Power distribution network topology identification and state estimation method based on message passing neural network and online deep learning

Abstract

The invention discloses a power distribution network topology identification and state estimation method based on a message passing neural network and online deep learning, and the method comprises the steps: training a neural network through historical measurement data, building a power distribution network topology identification model, combining the obtained topology information with corresponding power grid measurement data, and constructing a graph data set; taking the obtained graph data set as a training set to train a message passing neural network to obtain a power distribution network state estimation model; and finally, inputting the real-time measurement data into a state estimation model of the topology identification model to obtain a real-time state estimation result under the section. According to the method, the power distribution network topology can be accurately identified under limited measurement data, accurate state estimation is provided for a power distribution system, and technical support is provided for subsequent safety assessment and optimized operation.

Description

Technical Field

[0001] The present invention proposes a distribution network topology identification and state estimation method based on message passing neural network and online deep learning, which belongs to the field of AC power system. Background Art

[0002] As the penetration rate of distributed renewable energy (DRE) in power systems continues to increase, distribution networks are facing new challenges. Traditional distribution systems are based on unidirectional power flow, but with the access of DRE, power flow becomes bidirectional, resulting in frequent changes in the network topology. In order to ensure the safe, stable and economical operation of the distribution network, it is particularly important to obtain the state information and topology of the distribution system in real time and accurately. However, existing distribution network state estimation technologies face many limitations due to limited measurement data, strong model-driven dependence, and inability to adapt to topology changes.

[0003] To solve these problems, researchers have gradually turned to dynamic state estimation techniques and data-driven methods. Dynamic state estimation methods, such as the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF), can capture the temporal dynamic characteristics of the distribution network, but still rely on accurate physical models and sufficient measurement data. In contrast, data-driven methods use historical measurement data for training to construct a mapping relationship between grid measurement data and state variables. However, these methods usually ignore changes in topology or require a large amount of measurement data to ensure the accuracy of the estimation. Summary of the invention

[0004] Purpose of the invention: To address the above problems, the present invention provides a distribution network topology identification and state estimation method based on message passing neural network and online deep learning, which can accurately obtain real-time state information in the distribution network.

[0005] Technical solution, in order to achieve the above-mentioned invention object, the present invention proposes a distribution network topology identification and state estimation method based on message passing neural network and online deep learning, comprising the following steps:

[0006] Step 1: Establish a distribution network topology identification model, and use the smart meter or SCADA system in the distribution network to collect the voltage amplitude and power flow data of the distribution network in real time; input the real-time collected measurement data into the online neural network model, and after the model is trained with historical data, identify the current topology structure according to the input voltage and power data; introduce the hedge back propagation mechanism into the neural network, and dynamically adjust the network depth by weighted aggregation of the output results of each layer to ensure convergence in the face of different topology changes and identify the accurate topology structure;

[0007] Step 2: Construct a graph dataset, represent the nodes in the distribution network as nodes in the graph, and represent the lines in the distribution network as edges in the graph; use the node voltage amplitude, power injection characteristics and line power as edge characteristics and the topology information obtained by the topology identification model in step 1 to construct a graph dataset, and perform data normalization. This dataset will be used as training data for the message passing neural network;

[0008] Step 3: Establish a message passing neural network distribution network state estimation model, use the constructed graph data set to perform offline training of the message passing neural network, compare the state estimation values ​​of the system node voltage amplitude and branch phase angle difference output by the message passing graph neural network model with the true values ​​during the training process, and calculate the loss function, update the weights and parameters of the message passing graph neural network model through the back propagation algorithm, and obtain the message passing graph neural network model suitable for different topologies through multiple iterations;

[0009] Step 4, using the message passing neural network distribution network state estimation model to perform estimation, including: constructing the graph data of the current section based on the normalized current section measurement information and distribution network topology information obtained, and inputting the graph data of the current section into the message passing graph neural network model based on offline training completed in step 3, to obtain the state estimation value of the system node voltage amplitude and branch phase angle difference of the current section.

[0010] Furthermore, the output of the distribution network topology identification model established in step 1 is the disconnection state of the line, as follows:

[0011] (1.1) Switch line status

[0012]

[0013] In the formula, a i represents the state of the i-th switch line, 0 means the line is disconnected, 1 means the line is closed; y represents the output of the online neural network, and λ1 and λ2 are set to constant values; when the output of the neural network exceeds λ2, the line switch is considered to be closed; otherwise, the line switch is considered to be open; in this scenario, λ1=λ2=0.5;

[0014] (1.2) Measurement equation

[0015] The measurement equation of historical measurement information can be expressed as follows:

[0016] zh(x)=e

[0017] Where z represents the measurement vector, x represents the state variable vector, h(·) represents the AC or DC power flow equation, and e represents the measurement residual vector;

[0018] For AC systems, the measurements include node voltage amplitude measurement, node active injection power measurement, node reactive injection power measurement, branch active power measurement, and branch reactive power measurement. The power flow equation is as follows:

[0019]

[0020] θ ii' =θ i -θ i' {i,i'}∈N AC

[0021] Where P AC,ii’ and Q AC,ii’ Represent the branch active and reactive power of the branch connected to the i-th AC node and the i'th AC node, respectively, V i and V i’ Represent the voltage amplitude of the i-th AC node and the i'th AC node, θ ii’ represents the phase angle difference between the branches connected to the ith AC node and the i'th AC node, θ i and θ i’ Represent the phase angles of the ith AC node and the i'th AC node, P AC,i and Q AC,i They represent the active injection power and reactive injection power of the i-th AC node, respectively. and They represent the branch active power measurement error and branch reactive power measurement error of the branch connected to the i-th AC node and the i'th AC node, respectively. and They represent the node active power injection measurement error and node reactive power injection measurement error of the i-th AC node, respectively. AC represents the set of nodes in the AC system, G ii’ and B ii’ Respectively represent the real and imaginary parts of the admittance matrix of the branch connected to the ith AC node and the i'th AC node;

[0022] (1.3) The specific mechanism of hedging back propagation is as follows

[0023]

[0024] h 0 = z = [V,P node ,Q node ,P line ,Q line ]

[0025] In the formula, L represents the total number of layers of the neural network, w l and b l is the weight matrix and bias vector of the hidden layer l, Θ l and α l is the weight matrix and weight coefficient of the lth layer, represents the output of layer l, h l represents the activation value of the lth layer, h 0 Represents the initial activation value of the model, including voltage amplitude measurement V, node active power injection measurement P node and reactive power injection measurement Q node , the active power flow of the line is measured P line and reactive power flow measurement Q line , F(z) represents the output of the neural network.

[0026] Furthermore, in step 2, the graph data set is constructed by abstracting the electrical components of the distribution network system into a graph structure and normalizing the data. The specific process is as follows:

[0027] (2.1) Node definition and feature extraction

[0028] In the distribution network, power buses, load nodes, and distributed power sources are defined as nodes in the graph. The voltage amplitude and power state information of each node are input into the neural network as the feature vector of the node. The specific extraction steps are as follows:

[0029] Each bus and load node in the distribution network corresponds to a node in the diagram;

[0030] According to the data obtained from the SCADA system or smart meter, the voltage amplitude, active power and reactive power of each node are extracted, and the node characteristics are represented as a vector:

[0031] x i =[V i ,P i ,Q i ]

[0032] V i ∈V

[0033] P i ∈P node

[0034] Q i ∈Q node

[0035] In the formula, x i represents the characteristics of node i, V i represents the voltage amplitude measurement of node i, P i ,Q i denote the active and reactive power measurements of node i respectively;

[0036] (2.2) Edge definition and feature extraction

[0037] eg ij =[P ij ,Q ij ]

[0038] P ij ∈P line

[0039] Q ij ∈Q line

[0040] In the formula, eg ij represents the characteristics of edge ij, P ij represents the active power measurement of edge ij, Q ij represents the reactive power measurement of edge ij;

[0041] (2.3) Constructing a graph dataset

[0042] Graph data construction, based on the network structure obtained by topology recognition, generates a graph G(V,E), where V is a node set, representing the busbars and load nodes in the distribution network; E is an edge set, representing the power lines in the distribution network;

[0043] Node feature matrix, which stacks the feature vectors of all nodes into a matrix X, whose size is |V|×d, where |V| is the number of nodes and d is the dimension of the node feature;

[0044] The edge feature matrix stacks the feature vectors of each edge into a matrix E of size |E|×g, where |E| is the number of edges and g is the feature dimension of each edge;

[0045] (2.4) Data normalization

[0046]

[0047] In the formula, u i ' is the normalized historical measurement value of the ith i is the ith original historical measurement value, min(·) is the minimum value in a set of data, and max(·) is the maximum value in a set of data.

[0048] Furthermore, in step 3, the graph data is used for training, and the historical measurement data of the node voltage amplitude and the branch power are read, and the measurement data is collected by the SCADA measurement device; the read historical measurement is input into the message passing neural network model for training, and the target output of the model is the node voltage amplitude and the branch phase angle difference;

[0049] The message passing neural network is defined as follows:

[0050] For the i-th node, the following expression is given:

[0051]

[0052] In the formula, represents the information received by node i at time t+1; N(i) represents all neighboring nodes of node i; is the state of node i at time t; e ij is the edge feature between node i and node j; is the node message passing function at time t; this formula indicates that the information received by node i comes from and the state of the adjacent node j at time t and the edge features e connecting them ij ;Once the information is collected, the node status is updated;

[0053]

[0054] in, is the node update function, which takes and As input, generate the node state at time t+1 The node message passing function is as follows:

[0055]

[0056] in, is the learnable parameter matrix; is the activation function of the fully connected layer of the node at time t, is the edge feature of edge ij at time t;

[0057] Based on the above node modeling, the edges in the graph are modeled. The specific formula is as follows:

[0058]

[0059] in, represents the information received by edge ij at time t+1, and is the edge message passing function and state update function at time t, represents the state of edge ij at time t+1, and the edge message passing function is as follows:

[0060]

[0061] in, is the learnable parameter matrix, is the activation function of the fully connected layer at time t;

[0062] The final output of the model is:

[0063]

[0064] Where y represents the final output vector of the model, T represents the last time step of the message passing phase, R(·) is a fully connected layer whose parameters are determined by network training, and the state update function uses the gated recurrent unit GRU:

[0065]

[0066] Furthermore, in step 4, the target output of the model is the voltage amplitude of the node and the branch phase angle difference. The specific conversion formula of the branch phase angle difference and the node phase angle is as follows:

[0067] θ=(B T B) -1 B T Δθ

[0068] Where θ is the node voltage phase angle estimation vector, Δθ is the branch phase angle difference estimation vector, and B is the node-branch association matrix.

[0069] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0070] By introducing an online neural network (ONN), the present invention can quickly identify the real-time topology structure in the distribution network, which is particularly suitable for distribution network scenarios with frequent changes. Compared with the traditional state estimation method that relies on a fixed topology, the present invention does not need to rely on a static topology structure and can dynamically adjust when the topology changes, thereby improving the adaptability and flexibility of the system.

[0071] Improve state estimation accuracy: Message passing neural network (MPNN) is used for state estimation, and the node and line information of the distribution network is modeled as a graph structure, making full use of the physical topological characteristics of the system. MPNN can effectively integrate the information of nodes and edges, perform multiple rounds of iterative updates, and significantly improve the accuracy of distribution network state estimation, especially when the measurement data is sparse.

[0072] Dealing with the problem of insufficient measurement data: There are usually limited measurement points in the distribution network, which leads to great observability challenges for state estimation. This paper combines pseudo-measurement data with actual measurement data and uses deep learning models for topology identification and state estimation, reducing the reliance on highly redundant measurement data and effectively solving the problem of insufficient measurement data.

[0073] Improve computational efficiency: The MPNN in the present invention combines the advantages of graph neural networks and greatly improves the computational efficiency of state estimation in large-scale distribution networks by parallel computing the information transmission between nodes and edges. Compared with traditional weighted least squares (WLS) and mixed integer linear programming (MILP) methods, the present invention can significantly reduce computational complexity and shorten computational time, and is suitable for application in actual large-scale distribution systems.

[0074] Adaptive adjustment: The present invention introduces the HBP (Hedge Backpropagation) mechanism through a neural network, which can adaptively adjust the layer depth and weight parameters of the neural network, quickly adjust the model for different topologies, and ensure the accuracy and real-time performance of state estimation.

[0075] In view of the above problems, the present invention proposes a joint topology identification and state estimation method based on message passing neural network (MPNN) and online neural network (ONN). The method can quickly identify the real-time topology structure of the distribution network, and build a graph data set in combination with measurement data, and perform efficient state estimation through MPNN. The present invention significantly improves the accuracy and efficiency of distribution network state estimation in the case of frequent topology changes and insufficient measurement data. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of the method for realizing distribution network topology identification and state estimation based on message passing neural network and online deep learning described in the invention;

[0077] Figure 2 This is a 33-node three-phase balance test system diagram;

[0078] Figure 3 This is a 123-node three-phase unbalanced test system diagram;

[0079] Figure 4 is the loss function graph of the topology recognition model on the training set

[0080] Figure 5 This is the distribution diagram of weight coefficients of each layer after training and adjustment of the topology recognition model.

[0081] Figure 6Comparison of the state estimation results of three state estimation methods on the 123-node system with continuous-time section topology 3: (a) node voltage amplitude estimation error, (b) node voltage phase angle estimation error. DETAILED DESCRIPTION

[0082] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0083] like Figure 1 As shown, the present invention proposes a distribution network topology identification and state estimation method based on message passing neural network and online deep learning, comprising the following steps:

[0084] Step 1: Establish a distribution network topology identification model, and use the smart meter or SCADA system in the distribution network to collect the voltage amplitude and power flow data of the distribution network in real time; input the real-time collected measurement data into the online neural network model, and after the model is trained with historical data, identify the current topology structure according to the input voltage and power data; introduce the hedge back propagation mechanism into the neural network, and dynamically adjust the network depth by weighted aggregation of the output results of each layer to ensure convergence in the face of different topology changes and identify the accurate topology structure;

[0085] Step 2: Construct a graph dataset, represent the nodes in the distribution network as nodes in the graph, and represent the lines in the distribution network as edges in the graph; use the node voltage amplitude, power injection characteristics and line power as edge characteristics and the topology information obtained by the topology identification model in step 1 to construct a graph dataset, and perform data normalization. This dataset will be used as training data for the message passing neural network;

[0086] Step 3: Establish a message passing neural network distribution network state estimation model, use the constructed graph data set to perform offline training of the message passing neural network, compare the state estimation values ​​of the system node voltage amplitude and branch phase angle difference output by the message passing graph neural network model with the true values ​​during the training process, and calculate the loss function, update the weights and parameters of the message passing graph neural network model through the back propagation algorithm, and obtain the message passing graph neural network model suitable for different topologies through multiple iterations;

[0087] Step 4, using the message passing neural network distribution network state estimation model to perform estimation, including: constructing the graph data of the current section based on the normalized current section measurement information and distribution network topology information obtained, and inputting the graph data of the current section into the message passing graph neural network model based on offline training completed in step 3, to obtain the state estimation value of the system node voltage amplitude and branch phase angle difference of the current section.

[0088] Furthermore, the output of the distribution network topology identification model established in step 1 is the disconnection state of the line, as follows:

[0089] (1.1) Switch line status

[0090]

[0091] In the formula, a i represents the state of the i-th switch line, 0 means the line is disconnected, 1 means the line is closed; y represents the output of the online neural network, and λ1 and λ2 are set to constant values; when the output of the neural network exceeds λ2, the line switch is considered to be closed; otherwise, the line switch is considered to be open; in this scenario, λ1=λ2=0.5;

[0092] (1.2) Measurement equation

[0093] The measurement equation of historical measurement information can be expressed as follows:

[0094] zh(x)=e

[0095] Where z represents the measurement vector, x represents the state variable vector, h(·) represents the AC or DC power flow equation, and e represents the measurement residual vector;

[0096] For AC systems, the measurements include node voltage amplitude measurement, node active injection power measurement, node reactive injection power measurement, branch active power measurement, and branch reactive power measurement. The power flow equation is as follows:

[0097]

[0098] θ ii' =θ i -θ i' {i,i'}∈N AC

[0099] Where P AC,ii’ and Q AC,ii’ Represent the branch active and reactive power of the branch connected to the i-th AC node and the i'th AC node, respectively, V i and V i’ Represent the voltage amplitude of the i-th AC node and the i'th AC node, θ ii’ represents the phase angle difference between the branches connected to the ith AC node and the i'th AC node, θ i and θ i’ Represent the phase angles of the ith AC node and the i'th AC node, P AC,i and Q AC,i They represent the active injection power and reactive injection power of the i-th AC node, respectively. and They represent the branch active power measurement error and branch reactive power measurement error of the branch connected to the i-th AC node and the i'th AC node, respectively. and They represent the node active power injection measurement error and node reactive power injection measurement error of the i-th AC node, respectively. AC represents the set of nodes in the AC system, G ii’ and B ii’ Respectively represent the real and imaginary parts of the admittance matrix of the branch connected to the ith AC node and the i'th AC node;

[0100] (1.3) The specific mechanism of hedging back propagation is as follows

[0101]

[0102] h 0 = z = [V,P node ,Q node ,P line ,Q line ]

[0103] In the formula, L represents the total number of layers of the neural network, w l and b l is the weight matrix and bias vector of the hidden layer l, Θ l and α l is the weight matrix and weight coefficient of the lth layer, represents the output of layer l, h l represents the activation value of the lth layer, h 0 Represents the initial activation value of the model, including voltage amplitude measurement V, node active power injection measurement P node and reactive power injection measurement Q node , the active power flow of the line is measured P line and reactive power flow measurement Q line , F(z) represents the output of the neural network.

[0104] Furthermore, in step 2, the graph data set is constructed by abstracting the electrical components of the distribution network system into a graph structure and normalizing the data. The specific process is as follows:

[0105] (2.1) Node definition and feature extraction

[0106] In the distribution network, power buses, load nodes, and distributed power sources are defined as nodes in the graph. The voltage amplitude and power state information of each node are input into the neural network as the feature vector of the node. The specific extraction steps are as follows:

[0107] Each bus and load node in the distribution network corresponds to a node in the diagram;

[0108] According to the data obtained from the SCADA system or smart meter, the voltage amplitude, active power and reactive power of each node are extracted, and the node characteristics are represented as a vector:

[0109] x i =[V i ,P i ,Q i ]

[0110] V i ∈V

[0111] P i ∈P node

[0112] Q i ∈Q node

[0113] In the formula, x i represents the characteristics of node i, V i represents the voltage amplitude measurement of node i, P i ,Q i denote the active and reactive power measurements of node i respectively;

[0114] (2.2) Edge definition and feature extraction

[0115] eg ij =[P ij ,Q ij ]

[0116] P ij ∈P line

[0117] Q ij ∈Q line

[0118] In the formula, eg ij represents the characteristics of edge ij, P ij represents the active power measurement of edge ij, Q ij represents the reactive power measurement of edge ij;

[0119] (2.3) Constructing a graph dataset

[0120] Graph data construction, based on the network structure obtained by topology recognition, generates a graph G(V,E), where V is a node set, representing the busbars and load nodes in the distribution network; E is an edge set, representing the power lines in the distribution network;

[0121] Node feature matrix, which stacks the feature vectors of all nodes into a matrix X, whose size is |V|×d, where |V| is the number of nodes and d is the dimension of the node feature;

[0122] The edge feature matrix stacks the feature vectors of each edge into a matrix E of size |E|×g, where |E| is the number of edges and g is the feature dimension of each edge;

[0123] (2.4) Data normalization

[0124]

[0125] In the formula, u i ' is the normalized historical measurement value of the ith i is the ith original historical measurement value, min(·) is the minimum value in a set of data, and max(·) is the maximum value in a set of data.

[0126] Furthermore, in step 3, the graph data is used for training, and the historical measurement data of the node voltage amplitude and the branch power are read, and the measurement data is collected by the SCADA measurement device; the read historical measurement is input into the message passing neural network model for training, and the target output of the model is the node voltage amplitude and the branch phase angle difference;

[0127] The message passing neural network is defined as follows:

[0128] For the i-th node, the following expression is given:

[0129]

[0130] In the formula, represents the information received by node i at time t+1; N(i) represents all neighboring nodes of node i; is the state of node i at time t; e ij is the edge feature between node i and node j; is the node message passing function at time t; this formula indicates that the information received by node i comes from and the state of the adjacent node j at time t and the edge features e connecting them ij ;Once the information is collected, the node status is updated;

[0131]

[0132] in, is the node update function, which takes and As input, generate the node state at time t+1 The node message passing function is as follows:

[0133]

[0134] in, is the learnable parameter matrix; is the activation function of the fully connected layer of the node at time t, is the edge feature of edge ij at time t;

[0135] Based on the above node modeling, the edges in the graph are modeled. The specific formula is as follows:

[0136]

[0137] in, represents the information received by edge ij at time t+1, and is the edge message passing function and state update function at time t, represents the state of edge ij at time t+1, and the edge message passing function is as follows:

[0138]

[0139] in, is the learnable parameter matrix, is the activation function of the fully connected layer at time t;

[0140] The final output of the model is:

[0141]

[0142] Where y represents the final output vector of the model, T represents the last time step of the message passing phase, R(·) is a fully connected layer whose parameters are determined by network training, and the state update function uses the gated recurrent unit GRU:

[0143]

[0144] Furthermore, in step 4, the target output of the model is the voltage amplitude of the node and the branch phase angle difference. The specific conversion formula of the branch phase angle difference and the node phase angle is as follows:

[0145] θ=(B T B) -1 B T Δθ

[0146] Where θ is the node voltage phase angle estimation vector, Δθ is the branch phase angle difference estimation vector, and B is the node-branch association matrix.

[0147] Case Analysis

[0148] 1) Example description

[0149] The test system of the present invention is a 33-node three-phase balanced and 123-node three-phase unbalanced distribution network. The system structure is as follows: Figure 2 and Figure 3 As shown. In the 33-node system, nodes 18, 22, 25 and 33 are connected to distributed photovoltaics, and in the 123-node system, nodes 152, 197, 49 and 250 are connected to distributed photovoltaics. The topological configuration and measurement configuration of the above two test systems are as follows:

[0150] Table 1 Different test system topology configurations

[0151]

[0152] Table 2 Measurement configuration of different test systems

[0153]

[0154] 2) Distribution network topology identification model

[0155] First, the distribution network topology identification model is trained based on historical data. Figure 4 The loss function curves of the two test systems during the topology model training process are given. Topologies 1-4 are trained as historical known topologies, and the rest are new topologies to demonstrate the model recognition ability after dynamic adjustment of model training. It can be seen that as the training progresses, the loss of both models gradually decreases. The enhanced 33-bus system converges stably after 30 cycles, while the 123-bus system converges after 20 cycles. The test shows that after training with topo1-4, ONN can not only accurately recognize the trained topologies, but also show a certain ability in recognizing new topologies: the recognition accuracy of topo6 and topo7 reaches 100%, and the recognition accuracy of topo8 reaches 99%.

[0156] When the topology changes, the proposed training model can dynamically adjust the weight parameters of each layer, so that the model can quickly adapt to the new topological data. The model is configured with 5 layers, and the weight of each layer is initialized to 0.2. It can be seen that as the training progresses, the model adjusts the weight of each layer and finally stabilizes at Figure 5 By adjusting the model parameters, the model can quickly adapt to new topological structures and achieve accurate topological recognition for untrained topologies.

[0157] 3) Comparison of the algorithm of the present invention with other methods

[0158] There are two main comparison algorithms, WLS and deep neural network (DNN). Two comparison algorithms and the method proposed in this invention are used in each time section. The average absolute error (Average Absolute Error, AAE) is used to estimate the accuracy:

[0159]

[0160] Where: n is the number of state variables; x real represents the true value of power flow calculation, is the predicted output value of the model.

[0161] As shown in Table 3, in the 33-bus system, the proposed MPNN method has higher estimation accuracy compared with the WLS and DNN algorithms. In different topologies, the error value of MPNN is always low, indicating its effectiveness and accuracy in this system; compared with the traditional WLS method, the proposed MPNN method has a slightly lower estimation accuracy for the voltage amplitude in the IEEE 123 bus system, but still meets the required accuracy (within 1.8e-3). However, for the voltage phase angle estimation, the accuracy of the MPNN method is slightly higher than that of the WLS method. Figure 6 The estimation results of 50 consecutive time intervals of topo 3 in the IEEE 123 bus system are given.

[0162] Table 3 Comparison of state estimation errors of different methods under different test systems

[0163]

[0164] 4) Computational efficiency

[0165] There are many nodes in the distribution network, and they are scattered, so computational efficiency and time are one of the main criteria for ensuring the actual engineering application of the algorithm. As shown in Table 4, in a 33-node system, the estimation time of the WLS algorithm is 0.0824 seconds, while the estimation time of the Proposed algorithm is only 0.0249 seconds, which is an 81.92% improvement in efficiency. For a 123-node system, the computational time of WLS is significantly improved to 0.6873 seconds, while the proposed method only takes 0.0728 seconds, an improvement of 91.59%. These results show that as the scale of the system increases, the proposed method has a significant advantage in estimation efficiency, especially in large-scale systems, where the time growth of the proposed algorithm is much shorter than that of the WLS algorithm.

[0166] Table 4 Computation time of different state estimation methods

[0167]

Claims

1. A distribution network topology identification and state estimation method based on message passing neural network and online deep learning, characterized in that: The following steps are involved: Step 1: Establish a distribution network topology identification model, and use smart meters or SCADA systems in the distribution network to collect voltage amplitude and power flow data of the distribution network in real time; The real-time collected measurement data is input into the online neural network model. After the model is trained with historical data, it identifies the current topology based on the input voltage and power data. The hedge back propagation mechanism is introduced into the neural network. By weighted summarizing the output results of each layer, the network depth is dynamically adjusted to ensure convergence in the face of different topology changes and identify the accurate topology. Step 2: Construct a graph dataset, represent the nodes in the distribution network as nodes in the graph, and represent the lines in the distribution network as edges in the graph; use the node voltage amplitude, power injection characteristics and line power as edge characteristics and the topology information obtained by the topology identification model in step 1 to construct a graph dataset, and perform data normalization. This dataset will be used as training data for the message passing neural network; Step 3: Establish a message passing neural network distribution network state estimation model, use the constructed graph data set to perform offline training of the message passing neural network, compare the state estimation values ​​of the system node voltage amplitude and branch phase angle difference output by the message passing graph neural network model with the true values ​​during the training process, and calculate the loss function, update the weights and parameters of the message passing graph neural network model through the back propagation algorithm, and obtain the message passing graph neural network model suitable for different topologies through multiple iterations; Step 4, using the message passing neural network distribution network state estimation model to perform estimation, including: constructing the graph data of the current section based on the normalized current section measurement information and distribution network topology information obtained, and inputting the graph data of the current section into the message passing graph neural network model based on offline training completed in step 3, to obtain the state estimation value of the system node voltage amplitude and branch phase angle difference of the current section.

2. A distribution network topology identification and state estimation method based on message passing neural network and online deep learning according to claim 1, characterized in that: The output of the distribution network topology identification model established in step 1 is the disconnection state of the line, which is as follows: (1.1) Switch line status In the formula, a i Represents the state of the i-th switch line, 0 means the line is open, 1 means the line is closed; y represents the output of the online neural network, and λ1 and λ2 are set to constant values; when the output of the neural network exceeds λ2, the circuit breaker is considered closed; otherwise, the circuit breaker is considered open; in this scenario, λ1=λ2=0.5; (1.2) Measurement equation The measurement equation of historical measurement information can be expressed as follows: zh(x)=e Where z represents the measurement vector, x represents the state variable vector, h(·) represents the AC or DC power flow equation, and e represents the measurement residual vector; For AC systems, the measurements include node voltage amplitude measurement, node active injection power measurement, node reactive injection power measurement, branch active power measurement, and branch reactive power measurement. The power flow equation is as follows: i ii' =θ i -θ i' {i,i'}∈N AC Where P AC,ii’ and Q AC,ii’ Represent the branch active and reactive power of the branch connected to the i-th AC node and the i'th AC node, respectively, V i and V i’ Represent the voltage amplitude of the i-th AC node and the i'th AC node, θ ii’ represents the phase angle difference between the branches connected to the ith AC node and the i'th AC node, θ i and θ i’ Represent the phase angles of the ith AC node and the i'th AC node, P AC,i and Q AC,i They represent the active injection power and reactive injection power of the i-th AC node, respectively. and They represent the branch active power measurement error and branch reactive power measurement error of the branch connected to the i-th AC node and the i'th AC node, respectively. and They represent the node active power injection measurement error and node reactive power injection measurement error of the i-th AC node, respectively. AC represents the set of nodes in the AC system, G ii’ and B ii’ Respectively represent the real and imaginary parts of the admittance matrix of the branch connected to the ith AC node and the i'th AC node; (1.3) The specific mechanism of hedging back propagation is as follows h 0 =z=[V,P node ,Q node ,P line ,Q line ] In the formula, L represents the total number of layers of the neural network, w l and b l is the weight matrix and bias vector of the hidden layer l, Θ l and α l is the weight matrix and weight coefficient of the lth layer, represents the output of layer l, h l represents the activation value of the lth layer, h 0 Represents the initial activation value of the model, including voltage amplitude measurement V, node active power injection measurement P node and reactive power injection measurement Q node , the active power flow of the line is measured P line and reactive power flow measurement Q line , F(z) represents the output of the neural network.

3. A distribution network topology identification and state estimation method based on message passing neural network and online deep learning according to claim 2, characterized in that: In step 2, the graph dataset is constructed by abstracting the electrical components of the distribution network system into a graph structure and normalizing the data. The specific process is as follows: (2.1) Node definition and feature extraction In the distribution network, power buses, load nodes, and distributed power sources are defined as nodes in the graph. The voltage amplitude and power state information of each node are input into the neural network as the feature vector of the node. The specific extraction steps are as follows: Each bus and load node in the distribution network corresponds to a node in the diagram; According to the data obtained from the SCADA system or smart meter, the voltage amplitude, active power and reactive power of each node are extracted, and the node characteristics are represented as a vector: x i =[V i ,P i ,Q i ] In i ∈V P i ∈P node Q i ∈Q node In the formula, x i represents the characteristics of node i, V i represents the voltage amplitude measurement of node i, P i ,Q i denote the active and reactive power measurements of node i respectively; (2.2) Edge definition and feature extraction eg ij =[P ij ,Q ij ] P ij ∈P line Q ij ∈Q line In the formula, eg ij represents the characteristics of edge ij, P ij represents the active power measurement of edge ij, Q ij represents the reactive power measurement of edge ij; (2.3) Constructing a graph dataset Graph data construction, based on the network structure obtained by topology recognition, generates a graph G(V,E), where V is a node set, representing the busbars and load nodes in the distribution network; E is an edge set, representing the power lines in the distribution network; Node feature matrix, which stacks the feature vectors of all nodes into a matrix X, whose size is |V|×d, where |V| is the number of nodes and d is the dimension of the node feature; The edge feature matrix stacks the feature vectors of each edge into a matrix E of size |E|×g, where |E| is the number of edges and g is the feature dimension of each edge; (2.4) Data normalization In the formula, u i ' is the normalized historical measurement value of the ith i is the ith original historical measurement value, min(·) is the minimum value in a set of data, and max(·) is the maximum value in a set of data.

4. A distribution network topology identification and state estimation method based on message passing neural network and online deep learning according to claim 3, characterized in that: In step 3, the graph data is used for training, and the historical measurement data of the node voltage amplitude and the branch power are read, and the measurement data is collected by the SCADA measurement device; the read historical measurement is input into the message passing neural network model for training, and the target output of the model is the node voltage amplitude and the branch phase angle difference; The message passing neural network is defined as follows: For the i-th node, the following expression is given: In the formula, represents the information received by node i at time t+1; N(i) represents all neighboring nodes of node i; is the state of node i at time t; e ij is the edge feature between node i and node j; is the node message passing function at time t; this formula indicates that the information received by node i comes from and the state of the adjacent node j at time t and the edge features e connecting them ij ;Once the information is collected, the node status is updated; in, is the node update function, which takes and As input, generate the node state at time t+1 The node message passing function is as follows: in, is the learnable parameter matrix; is the activation function of the fully connected layer of the node at time t, is the edge feature of edge ij at time t; Based on the above node modeling, the edges in the graph are modeled. The specific formula is as follows: in, represents the information received by edge ij at time t+1, and is the edge message passing function and state update function at time t, represents the state of edge ij at time t+1, and the edge message passing function is as follows: in, is the learnable parameter matrix, is the activation function of the fully connected layer at time t; The final output of the model is: Where y represents the final output vector of the model, T represents the last time step of the message passing phase, R(·) is a fully connected layer whose parameters are determined by network training, and the state update function uses the gated recurrent unit GRU:

5. A distribution network topology identification and state estimation method based on message passing neural network and online deep learning according to claim 4, characterized in that: In step 4, the target output of the model is the voltage amplitude of the node and the branch phase angle difference. The specific conversion formula of the branch phase angle difference and the node phase angle is as follows: θ=(B T B) -1 B T Dth Where θ is the node voltage phase angle estimation vector, Δθ is the branch phase angle difference estimation vector, and B is the node-branch association matrix.

Images