A Public Building Carbon Emission Prediction Method Based on Multi-Source Data Processing

Abstract

The present invention discloses a public building carbon emission prediction method based on multi-source data processing, which relates to the technical field of carbon emission prediction. The present invention aims to effectively quantify behavior factors in a data-driven manner and use them as an additional type of multi-source data for training a prediction model. This enables the prediction model to further learn dynamic behavior factors on the basis of existing static building characteristics and environmental climate data, etc., so as to more comprehensively consider the impact of personnel activities and equipment usage in the building on energy consumption, and thus significantly improve the accuracy of the prediction model for public building carbon emission prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission prediction, and particularly to a public building carbon emission prediction method based on multi-source data processing. Background Art

[0002] The public building carbon emission prediction method based on multi-source data processing is a method that combines multiple data sources and technical means to accurately estimate the emissions of greenhouse gases such as carbon dioxide generated during the use of public buildings. In order to more accurately predict the carbon emissions of public buildings, existing methods usually follow a series of specific steps, and the following is a detailed description of these steps:

[0003] First, data collection is performed, including energy consumption data: by installing intelligent meters or sensors to monitor the usage of different types of energy in the building, including electricity, gas, heat, etc., and ensuring that the data can be transmitted to the data center at regular intervals; environmental and climate data: integrating information such as outdoor temperature, humidity, wind speed, and sunshine duration provided by local weather stations or third-party APIs to provide the necessary background conditions for understanding building energy consumption; building characteristic data: recording and organizing information about the static attributes of the building, such as area, height, orientation, building materials and their insulation performance, construction year, and design standards, etc.; operation data: using building automation systems (BAS), Internet of Things (IoT) devices, or other monitoring means to capture dynamic information such as the personnel activity patterns (working hours, visitor traffic) and equipment operation status (start and stop times of air conditioning systems, lighting systems) in the building;

[0004] Subsequently, preprocessing is performed on the collected data, including cleaning and verification: removing or correcting outliers, error records, and duplicates; confirming that the timestamps of all data points are correct; missing value processing: using interpolation methods to fill in blank data points, such as mean filling, nearest neighbor filling, or model-based estimation; standardization and normalization: converting various types of data into a unified format and unit for subsequent analysis; for numerical variables, standardization or normalization operations can be performed to improve algorithm efficiency;

[0005] Then, model selection is carried out. Usually, according to the nature of the problem and the characteristics of the data, a suitable modeling technique is selected, such as linear regression, decision tree, random forest, support vector machine (SVM), artificial neural network (ANN), long short-term memory network (LSTM), etc. After comparing the performance of different models and considering factors such as accuracy, computational complexity, training time, and interpretability, the most suitable model for the current task is selected, and usually an artificial neural network model is adopted;

[0006] Immediately proceed with model training, including dividing the dataset: divide the collected data into a training set, a validation set, and a test set to ensure that the model can perform well on known data and also have strong generalization ability on unknown data; hyperparameter tuning: use methods such as grid search, random search, or Bayesian optimization to find the optimal combination of hyperparameters to achieve the best performance; cross-validation: adopt methods such as K-fold cross-validation to reduce the risk of overfitting while ensuring the stability and reliability of the model;

[0007] Finally, once the model training is completed and passes the validation, it is deployed in the actual environment for real-time prediction of carbon emission trends in the next period of time.

[0008] Although the above methods provide a relatively comprehensive framework for predicting the carbon emissions of public buildings, there are still significant challenges in practice, especially in the following aspects:

[0009] Human behavior habits such as window ventilation and temperature setting have an important impact on building energy consumption. However, this part of unstructured information is ignored and not effectively incorporated into the existing prediction models, which to a certain extent reduces the accuracy of the existing prediction models for predicting the carbon emissions of public buildings.

[0010] Therefore, there is an urgent need in the existing technology for a technical solution for a public building carbon emission prediction method based on multi-source data processing. Summary of the Invention

[0011] To solve the above technical problems, the present invention provides a public building carbon emission prediction method based on multi-source data processing, specifically including the following steps:

[0012] Step S1, collect the energy consumption data, environmental climate data, building characteristic data, and operation data of public buildings, and define them as the first data. At the same time, collect the behavior factors of public buildings;

[0013] Step S2, perform quantization processing on the behavior factors of public buildings to obtain the second data of public buildings;

[0014] Step S2a, calculate the first score of each behavior factor of public buildings;

[0015] Step S2a1, for each behavior factor, respectively define at least two basic features, and construct a feature vector for each behavior factor based on at least two basic features;

[0016] Among them, the expression of the feature vector of each behavior factor is:

[0017] ;

[0018] In the formula, represents the feature vector of the i-th behavior factor; The k-th basic feature representing the i-th behavioral factor; k represents the number of basic features in the i-th behavioral factor;

[0019] Step S2a2: Based on historical data, define at least two influence factors for each basic feature in the feature vector of each behavioral factor;

[0020] Among them, the expression of the influence factor of each basic feature in the feature vector of each behavioral factor is:

[0021] ;

[0022] In the formula, represents the set of influence factors of the k-th basic feature in the feature vector of the i-th behavioral factor; represents the m-th influence factor of the k-th basic feature in the feature vector of the i-th behavioral factor; m represents the number of influence factors of the k-th basic feature in the feature vector of the i-th behavioral factor, and m≥2k, that is, each basic feature has at least two influence factors;

[0023] Step S2a3: Based on each basic feature in the feature vector of each behavioral factor and the influence factors of each basic feature in the feature vector of each behavioral factor, calculate the influence value of each basic feature in the feature vector of each behavioral factor;

[0024] Among them, the calculation formula of the influence value of each basic feature in the feature vector of each behavioral factor is:

[0025] ;

[0026] In the formula, represents the influence value of the k-th basic feature in the feature vector of the i-th behavioral factor; represents the function based on the j-th influence factor of the basic feature ; m represents the number of influence factors of the k-th basic feature in the feature vector of the i-th behavioral factor, and m≥2k, that is, each basic feature has at least two influence factors;

[0027] Step S2a4: Based on the influence value of each basic feature, calculate the first score of each behavioral factor by using the weighted summation method;

[0028] Among them, the calculation formula of the first score of each behavioral factor is:

[0029] ;

[0030] In the formula, represents the first score of the i-th behavioral factor; represents the weight of the k-th basic feature; represents the influence value of the k-th basic feature in the feature vector of the i-th behavior factor;

[0031] Step S2b: Divide each behavior factor into at least two categories according to different dimensions;

[0032] Step S2c: Calculate the second score of each category of behavior factors based on the first score of each behavior factor in each category of behavior factors;

[0033] Among them, the calculation formula for the second score of each category of behavior factors is:

[0034] ;

[0035] In the formula, represents the second score of the q-th category of behavior factors; represents the weight of the i-th behavior factor; represents the first score of the i-th behavior factor; represents that the i-th behavior factor belongs to the q-th category of behavior factors;

[0036] Step S2d: Calculate the comprehensive score of all behavior factors by integrating the second scores of each category of behavior factors;

[0037] Among them, the calculation formula for the comprehensive score of all behavior factors is:

[0038] ;

[0039] In the formula, represents the comprehensive score of all behavior factors; represents the weight of the q-th category of behavior factors; represents the second score of the q-th category of behavior factors; represents the number of behavior factor categories;

[0040] Step S2e: Statistically calculate the first score of each behavior factor, the second score of each category of behavior factors, and the comprehensive score of all behavior factors, and define them as the second data of the public building;

[0041] Step S3: Perform preprocessing on the first data and the second data, and the preprocessing includes cleaning, verification, missing value processing, standardization, and normalization;

[0042] Step S4: Build a carbon emission prediction model, and use the preprocessed first data and second data to train the carbon emission prediction model to obtain a trained carbon emission prediction model;

[0043] Step S5: Obtain the real-time data to be input, and input the real-time data into the carbon emission prediction model to obtain the carbon emission prediction result.

[0044] The embodiments of the present invention have the following technical effects:

[0045] The present invention aims to effectively quantify behavioral factors in a data-driven manner and use them as an additional type of multi-source data for training the prediction model. This enables the prediction model to further learn dynamic behavioral factors on the basis of existing static building characteristics and environmental climate data, etc., so as to more comprehensively consider the impact of personnel activities and equipment usage in the building on energy consumption, and thus significantly improve the accuracy of the prediction model for predicting carbon emissions of public buildings. Description of the Drawings

[0046] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0047] Figure 1 It is a flowchart of a method for predicting carbon emissions of public buildings based on multi-source data processing provided by an embodiment of the present invention. Specific Embodiments

[0048] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0049] Embodiment 1: As Figure 1 shown, the present invention provides a method for predicting carbon emissions of public buildings based on multi-source data processing, including the following steps:

[0050] Step S1: Collect the energy consumption data, environmental climate data, building characteristic data, and operation data of the public building, and define them as the first data. At the same time, collect the behavioral factors of the public building;

[0051] Details on the means of collecting energy consumption data, environmental climate data, building feature data, and operation data of public buildings will not be elaborated here. Instead, the focus is on how to collect the behavior factors of public buildings. Firstly, behavior sensors such as infrared sensors, cameras, and access control systems can be deployed inside the building to record behaviors such as the movement trajectories of people, window ventilation, and temperature adjustment settings. Secondly, questionnaires can be regularly distributed to the building users to understand their daily behavior patterns, such as working hours, rest times, and habits of using equipment. Or a mobile application can be developed, through which users can report their behaviors, such as turning lights on and off, adjusting air conditioner temperatures, etc.

[0052] Step S2: Perform quantization processing on the behavior factors of the public building to obtain the second data of the public building;

[0053] Step S2a: Calculate the first score of each behavior factor of the public building;

[0054] Step S2a1: For each behavior factor, define at least two basic features and construct a feature vector for each behavior factor based on at least two basic features;

[0055] It should be noted that the means of defining the basic features of behavior factors include but are not limited to: First, by analyzing the existing data using the clustering analysis method to identify the features highly correlated with energy consumption as the basic features of behavior factors; Second, using the traditional questionnaire survey method to understand the behavior patterns of building users, and extracting common behavior features therefrom as the basic features of behavior factors.

[0056] For common behavior factors in public buildings, the following basic features can be defined but are not limited to:

[0057] First, timestamp: Record the specific time when the behavior occurs, including date, hour, minute, etc. The main function is that behaviors in different time periods have different impacts on energy consumption. For example, weekdays and rest days, day and night, etc.; Second, frequency: The frequency of behavior occurrence, such as several times a day, several times a week, etc. The main function is that high-frequency behaviors may have a greater impact on energy consumption than low-frequency behaviors. For example, frequently turning lights on and off or adjusting air conditioner temperatures; Third, duration: The duration of each behavior, such as the time of each window ventilation, the time of each use of equipment, etc. The main function is that long-term behaviors usually consume more energy. For example, keeping the air conditioner or lighting on for a long time; Fourth, intensity: The intensity or strength of the behavior, such as the change range of the air conditioner temperature setting value, the light brightness, etc. The main function is that high-intensity behaviors may lead to higher energy consumption. For example, quickly adjusting the air conditioner temperature or using high-power equipment.

[0058] Among them, the expression of the feature vector of each behavior factor is:

[0059] ;

[0060] In the formula, represents the feature vector of the i-th behavior factor; represents the k-th basic feature of the i-th behavior factor; k represents the number of basic features in the i-th behavior factor;

[0061] Step S2a2: Based on historical data, define at least two impact factors for each basic feature in the feature vector of each behavior factor;

[0062] It should be noted that the means of defining impact factors based on historical data mainly analyzes historical data through a regression model, finds the quantitative relationship between each basic feature and energy consumption, and thus determines the impact factors;

[0063] For the above basic features, the following impact factors can be defined:

[0064] Impact factor of timestamp: Impact factor 1: Time periods in a day, morning, afternoon, evening, because energy consumption patterns are different in different time periods; Impact factor 2: Difference between weekdays and rest days, because people's behavior patterns vary greatly on weekdays and rest days;

[0065] Impact factor of frequency: Impact factor 1: Frequency of behavior occurrence, such as how many times a day, and high-frequency behaviors may lead to higher energy consumption; Impact factor 2: Cumulative number of behaviors. In the long run, the more the cumulative number of behaviors, the higher the total energy consumption;

[0066] Impact factor of duration: Impact factor 1: Duration of a single behavior, and long-duration behaviors usually consume more energy; Impact factor 2: Cumulative duration of behaviors, that is, the cumulative duration of behaviors within a period of time;

[0067] Impact factor of intensity: Impact factor 1: Intensity of behavior, such as the change range of temperature setting value, and high-intensity behaviors may lead to higher energy consumption; Impact factor 2: Severity of behavior, such as the difference between quickly adjusting the temperature and slowly adjusting the temperature, and severe behaviors may lead to higher instantaneous energy consumption.

[0068] Among them, the expression of the impact factor of each basic feature in the feature vector of each behavior factor is:

[0069] ;

[0070] In the formula, represents the set of impact factors of the k-th basic feature in the feature vector of the i-th behavior factor; The m-th influencing factor of the k-th basic feature in the feature vector representing the i-th behavior factor; m represents the number of influencing factors of the k-th basic feature in the feature vector of the i-th behavior factor, and m ≥ 2k, that is, each basic feature has at least two influencing factors;

[0071] Step S2a3: Based on each basic feature in the feature vector of each behavior factor and the influencing factors of each basic feature in the feature vector of each behavior factor, calculate the influence value of each basic feature in the feature vector of each behavior factor;

[0072] Among them, the calculation formula for the influence value of each basic feature in the feature vector of each behavior factor is:

[0073] ;

[0074] In the formula, represents the influence value of the k-th basic feature in the feature vector of the i-th behavior factor; represents the function of the j-th influencing factor based on the basic feature ; m represents the number of influencing factors of the k-th basic feature in the feature vector of the i-th behavior factor, and m ≥ 2k, that is, each basic feature has at least two influencing factors;

[0075] Step S2a4: Based on the influence value of each basic feature, calculate the first score of each behavior factor by using the weighted summation method;

[0076] Among them, the calculation formula for the first score of each behavior factor is:

[0077] ;

[0078] In the formula, represents the first score of the i-th behavior factor; represents the weight of the k-th basic feature; represents the influence value of the k-th basic feature in the feature vector of the i-th behavior factor;

[0079] Step S2b: Divide each behavior factor into at least two categories according to different dimensions;

[0080] It should be noted that among them, different dimensions can be divided according to the temporality of behavior factors such as weekdays / weekends, spatiality such as different rooms or areas, types such as office, meeting, leisure, and so on.

[0081] Step S2c: Based on the first score of each behavior factor in each category of behavior factors, calculate the second score of each category of behavior factors;

[0082] Among them, the calculation formula for the second score of each type of behavior factor is:

[0083] ;

[0084] In the formula, represents the second score of the q-th type of behavior factor; represents the weight of the i-th behavior factor; represents the first score of the i-th behavior factor; represents that the i-th behavior factor belongs to the q-th type of behavior factor;

[0085] Step S2d: Synthesize the second scores of each type of behavior factor, and calculate the comprehensive score of all behavior factors;

[0086] Among them, the calculation formula for the comprehensive score of all behavior factors is:

[0087] ;

[0088] In the formula, represents the comprehensive score of all behavior factors; represents the weight of the q-th type of behavior factor; represents the second score of the q-th type of behavior factor; represents the number of behavior factor categories;

[0089] It should be noted that in the above, the determination methods of the weights of each basic feature, each behavior factor, and each type of behavior factor are the same. Taking the determination method of the weight of the k-th basic feature as an example, first call the influence value of the k-th basic feature, then count and accumulate the influence values of all basic features in the feature vector of the i-th behavior factor to obtain the total influence value, and then perform a ratio process on the influence value of the k-th basic feature and this total influence value to obtain the weight of the k-th basic feature. By analogy, the determination methods of the weights of each behavior factor and each type of behavior factor are the same as the determination method of the weight of each basic feature above.

[0090] Step S2e: Statistically calculate the first score of each behavior factor, the second score of each type of behavior factor, and the comprehensive score of all behavior factors, and define them as the second data of the public building;

[0091] It should be noted that after calculating the first score of each behavior factor in this application, it is necessary to classify the behavior factors and further calculate the second score of each type of behavior factor and the comprehensive score of all behavior factors. This is because considering that behavior factors are complex and multi-dimensional, directly inputting the first score into the model may lose some important structural information. However, by classifying the behavior factors and calculating the second score and comprehensive score, a richer hierarchical representation can be provided for the model. The data at each level reflects behavior characteristics at different granularities. Among them, the first score reflects the influence of a single behavior factor, the second score reflects the comprehensive influence of multiple behavior factors within the same category, and the comprehensive score effectively reflects the overall influence of all behavior factors. This multi-level data representation helps the model to more comprehensively understand the influence of behavior factors on energy consumption, thereby effectively improving the accuracy of carbon emission prediction;

[0092] In addition, due to the diversity and dynamics of behavior factors in public buildings, directly using the first score cannot fully capture these characteristics. However, through classification and calculation of the second score and comprehensive score, the model can learn the associations and patterns between different types of behavior factors. This not only enhances the generalization ability of the model, enabling it to better handle unseen data, but also improves the adaptability of the model in different scenarios. For example, the activity patterns of people in an office building are very different on weekdays and weekends. Through classification processing, it can be ensured that the model can distinguish these two situations, thereby more accurately predicting the carbon emission changes in different time periods.

[0093] Step S3: Perform preprocessing on the first data and the second data, and the preprocessing includes cleaning, verification, missing value processing, standardization, and normalization;

[0094] Step S4: Build a carbon emission prediction model, and use the preprocessed first data and second data to train the carbon emission prediction model to obtain a trained carbon emission prediction model;

[0095] It should be noted that in the training stage of the carbon emission prediction model, its training process is basically the same as that of the existing model. The main difference is that the behavior factors after quantization processing, that is, the second data, are added. Therefore, the training process of the carbon emission prediction model will not be elaborated here. In the specific training process, first, the long short-term memory network model is selected as the main model architecture because it can effectively process time series data and has good dynamic behavior capture ability. Then, data preparation is carried out, including the first data composed of energy consumption data, environmental climate data, building feature data, and operation data, and the second data composed of the first score of each behavior factor, the second score of each type of behavior factor, and the comprehensive score of all behavior factors obtained by performing quantization processing on the behavior factors of public buildings based on the method of multi-source data processing. Since a series of preprocessing operations have been performed on the first and second data in step S3, the first and second data need to be divided into three parts for model training, etc. The three parts include the training set: the first part is 70% of the data for training the model, including: energy consumption data, environmental climate data, building feature data, operation data, and behavior factor data, that is, the second data. The second part is the validation set, that is, 15% of the data for adjusting hyperparameters and monitoring model performance, and the data types are the same as above. The third part is the test set, that is, 15% of the data for final evaluation of model performance, and the data types are the same as above. Subsequently, model initialization is carried out. This process should first define the number of layers and units. For example, a model containing two layers of LSTM units is defined, and each layer contains 64 LSTM units. Then, the weights are initialized, that is, the Xavier or He initialization method is used to initialize the model weights to ensure a good distribution of the initial state. Subsequently, a loss function for measuring the difference between the model prediction value and the true value is defined, and the calculation formula is: In the formula, represents the difference value between the model prediction value and the true value; represents the total number of samples, that is, the number of samples in the dataset; represents the true value of the b-th sample; represents the prediction value of the b-th sample.

[0096] This loss function is the core objective function of model training. The smaller its value, the closer the model prediction value is to the true value;

[0097] Subsequently, hyperparameter tuning is performed. Specifically, the grid search method is adopted to try different hyperparameter combinations, such as learning rates (0.001, 0.01, 0.1), batch sizes (32, 64, 128), the number of hidden layer units (32, 64, 128), etc., or the random search method is used to randomly select hyperparameter combinations within a large range for testing, etc. The purpose of hyperparameter tuning is to find the optimal model configuration to improve the prediction accuracy and generalization ability of the model. Common hyperparameters include learning rate, batch size, the number of hidden layer units, regularization parameters, etc. Immediately afterwards, model training is carried out. Specifically, first, the training data is input into the model in batches, and the model parameters are gradually updated until the set number of iterations or convergence conditions are reached. Each batch size should be adjusted according to the hardware resources, such as 32 or 64. Subsequently, the loss value and performance metrics on the validation set should be recorded regularly to ensure that the model does not overfit or underfit. Specifically, the learning process of the model should be visually demonstrated by plotting the loss curve and accuracy curve. Finally, when the value of the loss function stabilizes on the training set and the validation set, that is, the change amplitude is less than a certain threshold, such as 0.001, the model is considered to have converged. If the model has not fully converged but has reached the preset maximum number of iterations, such as 1000 times, the training is stopped, that is, the model training is completed. Subsequently, the K-fold cross-validation method is also required to reduce the contingency caused by data partitioning and improve the generalization ability and stability of the model. Specifically, the dataset is divided into K parts. Each time, K - 1 parts are used as the training set, and the remaining one part is used as the validation set. This is repeated K times, and the average result is used as the final evaluation metric. Commonly used values of K are 5 or 10. The evaluation metrics used during validation include but are not limited to mean absolute error, coefficient of determination, etc. Finally, when the model performs well on the validation set and the test set, it is saved in a file format such as.h5 or.pkl for subsequent loading and deployment. At the same time, this newly added data type provides the model with richer dynamic information, enhances the expression ability and generalization ability of the model, reduces noise and redundancy, and thus makes the prediction results more accurate and reliable.

[0098] Step S5: Obtain the real-time data to be input, and input the real-time data into the carbon emission prediction model to obtain the carbon emission prediction result.

[0099] It should be noted that when inputting real-time data into the carbon emission prediction model, it is necessary to first perform quantization processing on the real-time behavior factors therein. Subsequently, the quantized real-time behavior factors can be combined with other real-time data to form input data, which is then input into the carbon emission prediction model, and the carbon emission prediction model makes predictions to obtain the carbon emission prediction results. Among them, other real-time data refers to newly acquired energy consumption data, environmental climate data, etc.; the carbon emission prediction results can be the carbon emission trend in a future time period, such as predicting the change trend of the carbon dioxide emissions of a building in the next 24 hours, 7 days or a month; and specific numerical predictions, that is, providing specific carbon emission values, such as the predicted emissions per hour, per day or per month.

[0100] It should be noted that the terms used in the present invention are only for describing specific embodiments and do not limit the scope of the present application. As shown in the specification of the present invention, unless the context clearly indicates an exception, words such as "a", "an", "one" and / or "the" do not specifically refer to the singular and may also include the plural. The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or also includes elements inherent to such a process, method or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method or device including the said element.

[0101] It should also be noted that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. Unless otherwise clearly specified and limited, terms such as "installed", "connected", "coupled" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting carbon emissions from public buildings based on multi-source data processing, characterized in that: The following steps are involved: Step S1, collecting energy consumption data, environmental climate data, building characteristic data and operation data of public buildings, and defining them as first data, and collecting behavioral factors of public buildings at the same time; Step S2, performing quantitative processing on the behavior factors of the public building to obtain second data of the public building; Step S2a, calculating a first score of each behavior factor of a public building; Step S2a1: for each behavior factor, define at least two basic features respectively, and construct a feature vector for each behavior factor based on the at least two basic features; Step S2a2: Based on historical data, define at least two influencing factors for each basic feature in the feature vector of each behavior factor; Step S2a3, based on each basic feature in the feature vector of each behavior factor and the influence factor of each basic feature in the feature vector of each behavior factor, calculate the influence value of each basic feature in the feature vector of each behavior factor; Step S2a4: Based on the influence value of each basic feature, a first score of each behavior factor is calculated by weighted summation; Step S2b, dividing each behavioral factor into at least two categories according to different dimensions; Step S2c, calculating a second score for each type of behavior factor based on the first score of each behavior factor in each type of behavior factor; The calculation formula for the second score of each behavioral factor is: ； represents the second score of the qth category behavioral factor; represents the weight of the i-th behavioral factor; represents the first score of the ith behavioral factor; It means that the i-th behavior factor belongs to the q-th type of behavior factor; Step S2d, combining the second scores of each type of behavioral factors to obtain a comprehensive score of all behavioral factors; The formula for calculating the composite score of all behavioral factors is: ； In the formula, represents a composite score of all behavioral factors; represents the weight of the qth type of behavioral factor; represents the number of behavioral factor categories; Step S2e, counting the first score of each behavior factor, the second score of each type of behavior factor, and the comprehensive score of all behavior factors, and defining them as the second data of the public building; Step S3, performing preprocessing on the first data and the second data, wherein the preprocessing includes cleaning, verification, missing value processing, standardization and normalization; Step S4: constructing a carbon emission prediction model, and using the preprocessed first data and second data to train the carbon emission prediction model to obtain a trained carbon emission prediction model; Step S5: Acquire real-time data to be input, and input the real-time data into the carbon emission prediction model to obtain a carbon emission prediction result.

2. According to claim 1, a method for predicting carbon emissions from public buildings based on multi-source data processing is characterized in that: The expression of the characteristic vector of each behavioral factor is: ； In the formula, The feature vector representing the i-th behavioral factor; represents the kth basic feature of the ith behavioral factor; k represents the number of basic features in the ith behavioral factor.

3. The method for predicting carbon emissions from public buildings based on multi-source data processing according to claim 2 is characterized in that: The expression of the influencing factor of each basic feature in the feature vector of each behavioral factor is: ； In the formula, The set of influencing factors of the kth basic feature in the feature vector representing the i-th behavioral factor; Represents the mth influencing factor of the kth basic feature in the feature vector of the i-th behavioral factor; m represents the number of influencing factors of the kth basic feature in the feature vector of the i-th behavioral factor, and m≥2k, that is, each basic feature has at least two influencing factors.

4. The method for predicting carbon emissions from public buildings based on multi-source data processing according to claim 3 is characterized in that: The calculation formula for the influence value of each basic feature in the feature vector of each behavioral factor is: ； In the formula, The influence value of the kth basic feature in the feature vector representing the i-th behavioral factor; Represents the jth influencing factor of the kth basic feature in the feature vector of the i-th behavioral factor; m represents the number of influencing factors of the kth basic feature in the feature vector of the i-th behavioral factor, and m≥2, that is, each basic feature has at least two influencing factors.

5. The method for predicting carbon emissions from public buildings based on multi-source data processing according to claim 4 is characterized in that: The calculation formula for the first score of each behavioral factor is: ； In the formula, represents the first score of the ith behavioral factor; Represents the weight of the kth basic feature; Represents the influence value of the kth basic feature in the feature vector of the i-th behavioral factor.

Images