Carbon emission flow monitoring method, computer equipment and computer storage medium

Abstract

The invention relates to a carbon emission flow monitoring method, computer equipment and a computer storage medium, and belongs to the technical field of environment monitoring. According to the method, quality scoring is carried out on the multi-source environment detection data by using an adaptive fusion method, the weight corresponding to the environment detection data of each data source is determined according to the quality score, and weighted summation is carried out on the environment detection data of each data source by using the obtained weight to obtain a high-dimensional spatial-temporal feature; the obtained high-dimensional spatial-temporal features are effectively fused with multi-source environment detection data, the problem of unstable data quality is solved, the prediction precision can be improved when a prediction model is used for prediction subsequently, the method can adapt to data of various qualities, the stability of the prediction model is ensured, and thus the precision and real-time performance of carbon emission flow monitoring are improved.

Description

Technical Field

[0001] The invention relates to a carbon emission flow monitoring method, computer equipment and computer storage medium, belonging to the technical field of environmental monitoring. Background Art

[0002] As the global climate change problem becomes increasingly serious, reducing greenhouse gas emissions, especially carbon dioxide emissions, has become an important task for governments around the world. As the basis for reducing carbon emissions, carbon emission monitoring technology has received widespread attention in recent years. Traditional carbon emission monitoring methods mainly rely on fixed-point sensor monitoring, combined with manual statistics and model prediction to evaluate carbon emissions. Although such methods have a certain degree of accuracy in certain specific applications, they often have problems such as limited monitoring range and inability to effectively cope with large and complex scenarios. In addition, with the development of remote sensing technology and sensor networks, multi-source data for carbon emission monitoring is becoming more and more abundant. How to efficiently integrate and utilize these multi-source data and use deep learning technology to conduct intelligent analysis has become one of the research hotspots.

[0003] In the existing technology, although some machine learning models are introduced in carbon emission monitoring, the processing ability of complex spatiotemporal data is weak, and the spatiotemporal characteristics of multi-source data are not fully utilized. For example, the existing method cannot effectively capture the law of carbon emission changes from the spatiotemporal dimension, resulting in poor accuracy and real-time performance of the prediction results. For this reason, some people have proposed a machine learning model that considers time series and spatiotemporal dimensions for carbon emission monitoring, such as the Chinese patent application document with application publication number CN116882614A, which discloses a method and device for building a carbon emission prediction model, and specifically discloses that by screening the carbon emission influencing factor data, the key carbon emission influencing factors in the carbon emission influencing factor data are identified; the key carbon emission influencing factors are learned by using the constructed CNN-GRU model to predict carbon emissions. The CNN-GRU model also introduces an attention mechanism to obtain the attention right distribution value corresponding to the key carbon emission influencing factor data. It can be seen that the scheme published in the document can improve the prediction accuracy of the model, but the prediction accuracy of the scheme is still difficult to guarantee in the case of missing data or poor data quality, resulting in poor prediction stability, which in turn affects the accuracy of carbon emission flow monitoring. Summary of the invention

[0004] The purpose of the present invention is to provide a carbon emission flow monitoring method, a computer device and a computer storage medium to solve the problem of inaccurate carbon emission flow monitoring caused by the current carbon emission prediction being affected by data quality.

[0005] In order to solve the above technical problems, the present invention provides a carbon emission flow monitoring method, which comprises:

[0006] 1) Obtain multi-source environmental monitoring data related to carbon emission flows and pre-process them so that the representation, unit and format of environmental monitoring data from each data source are consistent;

[0007] 2) Detect the outliers and time series continuity of the preprocessed environmental detection data, evaluate the data quality based on the detection results to determine the weights corresponding to the environmental detection data of each data source, and use the obtained weights to perform weighted summation on the environmental detection data of each data source to obtain the fusion feature;

[0008] 3) Inputting the obtained fusion features into the constructed carbon emission prediction model for prediction, obtaining the carbon emission prediction value, and monitoring the carbon emission flow according to the carbon emission prediction value; the carbon emission prediction model is constructed by a deep learning model.

[0009] Furthermore, in the step 2), the outliers of the environmental monitoring data are detected by the 3σ principle, and the time series continuity of the environmental monitoring data is determined by the time difference between adjacent time sampling points of the environmental monitoring data. The quality of the preprocessed environmental monitoring data is evaluated by the proportion of outliers in the total data points and the proportion of discontinuous points in the total time points. The higher the data quality, the greater the corresponding weight, and the lower the data quality, the smaller the corresponding weight.

[0010] Furthermore, the weight corresponding to the environmental detection data of each data source is calculated using the following formula:

[0011]

[0012] where w i is the weight corresponding to the environmental detection data of the i-th data source, c i is the quality score of the environmental monitoring data of the ith data source, and e is a natural constant.

[0013] Furthermore, the fusion features input into the constructed carbon emission prediction model are the key spatiotemporal features obtained after adjustment using the spatiotemporal attention mechanism. The fusion features before adjustment are called high-dimensional spatiotemporal features. The process of determining the key spatiotemporal features is as follows:

[0014] The obtained high-dimensional spatiotemporal features are extracted and arranged at each time step and spatial position to obtain a high-dimensional spatiotemporal feature matrix;

[0015] Use high-dimensional spatiotemporal feature matrix to calculate temporal attention weight and spatial attention weight;

[0016] The key spatiotemporal features are calculated based on the obtained temporal attention weights and spatial attention weights. The calculation formula used for the key spatiotemporal features is:

[0017]

[0018] γ t，m =α t β m

[0019] Where K is the key spatiotemporal feature, M is the number of spatial locations, T is the number of time steps, and U is the t,m is the feature at time step t and spatial position m, β m is the attention weight of spatial position m, α t is the attention weight at time step t.

[0020] Furthermore, the deep learning model used in the carbon emission prediction model is CNN-GRU, and the prediction process includes:

[0021] The convolution kernel in CNN is used to extract the local spatial features of the key spatiotemporal features of the input, and the pooling operation is performed through the pooling layer in CNN to achieve feature downsampling, and the features after convolution and pooling operations are flattened into a one-dimensional vector;

[0022] Recursively process the one-dimensional vector using the reset gate and update gate in the GRU to obtain the final hidden state, and input it into the fully connected layer;

[0023] The fully connected layer predicts carbon emissions through a linear regression function, and the prediction results are:

[0024] y=H fc ·h T +b fc Among them, y is the predicted carbon emissions, H fc is the weight matrix of the fully connected layer, h T is the final hidden state, b fc is the bias term of the fully connected layer.

[0025] Furthermore, the monitoring of the carbon emission flow according to the carbon emission prediction value in step 3) includes evaluating the carbon emission of different regions using the carbon emission prediction value, and the calculation formula used is:

[0026]

[0027] Where Q g,q is the carbon emission of region q, A(t,s) is the spatiotemporal attention score of the high-dimensional spatiotemporal feature at time step t and spatial position s, T is the total number of time steps, and y r,t is the predicted value of carbon emissions at time step t.

[0028] Furthermore, the preprocessing in step 1) includes repairing or filling in missing or invalid data of the environmental detection data of each data source, and calibrating and synchronizing the timestamps of the environmental detection data of each data source.

[0029] Furthermore, the multi-source environmental monitoring data related to the carbon emission flow includes field sensor data, satellite remote sensing data and meteorological data.

[0030] The present invention also provides a computer device, comprising a processor, wherein the processor is used to execute a computer program to implement the steps of the carbon emission flow monitoring method of the present invention.

[0031] The present invention also provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the carbon emission flow monitoring method of the present invention are implemented.

[0032] The beneficial effects of the present invention are as follows: as an improved invention, the present invention uses an adaptive fusion method to perform quality scoring on multi-source environmental detection data, determines the weights corresponding to the environmental detection data of each data source according to the quality score, and uses the obtained weights to perform weighted summation on the environmental detection data of each data source to obtain high-dimensional spatiotemporal features. The obtained high-dimensional spatiotemporal features effectively fuse the multi-source environmental detection data, overcome the problem of unstable data quality, and enable the subsequent prediction accuracy to be improved when the prediction model is used for prediction, and can adapt to data of various qualities, thereby ensuring the stability of the prediction model, thereby improving the accuracy and real-time performance of carbon emission flow monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of the carbon emission flow monitoring method of the present invention;

[0034] Figure 2 It is a schematic diagram of the software modules used in the carbon emission flow monitoring method of the present invention. DETAILED DESCRIPTION

[0035] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.

[0036] The present invention introduces an adaptive fusion method and a spatiotemporal attention mechanism, uses the adaptive fusion method to perform quality scoring on multi-source environmental detection data, determines the weights corresponding to the environmental detection data of each data source according to the quality scores, and uses the obtained weights to perform weighted summation on the environmental detection data of each data source to obtain high-dimensional spatiotemporal features. The obtained high-dimensional spatiotemporal features effectively fuse the multi-source environmental detection data and overcome the problem of unstable data quality; this enables the prediction accuracy to be improved when the prediction model is used for subsequent prediction, and can adapt to data of various qualities, thereby ensuring the stability of the prediction model and improving the accuracy and real-time performance of carbon emission flow monitoring.

[0037] Embodiments of carbon emission flow monitoring methods

[0038] The present invention first obtains multi-source environmental detection data related to carbon emission flow and preprocesses it; then, the preprocessed multi-source environmental detection data is scored for quality, and the weights corresponding to the environmental detection data of each data source are determined according to the quality score, and the obtained weights are used to perform weighted summation on the environmental detection data of each data source to obtain fusion features; finally, the obtained fusion features are input into the constructed carbon emission prediction model for prediction, and the carbon emission prediction value is obtained to realize the monitoring of carbon emission flow; the carbon emission prediction model is constructed by a deep learning model. The specific implementation process of this method is as follows: Figure 1 As shown, the detailed description is given below.

[0039] 1. Obtain multi-source environmental monitoring data related to carbon emission flows and preprocess them.

[0040] The multi-source environmental monitoring data acquired by the present invention includes field sensor data, satellite remote sensing data and meteorological information. Since the units and formats of the data from each data source are different, in order to facilitate subsequent processing, the present invention needs to perform preprocessing of the data source in a standardized format, and uniformly convert the environmental monitoring data from different sources to ensure that the data representation, unit and format are consistent. Taking into account that each data source may be out of sync with time, the preprocessing of the present invention is to repair or fill in the missing or invalid data generated due to various reasons during the data collection process by deleting the records with more missing values; the timestamps from different data sources are calibrated and synchronized by interpolation to ensure that the time dimensions of different data sources are consistent.

[0041] 2. Based on the adaptive fusion method, the pre-processed environmental detection data from various data sources are fused to obtain high-dimensional spatiotemporal features.

[0042] The present invention performs quality assessment on the preprocessed multi-source environmental detection data, determines the weights corresponding to the environmental detection data of each data source according to the quality assessment results, and uses the obtained weights to perform weighted summation on the environmental detection data of each data source to obtain a fusion feature, which is a high-dimensional spatiotemporal feature.

[0043] This implementation evaluates the quality of preprocessed environmental monitoring data through a data integrity method, which includes checking outliers and checking time series continuity. The outliers of the environmental monitoring data are detected through the 3σ principle, and the time series continuity of the environmental monitoring data is determined by the time difference between adjacent time sampling points of the environmental monitoring data.

[0044] 1) Check for outliers, as follows:

[0045] The average value of the preprocessed environmental detection data is calculated using the mean formula, and the expression is:

[0046]

[0047] Among them, μ is the average value of the preprocessed environmental detection data, is the total number of preprocessed environmental detection data, x z is the value of the zth preprocessed environmental detection data, and z is the index variable of the preprocessed environmental detection data.

[0048] The standard deviation of the preprocessed environmental detection data is calculated using the standard deviation formula, and the expression is:

[0049]

[0050] Among them, σ is the standard deviation of the preprocessed environmental detection data; according to the 3σ principle, the values ​​of the preprocessed environmental detection data exceeding μ±3σ are marked as abnormal values, as follows:

[0051]

[0052] Calculate the proportion of outliers in the total data points as follows:

[0053] J=L / O

[0054] Among them, J is the proportion of outliers in the total data points, L is the number of outliers, and O is the total number of data points; the quality of the pre-processed environmental detection data is evaluated based on the proportion of outliers in the total data points, as follows: if the proportion of outliers is less than 5%, the quality of the pre-processed environmental detection data is high; if the proportion of outliers is between 5% and 20%, the quality of the pre-processed environmental detection data is medium; if the proportion of outliers is higher than 20%, the quality of the pre-processed environmental detection data is low. It should also be noted that the 3σ principle means that the probability of a data point falling outside the mean plus or minus 3 times the standard deviation is very small, and data points exceeding μ+3σ are usually regarded as outliers.

[0055] 2) Check the continuity of the time series, as follows:

[0056] Calculate the time difference between adjacent time points. The expression is:

[0057] Δf k =f k -f k-1

[0058] Where Δf k is the time difference between the kth time point and the k-1th time point, f k is the time of the kth time point, f k-1 is the time between the k-1th time points; according to the sampling frequency of the data, this embodiment sets a reasonable time difference range [Δt min,Δt max ].

[0059] Mark the points where the time difference is not within a reasonable time range as discontinuous points, as follows:

[0060]

[0061] Calculate the proportion of discontinuous points in the total time points as follows:

[0062]

[0063] Where D is the proportion of discontinuous points in the total time points, is the number of discontinuities, is the total number of time points. Based on the proportion of discontinuous points in the total time points, the quality of the preprocessed environmental detection data is evaluated as follows: if the proportion of discontinuous points is less than 5%, the quality of the preprocessed environmental detection data is high; if the proportion of discontinuous points is between 5% and 20%, the quality of the preprocessed environmental detection data is medium; if the proportion of discontinuous points is higher than 20%, the quality of the preprocessed environmental detection data is low.

[0064] 3) Conduct comprehensive quality assessment and weight determination

[0065] Based on the results of the evaluation of the proportion of outliers in the total data points and the results of the evaluation of the proportion of discontinuous points in the total time points, the quality of the pre-processed environmental detection data is comprehensively evaluated. The specific steps are as follows:

[0066] High quality: If all indicators in the above two aspects perform well (low proportion of outliers and low proportion of discontinuous points), the quality of the pre-processed environmental detection data can be considered to be high; Medium quality: If some indicators in the above two aspects perform well and some indicators perform generally, the quality of the pre-processed environmental detection data can be considered to be medium; Low quality: If multiple indicators in the above two aspects perform poorly (high proportion of outliers and high proportion of discontinuous points), the quality of the pre-processed environmental detection data can be considered to be low. Scoring can be performed according to the quality. For example, if low quality is assigned a score of 0.3, medium quality can be assigned a score of 0.6, and high quality can be assigned a score of 0.9. As another implementation method, a score can also be calculated based on the size of the outlier ratio and the size of the discontinuous point ratio.

[0067] The weights are assigned according to the quality assessment results of the environmental detection data of the preprocessed data source. The higher the data quality, the greater the corresponding weight, and the lower the data quality, the lower the corresponding weight. In this embodiment, the weights are assigned by Softmax and normalized. The calculation formula for the weights of the environmental detection data of each data source is:

[0068]

[0069] where w i is the weight corresponding to the environmental detection data of the i-th data source, c i is the quality score of the environmental detection data of the i-th data source, c j is the quality score of the environmental monitoring data of the jth data source, and e is a natural constant.

[0070] The weight c of the environmental detection data of each data source is obtained by i The fusion feature can be calculated. The weighted fusion calculation used in this embodiment is:

[0071]

[0072] Among them, G is the high-dimensional spatiotemporal feature, N is the number of environmental detection data sources, and w i is the normalized weight of the ith data source, E i is the feature of the ith data source. The fusion feature of each data source can be calculated through the above calculation formula, and the fusion feature is the high-dimensional spatiotemporal feature.

[0073] 3. Adjust high-dimensional spatiotemporal features through the spatiotemporal attention mechanism to obtain key spatiotemporal features.

[0074] Since the multi-source environmental detection data related to carbon emission flows obtained have spatiotemporal characteristics, in order to improve the performance and effect of the prediction model when processing spatiotemporal data, the present invention uses the spatiotemporal attention mechanism to process the high-dimensional spatiotemporal features to obtain key spatiotemporal features. The spatiotemporal attention mechanism refers to a method in which a neural network performs weighted processing on information at different positions and times according to the importance and relevance of the data when processing spatiotemporal data. This method can adaptively focus on important parts and ignore unimportant parts to improve processing efficiency and accuracy. This method measures the importance of different time steps and spatial positions to the prediction task by calculating the attention weights of the time and space dimensions respectively, and combines the two weights. The process of adjusting high-dimensional spatiotemporal features by the spatiotemporal attention mechanism is described below.

[0075] 1) Extract and arrange the feature data of high-dimensional spatiotemporal features at each time step and spatial position to construct a high-dimensional spatiotemporal feature matrix.

[0076] 2) The temporal attention weight is calculated through the spatiotemporal attention mechanism and the high-dimensional spatiotemporal feature matrix, and the expression is:

[0077]

[0078] Among them, α t is the attention weight at time step t, u tis the learnable feature vector at time step t, T represents the transpose, and W t is the learnable weight matrix at the time step, F t is the eigenvector at time step t, u t′ is the feature vector of other learnable time step t′, F t′ is the feature vector of other time steps t′, and T is the total number of time steps.

[0079] 3) The spatial attention weight is calculated through the spatiotemporal attention mechanism and the high-dimensional spatiotemporal feature matrix, and the expression is:

[0080]

[0081] Among them, β m is the attention weight of spatial position m, v m is the learnable spatial position feature vector, I s is the learnable weight matrix at the spatial position, P m is the eigenvector of the spatial position, m′ is the other spatial positions, and M is the total number of spatial positions.

[0082] 4) The temporal attention weight and the spatial attention weight are combined by element-wise multiplication to obtain the spatiotemporal attention weight.

[0083] It should be noted that the spatiotemporal attention weights are obtained by element-by-element multiplication, and the expression is:

[0084] γ t，m =α t β m

[0085] The key spatiotemporal features are calculated based on the spatiotemporal attention weights, and the expression is:

[0086]

[0087] Among them, K is the key spatiotemporal feature, γ t，m is the spatiotemporal attention weight, U t，m is the feature at time step t and spatial position m.

[0088] 4. Use deep learning models to build a prediction model for carbon emissions.

[0089] The prediction model in this embodiment is a combination of a convolutional neural network model CNN and a gated recurrent unit model GRU. CNN can effectively capture spatial dependencies and extract carbon emission characteristics in different regions, while GRU is good at processing time series data and capturing the changing trend of carbon emissions over time. Combining the two can simultaneously model spatiotemporal dependencies, thereby more accurately predicting carbon emissions. Both can automatically extract complex nonlinear features, reduce the need for manual feature engineering, and improve prediction accuracy and training efficiency.

[0090] Convolutional neural network CNN includes input layer, convolution layer, activation function and pooling layer, wherein the input layer is used to receive the key spatiotemporal features obtained in step 3; the convolution layer is the core of CNN, and is composed of multiple convolution kernels (also called filters), which slide on the high-dimensional spatiotemporal feature matrix and calculate the dot product to extract local spatial features; the activation function is usually used after the convolution layer to increase the nonlinearity of the network; the pooling layer is located after the convolution layer, which will reduce the dimension of the feature map to reduce the amount of calculation and avoid overfitting. The pooling operation generally includes maximum pooling and average pooling. The pooling layer in this embodiment uses the maximum pooling operation to downsample the local spatial features obtained by the convolution layer. In this embodiment, the key spatiotemporal features generate spatial feature vectors after multi-layer convolution and pooling operations of CNN; the spatial feature vectors are flattened to obtain the corresponding one-dimensional vectors, which are input into the gated recurrent unit model.

[0091] The gated recurrent unit GRU is a neural network structure for processing sequence data, including an update gate and a reset gate. The update gate is used to determine how much information of the previous hidden state needs to be forgotten, and the update gate can determine how much information of the current hidden state needs to be retained, as well as the weight ratio of the hidden state at the previous moment and the new candidate hidden state at the current moment. In this embodiment, the GRU processes the spatial features of the entire time series through recursive calculation; controls the recursive processing of the flattened spatial feature vector to forget past information through the reset gate; controls the introduction of current information through the update gate to generate the final hidden state; and inputs the final hidden state into the fully connected layer to obtain the predicted carbon emissions.

[0092] As other implementations, the CNN of the present invention may also sample other network models, such as Transformer, etc., and GRU may also adopt models such as LSTM, etc.

[0093] 5. Input key spatiotemporal features into the constructed prediction model to predict carbon emissions.

[0094] The key spatiotemporal feature K is input into the constructed prediction model, and the constructed model can be used to predict carbon emissions. The process is as follows: CNN uses convolution kernels to extract the local spatial features of the key spatiotemporal feature K, and flattens the convolution and pooling features into a one-dimensional vector through maximum pooling downsampling; GRU recursively processes the flattened key spatiotemporal feature K after convolution and pooling through reset gates and update gates, and inputs the final hidden state into the fully connected layer; the fully connected layer predicts carbon emissions through a linear regression function, expressed as:

[0095] y=H fc ·h T +b fc

[0096] Among them, y is the predicted carbon emissions, H fc is the weight matrix of the fully connected layer, h T is the final hidden state after processing based on the key spatiotemporal feature K, b fc is the bias term of the fully connected layer.

[0097] It should be noted that the final hidden state after the key spatiotemporal feature K is processed is obtained in the following process:

[0098] The key spatiotemporal feature K first extracts local spatial features through the convolution layer. The convolutional features are downsampled through the pooling layer to further reduce the data dimension. The downsampled features are input into the GRU recurrent neural network, and the temporal dependencies are captured through time series processing. In the processing of the recurrent network, as the time steps progress, a hidden state h containing spatiotemporal information is finally obtained. T , this hidden state will be used in the fully connected layer for the final carbon emissions prediction.

[0099] 6. Generate a carbon emission flow report based on the predicted carbon emissions.

[0100] 1) Predict carbon emissions through outlier detection and data smoothing.

[0101] In order to accurately understand the long-term trend of carbon emissions, the present invention also uses exponentially weighted moving average to reduce noise and short-term fluctuations in the data and highlight long-term trends and cyclical changes.

[0102] 2) Based on the predicted carbon emissions, calculate the carbon emissions growth rate, the expression is:

[0103]

[0104] Among them, ΔZ r is the carbon emission growth rate at time step t, y τ，t is the predicted carbon emissions at time step t, y r，t-1is the predicted carbon emissions at the previous time step t.

[0105] 3) Carbon emission analysis based on key spatiotemporal characteristics is used to evaluate carbon emissions in different regions. The expression is:

[0106]

[0107] Among them, Q g，q is the carbon emissions of region q, A (t，s) is the spatiotemporal attention score of the high-dimensional spatiotemporal feature at time step t and spatial position s, with T being the total number of time steps.

[0108] It should be noted that the spatiotemporal features of the extracted high-dimensional spatiotemporal features at each time step and spatial position are calculated through the self-attention mechanism. The self-attention mechanism uses the similarity between these features to generate weights. After normalization (such as Softmax), these weights become the spatiotemporal attention score A (t，s) .

[0109] 4) Generate a carbon emission flow report based on the processed predicted carbon emissions, carbon emission growth rate and regional carbon emissions.

[0110] On this basis, the carbon emission flow monitoring method of the present invention is implemented through software modules, such as Figure 2 As shown, it includes an acquisition and preprocessing module, an adaptive data fusion module, a spatiotemporal feature extraction module, a model building module, a model prediction module and a carbon emission flow report generation module.

[0111] The acquisition and preprocessing module is used to collect environmental monitoring data and preprocess the environmental detection data; the adaptive data fusion module is used to fuse the preprocessed environmental detection data based on the adaptive fusion method to obtain high-dimensional spatiotemporal features; the spatiotemporal feature extraction module is used to adjust the high-dimensional spatiotemporal features through the spatiotemporal attention mechanism to obtain key spatiotemporal features; the model construction module is used to build a spatiotemporal carbon emission prediction model based on the convolutional neural network model and the gated recurrent unit model; the model prediction module is used to input the key spatiotemporal features into the spatiotemporal carbon emission prediction model to obtain the carbon emission prediction results; the carbon emission flow report generation module is used to generate a carbon emission flow report based on the carbon emission prediction results.

[0112] Embodiments of computer equipment

[0113] This embodiment also provides a computer device, which is suitable for the carbon emission flow monitoring method of the present invention, including a processor, and the processor is used to execute computer executable instructions to implement the carbon emission flow monitoring method proposed in the above embodiment. The specific implementation method has been described in detail in the method embodiment and will not be repeated here.

[0114] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0115] Embodiments of computer storage media

[0116] The present invention also provides a computer storage medium, on which a computer program is stored, and when the program is executed by a processor, the method for monitoring carbon emission flow proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk.

[0117] In summary, by introducing an adaptive fusion method and a spatiotemporal attention mechanism, the present invention can effectively fuse multi-source environmental detection data, overcome the problem of unstable data quality, and accurately extract key spatiotemporal features; at the same time, by combining convolutional neural networks and gated recurrent unit models, an efficient spatiotemporal carbon emission prediction model is successfully constructed, thereby improving the accuracy and real-time performance of carbon emission flow monitoring, and ultimately achieving accurate prediction and report generation of carbon emission flows.

Claims

1. A method for monitoring carbon emission flow, characterized in that: The monitoring method includes: 1) Obtain multi-source environmental monitoring data related to carbon emission flows and pre-process them so that the representation, unit and format of environmental monitoring data from each data source are consistent; 2) Detect the outliers and time series continuity of the preprocessed environmental detection data, evaluate the data quality based on the detection results to determine the weights corresponding to the environmental detection data of each data source, and use the obtained weights to perform weighted summation on the environmental detection data of each data source to obtain the fusion feature; 3) Inputting the obtained fusion features into the constructed carbon emission prediction model for prediction, obtaining the carbon emission prediction value, and monitoring the carbon emission flow according to the carbon emission prediction value; the carbon emission prediction model is constructed by a deep learning model.

2. The carbon emission flow monitoring method according to claim 1, characterized in that: In the step 2), the outliers of the environmental monitoring data are detected by the 3σ principle. The time series continuity of the environmental monitoring data is determined by the time difference between adjacent time sampling points of the environmental monitoring data. The quality of the preprocessed environmental monitoring data is evaluated by the proportion of outliers in the total data points and the proportion of discontinuous points in the total time points. The higher the data quality, the greater the corresponding weight, and the lower the data quality, the smaller the corresponding weight.

3. The carbon emission flow monitoring method according to claim 2, characterized in that: The calculation formula for the weights of the environmental monitoring data from each data source is: where w i is the weight corresponding to the environmental detection data of the i-th data source, c i is the quality score of the environmental monitoring data of the ith data source, and e is a natural constant.

4. The carbon emission flow monitoring method according to claim 1, characterized in that: The fusion features input into the constructed carbon emission prediction model are the key spatiotemporal features obtained after adjustment using the spatiotemporal attention mechanism. The fusion features before adjustment are called high-dimensional spatiotemporal features. The process of determining the key spatiotemporal features is as follows: The fusion features obtained in step 2) are used as high-dimensional spatiotemporal features to extract and arrange feature data at each time step and spatial position to obtain a high-dimensional spatiotemporal feature matrix; Use high-dimensional spatiotemporal feature matrix to calculate temporal attention weight and spatial attention weight; The key spatiotemporal features are calculated based on the obtained temporal attention weights and spatial attention weights. The calculation formula used for the key spatiotemporal features is: c t，m =a t ·b m Where K is the key spatiotemporal feature, M is the number of spatial locations, T is the number of time steps, and U is the t，m is the feature at time step t and spatial position m, β m is the attention weight of spatial position m, α t is the attention weight at time step t.

5. The carbon emission flow monitoring method according to claim 4, characterized in that: The deep learning model used in the carbon emission prediction model is CNN-GRU, and the prediction process includes: The convolution kernel in CNN is used to extract the local spatial features of the key spatiotemporal features of the input, and the pooling operation is performed through the pooling layer in CNN to achieve feature downsampling, and the features after convolution and pooling operations are flattened into a one-dimensional vector; Recursively process the one-dimensional vector using the reset gate and update gate in the GRU to obtain the final hidden state, and input it into the fully connected layer; The fully connected layer predicts carbon emissions through a linear regression function, and the prediction results are: y=H fc ·h T +b fc Among them, y is the predicted carbon emissions, H fc is the weight matrix of the fully connected layer, h T is the final hidden state, b fc is the bias term of the fully connected layer.

6. The carbon emission flow monitoring method according to claim 4, characterized in that: The monitoring of the carbon emission flow according to the carbon emission prediction value in step 3) includes evaluating the carbon emission of different regions using the carbon emission prediction value, and the calculation formula used is: Where Q g,q is the carbon emission of region q, A(t,s) is the spatiotemporal attention score of the high-dimensional spatiotemporal feature at time step t and spatial position s, T is the total number of time steps, and y r,t is the predicted value of carbon emissions at time step t.

7. The carbon emission flow monitoring method according to any one of claims 1 to 6, characterized in that: The preprocessing in step 1) includes repairing or filling in missing or invalid data of the environmental detection data of each data source, and calibrating and synchronizing the timestamps of the environmental detection data of each data source.

8. The carbon emission flow monitoring method according to any one of claims 1 to 6, characterized in that: The multi-source environmental detection data related to carbon emission flow includes on-site sensor data, satellite remote sensing data and meteorological data.

9. A computer device comprising a processor, characterized in that: The processor is used to execute a computer program to implement the steps of the carbon emission flow monitoring method according to any one of claims 1 to 8.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the carbon emission flow monitoring method according to any one of claims 1 to 8 are implemented.

Images