Carbon emission prediction method and device, and storage medium

Abstract

This invention discloses a carbon emission prediction method, apparatus, and storage medium. The method includes: acquiring a multi-source dataset of a target area; preprocessing the multi-source dataset to obtain first two-dimensional feature maps of different categories; inputting the first two-dimensional feature maps of different categories into convolutional neural network branches of corresponding categories to obtain second two-dimensional feature maps of different categories; concatenating the second two-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and inputting the target three-dimensional feature map into a target convolutional neural network to obtain a fused feature map; and inputting the fused feature map into a multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions. This invention captures short-term fluctuations in pedestrian flow and equipment operation patterns within building units based on fine-grained local features and macro-trend features in the fused feature map, making the predicted carbon emission data more accurate.

Description

Technical Field

[0001] This invention relates to the field of carbon emission technology, and in particular to a carbon emission prediction method, apparatus and storage medium. Background Technology

[0002] Building units within integrated industrial parks (such as office buildings, factories, data centers, and public facilities) have become key scenarios for urban carbon emission reduction due to the cyclical fluctuations in personnel flow and equipment operation. Carbon emissions from these parks exhibit dynamic and complex characteristics. For example, during peak production periods, concentrated office hours, or seasonal control measures, high-load operation of air conditioning, lighting, elevators, and production equipment, coupled with the concentration of transportation, logistics, and personnel activities, can lead to a periodic surge in carbon emissions, significantly exceeding basic operational levels. Furthermore, traditional carbon emission prediction models for building units are mostly designed for steady-state scenarios such as offices and residences, making it difficult to adapt to building units with dynamic operational characteristics. Therefore, there is an urgent need for a carbon emission prediction method applicable to the carbon emission prediction scenarios of building units with dynamic operational characteristics.

[0003] In existing technologies, carbon emission prediction can be made by using statistical time series models based on historical data to mine the trends and periods of historical carbon emission data. However, this method has low sensitivity to multi-source dynamic fluctuation signals in building units (such as energy consumption changes caused by production scheduling, human activity, or seasonal regulation) and cannot correlate dynamic variables such as human behavior, resulting in inaccurate carbon emission prediction data. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] To address this, the present invention proposes a carbon emission prediction method. This method extracts features from first-dimensional feature maps of different categories obtained from multi-source datasets using a target convolutional neural network and different types of convolutional neural network branches to obtain a fused feature map. A multi-task collaborative prediction model then uses the fused feature map to obtain the target predicted value and confidence interval of carbon emissions. This allows the method to capture short-term fluctuations in pedestrian flow and equipment operation patterns within building units based on the fine-grained local features and macro-trend features in the fused feature map, making the predicted carbon emission data more accurate.

[0006] Another object of the present invention is to provide a carbon emission prediction device.

[0007] To achieve the above objectives, the present invention provides a carbon emission prediction method, the method comprising:

[0008] Acquire multi-source datasets of the target region; The multi-source dataset is preprocessed to obtain first two-dimensional feature maps of different categories; The first two-dimensional feature maps of different categories are input into the corresponding convolutional neural network branches to obtain the second two-dimensional feature maps of different categories; The second two-dimensional feature maps of different categories are concatenated to obtain the target three-dimensional feature map, and the target three-dimensional feature map is input into the target convolutional neural network to obtain the fused feature map; The fused feature map is input into the multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions.

[0009] The carbon emission prediction method of this invention may also have the following additional technical features: In one embodiment of the present invention, the multi-source dataset includes: Population density and distribution; Estimated values ​​of personnel activity intensity; Equipment energy consumption data; Carbon emission factors.

[0010] In one embodiment of the present invention, the preprocessing of the multi-source dataset to obtain first two-dimensional feature maps of different categories includes: The missing data in the multi-source dataset is processed using a hierarchical processing strategy to obtain first datasets of different categories. By introducing time embedding vectors into the first datasets of different categories, second datasets of different categories are obtained; The second datasets of different categories are subjected to feature normalization to obtain the third datasets of different categories; The third datasets of different categories are transformed to obtain the first two-dimensional feature maps of different categories.

[0011] In one embodiment of the present invention, the multi-task collaborative prediction model includes an MMoE module and multiple LSTM modules; the step of inputting the fused feature map into the multi-task collaborative prediction model to obtain the target predicted value of carbon emissions includes: The fused feature map is input into the MMoE module to obtain the first feature vectors of different subtasks; The first feature vector of each subtask is input into the LSTM module corresponding to each subtask to obtain the initial prediction value of each subtask. Based on the initial predicted values, the corresponding initial residual distribution is determined by fitting the residuals; Based on the initial predicted value, the initial residual distribution, and the preset confidence level, the target predicted value and confidence interval for carbon emissions are obtained.

[0012] In one embodiment of the present invention, the MMoE module includes multiple expert subnets and multiple gating units; the fused feature map is input into the MMoE module to obtain first feature vectors for different subtasks, including: The multiple expert subnets perform nonlinear transformations on the fused feature map to obtain multiple corresponding second feature vectors; Based on the multiple second feature vectors, the first feature vectors of different sub-tasks are obtained through the multiple gating units.

[0013] In one embodiment of the present invention, obtaining the target predicted value and confidence interval of carbon emissions based on the initial predicted value, the initial residual distribution, and a preset confidence level includes: Based on the initial residual distribution, determine the corresponding target residual distribution; Based on the initial predicted values, the target residual distribution, and the preset confidence level, the target predicted value and confidence interval for carbon emissions are determined.

[0014] To achieve the above objectives, another aspect of the present invention provides a carbon emission prediction device, the device comprising: The multi-source data acquisition module is used to acquire multi-source datasets of the target area. The data preprocessing module is used to preprocess the multi-source dataset to obtain first two-dimensional feature maps of different categories; The feature extraction module is used to input the first two-dimensional feature maps of different categories into the corresponding convolutional neural network branches to obtain the second two-dimensional feature maps of different categories; The multi-scale feature fusion module is used to stitch together the second two-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and input the target three-dimensional feature map into a target convolutional neural network to obtain a fused feature map; The multi-task collaborative prediction module is used to input the fused feature map into the multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions.

[0015] This invention discloses a carbon emission prediction method and apparatus. The method includes: acquiring a multi-source dataset of a target area; preprocessing the multi-source dataset to obtain first two-dimensional feature maps of different categories; inputting the first two-dimensional feature maps of different categories into convolutional neural network branches corresponding to those categories to obtain second two-dimensional feature maps of different categories; concatenating the second two-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and inputting the target three-dimensional feature map into a target convolutional neural network to obtain a fused feature map; and inputting the fused feature map into a multi-task collaborative prediction model to obtain a target predicted value and confidence interval for carbon emissions. Therefore, this invention can capture short-term fluctuations in pedestrian flow and equipment operation patterns within building units based on fine-grained local features and macro-trend features in the fused feature map, making the predicted carbon emission data more accurate.

[0016] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a carbon emission prediction method according to an embodiment of the present invention; Figure 2 This is a flowchart of a carbon emission prediction method according to an embodiment of the present invention; Figure 3 This is a structural diagram of a carbon emission prediction device according to an embodiment of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] A carbon emission prediction method and apparatus according to an embodiment of the present invention are described below with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of a carbon emission prediction method according to an embodiment of the present invention.

[0022] like Figure 1 As shown, the method includes: S1, Obtain the multi-source dataset of the target area; In one embodiment of the present invention, the target area can be a building unit with dynamic operational characteristics, such as a stadium, concert hall, or convention center.

[0023] In one embodiment of the present invention, the aforementioned multi-source dataset may include: Population density and distribution; Estimated values ​​of personnel activity intensity; Equipment energy consumption data; Carbon emission factors.

[0024] In one embodiment of the present invention, the methods for collecting data corresponding to the different datasets mentioned above also differ. Specifically, in one embodiment, infrared sensors can be deployed in key areas of the target area to collect personnel characteristics in real time, and the personnel density and distribution can be obtained based on these characteristics. For example, if the number of personnel in a certain key area is 15, the personnel density in that key area is obtained by dividing the number of personnel by the area of ​​the key area, and the number of personnel in that key area represents the distribution of that key area.

[0025] In one embodiment of the present invention, the activity trajectory, population density, dwell time and entry / exit status of people in a target area can be obtained through an intelligent access control system and video surveillance, and the activity intensity of people can be estimated based on the activity trajectory, population density, dwell time and entry / exit status of people in the target area.

[0026] Specifically, in one embodiment of the present invention, a linear model can be used to multiply the features corresponding to personnel activity trajectories, personnel density, dwell time, and entry / exit status with their corresponding weights and then sum them to obtain an estimated value of personnel activity intensity. In another embodiment of the present invention, a nonlinear model can be used to nonlinearly combine and learn the data corresponding to personnel activity trajectories, personnel density, dwell time, and entry / exit status to obtain an estimated value of personnel activity intensity. The aforementioned nonlinear model can be a machine learning model such as a neural network or a support vector machine.

[0027] In one embodiment of the present invention, the energy consumption data of the equipment can be obtained by monitoring the daily energy consumption data (such as electricity, water and gas consumption) of various equipment in the target area through smart meters or Internet of Things devices.

[0028] In one embodiment of the present invention, the carbon emission factors of various energy sources in the target area can be obtained from the national greenhouse gas emission factor database or the IPCC carbon emission factor database, based on the power sources (such as thermal power, hydropower, wind power, etc.) in the target area.

[0029] In one embodiment of the present invention, the data in the aforementioned multi-source dataset can correspond to different categories. Specifically, personnel density and distribution correspond to personnel distribution categories, estimated values ​​of personnel activity intensity correspond to personnel behavior categories, equipment energy consumption data correspond to energy consumption categories, and carbon emission factors correspond to environmental categories.

[0030] S2, preprocess the multi-source dataset to obtain the first two-dimensional feature maps of different categories; In one embodiment of the present invention, after obtaining the multi-source dataset of the target region through the above steps, the multi-source dataset can be preprocessed to obtain first two-dimensional feature maps of different categories.

[0031] Specifically, in one embodiment of the present invention, the method for preprocessing multi-source datasets to obtain first two-dimensional feature maps of different categories may include the following steps: S21, using a hierarchical processing strategy to process missing data in the multi-source dataset, to obtain the first dataset of different categories; S22, introduce time embedding vectors into the first dataset of different categories to obtain the second dataset of different categories; S23, perform feature normalization on the second datasets of different categories to obtain the third datasets of different categories; S24 transforms the third datasets of different categories to obtain the first two-dimensional feature maps of different categories.

[0032] In one embodiment of the present invention, the method for processing missing data in a multi-source dataset using a hierarchical processing strategy to obtain first datasets of different categories may include: determining the duration of consecutively missing data in the multi-source dataset; if the duration of consecutively missing data is less than or equal to a preset time, then linear interpolation is used to fill the missing data to obtain first datasets of different categories; if the duration of consecutively missing data is greater than the preset time, then the missing data is marked as missing values ​​to obtain first datasets of different categories. In one embodiment of the present invention, the preset time can be set as needed, such as 5 hours.

[0033] In one embodiment of the present invention, the data in the first dataset of different categories are time-labeled (e.g., weekdays, holidays, work schedules) to introduce time embedding vectors into the first dataset of different categories, thereby obtaining the second dataset of different categories. This can enhance the ability to capture periodic patterns and help subsequent models understand cross-day feature associations.

[0034] In one embodiment of the present invention, the numerical data in the second dataset of different categories are standardized by Min-Max, and the data in the second dataset of energy consumption category are logarithmically transformed to obtain the third dataset of different categories, so as to alleviate the long-tail distribution and ensure that the dimensions of each feature are consistent.

[0035] In one embodiment of the present invention, the third datasets of different categories are transformed to obtain first two-dimensional feature maps of different categories. Specifically, in one embodiment of the present invention, the data in the third dataset of each category are transformed by time step × feature dimension to obtain the first two-dimensional feature maps corresponding to different categories.

[0036] S3, input the first two-dimensional feature maps of different categories into the convolutional neural network branch of the corresponding category to obtain the second two-dimensional feature maps of different categories; In one embodiment of the present invention, the first two-dimensional feature map of each category is used as the input of the corresponding convolutional neural network branch, and convolution is performed with the corresponding convolutional kernel to obtain the second two-dimensional feature map of different categories, thereby extracting the features of each category.

[0037] For example, in one embodiment of the present invention, assuming that the personnel distribution category corresponds to the first convolutional neural network branch, the personnel behavior category corresponds to the second convolutional neural network branch, the energy consumption category corresponds to the third convolutional neural network branch, and the environment category corresponds to the fourth neural network branch, then the second two-dimensional feature map of the corresponding category can be obtained through the first convolutional neural network branch, the second convolutional neural network branch, the third convolutional neural network branch, and the fourth neural network branch.

[0038] S4. The second two-dimensional feature maps of different categories are stitched together to obtain the target three-dimensional stereo map, and the target three-dimensional stereo map is input into the target convolutional neural network to obtain the fused feature map; In one embodiment of the present invention, after obtaining second two-dimensional feature maps of different categories through the above steps, the second two-dimensional feature maps of different categories can be stitched together to obtain a target three-dimensional feature map, and the target three-dimensional feature map is input into a target convolutional neural network to obtain a fused feature map.

[0039] Specifically, in one embodiment of the present invention, second two-dimensional feature maps of different categories are concatenated along the channel dimension to obtain a target three-dimensional feature map, and a target convolutional neural network is used with convolutional kernels of different sizes (such as 3D convolutional kernels). 3, 7 7, 12 12) Extract multi-scale temporal features, and obtain a fused feature map by averaging the feature values ​​on each channel. This enables cross-scale feature extraction from fine-grained local features to macro trends, thereby improving the robustness of subsequent predictions.

[0040] S5 inputs the fused feature map into the multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions.

[0041] In one embodiment of the present invention, after obtaining the fused feature map through the above steps, the fused feature map can be input into a multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions.

[0042] In one embodiment of the present invention, the multi-task collaborative prediction model may include an MMoE module and multiple LSTM modules.

[0043] Specifically, in one embodiment of the present invention, the method of inputting the fused feature map into a multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions may include the following steps: S51, input the fused feature map into the MMoE module to obtain the first feature vector of different subtasks; S52, input the first feature vector of different subtasks into the LSTM module corresponding to each subtask to obtain the initial prediction value of each subtask; S53, Based on the initial predicted values, fit the residuals to determine the corresponding initial residual distribution; S54, based on the initial predicted value, the initial residual distribution and the preset confidence level, obtains the target predicted value and confidence interval of carbon emissions.

[0044] In one embodiment of the present invention, the MMoE module includes multiple expert subnets and multiple gating units, with each gating unit connected to each expert subnet. In another embodiment of the present invention, the method for inputting the fused feature map into the MMoE module to obtain the first feature vectors of different subtasks may include the following steps: S511, multiple expert subnets perform nonlinear transformations on the fused feature map to obtain multiple corresponding second feature vectors; S512 obtains the first feature vectors of different subtasks based on multiple second feature vectors through multiple gating units.

[0045] In one embodiment of the present invention, each of the above-mentioned multiple expert subnets performs a nonlinear transformation (activation function such as ReLU, Tanh or Sigmoid) on the fused feature map to obtain multiple corresponding second feature vectors, thereby enabling the learning of complex patterns in the fused feature map and extracting features from the fused feature map from different perspectives, and sharing feature parameters in multi-task learning.

[0046] It is important to note that, in one embodiment of the present invention, the input of the aforementioned expert subnet does not depend on a specific subtask, but rather extracts a general second feature vector from the fused feature map. This second feature vector can be shared and utilized by multiple tasks. Furthermore, different gating units can generate corresponding expert weights for different subtasks (such as predicting energy consumption), thereby outputting relevant first feature vectors for different subtasks, thus preserving task diversity while reducing overfitting.

[0047] In one embodiment of the present invention, each subtask (such as an energy consumption prediction task like predicting electricity consumption and water consumption) uses a corresponding independent LSTM module to perform time-series modeling on the subtask-specific features to obtain the corresponding initial prediction value for each subtask. In one embodiment of the present invention, the initial prediction value output by the LSTM module can be a carbon emission prediction value after a preset future time, which can be set as needed, such as one hour in the future.

[0048] In one embodiment of the present invention, after obtaining the initial predicted value for each subtask through the above steps, the true value corresponding to the initial predicted value in the historical data can be determined, and the initial residual distribution of the initial predicted value can be determined based on the residual between the initial predicted value and the true value. , ).

[0049] In one embodiment of the present invention, after obtaining the initial residual distribution of each subtask through the above steps, the method for obtaining the target predicted value and confidence interval of carbon emissions based on the initial predicted value, the initial residual distribution and the preset confidence level may include: determining the corresponding target residual distribution based on the initial residual distribution, and determining the target predicted value and confidence interval of carbon emissions based on each initial predicted value, the target residual distribution and the preset confidence level.

[0050] In one embodiment of the present invention, the method for determining the corresponding target residual distribution based on the initial residual distribution may include: determining the mean and variance of the residual distributions of all sub-tasks as the values ​​of the target residual distribution. and .

[0051] Furthermore, in one embodiment of the present invention, the initial predicted value for each of the above sub-tasks can be the energy consumption (e.g., electricity consumption) of each sub-task. In one embodiment of the present invention, the energy consumption of each sub-task is multiplied by its corresponding carbon emission factor and then weighted to obtain the target predicted value P for carbon emissions. C。 In one embodiment of the present invention, the weights can be set as needed, such as all being 1.

[0052] Furthermore, in one embodiment of the present invention, the preset confidence level can be set empirically, such as a confidence level of 95%, and the corresponding... = 1.96.

[0053] Furthermore, in one embodiment of the present invention, the target predicted value P of carbon emissions is obtained through the above steps. C In the target residual distribution and and the confidence level corresponding to Then, the confidence interval can be determined using the confidence interval formula as [[ ].

[0054] Furthermore, in one embodiment of the present invention, after obtaining the target predicted value and confidence interval of carbon emissions through the above steps, a dynamic optimization strategy can be generated based on the target predicted value and confidence interval. For example, in one embodiment of the present invention, if the target predicted value is greater than a preset threshold, the main sources of risk are identified by analyzing and determining the contribution rate of the uncertainty in the energy consumption prediction of each sub-task. For instance, if the contribution rate of the uncertainty in the air conditioning system reaches 65%, it is determined to be an air conditioning system risk, and the corresponding dynamic optimization strategy is to activate the building unit spray cooling system and adjust the air conditioning set temperature in different areas to cope with it. In one embodiment of the present invention, the contribution rate of the uncertainty in the energy consumption prediction of each sub-task is the percentage of the uncertainty in the energy consumption prediction of each sub-task. In the target The percentage.

[0055] According to an embodiment of the present invention, a carbon emission prediction method includes: acquiring a multi-source dataset of a target area; preprocessing the multi-source dataset to obtain first two-dimensional feature maps of different categories; inputting the first two-dimensional feature maps of different categories into convolutional neural network branches of corresponding categories to obtain second two-dimensional feature maps of different categories; concatenating the second two-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and inputting the target three-dimensional feature map into a target convolutional neural network to obtain a fused feature map; and inputting the fused feature map into a multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions. Therefore, the present invention can capture short-term fluctuations in pedestrian flow and equipment operation patterns in building units based on fine-grained local features and macro-trend features in the fused feature map, making the predicted carbon emission data more accurate.

[0056] Based on the above description Figure 2 A carbon emission prediction method proposed in this embodiment of the invention, such as... Figure 2As shown, a multi-source dataset of the target region is obtained; the multi-source dataset is preprocessed to obtain first two-dimensional feature maps of different categories; the first two-dimensional feature maps of different categories are input into a multi-scale feature extraction layer, and in the multi-scale feature extraction layer, the first two-dimensional feature maps of different categories are input into the corresponding CNN branches to obtain second two-dimensional feature maps of different categories, and the second two-dimensional feature maps of different categories are concatenated to obtain a target three-dimensional feature map, and the target three-dimensional feature map is input into a target convolutional neural network to obtain a fused feature map; the fused feature map is input into a prediction layer, and in the prediction layer, multiple expert subnetworks and multiple gating units in the MMoE module are used to obtain the first feature vectors of different sub-tasks, and the first feature vectors of different sub-tasks are input into the LSTM modules corresponding to each sub-task to obtain the initial prediction values ​​of each sub-task; the initial prediction values ​​of each sub-task are input into the output layer, and in the output layer, based on the initial prediction values ​​and initial residual distribution of each sub-task and the preset confidence level, the target prediction value and confidence interval of carbon emissions are obtained.

[0057] To achieve the above embodiments, such as Figure 3 As shown, this embodiment also provides a carbon emission prediction device 10, which includes a multi-source data acquisition module 301, a data preprocessing module 302, a feature extraction module 303, a multi-scale feature fusion module 304, and a multi-task collaborative prediction module 305. The multi-source data acquisition module 301 is used to acquire the multi-source dataset of the target area. Data preprocessing module 302 is used to preprocess multi-source datasets to obtain first two-dimensional feature maps of different categories; The feature extraction module 303 is used to input the first two-dimensional feature maps of different categories into the corresponding convolutional neural network branch to obtain the second two-dimensional feature maps of different categories. The multi-scale feature fusion module 304 is used to stitch together second-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and input the target three-dimensional feature map into the target convolutional neural network to obtain a fused feature map; The multi-task collaborative prediction module 305 is used to input the fused feature map into the multi-task collaborative prediction model to obtain the target predicted value and confidence interval of carbon emissions.

[0058] In one embodiment of the present invention, the above-mentioned multi-source dataset includes: Population density and distribution; Estimated values ​​of personnel activity intensity; Equipment energy consumption data; Carbon emission factors.

[0059] Furthermore, the aforementioned data preprocessing module 302 is specifically used for: A hierarchical processing strategy is used to process missing data in multi-source datasets to obtain first datasets of different categories. By introducing temporal embedding vectors into the first dataset for different categories, we obtain the second dataset for different categories. Feature normalization is performed on the second datasets of different categories to obtain the third datasets of different categories; The third datasets of different categories are transformed to obtain the first two-dimensional feature maps of different categories.

[0060] Furthermore, the aforementioned multi-task collaborative prediction model includes an MMoE module and multiple LSTM modules; the aforementioned multi-task collaborative prediction module 305 is specifically used for: The fused feature map is input into the MMoE module to obtain the first feature vectors of different subtasks; The first feature vector of each subtask is input into the LSTM module corresponding to each subtask to obtain the initial prediction value of each subtask. Based on the initial predicted values, the fitting residuals determine the corresponding initial residual distribution; Based on the initial predicted values, residual distribution, and preset confidence levels, the target predicted values ​​and confidence intervals for carbon emissions are obtained.

[0061] Furthermore, the aforementioned MMoE module includes multiple expert subnets and multiple gating units, and the aforementioned multi-task collaborative prediction module 305 is also used for: Multiple expert subnets perform nonlinear transformations on the fused feature map to obtain multiple corresponding second feature vectors; Based on multiple second feature vectors, first feature vectors for different subtasks are obtained through multiple gating units.

[0062] Furthermore, the aforementioned multi-task collaborative prediction module 305 is also used for: Based on the initial residual distribution, determine the corresponding target residual distribution; Based on the initial predicted values, the target residual distribution, and the preset confidence level, the target predicted value and confidence interval for carbon emissions are determined.

[0063] According to the carbon emission prediction device of the present invention, the device can extract features from the first two-dimensional feature maps of different categories obtained from multi-source datasets through a target convolutional neural network and different types of convolutional neural network branches to obtain a fused feature map. Based on the fused feature map, a multi-task collaborative prediction model is used to obtain the target predicted value and confidence interval of carbon emissions. Thus, based on the fine-grained local features and macro-trend features in the fused feature map, the device can capture the short-term fluctuations in human flow and the operation patterns of equipment in building units, making the predicted carbon emission data more accurate.

[0064] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0065] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A carbon emission prediction method characterized by, The method comprises the following steps: acquiring a multi-source data set of a target area; preprocessing the multi-source data set to obtain first two-dimensional feature maps of different categories; inputting the first two-dimensional feature maps of different categories into corresponding convolutional neural network branches to obtain second two-dimensional feature maps of different categories; splicing the second two-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and inputting the target three-dimensional feature map into a target convolutional neural network to obtain a fusion feature map; inputting the fusion feature map into a multi-task collaborative prediction model to obtain a target prediction value and a confidence interval of carbon emissions.

2. The method of claim 1, wherein, The multi-source data set comprises: personnel density and distribution; estimated value of personnel activity intensity; equipment energy consumption data; carbon emission factor.

3. The method of claim 2, wherein, The preprocessing of the multi-source data set to obtain first two-dimensional feature maps of different categories comprises the following steps: performing missing data processing on the multi-source data set using a hierarchical processing strategy to obtain first data sets of different categories; introducing time embedding vectors into the first data sets of different categories to obtain second data sets of different categories; performing feature normalization processing on the second data sets of different categories to obtain third data sets of different categories; converting the third data sets of different categories to obtain first two-dimensional feature maps of different categories.

4. The method of claim 1, wherein, The multi-task collaborative prediction model comprises an MMoE module and a plurality of LSTM modules; inputting the fusion feature map into the MMoE module to obtain first feature vectors of different sub-tasks; inputting the first feature vectors of different sub-tasks into corresponding LSTM modules of each sub-task to obtain initial prediction values of each sub-task; based on the initial prediction values, fitting residual error to determine the initial residual error distribution; based on the initial prediction values, the initial residual error distribution and a preset confidence level, a target prediction value and a confidence interval of carbon emissions are obtained. The MMoE module comprises a plurality of expert subnets and a plurality of gating units; inputting the fusion feature map into the MMoE module to obtain first feature vectors of different sub-tasks comprises: the plurality of expert subnets perform nonlinear transformation on the fusion feature map to obtain a plurality of second feature vectors; based on the plurality of second feature vectors, the first feature vectors of different sub-tasks are obtained through the plurality of gating units. Based on the initial prediction values, the initial residual error distribution and a preset confidence level, a target prediction value and a confidence interval of carbon emissions are obtained, which comprises:

5. The method of claim 4, wherein, based on the initial residual error distribution, a corresponding target residual error distribution is determined; based on each of the initial prediction values, the target residual error distribution and a preset confidence level, a target prediction value and a confidence interval of carbon emissions are determined. The method comprises the following steps:

6. The method of claim 4, wherein, a multi-source data acquisition module is configured to acquire a multi-source data set of a target area; a data preprocessing module is configured to preprocess the multi-source data set to obtain first two-dimensional feature maps of different categories; ​ 7. A carbon emission prediction device characterized by comprising: ​ ​ ​ a feature extraction module, configured to input the first two-dimensional feature maps of different categories into corresponding convolutional neural network branches of the categories to obtain second two-dimensional feature maps of different categories; a multi-scale feature fusion module, configured to splice the second two-dimensional feature maps of different categories to obtain a target three-dimensional feature map, and input the target three-dimensional feature map into a target convolutional neural network to obtain a fused feature map; a multi-task collaborative prediction module, configured to input the fused feature map into a multi-task collaborative prediction model to obtain a target prediction value and a confidence interval of carbon emissions.

8. The apparatus of claim 7, wherein, The multi-source data set comprises: personnel density and distribution; estimated value of personnel activity intensity; equipment energy consumption data; carbon emission factor. 9.An electronic device comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-6.

Images