Radial power distribution network topology modeling method and device based on intelligent electric meter data
Through smart meter data, the partial least squares method and deep neural network are used to solve the problem of difficult to obtain the topology structure of medium and low voltage distribution networks, and accurate distribution network modeling and line impedance estimation are achieved, which improves economic benefits and accurate analysis. sex.
Patent Information
- Application Number
- CN202410988698.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-29
AI Technical Summary
The topological structure and parameters of medium and low voltage distribution networks are difficult to accurately obtain. The existing methods rely on prior information or high equipment, which makes it difficult to model distribution networks, especially the topological modeling of radial distribution networks.
Based on smart meter data, the partial least squares method is used to obtain the sensitivity of the node power-voltage, establish a sensitivity matrix, identify the topological structure, and calculate the current data through a deep neural network, combine it with iterative solution method to accurately estimate the line impedance to realize the distribution network Accurate modeling.
No additional measurement equipment is required, and only relying on smart meter data is improved, which improves the accuracy and economicality of distribution network topology modeling and improves the accuracy of trend analysis.
Smart Images

Figure CN119051035A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of distribution network state estimation, and in particular relates to a radial distribution network topology modeling method and device based on smart meter data. Background Art
[0002] With the continuous increase in the penetration rate of new energy in the distribution network, the widespread access of distributed energy, mainly photovoltaic, energy storage and electric vehicles, has seriously affected the safe and stable operation of the distribution network. Unlike the existing data-driven methods that only involve power flow calculation, the distribution network topology modeling of the present invention includes topology identification and impedance estimation. Obtaining the distribution network topology structure and the impedance of the distribution line is a prerequisite for power flow analysis, state estimation, line loss assessment and fault location, which is also irreplaceable by the existing data-driven power flow calculation method.
[0003] For medium and low voltage distribution networks, accurate grid structures and parameters are usually difficult to obtain accurately, and relying on manpower to obtain topological structures and parameters is neither practical nor economical. Therefore, accurate modeling of distribution networks is difficult. At present, with the continuous promotion of smart meters and the continuous improvement of data acquisition technology, distribution network modeling based on historical household measurement data is particularly necessary. However, existing topological modeling methods mostly rely on prior topological information, that is, it is necessary to learn the existing topological structure data of the distribution network, or expensive monitoring equipment is required for topological modeling. The analysis of more economical smart meter data is generally limited to node voltage prediction, etc., and there is still little research on radial distribution network topological modeling. Summary of the invention
[0004] Purpose of the invention: The purpose of the present invention is to provide a radial distribution network topology modeling method based on smart meter data. First, according to the power and voltage data of the smart meter, the partial least squares method is used to obtain the power-voltage sensitivity of each node in the distribution network, and a sensitivity matrix is established; the topological structure of the distribution network is identified according to the sensitivity matrix; a distribution network flow calculation data model based on a deep neural network is established, and the partial derivatives of the node power-voltage are calculated according to the data model; the topological structure obtained by comprehensive identification, as well as the partial derivatives of the node power-voltage, are solved to solve the line impedance of the distribution network; an iterative solution method is used to perform error correction on the obtained line impedance, so as to realize accurate modeling of the distribution network topology based on smart meter data. The present invention is aimed at medium-voltage radial distribution networks without topological information and network parameters, and realizes topological drawing and network parameter estimation only based on smart meter data; at the same time, the line impedance is corrected by the same iterative solution method, and has a high distribution network topology modeling accuracy.
[0005] Technical solution: A radial distribution network topology modeling method based on smart meter data of the present invention comprises the following steps:
[0006] Step 1: Obtain the power and voltage history data of the smart meter, use the partial least squares method to obtain the power-node sensitivity of each node, and construct the power-voltage sensitivity matrix of each node in the distribution network;
[0007] Step 2: Based on the power-voltage sensitivity matrix of each node in the distribution network, the relationship between the distribution network topology and the sensitivity matrix is analyzed;
[0008] Step 3: Identify the topology of the distribution network based on the node power-voltage sensitivity matrix;
[0009] Step 4: Based on the power and voltage historical data of the smart meter, a deep neural network model for distribution network power flow calculation is established to obtain the partial derivative of power-voltage of the distribution network node;
[0010] Step 5: Comprehensively identify the distribution network topology and power-voltage partial derivatives to estimate the distribution network line impedance;
[0011] Step 6: Through the iterative solution method, the error correction of the acquired line impedance is carried out to achieve accurate modeling of the distribution network based on the smart meter data.
[0012] Furthermore, step 1 obtains the power and voltage history data of the smart meter, uses the partial least squares method to obtain the power-node sensitivity of each node, and constructs the power-voltage sensitivity matrix of each node in the distribution network. Specifically:
[0013] The forward-backward substitution power flow calculation equation of the radial distribution network is expressed as:
[0014]
[0015] Where a and b are any two adjacent nodes in the distribution network, U a and U b The voltage values of the front-end node a and the back-end node b are respectively, a-b and x a-b are the resistance and reactance of line ab, p a→b and q a→b are the active and reactive powers flowing from the front-end node a to the rear-end node b through the line ab. From the power flow equation of formula (1), it can be seen that there is a nonlinear mapping relationship between the node voltage and the node power in the distribution network.
[0016] According to the power and voltage data measured by the smart meter, the partial least squares method is used to fit the power-node sensitivity of each node. The node power-voltage sensitivity matrix K is constructed based on the obtained sensitivity values, which is:
[0017]
[0018] Furthermore, step 2 is to analyze the relationship between the topological structure of the distribution network and the sensitivity matrix based on the power-voltage sensitivity matrix of each node in the distribution network. First, based on the characteristics of the radial distribution network, any two nodes i and j have the following three topological relationships: j is the front-end node of i, j is the back-end node of i, and j is the parallel node of i; the sensitivity relationship between node power and voltage is analyzed based on the above three topological relationships. Specifically, the following steps are included:
[0019] Step 2.1, j is the front-end node of i
[0020] The voltage drop between the root node 0 and any node i is expressed as the sum of the voltage drops of each adjacent node on the path from the root node to node i:
[0021]
[0022] Among them, B i is the set of lines from the root node to node i, a and b belong to B i The voltage of the root node is set to a fixed value. If the active power of node i changes slightly by Δp i →0 (reactive power remains unchanged), then the voltage change at nodes i and j is expressed as:
[0023]
[0024] Since j is the front node of i, it is not difficult to get And the sensitivity of the voltages at nodes i and j to the active power at node i is as follows:
[0025]
[0026] Step 2.2, j is the backend node of i
[0027] When j is the back-end node of i, the voltage drop at node j can be divided into the following two parts:
[0028]
[0029] Among them, M a→b =r a-b p a→b +x a-b q a→b , assuming c is the back-end node of i and is adjacent to it, then the voltage drop from i to c should be: U i -U c =M i→c / U i ; Since the power change only occurs at node i, the power flowing from node i to node c remains unchanged, M i→c is a constant, so the voltage change at node c is:
[0030]
[0031] Among them U c ′ and U i ′ are the voltages of nodes c and i after the power change at node i. Similarly, if j is any back-end node of node i, then:
[0032]
[0033] Among them B j -B i is the set of routes from node i to node j, U j-1 is the front node adjacent to node j. When nodes i and j are located on the same branch, the sensitivity is higher than k. ji / k ii It increases as the topological position of node j moves backward;
[0034] Step 2.3, j is the parallel node of i
[0035] When j is a parallel node of i, j and i belong to different branches and have the same front-end connection node d, the sensitivity of the voltage of node j to the power of node i can be expressed as:
[0036]
[0037] Among them, M a→b / U a is the voltage drop from node a to its adjacent node b, which is very small, (1+M a→b / U a U a ′)≈1, and k ji / k ii <1.
[0038] Furthermore, step 3 is specifically as follows: based on the node power-voltage sensitivity matrix, identifying the topology of the distribution network.
[0039] It can be seen from formula (2) that the sensitivity vector of the active power of node i to the voltage of each node can be expressed as K i =[k p1i k p2i …k pni ] T , n is the number of load nodes. Analyze the mutual sensitivity k of power-voltage of each node ji The relationship between the self-sensitivity k ii , define the active power sensitivity ratio matrix as Based on this, the topological structure of the distribution network is determined. The specific steps include:
[0040] Step 3.1: Determine the end nodes of each branch road
[0041] Sensitivity ratio k ji / k ii As the topological position of node j moves away from the root node, it increases. For a node i, if the i-th column vector R i(n×1) The maximum value of is the element in the e-th row, then e is determined to be its terminal node. Each terminal node corresponds to a branch road, and the branch road to which the terminal node e belongs is named branch e;
[0042] Step 3.2: Determine the nodes of each branch road
[0043] Only when j is the backend node of i, k ji / k ii > 1, assuming that node j has been determined to be an end node, if the maximum value of the i-th column vector is R ji And R ji > 1, then i belongs to branch j if there is another terminal node k and R ji >R ki >1, then i belongs to the combined path of branches j and k, but is temporarily classified as branch j during classification;
[0044] For general distribution network conditions, the parallel node ratio k ji / k ii <1, but the following special cases may occur: the impedance of a branch line r d-k 、x d-k Very large, the voltage drop is large, and k ki / k ii >1, so it is also necessary to confirm the node of each branch according to the voltage;
[0045] For node i and its back-end node j, the voltage drop and sensitivity ratio can be approximated as:
[0046]
[0047] Therefore, if i belongs to branch j, we should have On the contrary, if This means that k is a parallel node of i and needs to be excluded;
[0048] For a normal distribution network without distributed generation, the voltage of the same branch decreases as the topological position moves backward, so the nodes belonging to the same branch can be ranked according to the voltage measured in the historical data;
[0049] Step 3.3, select the root node and main path
[0050] The load state of the distribution network can be divided into two cases: S1) All nodes are load nodes, then the power sensitivity ratio of each node i to the voltage sensitivity ratio of the root node r will be k ir / k rr ≥1, so if the sensitivity is greater than the minimum element min(R r(n×1) )=1, then node r is selected as the root node of the distribution network; S2) only the branch road contains load nodes, at this time, the sensitivity ratio matrix R is the minimum element min(R i(n×1) )<1, a measurable root node needs to be selected. The measurable root node r should be the node with the least elements that make its column vector less than 1. The standard of the root node r is:
[0051]
[0052] Finally, the branch road to which the root node r belongs is determined as the main road;
[0053] Step 3.4: Select connection nodes and paths
[0054] Each node has been classified into branches and sorted. For node i that belongs to the combined branch of branch j and branch k, it has two branch paths, which can be recorded as:
[0055]
[0056] Where O(i) is the set of terminal nodes in the path of node i, o(i) is the number of paths of node i, for a node l belonging to branch j, if:
[0057]
[0058] This means that branch k is connected to branch j via connection node l.
[0059] Furthermore, step 4 is specifically as follows: based on the power and voltage historical data of the smart meter, a fully connected deep neural network model for distribution network power flow calculation is established, and the partial derivatives of power-voltage of the distribution network nodes are obtained accordingly.
[0060] A fully connected deep neural network consists of multiple layers of neurons, which can be expressed as:
[0061] Y pf =f (m) (f (m-1) (…f (1) (X pf )) (16)
[0062]
[0063]
[0064] Among them, X pf and Y pf are the input of the fully connected deep neural network and the output vector of the last layer, respectively. (l) represents the feedforward transfer function of the l-th layer of neurons, represents the input of this layer, W l and b l Represents the weight matrix and bias vector of the l-th layer of neurons to the next layer of neurons, R l Represents the activation function of this layer. Specifically, the whole function takes the power matrix [P,Q] of each node as the training input and the voltage amplitude [U] of each node as the training output;
[0065] Taking the m-layer fully connected neural network as an example, the node power and voltage of the distribution network are expressed as:
[0066] Y=R m (W m R m-1 (......(W2R1(W1X+b1)+b2)+......+b m-1 )+b m ) (19)
[0067]
[0068] Among them, μ and σ are parameters for data normalization, and the subscripts U and PQ represent voltage and active power and reactive power, respectively. Based on equations (19) to (21), the partial derivative of input active power with respect to output voltage is obtained:
[0069]
[0070] Furthermore, step 5 is to comprehensively identify the distribution network topology and power-voltage partial derivatives to estimate the distribution network line impedance. The details are as follows:
[0071] For the line between adjacent nodes a and b in the radial distribution network, the relationship between voltage, impedance and power is shown in equation (1). If the active power of the rear node b has a small change Δp b →0, then the new relationship between the nodes is:
[0072]
[0073] Combining (1) and (23), the resistance of line segment ab is:
[0074]
[0075] When each change Δp b ,ΔU a ,ΔU b→0, the difference ΔU a / Δp b It can be approximated as a partial derivative Ignoring its higher-order terms, the impedance between nodes a and b can be expressed as:
[0076]
[0077] Furthermore, step 6 is: by using an iterative solution method, error correction is performed on the acquired line impedance to achieve accurate modeling of the distribution network based on smart meter data. The specific steps are as follows:
[0078] Combining the line impedance obtained by formula (25) with the identified topological structure, the power flow equation is established by forward-backward substitution method. The Jacobian matrix shown in formula (26) is established based on the power flow equation:
[0079]
[0080] Among them, Δr i~j With Δx i~j is the impedance correction value to be solved iteratively, and the left side of the equal sign is the deviation between the power flow calculated voltage and the actual voltage, ΔU i.q With ΔU i.p is the voltage error component that takes into account the effects of resistance and reactance.
[0081] Since the number of variables on both sides of the Jacobian matrix must be equal, and the number of impedance values to be solved is 2n, and the number of voltage values is n, the following processing is performed: Assuming that the voltage of the adjacent front-end node of node i is accurate, the forward and backward method of power flow calculation shows that the calculation result of node i is:
[0082]
[0083] Then the voltage difference due to impedance error can be expressed as:
[0084]
[0085] The voltage error can be expressed as the sum of the calculated differences caused by the resistance error, that is, the reactance error, and its individual components can be expressed as:
[0086]
[0087] Among them, ΔU p is the voltage error component caused by the resistance error, ΔU q is the voltage error component caused by the reactance error, and p and q are the active and reactive powers injected into the i node from the i-1 node, respectively.
[0088] For the load and voltage status in a certain historical data, the above iterative solution is used and the voltage MSE value is used as the evaluation index to iterate the initial impedance to the target accuracy to obtain the final impedance value. The specific steps are as follows:
[0089] 1) Substitute the initial impedance value into the power flow calculation program, solve the voltage of each node, and compare it with the historical voltage data to obtain the voltage error vector [ΔU1ΔU2…ΔU n ] T ;
[0090] 2) According to formula (29), the voltage error component is expanded into resistance and reactance components and substituted into formula (26). At this time, the voltage error component on the left side of the equal sign of formula (26) and the Jacobian matrix on the right side of the equal sign are known, and the resistance and reactance correction vectors are to be calculated;
[0091] 3) Solve the resistance and reactance correction vectors and superimpose them with the initial impedance value, and substitute the correction results into the power flow calculation program again to solve the MSE value of the corrected voltage error;
[0092] 4) Repeat the above steps until the voltage error MSE value reaches the target accuracy.
[0093] The present invention also discloses a radial distribution network topology modeling device based on smart meter data, comprising:
[0094] The sensitivity matrix construction module uses the partial least squares method to fit the power-node sensitivity of each node based on the historical power and voltage data measured by the smart meter, and constructs the sensitivity matrix accordingly.
[0095] The topology drawing module analyzes the mathematical relationship between nodes in the sensitivity matrix based on the power-voltage sensitivity matrix of each node in the distribution network, thereby obtaining the corresponding topological relationship.
[0096] A fully connected deep neural network power flow calculation module is used to establish a fully connected deep neural network. The historical power data of the smart meter is used as input and the historical voltage data is used as output. The fully connected deep neural network is trained to obtain a fully connected deep neural network power flow calculation model.
[0097] The line impedance estimation module is used to solve the impedance parameters of each line section and correct them to obtain more accurate distribution network impedance parameters.
[0098] Furthermore, the radial distribution network topology modeling device also includes an analysis and evaluation module for analyzing and evaluating the impedance estimation results and the correction effects, and also analyzing the voltage estimation accuracy based on the method of the present invention.
[0099] Compared with the prior art, the present invention has the following beneficial effects:
[0100] In order to solve the problems of variable topological structures of current medium and low voltage distribution networks and difficulty in measuring line parameters, the present invention proposes a method for topological identification based on smart meter data by analyzing the mathematical relationship between power-voltage sensitivity and topological structure.
[0101] The parameter correction method proposed in the present invention estimates and corrects the line parameters according to the identified topological structure, further improving the accuracy of the topological modeling of the distribution network and the accuracy of the distribution network flow analysis based on the smart meter data. The present invention only needs to know the power and voltage data of the smart meter, without the need for additional measurement equipment, and can bring significant economic benefits to the informatization of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS
[0102] Figure 1 Flow chart of the present invention
[0103] Figure 2 It is the topological structure diagram of radial distribution network;
[0104] Figure 3 This is the topology diagram of the IEEE-33 system of the example system of Example 1.
[0105] Figure 4 The first impedance identification result of the example system under the method described in Example 1;
[0106] Figure 5 Iteration effect of the method described in Example 1;
[0107] Figure 6 Comparison of the accuracy before and after error correction under the method described in Example 1; DETAILED DESCRIPTION
[0108] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0109] Embodiment 1:
[0110] like Figure 1 A radial distribution network topology modeling method based on smart meter data is shown in FIG. Figure 2 shown.
[0111] The following steps are involved:
[0112] Step 1: Obtain the historical power and voltage data of the smart meter, use the partial least squares method to obtain the power-node sensitivity of each node, and construct the power-voltage sensitivity matrix of each node in the distribution network.
[0113] by Figure 3Taking the IEEE 33-node system shown as an example, the load data of all nodes in the system are selected from an actual distribution network in my country in 2021. The historical data (P, Q) measured by the smart meter are substituted into the IEEE 33-node system (15-minute interval, 96 groups / day, 365 days) for flow calculation to obtain the voltage value of each node, and obtain the historical data set {P, Q, U}.
[0114] The forward-backward substitution power flow calculation equation of the radial distribution network is expressed as:
[0115]
[0116] Where a and b are any two adjacent nodes in the distribution network, U a and U b The voltage values of the front-end node a and the back-end node b are respectively, a-b and x a-b are the resistance and reactance of line ab, p a→b and q a→b are the active and reactive powers flowing from the front-end node a to the rear-end node b through the line ab. From the power flow equation of formula (1), it can be seen that there is a nonlinear mapping relationship between the node voltage and the node power in the distribution network.
[0117] According to the power and voltage data measured by the smart meter, the partial least squares method is used to fit the power-node sensitivity of each node. The node power-voltage sensitivity matrix K is constructed based on the obtained sensitivity values, which is:
[0118]
[0119] Step 2: Based on the power-voltage sensitivity matrix of each node in the distribution network, analyze the relationship between the distribution network topology and the sensitivity matrix.
[0120] Based on the characteristics of the radial distribution network, any two nodes i and j have the following three topological relationships: j is the front-end node of i, j is the back-end node of i, and j is the parallel node of i. The sensitivity relationship between node power and voltage is analyzed based on the above three topological relationships, which specifically includes the following steps:
[0121] Step 2.1, j is the front-end node of i
[0122] The voltage drop between the root node 0 and any node i is expressed as the sum of the voltage drops of each adjacent node on the path from the root node to node i:
[0123]
[0124] Among them, B i is the set of lines from the root node to node i, a and b belong to B iThe voltage of the root node is set to a fixed value. If the active power of node i changes slightly by Δp i →0 (reactive power remains unchanged), then the voltage change at nodes i and j is expressed as:
[0125]
[0126] Since j is the front node of i, it is not difficult to get And the sensitivity of the voltages at nodes i and j to the active power at node i is as follows:
[0127]
[0128] Step 2.2, j is the backend node of i
[0129] When j is the back-end node of i, the voltage drop at node j can be divided into the following two parts:
[0130]
[0131] Among them, M a→b =r a-b p a→b +x a-b q a→b , assuming c is the back-end node of i and is adjacent to it, then the voltage drop from i to c should be: U i -U c =M i→c / U i ; Since the power change only occurs at node i, the power flowing from node i to node c remains unchanged, M i→c is a constant, so the voltage change at node c is:
[0132]
[0133] Among them U c ′ and U i ′ are the voltages of nodes c and i after the power change at node i. Similarly, if j is any back-end node of node i, then:
[0134]
[0135] Among them B j -B i is the set of routes from node i to node j, U j-1 is the front node adjacent to node j. When nodes i and j are located on the same branch, the sensitivity is higher than k. ji / k ii It increases as the topological position of node j moves backward;
[0136] Step 2.3, j is the parallel node of i
[0137] When j is a parallel node of i, j and i belong to different branches and have the same front-end connection node d, the sensitivity of the voltage of node j to the power of node i can be expressed as:
[0138]
[0139] Among them, M a→b / U a is the voltage drop from node a to its adjacent node b, which is very small, (1+M a→b / U a U a ′)≈1, and k ji / k ii <1.
[0140] Implement the above steps and perform topology identification for IEEE-33. The results are as follows: Figure 3 The topology is consistent.
[0141] Step 3: Identify the topology of the distribution network based on the node power-voltage sensitivity matrix;
[0142] It can be seen from formula (2) that the sensitivity vector of the active power of node i to the voltage of each node can be expressed as K i =[k p1i k p2i …k pni ] T , n is the number of load nodes. Analyze the mutual sensitivity k of power-voltage of each node ji The relationship between the self-sensitivity k ii , define the active power sensitivity ratio matrix as Based on this, the topological structure of the distribution network is determined. The specific steps include:
[0143] Step 3.1: Determine the end nodes of each branch road
[0144] Sensitivity ratio k ji / k ii As the topological position of node j moves away from the root node, it increases. For a node i, if the i-th column vector R i(n×1) The maximum value of is the element in the e-th row, then e is determined to be its terminal node. Each terminal node corresponds to a branch road, and the branch road to which the terminal node e belongs is named branch e;
[0145] Step 3.2: Determine the nodes of each branch road
[0146] Only when j is the backend node of i, k ji / k ii > 1, assuming that node j has been determined to be an end node, if the maximum value of the i-th column vector is Rji And R ji > 1, then i belongs to branch j if there is another terminal node k and R ji >R ki >1, then i belongs to the combined path of branches j and k, but is temporarily classified as branch j during classification;
[0147] For general distribution network conditions, the parallel node ratio k ji / k ii <1, but the following special cases may occur: the impedance of a branch line r d-k 、x d-k Very large, the voltage drop is large, and k ki / k ii >1, so it is also necessary to confirm the node of each branch according to the voltage;
[0148] For node i and its back-end node j, the voltage drop and sensitivity ratio can be approximated as:
[0149]
[0150] Therefore, if i belongs to branch j, we should have On the contrary, if This means that k is a parallel node of i and needs to be excluded;
[0151] For a normal distribution network without distributed generation, the voltage of the same branch decreases as the topological position moves backward, so the nodes belonging to the same branch can be ranked according to the voltage measured in the historical data;
[0152] Step 3.3, select the root node and main path
[0153] The load state of the distribution network can be divided into two cases: S1) All nodes are load nodes, then the power sensitivity ratio of each node i to the voltage sensitivity ratio of the root node r will be k ir / k rr ≥1, so if the sensitivity is greater than the minimum element min(R r(n×1) )=1, then node r is selected as the root node of the distribution network; S2) only the branch road contains load nodes, at this time, the sensitivity ratio matrix R is the minimum element min(R i(n×1) )<1, a measurable root node needs to be selected. The measurable root node r should be the node with the least elements that make its column vector less than 1. The standard of the root node r is:
[0154]
[0155] Finally, the branch road to which the root node r belongs is determined as the main road;
[0156] Step 3.4: Select connection nodes and paths
[0157] Each node has been classified into branches and sorted. For node i that belongs to the combined branch of branch j and branch k, it has two branch paths, which can be recorded as:
[0158]
[0159] Where O(i) is the set of terminal nodes in the path of node i, o(i) is the number of paths of node i, for a node l belonging to branch j, if:
[0160]
[0161] This means that branch k is connected to branch j via connection node l.
[0162] 20,000 sets of data are selected from the above historical data, 80% of which are randomly selected for deep neural network (DNN) model training, and 20% are used for testing. During training, {P, Q} is used as input and {U} is used as output. The present invention uses a three-layer DNN network (m=3), and its activation function is Leakyrelu. The parameters of DNN are shown in Table 1.
[0163] Table 1 Deep neural network structure and main parameters
[0164]
[0165] Step 4: Based on the power and voltage historical data of the smart meter, a fully connected deep neural network model for distribution network power flow calculation is established to obtain the partial derivative of power-voltage of the distribution network node;
[0166] A fully connected deep neural network consists of multiple layers of neurons, which can be expressed as:
[0167] Y pf =f (m) (f (m-1) (...f (1) (X pf )) (16)
[0168]
[0169] Among them, X pf and Y pf are the input of the fully connected deep neural network and the output vector of the last layer, respectively. (l) represents the feedforward transfer function of the l-th layer of neurons, represents the input of this layer, W l and b l Represents the weight matrix and bias vector of the l-th layer of neurons to the next layer of neurons, R lRepresents the activation function of this layer. Specifically, the whole function takes the power matrix [P,Q] of each node as the training input and the voltage amplitude [U] of each node as the training output;
[0170] Taking the m-layer fully connected neural network as an example, the node power and voltage of the distribution network are expressed as:
[0171] Y=R m (W m R m-1 (......(W2R1(W1X+b1)+b2)+......+b m-1 )+b m ) (19)
[0172]
[0173] Among them, μ and σ are parameters for data normalization, and the subscripts U and PQ represent voltage and active power and reactive power, respectively. Based on equations (19) to (21), the partial derivative of input active power with respect to output voltage is obtained:
[0174]
[0175] Step 5: Comprehensively identify the distribution network topology and power-voltage partial derivatives to estimate the distribution network line impedance;
[0176] For the line between adjacent nodes a and b in the radial distribution network, the relationship between voltage, impedance and power is shown in equation (1). If the active power of the rear node b has a small change Δp b →0, then the new relationship between the nodes is:
[0177]
[0178] Combining (1) and (23), the resistance of line segment ab is:
[0179]
[0180] When each change Δp b ,ΔU a ,ΔU b →0, the difference ΔU a / Δp b It can be approximated as a partial derivative Ignoring its higher-order terms, the impedance between nodes a and b can be expressed as:
[0181]
[0182] 100 sets of data are randomly selected from the test set data, and the impedance is estimated by the above method. In order to make the historical data consistent with the actual measurement, 0.2% Gaussian white noise is injected into all the data. The above method is used for multiple identifications, and the average of the multiple identification results is used as the final line resistance and reactance results, as shown in Figure 4 The error is defined as The relative errors between the estimated values and the actual values are shown in Table 2.
[0183] Table 2 Impedance estimation error
[0184]
[0185] Step 6: Through the iterative solution method, the error correction of the acquired line impedance is carried out to achieve accurate modeling of the distribution network based on the smart meter data.
[0186] Combining the line impedance obtained by formula (25) with the identified topological structure, the power flow equation is established by forward-backward substitution method. The Jacobian matrix shown in formula (26) is established based on the power flow equation:
[0187]
[0188] Among them, Δr i~j With Δx i~j is the impedance correction value to be solved iteratively, and the left side of the equal sign is the deviation between the power flow calculated voltage and the actual voltage, ΔU i.q With ΔU i.p is the voltage error component that takes into account the effects of resistance and reactance.
[0189] Since the number of variables on both sides of the Jacobian matrix must be equal, and the number of impedance values to be solved is 2n, and the number of voltage values is n, the following processing is performed: Assuming that the voltage of the adjacent front-end node of node i is accurate, the forward and backward method of power flow calculation shows that the calculation result of node i is:
[0190]
[0191] Then the voltage difference due to impedance error can be expressed as:
[0192]
[0193] The voltage error can be expressed as the sum of the calculated differences caused by the resistance error, that is, the reactance error, and its individual components can be expressed as:
[0194]
[0195] Among them, ΔU p is the voltage error component caused by the resistance error, ΔU qis the voltage error component caused by the reactance error, and p and q are the active and reactive powers injected into the i node from the i-1 node, respectively.
[0196] The impedance error correction is performed for the IEEE-33 node system. The initial impedance is iterated 6 times and the results are shown in Table 3 and Figure 5-Figure 6 The corrected impedance error results and the voltage accuracy of the system are shown.
[0197] Table 3 Impedance error before and after correction
[0198] Impedance error (%) Maximum average value Before correction 2.11 0.52 After correction 1.52 0.21
[0199] Embodiment 2
[0200] This embodiment proposes a radial distribution network topology modeling device based on smart meter data, including:
[0201] The sensitivity matrix construction module uses the partial least squares method to fit the power-node sensitivity of each node based on the historical power and voltage data measured by the smart meter, and constructs the sensitivity matrix accordingly.
[0202] The topology drawing module analyzes the mathematical relationship between nodes in the sensitivity matrix based on the power-voltage sensitivity matrix of each node in the distribution network, thereby obtaining the corresponding topological relationship.
[0203] A fully connected deep neural network power flow calculation module is used to establish a fully connected deep neural network. The historical power data of the smart meter is used as input and the historical voltage data is used as output. The fully connected deep neural network is trained to obtain a fully connected deep neural network power flow calculation model.
[0204] The line impedance estimation module is used to solve the impedance parameters of each line section and correct them to obtain more accurate distribution network impedance parameters.
[0205] The radial distribution network topology modeling device also includes an analysis and evaluation module for analyzing and evaluating the impedance estimation result and the correction effect, and also analyzing the voltage estimation accuracy based on the method of the present invention.
[0206] For the specific implementation scheme of each module of the system of the present invention, please refer to the specific implementation process of the method described in Example 1.
[0207] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0208] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0209] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0210] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0211] The above description is only a preferred specific implementation manner of the present invention. It should be pointed out that the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A radial distribution network topology modeling method based on smart meter data, characterized in that: The steps include: Step 1: Obtain the power and voltage history data of the smart meter, use the partial least squares method to obtain the power-node sensitivity of each node, and construct the power-voltage sensitivity matrix of each node in the distribution network; Step 2: Based on the power-voltage sensitivity matrix of each node in the distribution network, the relationship between the distribution network topology and the sensitivity matrix is analyzed; Step 3: Identify the topology of the distribution network based on the node power-voltage sensitivity matrix; Step 4: Based on the power and voltage historical data of the smart meter, a deep neural network model for distribution network power flow calculation is established to obtain the partial derivative of power-voltage of the distribution network node; Step 5: Comprehensively identify the distribution network topology and power-voltage partial derivatives to estimate the distribution network line impedance; Step 6: Through the iterative solution method, the error correction of the acquired line impedance is carried out to achieve accurate modeling of the distribution network based on the smart meter data.
2. A radial distribution network topology modeling method based on smart meter data according to claim 1, characterized in that: Step 1 is as follows: The forward-backward substitution power flow calculation equation of the radial distribution network is expressed as: Among them, a and b are any two adjacent nodes in the distribution network, U a and U b The voltage values of the front-end node a and the back-end node b are respectively, a-b and x a-b are the resistance and reactance of line ab, p a→b and q a→b are respectively the active and reactive power flowing from the front-end node a to the rear-end node b through the line ab. From the power flow equation of formula (1), it can be seen that there is a nonlinear mapping relationship between the node voltage and the node power in the distribution network; According to the power and voltage data measured by the smart meter, the partial least squares method is used to fit the power-node sensitivity of each node, and the node power-voltage sensitivity matrix K is constructed according to the obtained sensitivity values, which is specifically:
3. A radial distribution network topology modeling method based on smart meter data according to claim 1, characterized in that: Step 2 is to analyze the relationship between the topological structure of the distribution network and the sensitivity matrix based on the power-voltage sensitivity matrix of each node in the distribution network; specifically: based on the characteristics of the radial distribution network, any two nodes i and j have the following three topological relationships: j is the front-end node of i, j is the back-end node of i, and j is the parallel node of i; according to the above three topological relationships, the sensitivity relationship of node power-voltage is analyzed, which specifically includes the following steps: Step 2.1, j is the front-end node of i The voltage drop between the root node 0 and any node i is expressed as the sum of the voltage drops of each adjacent node on the path from the root node to node i: Among them, B i is the set of lines from the root node to node i, a and b belong to B i The voltage of the root node is set to a fixed value. If the active power of node i changes slightly by Δp i →0, the reactive power remains unchanged, then the voltage change of nodes i and j is expressed as: Since j is the front node of i, it is not difficult to get And the sensitivity of the voltages at nodes i and j to the active power at node i is as follows: Step 2.2, j is the backend node of i When j is the back-end node of i, the voltage drop at node j can be divided into the following two parts: Among them, M a→b =r a-b p a→b +x a-b q a→b , assuming c is the back-end node of i and is adjacent to it, then the voltage drop from i to c should be: U i -U c =M i→c / U i ; Since the power change only occurs at node i, the power flowing from node i to node c remains unchanged, M i→c is a constant, so the voltage change at node c is: Among them U c ′ and U i ′ are the voltages of nodes c and i after the power change at node i. Similarly, if j is any back-end node of node i, then: Among them B j -B i is the set of routes from node i to node j, U j-1 is the front node adjacent to node j. When nodes i and j are located on the same branch, the sensitivity is higher than k. ji / k ii It increases as the topological position of node j moves backward; Step 2.3, j is the parallel node of i When j is a parallel node of i, j and i belong to different branches and have the same front-end connection node d, the sensitivity of the voltage of node j to the power of node i can be expressed as: Among them, M a→b / U a is the voltage drop from node a to its adjacent node b, which is very small, (1+M a→b / U a U′ a )≈1, and k ji / k ii <1.
4. A radial distribution network topology modeling method based on smart meter data according to claim 1, characterized in that: Step 3 is specifically as follows: based on the node power-voltage sensitivity matrix, identify the topology of the distribution network; It can be seen from formula (2) that the sensitivity vector of the active power of node i to the voltage of each node can be expressed as K i =[k p1i k p2i …k pni ] T , n is the number of load nodes; analyze the power-voltage mutual sensitivity k of each node ji The relationship between the self-sensitivity k ii , define the active power sensitivity ratio matrix as Based on this, the topological structure of the distribution network is determined; specifically, the following steps are included: Step 3.1: Determine the end nodes of each branch road Sensitivity ratio k ji / k ii As the topological position of node j moves away from the root node, it increases. For a node i, if the i-th column vector R i(n×1) The maximum value of is the element in the e-th row, then e is determined to be its terminal node. Each terminal node corresponds to a branch road, and the branch road to which the terminal node e belongs is named branch e; Step 3.2: Determine the nodes of each branch road Only when j is the backend node of i, k ji / k ii > 1, assuming that node j has been determined to be an end node, if the maximum value of the i-th column vector is R ji And R ji > 1, then i belongs to branch j if there is another terminal node k and R ji >R ki >1, then i belongs to the combined path of branches j and k, but is temporarily classified as branch j during classification; For general distribution network conditions, the parallel node ratio k ji / k ii <1, but the following special cases may occur: the impedance of a branch line r d-k 、x d-k Very large, the voltage drop is large, and k ki / k ii >1, so it is also necessary to confirm the node of each branch according to the voltage; For node i and its back-end node j, the voltage drop and sensitivity ratio can be approximated as: Therefore, if i belongs to branch j, we should have On the contrary, if This means that k is a parallel node of i and needs to be excluded; For a normal distribution network without distributed generation, the voltage of the same branch decreases as the topological position moves backward, so the nodes belonging to the same branch can be ranked according to the voltage measured in the historical data; Step 3.3: Select the root node and main path The load state of the distribution network can be divided into two cases: S1) All nodes are load nodes, then the power sensitivity ratio of each node i to the voltage sensitivity ratio of the root node r will be k ir / k rr ≥1, so if the sensitivity is greater than the minimum element min(R r(n×1) )=1, then node r is selected as the root node of the distribution network; S2) Only the branch road contains the load node, at this time, the sensitivity ratio matrix R is the minimum element min(R i(n×1) )<1, a measurable root node needs to be selected; the measurable root node r should be the node with the least elements that make its column vector less than 1. The standard of the root node r is: Finally, the branch road to which the root node r belongs is determined as the main road; Step 3.4: Select connection nodes and paths Each node has been classified into branches and sorted. For node i that belongs to the combined branch of branch j and branch k, it has two branch paths, which can be recorded as: Where O(i) is the set of terminal nodes in the path of node i, o(i) is the number of paths of node i, for a node l belonging to branch j, if: This means that branch k is connected to branch j via connection node l.
5. A radial distribution network topology modeling method based on smart meter data according to claim 1, characterized in that: Step 4 is as follows: based on the power and voltage history data of the smart meter, a fully connected deep neural network model for distribution network power flow calculation is established, and the partial derivative of power-voltage of the distribution network node is obtained accordingly; A fully connected deep neural network consists of multiple layers of neurons, which can be expressed as: Y pf =f (m) (f (m-1) (...f (1) (X pf )) (16) Among them, X pf and Y pf are the input of the fully connected deep neural network and the output vector of the last layer, respectively. (l) represents the feedforward transfer function of the l-th layer of neurons, represents the input of this layer, W l and b l Represents the weight matrix and bias vector of the l-th layer of neurons to the next layer of neurons, R l Represents the activation function of this layer. Specifically, the whole function takes the power matrix [P,Q] of each node as the training input and the voltage amplitude [U] of each node as the training output; Taking the m-layer fully connected neural network as an example, the node power and voltage of the distribution network are expressed as: Y=R m (W m R m-1 (......(W2R1(W1X+b1)+b2)+......+b m-1 )+b m ) (19) Among them, μ and σ are parameters for data normalization, and the subscripts U and PQ represent voltage and active power and reactive power, respectively. Based on equations (19) to (21), the partial derivative of input active power with respect to output voltage is obtained:
6. A radial distribution network topology modeling method based on smart meter data according to claim 1, characterized in that: Step 5 is to comprehensively identify the distribution network topology and power-voltage partial derivatives to estimate the distribution network line impedance; the details are as follows: For the line between adjacent nodes a and b in the radial distribution network, the relationship between voltage, impedance and power is shown in equation (1). If the active power of the rear node b has a small change Δp b →0, then the new relationship between the nodes is: Combining (1) and (23), the resistance of line segment ab is: When each change Δp b ,ΔU a ,ΔU b →0, the difference ΔU a / Δp b It can be approximated as a partial derivative Ignoring its higher-order terms, the impedance between nodes a and b can be expressed as:
7. A radial distribution network topology modeling method based on smart meter data according to claim 1, characterized in that: Step 6 is: by using an iterative solution method, error correction is performed on the acquired line impedance to achieve accurate modeling of the distribution network based on smart meter data; the specific steps are as follows: Combining the line impedance obtained by formula (25) with the identified topological structure, the power flow equation is established by forward-backward substitution method. The Jacobian matrix shown in formula (26) is established based on the power flow equation: Among them, Δr i~j With Δx i~j is the impedance correction value to be solved iteratively, and the left side of the equal sign is the deviation between the power flow calculated voltage and the actual voltage, ΔU i.q With ΔU i.p is the voltage error component considering the influence of resistance and reactance; Since the number of variables on both sides of the Jacobian matrix must be equal, and the number of impedance values to be solved is 2n, and the number of voltage values is n, the following processing is performed: Assuming that the voltage of the adjacent front-end node of node i is accurate, the forward and backward method of power flow calculation shows that the calculation result of node i is: Then the voltage difference due to impedance error can be expressed as: The voltage error can be expressed as the sum of the calculated differences caused by the resistance error, that is, the reactance error, and its individual components can be expressed as: Among them, ΔU p is the voltage error component caused by the resistance error, ΔU q is the voltage error component caused by the reactance error, p and q are the active and reactive powers injected from the i-1 node into the i node, respectively; For the load and voltage status in a certain historical data, the above iterative solution is used and the voltage MSE value is used as the evaluation index to iterate the initial impedance to the target accuracy to obtain the final impedance value.
8. A radial distribution network topology modeling method based on smart meter data according to claim 7, characterized in that: For the load and voltage state in a certain historical data, the above iterative solution is used and the voltage MSE value is used as the evaluation index to iterate the initial impedance to the target accuracy, that is, to obtain the final impedance value. The specific steps are as follows: the initial impedance value is substituted into the prepared power flow calculation program to solve the voltage of each node and compare it with the historical voltage data to obtain the voltage error vector [ΔU1ΔU2…ΔU n ] T ; Step 6.1, according to formula (29), expand the voltage error component into resistance and reactance components and substitute them into formula (26). At this time, the voltage error component on the left side of the equal sign of formula (26) and the Jacobian matrix on the right side of the equal sign are known, and the resistance and reactance correction vectors are to be calculated; Step 6.2, solve the resistance and reactance correction vectors and superimpose them with the initial impedance value, and substitute the correction results into the power flow calculation program again to solve the MSE value of the corrected voltage error; Step 6.3: Repeat the above steps until the voltage error MSE value reaches the target accuracy.
9. A radial distribution network topology modeling device based on smart meter data, characterized in that: include: The sensitivity matrix construction module uses the partial least squares method to fit the power-node sensitivity of each node based on the historical power and voltage data measured by the smart meter, and then constructs the sensitivity matrix; The topology drawing module analyzes the mathematical relationship between nodes in the sensitivity matrix based on the power-voltage sensitivity matrix of each node in the distribution network, thereby obtaining the corresponding topological relationship; A fully connected deep neural network power flow calculation module is used to establish a fully connected deep neural network. The historical power data of the smart meter is used as input and the historical voltage data is used as output. The fully connected deep neural network is trained to obtain a fully connected deep neural network power flow calculation model. The line impedance estimation module is used to solve the impedance parameters of each line section and correct them to obtain more accurate distribution network impedance parameters.
10. A radial distribution network topology modeling device based on smart meter data according to claim 9, characterized in that: The radial distribution network topology modeling device also includes an analysis and evaluation module for analyzing and evaluating the impedance estimation result and the correction effect, and also analyzing the voltage estimation accuracy based on the method of the present invention.
Citation Information
Cited By
Low-voltage transformer area topology modeling method and system based on hierarchical clustering
CN119627912A
Power distribution network topology intelligent identification method based on deep convolutional network
CN122020098A
Transformer topology identification method and device based on voltage sensitivity matrix reconstruction
CN122869306A
