Comprehensive energy system rapid energy flow analysis method based on physical mechanism guidance and multi-head map attention mechanism

Abstract

The invention discloses an integrated energy system rapid energy flow analysis method based on physical mechanism guidance and a multi-head graph attention mechanism, and the method comprises an integrated energy system node and topological feature extraction method based on a graph neural network, which is used for constructing multi-energy flow node information and graph representation of an integrated energy system; constructing an energy flow convergence discrimination model based on a long short-term memory neural network algorithm, wherein the energy flow convergence discrimination model is used for discriminating whether the real-time energy flow to be predicted converges; an energy flow prediction method based on multi-head graph attention and physical mechanism guidance solves the problems that a traditional energy flow analysis method is low in calculation efficiency and a conventional machine learning method does not consider actual physical constraints, meanwhile, a multi-head graph attention mechanism is adopted to expand the visual field of attention, the learning ability of the relation between different nodes of a model is enhanced, and the learning efficiency is improved. Therefore, the operation efficiency and the result accuracy of subsequent energy flow prediction are ensured. According to the method, on the basis of historical operation data of a comprehensive energy system, a graph neural network is used for constructing and representing the relation between nodes and branches, a judgment model is constructed through a long-short-term memory neural network, and whether an energy flow prediction model carries out energy flow prediction or not is guided; and a rapid and accurate energy flow analysis result can be provided for operation scheduling of the integrated energy system.

Description

Technical Field

[0001] The application of the present invention belongs to the field of integrated energy systems, and the technical means of the present invention belongs to the field of time series data analysis and machine learning, and also belongs to the field of artificial intelligence technology, especially the field of power system flow analysis and integrated energy system energy flow analysis involving graph neural networks. Background Art

[0002] Under the background of "dual carbon goals", the proportion of high-proportion renewable energy in the integrated energy system has increased, and the construction of an integrated energy system with wind power, photovoltaics and other new energy as the main body has become an inevitable trend. Among them, the volatility and uncertainty of wind and solar power have brought huge challenges to the operation, dispatching and control of the integrated energy system. Accurate prediction of new energy power generation can provide reliable decision-making information for the dispatching and control of integrated energy systems at different levels, thereby ensuring the safe, stable, economical and green operation of the system. At the same time, the electric load in the power system, the thermal load in the thermal system and the natural gas load in the gas system change over time, making the integrated energy system present a certain degree of certainty. The uncertainty of a single energy system in the integrated energy system is transmitted to other systems through energy coupling devices, resulting in changes in the operating state of the system, which in turn affects the overall operating state of the integrated energy system. However, the traditional multi-energy flow calculation can only give the energy flow distribution of the system within a time section, and cannot give the continuous operating state of the system under the influence of uncertainty factors, and the calculation efficiency is low. Therefore, it is urgent to carry out research on fast multi-energy flow calculation methods considering the influence of multiple uncertainty factors for integrated energy systems.

[0003] How to establish a fast and accurate energy flow analysis method for integrated energy systems is a difficult problem currently faced by researchers. Traditional energy flow analysis methods for integrated energy systems generally use probabilistic energy flow analysis methods, which are mainly divided into three categories: simulation method, approximate method and analytical method. When considering the uncertainty factors in the integrated energy system, the advantages and disadvantages of different methods in various situations are as follows.

[0004] (1) Simulation method. The simulation method is mainly represented by the Monte Carlo simulation method (MCSM). The main disadvantages are long calculation time, high model complexity, sensitivity to initial conditions and parameters, and difficulty in capturing the long-term dynamic behavior and global optimization of the system. These disadvantages may limit the accuracy and reliability of the simulation results, especially when dealing with large-scale and highly complex systems.

[0005] (2) Approximate methods. Approximate methods are mainly represented by point estimation method, first-order second moment method and traceless transformation method. Their main disadvantages are limited accuracy, narrow scope of application, strong model dependence, difficulty in verification and poor robustness in the face of system parameter fluctuations or uncertainties. These disadvantages may lead to inaccurate and unreliable calculation results, affecting the final system analysis and optimization decision.

[0006] (3) The main disadvantage of analytical methods (including convolution method and semi-invariant method) in calculating the energy flow of integrated energy systems is the high computational complexity. Especially when dealing with high-dimensional and multivariable systems, accurate mathematical models and a large amount of computing resources are usually required, resulting in a significant increase in computing time and memory usage. In addition, analytical methods have strict assumptions on the model, and nonlinear and uncertain factors in the actual system may be difficult to accurately model, thus affecting the accuracy and reliability of the calculation results. Summary of the invention

[0007] In order to overcome the deficiencies of the above-mentioned prior art, the ultimate purpose of the present invention is to provide a method for rapid energy flow analysis of an integrated energy system based on machine learning, which adopts a multi-head graph attention mechanism and a physical guidance mechanism, and constructs the node and branch relationship in the actual integrated energy system through a graph network, solving the problem that the classical neural network model does not consider boundary attributes, and at the same time improves the accuracy of energy flow prediction by guiding the model training method through physical constraints. The specific technical route of the present invention includes the following steps:

[0008] Step 1: Construct an input feature set based on the data collected by each sensor in the integrated energy system. In the power system flow calculation, the input features of the PQ node are the injected active power P and reactive power Q, the input features of the PV node are the injected active power P and the voltage amplitude V, and the input features of the slack (balance node) node are the voltage amplitude V and the phase angle θ. In the energy flow analysis of the natural gas system, similar input features are used. The input features of the pressure-flow node are the injected flow and node pressure, the input features of the pressure-production node are the node pressure and natural gas production, and the input features of the balance node are the node pressure and reference flow. In the energy flow analysis of the thermal system, the input features of the temperature-flow node are the injected heat load and node temperature, the input features of the temperature-flow node are the node temperature and flow, and the input features of the balance node are the node temperature and reference flow. All the above node features are combined into a node feature matrix H. In the power system, conductance and admittance are used as the edge features of the graph, in the natural gas system, friction loss and heat loss are used as edge features, and in the thermal system, thermal conductivity, pressure loss and heat loss coefficient are used as edge features to construct the edge feature matrix. At the same time, the node adjacency matrix A is used to represent the topological relationship between nodes and branches in the topological graph.

[0009] Step 2: Construct an energy flow convergence algorithm based on the LSTM algorithm, calculate the energy flow from the power system, thermal system, natural gas system and the overall level of the integrated energy system, and determine whether the energy flow equation has a real number solution, that is, whether the actual integrated energy system can operate normally under this operating condition.

[0010] Step 3: Build an energy flow analysis model based on graph neural network, embed the topological structure information and physical constraints of the integrated energy system into the GCN network, so that the graph neural network can capture the dynamic relationship of time series data under the guidance of physical constraints during the training process, and conform to physical equations such as Ohm's law and the law of conservation of energy, thereby enhancing the model's learning ability at the physical knowledge level and improving the model's prediction accuracy.

[0011] Compared with the prior art, the present invention proposes an innovative method based on graph neural network to solve the problems of low computational efficiency, high sensitivity of model parameters, complex calculations, limited accuracy, etc. in the energy flow analysis of integrated energy systems. By inputting the topological information and node information in the actual integrated energy system into the graph neural network, the dynamic characteristics of different energy flows in the system can be captured more accurately. By effectively integrating the data input characteristics and topological relationships of electricity, natural gas, and thermal systems, the present invention solves the problem that the classical neural network model cannot effectively handle boundary attributes and cannot maintain physical constraints. At the same time, the LSTM-based energy flow convergence algorithm is used to solve the problem of judging whether the energy flow converges, further improving the calculation degree and accuracy of the energy flow prediction, so that the integrated energy system can achieve more efficient and accurate energy flow analysis and judgment under complex operating conditions.

[0012] Beneficial effects of the present invention:

[0013] The present invention comprehensively utilizes the historical operation data of the integrated energy system and the static data of the integrated energy system and other related information to realize rapid power flow convergence and prediction in the integrated energy system, which can solve the problems of low computational efficiency and limited accuracy in traditional energy flow calculations, and can provide fast and effective data support for the scheduling and decision-making of the integrated energy system, thereby maximizing the management timeliness of scheduling decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of an energy flow convergence model based on an LSTM algorithm used in a fast energy flow analysis method for an integrated energy system based on physical mechanism guidance and multi-head graph attention mechanism of the present invention;

[0015] Figure 2 The energy flow prediction flow chart based on graph neural network mentioned in the method for rapid energy flow analysis of integrated energy system based on physical mechanism guidance and multi-head graph attention mechanism of the present invention;

[0016] Figure 3This is the overall technical roadmap for energy flow convergence and prediction mentioned in the fast energy flow analysis method for an integrated energy system based on physical mechanism guidance and multi-head graph attention mechanism. DETAILED DESCRIPTION

[0017] In order to make the objects, advantages and features of the present invention more obvious, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, the specific steps in a specific embodiment of the present invention are as follows:

[0019] Step A. Obtain the historical operation data and system static information of the integrated energy system, build a training feature set for the energy flow prediction model, and obtain the real-time energy flow feature sequence of the integrated energy system to be predicted based on the real-time operation data of the integrated energy system.

[0020] The specific steps include: collecting historical operating data of the integrated energy system, including time series data of different energy flows such as electricity, heat, and natural gas. The data should cover relevant variables such as the system's load, power generation, energy storage, and energy conversion efficiency. Obtaining static information related to the system structure and parameters, such as the capacity, efficiency, topology, and energy conversion coefficient of the equipment, which helps to understand the configuration and operating characteristics of the system.

[0021] The second step is to perform feature engineering. In addition to the historical operating characteristics of the system itself, time periods such as hours, days, weeks, and months are collected, and meteorological characteristics such as temperature, air pressure, and humidity that have a great impact on the operating characteristics of the system are considered. The system operating characteristics are used as the main features, and the external features (time features, meteorological features) are used as auxiliary features to construct a feature set. The real-time features of the nodes inside the system are combined to obtain the energy flow feature sequence to be predicted X = [X1, X2, X3, ...], where X1 represents the feature set of the first node, and so on. At the same time, conductance and admittance are used as edge features of the graph. In the natural gas system, friction loss and heat loss are used as edge features, while in the thermal system, thermal conductivity, pressure loss, and heat loss coefficient are used as edge features. The edge feature matrix L = [L1, L2, L3, ...] is constructed, where L1 is the feature set of the first edge, and so on. Based on step A, the input feature set of the basic energy flow prediction model is obtained.

[0022] Step B. Construct an energy flow convergence model based on the LSTM algorithm to calculate whether each subsystem and the overall integrated energy system have converged to a real number solution. If converged, the data is input into the prediction model for the next step of integrated energy system energy flow prediction. If not, the output system does not converge. Figure 1 The specific steps are as follows:

[0023] First, the spatiotemporal features are extracted through the graph convolutional neural network based on the real-time node feature matrix and edge feature matrix constructed in the previous step A. Specifically, the graph-level features are extracted through the heterogeneous graph attention mechanism and graph pooling and input into the subsequent LSTM discriminant model for energy flow convergence. The mainstream graph neural network realizes the information transmission of network topology through the adjacency matrix, node matrix, and edge feature matrix. In the forward propagation process of the graph neural network, the input node feature matrix H is first subjected to a graph convolution operation, and the attention weight coefficient aij between nodes is calculated by combining the adjacency matrix A and the edge feature matrix L through the graph attention mechanism. Specifically, the calculation formula of the graph attention layer is as follows:

[0024]

[0025] a ij is the attention weight between nodes i and j, a is a learned attention vector, W is the weight matrix, || represents the vector concatenation operation, and h represents the node feature matrix of the input neural network.

[0026] After the graph convolution layer and the graph attention layer are processed, the new node feature matrix (H′∈R N×d ). In order to aggregate the features of all nodes in the graph into a global graph-level representation, a global graph pooling method is used. The global pooling operation can aggregate the features of all nodes in the graph and output a global representation (g∈R) of size (d). d ). The calculation formula for global graph average pooling is as follows:

[0027] [g j =max i∈V (h′ ij ),j=1,2,…,d]

[0028] Among them, g j is the maximum value of the graph in the jth feature dimension, h' ij is the value of node i on the jth feature dimension, and V is the number of nodes.

[0029] The global feature vector g∈R obtained by pooling operation d It can be used as the input of the subsequent discriminant model LSTM to further determine the energy flow at the graph level. Based on the graph-based time series data, LSTM is used for binary classification. The input of each time step is the feature extracted from the graph (usually a global graph representation with a dimension of d. The model needs to predict whether the current energy flow converges based on the feature sequence of these time steps. If it converges, the label value is output as 1, and if it does not converge, the label value is output as 0.

[0030] LSTM is mainly composed of cell state, input gate, output gate, and forget gate. Its calculation process is as follows:

[0031] First, the input gate is used to determine the new information input to the memory cell to update the cell state. The function expression is as follows:

[0032] i t =σ(W i [h t-1 ,x t ]+b i ) (1)

[0033] g t =tanh(W C [h t-1 ,x t ]+b C ) (2)

[0034] C t =f t C t-1 +i t g t (3)

[0035] Where: σ is the sigmoid function, which makes the output value range 0-1; i t is the input gate parameter; W i is the input gate weight; W C is the cell state weight; b i is the bias matrix of the input gate; b C is the bias matrix of the cell state; g t is the input state of the memory cell; tanh is the hyperbolic tangent function; C t is the cell state; C t-1 is the cell state at the previous moment; f t is the forget gate parameter; h t-1 is the hidden layer state at the previous moment; x t The input at the current moment.

[0036] Secondly, the output gate is used to filter the information of the memory unit and control the model output. The function expression is as follows:

[0037] ο t =σ(W ο [h t-1 ,x t ]+b ο ) (4)

[0038] h t =ο t tanh(C t ) (5)

[0039] In the formula: t is the output gate parameter; W οis the output gate weight; b ο is the bias matrix of the output gate; h t It is the hidden layer state at the current moment.

[0040] Finally, the forget gate selects the information to be forgotten and retained at time t-1, and the output value is between 0 and 1. The function expression is as follows:

[0041] f t =σ(W f [h t-1 ,x t ]+b f ) (6)

[0042] Where: W f is the weight of the forget gate; b f is the bias matrix of the forget gate.

[0043] The LSTM layer passes information through time steps and eventually outputs the hidden state h at each time step. t . For classification tasks, the hidden state of the last time step is usually used as the output feature of the model because it contains the context information of the entire sequence. After the LSTM layer is calculated, a linear fully connected layer is used to perform the energy flow convergence (binary classification) task, converting the implicit information output of the LSTM layer into a category prediction. The hidden state ht of the LSTM output is used as the input of the fully connected layer for classification, and ht is mapped to the category space. For the convergence task (0 or 1), the following linear transformation is used:

[0044] z=W fc ·h T +b fc

[0045] Among them, W fc is the weight of the fully connected layer, b fc is the bias term, h T is the last hidden state of LSTM. Sigmoid is used as the activation function to map the output z of the model to a probability value between [0,1].

[0046]

[0047] in, is the probability of the classification result (the probability of belonging to category 1). In the training process of the LSTM classification model, the binary cross entropy loss is used as the loss function to measure the difference between the probability of the model output and the actual label. Given the true label y∈0,1, the cross entropy loss calculation formula is:

[0048]

[0049] The goal of the model is to minimize this loss function so that As close as possible to the real label y. Through the LSTM convergence judgment model, it can be obtained whether the energy flow distribution of the comprehensive energy system to be predicted converges with a real number solution. If it converges, the next step of energy flow prediction is carried out.

[0050] Step C. Construct an integrated energy system energy flow prediction model based on graph convolutional neural network and multi-task learning

[0051] The prediction process is as follows Figure 2 As shown, the specific steps are as follows:

[0052] First, the node features and edge feature sets are normalized separately. By normalizing the node features and edge features, the training of the neural network can be accelerated. The normalization calculation formula is as follows:

[0053]

[0054] Among them, x i is the original data in the input feature, x i ′ is the normalized data, μ i is the mean value, σ i is the standard deviation, and m is the minimum value. The goal of energy flow prediction is to predict the energy flow (electricity flow, heat flow, natural gas flow, etc.) of each node in the integrated energy system. This process involves spatiotemporal prediction problems, so it is necessary not only to consider the immediate status of each node in the system, but also to capture the dependency of the time dimension.

[0055] The prediction model is GNN (graph neural network), and its framework and process are as follows:

[0056] 1. Graph representation construction: First, the graph data is represented as a set of nodes and edges. Each node v∈V will have a feature vector x v , represents the state information of the node, each edge e∈E will have a feature vector x e , represents the transmission characteristics between nodes (such as transmission loss, transmission capacity, etc.). Since energy flow prediction is usually a time series problem, the characteristics of nodes and edges will change over time. Therefore, the characteristics of each node and edge can be expressed as a time series x v (t) and x e (t), where t represents the time step.

[0057] 2. Graph Convolution (GCN): GCN is a neural network based on convolution operation, which shares parameters between node neighbors and updates the representation of nodes by aggregating the features of neighboring nodes. The formula of graph convolution is as follows:

[0058]

[0059] in, is the representation of node v at layer l. N(v) is the set of neighbor nodes of node v. vu is the normalization constant between node v and its neighbor node u. (l) is the weight matrix of the lth layer, b (l) is the bias term and σ is the activation function.

[0060] 3. Temporal convolution layer: In order to capture time dependency, a temporal convolution layer can be introduced on the basis of GNN to process the time series features of each node. The formula of the temporal convolution layer is as follows:

[0061] hv (l) (t) = σ(∑k = 0 K-1 W (l) ·x v (tk)+b (l) )

[0062] Where: x v (tk) represents the feature of node v at time step tk. K is the size of the convolution kernel, which controls the length of the time window and is defined as 16.

[0063] 4. Graph Attention Mechanism: The self-attention mechanism can dynamically adjust the relationship between nodes, so that each node can selectively pay attention to different neighboring nodes and better understand and simulate the local structure of the graph during information transmission.

[0064] The attention mechanism is added to the information aggregation process of the graph convolutional neural network, so that the node will not only aggregate information from neighboring nodes, but also adjust the weight according to the "importance" of the neighboring nodes. Weighted aggregation is performed by calculating the attention coefficient between the node and the neighboring nodes, and the updated node features are transformed nonlinearly (ReLU activation function) to obtain a new node representation. Finally, the representation ability of the node is enhanced by stacking multiple graph attention layers, similar to the traditional GNN layer stacking.

[0065] In order to weight neighbor nodes, we first need to calculate the attention coefficient between nodes. For node v and neighbor node u, calculate the attention coefficient α between them. vu , the formula is as follows:

[0066]

[0067] Where: h v and h uis the feature representation of nodes v and u. W is the learned weight matrix used to linearly transform node features. a is the weight vector used to calculate the attention coefficient. | represents the vector concatenation operation. LeakyReLU is an activation function with a negative slope to enhance the model's response to smaller inputs. The normalization operation exponentially normalizes the attention coefficients of all neighbor nodes and the target node to ensure that the sum of all weights is 1.

[0068] Based on the calculated attention coefficient α vu , node v will perform weighted aggregation based on the information of its neighboring nodes. The formula is as follows:

[0069]

[0070] in: is the representation of node v at layer l. is the representation of neighbor node u at the l-1th layer. vu is the attention weight of neighbor node u to node v.

[0071] In order to enhance the expressiveness of the model, the single-head attention is improved to a multi-head attention mechanism. That is, multiple independent attention heads are used, and the output of each head is concatenated or averaged. The purpose of this is to enable the model to understand the relationship between nodes from multiple different perspectives, and to understand the coupling relationship between different energy nodes in the integrated energy system from different perspectives. The formula of the multi-head attention mechanism is as follows:

[0072]

[0073] Where: K is the number of attention heads. is the attention coefficient of the i-th attention head. W (i) is the weight matrix of the (i)th attention head. | denotes concatenating the outputs of all heads.

[0074] The output of each layer can be transformed by a nonlinear activation function, and the activation function used is LeakyReLU. After attention aggregation and weighting, the output can be linearly transformed or dimensionally reduced to obtain new node features. Global average pooling is used to achieve the global representation of the graph. The calculation formula is:

[0075]

[0076] where |V| is the number of nodes in the graph, h G It is a representation of a graph.

[0077] After obtaining the global representation of the graph, the high-dimensional graph representation is linearly transformed into a lower-dimensional vector space to reduce the consumption of computing resources. The calculation formula is:

[0078]

[0079] Where W out is the reduced weight matrix.

[0080] 5. Physically driven loss function: The energy flow prediction model based on graph neural network realizes the modeling of the mapping relationship between system state information and energy flow prediction results. However, only using this structure cannot guarantee that the prediction results will meet the physical mechanism constraints such as node power balance in the system, which is also the main shortcoming of the existing data-driven model. In order to improve this shortcoming of the existing data-driven model in the training process, the concept of physical constraint loss function of physical information neural network is borrowed, and the loss function in the graph neural network training process is divided into data loss function and physical loss function. The data loss function uses mean square error:

[0081]

[0082] N is the number of nodes, T is the number of time steps, y v (t) is the actual energy flow value.

[0083] Physical Constraint Loss Function Here, we take the power subsystem as an example and use node power balance as the physical loss function. The specific physical loss function is as follows:

[0084] L=(1-α)L MSE +α(L physic_p +L physic_q )

[0085]

[0086]

[0087] Where: L physic_pj and L physic_qi They respectively represent the degree to which the active power and reactive power of node i in the model prediction results violate the node power balance equation, and α represents the regularization weight. The node power balance equation represented by the above formula is shifted and squared to obtain L physic_pj and L physic_qi The calculation formula is:

[0088]

[0089]

[0090] The above formula contains the topological structure information of the power system nodes, which is used as a regular term to characterize the connection relationship between the nodes in the power system.

[0091] 5. Prediction: Energy flow prediction tasks are divided into two categories according to the prediction step: single-step prediction and multi-step prediction. Single-step prediction is the simplest case, while multi-step prediction is divided into two types: direct multi-step prediction based on fixed historical data and recursive multi-stage prediction based on the latest prediction value. In order to perform multi-step energy flow rolling prediction, two graph convolution blocks are stacked and multi-layer attributes are fused to capture the spatiotemporal pattern of the data. Then, the output module is built with a temporal convolution layer and two fully connected layers to directly obtain the load forecast of all nodes in the future step. The overall technical route of energy flow convergence and prediction is as follows: Figure 3 shown.

[0092] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the essence and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention, and the patent protection scope of the present invention should be defined by the claims. The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A fast energy flow analysis method for integrated energy systems based on physical mechanism guidance and multi-head graph attention mechanism, characterized in that: The steps include: Construct node feature sets and edge feature sets through data collected by sensors in the integrated energy system; Graph neural network and graph attention mechanism are used to extract graph features and provide data preparation for the energy flow convergence model; Construct an energy flow convergence judgment model based on the LSTM algorithm, calculate the energy flow from the power system, thermal system, natural gas system and the overall level of the integrated energy system, and judge whether the energy flow equation has a real number solution and whether it converges; Using multi-head graph attention mechanism to improve the training process of graph neural network; Establish physical loss functions for power systems, natural gas systems, and thermal systems to guide the training process of graph neural networks; Construct a fast energy flow analysis method for integrated energy systems based on graph neural networks.

2. The method for rapid energy flow analysis of an integrated energy system according to claim 1, characterized in that When constructing the LSTM convergence model, in the forward propagation process of the graph neural network, the input node feature matrix H is firstly subjected to graph convolution operation, and the attention weight coefficient a between nodes is calculated by combining the adjacency matrix A and the edge feature matrix L through the graph attention mechanism. ij , to form a graph representation and extract graph-level features; The global feature vector g∈R extracted by the graph neural network d As the input of the subsequent discriminant model LSTM, further energy flow judgment at the graph level is performed. Based on the graph-based time series data, LSTM is used for binary classification to determine whether it has converged.

3. The method for rapid energy flow analysis of an integrated energy system according to claim 1, characterized in that: A multi-graph attention mechanism and physical guidance are used to guide the training of graph neural networks to construct a fast energy flow analysis method for energy systems.

4. The method for rapid energy flow analysis of an integrated energy system according to claim 1, characterized in that: The attention mechanism adopted is a multi-graph attention mechanism improved based on the graph attention mechanism; In order to enhance the expressiveness of the model, the single-head attention is improved to a multi-head attention mechanism. That is, multiple independent attention heads are used, and the output of each head is concatenated or averaged. The purpose of this is to enable the model to understand the relationship between nodes from multiple different perspectives, and to understand the coupling relationship between different energy nodes in the integrated energy system from different perspectives. The formula of the multi-head attention mechanism is as follows: Where: K is the number of attention heads. is the attention coefficient of the i-th attention head. W (i) is the weight matrix of the (i)th attention head. | denotes concatenating the outputs of all heads.

5. The method for rapid energy flow analysis of an integrated energy system according to claim 1, characterized in that: The physical loss function based on the physical information neural network is used to guide the learning of the graph neural network, so that the learning process of the graph neural network is more in line with the actual physical equation constraints; Drawing on the concept of physical constraint loss function of physical information neural network, the loss function in the training process of graph neural network is divided into data loss function and physical loss function. The data loss function uses mean square error: N is the number of nodes, T is the number of time steps, y v (t) is the actual energy flow value. Physical Constraint Loss Function Here, we take the power subsystem as an example and use node power balance as the physical loss function. The specific physical loss function is as follows: L=(1-α)L MSE +α(L physic_p +L physic_q ) Where: L physic_pj and L physic_qi They respectively represent the degree to which the active power and reactive power of node i in the model prediction results violate the node power balance equation, and α represents the regularization weight.

6. The method for rapid energy flow analysis of an integrated energy system according to claim 1, characterized in that: The physical constraints in the actual integrated energy system are taken into account and added to the training process of the graph neural network in the form of a loss function, so that the graph neural network can enhance the fitting of physical equations.

7. The method for rapid energy flow analysis of an integrated energy system according to claim 1, characterized in that: The proposed fast energy flow analysis method for the integrated energy system is calculated based on the historical data of the integrated energy system. It is a data-driven machine learning method that can greatly improve the computational efficiency while taking into account the physical constraint equations of the system.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images