error of training and verification, which can ensure its prediction effect and practical application value. Specific embodiment:

[0076] The method of the present invention is applied to the research on the accurate high-frequency prediction of domestic waste in City A from 2016 to 2021. Based on data such as night lights, mobile phone signaling data, and garbage collection and transportation vehicle information, the method of the present invention is verified to timely configure and manage domestic waste-related treatment or recycling resources in advance by integrating and utilizing multi-source data. The specific implementation process is as follows:

[0077] (1) Data collection

[0078] First, relevant data of City A are collected, including daily nighttime light data of City A from 2016 to 2021, EVI index data of City A, population distribution data obtained through mobile devices, and detailed data of garbage collection and transportation vehicles at the transfer station of City A. The nighttime light data and EVI index data can be clipped using the administrative boundary contours of the township level in the study area to obtain relevant data, and the data are uniformly processed by ArcGIS10.7 software.

[0079] (2) Processing night light data

[0080] The population density is analyzed by point density. The population density threshold is set. If the population density is less than the threshold, it is a region with sparse population distribution. At the same time, a threshold is set for the night light intensity. If the region is marked as a region with sparse population distribution but the night light intensity exceeds the threshold, it is marked. When the region is subsequently input into the model, the weight of the night light intensity in the region is reduced.

[0081] (3) Calculation of EANTLI index

[0082] Taking the service area of ​​each garbage transfer station as the object, the EANTLI index is calculated by combining the night light index and EVI index in the area through a formula.

[0083] (4) CNN processing EANTLI index

[0084] The EANTLI image is cut and normalized according to the determined grid. After the convolution kernel is determined, the high-order features in the image are extracted through the convolution layer, and the maximum value pooling method is used to further extract features to facilitate the input of the subsequent LSTM model.

[0085] (5) Determine the amount of garbage

[0086] The garbage production in each area during the day and at night is calculated by collecting parameters such as the working hours and driving distances of the garbage collection and transportation vehicles at various transfer stations in City A.

[0087] (6) Data Grouping

[0088] The final population distribution data, garbage production data, and EANTLI data are respectively mapped to each area and divided into training set, test set, and validation set in a ratio of 7:2:1.

[0089] (7) Build and optimize the model

[0090] Using the population distribution data and EANTLI data in the above training set as independent variables and garbage output as the dependent variable, we continuously change the hyperparameters and model architecture to obtain the hyperparameters and model architecture with the minimum mean square error and finally determine the model.

[0091] (8) Model verification

[0092] For each grid, the population distribution data and EANTLI data were input as independent variables, compared with the actual garbage output, and the mean square error was calculated to see that the final mean square error of the validation set was close to the mean square error of the previous training and test sets. This shows that the model method can ensure its prediction effect and has practical application value.

[0093] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A method for predicting urban garbage generation based on signaling remote sensing fusion, characterized in that: include: Establish training sets, including training set 1 and training set 2. The input data in training set 1 includes the daytime population distribution data of the city for N days, and the output data is the daytime garbage generation on the N+1th day; the input data in training set 2 includes the nighttime population distribution data and nighttime light index characteristics for N days, and the output data is the nighttime garbage generation on the N+1th day; Among them, the daytime population distribution data and nighttime population distribution data are determined by the mobile phone signaling data of residents in the area, and the night light index is determined based on the night light data in the remote sensing image; Establish a daytime garbage prediction model and a nighttime garbage prediction model, using a neural network or deep learning model to implement them; Use training set 1 and training set 2 to train the daytime garbage prediction model and nighttime garbage prediction model respectively; The corresponding input data of the previous N days in the urban test area are input into the trained daytime garbage prediction model and nighttime garbage prediction model. The daytime garbage prediction model and the nighttime garbage prediction model output the daytime garbage generation and the nighttime garbage generation respectively.

2. The method for predicting urban garbage generation based on signaling remote sensing fusion according to claim 1 is characterized in that: Methods for building a training set include: Divide the urban space into grids, and determine the daytime and nighttime population distribution data in each grid according to the time and the mobile phone signaling data of the residents in the grid; Obtain the night light data of each grid from remote sensing data, and process the night light data to obtain the light index characteristics; According to the amount of garbage at each transfer station, the daytime and nighttime garbage generation in the transfer station service area is determined, and then the daytime and nighttime garbage generation in each grid in the city is determined.

3. The method for predicting urban garbage generation based on signaling remote sensing fusion according to claim 2 is characterized in that: The night light index determination method is: Acquire nighttime light data of each grid from remote sensing data, wherein the nighttime light data is the NTL value of the grid, and correct the NTL value of each grid; Calculate the EANTLI value for each grid: Where EANTLI is the lighting index, NTL is norm is the normalized night light intensity, EVI is the enhanced vegetation index, and NTL is the corrected night light intensity; According to the EANTLI value of each grid, the EANTLI image of the city is obtained; The EANTLI image is input into the convolutional neural network, and the features of EANTLI are extracted as the night light index features.

4. The method for predicting urban garbage generation based on signaling remote sensing fusion according to claim 3 is characterized in that: Method for correcting the NTL value of each grid: Set the population density threshold and the night light intensity threshold, determine the population density based on the population distribution data of each grid, and update the night light intensity of the grid that is less than the population density threshold and greater than the night light intensity threshold to the average of the night light intensity in the surrounding grids.

5. The method for predicting urban garbage generation based on signaling remote sensing fusion according to claim 2 is characterized in that: Daily garbage generation in the transfer station service area G day_total and nighttime garbage generation G night_total How to obtain: G day_per_vehicle =H single ×G per_household ×T day G day_total =N day ×G day_per_vehicle G night_per_vehicle =H single ×G per_household ×T night G night_total =N night ×G night_per_vehiclr In the formula, G day_per_vehicle , G night_per_vehicle represents the amount of garbage collected by each vehicle during the day and at night, respectively, single Indicates the number of households served by each vehicle in a single trip, G per_household represents the amount of garbage generated by each household, T day , T night Represent the number of trips per vehicle during the day and at night, respectively.

6. The method for predicting urban garbage generation based on signaling remote sensing fusion according to claim 3 is characterized in that: Both the daytime garbage prediction model and the nighttime garbage prediction model are implemented using the long short-term memory neural network LSTM. Both the daytime garbage prediction model and the nighttime garbage prediction model use the Sequential class to build a sequential model, add three LSTM layers, and set 128, 64, and 32 neurons in each LSTM layer respectively.

7. The method for predicting urban garbage generation based on signaling remote sensing fusion according to claim 6 is characterized in that: Add a Dropout layer between LSTM layers.

8. A computer-readable storage device storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting urban garbage generation based on signaling remote sensing fusion as described in any one of claims 1 to 7 are implemented.

9. A device for predicting the amount of urban garbage generated based on signaling remote sensing fusion, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that: The processor executes the computer program to implement the steps of the method for predicting urban garbage generation based on signaling remote sensing fusion as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for predicting urban garbage generation based on signaling remote sensing fusion as described in any one of claims 1 to 7 are implemented.

Images