Road carbon emission prediction method based on Informer model

Abstract

The invention discloses a road carbon emission prediction method based on an Informer model, and belongs to the technical field of environmental monitoring, and the method comprises the steps: constructing an Informer model which comprises a dynamic attention mechanism unit, a static spatial dependence perception unit and a multi-source information fusion gate unit; capturing a dynamic spatial dependency matrix of the road data through a dynamic attention mechanism unit; capturing a static spatial dependency matrix of the road data through a static spatial dependency sensing unit; fusing the dynamic spatial dependency matrix and the static spatial dependency matrix through a multi-source information fusion gate unit to obtain fusion spatial features of the road data; determining a road carbon emission prediction result of the road data based on the fused spatial features; the dynamic spatial features and the static spatial features are fused to obtain the fused spatial features of the road data, and the dynamic spatial features of the road data are captured.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and in particular to a method for predicting road carbon emissions based on the Informer model. Background Art

[0003] However, traditional carbon emission prediction models have many defects, perform poorly in capturing the spatio-temporal dependence of data, resulting in unsatisfactory prediction accuracy. Especially when dealing with multi-source data, the dynamic spatial changes of traffic conditions cannot be effectively considered, leading to unstable prediction performance.

[0004] Therefore, in the process of predicting road carbon emissions in the prior art, there is a problem that it is difficult to effectively capture dynamic spatial features. Summary of the Invention

[0005] In view of this, it is necessary to provide a method for predicting road carbon emissions based on the Informer model to solve the problem that it is difficult to effectively capture dynamic spatial features in the process of predicting road carbon emissions in the prior art.

[0006] To solve the above problems, the present invention provides a method for predicting road carbon emissions based on the Informer model, including: Construct an Informer model, which includes a dynamic attention mechanism unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit; Capture the dynamic spatial dependence matrix of road data through the dynamic attention mechanism unit; Capture the static spatial dependence matrix of road data through the static spatial dependence perception unit; Fuse the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial features of road data; Determine the road carbon emission prediction result of road data based on the fused spatial features.

[0007] In a possible implementation manner, the dynamic attention mechanism unit includes a three-layer feedforward neural network, a query subspace, a key subspace, and a value subspace; capturing the dynamic spatial dependence matrix of road data through the dynamic attention mechanism unit includes: Normalize the road data according to the query subspace and the key subspace to obtain the dynamic spatial dependence matrix of the road data; Multiply the data in the value subspace by the dynamic spatial dependence matrix to obtain the node feature matrix of the road data; Activate and update the node feature matrix according to the three-layer feedforward neural network to obtain the final spatial dependence feature of the road data; Determine the dynamic spatial dependence matrix based on the node feature matrix and the final spatial dependence feature.

[0008] In a possible implementation, the static spatial dependence perception unit includes a normalization module and a static dependence module; capturing the static spatial dependence matrix of road data through the static spatial dependence perception unit includes: Normalize the road data through the normalization module to obtain the normalized Laplacian matrix of the road data; Perform dependence calculation on the normalized Laplacian matrix and the road data through the static dependence module to obtain the static spatial dependence matrix.

[0009] In a possible implementation, the multi-source information fusion gate unit includes an activation function module; fusing the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial feature of the road data, including: Determine the dynamic weight of the dynamic spatial dependence matrix and the static weight of the static spatial dependence matrix according to the activation function module respectively; Perform weighted calculation on the dynamic spatial dependence matrix and the static spatial dependence matrix according to the dynamic weight and the static weight to obtain the fused spatial feature of the road data.

[0010] In a possible implementation, the Informer model further includes a cross-time unit selection module; the cross-time unit selection module includes at least two transformation sub-modules based on multi-layer perceptrons, and the transformation sub-modules based on multi-layer perceptrons are respectively arranged at the start position and the end position of the cross-time unit selection module; Select the cross-time feature of the road data through the transformation sub-module based on the multi-layer perceptron.

[0011] In a possible implementation, the Informer model further includes a feature selection module, the feature selection module includes a gated residual network and a normalization layer; the gated residual network includes a gating layer; Obtain the road feature weight of the road data according to the gated residual network and the normalization layer; The gating layer adjusts the input of the gated residual network based on the gating mechanism to obtain the gated residual gated feature; Assign weights to the gated residual gated feature to obtain the updated gated residual feature of the gating layer; Determine the feature selection result of the feature selection module according to the updated gated residual feature and the road feature weight.

[0012] In one possible implementation, the Informer model includes a distillation mechanism encoding layer, the distillation mechanism encoding layer includes a multi-head attention mechanism module and an attention distillation mechanism module, and the attention distillation mechanism module is arranged between the multi-head attention mechanism modules; after fusing the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial feature of the road data, it further includes: Transmit the weights of the multi-head attention mechanism module to the last layer through the attention distillation mechanism module to obtain the road feature of the road data; Among them, the multi-head attention mechanism module includes a dynamic attention mechanism module unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit.

[0013] In one possible implementation, the Informer model includes an input layer, an embedding layer, a distillation mechanism encoding layer, a decoding layer, and an output layer; The decoding layer includes a masked attention mechanism module, and the masked attention mechanism module includes a masking mechanism sub-module.

[0014] In one possible implementation, the embedding layer includes a time encoding, and the time encoding is embedded at the position encoding of the embedding layer.

[0015] To solve the above problems, the present invention also provides a road carbon emission prediction device based on the Informer model, including: An Informer model construction module for constructing an Informer model, and the Informer model includes a dynamic attention mechanism unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit; A dynamic spatial dependence matrix acquisition module for capturing the dynamic spatial dependence matrix of road data through the dynamic attention mechanism unit; A static spatial dependence matrix acquisition module for capturing the static spatial dependence matrix of road data through the static spatial dependence perception unit; A fused spatial feature acquisition module for fusing the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial feature of the road data; A road carbon emission prediction module for determining the road carbon emission prediction result of road data based on the fused spatial feature.

[0016] The beneficial effects of adopting the above embodiments are as follows: The present invention provides a method for predicting road carbon emissions based on the Informer model. By means of a dynamic attention mechanism unit, a dynamic spatial dependence matrix of road data is extracted, realizing the extraction of dynamic spatial features of road data; by means of a static spatial dependence perception unit, a static spatial dependence matrix of road data is extracted, realizing the extraction of static spatial features of road data; through a multi-source information fusion gate unit, the dynamic spatial features and the static spatial features are fused to obtain the fused spatial features of road data, realizing the fusion of dynamic and static correlations to capture the spatial correlations of road data changing over time, and effectively realizing the capture of dynamic spatial features of road data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flowchart of an embodiment of the method for predicting road carbon emissions based on the Informer model provided by the present invention; Figure 2 It is a schematic flowchart of an embodiment of capturing a dynamic spatial dependence matrix provided by the present invention; Figure 3 It is a schematic structural diagram of an embodiment of a dynamic capture self-attention mechanism provided by the present invention; Figure 4 It is a schematic flowchart of an embodiment of capturing a static spatial dependence matrix provided by the present invention; Figure 5 It is a schematic structural diagram of an embodiment of cross-feature and cross-time unit selection comparison effect provided by the present invention; Figure 6 It is a comparison diagram of an embodiment of the relationship model and the actual situation of traffic flow and speed provided by the present invention; Figure 7 It is a schematic structural diagram of an embodiment of a feature selection module provided by the present invention; Figure 8 It is a schematic flowchart of an embodiment of determining the Informer model provided by the present invention; Figure 9 It is a schematic framework structural diagram of an embodiment of an improved Informer encoder provided by the present invention; Figure 10 It is a schematic overall structural diagram of an embodiment of the Informer model provided by the present invention; Figure 11 It is a schematic block diagram of an embodiment of a device for predicting road carbon emissions based on the Informer model provided by the present invention; Figure 12 It is a schematic block diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The preferred embodiments of the present invention will be specifically described below with reference to the accompanying drawings. The accompanying drawings form a part of the present invention and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0019] Before presenting the embodiments, the Informer model is paraphrased as follows: The Informer model specifically refers to a model used in the field of deep learning, especially in time series prediction. Its name "Informer" is closely related to the function of the model, that is, "providing information" or "informing the future".

[0020] To solve the problem that it is difficult to effectively capture dynamic spatial features in the process of predicting road carbon emissions in the prior art, the present invention provides a road carbon emission prediction method, device, electronic device, and computer-readable storage medium based on the Informer model, which will be described in detail below.

[0021] Before introducing the embodiments, it should be noted that: The road carbon emission prediction method based on the Informer model in the embodiments of the present invention can be implemented in devices such as desktop computers, notebooks, tablets, and laptop computers. Any of the above devices stores a program compiled by the road carbon emission prediction method based on the Informer model. When any of the above devices is started, the program is called, and then the road carbon emission prediction method based on the Informer model is implemented.

[0022] As Figure 1 shown, Figure 1 is a schematic flowchart of an embodiment of the road carbon emission prediction method based on the Informer model provided by the present invention. The road carbon emission prediction method based on the Informer model includes: S101: Construct an Informer model, where the Informer model includes a dynamic attention mechanism unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit; S102: Capture the dynamic spatial dependence matrix of road data through the dynamic attention mechanism unit; S103: Capture the static spatial dependence matrix of road data through the static spatial dependence perception unit; S104: Fuse the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial features of road data; S105: Determine the road carbon emission prediction result of road data based on the fused spatial features.

[0023] In this embodiment, the dynamic spatial dependence matrix of road data is extracted through the dynamic attention mechanism unit, realizing the extraction of the dynamic spatial features of road data; the static spatial dependence matrix of road data is extracted through the static spatial dependence perception unit, realizing the extraction of the static spatial features of road data; the dynamic spatial features and the static spatial features are fused through the multi-source information fusion gate unit to obtain the fused spatial features of road data, realizing the fusion of dynamic and static correlations to capture the spatial correlations of road data over time, and effectively realizing the capture of the dynamic spatial features of road data.

[0024] It should be noted that the hardware entities for constructing the Informer model provided in the embodiments of this application include high-performance computing servers, multi-core CPUs, high-performance GPUs, large-capacity memories, high-speed storage devices, stable power supplies and network connections, and effective heat dissipation systems, etc. The embodiments of this application do not impose any restrictions on the specific type of the Informer model.

[0025] In addition, the road carbon emission prediction method based on the Informer model provided in the embodiments of this application can be applied to a road carbon emission prediction system based on the Informer model. The road carbon emission prediction system based on the Informer model can be a software system running on a terminal device, such as: model generation software, specifically Autodesk Revit, Graphisoft ArchiCAD, Bentley OpenRoads, SketchUp, Rhino, Vectorworks or AutoCAD; the terminal device can be a server, a tablet computer, an Augmented Reality (AR) / Virtual Reality (VR) device, a laptop computer, an Ultra-Mobile Personal Computer (UMPC), a netbook, a Personal Digital Assistant (PDA), a mobile phone and other terminal devices. The embodiments of this application do not impose any restrictions on the specific type of the terminal device.

[0026] As a preferred embodiment, in S102, the dynamic attention mechanism unit includes a three-layer feedforward neural network, a query subspace, a key subspace and a value subspace; in order to capture the dynamic spatial dependence matrix of road data through the dynamic attention mechanism unit, as Figure 2 shown, Figure 2 is a schematic flowchart of an embodiment for capturing the dynamic spatial dependence matrix provided by the present invention, including: S201: Normalize the road data according to the query subspace and the key subspace to obtain the dynamic spatial dependence matrix of the road data; S202: Multiply the data in the value subspace by the dynamic spatial dependency matrix to obtain the node feature matrix of the road data; S203: Activate and update the node feature matrix according to a three-layer feedforward neural network to obtain the final spatial dependency feature of the road data; S204: Determine the dynamic spatial dependency matrix based on the node feature matrix and the final spatial dependency feature.

[0027] To intuitively describe the process of capturing the dynamic spatial dependency matrix, as Figure 3 shown, Figure 3 is a schematic structural diagram of an embodiment of the dynamic capture self-attention mechanism provided by the present invention. Through the query matrix Q i , the key matrix K i and the value matrix V i ( i =1，...，n ) perform data fusion on the multi-source data to achieve the acquisition of the dynamic spatial features of the road data.

[0028] In a specific embodiment, the dynamic spatial dependency matrix is introduced, and its calculation method is:

[0029] wherein, refers to the normalization process, is a sparse matrix with the same dimension as the query matrix Q i , is the inverse matrix of the key matrix K i , d is the number of input feature dimensions.

[0030] Through and the value matrix the node feature matrix is obtained, and the calculation formula is:

[0031] After that, is further updated through a three-layer feedforward neural network to obtain the final spatial dependency , and the calculation formula is:

[0032] wherein, is the residual connection of the multi-source data ; , and are the weight matrices of a three - layer feed - forward neural network.

[0033] Finally, add and to obtain the dynamic spatial dependence matrix .

[0034] In this embodiment, by fusing the dynamic features of road data to adapt to the dynamic spatial correlation, when dealing with a complex road network system with a non - linear topological structure and traffic states, the accuracy of data feature acquisition can be guaranteed.

[0035] Furthermore, in S103, the static spatial dependence perception unit includes a normalization module and a static dependence module; in order to capture the static spatial dependence matrix of road data through the static spatial dependence perception unit, as Figure 4 shown, Figure 4 is a schematic flowchart of an embodiment for capturing the static spatial dependence matrix provided by the present invention, including: S401: Normalize the road data through the normalization module to obtain the normalized Laplacian matrix of the road data; S402: Perform dependence calculation on the normalized Laplacian matrix and the road data through the static dependence module to obtain the static spatial dependence matrix.

[0036] In a specific embodiment, for the constructed road network model , in order to obtain the static spatial dependence of the topological model, first, obtain the normalized Laplacian matrix of the model for learning the static topological adjacency information, and its calculation formula is:

[0037] In the formula, is the identity matrix; is the degree matrix of the distance matrix ; is with the reciprocal of the square root of each element (only non - zero elements on the diagonal). Combining with the data , the final static spatial dependence matrix is obtained, and its calculation formula is:

[0038] In the formula, represents 's th column; is the number of input feature dimensions; is the weight learned through the model; is the Chebyshev polynomial; is the order of the polynomial.

[0039] In this embodiment, by converting the static topological adjacency information of road data, the static spatial dependence characteristics of road data are captured, which can effectively prevent feature loss.

[0040] Furthermore, in S104, the multi-source information fusion gate unit includes an activation function module; in order to fuse the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial features of road data, specifically, first, the dynamic weight of the dynamic spatial dependence matrix and the static weight of the static spatial dependence matrix are respectively determined according to the activation function module; then, according to the dynamic weight and the static weight, weighted calculations are performed on the dynamic spatial dependence matrix and the static spatial dependence matrix to obtain the fused spatial features of road data.

[0041] In a specific embodiment, the dynamic weight has the following calculation formula:

[0042] The static weight has the following calculation formula:

[0043] The fused spatial feature is:

[0044] In summary, the present application forms an improved attention mechanism through the dynamic attention mechanism unit, the static spatial dependence perception unit, and the multi-source information fusion gate unit to better learn spatial correlation and improve the model accuracy.

[0045] Furthermore, in order to improve the prediction accuracy of the Informer model, it is also possible to start from the perspective of learning the temporal correlation and long-term trend of multi-source data of topological-level urban roads, and improve the model generalization ability by extracting cross-time features.

[0046] Specifically, the Informer model further includes a cross-time unit selection module, as Figure 5 shown, Figure 5 is a structural schematic diagram of an embodiment of the cross-feature and cross-time unit selection comparison effect provided by the present invention. The cross-time unit selection module includes at least two transformation sub-modules based on a multi-layer perceptron, and the transformation sub-modules based on the multi-layer perceptron are respectively arranged at the start position and the end position of the cross-time unit selection module; The cross-time features of road data are selected through the transformation sub-module based on the multi-layer perceptron.

[0047] It should be noted that, in order to achieve cross-time feature selection, this application adds a transformation module based on a multi-layer perceptron at the start and end positions of the model. Taking the historical data of a single road node as an example, its prediction target is , and the brief training process of the entire model is as follows:

[0048] In the formula, represents the cross-time training unit of the -dimensional feature in the dimension, and the subscript represents the batch index; represents the layer where this encoder is located.

[0049] Due to the change in the input feature dimension, the multi-head self-attention mechanism becomes capable of capturing the temporal correlations within the sequence. And the multi-layer perceptron neurons in the feed-forward neural network can further mine data features, such as oscillation amplitude, time periodicity, etc., through the improved training unit. There are obvious correlations among some characteristics of multi-source topological traffic data.

[0050] In this embodiment, by mining the temporal correlation features of road data, the accuracy of the predicted result of road carbon emissions can be better guaranteed.

[0051] Specifically, taking traffic flow and average speed as an example, the relationship between the two can be described by the Greenshields model. The model schematic diagram and real data are as Figure 6 shown, Figure 6 which is a comparison diagram of the relationship model between traffic flow and speed provided by the present invention and an actual example.

[0052] From Figure 6 , it can be seen that when the speed is low, the traffic flow and speed are positively correlated, but when the speed reaches a certain value, the traffic flow and speed are negatively correlated. This is because at lower speeds, the distance between vehicles in front and behind is smaller, allowing more vehicles to occupy the road, and the traffic flow increases. However, at higher speeds, the increased distance between vehicles results in fewer vehicles that can be accommodated on the road, and the traffic flow decreases accordingly. In addition, according to the carbon emission measurement process, the magnitude of carbon emissions is highly correlated with some data such as traffic flow and average speed in multi-dimensional data, making the feature importance different.

[0053] Based on the above analysis, it can be known that there is a coupling relationship among multi-dimensional data and there are differences in the contribution degrees to the result. Therefore, in order to mine the coupling and importance of feature variables, the Informer model also includes a feature selection module, as Figure 7 shown, Figure 7Schematic diagram of a structural embodiment of the feature selection module provided by the present invention. The feature selection module includes a gated residual network (GRN) and a normalization layer softmax. The gated residual network GRN includes a gated layer GLU. Further, in order to select the processing features of the Informer model according to the feature selection module, as Figure 8 shown Figure 8 Schematic diagram of a process embodiment of the determination of the Informer model provided by the present invention, including: S801: Obtain the road feature weights of the road data according to the gated residual network and the normalization layer; S802: The gated linear unit adjusts the input of the gated residual network based on the gating mechanism to obtain the gated residual gated feature; S803: Weight the gated residual gated feature to obtain the updated gated residual feature of the gated linear unit; S804: Determine the feature selection result of the feature selection module according to the updated gated residual feature and the road feature weight.

[0054] In a specific embodiment, the road features of the input road data first pass through module and operation to obtain the weights of each feature , and then weight the transformed features to complete feature selection. The specific calculation process is as follows:

[0055]

[0056] In the formula, is the output of the feature selection module at time

[0057] The calculation method of

[0058] In the formula, represents the gated linear unit; and represent the learnable weights and biases. Based on the gating mechanism, the model input is adjusted to improve the flexibility of the model. The calculation method is:

[0059] In the formula, is the input of the gated layer; and represent the learnable weights and biases of the model; denotes the Hadamard product.

[0060] In this embodiment, by specifically screening the processing features of the Informer model, the accuracy of the road carbon emission prediction results can be better guaranteed.

[0061] As a preferred embodiment, in order to accurately describe the relationship between each module in the Informer model, as Figure 9 shown, Figure 9 FIG. is a schematic structural diagram of an embodiment of an improved Informer encoder provided by the present invention. The improved Informer encoder includes a cross-time training unit selection module, a single-feature improved Informer encoder layer, and a multi-source feature selection network. By performing cross-time feature extraction on the input features of road data, the temporal viscosity between features is enhanced. By fusing dynamic and static features, the reliability of the fused features is improved. By performing multi-source feature selection, reliable features are obtained specifically, realizing the optimization processing of road data features from three perspectives to ensure the accuracy of road carbon emission prediction results.

[0062] Furthermore, as Figure 9 shown, the Informer model includes a distillation mechanism encoding layer. Among them, the distillation mechanism encoding layer includes a multi-head attention mechanism module and an attention distillation mechanism module, and the attention distillation mechanism module is arranged between the multi-head attention mechanism modules; after fusing the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial features of road data, the following further includes: transferring the weights of the multi-head attention mechanism module to the last layer through the attention distillation mechanism module to obtain the road features of road data; Among them, the multi-head attention mechanism module includes a dynamic attention mechanism module unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit.

[0063] In a specific embodiment, the model complexity is further reduced by adding an attention distillation mechanism between the multi-head attention blocks to increase efficiency, and the implementation formula is:

[0064] In the formula, is the output of the current ProbSparse self-attention mechanism layer; denotes the max pooling operation; denotes the exponential linear unit activation function; denotes the one-dimensional convolution; is the output of the previous ProbSparse self-attention mechanism layer.

[0065] As a preferred embodiment, in order to completely describe the structure of the Informer model, as Figure 10 shown, Figure 10 FIG. 4 is a schematic diagram of the overall structure of an embodiment of the Informer model provided by the present invention. The Informer model 1000 includes an input layer, an embedding layer, a distillation mechanism encoding layer, a decoding layer, and an output layer; The decoding layer includes a masked attention mechanism module, and the masked attention mechanism module includes a masking mechanism sub-module.

[0066] In this embodiment, the decoding layer of the Informer model introduces a masking mechanism in the ProbSparse attention mechanism to avoid the autoregressive phenomenon by masking future traffic feature data. In addition, the Informer model adopts a single-step decoding structure and can obtain all prediction results at one time.

[0067] Furthermore, in order to better explore the temporal correlation of data, the embedding layer includes temporal encoding, and the temporal encoding is embedded at the position encoding of the embedding layer.

[0068] Specifically, in order to capture the global temporal information existing in the feature data, the Informer model additionally introduces temporal encoding at the position encoding. By embedding timestamp information such as week, month, and year, the model can better explore the temporal correlation of data.

[0069] Through the above method, the dynamic spatial dependence matrix of road data is extracted through the dynamic attention mechanism unit, realizing the extraction of the dynamic spatial features of road data; the static spatial dependence matrix of road data is extracted through the static spatial dependence perception unit, realizing the extraction of the static spatial features of road data; the dynamic spatial features and static spatial features are fused through the multi-source information fusion gate unit to obtain the fused spatial features of road data, realizing the fusion of dynamic and static correlations to capture the spatial correlation of road data changing over time, and effectively realizing the capture of the dynamic spatial features of road data.

[0070] To solve the above problems, the present invention also provides a road carbon emission prediction device based on the Informer model, as Figure 11 shown, Figure 11 FIG. 5 is a structural block diagram of an embodiment of the road carbon emission prediction device based on the Informer model provided by the present invention. The road carbon emission prediction device 1100 based on the Informer model includes: An Informer model construction module 1101, configured to construct an Informer model, where the Informer model includes a dynamic attention mechanism unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit; The dynamic spatial dependence matrix acquisition module 1102 is configured to capture the dynamic spatial dependence matrix of road data through a dynamic attention mechanism unit; The static spatial dependence matrix acquisition module 1103 is configured to capture the static spatial dependence matrix of road data through a static spatial dependence perception unit; The fused spatial feature acquisition module 1104 is configured to fuse the dynamic spatial dependence matrix and the static spatial dependence matrix through a multi-source information fusion gate unit to obtain the fused spatial features of road data; The road carbon emission prediction module 1105 is configured to determine the road carbon emission prediction result of road data based on the fused spatial features.

[0071] As Figure 12 shown, the present invention also correspondingly provides an electronic device 1200, which includes a processor 1201, a memory 1202, and a display 1203. Figure 12 Only some components of the electronic device 1200 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0072] In some embodiments, the processor 1201 may be a central processing unit (CPU), a microprocessor, or other data processing chips, and is configured to run the program code stored in the memory 1202 or process data, such as the road carbon emission prediction method based on the Informer model in the present invention.

[0073] In some embodiments, the processor 1201 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 1201 may be local or remote. In some embodiments, the processor 1201 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination of the above.

[0074] In some embodiments, the memory 1202 may be an internal storage unit of the electronic device 1200, such as the hard disk or memory of the electronic device 1200. In some other embodiments, the memory 1202 may also be an external storage device of the electronic device 1200, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 1200.

[0075] Further, the memory 1202 may include both the internal storage unit of the electronic device 1200 and external storage devices. The memory 1202 is used to store the application software installed in the electronic device 1200 and various types of data.

[0076] In some embodiments, the display 1203 may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. The display 1203 is used to display the information of the electronic device 1200 and to display the visual user interface. The components 1201 - 1203 of the electronic device 1200 communicate with each other through the system device bus.

[0077] In one embodiment, when the processor 1201 executes the road carbon emission prediction program based on the Informer model in the memory 1202, the following steps may be implemented: Construct an Informer model, where the Informer model includes a dynamic attention mechanism unit, a static spatial dependence perception unit, and a multi-source information fusion gate unit; Capture the dynamic spatial dependence matrix of the road data through the dynamic attention mechanism unit; Capture the static spatial dependence matrix of the road data through the static spatial dependence perception unit; Fuse the dynamic spatial dependence matrix and the static spatial dependence matrix through the multi-source information fusion gate unit to obtain the fused spatial features of the road data; Determine the road carbon emission prediction result of the road data based on the fused spatial features.

[0078] It should be understood that when the processor 1201 executes the road carbon emission prediction program based on the Informer model in the memory 1202, in addition to the above functions, other functions may also be implemented. For details, refer to the description of the corresponding method embodiments above.

[0079] Furthermore, the embodiments of the present invention do not specifically limit the type of the mentioned electronic device 1200. The electronic device 1200 can be a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop, or other portable devices. Exemplary embodiments of the portable device include, but are not limited to, portable devices equipped with IOS, android, microsoft, or other operating system devices. The above portable devices can also be other portable devices, such as a laptop with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 1200 may not be a portable device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0080] Correspondingly, the embodiments of the present invention further provide a computer-readable storage medium. The computer-readable storage medium is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the road carbon emission prediction method based on the Informer model provided by the above various method embodiments can be realized.

[0081] Those skilled in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0082] The above has introduced in detail the road carbon emission prediction method based on the Informer model provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A road carbon emission prediction method based on the Informer model, characterized in that: include: Constructing an Informer model, wherein the Informer model includes a dynamic attention mechanism unit, a static spatial dependency perception unit, and a multi-source information fusion gate unit; Capturing a dynamic spatial dependency matrix of road data through the dynamic attention mechanism unit; capturing a static spatial dependency matrix of the road data by the static spatial dependency perception unit; The dynamic spatial dependency matrix and the static spatial dependency matrix are fused by the multi-source information fusion gate unit to obtain the fused spatial features of the road data; A road carbon emission prediction result of the road data is determined based on the fused spatial features.

2. The road carbon emission prediction method based on the Informer model according to claim 1 is characterized in that: The dynamic attention mechanism unit includes a three-layer feedforward neural network, a query subspace, a key subspace, and a value subspace; the dynamic spatial dependency matrix of the road data captured by the dynamic attention mechanism unit includes: Normalizing the road data according to the query subspace and the key subspace to obtain a dynamic spatial dependency matrix of the road data; Multiplying the data of the value subspace with the dynamic space dependency matrix to obtain a node feature matrix of the road data; Activating and updating the node feature matrix according to the three-layer feedforward neural network to obtain the final spatial dependency features of the road data; The dynamic spatial dependency matrix is ​​determined based on the node feature matrix and the final spatial dependency features.

3. The road carbon emission prediction method based on the Informer model according to claim 1 is characterized in that: The static spatial dependency perception unit includes a normalization module and a static dependency module; The capturing of the static spatial dependency matrix of the road data by the static spatial dependency perception unit comprises: The road data is normalized by the normalization module to obtain a normalized Laplace matrix of the road data; The static spatial dependency matrix is ​​obtained by performing dependency calculation on the normalized Laplace matrix and the road data through the static dependency module.

4. The road carbon emission prediction method based on the Informer model according to claim 1 is characterized in that: The multi-source information fusion gate unit includes an activation function module; the multi-source information fusion gate unit fuses the dynamic spatial dependency matrix and the static spatial dependency matrix to obtain the fusion spatial features of the road data, including: Determining the dynamic weights of the dynamic spatial dependency matrix and the static weights of the static spatial dependency matrix respectively according to the activation function module; According to the dynamic weight and the static weight, weighted calculation is performed on the dynamic spatial dependency matrix and the static spatial dependency matrix to obtain the fused spatial features of the road data.

5. The road carbon emission prediction method based on the Informer model according to claim 1 is characterized in that: The Informer model also includes a cross-time unit selection module; the cross-time unit selection module includes at least two conversion submodules based on multi-layer perceptrons, and the conversion submodules based on multi-layer perceptrons are respectively arranged at the start position and the end position of the cross-time unit selection module; The cross-time features of the road data are selected through the conversion submodule based on the multi-layer perceptron.

6. The road carbon emission prediction method based on the Informer model according to claim 1 is characterized in that: The Informer model also includes a feature selection module, which includes a gated residual network and a normalization layer; the gated residual network includes a gated layer; Acquire a road feature weight of the road data according to the gated residual network and the normalization layer; The gating layer adjusts the input of the gated residual network based on a gating mechanism to obtain a gated residual gating feature; Weighting the gated residual gating feature to obtain an updated gated residual feature of the gated layer; The feature selection result of the feature selection module is determined according to the updated gated residual feature and the road feature weight.

7. The road carbon emission prediction method based on the Informer model according to claim 1 is characterized in that: The Informer model includes a distillation mechanism encoding layer, the distillation mechanism encoding layer includes a multi-head attention mechanism module and an attention distillation mechanism module, and the attention distillation mechanism module is arranged between the multi-head attention mechanism modules; after fusing the dynamic spatial dependency matrix and the static spatial dependency matrix through the multi-source information fusion gate unit to obtain the fused spatial features of the road data, it also includes: The weight of the multi-head attention mechanism module is transferred to the last layer through the attention distillation mechanism module to obtain the road features of the road data; Among them, the multi-head attention mechanism module includes the dynamic attention mechanism module unit, the static spatial dependency perception unit and the multi-source information fusion gate unit.

8. The road carbon emission prediction method based on the Informer model according to claim 7 is characterized in that: The Informer model includes an input layer, an embedding layer, the distillation mechanism encoding layer, a decoding layer and an output layer; The decoding layer includes a masked attention mechanism module, and the masked attention mechanism module includes a mask mechanism sub-module.

9. The road carbon emission prediction method based on the Informer model according to claim 8 is characterized in that: The embedding layer includes a time code embedded at a position code of the embedding layer.

10. A road carbon emission prediction device based on the Informer model, characterized in that: include: An informer model construction module, used to construct an informer model, wherein the informer model includes a dynamic attention mechanism unit, a static spatial dependency perception unit, and a multi-source information fusion gate unit; A dynamic spatial dependency matrix acquisition module, used to capture the dynamic spatial dependency matrix of road data through the dynamic attention mechanism unit; A static spatial dependency matrix acquisition module, used for capturing the static spatial dependency matrix of the road data through the static spatial dependency perception unit; A fusion spatial feature acquisition module, used for fusing the dynamic spatial dependency matrix and the static spatial dependency matrix through the multi-source information fusion gate unit to obtain the fusion spatial features of the road data; A road carbon emission prediction module is used to determine a road carbon emission prediction result of the road data based on the fused spatial features.

Images