Power distribution network line loss calculation method based on terminal data power flow fitting

Abstract

The invention discloses a power distribution network line loss calculation method based on terminal data power flow fitting, and belongs to the technical field of intelligent power distribution networks, and the method comprises the steps: S1, building a missing data deduction model through employing graph attention and a convolutional neural network based on the structural characteristics of a power distribution network and historical measurement data of the power distribution network; s2, taking the sparse measurement data of the power distribution network as input of a missing data deduction model, mapping initial features of nodes of the power distribution network by using the missing data deduction model to obtain a first mapping result, and obtaining a first attention weight of a branch of the power distribution network based on the first mapping result, obtaining a final feature of the node based on the first mapping result and the first attention weight; and S3, deducing first missing data based on the final features, and calculating the bus loss of the power distribution network according to the first missing data. The technical problem that it is difficult to fully and accurately obtain measurement data is solved, the data integrity and data accuracy are improved, and then the power distribution network line loss calculation accuracy is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent distribution network, and in particular to a method for calculating line loss of a distribution network based on terminal data flow fitting. Background Art

[0002] With the continuous growth of global energy demand and the rapid development of renewable energy, the power system is transforming towards a more intelligent, distributed and green direction. In this context, the distribution network is an important link between power sources and users in the power system. Its efficient operation is of great significance to the reliability, economy and sustainability of the power system. However, the widespread line loss problem in the distribution network not only causes energy waste, but also directly affects the operation efficiency and economic benefits of the power grid. Therefore, accurate calculation and evaluation of distribution network line losses have become an important basis for optimizing grid dispatch and energy conservation and emission reduction. For example, a distribution network line loss calculation method and system with patent number CN110212521A calculates the per-unit value of the distribution transformer daily load curve based on the proportional coefficient of each type of load in the typical load under the distribution transformer and the per-unit value based on the average load of the typical load curve. The active power of the distribution transformer at time t is obtained by dividing the value and the calculated daily power consumption of the distribution transformer by the load at the time point, and the reactive power of the distribution transformer at time t is calculated. The power flow is calculated based on the forward-backward method, the branch active power and reactive power are updated, the node voltage and voltage correction are calculated, and the convergence is checked. If converged, the power flow distribution is obtained according to the power flow calculation, and the distribution network loss is calculated. Although the above scheme takes into account the differences between different typical load curves and improves the calculation accuracy to a certain extent, the above scheme relies on accurate load curves, proportional coefficients and power consumption data. In the actual power system, the above data needs to be collected through various sensors and measuring equipment. However, in the process of transmitting the collected data to the data processing equipment for data processing, due to the influence of transmission distance or interference factors, the data is lost during the transmission process, which leads to insufficient data acquisition. In addition, due to various factors, the coverage of some areas is insufficient, which also leads to insufficient data acquisition. It can be seen that how to fully and accurately obtain measurement data is a technical problem that is difficult to overcome at present. Summary of the invention

[0003] In response to the technical problem that it is difficult to fully and accurately obtain measurement data in the prior art, the present invention provides a distribution network line loss calculation method based on terminal data flow fitting, by using a missing data deduction model constructed by graph attention and convolutional neural network to obtain a first mapping result, obtaining a first attention weight through the first mapping result, obtaining the final characteristics of the node based on the obtained first mapping result and the first attention weight, and accurately deducing the first missing data through the final characteristics of the node, which solves the technical problem that it is difficult to fully and accurately obtain measurement data, improves data integrity and data accuracy, and thus significantly improves the accuracy of distribution network line loss calculation.

[0004] In order to solve the above technical problems, the present invention provides a method for calculating line loss of a distribution network based on terminal data flow fitting, comprising the following steps: S1: Based on the structural characteristics of the distribution network and the historical measurement data of the distribution network, a missing data inference model is constructed using graph attention and convolutional neural network; S2: using the sparse measurement data of the distribution network as the input of the missing data inference model, using the missing data inference model to map the initial features of the nodes of the distribution network to obtain a first mapping result, obtaining a first attention weight of the branch of the distribution network based on the first mapping result, and obtaining the final feature of the node based on the first mapping result and the first attention weight; S3: Deducing first missing data based on the final features, and calculating the distribution network bus loss according to the first missing data.

[0005] After adopting the above technical solution, the present invention has the following advantages: The combination of graph attention and convolutional neural network can make full use of the structural characteristics and historical measurement data of the distribution network to build an efficient missing data deduction model, so that the missing data deduction model can accurately map the characteristics of the nodes. At the same time, the first attention weight of the branch is obtained through the accurate first mapping result, which is convenient for identifying the importance of the branch that affects the missing data deduction, and then combining the node characteristics with the importance of the branch that affects the missing data deduction to obtain the final characteristics of the node, and deduce the first missing data according to the final characteristics of the node, which improves the comprehensiveness of the missing data deduction, and then improves the accuracy of the deduced missing data, thereby improving the accuracy of the distribution network line loss calculation; The missing data inference model constructed by graph attention can also dynamically adjust the attention weight of the branch to adapt to the data changes under different operating conditions in the distribution network, thereby improving the flexibility and adaptability of the missing data inference model and further improving the accuracy of the distribution network line loss calculation; The technical problem of difficulty in obtaining sufficient and accurate measurement data is solved.

[0006] Preferably, the S1 comprises: S11: According to the structural characteristics of the distribution network, an initial missing data deduction model including the nodes and branches is constructed using graph attention and convolutional neural network; S12: Use historical measurement data of the distribution network to train the initial missing data deduction model, and use the successfully trained initial missing data deduction model as the missing data deduction model.

[0007] Preferably, the S12 includes: S12a: Inputting historical measurement data of the distribution network into the initial missing data deduction model to obtain a second mapping result, and obtaining a second attention weight of the branch according to the second mapping result; S12b: based on expert experience, obtain the rationality of the second attention weight in the process of deducing the second missing data. If the rationality meets the preset conditions, execute S12c; otherwise, adjust the parameters in the initial missing data deduction model based on the rationality, and use the distribution network historical measurement data corresponding to the second attention weight whose rationality does not meet the preset conditions as the distribution network historical measurement data, and execute S12a; S12c: Obtain second missing data based on the second mapping result and the second attention weight, obtain predicted line loss of the distribution network based on the second missing data, compare the predicted line loss with the actual line loss, and if the comparison result is within a preset range, it indicates that the initial missing data deduction model training is successful.

[0008] In this scheme, the rationality of the second attention weight is judged by expert experience. After the judgment is successful, the predicted line loss is compared with the actual line loss. If the comparison is successful, it means that the initial missing data deduction model is trained successfully, which provides double protection for the acquisition of the missing data deduction model and improves the accuracy of the missing data deduction model. When the judgment is unsuccessful, the parameters in the missing data deduction model are adjusted according to the rationality, which improves the pertinence of the parameter adjustment and avoids the problem of inefficiency caused by blind adjustment. At the same time, after the adjustment is completed, the historical measurement data of the distribution network corresponding to the second attention weight that does not meet the preset conditions is used as the historical measurement data of the distribution network, and S12a is executed. Through continuous iteration, the obtained second attention weight is gradually improved and accurate, and the parameters are adjusted according to the rationality of the gradually improved and accurate second attention weight, which further improves the pertinence of the parameter adjustment and improves the calculation efficiency and accuracy of the distribution network line loss.

[0009] Preferably, in S2, the expression for obtaining the first mapping result by mapping the initial characteristics of the nodes of the distribution network using the missing data inference model is: In the formula, represents the first mapping result of node i in the missing data inference model at the lth layer, W (l) represents the learnable weight matrix of the lth layer, Represents the initial features of the node i represented at the lth layer in the missing data inference model.

[0010] Preferably, in S2, the obtaining of the first attention weight of the branch of the distribution network based on the first mapping result includes: obtaining the edge feature of the branch based on the first mapping result, and obtaining the first attention weight based on the edge feature.

[0011] Preferably, the expression for obtaining the edge feature of the branch based on the first mapping result is: In the formula, ReLU() represents the first activation function, γ (l) represents the learnable parameters of the lth layer, represents the first mapping result of node j in the missing data inference model at the lth layer, A ij represents the connection relationship between the node i and the node j, Represents the edge features of the ij branch formed by the node i and the node j in the branch at the lth layer in the missing data inference model.

[0012] Preferably, the expression for obtaining the first attention weight based on the edge feature is: In the formula, represents the first attention weight of the ij branch in the missing data inference model at the lth layer, N(i) represents the set of neighbor nodes in the node that are associated with the node i, Represents the edge features of the ik branch formed by the node i in the branch and the node k in the node in the missing data inference model at the lth layer.

[0013] Preferably, in S2, obtaining the final feature of the node based on the first mapping result and the first attention weight includes: Performing a convolution operation based on the missing data inference model to obtain process characteristics of the node; Obtain final features based on process features; The expression of the convolution operation based on the missing data inference model to obtain the process characteristics is: In the formula, is the process feature obtained by the node i after the attention mechanism of the lth layer in the missing data inference model, and σ() represents the second activation function.

[0014] Preferably, the expression for obtaining the final feature based on the process feature is: In the formula, represents the final feature of the node i at the lth layer in the missing data inference model, u (l) (i) represents the weight of the node i after convolution with the lth convolution kernel in the missing data inference model, and N represents the number of the nodes.

[0015] Preferably, in S3, the expression for calculating the distribution network bus loss according to the first missing data is: In the formula, S loss is the total loss of the distribution network, P loss represents the active power loss of the branch, Q loss represents the reactive power loss of the branch, ε represents the set of branches, R ij represents the resistance of the ij branch, X ij represents the reactance of the ij branch, I ij represents the current of the ij branch; S ij | represents the complex power of the ij branch, |V i | represents the voltage amplitude of the node i, P ij and Q ij They respectively represent the injected active power and injected reactive power of the ij branch.

[0016] Beneficial effects of this program: The combination of graph attention and convolutional neural network can make full use of the structural characteristics and historical measurement data of the distribution network to build an efficient missing data deduction model, so that the missing data deduction model can accurately map the characteristics of the nodes. At the same time, the first attention weight of the branch is obtained through the accurate first mapping result, which is convenient for identifying the importance of the branch that affects the missing data deduction, and then combining the node characteristics with the importance of the branch that affects the missing data deduction to obtain the final characteristics of the node, and deduce the first missing data according to the final characteristics of the node, which improves the comprehensiveness of the missing data deduction, and then improves the accuracy of the deduced missing data, thereby improving the accuracy of the distribution network line loss calculation; The missing data inference model constructed by graph attention can also dynamically adjust the attention weight of the branch to adapt to the data changes under different operating conditions in the distribution network, thereby improving the flexibility and adaptability of the missing data inference model and further improving the accuracy of the distribution network line loss calculation; The rationality of the second attention weight is judged by expert experience. After the judgment is successful, the predicted line loss is compared with the actual line loss. If the comparison is successful, it means that the initial missing data deduction model is trained successfully, which provides double protection for the acquisition of the missing data deduction model and improves the accuracy of the missing data deduction model. When the judgment is unsuccessful, the parameters in the missing data deduction model are adjusted according to the rationality, which improves the pertinence of the parameter adjustment and avoids the problem of inefficiency caused by blind adjustment. At the same time, after the adjustment is completed, the historical measurement data of the distribution network corresponding to the second attention weight that does not meet the preset conditions is used as the historical measurement data of the distribution network, and S12a is executed. Through continuous iteration, the acquired second attention weight is gradually improved and accurate, and the parameters are adjusted according to the rationality of the gradually improved and accurate second attention weight, which further improves the pertinence of the parameter adjustment, and further improves the calculation accuracy while improving the calculation efficiency of the distribution network line loss. The technical problem of difficulty in fully and accurately obtaining measurement data is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0018] Figure 1 It is a flow chart of the method for calculating line loss of distribution network based on terminal data power flow fitting of the present invention; Figure 2 This is a comparison chart of the effects of the distribution network line loss calculation method based on terminal data flow fitting of the present invention and other methods for missing data deduction. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0020] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the operations (or steps) as sequential processes, many of the operations (or steps) therein can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the drawings; the process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0021] Embodiment 1: like Figure 1 As shown, the distribution network line loss calculation method based on terminal data flow fitting includes the following steps: S1: Based on the structural characteristics of the distribution network and the historical measurement data of the distribution network, a missing data inference model is constructed using graph attention and convolutional neural network.

[0022] The S1 includes: S11: According to the structural characteristics of the distribution network, an initial missing data deduction model including the nodes and branches is constructed using graph attention and convolutional neural network; S12: Use historical measurement data of the distribution network to train the initial missing data deduction model, and use the successfully trained initial missing data deduction model as the missing data deduction model.

[0023] The S12 includes: S12a: Inputting historical measurement data of the distribution network into the initial missing data deduction model to obtain a second mapping result, and obtaining a second attention weight of the branch according to the second mapping result; S12b: based on expert experience, obtain the rationality of the second attention weight in the process of deducing the second missing data. If the rationality meets the preset conditions, execute S12c; otherwise, adjust the parameters in the initial missing data deduction model based on the rationality, and use the distribution network historical measurement data corresponding to the second attention weight whose rationality does not meet the preset conditions as the distribution network historical measurement data, and execute S12a; S12c: Obtain second missing data based on the second mapping result and the second attention weight, obtain predicted line loss of the distribution network based on the second missing data, compare the predicted line loss with the actual line loss, and if the comparison result is within a preset range, it indicates that the initial missing data deduction model training is successful.

[0024] In this embodiment, the structural characteristics of the distribution network are a network architecture composed of the nodes in the distribution network and the connection relationship between these nodes formed by the power lines. The historical measurement data of the distribution network are the active and reactive power injected into the node, the active and reactive power injected into the branch, the node voltage amplitude, etc. According to the structural characteristics of the distribution network, a graph G = {V, E} is defined to represent the distribution network, where V = {v1, ..., v N} and E represent the number of nodes and branches in the distribution network respectively. For a node m in the distribution network, z m Represents the sparse measurement data of node m, including node voltage amplitude, node injected active power and reactive power, branch injected active power and reactive power, etc. m =[z m (k0),ε,z m (k2),ε,…,z m (k n )], where z m (k n ) represents k n The measurement data of node m at time instant, ε represents the missing measurement data of node m at a certain time instant. Missing data deduction is the sparse measurement data of the distribution network. Deduction refers to the use of sparse measurement data to deduce and fill in the missing data to obtain all the measurement data z m_all , z m_all =[z m (k0),z m (k1),z m (k2),z m (k3),…,z m (k n )].

[0025] Specifically, based on expert experience, the rationality of the second attention weight in the process of deducing the second missing data is obtained. If the rationality meets the preset conditions, S12c can be executed as follows: if the second attention weights of branch A, branch B and branch C obtained through the second mapping result are 0.1, 0.4 and 0.5 respectively, but after analysis through expert experience, it is found that branch A is more important in the process of deducing the second missing data, then the rationality does not meet the preset conditions. The preset conditions can be flexibly adjusted according to needs. If it is necessary to maximize the accuracy of the missing data deduction model, in addition to judging whether the preset conditions are met based on the size of the second attention weight of the branch, it can also be used when the second attention weights of all branches meet the expert experience analysis. In addition to the obtained weight relationship, the difference in the second attention weights of each branch is strictly set. If the second attention weights of branch A, branch B and branch C obtained through the second mapping result are 0.1, 0.4 and 0.5 respectively, then after analysis through expert experience, it is found that the importance of branch A, branch B and branch C gradually increases in the process of deducing the second missing data, but after analysis through expert experience, it is found that the importance of branch A and branch B in the process of deducing the second missing data is slightly different, while the importance of branch B and branch C in the process of deducing the second missing data is relatively different. At this time, the difference in the second attention weights of branch A, branch B and branch C does not meet the above-mentioned importance gap, and the reasonableness in this case does not meet the preset conditions. The preset range is also flexibly set according to demand. If the line loss calculation needs to be maximized, the preset range is set relatively small. The rationality of the second attention weight is judged by expert experience. After the judgment is successful, the predicted line loss is compared with the actual line loss. If the comparison is successful, it means that the initial missing data deduction model is trained successfully, which provides double protection for the acquisition of the missing data deduction model and improves the accuracy of the missing data deduction model. When the judgment is unsuccessful, the parameters in the missing data deduction model are adjusted according to the rationality, which improves the pertinence of the parameter adjustment and avoids the problem of inefficiency caused by blind adjustment. At the same time, after the adjustment is completed, the distribution network historical measurement data corresponding to the second attention weight that does not meet the preset conditions is used as the distribution network historical measurement data, and S12a is executed. Through continuous iteration, the acquired second attention weight is gradually improved and accurate, and the parameters are adjusted according to the rationality of the gradually improved and accurate second attention weight, which further improves the pertinence of the parameter adjustment and improves the efficiency and accuracy of the distribution network line loss calculation.

[0026] S2: Using the sparse measurement data of the distribution network as the input of the missing data inference model, using the missing data inference model to map the initial features of the nodes of the distribution network to obtain a first mapping result, obtaining a first attention weight of the branch of the distribution network based on the first mapping result, and obtaining the final features of the node based on the first mapping result and the first attention weight.

[0027] In S2, the expression for obtaining the first mapping result by mapping the initial characteristics of the nodes of the distribution network using the missing data inference model is: In the formula, represents the first mapping result of node i in the missing data inference model at the lth layer, W (l) represents the learnable weight matrix of the lth layer, Represents the initial features of the node i represented at the lth layer in the missing data inference model.

[0028] In S2, the obtaining of a first attention weight of a branch of the distribution network based on the first mapping result includes: The edge feature of the branch is obtained based on the first mapping result, and the first attention weight is obtained based on the edge feature.

[0029] The expression for obtaining the edge feature of the branch based on the first mapping result is: In the formula, ReLU() represents the first activation function, γ (l) represents the learnable parameters of the lth layer, represents the first mapping result of node j in the missing data inference model at the lth layer, A ij represents the connection relationship between the node i and the node j, Represents the edge features of the ij branch formed by the node i and the node j in the branch at the lth layer in the missing data inference model.

[0030] The expression for obtaining the first attention weight based on the edge feature is: In the formula, represents the first attention weight of the ij branch in the missing data inference model at the lth layer, N(i) represents the set of neighbor nodes in the node that are associated with the node i, Represents the edge features of the ik branch formed by the node i in the branch and the node k in the node in the missing data inference model at the lth layer.

[0031] In S2, obtaining the final feature of the node based on the first mapping result and the first attention weight includes: Performing a convolution operation based on the missing data inference model to obtain process characteristics of the node; Obtain final features based on process features; The expression of the convolution operation based on the missing data inference model to obtain the process characteristics is: In the formula, is the process feature obtained by the node i after the attention mechanism of the lth layer in the missing data inference model, and σ() represents the second activation function.

[0032] The expression for obtaining the final feature based on the process feature is: In the formula, represents the final feature of the node i at the lth layer in the missing data inference model, u (l) (i) represents the weight of the node i after convolution with the lth convolution kernel in the missing data inference model, and N represents the number of the nodes.

[0033] In this embodiment, X i The initial characteristics of node i are input characteristics. The initial characteristics include the active and reactive power injected into the node, the active and reactive power injected into the branch, and the node voltage amplitude. The topological connection matrix of the distribution network is A N×N , A ij =0 means that nodes i and j are not connected, A ij =1 means that nodes i and j are connected. The association relationship is a connection relationship. If nodes i and j are connected, it means that nodes i and j are associated, otherwise they are not associated.

[0034] In this embodiment, the first activation function ReLU() only involves a comparison operation and possible multiplication, which makes the calculation speed of ReLU in large-scale neural networks very fast. At the same time, compared with traditional activation functions such as Sigmoid and Tanh, ReLU has a constant derivative (derivative is 1) in the positive range, which helps to maintain the stability of the gradient during back propagation, thereby alleviating the gradient disappearance problem. Therefore, edge features are obtained by ReLU(), which improves the acquisition efficiency and accuracy. The second activation function σ() is one of tanh, sigmoid, ReLU, PReLU, ELU, softplus, softmax, and swish, which is selected according to needs. The missing data inference model is used to accurately map the characteristics of the node. At the same time, the edge characteristics of the branch are obtained through the accurate first mapping result and the first activation function, which improves the accuracy and acquisition efficiency of the obtained edge characteristics. The first attention weight is obtained through the edge characteristics, which facilitates the identification of the importance of the branch that has an impact on the missing data deduction, and then the node characteristics are combined with the importance of the branch that has an impact on the missing data deduction to obtain the final node characteristics, and the first missing data is deduced according to the final node characteristics, which improves the comprehensiveness of the missing data deduction, and then improves the accuracy of the deduced missing data, thereby improving the accuracy of the distribution network line loss calculation.

[0035] S3: Deducing first missing data based on the final features, and calculating the distribution network bus loss according to the first missing data.

[0036] In S3, the expression for calculating the distribution network bus loss according to the first missing data is: In the formula, S loss is the total loss of the distribution network, P loss represents the active power loss of the branch, Q loss represents the reactive power loss of the branch, ε represents the set of branches, R ij represents the resistance of the ij branch, X ij represents the reactance of the ij branch, I ij represents the current of the ij branch; S ij | represents the complex power of the ij branch, |V i | represents the voltage amplitude of the node i, P ij and Q ij They respectively represent the injected active power and injected reactive power of the ij branch.

[0037] In this embodiment, the first missing data and the second missing data are both voltage amplitude and phase angle characteristics of the node. The mapping relationship of obtaining the missing data through the power flow variables in the final characteristics can be expressed by the function as follows: F:{P i ,Q i ,P ij ,Q ij}→{V i ,θ i} Among them, P i and Q i denote the injected active and reactive power of node i, respectively, P ij and Q ij Respectively represent the injected active and reactive power of branch ij, V i and θ i They represent the voltage amplitude and phase angle of node i respectively.

[0038] In this embodiment, if Figure 2 As shown, the IEEE57 node standard system is used to carry out simulation tests, and the present invention is compared with the deep neural network and the weighted least squares method. The mean absolute error (MAE) and the root mean square error (RMSE) are used as evaluation indicators to evaluate the data deduction effects of different algorithms on the node voltage amplitude. After comparison, it can be seen that the deduction results of the present invention are closer to the actual values ​​and have higher deduction accuracy. Through advanced technologies such as graph neural networks, the missing data can be accurately deduced, and the complex relationship between the flow variables can be captured, thereby optimizing the operation efficiency of the distribution network and reducing energy waste. At the same time, it also promotes the intelligent development of the power system, helps to achieve low-carbon and green energy goals, and improves the economy and sustainability of the distribution network.

[0039] The specific implementation described above is a preferred implementation of the distribution network line loss calculation method based on terminal data flow fitting of the present invention, and is not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to this specific implementation. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for calculating line loss in a distribution network based on terminal data flow fitting, characterized in that: The following steps are involved: S1: Based on the structural characteristics of the distribution network and the historical measurement data of the distribution network, a missing data inference model is constructed using graph attention and convolutional neural network; S2: using the sparse measurement data of the distribution network as the input of the missing data inference model, using the missing data inference model to map the initial features of the nodes of the distribution network to obtain a first mapping result, obtaining a first attention weight of the branch of the distribution network based on the first mapping result, and obtaining the final feature of the node based on the first mapping result and the first attention weight; S3: Deducing first missing data based on the final features, and calculating the distribution network bus loss according to the first missing data.

2. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 1 is characterized in that: The S1 includes: S11: According to the structural characteristics of the distribution network, an initial missing data deduction model including the nodes and branches is constructed using graph attention and convolutional neural network; S12: Use historical measurement data of the distribution network to train the initial missing data deduction model, and use the successfully trained initial missing data deduction model as the missing data deduction model.

3. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 2 is characterized in that: The S12 includes: S12a: Inputting historical measurement data of the distribution network into the initial missing data deduction model to obtain a second mapping result, and obtaining a second attention weight of the branch according to the second mapping result; S12b: based on expert experience, obtain the rationality of the second attention weight in the process of deducing the second missing data. If the rationality meets the preset conditions, execute S12c; otherwise, adjust the parameters in the initial missing data deduction model based on the rationality, and use the distribution network historical measurement data corresponding to the second attention weight whose rationality does not meet the preset conditions as the distribution network historical measurement data, and execute S12a; S12c: Obtain second missing data based on the second mapping result and the second attention weight, obtain predicted line loss of the distribution network based on the second missing data, compare the predicted line loss with the actual line loss, and if the comparison result is within a preset range, it indicates that the initial missing data deduction model training is successful.

4. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 1 is characterized in that: In S2, the expression for obtaining the first mapping result by mapping the initial characteristics of the nodes of the distribution network using the missing data inference model is: In the formula, represents the first mapping result of node i in the missing data inference model at the lth layer, W (l) represents the learnable weight matrix of the lth layer, Represents the initial features of the node i represented at the lth layer in the missing data inference model.

5. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 4 is characterized in that: In S2, obtaining a first attention weight of a branch of the distribution network based on the first mapping result includes: The edge feature of the branch is obtained based on the first mapping result, and the first attention weight is obtained based on the edge feature.

6. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 5 is characterized in that: The expression for obtaining the edge feature of the branch based on the first mapping result is: In the formula, ReLU() represents the first activation function, γ (l) represents the learnable parameters of the lth layer, represents the first mapping result of node j in the missing data inference model at the lth layer, A ij represents the connection relationship between the node i and the node j, Represents the edge features of the ij branch formed by the node i and the node j in the branch at the lth layer in the missing data inference model.

7. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 6 is characterized in that: The expression for obtaining the first attention weight based on the edge feature is: In the formula, represents the first attention weight of the ij branch in the missing data inference model at the lth layer, N(i) represents the set of neighbor nodes in the node that are associated with the node i, Represents the edge features of the ik branch formed by the node i in the branch and the node k in the node in the missing data inference model at the lth layer.

8. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 7 is characterized in that: In S2, obtaining the final feature of the node based on the first mapping result and the first attention weight includes: Performing a convolution operation based on the missing data inference model to obtain process characteristics of the node; Obtain final features based on process features; The expression of the convolution operation based on the missing data inference model to obtain the process characteristics is: In the formula, is the process feature obtained by the node i after the attention mechanism of the lth layer in the missing data inference model, and σ() represents the second activation function.

9. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 8, characterized in that: The expression for obtaining the final feature based on the process feature is: In the formula, represents the final feature of the node i at the lth layer in the missing data inference model, u (l) (i) represents the weight of the node i after convolution with the lth convolution kernel in the missing data inference model, and N represents the number of the nodes.

10. The method for calculating line loss in a distribution network based on terminal data power flow fitting according to claim 9, characterized in that: In S3, the expression for calculating the distribution network bus loss according to the first missing data is: In the formula, S loss is the total loss of the distribution network, P loss represents the active power loss of the branch, Q loss represents the reactive power loss of the branch, ε represents the set of branches, R ij represents the resistance of the ij branch, X ij represents the reactance of the ij branch, I ij represents the current of the ij branch; |S ij | represents the complex power of the ij branch, |V i | represents the voltage amplitude of the node i, P ij and Q ij They respectively represent the injected active power and injected reactive power of the ij branch.

Images