Building-level living source CO2 emission prediction method based on multi-source data and deep learning

Abstract

The invention relates to the fields of carbon list compilation, deep learning technology and the like, and discloses a building-level living source CO2 emission prediction method based on multi-source data and deep learning, and the method comprises the steps: obtaining a multi-source data set which is high-temporal-spatial-resolution grid population activity level data, building vector data, regional POI data and regional meteorological data of a calculation region; preprocessing the multi-source data set, wherein the preprocessing comprises grid population data space filling, grid distribution, building information association and building and POI data combination; constructing an hourly grid population activity level prediction model based on weather and POI features by using a space-time diagram convolutional network algorithm; the grid population activity level is distributed according to the grid building volume proportion, and the building scale population activity level is obtained; and building a building-scale living source CO2 emission accounting model in combination with the living source CO2 emission basic survey data, and outputting a building-scale living source CO2 emission prediction result.

Description

Technical Field

[0001] The present invention relates to the fields of carbon inventory compilation, deep learning technology, and specifically refers to a building-level living source CO2 emission prediction method based on multi-source data and deep learning. Background Art

[0002] Climate change is a global problem facing mankind. Compiling a CO2 inventory is a fundamental task to control greenhouse gas emissions and address climate change. An important attribute of quantifying CO2 emissions is the spatial and temporal scales considered. Currently, some international and national institutions collect and report emissions at national and global levels, and the most common spatial and temporal distribution is concentrated on the annual time scale and national spatial scale.

[0003] Currently, studies on quantifying CO2 emissions at the urban landscape scale usually stay at the level of the entire city or census area; even though many studies have quantified CO2 emissions at the scale of the entire county or city, and some studies have tried smaller spatial scales by quantifying emissions at the census area or community level, there are not many studies on CO2 emissions at the scale of a single building; with the development of computer technology and big data technology, advanced machine learning methods have made it possible to explore the spatiotemporal changes in regional population activities;

[0004] Machine learning is firstly used more and more in the simulation of atmospheric pollutant concentrations. For example, the Bayesian Maximum Entropy Model (BME) and Random Forest (RF) and other machine learning methods are used to establish the relationship between the pollutant concentration data of the monitoring site and the relevant geographical and economic elements, and simulate the spatial distribution of concentrations in a certain area. This method is both scientific and adaptable to the real-time calculation of high-resolution inventories. Compared with traditional regression methods, machine learning methods can avoid errors caused by simple model relationship assumptions, so that complex and large data sets can be trained more efficiently and accurately and the relationship between variables can be established; deep learning, as a special method and extension of machine learning, uses a hierarchical neural network model for learning, which can be used to process complex and large-scale data.

[0005] Therefore, we urgently need to provide a new building-level life source CO2 emission prediction method based on multi-source data and deep learning to solve the problem of quantifying CO2 emissions at the urban landscape scale. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned technology and provide a building-level living source CO2 emission prediction method based on multi-source data and deep learning.

[0007] In order to solve the above technical problems, the technical solution provided by the present invention is a building-level life source CO2 emission prediction method based on multi-source data and deep learning:

[0008] Step S1: Acquire a multi-source data set, wherein the data set is high spatiotemporal resolution grid population activity level data, building vector data, regional POI data, and regional meteorological data in the calculation area;

[0009] Step S2: preprocessing the multi-source data set, including grid population data space filling, grid allocation, building information association, and building and POI data combination;

[0010] Step S3: constructing the hourly grid population activity level prediction model based on meteorological and POI features using a spatiotemporal graph convolutional network algorithm;

[0011] Step S4: allocating the grid population activity level by the grid building volume ratio to obtain the building scale population activity level;

[0012] Step S5: Build the building-scale domestic CO2 emission accounting model in combination with the basic survey data on domestic CO2 emissions, and output the building-level domestic CO2 emission prediction results.

[0013] As an improvement, the population activity level grid data of the calculation area is obtained by a set of population activity level grid data of the area based on mobile phone signaling throughout the year, with a time resolution of hours and a spatial resolution of 200m×200m.

[0014] As an improvement, the original multi-source data is pre-processed by data cleaning, including grid population data space filling, grid allocation, building information association, and building and POI data combination to obtain the hourly population activity level data of the calculation area grid and the building information data.

[0015] As an improvement, the deep learning network model of the hourly grid population activity level prediction model is a spatiotemporal graph convolutional network model STGCN, which is used to obtain the spatiotemporal correlation between the hourly population activity level grid data and meteorological data; wherein, the neural network that processes time and space is a spatiotemporal convolutional layer.

[0016] As an improvement, the temporal convolution block uses GLU to implement 1D causal convolution, and the spatial convolution block uses Chebyshev approximation and first-order approximate graph convolution.

[0017] As an improvement, the grid population activity level is distributed by grid building volume ratio to obtain the building scale population activity level.

[0018] As an improvement, the basic survey data of CO2 emissions from domestic sources is obtained through field basic investigation, specifically the hourly gas consumption coefficient of the residential / commercial population.

[0019] As an improvement, the building-scale domestic source CO2 emission accounting model is constructed using the emission factor method.

[0020] Compared with the prior art, the advantages of the present invention are: providing a building-level domestic source CO2 emission prediction method based on multi-source data and deep learning, and establishing spatiotemporal correlation between data by building a deep learning network model, which helps to construct a high-resolution greenhouse gas emission inventory, improve the resolution ability of greenhouse gas emission inventory compilation, help identify emission sources and monitor emission reduction effects, and promote emission reduction actions. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the method for predicting CO2 emissions from building-level domestic sources based on multi-source data and deep learning of the present invention.

[0022] Figure 2 It is a deep learning network model framework diagram constructed according to the building-level living source CO2 emission prediction method based on multi-source data and deep learning of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0024] Figure 1 A schematic flow chart of a method for predicting CO2 emissions from building-level living sources based on multi-source data and deep learning according to an embodiment of the present invention is shown. The method for predicting CO2 emissions from building-level living sources based on multi-source data and deep learning comprises:

[0025] Step S1: Acquire a multi-source data set, wherein the data set is high spatiotemporal resolution grid population activity level data, building vector data, regional POI data, and regional meteorological data in the calculation area;

[0026] Step S2: preprocessing the multi-source data set, including grid population data space filling, grid allocation, building information association, and building and POI data combination;

[0027] Step S3: constructing the hourly grid population activity level prediction model based on meteorological and POI features using a spatiotemporal graph convolutional network algorithm;

[0028] Step S4: allocating the grid population activity level by the grid building volume ratio to obtain the building-level population activity level;

[0029] Step S5: Build the building-scale domestic CO2 emission accounting model in combination with the basic survey data on domestic CO2 emissions, and output the building-level domestic CO2 emission prediction results.

[0030] In this embodiment, the multi-source data set is the population activity level data of the computing area with high spatiotemporal resolution, building vector data, regional POI data, and regional meteorological data.

[0031] like Figure 2 As shown, the constructed deep learning network model is a spatiotemporal graph convolutional network model; among them, the neural network that processes time and space is the spatiotemporal convolutional layer.

[0032] The building-level domestic source CO2 emission prediction method based on multi-source data and deep learning provided by the present invention, by building a deep learning network model and establishing the spatiotemporal correlation between data, is helpful to construct a high-resolution greenhouse gas emission inventory, improve the resolution ability of greenhouse gas emission inventory compilation, help identify emission sources and monitor emission reduction effects, and promote emission reduction actions.

[0033] In one embodiment of the present invention, the multi-source data set is high spatiotemporal resolution grid population activity level data, building vector data, regional POI data, and regional meteorological data in the computing area.

[0034] In one embodiment of the present invention, the spatiotemporal graph convolutional network model includes a three-layer network structure; the first and second layers are network structures for processing spatiotemporal correlation, and the third layer is a fully connected layer.

[0035] In this embodiment, the main function of processing the spatiotemporal correlation network structure is to extract spatiotemporal features of the input data in order to better capture the complex relationships and patterns in the data.

[0036] In one embodiment of the present invention, in step S3, in the network structure for processing spatiotemporal correlation, the temporal convolution block uses GLU to implement one-dimensional causal convolution, and the spatial convolution block uses Chebyshev approximation and first-order approximate graph convolution.

[0037] In this embodiment, the spatiotemporal graph convolutional network model task is a regression task, so the number of features designed for the output layer is the number of grids, and the feature representation is the grid population activity level.

[0038] In one embodiment of the present invention, the step S4 specifically includes: counting the proportion of building volumes in the grids of the calculation area, and merging the same buildings in different grids.

[0039] In this embodiment, the specific steps of predicting CO2 emissions from living sources at the building level are described. The population activity level at the building scale can be obtained by predicting the grid population activity level and the building volume ratio in the grid through time series and meteorological data prediction. Then, the hourly gas consumption coefficient of the residential / commercial population obtained through survey statistics can be used to calculate the CO2 emissions at the building level according to the emission factor method, as shown in the following formula:

[0040] Emissions=∑EF i ×Activity i ×k×h

[0041] Where E is the carbon dioxide emission, kg; EF is the emission factor, kgCO2 / m3; Activity is the activity level of the building population; K is the hourly residential / commercial per capita gas consumption coefficient; h is the building population activity level coefficient, and the empirical coefficient is 0.9; i is the fuel type.

[0042] The technical solution of the present invention will be demonstrated with a specific embodiment below.

[0043] The specific implementation steps of the building-level life source CO2 emission prediction method based on multi-source data and deep learning in this embodiment are as follows:

[0044] (1) Obtain multi-source data sets; the multi-source data sets include high-temporal and spatial resolution grid population activity level data, building vector data, regional POI data, and regional meteorological data in the calculation area;

[0045] (2) Preprocessing the dataset: We perform spatial filling on the noisy population activity data grid, associate the building vector information, resample the grid to the calculation area, distinguish the POI types in the grid, and count the number of different types of POI data; further, we merge the grid population activity level data, regional POI data, and regional meteorological data to form a basic dataset;

[0046] (3) 80% of the preprocessed data set is extracted as the training set, and 20% of the preprocessed data set is extracted as the test set; the training set contains a large amount of data sample data, which is used for parameter optimization and tuning of the model; the test set contains sample data that is not repeated in the training set, which is used to test the generalization ability of the model;

[0047] (4) Building a deep learning network model; the deep learning network model is a spatiotemporal graph convolutional network model, which includes a three-layer network structure; the first and second layers are network structures for processing spatiotemporal correlation, and the third layer is a fully connected layer; the neural network for processing spatiotemporal is the spatiotemporal convolutional layer, and the network structure is a temporal convolution block-spatial convolution block-temporal convolution block; by combining graph convolution and gated temporal convolution to capture the spatial and temporal dependencies of data, this architecture allows the model to process large-scale networks with higher efficiency; the features processed by the spatiotemporal convolution layer are passed to the fully connected layer to obtain the grid population activity level results in time series;

[0048] In this specific embodiment, the temporal convolution block uses GLU to implement 1D causal convolution, and the spatial convolution block uses Chebyshev approximation and first-order approximate graph convolution;

[0049] (5) Model training: input the training set into the constructed deep learning network model for training, and save the final optimized deep learning network model; input the test set into the final optimized deep learning network model to predict the grid population activity level, and obtain the predicted results of the grid population activity level, and use the mean absolute error (MAE) and determination coefficient (R2) as performance indicators for evaluating the constructed deep learning network model.

[0050] Among them, MAE is the average of the absolute values ​​of the difference between the predicted value and the actual value, which is used to measure the average error between the predicted value and the true value. The formula is as follows:

[0051]

[0052] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0053] R2 is a statistical indicator that evaluates the difference between the simulated value and the true value in the test set. It is used to measure the goodness of fit of the regression model. The formula is as follows:

[0054]

[0055] Where R 2 represents the coefficient of determination; y i represents the i-th observation value; represents the i-th predicted value; represents the mean of all observations.

[0056] In this specific embodiment, MAE is 84.55, R 2 It is 0.79, which shows that the prediction model has good generalization ability and the deep learning model has achieved good results.

[0057] (6) Building-level CO2 emission accounting for domestic sources. Population activity levels are allocated based on the proportion of building volumes within the grid. Furthermore, the basic survey data on domestic CO2 emissions are used to obtain the building-scale gas consumption. The building-scale gas consumption is multiplied by the fuel combustion CO2 emission coefficient to calculate the building-level domestic CO2 emissions. By matching building information, a high-resolution greenhouse gas emission inventory is constructed.

[0058] (7) Building-level CO2 emission prediction: By acquiring real-time meteorological data, updating the grid population activity level, and then calculating CO2 emissions using the above method, the purpose of building-level CO2 emission prediction can be achieved.

[0059] In the description of the embodiments of the present invention, it should be noted that if the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invented product is usually placed when used. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0060] In addition, the terms "horizontal", "vertical", "overhanging" and the like do not mean that the components must be absolutely horizontal or overhanging, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0061] In the description of the embodiments of the present invention, "plurality" means at least 2.

[0062] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0063] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A building-level life source CO2 emission prediction method based on multi-source data and deep learning, characterized by: include Step S1: Acquire a multi-source data set, wherein the data set is high spatiotemporal resolution grid population activity level data, building vector data, regional POI data, and regional meteorological data in the calculation area; Step S2: preprocessing the multi-source data set, including grid population data space filling, grid allocation, building information association, and building and POI data combination; Step S3: constructing the hourly grid population activity level prediction model based on meteorological and POI features using a spatiotemporal graph convolutional network algorithm; Step S4: allocating the grid population activity level by the grid building volume ratio to obtain the building scale population activity level; Step S5: Build the building-scale domestic CO2 emission accounting model in combination with the basic survey data on domestic CO2 emissions, and output the building-level domestic CO2 emission prediction results.

2. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 1 is characterized in that: The population activity level grid data of the calculation area is obtained, which is a set of population activity level grid data of the area based on mobile phone signaling throughout the year, with a time resolution of hours and a spatial resolution of 200m×200m.

3. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 1 is characterized in that: The original multi-source data is subjected to data cleaning pre-processing, including grid population data space filling, grid allocation, building information association, and building and POI data combination, to obtain the hourly population activity level data of the calculation area grid and the building information data.

4. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 1 is characterized in that: The deep learning network model of the hourly grid population activity level prediction model is a spatiotemporal graph convolutional network model STGCN, which is used to obtain the spatiotemporal correlation between the hourly population activity level grid data and meteorological data; wherein the neural network that processes time and space is a spatiotemporal convolutional layer.

5. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 4 is characterized in that: The temporal convolution block uses GLU to implement 1D causal convolution, and the spatial convolution block uses Chebyshev approximation and first-order approximate graph convolution.

6. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 1 is characterized in that: The grid population activity level is distributed by the grid building volume ratio to obtain the building scale population activity level.

7. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 1 is characterized in that: The basic survey data on CO2 emissions from domestic sources is obtained through field basic investigations, specifically the hourly gas consumption coefficient of the residential / commercial population.

8. The method for predicting building-level life source CO2 emissions based on multi-source data and deep learning according to claim 1 is characterized by: The building-scale living source CO2 emission accounting model is constructed using the emission factor method.

Images