A method, device, equipment and storage medium for predicting regional heat load

Abstract

The present application provides an interpretable regional heat load prediction method, device, equipment and storage medium, which relates to the field of smart heating technology. The method includes: obtaining the heating load time series data of multiple target customers in the area within a historical time period; after pre-processing the heating load time series data, inputting a pre-trained heating load prediction model to obtain the predicted heating load data of multiple target customers in the area within a future set time period; the heating load prediction model is obtained by training an interpretable graph neural network with samples of historical heating load time series data of multiple target customers in the area. The heating load prediction results of the present application are interpretable. In the face of complex real-time changes, it can explain the reasons for the prediction results and the inherent laws of time series data fluctuations. Compared with traditional heating load predictions, it can improve the accuracy of heat load predictions for multiple customers in the area.

Description

Technical Field

[0001] The present application relates to the field of smart heating technology, and specifically to an interpretable regional heat load prediction method, device, equipment and storage medium. Background Art

[0002] We have always focused on sustainable energy development and energy conservation and emission reduction, with energy transformation and the construction of smart energy systems being key development areas. As a key component of smart energy systems, interpretable regional heat load forecasting can provide important support and guidance for energy management and energy conservation and emission reduction.

[0003] Disadvantages of traditional heat load forecasting technology: Traditional heat load forecasting technology usually uses black box models for prediction, which makes it difficult to explain the reasons and internal laws of the prediction results and lacks explainability; traditional heat load forecasting technology usually uses empirical models or traditional statistical models for heat load prediction, which is difficult to accurately predict for complex nonlinear systems and has low prediction accuracy; traditional heat load forecasting technology usually uses static models for prediction, which is difficult to cope with complex real-time changes, such as changes in user energy consumption behavior. Summary of the Invention

[0004] The purpose of this application is to provide an interpretable regional heat load forecasting method, device, electronic device and storage medium, thereby improving the accuracy of heat load forecasting for multiple customers in a region.

[0005] In a first aspect, an embodiment of the present application provides an interpretable regional heat load prediction method, comprising:

[0006] Obtain heating load time series data for multiple target customers in the region over a historical period;

[0007] After pre-processing the heating load time series data, the data is input into a pre-trained heating load forecasting model to obtain forecast heating load data for multiple target customers in the area within a set time period in the future;

[0008] The heating load prediction model is obtained by training an interpretable graph neural network with historical heating load time series data samples of multiple target customers in the area.

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

[0010] Obtain historical heating load data of multiple target customers in the area, and compile statistics into time series within the same time interval to obtain historical heating load time series data samples of multiple target customers in the area;

[0011] After performing the preprocessing on the historical heating load time series data samples, the historical heating load time series data samples are converted into heating load supervised data;

[0012] An objective function is set, and the supervised heating load data is input into the interpretable graph neural network for training, and the heating load prediction model is obtained after a preset training cutoff condition is reached.

[0013] In some embodiments of the present application, converting the historical heating load time series data samples into supervised heating load data includes:

[0014] The first several variables in the historical heating load time series data sample are used as the input variables of the model, and the last variable is aggregated as the output of the model to obtain the supervised data of the heating load.

[0015] In some embodiments of the present application, the preprocessing includes maximum processing, minimum processing, data averaging and normalization processing.

[0016] In some embodiments of the present application, the interpretable graph neural network includes a sequentially connected spatiotemporal convolution component, a spatiotemporal memory enhancement component, a multi-component fusion component, and a predictor;

[0017] The spatiotemporal convolution component includes: a multi-layer graph convolution module in the time dimension and a single-layer convolution module in the space dimension. The multi-layer graph convolution module in the time dimension is used to extract time correlation from time series data, and the single-layer convolution module in the space dimension is used to extract spatial correlation from time series data.

[0018] The spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatial attention module is used to extract local dynamic correlations in the spatial dimension from time series data, and the recurrent neural network module is used to further extract temporal correlations from time series data.

[0019] The multi-component fusion component is used to combine the inputs;

[0020] The predictor is used to make a final prediction based on the output data of the multi-component fusion component.

[0021] In some embodiments of the present application, the recurrent neural network module adopts a gated recursive unit.

[0022] In some embodiments of the present application, the objective function is set to a mean square error loss function, and an L2 regularization term is used to prevent overfitting.

[0023] In a second aspect, an embodiment of the present application provides an interpretable regional heat load prediction device, comprising:

[0024] An acquisition module is used to obtain the heating load time series data of multiple target customers in the area during a historical period;

[0025] A prediction module is used to pre-process the heating load time series data and input it into a pre-trained heating load prediction model to obtain predicted heating load data for multiple target customers in the area within a set time period in the future;

[0026] The heating load prediction model is obtained by training an interpretable graph neural network with historical heating load time series data samples of multiple target customers in the area.

[0027] In some embodiments of the present application, the device further includes: a model training module for pre-training the heating load prediction model in the following manner:

[0028] Obtain historical heating load data of multiple target customers in the area, and compile statistics into time series within the same time interval to obtain historical heating load time series data samples of multiple target customers in the area;

[0029] After performing the preprocessing on the historical heating load time series data samples, the historical heating load time series data samples are converted into heating load supervised data;

[0030] An objective function is set, and the supervised heating load data is input into the interpretable graph neural network for training, and the heating load prediction model is obtained after a preset training cutoff condition is reached.

[0031] In some embodiments of the present application, the model training module is specifically used to:

[0032] The first several variables in the historical heating load time series data sample are used as the input variables of the model, and the last variable is aggregated as the output of the model to obtain the supervised data of the heating load.

[0033] In some embodiments of the present application, the preprocessing includes maximum processing, minimum processing, data averaging and normalization processing.

[0034] In some embodiments of the present application, the interpretable graph neural network includes a sequentially connected spatiotemporal convolution component, a spatiotemporal memory enhancement component, a multi-component fusion component, and a predictor;

[0035] The spatiotemporal convolution component includes: a multi-layer graph convolution module in the time dimension and a single-layer convolution module in the space dimension. The multi-layer graph convolution module in the time dimension is used to extract time correlation from time series data, and the single-layer convolution module in the space dimension is used to extract spatial correlation from time series data.

[0036] The spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatial attention module is used to extract local dynamic correlations in the spatial dimension from time series data, and the recurrent neural network module is used to further extract temporal correlations from time series data.

[0037] The multi-component fusion component is used to combine the inputs;

[0038] The predictor is used to make a final prediction based on the output data of the multi-component fusion component.

[0039] In some embodiments of the present application, the recurrent neural network module adopts a gated recursive unit.

[0040] In some embodiments of the present application, the objective function is set to a mean square error loss function, and an L2 regularization term is used to prevent overfitting.

[0041] In a third aspect, the present application provides an electronic device comprising: 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 described in the first aspect.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium having computer-readable instructions stored thereon, and the computer-readable instructions can be executed by a processor to implement the method described in the first aspect.

[0043] Compared with the existing technology, the interpretable regional heat load prediction method provided by this application obtains the heating load time series data of multiple target customers in the region within a historical time period; after pre-processing the heating load time series data, it inputs a pre-trained heating load prediction model to obtain the predicted heating load data of multiple target customers in the region within a future set time period; the heating load prediction model is obtained by training an interpretable graph neural network with samples of historical heating load time series data of multiple target customers in the region. It can be seen that the heating load prediction results of this application are interpretable. In the face of complex real-time changes, it can explain the reasons for the prediction results and the inherent laws of time series data fluctuations. Compared with traditional heating load prediction, it can improve the accuracy of heat load prediction for multiple customers in the region. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0045] Figure 1 A flow chart of an interpretable regional heat load prediction method provided by the present application is shown;

[0046] Figure 2 A flow chart of a heating load prediction model training method provided by the present application is shown;

[0047] Figure 3 A schematic diagram of the prediction process of a heating load prediction model provided by the present application is shown;

[0048] Figure 4 A schematic diagram of an interpretable regional heat load prediction device provided by the present application is shown;

[0049] Figure 5 A schematic diagram of an electronic device provided by the present application is shown. DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0051] It should be noted that, unless otherwise specified, the technical or scientific terms used in this application should have the common meanings understood by those skilled in the art to which this application belongs.

[0052] In addition, the terms "first" and "second" are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0053] Please refer to Figure 1 , Figure 1 This is a flow chart of an interpretable regional heat load prediction method provided in an embodiment of the present application, including the following S101 to S102:

[0054] S101. Obtain heating load time series data of multiple target customers in a region within a historical time period.

[0055] The heating load time series data is a time series data consisting of multiple heating load data of multiple target customers in the area at a set time interval. The time interval can be set to a specific time interval such as hour, day, week or month.

[0056] S102: After pre-processing the heating load time series data, the pre-trained heating load prediction model is input to obtain predicted heating load data of multiple target customers in the area within a set time period in the future.

[0057] Among them, the heating load prediction model is obtained by training an interpretable graph neural network with historical heating load time series data samples of multiple target customers in the area.

[0058] Optionally, the above-mentioned data preprocessing operations include: maximum value processing, minimum value processing, data averaging and normalization processing.

[0059] After preprocessing the heating load time series data, the pre-trained heating load forecasting model is input to obtain the forecasted heating load data of multiple target customers in the area within a set time period in the future.

[0060] The following describes how to pre-train a heating load prediction model. Specifically, the heating load prediction model can be pre-trained in the following ways: Figure 2 As shown, it includes steps S201 to S203:

[0061] S201. Obtain historical heating load data of multiple target customers in the area, and compile statistics into time series within the same time interval to obtain historical heating load time series data samples of the multiple target customers in the area.

[0062] The historical heating load data of multiple target customers in the region are collected, and time intervals are set. The historical heating load data are statistically analyzed in chronological order to form time series within the same time interval to obtain initial multivariate time series data. The multivariate time series data is a sample of the historical heating load time series data of multiple target customers in the region. The specific implementation steps are as follows:

[0063] Determine the time interval: Based on the time span of the data and the time granularity required for statistics, determine the time interval, which can be set to hours, days, weeks, or months. Time-align the raw data: For each customer's historical heating load data, align it according to the time interval, that is, fill the data at the corresponding time point. Merge the data of multiple customers: Merge the data of all customers in chronological order to obtain a multivariate time series dataset, in which each time point contains the heating load data of each customer.

[0064] S202: After preprocessing the historical heating load time series data samples, convert the historical heating load time series data samples into supervised heating load data.

[0065] Specifically, {t1, t2, ..., t n The data within the time interval represented by} is aggregated into statistical data of a clearly calibrated moment, and the initial input sequence is generated by traversing the data samples.

[0066] In {t1,t2,…,t n}time interval to collect heating load data of each customer in the area, and use m T ={m1,m2,…,m T} represents the time series of the historical heating variable m of each customer in the area within T time intervals. Represents the input set of heating load data of D customers at time t.

[0067] Furthermore, in order to reduce the significant data range differences between the data, this application uses the Min-Max method to normalize the data and compress it to the interval [0,1] to speed up the training of the model.

[0068] The formulas for normalization and denormalization are as follows:

[0069]

[0070] z=z'(max(z)-min(z))+min(z);

[0071] in, Denotes a set of samples, N denotes the number of observed samples, z' denotes the normalized data, min(·) and max(·) denote the minimum and maximum values ​​of the input vector, respectively. The denormalization formula is used to restore the predicted value of the model output to the predicted heating load data.

[0072] Specifically, in S202, the historical heating load time series data samples are converted into heating load supervised data, including:

[0073] The first several variables in the historical heating load time series data sample are used as the input variables of the model, and the last variable is aggregated as the output of the model to obtain the supervised data of the heating load.

[0074] Specifically, in order to learn temporal characteristics from time series, this application converts time series data into supervised data. Given a time series data, during the conversion process, the first several variables are used as input variables of the model, and the last variable is considered as the output variable of the model.

[0075] By collecting heating load data from each customer in the area and converting it into supervised data, an interpretable deep learning model is trained using the supervised data. Finally, the trained model can provide more accurate predictions.

[0076] S203 , setting an objective function, inputting the supervised heating load data into the interpretable graph neural network for training, and obtaining the heating load prediction model after reaching a preset training cutoff condition.

[0077] Specifically, the interpretable graph neural network includes a sequentially connected spatiotemporal convolution component, a spatiotemporal memory enhancement component, a multi-component fusion component, and a predictor.

[0078] The spatiotemporal convolution component includes: a multi-layer graph convolution module on the time dimension and a single-layer convolution module on the spatial dimension. The multi-layer graph convolution module on the time dimension is used to extract time correlation from time series data, and the single-layer convolution module on the spatial dimension is used to extract spatial correlation from time series data; the spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatial attention module is used to extract local dynamic correlation in the spatial dimension from time series data, and the recurrent neural network module is used to further extract time correlation from time series data; the multi-component fusion component is used to combine each input; the predictor is used to make a final prediction based on the output data of the multi-component fusion component.

[0079] Please refer to Figure 3 ,The following is a detailed introduction to the heating load prediction model.

[0080] The spatiotemporal convolution component includes a multi-layer graph convolution network (MGCN) module along the time dimension and a single-layer convolution module in the spatial dimension.

[0081] Electricity meters recording customer heating data with a correlation matrix form a mesh network that can be defined as a directed graph:

[0082]

[0083] in, express A∈R is a finite set of nodes (i.e., power), E represents the set of edges, and A∈R N×N Representation diagram The adjacency matrix of .

[0084] For each node Based on the Pearson correlation matrix C between target customers, select the top k related nodes To build connections between nodes in The graph structure exploits the underlying topological properties of multivariate time series inputs.

[0085] The correlation coefficients between each specific factor and heating usage data are calculated from the perspective of time difference. The calculation formula is:

[0086]

[0087] Where C∈R N×N represents the calculated Pearson correlation matrix, L represents the length of the time series data, represents the heat supply time series data of the i-th target customer at time t, Represents the mean of the hot supply time series data of the i-th target customer.

[0088] The MGCN module is designed to transform graph data into an algebraic form based on spectral graph theory and discover topological properties, including connectivity and other patterns in the graph structure. These graph properties can be obtained by analyzing the Laplacian matrix spectral graph. The normalized Laplacian matrix can be formulated as:

[0089]

[0090] Among them, L p ∈R T×T represents the normalized Laplace matrix; D∈R T×T is a diagonal matrix, D (i,i) =∑ j A (i,j) ;I unit Represents the identity matrix.

[0091] The embedding process of the MGCN module can be described as follows:

[0092]

[0093]

[0094] in, represents the graph convolution operation, Denotes an adjacency matrix with inserted self-loops, M∈R B×L×N represents the input time series, R g,o Represents the final output of the MGCN module, ⊙ represents the corresponding element multiplication, and σ(·) represents the rectified linear unit (ReLU) activation function. ε g is a binary mask used to forget some parameters; W g,θ and W g,o Both represent learnable weight matrices, b gRepresents the bias parameter of the model.

[0095] After capturing the adjacent time step information of each node on the graph in the temporal dimension, a single-layer convolutional module in the spatial dimension is further integrated to highlight the important pattern information between different target customers.

[0096] Spatial convolution can be formulated as follows:

[0097]

[0098] Among them, R c,o represents the generative representation of spatiotemporal convolution, W c,o represents the learnable parameters of the convolution kernel, ε c Represents a binary mask matrix used to forget some parameters to prevent the network from overfitting.

[0099] In this embodiment, a multi-layer graph convolution module in the time dimension is used to extract the temporal correlation from the adjacent time steps of each electricity meter, and a single-layer convolution module in the spatial dimension is used to capture the spatial correlation from all electricity meters.

[0100] The spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatiotemporal memory enhancement component is designed to further capture the dynamic spatial and temporal correlations of the input.

[0101] The electricity meters that record customer heating data form a mesh network, and there is a certain amount of dynamic interaction between each meter. The spatial attention module is applied to exploit the local dynamic correlation between meters in the spatial dimension. The specific process is as follows:

[0102]

[0103]

[0104] in, Indicates the attention score that dynamically adjusts the influence weight in the spatial dimension, represents the product of the spatial attention mechanism, and represents the learnable weight matrix, b g Represents the learnable bias parameter.

[0105] The output of spatial attention is input into a recurrent neural network module, which can be based on a gated recurrent unit (GRU) to further extract temporal correlation:

[0106]

[0107]

[0108]

[0109] h t =(1-u t )⊙h t-1 +u t ⊙z t ;

[0110] Among them, e t ,u t and z t Represent the internal states of the reset gate, update gate, and cell state at time step t, respectively. t Represents the output of the recurrent neural network module. r,e , W r,u and W r,z Represents the weighting matrix. b r,e , b r,u and b r,z represents the corresponding bias parameter. Indicates the sigmoid activation function, and [;] indicates the merge operation. To ensure alignment with the desired output shape, Emb is used in the time dimension and spatial dimension respectively. tmp (·) and Emb spt (·) Map the output of the recurrent neural network module to the embedding space:

[0111]

[0112] in, represents the final output of the spatiotemporal memory enhancement component, ε r Represents a binary mask matrix.

[0113] The multi-component fusion component is used to combine the inputs. Nonlinear modules can effectively capture potential dependencies in time or space dimensions. However, excessive nonlinearity may lead to gradient vanishing or exploding problems, which in turn will have an adverse effect on prediction accuracy. In order to improve the robustness of the model, in this embodiment, the linear representation is combined with the representation of the output of the recurrent neural network module. Combine to achieve the final prediction:

[0114]

[0115] in, represents the predicted value, Indicates the element-wise addition operation, W f,o is a learnable weight parameter, b f is the corresponding bias parameter.

[0116] The predictor is used to make the final prediction based on the output data of the multi-component fusion component.

[0117] To train the predictor, we set the model parameters such as the objective function, the time window size T, the number of feature maps of the CNN, the value of λ in the activation function σ(·), and the hidden state parameter H of the GRU neural network, and then train the model.

[0118] Specifically, the objective function can be set as the mean squared error loss function (MSE), and the L2 regularization term is used to prevent overfitting. The objective function can be defined as:

[0119]

[0120] Among them, K is the length of the training sample; N is the dimension of the target data; L is the prediction step; λ is a constant, which represents the regularization coefficient; Θ is the total number of weight parameters in the predictor, Represents the weight parameters of the model.

[0121] Using the inverse normalization function, the model output is restored to the predicted heating load data for multiple target customers in the area. Based on the model output, the above model parameters are adjusted to obtain the optimal heating load forecast model.

[0122] After obtaining the heating load prediction model, the heating load time series data of multiple target customers in the area can be input into the heating load prediction model to predict the heating load data of multiple target customers in the area within a set time period in the future. The regional heat load prediction process can be referred to Figure 3 As shown, the prediction results can be provided to the heating system to make accurate production plans and carry out precise regional heating.

[0123] In order to provide real-time and accurate heating forecasts for the region, this application proposes an interpretable regional heat load forecasting method. This method collects the original data on heating usage of each user in the region and converts it into supervised data. The supervised data is used to train an interpretable graph neural network to obtain a heating load forecasting model, and finally the trained model is used to provide accurate forecasts.

[0124] The interpretable regional heat load prediction method provided by the embodiment of the present application has the following beneficial effects:

[0125] The heating load prediction model of the present application is an interpretable deep learning model. The deep learning model can be used to efficiently learn and model data, so that more accurate prediction results can be obtained, and the prediction results are highly accurate. The deep learning model of the present application can quickly process large amounts of data and can perform real-time calculations during inference, and is highly efficient. The deep learning model established by the present application through target customer correlation can provide an explanation of the prediction results, and can analyze the decision-making process and key features of the model, so as to better understand the prediction results of the model. The deep learning model of the present application can enhance the learning ability of the model by increasing the number of layers and nodes, and can also improve the performance of the model by enhancing training data, and has high scalability. The deep learning model of the present application can handle the complex relationships between multiple input features, and is also robust to noise and outliers in the input data.

[0126] The interpretable regional heat load forecasting method based on deep learning can provide important technical support for the construction of smart energy systems, optimal energy scheduling, energy cleaning and emission reduction, and can also provide important guarantees for energy security.

[0127] In the above embodiment, an interpretable regional heat load prediction method is provided. Correspondingly, the present application also provides an interpretable regional heat load prediction device 10. The interpretable regional heat load prediction device provided in the embodiment of the present application can implement the above interpretable regional heat load prediction method. The interpretable regional heat load prediction device can be implemented by software, hardware, or a combination of software and hardware. For example, the interpretable regional heat load prediction device can include integrated or separate functional modules or units to perform the corresponding steps in the above methods. Please refer to Figure 4 As shown, including:

[0128] An acquisition module 101 is used to acquire heating load time series data of multiple target customers in a region within a historical time period;

[0129] The prediction module 102 is configured to pre-process the heating load time series data and input the pre-trained heating load prediction model to obtain the predicted heating load data of multiple target customers in the area within a set time period in the future;

[0130] The heating load prediction model is obtained by training an interpretable graph neural network with historical heating load time series data samples of multiple target customers in the area.

[0131] In some embodiments of the present application, the device further includes: a model training module for pre-training the heating load prediction model in the following manner:

[0132] Obtain historical heating load data of multiple target customers in the area, and compile statistics into time series within the same time interval to obtain historical heating load time series data samples of multiple target customers in the area;

[0133] After performing the preprocessing on the historical heating load time series data samples, the historical heating load time series data samples are converted into heating load supervised data;

[0134] An objective function is set, and the supervised heating load data is input into the interpretable graph neural network for training, and the heating load prediction model is obtained after a preset training cutoff condition is reached.

[0135] In some embodiments of the present application, the model training module is specifically used to:

[0136] The first several variables in the historical heating load time series data sample are used as the input variables of the model, and the last variable is aggregated as the output of the model to obtain the supervised data of the heating load.

[0137] In some embodiments of the present application, the preprocessing includes maximum processing, minimum processing, data averaging and normalization processing.

[0138] In some embodiments of the present application, the interpretable graph neural network includes a sequentially connected spatiotemporal convolution component, a spatiotemporal memory enhancement component, a multi-component fusion component, and a predictor;

[0139] The spatiotemporal convolution component includes: a multi-layer graph convolution module in the time dimension and a single-layer convolution module in the space dimension. The multi-layer graph convolution module in the time dimension is used to extract time correlation from time series data, and the single-layer convolution module in the space dimension is used to extract spatial correlation from time series data.

[0140] The spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatial attention module is used to extract local dynamic correlations in the spatial dimension from time series data, and the recurrent neural network module is used to further extract temporal correlations from time series data.

[0141] The multi-component fusion component is used to combine the inputs;

[0142] The predictor is used to make a final prediction based on the output data of the multi-component fusion component.

[0143] In some embodiments of the present application, the recurrent neural network module adopts a gated recursive unit.

[0144] In some embodiments of the present application, the objective function is set to a mean square error loss function, and an L2 regularization term is used to prevent overfitting.

[0145] The interpretable regional heat load prediction device provided in the embodiment of the present application and the interpretable regional heat load prediction method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented therein.

[0146] An embodiment of the present application also provides an electronic device corresponding to the method provided in the aforementioned embodiment. The electronic device may be an electronic device for a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the above-mentioned interpretable regional heat load prediction method.

[0147] Please refer to Figure 5 , which shows a schematic diagram of an electronic device provided by some embodiments of the present application. Figure 5 As shown, the electronic device 20 includes: a processor 200, a memory 201, a bus 202 and a communication interface 203, and the processor 200, the communication interface 203 and the memory 201 are connected through the bus 202; the memory 201 stores a computer program that can be run on the processor 200, and when the processor 200 runs the computer program, it executes the phishing email tracing method provided by any of the aforementioned embodiments of the present application.

[0148] The memory 201 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The system network element communicates with at least one other network element via at least one communication interface 203 (which may be wired or wireless), and may utilize the Internet, a wide area network, a local area network, a metropolitan area network, or the like.

[0149] Bus 202 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, and the like. Memory 201 is used to store programs, and processor 200 executes the programs upon receiving execution instructions. The phishing email tracing method disclosed in any of the aforementioned embodiments of the present application may be applied to or implemented by processor 200.

[0150] The processor 200 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor 200 or by software instructions. The above 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 field 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 various 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 method disclosed in conjunction with the embodiments of this application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 201 , and the processor 200 reads the information in the memory 201 and completes the steps of the above method in combination with its hardware.

[0151] The electronic device provided in the embodiment of the present application and the interpretable regional heat load prediction method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented by them.

[0152] An embodiment of the present application also provides a computer-readable storage medium corresponding to the interpretable regional heat load prediction method provided by the aforementioned embodiment, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it will execute the interpretable regional heat load prediction method provided by any of the aforementioned embodiments.

[0153] 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 or magnetic storage media, which are not listed here one by one.

[0154] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the interpretable regional heat load prediction method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and they should all be included in the scope of the claims and description of the present application.

Claims

1. A method for predicting regional heat loads that can be interpreted, characterized in that: include: Obtain heating load time series data for multiple target customers in the region over a historical period; After pre-processing the heating load time series data, the data is input into a pre-trained heating load forecasting model to obtain forecast heating load data for multiple target customers in the area within a set time period in the future; The heating load prediction model is obtained by training an interpretable graph neural network using historical heating load time series data samples of multiple target customers in the area; The method pre-trains the heating load prediction model in the following manner: Obtain historical heating load data of multiple target customers in the area, and compile statistics into time series within the same time interval to obtain historical heating load time series data samples of multiple target customers in the area; After performing the preprocessing on the historical heating load time series data samples, the historical heating load time series data samples are converted into heating load supervised data; Setting an objective function, inputting the supervised heating load data into the interpretable graph neural network for training, and obtaining the heating load prediction model after reaching a preset training cutoff condition; The interpretable graph neural network includes a sequentially connected spatiotemporal convolution component, a spatiotemporal memory enhancement component, a multi-component fusion component, and a predictor; The spatiotemporal convolution component includes: a multi-layer graph convolution module in the time dimension and a single-layer convolution module in the space dimension. The multi-layer graph convolution module in the time dimension is used to extract time correlation from time series data, and the single-layer convolution module in the space dimension is used to extract spatial correlation from time series data. The spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatial attention module is used to extract local dynamic correlations in the spatial dimension from time series data, and the recurrent neural network module is used to further extract temporal correlations from time series data. The multi-component fusion component is used to combine the inputs; The predictor is used to make a final prediction based on the output data of the multi-component fusion component.

2. The method according to claim 1, characterized in that The converting the historical heating load time series data samples into heating load supervised data includes: The first several variables in the historical heating load time series data sample are used as the input variables of the model, and the last variable is aggregated as the output of the model to obtain the supervised data of the heating load.

3. The method according to claim 1, characterized in that The preprocessing includes maximum value processing, minimum value processing, data averaging and normalization processing.

4. The method according to claim 1, wherein The recurrent neural network module adopts a gated recursive unit.

5. The method according to claim 1, wherein The objective function is set as the mean square error loss function, and the L2 regularization term is used to prevent overfitting.

6. An interpretable regional heat load prediction device, characterized in that: include: An acquisition module is used to obtain the heating load time series data of multiple target customers in the area during a historical period; A prediction module is used to pre-process the heating load time series data and input it into a pre-trained heating load prediction model to obtain predicted heating load data for multiple target customers in the area within a set time period in the future; The heating load prediction model is obtained by training an interpretable graph neural network using historical heating load time series data samples of multiple target customers in the area; The device pre-trains the heating load prediction model in the following manner: Obtain historical heating load data of multiple target customers in the area, and compile statistics into time series within the same time interval to obtain historical heating load time series data samples of multiple target customers in the area; After performing the preprocessing on the historical heating load time series data samples, the historical heating load time series data samples are converted into heating load supervised data; Setting an objective function, inputting the supervised heating load data into the interpretable graph neural network for training, and obtaining the heating load prediction model after reaching a preset training cutoff condition; The interpretable graph neural network includes a sequentially connected spatiotemporal convolution component, a spatiotemporal memory enhancement component, a multi-component fusion component, and a predictor; The spatiotemporal convolution component includes: a multi-layer graph convolution module in the time dimension and a single-layer convolution module in the space dimension. The multi-layer graph convolution module in the time dimension is used to extract time correlation from time series data, and the single-layer convolution module in the space dimension is used to extract spatial correlation from time series data. The spatiotemporal memory enhancement component includes: a spatial attention module and a recurrent neural network module. The spatial attention module is used to extract local dynamic correlations in the spatial dimension from time series data, and the recurrent neural network module is used to further extract temporal correlations from time series data. The multi-component fusion component is used to combine the inputs; The predictor is used to make a final prediction based on the output data of the multi-component fusion component.

7. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that Computer-readable instructions are stored thereon, and the computer-readable instructions can be executed by a processor to implement the method according to any one of claims 1 to 5.

Images