A method and apparatus for predicting regional heat load based on sparse dynamic graph neural networks

The heat load prediction model constructed by sparse dynamic graph neural network solves the problem of insufficient accuracy in heat load prediction in district heating systems, and achieves more efficient and accurate heat load prediction, with high efficiency, interpretability and robustness.

CN117725962BActive Publication Date: 2026-11-17JIMEI UNIV +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311470447.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-07
Publication Date
2026-11-17
Estimated Expiration
2043-11-07

AI Technical Summary

Technical Problem

Existing regional heat load forecasting methods are not accurate enough when dealing with complex and dynamic heating systems, and cannot meet the accuracy requirements of heating demand.

Method used

A sparse dynamic graph neural network is adopted. By acquiring time-series data of heat load and external meteorological factors in the target area, the heat load prediction model is constructed through training using sparse graph learning module, spatiotemporal memory enhancement module and time fusion module, so as to achieve efficient learning and modeling of data.

Benefits of technology

It improves the accuracy and efficiency of regional heat load forecasting, better captures local patterns and long-term dependencies in time-series data, provides interpretable forecast results, and is robust and scalable.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117725962B_ABST
    Figure CN117725962B_ABST
Patent Text Reader

Abstract

This application provides a method and apparatus for regional heat load prediction based on a sparse dynamic graph neural network, relating to the field of smart heating technology. The method includes: acquiring time-series heat load data and external meteorological factor data of a target area over a historical period; preprocessing the heat load time-series data and external meteorological factor time-series data, and then inputting them into a pre-trained heat load prediction model to obtain predicted heat load data for the target area over a future set time period; wherein the heat load prediction model is trained on a sparse dynamic graph neural network based on historical heat load time-series data samples and historical external meteorological factor time-series data samples from multiple regions. Therefore, this application, through its heat load prediction model based on a sparse dynamic graph neural network, can efficiently learn and model data, thereby obtaining more accurate regional heat load prediction results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of smart heating technology, specifically to a method and device for predicting regional heat load based on sparse dynamic graph neural networks. Background Technology

[0002] District heating is a centralized energy distribution system that provides heat to residential and commercial buildings from a common heat source. It is widely used in urban areas around the world as a sustainable and efficient solution for meeting heating needs.

[0003] However, the optimized operation and management of district heating systems rely on accurate forecasting of heat load, which refers to the amount of heat required by users within a given timeframe. Due to the complexity and dynamism of district heating networks, heat load forecasting is a challenging task involving numerous factors and uncertainties. Traditional district heat load forecasting methods employ various statistical and machine learning approaches. However, these methods often fall short of accuracy in the face of the complexity and dynamism of district heat demand. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, electronic device, and storage medium for predicting regional heat load based on a sparse dynamic graph neural network, thereby improving the accuracy of regional heat load prediction.

[0005] In a first aspect, embodiments of this application provide a method for predicting regional heat load based on a sparse dynamic graph neural network, including:

[0006] Acquire time-series data of heat load and external meteorological factors for the target area within a historical time period;

[0007] After preprocessing the heat load time series data and the external meteorological factor time series data, the data are input into a pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future within a set time period.

[0008] The heat load prediction model is obtained by training a sparse dynamic graph neural network based on historical heat load time series data samples from multiple regions and historical external meteorological factor time series data samples.

[0009] In some embodiments of this application, the heat load prediction model is pre-trained in the following manner:

[0010] Historical heat load data and external meteorological factor data for multiple regions were acquired and statistically analyzed into time series within the same time interval, resulting in historical heat load time series data samples and historical external meteorological factor time series data samples for multiple regions.

[0011] After performing the preprocessing on the historical heat load time series data samples and the historical external meteorological factor time series data samples, the historical heat load time series data samples and the historical external meteorological factor time series data samples are divided into batches of data of equal length.

[0012] Set an objective function, input the batch data obtained by preprocessing the historical heat load time series data samples and the historical external meteorological factor time series data samples into the sparse dynamic graph neural network for training, and obtain the heat load prediction model after reaching the preset training cutoff condition.

[0013] In some embodiments of this application, the preprocessing involves dividing the time-series data into multiple batches.

[0014] In some embodiments of this application, the sparse dynamic graph neural network includes a sequentially connected sparse graph learning module, a spatiotemporal memory enhancement module, and a temporal fusion module;

[0015] The sparse graph learning module includes: a sparse graph construction component and a temporal graph learning component;

[0016] The spatiotemporal memory enhancement module includes: a spatiotemporal convolution component and a global attention mechanism component;

[0017] The time fusion module uses a skip connection architecture.

[0018] In some embodiments of this application, the objective function is set as the mean squared error loss function, and an L2 regularization term is used to prevent overfitting.

[0019] Secondly, embodiments of this application provide a regional heat load prediction device based on a sparse dynamic graph neural network, comprising:

[0020] The acquisition module is used to acquire time-series data of heat load and external meteorological factors for the target area over a historical period.

[0021] The prediction module is used to preprocess the heat load time series data and the external meteorological factor time series data, and then input them into the pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future within a set time period.

[0022] The heat load prediction model is obtained by training a sparse dynamic graph neural network based on historical heat load time series data samples from multiple regions and historical external meteorological factor time series data samples.

[0023] In some embodiments of this application, the apparatus further includes:

[0024] The training module is used to pre-train the heat load prediction model in the following manner:

[0025] Historical heat load data and external meteorological factor data for multiple regions were acquired and statistically analyzed into time series within the same time interval, resulting in historical heat load time series data samples and historical external meteorological factor time series data samples for multiple regions.

[0026] After performing the preprocessing on the historical heat load time series data samples and the historical external meteorological factor time series data samples, the historical heat load time series data samples and the historical external meteorological factor time series data samples are divided into batches of data of equal length.

[0027] Set an objective function, input the batch data obtained by preprocessing the historical heat load time series data samples and the historical external meteorological factor time series data samples into the sparse dynamic graph neural network for training, and obtain the heat load prediction model after reaching the preset training cutoff condition.

[0028] In some embodiments of this application, the preprocessing involves dividing the time-series data into multiple batches.

[0029] In some embodiments of this application, the sparse dynamic graph neural network includes a sequentially connected sparse graph learning module, a spatiotemporal memory enhancement module, and a temporal fusion module;

[0030] The sparse graph learning module includes: a sparse graph construction component and a temporal graph learning component;

[0031] The spatiotemporal memory enhancement module includes: a spatiotemporal convolution component and a global attention mechanism component;

[0032] The time fusion module uses a skip connection architecture.

[0033] In some embodiments of this application, the objective function is set as the mean squared error loss function, and an L2 regularization term is used to prevent overfitting.

[0034] Thirdly, this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method as described in the first aspect.

[0035] Fourthly, this application provides a computer-readable storage medium having computer-readable instructions stored thereon, which can be executed by a processor to implement the method described in the first aspect.

[0036] Compared to existing technologies, the regional heat load prediction method based on sparse dynamic graph neural networks provided in this application obtains time-series heat load data and external meteorological factor data of the target area over a historical period. After preprocessing the heat load time-series data and external meteorological factor time-series data, the data is input into a pre-trained heat load prediction model to obtain the predicted heat load data of the target area over a future set time period. The heat load prediction model is trained on the sparse dynamic graph neural network based on historical heat load time-series data samples and historical external meteorological factor time-series data samples from multiple regions. Therefore, this application's heat load prediction model based on sparse dynamic graph neural networks can efficiently learn and model data, thereby obtaining more accurate regional heat load prediction results. Attached Figure Description

[0037] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0038] Figure 1 The flowchart of a regional heat load prediction method based on a sparse dynamic graph neural network provided in this application is shown.

[0039] Figure 2 A flowchart of a heat load prediction model training method provided in this application is shown;

[0040] Figure 3 A schematic diagram illustrating the prediction process of a heat load prediction model provided in this application is shown.

[0041] Figure 4 A schematic diagram illustrating the prediction process of another heat load prediction model provided in this application is shown;

[0042] Figure 5 This invention provides a schematic diagram of a regional heat load prediction device based on a sparse dynamic graph neural network.

[0043] Figure 6 A schematic diagram of an electronic device provided in this application is shown. Detailed Implementation

[0044] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0045] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.

[0046] Furthermore, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to those processes, methods, products, or apparatuses.

[0047] Please refer to Figure 1 , Figure 1 This is a flowchart of a regional heat load prediction method based on a sparse dynamic graph neural network provided in an embodiment of this application, including the following S101 to S102:

[0048] S101. Obtain the time series data of heat load and external meteorological factors for the target area within a historical time period.

[0049] The heat load time series data consists of multiple heat load data points for the target area arranged at set time intervals. These time intervals can be set to specific intervals such as hours, days, weeks, or months. The heat load data can be observations from heat meters.

[0050] External meteorological factor time series data consists of multiple meteorological data points for the target area arranged at set time intervals. The time interval settings are the same as those for heat load time series data.

[0051] S102. After preprocessing the heat load time series data and the external meteorological factor time series data, input them into the pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future set time period.

[0052] The heat load prediction model is obtained by training a sparse dynamic graph neural network based on historical heat load time series data samples from multiple regions and historical external meteorological factor time series data samples.

[0053] Optionally, the above-mentioned data preprocessing operation can be to divide the time series data into multiple batches, each batch being a fixed-length subsequence that can be directly passed to the heat load prediction model for training and prediction. Its advantage is that it can capture local patterns and long-term dependencies in the time series data, thereby improving the accuracy of prediction.

[0054] After preprocessing the time series data of heat load and external meteorological factors, the data are input into a pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future within a set time period.

[0055] The following describes how to pre-train a heat load prediction model. Specifically, the heat load prediction model can be pre-trained as follows: Figure 2 As shown, steps S201 to S203 are included:

[0056] S201. Obtain historical heat load data and external meteorological factor data for multiple regions, and statistically analyze them into time series within the same time interval to obtain historical heat load time series data samples and historical external meteorological factor time series data samples for multiple regions.

[0057] Specifically, historical heat load records include heat load data and external meteorological factor data. Statistical analysis of these historical heat load records yields a total of T data points, representing Q regions. The symbol X is used to represent these regions. q =x q,1 ,x q,2 ,…,x q,T Z represents a sample of historical heat load time series data for a region. q =z q,1 ,z q,2 ,…,z q,T This represents a sample of historical external meteorological factors for a region.

[0058] S202. After performing the preprocessing on the historical heat load time series data sample and the historical external meteorological factor time series data sample, the historical heat load time series data sample and the historical external meteorological factor time series data sample are divided into batches of data of equal length.

[0059]

[0060] in, This represents heat load data over a specific time period. The heat load data represents a single time step, where s∈[1,L] is the time step, n∈[1,N] represents the time slot, and N=[T / P] is defined as the upper limit of T÷P, representing the total number of batch data obtained after processing. The set of spliced ​​observations can be represented as... If T is not an exact multiple of P, use the final value X.t Padding is used to ensure even-numbered division. To extract representations from each time slot, the temporal information in the batch data tensor is encoded using a linear transformation. This process can be represented as:

[0061]

[0062] in, W represents the final output of a time-sparse process. enc This represents the learnable weight parameters. Time sparsity effectively reduces the count of input tokens from T to approximately [value missing]. This reduces memory consumption and computational complexity within the temporal graph learning component, and this reduction optimizes training time and memory usage.

[0063] The goal of sparse relationships is to selectively prune the relationships between nodes in the graph, which is accomplished using the k-nearest neighbor method. Specifically, during the graph structure construction process, through input... The dot product is used to calculate node similarity, expressed as:

[0064]

[0065] Where exp(·) represents the exponential function, It is a normalized similarity matrix For each node, select the weight matrix. The top-k maximum values ​​in the graph. k represents the sparsity of edges in the graph, calculated as k = ∈ N. 2 Adjacency matrix The corresponding item is set to 1, and all other items are set to 0. This method simplifies the graph structure and does not rely on predefined relationships. As the input time window X... t-T+1:t As the nodes change, the relationships between them also change, forming a dynamic adjacency matrix. The generated graph changes over time and can be represented as follows: In the picture, Represents a set of nodes. Each node It corresponds to a different time slot. It is a set of edges, where each edge e i,j ∈ε T Time slot v was established i With v j The temporal relationship or dependency between them. This temporal connectivity can also be represented by an adjacency matrix. express.

[0066] During the temporal graph learning phase, the temporal graph learning component applies graph convolution operations to batch data input and dynamic...

[0067] Specifically, preprocessing can be performed using a patching component. The patching component divides the time series into batches of data of equal length and assigns a label to each batch of data according to its pattern, transforming historical heat load time series data samples and historical external meteorological factor time series data samples into supervised data.

[0068] S203. Set the objective function, input the batch data obtained by preprocessing the historical heat load time series data samples and the historical external meteorological factor time series data samples into the sparse dynamic graph neural network for training, and obtain the heat load prediction model after reaching the preset training cutoff condition.

[0069] Specifically, the sparse dynamic graph neural network includes a sequentially connected sparse graph learning module, a spatiotemporal memory enhancement module, and a temporal fusion module.

[0070] The sparse graph learning module includes a sparse graph construction component and a temporal graph learning component; the spatiotemporal memory enhancement module includes a spatiotemporal convolution component and a global attention mechanism component; the temporal fusion module uses a skip connection architecture.

[0071] Please refer to Figure 3 The following section provides a detailed introduction to the heat load prediction model.

[0072] The heat load prediction model constructs a dynamic graph for each batch of data based on the spatiotemporal dependencies of the data and the global context.

[0073] 1) Sparse Graph Learning Module: The sparse graph learning module learns the temporal dependencies between regions by constructing a sparse dynamic graph based on historical data. This is achieved by setting... The graph structure representing a sparse dynamic graph. Among them, For a set of nodes, each node Corresponding to a different time point X t The edge set is denoted as ε, and is determined by e. i,j Each edge is composed of individual edges, ∈ε. Each edge is used at time step v. i and v j Establish temporal or dependency relationships between them. These complex relationships can also be represented using adjacency matrices. To express.

[0074] The sparse graph learning module consists of two stages: the sparse graph construction stage and the temporal graph learning stage.

[0075] During the sparse graph construction phase, the sparse graph construction component extracts heat table observations to condense sequence information and establish meaningful node relationships. The connections between nodes are dynamically and adaptively pruned to reflect time changes, retaining only the most relevant connections.

[0076] This application employs a batch processing strategy to preprocess heat table observations before merging them into the temporal plot learning component. Let P be a constant representing the batch data length. Using this batch processing strategy, each group of observations with consecutive time steps of P is grouped into a single time slot, as follows:

[0077]

[0078] The adjacency matrix. Graph convolution operations use learnable weights, biases, and activation functions to update node features. The graph convolution operation can be represented as:

[0079] in,* G This represents the graph convolution operation. This represents the input data after time sparsity processing. It is a self-loop adjacency matrix. u It is an identity matrix. D represents a diagonal matrix. and D i,j =0, for i≠j, and These are the learnable weights and biases, respectively.

[0080] To aggregate information from a broader set of neighbors and enhance the model's ability to capture complex relationships in the graph, two graph convolutional components are stacked to produce the final representation:

[0081]

[0082] in, This represents the generated representation for the temporal graph learning component. σ(·) denotes the linear rectified function (ReLU). M g (·) represents the dropout mask, and γ is the dropout rate. BN(·) represents the batch normalization layer used to stabilize and accelerate the training process. W g It is a learnable weight matrix. ⊙ represents element-wise multiplication.

[0083] The temporal graph learning component learns the temporal dependencies between regions by updating the node features of regions based on historical data and the dynamic graph structure. The component outputs a set of node representations that capture the spatiotemporal information and global context of the data.

[0084] 2) Spatiotemporal Memory Enhancement Module: The spatiotemporal memory enhancement module includes a spatiotemporal convolution component and a global attention mechanism component, integrating spatiotemporal convolution and global attention mechanisms. While graph neural networks capture the topological relationships within a graph structure, they may not fully explain complex temporal dynamics and subtle spatial variations. In contrast, spatiotemporal convolution excels at modeling spatial and temporal dependencies, which is crucial when dealing with dynamically evolving graph structures and temporally varying node or edge attributes. The integration of spatiotemporal convolution partially alleviates the inherent limitations of traditional graph learning. Furthermore, this combined approach not only enhances the expressive power of the heat load prediction model, enabling the extraction of more complex features, but also provides a regularization form that potentially prevents overfitting. The generated representation is as follows:

[0085]

[0086] Among them, R c This represents the output of the spatiotemporal convolution. W c These are the learnable parameters of the convolution kernel. * represents the regular convolution operation. M c (γ) represents the dropout mask generated based on the dropout rate of γ.

[0087] Spatiotemporal convolution can capture the inherent spatial and temporal dependencies in data, but its ability to focus on specific temporal patterns or key events may be limited. To address this limitation and highlight key time series and features, this application integrates a global attention mechanism to enhance the sensitivity of the heat load prediction model to key spatiotemporal changes:

[0088]

[0089]

[0090]

[0091] Among them, R α,o This indicates the generation of the global attention mechanism. S α The attention score is represented by , and softmax(·) is the softmax function used to normalize these scores. α It is the weight matrix, b α Corresponding to the bias term. M α (γ) represents the dropout mask, and the dropout rate is γ.

[0092] 3) Temporal Fusion Module: Excessive nonlinearity can lead to dynamic instability during training, resulting in oscillating loss and convergence challenges. To address this challenge, the temporal fusion module employs a skip connection architecture. This architecture effectively combines the original input information with the enhanced representation obtained from the spatiotemporal memory enhancement stage to obtain the final prediction result. Prior to this stage, this application first constructs a decoder consisting of dual embedding layers, each designed with a fully connected network (FCN). The embedding structure aims to map the condensed feature representation back to the desired target output shape:

[0093] O = FCN spat (FCN temp (R α,o )),

[0094] in, This indicates the generation of the decoder. FCN spat (·) and FCN temp (·) represents the embedding layer, which decodes the feature matrix in both the variable and time dimensions. To account for external meteorological factors, the decoder output can be expressed as:

[0095] O = FCN spat (FCN temp ([R α,o ||Emb(Z)])),

[0096] Where Z represents the input external meteorological factor data, and [·||·] represents the concatenation operation. Emb(·) indicates that it is used to embed the external input to R. α,o Linear transformation for shape alignment.

[0097] Subsequently, a skip connection architecture is used to integrate information from the linear representation of the original heat table observations and the condensed feature representation generated by the spatiotemporal memory enhancement module:

[0098]

[0099] in, It is the predicted value for the next τ steps. This represents element-wise addition. μ:t W represents the observation value at a past time step of μ. o and b o These are trainable parameters responsible for performing a linear transformation on the input data to align it with the target output shape.

[0100] Specifically, the objective function chosen to optimize the model parameters W can be set as the Mean Squared Error (MSE) loss function, which quantifies the difference between the predicted and actual data and uses an L2 regularization term to prevent overfitting. The combined loss function L(W) is described in the following form:

[0101]

[0102] Where H represents the total number of heat meters, and λ is the regularization parameter. The loss of each heat meter is calculated, summarized, and then averaged to form a comprehensive loss measurement of all heat meter observations.

[0103] Adjust the model parameters based on the model output to obtain the optimal heat load prediction model.

[0104] After obtaining the heat load prediction model, the predicted heat load data for the target area in a future set time period can be predicted by inputting the historical heat load time series data and external meteorological factor time series data of the target area into the heat load prediction model. The heat load prediction process for this area can be found in [reference needed]. Figure 3 As shown, the predicted results can be provided to the heating system to create accurate production plans for precise district heating, or they can be provided to other target users.

[0105] like Figure 4 As shown, it can collect regional historical heat load records consisting of historical heat load data and historical external meteorological factor data in real time, preprocess the data, then train the model and make real-time predictions, send the prediction results to the target user, and then obtain the real values ​​fed back by the target user in real time, thereby optimizing the model.

[0106] To provide real-time and accurate heating forecasts for a region, this application proposes a regional heat load forecasting method based on a sparse dynamic graph neural network. This method collects historical heat load records for the region and combines them with historical external factor records to statistically obtain a data matrix characterized by the regional heat load. This matrix is ​​then transformed into supervised data, which is used to train the neural network. Finally, the trained model is used to provide accurate forecasts.

[0107] The regional heat load prediction method based on sparse dynamic graph neural network provided in this application has the following beneficial effects:

[0108] a) High precision

[0109] This application utilizes a sparse dynamic graph neural network model to efficiently learn and model data, thereby obtaining more accurate prediction results.

[0110] b) High efficiency

[0111] The sparse dynamic graph neural network model can quickly process large amounts of data and has high computational efficiency during inference.

[0112] c) Explainability

[0113] Sparse dynamic graph neural network models can provide interpretations of prediction results, analyze the model's decision-making process and key features, and thus better understand the model's predictions.

[0114] d) Scalability

[0115] Sparse dynamic graph neural network models can enhance their learning ability by increasing the number of layers and nodes, and their performance can also be improved by enhancing the training data.

[0116] e) Robustness

[0117] The sparse dynamic graph neural network model can handle the complex relationships between multiple input features and is also robust to noise and outliers in the input data.

[0118] In the above embodiments, a regional heat load prediction method based on a sparse dynamic graph neural network is provided. Correspondingly, this application also provides a regional heat load prediction device 10 based on a sparse dynamic graph neural network. The regional heat load prediction device based on a sparse dynamic graph neural network provided in this application can implement the above-described regional heat load prediction method based on a sparse dynamic graph neural network. This regional heat load prediction device based on a sparse dynamic graph neural network can be implemented by software, hardware, or a combination of both. For example, the regional heat load prediction device based on a sparse dynamic graph neural network may include integrated or separate functional modules or units to perform the corresponding steps in the above methods. Please refer to... Figure 5 As shown, it includes:

[0119] The acquisition module 101 is used to acquire the time series data of heat load and external meteorological factors of the target area in a historical period.

[0120] Prediction module 102 is used to preprocess the heat load time series data and external meteorological factor time series data, and then input them into a pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future set time period.

[0121] The heat load prediction model is obtained by training a sparse dynamic graph neural network based on historical heat load time series data samples from multiple regions and historical external meteorological factor time series data samples.

[0122] In some embodiments of this application, the apparatus further includes:

[0123] The training module is used to pre-train the heat load prediction model in the following manner:

[0124] Historical heat load data and external meteorological factor data for multiple regions were acquired and statistically analyzed into time series within the same time interval, resulting in historical heat load time series data samples and historical external meteorological factor time series data samples for multiple regions.

[0125] After performing the preprocessing on the historical heat load time series data samples and the historical external meteorological factor time series data samples, the historical heat load time series data samples and the historical external meteorological factor time series data samples are divided into batches of data of equal length.

[0126] Set an objective function, input the batch data obtained by preprocessing the historical heat load time series data samples and the historical external meteorological factor time series data samples into the sparse dynamic graph neural network for training, and obtain the heat load prediction model after reaching the preset training cutoff condition.

[0127] In some embodiments of this application, the preprocessing involves dividing the time-series data into multiple batches.

[0128] In some embodiments of this application, the sparse dynamic graph neural network includes a sequentially connected sparse graph learning module, a spatiotemporal memory enhancement module, and a temporal fusion module;

[0129] The sparse graph learning module includes: a sparse graph construction component and a temporal graph learning component;

[0130] The spatiotemporal memory enhancement module includes: a spatiotemporal convolution component and a global attention mechanism component;

[0131] The time fusion module uses a skip connection architecture.

[0132] In some embodiments of this application, the objective function is set as the mean squared error loss function, and an L2 regularization term is used to prevent overfitting.

[0133] The regional heat load prediction device based on sparse dynamic graph neural network provided in this application embodiment and the regional heat load prediction method based on sparse dynamic graph neural network provided in this application embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.

[0134] This application also provides an electronic device corresponding to the method provided in the foregoing embodiments. The electronic device may be an electronic device for a client, such as a mobile phone, laptop, tablet computer, desktop computer, etc., to execute the above-described regional heat load prediction method based on sparse dynamic graph neural network.

[0135] Please refer to Figure 6 This illustrates a schematic diagram of an electronic device provided by some embodiments of this application. For example... Figure 6 As shown, the electronic device 20 includes: a processor 200, a memory 201, a bus 202, and a communication interface 203. The processor 200, the communication interface 203, and the memory 201 are connected via the bus 202. The memory 201 stores a computer program that can run on the processor 200. When the processor 200 runs the computer program, it executes the phishing email tracing method provided in any of the foregoing embodiments of this application.

[0136] The memory 201 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 203 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0137] Bus 202 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory 201 is used to store programs. After receiving an execution instruction, the processor 200 executes the program. The phishing email tracing method disclosed in any of the foregoing embodiments of this application can be applied to the processor 200, or implemented by the processor 200.

[0138] The processor 200 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 200 or by instructions in software form. The processor 200 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 201. The processor 200 reads the information in memory 201 and, in conjunction with its hardware, completes the steps of the above method.

[0139] The electronic device provided in this application embodiment and the regional heat load prediction method based on sparse dynamic graph neural network provided in this application embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.

[0140] This application also provides a computer-readable storage medium corresponding to the regional heat load prediction method based on sparse dynamic graph neural network provided in the foregoing embodiments, wherein a computer program (i.e., program product) is stored thereon. When the computer program is run by a processor, it executes the regional heat load prediction method based on sparse dynamic graph neural network provided in any of the foregoing embodiments.

[0141] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0142] The computer-readable storage medium provided in the above embodiments of this application and the regional heat load prediction method based on sparse dynamic graph neural network provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application.

Claims

1. A method for predicting regional heat load based on a sparse dynamic graph neural network, characterized in that, include: Acquire time-series data of heat load and external meteorological factors for the target area within a historical time period; After preprocessing the heat load time series data and the external meteorological factor time series data, the data are input into a pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future within a set time period. The heat load prediction model is obtained by training a sparse dynamic graph neural network based on historical heat load time series data samples from multiple regions and historical external meteorological factor time series data samples. The method pre-trains the heat load prediction model in the following manner: Historical heat load data and external meteorological factor data for multiple regions were acquired and statistically analyzed into time series within the same time interval, resulting in historical heat load time series data samples and historical external meteorological factor time series data samples for multiple regions. After performing the preprocessing on the historical heat load time series data samples and the historical external meteorological factor time series data samples, the historical heat load time series data samples and the historical external meteorological factor time series data samples are divided into batches of data of equal length. Set an objective function, input the batch data obtained by preprocessing the historical heat load time series data samples and the historical external meteorological factor time series data samples into the sparse dynamic graph neural network for training, and obtain the heat load prediction model after reaching the preset training cutoff condition. The sparse dynamic graph neural network includes a sequentially connected sparse graph learning module, a spatiotemporal memory enhancement module, and a temporal fusion module. The sparse graph learning module includes: a sparse graph construction component and a temporal graph learning component; The spatiotemporal memory enhancement module includes: a spatiotemporal convolution component and a global attention mechanism component; The time fusion module uses a skip connection architecture.

2. The method according to claim 1, characterized in that, The preprocessing involves dividing the time-series data into multiple batches.

3. The method according to claim 1, characterized in that, The objective function is set as the mean squared error loss function, and an L2 regularization term is used to prevent overfitting.

4. A regional heat load prediction device based on a sparse dynamic graph neural network, characterized in that, include: The acquisition module is used to acquire time-series data of heat load and external meteorological factors for the target area over a historical period. The prediction module is used to preprocess the heat load time series data and the external meteorological factor time series data, and then input them into the pre-trained heat load prediction model to obtain the predicted heat load data of the target area in the future within a set time period. The heat load prediction model is obtained by training a sparse dynamic graph neural network based on historical heat load time series data samples from multiple regions and historical external meteorological factor time series data samples. The device further includes: The training module is used to pre-train the heat load prediction model in the following manner: Historical heat load data and external meteorological factor data for multiple regions were acquired and statistically analyzed into time series within the same time interval, resulting in historical heat load time series data samples and historical external meteorological factor time series data samples for multiple regions. After performing the preprocessing on the historical heat load time series data samples and the historical external meteorological factor time series data samples, the historical heat load time series data samples and the historical external meteorological factor time series data samples are divided into batches of data of equal length. Set an objective function, input the batch data obtained by preprocessing the historical heat load time series data samples and the historical external meteorological factor time series data samples into the sparse dynamic graph neural network for training, and obtain the heat load prediction model after reaching the preset training cutoff condition. The sparse dynamic graph neural network includes a sequentially connected sparse graph learning module, a spatiotemporal memory enhancement module, and a temporal fusion module. The sparse graph learning module includes: a sparse graph construction component and a temporal graph learning component; The spatiotemporal memory enhancement module includes: a spatiotemporal convolution component and a global attention mechanism component; The time fusion module uses a skip connection architecture.

5. The apparatus according to claim 4, characterized in that, The preprocessing involves dividing the time-series data into multiple batches.

6. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method as claimed in any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that, It stores computer-readable instructions that can be executed by a processor to implement the method as described in any one of claims 1 to 3.

Citation Information

Patent Citations

  • Regional heating load prediction method, device and equipment and storage medium

    CN115796382A

  • Multi-energy load prediction method based on space-time diagram neural network

    CN115983448A