CN115099450A

CN115099450A - Family carbon emission monitoring and accounting platform based on fusion model

Info

Publication number: CN115099450A
Application number: CN202210430826.6A
Authority: CN
Inventors: 庞峥琦; 孔英
Original assignee: Shenzhen International Graduate School of Tsinghua University
Current assignee: Shenzhen International Graduate School of Tsinghua University
Priority date: 2022-04-22
Filing date: 2022-04-22
Publication date: 2022-09-23

Abstract

The invention provides a home carbon emission monitoring method based on a fusion model, which comprises the following steps: s1, establishing a database and a prediction model, and extracting features by using a Lasso algorithm aiming at CFPS data; s2, performing parameter adjustment by using the XGboost prediction fusion model, performing classification prediction on high-carbon emission and low-carbon emission, and calculating according to different models respectively; and S3, calculating and generating carbon emission data of the user. The invention mixes the network model structure, can predict the future family carbon emission by inputting the family characteristics, and realizes the accounting and monitoring of the family carbon emission. Compared with other prediction models, the integrated learning prediction method based on the family carbon emission is higher in accuracy rate, achieves family ranking, family emission reduction and family carbon emission accounting, and is beneficial to exciting low-carbon life of users and completing dynamic monitoring accounting of the family carbon emission. The Lasso-XGboost fusion model provided by the invention can better predict the family carbon emission in a classified manner. The individual learner of the fusion model has high accuracy, great diversity and good fusion.

Description

Family carbon emission monitoring and accounting platform based on fusion model

Technical Field

The invention relates to the technical field of carbon emission monitoring, in particular to a method and a device for monitoring household carbon emission.

Background

Data collection and platform construction of carbon emission introduced by an Internet plus means are set as trends, the size of a data collection unit and a data set is very critical, and the accuracy of a final analysis result is influenced by a data processing mode and a prediction model.

Direct and indirect carbon emission accounting and implementation monitoring and early warning of the whole industrial chain of all activities of the family are realized. Carbon emissions have been a major variable in the measure of environmental impact. Energy consumption of goods and services required by households is involved in the investment and production of various industries. Every consumption behavior of the family consumes energy directly or indirectly. In china, with the development and structural changes of economy, the proportion and influence of household consumption in the overall resource consumption are expanding year by year. In 2018, the requirements of industrial departments on environmental monitoring are getting stricter, the corresponding carbon emission is accelerated and slowed down, the corresponding measures for carbon emission of families serving as consumption demand ends are worth paying attention, and the family departments have larger carbon reduction potential.

Literature investigations show that: the feature analysis method suitable for large sample prediction has remarkable characteristics, and needs to be improved according to a research framework when a specific algorithm model is applied. For example, Zhao research finds that due to the overfitting problem caused by low-proportion training data of batteries, the average Absolute Percentage Error (MAPE) of analysis models such as traditional XGBoost (extreme gradient boost) and Artificial Neural Network (ANN) is high, and therefore, the application of large-scale prediction of energy demand behind electric vehicles needs to be improved and optimized.

The scholars usually adopt scenario simulation assumptions based on more limiting conditions for household carbon emission prediction. If the household consumption is directly recalculated in proportion after the assumed moderate income group is expanded in proportion, the predicted reference scene setting is more rigid and the prediction error is easily caused by the assumed error. Scholars mainly use prediction based on a metering model to propose reduction of Household Carbon footprints (HCEs) and adoption of more environmentally friendly lifestyle choices. However, estimating the effectiveness of these actions to actually reduce carbon emissions requires consideration of both lifestyle choices and household characteristics. Therefore, the machine learning model is more suitable for large-scale classification regression prediction than the common metering model, can reduce uncertainty caused by a large amount of assumptions of the metering model, and can increase prediction accuracy, but needs a time sequence with a larger sample size.

Ensemble learning is mainly by adjusting the weight network results, but for a deep learning model with only 5-stage panels, the influence of classification on family individuals should be considered too instantaneous, while the LSTM model focuses on the long-term model and is therefore not applicable. The invention has been published: application number of carbon number data processing, interaction and display method, electronic device and storage medium: CN202210012605.7, it is proposed that the carbon footprint can be well characterized due to the energy handling link. And the channel of data collection contains the carbon emission converted from the user's bicycle data and the like, but the carbon emission reduction incentive for the user is not in the unit of family and does not relate to future carbon emission prediction.

It is noted that the information disclosed in the above background section is only for understanding of the background of the present application and therefore may include information that does not constitute prior art that is known to a person of ordinary skill in the art.

Disclosure of Invention

The invention aims to overcome the defects of the background art and provide a method and a device for monitoring household carbon emission so as to improve the accuracy of predicting the future carbon emission of a household.

In order to achieve the purpose, the invention adopts the following technical scheme:

a family carbon emission monitoring method based on a fusion model comprises the following steps:

s1, establishing a database and a prediction model, and extracting features by using a Lasso algorithm aiming at CFPS data;

s2, performing parameter adjustment by using the XGboost prediction fusion model, performing classification prediction on high-carbon emission and low-carbon emission, and calculating according to different models respectively;

s3, calculating and generating carbon emission data of the user: and calculating the carbon emission of the family in various types of consumption by the combination of the conversion of the coefficient of the input-output matrix and the carbon emission factor of the primary secondary energy through the collected family consumption data, wherein the carbon emission is the carbon emission of the family in the current year in total.

In some embodiments, the model is a multi-zone input-output model in step S1, incorporating a multi-zone input-output analysis framework that embeds the cross-zone carbon emissions for the consumption of different zones of production and services in another zone, and the interaction between the households in different zones.

In some embodiments of the present invention, the,

in the multi-zone input-output model, the specific formula is as follows:

wherein A represents a coefficient matrix of input and output, n industries in the region, s represents the region, X represents input, and y represents demand.

In some embodiments of the present invention, the,

the formula of the domestic consumption carbon emission coefficient is as follows:

wherein, EC _Hj Represents the carbon emissions, OP, of different types of consumption departments _Hj The sales output value of a consumption department is represented, I is a unit vector matrix, and A is a coefficient matrix of an inter-area input-output table;

the domestic carbon emissions, which are thus summed up for the carbon emissions consumed and for the direct carbon emissions, are in kg.

In some embodiments of the present invention, the,

besides collecting and sorting the original data of the CFPS, the data input by the user voluntarily is also obtained.

In some embodiments of the present invention, the,

the method comprises the step of stimulating the low carbon behavior according to data input by a user voluntarily, or generating the ranking at the same time of generating the household carbon emission reduction data.

In some embodiments of the present invention, the,

in step S1, in the objective function of the Lasso model, the loss function with penalty term is as follows:

wherein m is the number of samples, k is a parameter, yi is all dependent variables of the family, λ is a penalty term weight, wo represents a weight, wj represents a penalty term of the parameter, and xij contains all covariates;

and (4) deleting partial features of the punishment item of the Lasso loss function, and performing XGboost on the features after the deletion to predict the carbon emission height.

In some embodiments of the present invention, the,

in step S2, the XGboost prediction function is as follows:

where Yi is the predicted value for i samples, fi (xi) denotes the prediction of the ith sample by i trees, K denotes K trees,

is the maximum estimated improvement relative to constant fit and squared error risk over the entire area;

the XGboost can realize high and low carbon emission prediction aiming at family characteristics.

In some embodiments of the present invention, the,

in step S1, the model is a LASSO-XGBOOST ensemble learning model trained based on CFPS.

In some embodiments of the present invention, the,

the training process comprises: the method comprises the steps of firstly, carrying out time sequence characteristic analysis on family carbon emission historical data, extracting characteristic indexes based on the historical data, deeply excavating the characteristics of the historical data in a family, constructing an input data set of a model based on the characteristic indexes and the historical data, and then training a family carbon emission short-term prediction model by means of multi-task learning theory and multi-element coupling information based on a deep learning algorithm.

A home carbon emission monitoring device based on a fusion model, comprising a processor and a memory, the memory having stored therein a computer program, characterized in that the computer program is processable by the processor to perform the above method.

The invention has the following beneficial effects:

the household carbon emission monitoring method provided by the invention mixes a network model structure, can predict future household carbon emission by inputting household characteristics, and realizes accounting and monitoring of household carbon emission. The Lasso-XGboost fusion model provided by the invention can better predict the family carbon emission in a classified manner. The individual learner of the fusion model has high accuracy, great diversity and good fusion. Lasso regression is adopted for iteration for multiple times, if unremarkable covariates are filtered in advance, shrinkage punishment is carried out by using an L1 norm, lambda enabling MSPE (pure mean square error) to be minimum is selected, XGboost is firstly subjected to feature screening by Lasso and then is subjected to prediction, and the result is more regularized. Compared with other prediction models, the integrated learning prediction method based on the prediction model is better in accuracy rate.

In addition, in some embodiments, the method realizes household ranking, household emission reduction and household carbon emission accounting, and is beneficial to exciting low-carbon life of users and completing dynamic monitoring and accounting of household carbon emission.

Drawings

FIG. 1 is a schematic diagram of the operation of a home carbon emissions database in an embodiment of the present invention;

FIG. 2 is a schematic illustration of carbon emissions reduction behavioral encouragement for high carbon emitting households in an embodiment of the invention;

FIG. 3 is a working schematic diagram of a Lasso-XGboost fusion model in the embodiment of the present invention;

FIG. 4a is a diagram of a blending model confusion matrix for the prediction results of city samples according to an embodiment of the present invention;

FIG. 4b is a graph of a fusion model confusion matrix for the prediction results of rural samples in an embodiment of the present invention;

FIG. 5a is a KNN cluster map for a city sample in an embodiment of the invention;

FIG. 5b is a KNN cluster map for rural samples in an embodiment of the present invention;

FIG. 6a is a decision tree regression ROC plot for a city sample in an embodiment of the present invention;

FIG. 6b is a decision tree regression ROC plot for rural samples in an embodiment of the present invention;

FIG. 7a is a diagram illustrating a leader board interface of the applet app in an embodiment of the present invention;

FIG. 7b is a schematic carbon product interface diagram of the applet app in an embodiment of the present invention;

FIG. 7c is a schematic view of my interface of an applet app in an embodiment of the invention;

FIG. 8 is a flowchart illustrating the operation of a front-end separation architecture for a product in an embodiment of the present invention;

FIG. 9 is a schematic diagram of the My carbon footprint interface of the applet app in an embodiment of the invention;

FIG. 10 is a schematic diagram of a data collection interface of an applet app in an embodiment of the invention;

FIG. 11 is a flow chart of an algorithm for home carbon emissions monitoring accounting in an embodiment of the present invention;

fig. 12 is a flowchart of coefficient calculation for home carbon emission monitoring accounting in the embodiment of the present invention.

Detailed Description

The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary in nature and is not intended to limit the scope of the invention or its application.

Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of the present invention, "a plurality" means two or more unless specifically defined otherwise.

The invention concept of the invention is as follows:

the embodiment of the invention provides a fusion model-based home carbon emission monitoring method and device, which use a plurality of vectors to represent characteristics of different levels of users. Two accounting and incentive based modules are designed and attempt to non-invasively combine user data with existing databases during interaction, completing performance overrides of traditional products. The design fusion module balances by considering recently interacted items, which is able to calculate the number of recalls for different interest vectors and does not require the participation of the target item. Therefore, the Lasso-XGboost fusion network provided by the invention can better classify and predict.

Through market research, the current low-carbon knowledge science popularization channels of China are converged into 3, the Chinese textbook in the middle school classroom advocates 'low-carbon life' in a unified way, and academic paper knowledge points are also visualized into posters for propaganda. Products are on the market that focus more on individual rather than family behavior. At present, the relevant internet products for realizing profitability of carbon emission mainly focus on the carbon financial fields of carbon tax, carbon transaction and the like among enterprises, and pay less attention to the carbon emission of household consumption.

The Chinese Family tracking survey (CFPS) itself counted 5 years for a fixed 13000 Family, and thus has a database. But because only 5 years are Short panels, the deep learning model for only 5-phase panels should consider that the impact of classification on family individuals is too transient, while the LSTM (Long Short-Term Memory) model is focused on the Long-Term model and is therefore not applicable. The Transformer model is not preferred because of the huge data size of the prediction of 6 ten thousand families. If stacking integration is used to overlay and combine multiple LSTM networks, a complex tuning process is required. On the other hand, if Adaboost (adaptive boosting classifier) training is used first and then LSTM prediction is used, wherein data are processed according to the mean value, the uniqueness of partial family carbon emission consumption classes is lost, and the features are averaged.

The invention classifies and summarizes the total carbon emission of the building sites in the region and analyzes the total carbon emission, can simulate the carbon emission trend under different situations, and can improve the government management efficiency by the proposed carbon emission reduction scheme.

The embodiment of the invention provides a home carbon emission accounting, classifying and predicting method based on a Lasso (Least absolute geographic and selection operator) fusion model, belonging to the technical field of data processing and specifically comprising the following steps of: extracting a plurality of family features; and (3) constructing a characteristic equation according to comparison among all families, realizing characteristic screening by using a Lasso method, and constructing a database according to all characteristics. Database-linked sql (Structured Query Language) databases can be updated and extracted for analysis at any time.

The fusion model is used as a model integration mode and is applied to the actual energy economy field, and the fusion model using Lasso for feature screening has higher prediction accuracy in a training set and a testing set than the original algorithm, so that the method has better extrapolation performance.

XGBoost is applicable to class prediction problems where the input is assigned a class or label. They are also applicable to regression prediction problems, where the number of real values is predicted given a set of inputs. The invention mixes the network model structure, can predict the future family carbon emission by inputting family characteristics, and completes the accounting and prediction of the family carbon emission through the database interaction. Compared with other prediction-based models, the ensemble learning encodes the original family carbon emission high and low classification, but the self-contained database only has a 5-year time series, so that the effect of using the self-contained backtracking model is not ideal. The invention collects data through platforms such as small programs and the like to update the database, and meanwhile, the data can be used as a new data set to test the prediction accuracy of the deep learning model, so that the integrated learning prediction accuracy of the invention is better.

The home carbon emission value obtained by averaging the original macroscopic data cannot represent an individual, the micro general survey data of China is not provided with a database for designing energy consumption, an IO-EA-expenditure method (input-output consumption expenditure method) is adopted, and the method is a method for calculating the energy intensity behind consumption data by combining household survey data and input-output model analysis and can be used for effectively accounting the home carbon emission. The analysis structure shows the interaction among industrial chains of various domestic consumption departments in China, and mainly reflects on two aspects of economic conversion and production consumption. Meanwhile, the research is combined with a multi-region input-output analysis framework, the model of the research is embedded with the cross-region carbon emission of the production and service of different regions in another region, the interaction among families of different regions is considered, and the research is effective supplement to the original microscopic data. In a multi-region input-output IO model, i.e. 25 provinces of separate IO, inter-provincial carbon emissions shifts are also considered. A represents the coefficient matrix of input and output, n trades in the area, s represents the area, X represents the input, y represents the demand, and the concrete formula is as follows:

the carbon Emission list of the different industries in China 25 province covered by China Family tracking survey (CFPS) is obtained by combining Chinese carbon accounting databases (China Emission Accounts and data sets, CEADs) and data of 2010-2018 years in the IPCC greenhouse gas Emission list guide. According to the urban and rural resident consumption price index itemized index published by the national statistical bureau, the expenditure and income values of eight years are adjusted to the 2010 unchanged price. Then converting by combining with an input-output table to finally obtain the corresponding domestic consumption carbon emission coefficient A _CE The formula is as follows:

wherein, EC _Hj Representing carbon emission of different consumption departments, calculating direct energy consumption data of the departments, and referring to annual book of Chinese energy statistics, OP of 2010-2018 _Hj The sales yield value of a consumption department is represented, I is a unit vector matrix, A is a coefficient matrix of an inter-area input-output table, and 2010, 2012 and 2017 are referred to respectivelyThe input-output table between 31 provinces and cities in China.

After the household carbon emission accounting is completed, a household carbon emission database is established, as shown in fig. 1, the household carbon emission database comprises a carbon emission data acquisition unit and a carbon emission data analysis unit; the carbon emission data acquisition unit can acquire data voluntarily input by a user by taking a small program as a platform except for collecting and sorting original data of the CFPS, and generates ranking by means of low-carbon behavior motivation and the like or household carbon emission reduction data at the same time; the carbon emission data analysis unit includes single family carbon emission accounting and fusion model analysis, and summarizes a plurality of factors affecting the family carbon emission, including family population, house area, number of people eating at home, total family income, and the like. Through comparison and analysis with a plurality of machine learning model results such as KNN (K neighbor) and DT (decision tree), as shown in the following table 1, the Lasso-XGboost fusion model is found to be optimal in effect and is selected.

TABLE 1

CFPS is divided into two samples of cities and villages which are measured and calculated together for 5 years, and multiple indexes such as accuracy, recall rate, F1 and ROC are compared. The household carbon emission is divided into two types of high carbon emission (represented by 1) and low carbon emission (represented by 0) according to the median, and the index formula of the main model evaluation for the two-classification problem is as follows, wherein the support is the occurrence number of each label, the true case calculated according to the mean value is predicted to be a true case and defined as TP, the false case is predicted to be TN, the false case is predicted to be a true case and FP, and the true case is predicted to be a false case and is referred to as FN. The macro-average is obtained by averaging each category, the micro-average does not distinguish the sample categories, and the weighted average considers the distribution of different tag numbers Support. The precision rate and the recall rate are mutually restricted by the adjustment of the super-parameters, the precision rate is high, the recall rate is low, and the precision rate and the recall rate need to be balanced.

And simultaneously, the family relation library and the user individual data are matched to be used as a test set for testing. The back-end carbon data processing platform of the product can provide carbon accounts and corresponding data processing services for families, such as managing source data acquisition equipment, editing carbon emission main body information, browsing real-time carbon emission records, performing timed excitation and certain adjustment according to family characteristics.

The carbon emission prediction module performs two-class prediction on the future emission situation to obtain the carbon emission level of the family, and performs carbon emission reduction behavior encouragement (such as walking encouragement) on the high-carbon-emission family, as shown in fig. 2, the amount of carbon emission reduction of the family can be predicted according to the number of steps of walking every day of WeChat exercise. And abnormal conditions in the results are predicted through data detection and analysis, and finally, the acquired data, the carbon emission real-time calculation results and the abnormal condition analysis results are displayed in real time through a small program interface.

The embodiment of the invention provides a family carbon emission monitoring system based on a fusion model, which comprises an acquisition module, a data module, a display module and a public module. The acquisition module is used for displaying an interface, inputting information and recording all consumption data of a family; the data module performs data transmission, data storage and data processing; the display module realizes a human-computer interaction function; the public module can compare the carbon footprints of itself and others and rank the carbon emissions. And a public module of the product is connected with urban subway platform data to record low-carbon trip mileage and stimulate emission reduction behaviors.

Lasso (Least absolute shrinkage and selection operator), XGBoost (eXtreme Gradient Boosting) fusion model

The Lasso-XGboost fusion model provided by the embodiment of the invention is a LASSO-XGBOOST integrated learning model trained based on CFPS, and realizes the training of a short-term prediction model of the household carbon emission. The working principle of the model is shown in fig. 3, firstly, all CFPS characteristics are input, a loss function with a penalty term is used for characteristic screening, the screened characteristics are put into an XGboost to predict whether the carbon emission is high or low in the future, wherein the high carbon emission is 1, the low carbon emission is 0, the two-classification problem is converted according to the median value of the household carbon emission, and the household carbon emission in rural areas of cities is remarkably different, so that the two samples are divided into two samples in rural areas, and model prediction is respectively adopted. In the parameter adjusting process, as the parameter n _ estimators is larger, the model learning capability is stronger but the overfitting is easy, the parameter n-estimators with the highest accuracy is selected to be 100 according to the learning curve; the eta (estimated time of arrival) in the training set is taken as a value in 0-1, a scoring function is established, the running time is calculated through the time function, the learning rate is adjusted, whether the running time and the model are converged is judged, and finally the eta is calculated to be 0.1.

Lasso-XGboost fusion model confusion matrix

The XGboost result obtained after feature screening is carried out by using Lasso is obviously better, the fusion model can grasp key factors influencing household carbon emission better, and the prediction accuracy is higher. Predictions for city and rural samples, respectively, where 0 represents low carbon emissions and 1 represents high carbon emissions. The confusion matrix (also called error matrix) allows to visualize the tree, a confusion matrix for the predicted results for city and rural samples, respectively, as fig. 4a (city samples) and fig. 4b (rural samples), where 0 represents low carbon emissions and 1 represents high carbon emissions. According to the depth of the color, the prediction accuracy is shown, and it can be seen that most of high-carbon emissions in the model in the urban sample are accurately predicted, compared with most of low-carbon emissions in the rural sample, the prediction accuracy is improved.

The model optimizes the model by using a five-fold cross validation method, and the data are divided into 7 parts each time: and 3, dividing the training set and the test set in proportion, wherein the divided data are not repeated, the operation is performed for 5 times in turn, and the final error is the error average value of ten times of cross validation, so that the influence of abnormal data on the household carbon emission prediction result can be reduced. The home carbon emissions were divided into high and low according to median and the total samples were randomly divided into 70% as the "training set" and 30% as the "test set" and quintupled cross-validation was applied in the training set resampling of sample sizes 13798, 13473, 12881, 13516 and 13863. The XGBoost tree 89.46% is also better than the decision tree 68.71% in accuracy.

K nearest neighbor model

The method comprises the steps of performing unique hot coding on a large number of category attributes of an original database of China family tracing survey (CFPS) to complete basic data analysis, converting a large number of numerical values into character strings, performing special treatment on the mean filling null value of each column if the distribution of unordered categories is distorted, and normalizing the numerical values of all characteristics by using log1 p. After the KNN model is run, it is found that K ═ 5, and a chebyshev distance is used to obtain a cluster map of the influence factors of the city and rural samples, as shown in fig. 5a (city sample) and fig. 5b (rural sample).

The clustering is mainly based on the principle that the Chebyshev distance between samples is calculated through the characteristics of urban and rural family samples respectively, the similarity degree is further judged, the distance between the longest nodes in the default cluster is compared and merged, and finally 3 clusters are formed by segmenting the 3 clusters, the distance of the overall calculation result does not exceed 30 as shown by an x axis, and the difference between the sample clusters is not obvious. The combination of the analysis of the predicted result can find that: fuel type to reflect the direct energy consumption structure of the home. From a transportation perspective, the number of cars has a significant positive impact on family carbon emissions, with behavioural aspects being higher, the higher the income, the lower the mental profile index of family members, and the higher the per-capita emissions for a particular family. In terms of housing conditions, the higher the city grade, the higher the per-capita emissions.

Although whether or not cars are used and total expenses are classified into one category in the city sample, cars are listed in the countryside, so that the comprehensive consideration is that cars are classified into family activities together with whether or not workers are out of service and the number of people who eat at home when the database is subsequently built. The indicator function indicates that the KNN algorithm is not sensitive to outliers but very sensitive to local structures. Other models therefore need to be tried.

Decision Tree regression ROC curve

The decision tree is used as a method which is mainly used for classification prediction in a supervised learning algorithm and does not need any premise hypothesis, is suitable for family classification and regression according to key influence factors of energy requirements, can generally obtain higher accuracy compared with common measurement multivariate regression, and is more suitable for binary classification problems. The tree model sequentially recursively divides data according to an information gain theory until reaching leaf nodes by meeting conditions to obtain entropy values, and the recursive division and splitting respectively adopt a regression tree and a classification tree for continuous variables and classification variables. The classification tree maximizes Entropy (Entropy) while the regression tree adjusts Mean Squared Error (Mean Squared Error).

Unlike regression analysis, the tree-based approach aims to identify different subgroups of samples. Its hierarchical nature does not allow the estimation of the net effect of a single argument. To illustrate this, income, dining out, mortgage loan and education are related to energy demand, with statistical significance in linear regression. However, they are less important in predicting energy demand in tree models than income and mental health. One advantage of the tree model is that the tree is easy to interpret and produces a set of rules that can be visualized as each path from the root of the tree to one of the leaves can be converted into a clear rule, thereby improving user intelligibility and possibly accuracy.

Decision tree models can be identified among a large number of input attributes based on their correlation with predicted output. Based on the adjusted p-value obtained from classical statistical tests, the importance measure does not make any assumptions of gaussian, linear or independence of the regression analysis, and is able to detect multivariate effects, i.e. properties that are related only by interaction with others.

The economics combines top-down and bottom-up methods, and the micro 6 ten thousand data are matched with the macro 25 input-output tables 37 departments one by one. In 2018, the requirements of industrial departments on environmental protection supervision become stricter, the corresponding carbon emission is accelerated and slowed down, the corresponding measures for carbon emission of families serving as consumption demand ends are worth paying attention, and the family departments have larger carbon reduction potential. Difficulties in research include that families are not a completely closed economy and individual preferences are difficult to quantify.

And (4) standardizing variables to eliminate dimension influence, and performing binary prediction on the medium-grade classified carbon emission into high and low classes. The comprehensive evaluation of sensitivity and specificity is shown in fig. 6a (urban sample) and fig. 6b (rural sample), and the size of the area under the curve surrounded by two coordinate axes is used for selecting the optimal model, wherein the y-axis represents the True Positive Rate (TPR) to represent the sensitivity of the model, and the x-axis represents the False Positive Rate (FPR). The AUC value of the urban sample is 0.963 slightly larger than 0.945 of the AUC of the rural sample, but both models are larger than 0.9, so that the application of the decision tree model in the urban and rural samples on the premise of large samples shows higher accuracy. Although decision trees are well interpretable, they are easily over-fit because the features of the training set recursion are dense. Other model comparisons are therefore also required.

In many prediction problems, missing values exist in the data, which will affect the prediction accuracy of the method. It is believed that using a hybrid approach may have better predictive performance than a single technique. The decision tree is a single tree, while the XGBoost tree belongs to ensemble learning with multiple trees. While decision tree algorithms are effective in providing human-readable classification rules, they are also "greedy," resulting in an over-fit training set with low bias but high variance, and generally worse than ensemble learning. In terms of scalability, the decision tree cannot be parallelized with only one tree, but the XGBoost tree can be parallelized so that each processor runs one tree on the cluster. While decision trees have superior interpretability to XGBoost, XGBoost has the disadvantage that it is difficult to interpret the relationship between energy requirements and predictors, and it is impractical to investigate or visualize the structure of all trees in a forest in large-scale home sample exploration. Therefore, a machine learning fusion model is applied, and a Lasso-XGboost model is provided to realize analysis of carbon emission of Chinese families, so that accuracy of family carbon emission classification, driving factor analysis and prediction is improved.

In summary, although the result of the decision tree is intuitive and vivid and has strong interpretability, the training set is dense in recursive features and is easy to be over-fitted, so that the precision of the test result is low. The K neighbor model has too high dependence on local specific data by using a clustering mode, so that the Lasso-XGboost fusion model is introduced into the research.

Feature selection based on Lasso

Linear regression models are widely used to estimate the effect of covariates on a given dependent variable. However, for models with a large number of covariates, such as decision trees, which are prone to problems such as overfitting and multicollinearity, the survey data of Chinese Family Pursuit Survey (CFPS) yields sparse data in the family carbon emission model, and only a small fraction of the independent variables play an important role in the model. The objective function of the Lasso model optimizes the intercept and coefficient of covariates, and the loss function with penalty term is as follows:

where m is the number of samples, k is a parameter, yi is all dependent variables of the family, λ is the penalty term weight, w _o Representing weight, wj representing penalty item of parameter, xij containing all covariates; and (4) deleting partial features of the punishment item of the Lasso loss function, and performing XGboost on the features after the deletion to predict the carbon emission height.

And (4) finding a corresponding lambda value (penalty value) with the minimum likelihood deviation value according to the statistic (likelihood deviation value) corresponding to each lambda value (penalty value) in the Lasso coefficient screening process, thereby confirming the coefficient corresponding to each variable (if the coefficient is not 0, the corresponding variable is included in the model).

Extreme gradient lifting XGboost

The XGboost is an addition model integrating a plurality of decision trees: the residual data of the previous tree builds the next tree. When the nodes of the tree are selected, the characteristics related to calculation (and screened by the Lasso model) are divided by the division points, wherein the loss function L is unknown, XGboost extreme gradient promotion has the functions of extensible cache access and the like, a plurality of trees similar to random forests are constructed through bagging in an independent mode, a new tree is added on the basis, and the constructed trees are supplemented through promotion, so that the method is efficient, flexible and portable. The prediction function is as follows:

where Yi is the predicted value for i samples, fi (xi) indicates that the i-th sample is predicted by i trees, and K indicates K trees.

Is the maximum estimated improvement with respect to constant fit over the entire area and the risk of squared error. The XGboost can realize high and low carbon emission prediction aiming at family characteristics.

An app (application, mobile phone software) is designed on the basis, and as shown in fig. 7a, 7b and 7c, the app comprises three interfaces of a leader board, carbon credits and my, wherein a calculator button can be linked with a household carbon emission small program, and background of the calculator button and the household carbon emission small program can be used for data sharing and analysis through a cloud server at the same time to calculate the household emission reduction. The user database of the family is generated to the rear end together with the carbon emission calculated by the applet by the automatically calculated displacement reduction of the family by the app, and is analyzed together with the cfps database in the sql to initially form an integrated platform. For carbon emission reduction, as shown in fig. 9, according to the carbon emission of 0.78kg at 1 degree electricity, data such as 30g of carbon emission can be reduced by walking for 1 kilometer, the family is set to walk for 1 kilometer, participates in attendance and card-punching, one-degree electricity saving activities are performed, the energy value of the obtained app is shared to exchange for 0.91kg of carbon emission, a record of carbon emission reduction of 8.74kg of the family is generated, and the carbon emission data of each day is summed to obtain annual emission reduction data. Taking a carbon emission source as an energy consumption device as an example, the source data acquisition device may acquire energy consumption of the energy consumption device, for example, the carbon emission source data acquired by the intelligent water meter is water consumption, and an interface of data acquisition in app is shown in fig. 10, but a carbon emission value behind the source data is obtained by a carbon footprint accounting method of a corresponding economic department. It is further noted that households may be both carbon emission sources and carbon abatement sources, depending on behavioral incentives with psychological cues generated for the ranking of households as a unit. The product uses a front-end and back-end separation architecture, a front-end VUE framework and a back-end Fastapi framework, as shown in FIG. 8.

Example 1:

as shown in fig. 11, the embodiment of the present invention provides a home carbon emission monitoring method based on a fusion model, where monitoring refers to generation of high/low home carbon emission and ranking, and if there is a change in home characteristics, the model can predict the high/low home carbon emission. Fig. 12 is a flowchart of coefficient calculation therein.

The method comprises two parts of home carbon emission accounting and binary classification prediction, wherein the home carbon emission accounting part comprises the following steps:

e1, acquiring 5-year data of 2010-2018 at the CFPS official website;

e2, performing Stata software processing on the 5-year Data, matching family and personal Data through an open instruction and a merge instruction, integrating Panel Data (Panel Data) in a cross-year mode, cleaning the Data through python characteristic engineering, and synthesizing the variables into 22 variables;

e3, acquiring the consumption 8-type data of 2010-2018 in the CFPS official website;

e4, performing matrix calculation on the consumption 8-type data, performing Input-Output table coefficient conversion through MRIO (Multi Regional Input Output), and calculating carbon emission of 6 ten thousand families after the consumption department corresponds to the economic department, wherein the carbon emission corresponds to 22 variables in the step E2 one by one;

the two-classification prediction part comprises the following steps:

e5, classifying samples of urban families and rural families for discussion;

e6, extracting features through Lasso, deleting five features of wage income, fuel cost, heating cost, electric charge and pension, classifying the remaining water sources, cooking fuel, neighborhood harmony degree, whether to go out to service workers, whether to own automobiles and government subsidies by families, the number of persons eating at home, the mental features of family members, the area of the family, the harmony degree among the family members, the scale of the population of the family, the total income of the family and the total expense of the family, wherein 13 core variables are selected;

e7, comparing and analyzing results of multiple machine learning models such as KNN (K neighbor) and DT (decision tree), finding that multiple indexes such as F1, accuracy, recall rate and ROC of the Lasso-XGboost fusion model are superior to those of multiple machine learning models such as KNN (K neighbor) and DT (decision tree), selecting the Lasso-XGboost fusion model to predict whether carbon emission is high or low in the future (binary problem), optimizing the model by using a five-fold cross validation method, and enabling data to be 7: 3, dividing a training set and a test set according to the proportion, selecting the parameter n-estimators with the highest accuracy as 100 according to the learning curve, and calculating eta as 0.1 through a time function;

e8, putting the 13 core variables in the step E6 into XGboost, HPO (Hyper-parameter optimization) parameter adjustment, and realizing the high-low classification prediction of the household carbon emission.

The method is specifically described as follows:

2010-2018 year data are downloaded from a downloadable database of a CFPS public network, no family and adult data exist in 2020, 22 variables are selected according to literature research, the inter-year data are integrated and matched with families and individuals, and the encoding names of the inter-year variables are integrated as shown in Table 2.

TABLE 2

Meanwhile, CFPS includes consumption of households (average expenses in different categories per year) in addition to household characteristics, and consumption of households includes 8 major types of consumption, as shown in table 3. After the sorting and integration, the carbon emission of a single family can be obtained by corresponding to the input-output IO table, and the carbon emission of each family can be calculated by 62000 families.

TABLE 3

Classification	Specifically comprises
		Food product	Daily meal fee and meal fee for going out
Clothes and coat	Shoes and hat for clothes, trousers
		Residence	Property, house maintenance, house renting
Articles for daily use	Furniture, durable goods, household electrical appliances, automobile purchase fee and daily necessities
		Traffic communication	Post and telecommunications communication, traffic communication, local traffic fee
Entertainment for culture and educationMusic instrument	Culture, education and tourism
		Medical treatment	Medical treatment and health care
Others	For economic help, social donation and the like of relatives

Each category corresponds to a coefficient from the IO table after integration,

wherein A represents the coefficient matrix of input and output, n trades in the region, s represents the region, X represents the input, and y represents the demand.

The cross-regional carbon emission of production and service consumption in different regions in another region is embedded in the input-output model matrix coefficient, and the interaction among families in different regions is considered.

For example, the input-output tables of 25 provinces in 2010 correspond to eight major categories of CFPS households.

Table 4 is the coordination matrix for the consumption department and the economic department.

TABLE 4

Coefficient of carbon emission for domestic consumption A _CE The formula is as follows:

wherein, EC _Hj Represents the carbon emissions, OP, of different types of consumption departments _Hj And (3) representing the sales output value of a consumption department, wherein I is a unit vector matrix, and A is a coefficient matrix of an inter-area input-output table.

Thus, the coefficient of each family is obtained by carrying out coordinated averaging on 25 provinces of each year, and the carbon emission of the family in the current year is obtained by multiplying the coefficient by the 8-family large-class consumption. The corresponding carbon emission data are obtained after conversion, coefficients Ace calculated in 2010 are 0.11 of food (food), 0.24 of clothing (stress), 0.51 of residence (house), 0.31 of durable goods (daily), 0.07 of medical (med), 0.19 of traffic (trco), 0.23 of educational entertainment (eec) and 0.07 of others (other), detailed data are shown in a table 5,

TABLE 5

Food product	Clothes and coat	Housing	Durable goods	Medical treatment	Traffic system	Education culture entertainment	Others
								14210	3000	2430	2600	1000	600	5000	2000

The specific calculation process is as follows: 14210 × 0.11+3000 × 0.24+2430 × 0.51+2600 × 0.31+1000 × 0.07+600 × 0.19+5000 × 0.23+2000 × 0.07 ═ 4972.1kg

The calculation result shows that the carbon emission of the family in 2010 is 4972.1 kg.

Here, home carbon emission accounting is completed, followed by prediction based on home characteristics.

Meanwhile, a database of the system trains a LASSO-XGBOOST ensemble learning model based on CFPS to predict whether the carbon emission is high or low in the future (a binary problem), the training model firstly screens through Lasso and Lasso to remove 5 characteristics of wage income, fuel cost, heating cost, electricity cost and pension, and calculates through a penalty term loss function,

its loss function with penalty term is as follows:

wherein m is the number of samples, k is a parameter, yi is all dependent variables of the family, λ is a penalty term weight, wo represents a weight, wj represents a penalty term of the parameter, and xij contains all covariates.

The CFPS selects 22 family and personal characteristics according to the literature, and the family codes, provinces, years and urban and rural classifications are not considered as simple characteristics to be taken into a prediction model and are considered independently. After five variables of wage income, fuel cost, heating cost, electric cost and pension are deleted, 13 core variables of the rest water sources, cooking fuel, neighborhood harmony degree, whether a householder goes out, whether a family has a car, government subsidies, the number of people eating at home, the mental features of family members, the family area, the harmony degree among the family members, the family population scale, the total family income and the total family expenditure are put into XGboost.

TABLE 6

Considering the deviation (bias) of Lasso as the contraction estimator, the results of the "Post Lasso" estimator were reported, i.e. only using Lasso for variable screening, five variables of payroll income, fuel cost, heating cost, electricity cost, pension not in the table were eliminated, and the Lasso coefficient table is shown in table 6.

Meanwhile, the payroll income coincides with the total household income considered, and the total household income not only comprises a plurality of income such as payroll income and interest income. The pension is also part of the government subsidy, which also includes land collection, house removal, etc. So the elimination of these features does not affect the response of the corresponding social factors in the final prediction.

The XGboost prediction function is as follows:

where Yi is the predicted value for i samples, fi (xi) denotes the prediction of the i-th sample by the i trees, K denotes the K trees,

is the maximum estimated improvement relative to the constant fit and squared error risk over the entire area.

The model provided by the embodiments of the present invention may enable migration at a new data set because if the family characteristics in 2030 were known, the family carbon emissions in 2030 could be predicted.

The application comprises the following steps:

existing products such as apps collect data independently or collect data for individuals, and the saas integration and block chain synchronization are not performed on industrial chain data, so that the analysis is single, and the actual effect of the obtained carbon reduction incentive is not enough.

The embodiment of the invention also provides a home carbon emission monitoring and exciting system based on deep learning, which comprises a database accounting and analyzing system, a user interaction and exciting system and a home carbon emission game system, wherein the home carbon emission monitoring and exciting system is linked through a saas (Software as a Service) block chain. The reduced carbon emissions of the user correspond to a substantial reward, such as a central line of digital currency, or a virtual reward, such as an amount of effort, etc.

The embodiment of the invention also provides a system for predicting carbon emission based on various structural features of a family, which comprises traversing historical data by a proper sliding window, extracting a change trend and obtaining a family feature index containing multiple time scales. The method comprises the steps of deeply mining the characteristics of historical data in a family, inputting a data set based on characteristic indexes, and then training a family carbon emission prediction model. Like the carbon emissions of businesses for carbon trading and carbon tax research, carbon emissions in units of families face computational challenges on traditional thinking, and even calculations may need to be performed in an interactive scenario, such as while browsing. The present invention provides a system and algorithm to present to the customer all direct indirect consumption of the family and full life cycle carbon emissions classification and tips behind the product, while rewarding and interactively promoting low carbon behavior.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The background of the present invention may contain background information related to the problem or environment of the present invention and does not necessarily describe the prior art. Accordingly, the inclusion in the background section is not an admission of prior art by the applicant.

The foregoing is a more detailed description of the present invention in connection with specific/preferred embodiments, and it is not to be construed that the specific embodiments of the present invention are limited to those descriptions. It will be apparent to those skilled in the art that numerous alterations and modifications can be made to the described embodiments without departing from the inventive concepts herein, and such alterations and modifications are to be considered as within the scope of the invention. In the description herein, references to the description of the term "one embodiment," "some embodiments," "preferred embodiments," "an example," "a specific example," or "some examples" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Those skilled in the art will be able to combine and combine features of different embodiments or examples and features of different embodiments or examples described in this specification without contradiction. Although embodiments of the present invention and their advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the scope of the claims.

Claims

1. A home carbon emission monitoring method based on a fusion model is characterized by comprising the following steps:

2. The method for monitoring home carbon emissions according to claim 1, wherein in step S1, the model is a multi-region input-output model, which incorporates a multi-region input-output analysis framework, and which embeds the cross-region carbon emissions of different regions for production and service consumption in another region, and the interaction between homes in different regions.

3. The method for monitoring home carbon emissions according to claim 2, wherein the multi-zone input-output model has the following specific formula:

4. A home carbon emission monitoring method according to claim 3, wherein the home carbon consumption coefficient is formulated as follows:

5. A home carbon emission monitoring method according to claim 1, wherein data voluntarily input by the user is obtained in addition to the collection and consolidation of the CFPS original data.

6. The home carbon emission monitoring method of claim 5, further comprising the step of stimulating low carbon behavior based on data voluntarily input by a user, or generating a ranking while generating home carbon emission reduction data.

7. The home carbon emission monitoring method of claim 1, wherein in step S1, the loss function with penalty term in the objective function of the Lasso model is as follows:

and deleting partial features of the punishment item of the Lasso loss function, and performing XGboost on the removed features to predict the carbon emission height.

8. The home carbon emission monitoring method of claim 1, wherein the XGboost prediction function in step S2 is as follows:

9. The home carbon emission monitoring method of claim 1, wherein in step S1, the model is a model of LASSO-XGBOOST ensemble learning trained based on CFPS.

10. The home carbon emission monitoring method of claim 1, wherein the training process comprises: the method comprises the steps of firstly, carrying out time sequence characteristic analysis on family carbon emission historical data, extracting characteristic indexes based on the historical data, deeply excavating the characteristics of the historical data in a family, constructing an input data set of a model based on the characteristic indexes and the historical data, and then training the family carbon emission short-term prediction model based on a deep learning algorithm by taking multi-task learning theories and multi-element coupling information into consideration.

11. A fusion model based home carbon emission monitoring device comprising a processor and a memory, the memory having stored therein a computer program, characterized in that the computer program is processable by the processor to perform the method according to claims 1-10.