A method and system for urban carbon emission prediction and control based on deep learning

Abstract

The present invention discloses a method and system for urban carbon emission prediction and control based on deep learning, which relates to the technical field of carbon emission prediction. It includes obtaining carbon content data released by residents in public areas based on smoking behavior, obtaining action data of residents' smoking behavior in public areas based on the carbon content data, where the action data is composed of the moving positions of limb nodes, and obtaining the target population in public areas based on the carbon content data and the action data. By comparing and analyzing the action data of the target population with that of other target individuals, the present invention can extract differential data from the behavior patterns and use these differential data to adjust the values of the correction data, thereby improving the accuracy and real-time performance of carbon emission prediction. In addition, by automatically adjusting the correction data and comparing it with the real-time action data, the future smoking behavior and carbon emission settings of residents can be predicted, improving the accuracy of carbon emission prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission prediction, and specifically to a method and system for urban carbon emission prediction and control based on deep learning. Background Art

[0002] The scope of urban carbon emissions generally refers to the greenhouse gas emissions generated by human activities within the urban area, especially carbon dioxide, as well as other greenhouse gases (such as methane, nitrogen oxides, etc.). These emission sources can be classified from multiple fields and usually include the following major categories: urban energy consumption, industrial emissions, transportation, agricultural emissions, waste management, land use and urban development, and residents' living and consumption behaviors.

[0003] After retrieval, the Chinese invention patent with the publication number "CN116596095A" discloses a "training method and device for a carbon emission prediction model based on machine learning". This application obtains multiple initial training sets of initial enterprise training samples related to carbon emissions and corresponding label data for each enterprise; for any initial training set, based on each minority-class initial sample in the initial training set and the Euclidean distance between each minority-class initial sample and each majority-class initial sample, the current training set is obtained; based on a preset graph convolutional network, feature extraction is performed on each enterprise training sample in the current training set to obtain the graph feature vector corresponding to the current training set; a preset loss function is used to input the graph feature vectors corresponding to different current training sets into a multi-layer neural network classifier to be trained to obtain a trained multi-task prediction model.

[0004] Smoking behavior, as an existing and specific lifestyle in residents' living and consumption behaviors, will exacerbate the urban carbon emission problem from the perspective of lifestyle. However, the above-mentioned disclosed prediction methods and similar methods mainly focus on carbon emission prediction at the enterprise or industrial level. Although they can accurately capture enterprise carbon emission data, they lack attention to the dynamic impact of individual behaviors on carbon emissions. Therefore, the prediction accuracy is limited during actual operation. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for urban carbon emission prediction and control based on deep learning to solve the problems raised in the above background art.

[0006] In the first aspect, to solve the above problems, a method for urban carbon emission prediction and control based on deep learning is proposed, including:

[0007] Obtaining the carbon content data released by residents in public areas based on their smoking behavior;

[0008] Based on the carbon content data, obtain the action data of the smoking behavior of residents in the public area, where the action data is composed of the moving positions of the limb nodes;

[0009] Based on the carbon content data and the action data, obtain the target population in the public area;

[0010] Obtain the first target action data, where the first target is any one within the target population;

[0011] Obtain the second target action data, where the second target is any one among the public area population whose carbon content data matches the smoking behavior;

[0012] Interact the first target action data and the second target action data, and obtain the difference data, where the difference data is used to analyze the smoking behavior pattern of the target population in the public area;

[0013] Analyze the difference data and obtain additional data;

[0014] The additional data adjusts the action data based on the smoking behavior and obtains the corrected data;

[0015] Based on the comparison between the corrected data and the real-time action data, predict the carbon emission actions of residents based on the smoking behavior;

[0016] Based on the corrected data and the carbon content data, formulate control rules, and the control rules are executed based on the matching degree between the action data of the target object, the corrected data, and the carbon content data.

[0017] As a further optimization of this technical solution, the method for obtaining the action data of the smoking behavior of residents in the public area includes:

[0018] Based on the image acquisition technology, mark the limb nodes of the residents, where the limb nodes include: hand nodes, arm nodes, and head nodes;

[0019] Obtain the time series of the carbon content data;

[0020] Based on the time series, obtain the motion trajectory and motion frequency of the limb nodes;

[0021] Based on the duration of the carbon content data, the duration of the motion trajectory, and the motion frequency, construct a smoking behavior determination formula;

[0022] Based on the smoking behavior determination formula, determine whether the obtained action data matches the smoking behavior.

[0023] As a further optimization of this technical solution, the smoking behavior determination formula:

[0024] ;

[0025] where P smokingUsed to represent the determination probability of smoking behavior, and the value range is usually between 0 and 1. The closer it is to 1, the greater the possibility of smoking, C emission Used to represent the known carbon emissions, f motion Used to represent the movement frequency of limb nodes, f threshold Used to represent the movement frequency threshold of smoking actions, f threshold Set based on the observation and analysis of normal smoking behavior, t carbon Used to represent the duration of carbon content data, t motion Used to represent the duration of the movement trajectory. β is an artificially set adjustment coefficient used to adjust the correlation between carbon emissions and smoking behavior;

[0026] Based on P smoking Determine whether the acquired action data matches the smoking behavior based on the value of P.

[0027] As a further preference of this technical solution, the interaction method of the first target action data and the second target action data:

[0028] Based on image acquisition technology, obtain the half-body images of the first target and the second target respectively;

[0029] Stack the quantities of the first target half-body image and the second target half-body image based on time series, and obtain the image sequences of the first target and the second target within the same time period;

[0030] Analyze the movement trajectories of limb nodes in the first image sequence and the second image sequence;

[0031] Analyze the difference value between the first movement trajectory and the second movement trajectory based on the movement trajectory analogy formula;

[0032] Obtain the feature data of the first target half-body image and the second target half-body image. The feature data includes: environmental data and target action data;

[0033] Integrate the difference value and the feature data to form difference data.

[0034] As a further preference of this technical solution, construct a two-dimensional coordinate system based on the half-body image. The two-dimensional coordinate system is used to record the movement trajectories of target limb nodes;

[0035] At each time point t within the time series, the position of the limb node of the target is represented as S(t)=(X (t) ,Y (t) ). The movement trajectory of the first target is S 1 ={S 1 (t) / t=t 0 ,t 1 ,……t n}, and the movement trajectory of the second target is S2 ={S 2 (t) / t=t 0 ,t 1 ,……t n};

[0036] The acquisition formula for the distance d(t) between the first target limb node position and the second target limb node position is as follows:

[0037] , where [x 1 (t), y 1 (t)] is the limb node position of the first target at time point t, and [x 2 (t), y 2 (t)] is the limb node position of the second target at time point t. The motion trajectory analogy formula is:

[0038] , where D weighted is the difference value between the first target behavior data and the second target behavior data.

[0039] As a further preferred embodiment of this technical solution, the method for obtaining carbon content data includes:

[0040] Obtain the public area based on the smoking behavior of urban residents;

[0041] Collect the gas content in the public area per unit time;

[0042] Obtain carbon content data based on the gas content and the urban tobacco combustion carbon emission formula;

[0043] The urban tobacco combustion carbon emission formula is:

[0044] ;

[0045] The E carbon,total is used to represent the carbon content data of tobacco combustion carbon emissions in the public area. n is the number of gas types. C i and EF i are respectively the content and carbon emission factor of the i-th gas. T represents the unit time, and A represents the public area based on smoking behavior.

[0046] As a further preferred embodiment of this technical solution, the method for comparing the correction data and the real-time action data includes:

[0047] Integrate and align the correction data and the real-time action data based on the time series;

[0048] Obtain the feature vectors that match the correction data and the real-time action data;

[0049] Retrieve the features within the action data of residents' smoking behavior in the common area based on the feature vectors.

[0050] In a second aspect, to further improve the above technical solution, a system using the above method for predicting and controlling urban carbon emissions based on deep learning is also proposed, and it includes: a data acquisition module for collecting real-time action data of the target population and carbon content data in the public area;

[0051] A data processing module for processing and analyzing the data collected by the data acquisition module, including preprocessing the data, feature extraction, and pattern recognition;

[0052] A data analysis module for analyzing the data processed by the processing module, identifying the smoking behavior patterns in the public area, and adjusting the action data based on the smoking behavior;

[0053] A prediction and control module for formulating control rules according to the prediction results of the data analysis module, and adjusting the ventilation system or other environmental regulation facilities in the public area based on the control rules to reduce carbon emissions;

[0054] A user interface module for displaying real-time data and prediction results, providing user interaction functions, enabling managers to monitor the system status and manually adjust the control rules.

[0055] As a further preference of this technical solution, the data acquisition module includes: an image acquisition unit and a gas content acquisition unit. The image acquisition unit captures the behavior images of residents in the public area based on image acquisition technology, and the gas content acquisition unit is used to collect the gas content in the public area per unit time;

[0056] The data processing module includes: an action data acquisition unit, a carbon content data calculation unit, and a target population identification unit. The action data acquisition unit marks the limb nodes of residents based on image acquisition technology and obtains the movement trajectories and movement frequencies of these nodes to identify smoking behavior. The carbon content data calculation unit is used to calculate the carbon content data of tobacco combustion in the public area according to the collected gas content and the urban tobacco combustion carbon emission formula. The target population identification unit identifies the target population in the public area based on the carbon content data and the action data;

[0057] The data analysis module includes: a difference data acquisition unit and a correction data generation unit. The difference data acquisition unit obtains difference data by comparing the action data of different individuals within the target population and analyzes the smoking behavior patterns. The correction data generation unit adjusts the action data based on the smoking behavior based on the difference data and additional data to generate correction data;

[0058] The prediction control module includes: a carbon emission prediction unit and a control rule formulation unit. The carbon emission prediction unit predicts the carbon emission actions of residents based on smoking behavior by comparing the correction data and the real-time action data. The control rule formulation unit formulates control rules based on the correction data and the carbon content data.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] This method and system for predicting and controlling urban carbon emissions based on deep learning collect carbon content data and action data to identify the target population in public areas, and obtain difference data by comparing the action data of the target individual with that of the other targets, and analyze the smoking behavior patterns differentiated by the difference data. However, by deeply analyzing the smoking behavior patterns of the target population and combining the difference data, the carbon emission characteristics of different smoking individuals are identified. By collecting various data and setting based on deep learning, not only can the smoking behavior of the target population be predicted, but also personalized control strategy adjustment can be made based on the difference data;

[0061] In addition, the present invention compares and analyzes the action data of the target population with that of the other target individuals, realizes the extraction of difference data from the behavior patterns, and uses these difference data to adjust the value of the correction data, thereby improving the accuracy and real-time performance of carbon emission prediction;

[0062] Finally, based on the analysis of the behavior patterns, the present invention further analyzes the difference data, automatically adjusts the correction data, and compares it with the real-time action data, so as to predict the future smoking behavior and carbon emission settings of residents, which not only improves the accuracy of carbon emission prediction, but also promotes the dynamic optimization of control rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is the operation flow chart of the method of the present invention;

[0064] Figure 2 is the module composition diagram of the system of the present invention;

[0065] Figure 3 is the auxiliary diagram in the specific implementation process of step S101 of the present invention;

[0066] Figure 4 is the two-dimensional coordinate system diagram constructed based on the half-body image of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] Before understanding the technical solutions proposed by the present invention, it should be clear that when the technical solutions proposed by the present invention are applied to the existing technical field, they can be applied to any surveillance device with image acquisition technology installed in public places. For example, they can be integrated into traffic monitoring cameras, shopping mall security monitoring systems, and intelligent street lamp devices. In addition, it should be noted that in order to further improve the prediction accuracy, when actually put into use, it can also cooperate with the apps of mobile devices (mobile phones or smart watches) commonly used by residents, so that these apps can obtain the urban location of residents without causing information leakage, and be used to determine the high-emission public areas with frequent smoking behaviors. Thus, by analyzing the movement trajectories and behavior habits of residents, potential high-carbon emission areas can be identified, and data support can be provided for urban planning and environmental management to achieve more scientific and effective carbon emission control.

[0069] Reference Figure 1 As can be seen, the present invention proposes a method for predicting and controlling urban carbon emissions based on deep learning, including: step S100 - step S800.

[0070] Step S100: Obtain the carbon content data released by residents in public areas based on smoking behaviors.

[0071] Specifically, the method for obtaining the carbon content in step S100 includes: step S101 - step S103.

[0072] Specifically, step S101: Obtain the public area area of urban residents based on smoking behaviors.

[0073] It should be added that for the determination and calculation of the public area area in step S101, the number of surveillance devices implanted with the disclosed method of the present invention in the public area and the distance between two adjacent surveillance devices are obtained.

[0074] Specifically, reference Figure 3It can be seen that it is assumed that 3 monitoring devices are installed in different areas of the commercial street (such as the entrance passage, snack area, and rest area) respectively. The monitoring range of these monitoring devices is 50 square meters. Among them, A is located at the entrance of the commercial area, covering the entire entrance passage of the commercial street. Monitoring device B is located in the snack area of the commercial street, covering the main passage of this area. Monitoring device C is located in the rest area of the commercial street, covering a quarter of the area. At this time, the monitoring ranges of monitoring devices A, B, and C are 50 square meters, 50 square meters, and 12.5 square meters respectively. By calculating the sum of the monitoring ranges of these monitoring devices, the public area is calculated to be 112.5 square meters.

[0075] Step S102: Collect the gas content in the public area based on the unit time.

[0076] It should be clear that in the specific implementation of step S102, the unit time is 2.5 min, and the gas content is obtained by real-time monitoring through gas sensors installed in the monitoring devices in the prior art. These sensors can specifically be gas-sensitive smoke sensors in the prior art, and mainly obtain the gas content in the public area by detecting the smoke particle concentration in the air.

[0077] Step S103: Obtain carbon content data based on the gas content and the urban tobacco combustion carbon emission formula.

[0078] It should be clear that the urban tobacco combustion carbon emission formula disclosed in step S103 is:

[0079] , where E carbon,total is used to represent the carbon content data of tobacco combustion carbon emissions in the public area, n is the number of gas types, C i and EF i are the content and carbon emission factor of the i-th gas respectively, T represents the unit time, and A represents the public area based on smoking behavior.

[0080] Specifically, when the urban tobacco combustion carbon emission formula is applied to an actual place, it is assumed that in the public area of the commercial street, if the monitoring range of monitoring device A is used as a reference, if the smoke particle concentration detected in this area within the unit time is C i , the specific value is 1000 μg / m³, and the public area A caused by smoking behavior in this area is 100 square meters. It is assumed that the carbon emission factor EF i corresponding to the smoke particle concentration C i is 0.001 g / μg, and this gas is the only detected gas type. Then the calculation result of E carbon,total is 1000 μg / m³ × 0.001 g / μg × 2.5 min = 2.5 g / m 3, therefore, in the public area of this commercial street, the carbon emission from tobacco combustion per unit time is 2.5 g / m 3 .

[0081] Step S200: Based on the carbon content data, obtain the motion data of the smoking behavior of residents in the public area.

[0082] It should be clear that in Step S200, the motion data is composed of the moving positions of the limb nodes. In addition, the method for obtaining the motion data of the smoking behavior of residents in the public area in Step S200 includes: Step S201 - Step S204.

[0083] Specifically, Step S201: Mark the limb nodes of residents based on image acquisition technology.

[0084] It should be clear that in Step S201, the limb nodes include: hand nodes, arm nodes, and head nodes. In addition, the image acquisition technology in Step S201 is a mature technical method in the current image field. Specifically in the present invention, it is mainly set in the monitoring device. When the monitoring device is actually running, it marks the limb nodes of residents by identifying and tracking the positions of the limb nodes.

[0085] Step S202: Obtain the time series of the carbon content data.

[0086] It should be clear that Step S202 can correspond the obtained carbon content data with the motion data of the smoking behavior of residents in terms of time, so as to accurately analyze the relationship between the smoking behavior of residents and the carbon emissions.

[0087] Specifically, the acquisition of the time series in Step S202 is achieved by presetting and installing a clock synchronization function in the monitoring device, ensuring the accuracy of data collection. Among them, the clock synchronization function is specifically achieved by installing a GPS module in the monitoring device in the present invention. The GPS module can be synchronized with the global positioning system, so as to ensure that the time information of the monitoring device is consistent with the global standard time. In this way, even when multiple monitoring devices are working simultaneously, it can ensure that the data collected by all devices has a unified timestamp, facilitating subsequent data analysis and processing.

[0088] Step S203: Based on the time series, obtain the motion trajectory and motion frequency of the limb nodes.

[0089] It should be clear that in Step S203, obtaining the motion trajectory and motion frequency of the limb nodes through the time series is achieved by the image capture function in the existing technology.

[0090] Specifically, in step S201, limb nodes are marked, and in step S202, a time series of carbon content data is obtained. Since both step S201 and step S202 are in the same area (public area), by enabling the monitoring device to have an image capture function and combining the movement trajectory and frequency information of the obtained limb nodes with the time series of carbon content data, the correlation between the activity patterns of residents in the public area and carbon emissions is analyzed, thereby providing more accurate data support for the prediction and control of urban carbon emissions.

[0091] Step S204: Construct a smoking behavior determination formula based on the duration of carbon content data, the duration of the movement trajectory, and the movement frequency, and determine whether the obtained action data matches the smoking behavior based on the smoking behavior determination formula.

[0092] It should be added that the smoking behavior determination formula proposed in step S204 of the present invention is:

[0093] ;

[0094] where P smoking is used to represent the determination probability of smoking behavior, and its value range is usually between 0 and 1. The closer it is to 1, the greater the possibility of smoking. C emission is used to represent the known carbon emissions, f motion is used to represent the movement frequency of limb nodes, f threshold is used to represent the movement frequency threshold of smoking actions, f threshold is set based on the observation and analysis of normal smoking behavior, t carbon is used to represent the duration of carbon content data, t motion is used to represent the duration of the movement trajectory, and β is an artificially set adjustment coefficient used to adjust the correlation degree between carbon emissions and smoking behavior.

[0095] Specifically, when the smoking behavior determination formula is actually used, when the specific value of C emission is 10.32 g, the value of f motion is 0.5 Hz, the value of t carbon is 15 min, the value of t motion is 10 min, f threshold is set to 0.4 Hz, and β is taken as 0.6. According to these parameters, the calculated P smoking = 0.6×10.32 g×0.5 Hz / 0.4 Hz×(15 min / 10 min) = 0.934. According to the calculation result, the value of P smoking is close to 1, indicating that the action data highly matches the smoking behavior. Therefore, it can be determined that the action data indeed reflects the smoking behavior.

[0096] Step S300: Obtain the target population within the public area based on the carbon content data and action data.

[0097] It should be clear that the specific method for step S300 to obtain the target population is to analyze the carbon content data and action data within the public area to identify and classify the behavior patterns of the population. First, collect the carbon emission data within the public area, which comes from fixed monitoring devices. Then, combine the action data, which includes limb movements, moving speeds, direction changes, etc. By identifying possible smoking behaviors, distinguish which populations may have carried out smoking behaviors within a specific time, so as to achieve the positioning of the target population, which not only improves the efficiency of data processing but also enhances the control ability of carbon emission behaviors in the public area.

[0098] Step S400: Obtain the difference data.

[0099] It should be clear that the difference data in step S400 is used to analyze the smoking behavior patterns of the target population within the public area. Specifically, in the present invention, the method for obtaining the difference data is to collect the action data of the first target, where the first target refers to any individual in the target group, collect the action data of the second target, and the second target specifically refers to any individual with a smoking behavior that matches the carbon content data among the public area population. Perform interactive analysis on the action data of the first target and the action data of the second target, and then extract the difference data between the two.

[0100] It should be supplemented that the method for performing interactive analysis on the action data of the first target and the action data of the second target includes: step S401 - step S406.

[0101] Step S401: Respectively obtain the half-body images of the first target and the second target based on image acquisition technology. It should be supplemented that the action data is included in the half-body images.

[0102] Step S402: Stack the quantities of the half-body images of the first target and the second target based on time series, and obtain the image sequences of the first target and the second target within the same time period.

[0103] It should be noted that step S402 is used to ensure the analysis of the behavior patterns of the target individuals under the same time background.

[0104] Step S403: Analyze the movement trajectories of the limb nodes in the first image sequence and the second image sequence.

[0105] Step S404: Analyze the difference value between the first movement trajectory behavior data and the second movement trajectory behavior data based on the movement trajectory analogy formula.

[0106] Step S405: Obtain the feature data of the first target half-body image and the second target half-body image. The feature data includes: environmental data and target action data.

[0107] Step S406: Integrate the difference value and the feature data to form difference data.

[0108] It should be noted that in step S406, the integration of the difference value and the feature data uses the assignment method, that is, the difference data is generated by adjusting the numerical value of the feature data.

[0109] In addition, it should be clearly pointed out that between step S401 and step S402, preprocessing must be carried out to analyze the movement trajectory of the limb nodes.

[0110] Specifically, a two-dimensional coordinate system is constructed based on the half-body image. This coordinate system is used to track and record the movement trajectory of the target limb nodes. It should be noted that the two-dimensional coordinate system is implemented through the graphic drawing function of existing computer programs. The abscissa and ordinate of the two-dimensional coordinate system respectively represent the positions of two different limb nodes in the action data.

[0111] It should be added that at each time point t in the time series, the position of the target limb node is S(t)=(X (t) ,Y (t) ), indicating that the movement trajectory of the first target is S 1 ={S 1 (t) / t=t 0 ,t 1 ,……t n}, and the movement trajectory of the second target is S 2 ={S 2 (t) / t=t 0 ,t 1 ,……t n};

[0112] The acquisition formula for the distance d(t) between the first target limb node position and the second target limb node position is:

[0113] , where [x 1 (t), y 1 (t)] is the limb node position of the first target at time point t, [x 2 (t), y 2 (t)] is the limb node position of the second target at time point t, and the movement trajectory analogy formula is:

[0114] , where D weighted is the difference value between the first target behavior data and the second target behavior data.

[0115] It should be further noted that with reference to Figure 4 , when the motion trajectory analogy formula is actually running, set the time point t 0， At this time, the position S of the first target limb node 1 (t 0 ) = (X 1(t0) , Y 1(t0) ) = (3, 4), and the position S of the second target limb node 2 (t 0 ) = (X 2(t0) , Y 2(t0) ) = (7, 1). At this time, .

[0116] Since there is only one time point t 0 in this case, therefore the value of weighted is approximately 1, so D 0 = 1×d(t

[0117] In addition, when the position of the limb node is negative, the position S of the first target limb node 1 (t 1 ) = (-2, 3). At this time, the value of -2 represents that the position of the first target limb node is shifted backward relative to the position of the limb node based on the smoking behavior. The position of the second target limb node is S 2 (t 2 ) = (X 2(t0) , Y 2(t0) ) = (8, -2). At this time, the value of the distance d(t 1 ) between the position of the first target limb node and the position of the second target limb node is approximately , so the behavior difference point is 11 unit nodes.

[0118] Step S500: Analyze the difference data and obtain additional data.

[0119] It should be noted that in step S500, the additional data includes the difference value and the environmental data between the first target and the second target.

[0120] Step S600: Adjust the action data based on the smoking behavior with the additional data and obtain the corrected data.

[0121] It should be supplemented that in step S600, adjusting the action data based on the smoking behavior with the additional data is to finely adjust the position of the limb node in the action data through the positive or negative of the difference value.

[0122] Specifically, when the difference value is positive, it indicates that the position of the first target limb node is ahead of the second target. In this case, the action data of the first target needs to be adjusted backward. Conversely, when the difference value is negative, it means that the position of the first target limb node lags behind the second target, and the action data of the first target needs to be adjusted forward. It should be added that the acquisition of the correction data is to ensure the accuracy of the action data, so as to more accurately predict and control urban carbon emissions.

[0123] Step S700: Predict the carbon emission actions of residents based on smoking behavior based on the comparison of correction data and real-time action data.

[0124] It should be clear that in step S700, the comparison method for the correction data and the real-time action data includes steps S701 - S703.

[0125] Specifically, step S701: Integrate and align the correction data and the real-time action data based on the time series.

[0126] It should be added that step S701 is used to integrate the data collected at different time points to ensure the consistency and comparability of the data, which is achieved by aligning the correction data and the real-time action data with the same time stamp.

[0127] Step S702: Obtain the feature vectors that match the correction data and the real-time action data.

[0128] It should be added that step S702 is used to extract the common feature information in the two sets of data. Specifically, step S702 is completed through the feature extraction algorithm in the existing technology. The purpose is to find the key features that can represent the similarity of the two sets of data. In this way, the corresponding relationship between the correction data and the real-time action data can be effectively identified, and then accurate input can be provided for the subsequent prediction model.

[0129] Step S703: Retrieve the features in the smoking behavior action data of residents in the public area based on the category of the feature vector.

[0130] It should be noted that step S703 aims to identify the smoking behavior patterns that match the feature vector by analyzing the smoking behavior action data of residents in the public area.

[0131] Step S800: Formulate control rules based on the correction data and the carbon content data.

[0132] It should be noted that the control rules are executed based on the matching degree of the action data of the target object with the correction data and the carbon content data.

[0133] In addition, it should be supplemented that in step S800, the control rules include: determining the action data of the target object, including age, gender, and occupation information, analyzing the correlation between the correction data and the carbon content data to determine the impact degree of different action data on carbon emissions, formulating corresponding control measures, such as restricting smoking behavior in specific areas or imposing economic penalties on high-emission behaviors, implementing a dynamic adjustment mechanism, and continuously optimizing the control rules according to the feedback of real-time monitoring data and prediction models.

[0134] It should be supplemented that the execution of the control rules based on the matching degree of the action data, correction data, and carbon content data of the target object is based on the smoking behavior determination formula in step S204 and the method for obtaining the difference value in step S400 to achieve the determination of the matching degree.

[0135] Reference Figure 2 As can be seen, to improve the above technical solution, the present invention also proposes a deep learning-based urban carbon emission prediction and control system, including: a data collection module for collecting real-time action data of the target population and carbon content data in public areas;

[0136] a data processing module for processing and analyzing the data collected by the data collection module, including preprocessing the data, extracting features, and pattern recognition;

[0137] a data analysis module for analyzing the data processed by the processing module to identify smoking behavior patterns in public areas and adjusting the action data based on smoking behavior;

[0138] a prediction and control module for formulating control rules according to the prediction results of the data analysis module and adjusting the ventilation system or other environmental adjustment facilities in public areas based on the control rules to reduce carbon emissions;

[0139] a user interface module for displaying real-time data and prediction results, providing user interaction functions, enabling management personnel to monitor the system status and manually adjust the control rules.

[0140] It should be supplemented that the data collection module includes: an image collection unit and a gas content collection unit. The image collection unit captures the behavior images of residents in public areas based on image collection technology, and the gas content collection unit is used to collect the gas content in public areas per unit time.

[0141] The data processing module includes: an action data acquisition unit, a carbon content data calculation unit, and a target population identification unit. The action data acquisition unit marks the limb nodes of residents based on image acquisition technology, and obtains the movement trajectories and movement frequencies of these nodes, so as to identify smoking behaviors. The carbon content data calculation unit is used to calculate the carbon content data of tobacco combustion in the public area according to the collected gas content and the urban tobacco combustion carbon emission formula. The target population identification unit identifies the target population in the public area based on the carbon content data and the action data;

[0142] The data analysis module includes: a difference data acquisition unit and a correction data generation unit. The difference data acquisition unit obtains difference data by comparing the action data of different individuals within the target population, and analyzes the smoking behavior pattern. The correction data generation unit adjusts the action data based on smoking behaviors based on the difference data and additional data to generate correction data;

[0143] The prediction and control module includes: a carbon emission prediction unit and a management and control rule formulation unit. The carbon emission prediction unit predicts the carbon emission actions of residents based on smoking behaviors based on the comparison of the correction data and the real-time action data. The management and control rule formulation unit formulates management and control rules based on the correction data and the carbon content data.

[0144] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended embodiments and their equivalents.

Claims

1. A method for predicting and controlling urban carbon emissions based on deep learning, characterized in that: include: Obtain data on the carbon content released by residents in public areas based on their smoking behavior; Based on the carbon content data, the action data of residents’ smoking behavior in public areas is obtained. The action data consists of the moving positions of limb nodes. Acquire target groups in public areas based on carbon content data and motion data; Acquire first target action data, where the first target is any one in the target population; Acquire the action data of the second target, where the second target is any one of the people in the public area whose carbon content data matches the smoking behavior; Interacting the first target action data and the second target action data, and obtaining difference data, the difference data being used to analyze the smoking behavior pattern of the target population in the public area; Parse the difference data and obtain additional data; Additional data adjusts the action data based on smoking behavior and obtains correction data; Predict residents' carbon emission actions based on smoking behavior based on the comparison of correction data and real-time action data; Based on the correction data and carbon content data, control rules are formulated, and the control rules are executed based on the matching degree between the target object's action data and the correction data and carbon content data; The method for acquiring action data of residents' smoking behavior in public areas includes: Mark residents’ limb nodes based on image acquisition technology, including hand nodes, arm nodes and head nodes; Get a time series of carbon content data; Obtain the motion trajectory and motion frequency of limb nodes based on time series; A smoking behavior determination formula is constructed based on the duration of carbon content data, the duration of movement trajectory, and the movement frequency; Determine whether the acquired action data matches the smoking behavior based on the smoking behavior determination formula; Interaction method between the first target action data and the second target action data: Based on the image acquisition technology, half-body images of the first target and the second target are respectively acquired; Based on the time series, the number of half-body images of the first target and the number of half-body images of the second target are accumulated, and the image sequence of the first target and the second target in the same time period is obtained; analyzing the motion trajectories of the limb nodes in the first image sequence and the second image sequence; Analyze the difference between the first motion trajectory and the second motion trajectory based on the motion trajectory analogy formula; Acquire feature data of the first target half-body image and the second target half-body image, the feature data including: environment data and target action data; Integrate the difference value and feature data to form difference data; The comparison method between the correction data and the real-time motion data includes: Integrate alignment correction data and real-time motion data based on time series; Obtain feature vectors that match the correction data and real-time motion data; Retrieve features from the smoking behavior action data of residents in public areas based on the categories of feature vectors.

2. According to claim 1, a method for predicting and controlling urban carbon emissions based on deep learning is characterized in that: The smoking behavior determination formula: ； Where P smoking It is used to indicate the probability of smoking behavior. The value range is usually between 0 and 1. The closer to 1, the greater the possibility of smoking. emission Used to represent known carbon emissions, f motion Used to represent the movement frequency of the limb nodes, f threshold The motion frequency threshold used to indicate the smoking action, f threshold Based on the observation and analysis of normal smoking behavior, t carbon The duration used to represent carbon content data, t motion It is used to represent the duration of the movement trajectory, and β is an artificially set adjustment coefficient used to adjust the correlation between carbon emissions and smoking behavior; Based on P smoking The numerical value of determines whether the acquired action data matches the smoking behavior.

3. According to claim 1, a method for predicting and controlling urban carbon emissions based on deep learning is characterized in that: Methods for obtaining carbon content data include: Obtain the area of ​​public areas based on smoking behavior of urban residents; Collect gas content in public areas based on unit time; Obtain carbon content data based on gas content and urban tobacco combustion carbon emission formula; The carbon emission formula for urban tobacco combustion is: ； The E carbon,total It is used to represent the carbon content data of tobacco combustion carbon emissions in public areas, where n is the number of gas types, C i and EF i are the content of the i-th gas and the carbon emission factor, respectively. T represents the unit time, and A represents the area of ​​the public area based on smoking behavior.

4. A deep learning-based urban carbon emission prediction and control system, using a deep learning-based urban carbon emission prediction and control method according to any one of claims 1 to 3, characterized in that: include: Data collection module, used to collect real-time action data of target groups and carbon content data in public areas; The data processing module is used to process and analyze the data collected by the data acquisition module, including preprocessing, feature extraction and pattern recognition of the data; A data analysis module, used to analyze the data processed by the processing module, identify smoking behavior patterns in public areas, and adjust action data based on smoking behavior; The prediction control module is used to formulate control rules according to the prediction results of the data analysis module, and adjust the ventilation system or other environmental conditioning facilities in the public area based on the control rules to reduce carbon emissions; The user interface module is used to display real-time data and prediction results, provide user interaction functions, and enable managers to monitor system status and manually adjust control rules.

5. The urban carbon emission prediction and control system based on deep learning according to claim 4 is characterized by: The data acquisition module includes: an image acquisition unit and a gas content acquisition unit, the image acquisition unit captures the behavior images of residents in the public area based on image acquisition technology, and the gas content acquisition unit is used to collect the gas content of the public area based on unit time; The data processing module includes: a motion data acquisition unit, a carbon content data calculation unit and a target population identification unit. The motion data acquisition unit marks the limb nodes of residents based on image acquisition technology, and obtains the motion trajectory and motion frequency of these nodes, so as to identify smoking behavior. The carbon content data calculation unit is used to calculate the carbon content data of tobacco burning in the public area according to the collected gas content and the urban tobacco burning carbon emission formula. The target population identification unit identifies the target population in the public area based on the carbon content data and the motion data. The data analysis module includes: a difference data acquisition unit and a correction data generation unit. The difference data acquisition unit acquires difference data by comparing the action data of different individuals in the target population and analyzes the smoking behavior pattern. The correction data generation unit adjusts the action data based on the smoking behavior based on the difference data and the additional data to generate correction data. The prediction and control module includes: a carbon emission prediction unit and a management and control rule formulation unit. The carbon emission prediction unit predicts the carbon emission actions of residents based on smoking behavior based on the comparison of correction data and real-time action data. The management and control rule formulation unit formulates management and control rules based on the correction data and carbon content data.

Images