Active power distribution network line loss situation awareness method and device

Abstract

The invention discloses an active power distribution network line loss situation awareness method and device. The method comprises the following steps: acquiring operation data of an active power distribution network; performing preliminary identification processing based on the operation data and a global topology identification model to obtain a preliminary topology identification result of the active power distribution network; performing partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active power distribution network; wherein the final topology identification result comprises the topology identification result of each partition; based on the final topology identification result, determining an operation factor of each grid-connected main body influencing the line loss of the active power distribution network, and based on the operation factor, obtaining a line loss prediction key index; inputting the line loss prediction key index into a line loss prediction model for prediction to obtain a prediction result of the active power distribution line loss situation; according to the invention, effective perception prediction of the active power distribution network line loss is realized, and the generalization ability of power distribution network line loss situation perception under the access of multiple grid-connected main bodies is improved.

Description

Technical Field

[0001] The invention relates to a method and a device for sensing line loss situation in an active power distribution network, belonging to the technical field of active power distribution networks. Background Art

[0002] With the large-scale access of various flexible load resources to the distribution network, the grid structure and operation mode of the distribution network have undergone great changes. The flow direction has become a two-way flow, and the uncertainty has also greatly increased, making the line loss changes in the distribution network more complicated, bringing new challenges to the economic and low-carbon operation of the distribution network. As a key link in reducing costs and increasing efficiency and helping my country achieve its dual carbon goals, line loss still has many problems such as difficulty in perceiving and analyzing, and urgent need for refined management due to the difficulty in processing massive measurement data and the complex and changeable operation status of the distribution network. The increasing penetration rate of various grid-connected entities such as distributed power sources and electric vehicle charging piles has made the line loss change mechanism in the active distribution network more complicated, and the number of influencing factors has increased, making it difficult for traditional distribution network line loss analysis methods to fully explore the line loss characteristics in the active distribution network environment. Line loss analysis methods suitable for active distribution networks have received widespread attention.

[0003] As a technical means that integrates perception, understanding and prediction, situation awareness technology can improve the ability to collect and analyze multi-source data, accurately understand the current status and trend, improve the overall observability, and help formulate control strategies. Its concept was first proposed in the military field and has now been widely used in risk assessment, intelligent transportation, power system security and other aspects. Situation awareness in power systems can be divided into three levels: situation awareness layer, situation understanding layer and situation prediction layer. Environmental data is extracted in the situation awareness layer, data information is mined and analyzed in the situation understanding layer, and finally the prediction of future development trends is achieved in the situation prediction layer. At present, the application of situation awareness technology in the field of power distribution is being vigorously developed, and it is gradually penetrating into various aspects such as dispatching, control and monitoring of intelligent distribution networks. Using situation awareness technology to improve observability and extract key information from multi-source data can effectively deal with the uncertainty factors of active distribution networks and achieve refined operation and maintenance goals. However, the uncertainty and volatility of the output and load brought by each grid-connected entity have put forward higher requirements on the accuracy of situational awareness. The situational awareness technology directly applied to the field of distribution network line loss control still has many shortcomings. The existing situational awareness technology shows poor generalization ability when facing the flexible access of grid-connected entities and changes in the distribution network environment. It is difficult to fully understand the different coupling mechanisms between the access operation of each grid-connected entity and the line loss, summarize the changing laws of line loss, and the line loss prediction model that lacks consideration of the dynamic correlation characteristics between the operation of the grid-connected entity and the line loss has low prediction accuracy, and cannot truly and effectively reflect the distribution network operation trend and make timely strategic responses to the changes of the grid-connected entity.

[0004] In summary, how to explore the changing characteristics of line loss in active distribution networks and further combine situational awareness technology to achieve effective perception and analysis of line loss in active distribution network environments with complex and changeable operating conditions requires in-depth and systematic research. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for sensing the line loss situation of an active distribution network. The method and device obtain the final topology identification result of the active distribution network based on global topology identification and partition topology identification, and obtain key indicators for line loss prediction. Based on the key indicators, the line loss of the active distribution network is predicted using a constructed line loss prediction model, thereby achieving effective sensing and prediction of the line loss of the active distribution network and improving the generalization capability of the line loss situation perception of the distribution network under the access of multiple grid-connected entities.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a method for actively sensing line loss in a power distribution network, comprising:

[0008] Obtaining the operating data of active distribution networks;

[0009] Performing preliminary identification processing based on the operating data and the global topology identification model to obtain preliminary topology identification results of the active distribution network;

[0010] Performing partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active distribution network; wherein the final topology identification result includes the topology identification results of each partition;

[0011] Based on the final topology identification result, determining the operating factors of each grid-connected entity that affect the line loss of the active distribution network, and based on the operating factors, obtaining key indicators for line loss prediction;

[0012] The key indicators for line loss prediction are input into a line loss prediction model for prediction, and a prediction result of the line loss situation of the active distribution network is obtained.

[0013] Optionally, the preliminary topology identification result is subjected to a partition identification process to obtain a final topology identification result of the active distribution network, including:

[0014] After dividing the preliminary topology identification result into regions, the regions are input into a partition topology identification model to obtain a partition topology identification result;

[0015] If the partition topology identification result is the same as the preliminary topology identification result, the output partition topology identification result is the final topology identification result of the active distribution network;

[0016] If the partition topology identification result is different from the preliminary topology identification result, the loop steps are iteratively executed until the preset conditions are met to obtain the final topology identification result; wherein the loop steps include: after dividing the partition topology identification result into regions, inputting it into the partition topology identification model to obtain the current partition topology identification result; using the current partition topology identification result to update the partition topology identification result; the preset condition is: the current partition topology identification result is the same as the partition topology identification result.

[0017] Optionally, dividing the topology identification result into regions includes:

[0018] Based on the topology identification results, the active distribution network topology is determined and the power flow calculation is performed to obtain the power correction equation;

[0019] The sensitivity matrix is ​​obtained according to the power correction equation, and the electrical distance between each node in the active distribution network and each distributed generation access node is calculated;

[0020] Combining the calculation result of the electrical distance with the hierarchical clustering algorithm, each distributed power source connected to the active distribution network is taken as the cluster center of the sub-partition, and the area is divided based on the partition restriction conditions; the partition restriction conditions include: the number of nodes in each sub-partition is at least 2 and cannot exceed half of the total number of nodes, the nodes in each sub-partition cannot be repeated, and all nodes in the active distribution network are included in each sub-partition.

[0021] Optionally, based on the final topology identification result, determining the operating factors of each grid-connected entity that affect the line loss of the active distribution network includes:

[0022] Based on the final topology identification results of the active distribution network, the line loss changes before and after the multiple grid-connected entities are connected to the active distribution network are calculated by theoretical formula, and the line loss changes before and after the grid-connected entities are connected to the active distribution network are obtained;

[0023] Combined with the line loss variation, the line loss of each grid-connected entity connected to the active distribution network is simulated to determine the line loss variation law caused by each grid-connected entity connected to the active distribution network;

[0024] According to the change law, and in combination with the actual topological structure and operating status of the active distribution network, the operating factors of each grid-connected entity that affect the line loss of the active distribution network are determined; the operating factors include: access location, grid-connected capacity, operating mode, and power factor.

[0025] Optionally, based on the operating factors, key indicators for line loss prediction are obtained, including:

[0026] Combined with the operational factors of each grid-connected entity that affect the line loss of the active distribution network, the correlation characteristics between the grid-connected entity output data, load data and the active distribution network flow data are determined;

[0027] According to the correlation characteristics, an active distribution network line loss prediction index system is established;

[0028] The key indicators for line loss prediction are extracted by using the active distribution network line loss prediction index system and the improved grey correlation analysis method.

[0029] Optionally, the line loss prediction index system includes:

[0030] Line attribute indicators are the basic parameters of the active distribution network, including: trunk line length, total line length, number of distribution transformers, and distribution transformer models;

[0031] Basic operation characteristic indicators are daily operation parameters of active distribution network, including active power supply, reactive power supply, and historical line loss;

[0032] Management factor indicators are non-technical factors that cause line losses, including: equipment aging, meter reading accuracy, and operating years;

[0033] The subject influencing indicators are the operating factors that affect the line loss after each grid-connected subject is connected to the active distribution network, including: grid-connected capacity, access location, operating mode, and power factor.

[0034] Optionally, the improved grey relational analysis method includes:

[0035] The line loss of active distribution network is taken as the reference sequence, and each line loss index in the line loss prediction index system is taken as the comparison sequence. After normalizing the data, the grey correlation coefficient is obtained.

[0036] Based on the grey correlation coefficient, the grey correlation of each line loss index is obtained;

[0037] The values ​​of each line loss index are converted into probability distribution, and the entropy values ​​of each line loss index are weighted averaged to obtain the weight coefficient;

[0038] Combining the grey correlation degree of each line loss index and the weight coefficient, a weighted correlation degree is obtained;

[0039] The indicators are sorted according to the weighted correlation to determine the influence of each line loss indicator and extract the key indicators for line loss prediction.

[0040] Optionally, the formula for the weighted association degree is as follows:

[0041] ;

[0042] ;

[0043] ;

[0044] in: is the weighted correlation, P tk is the proportion of the kth element of line loss index t to the index, n is the number of elements contained in the line loss index, d is the number of indicators, E t is the entropy value of line loss index t, is the weight coefficient of line loss index t; r t is the grey relational degree of line loss index t.

[0045] Optionally, the method for acquiring the line loss prediction model includes:

[0046] Acquire a key indicator data set, wherein the key indicator data set includes a training set and a test set;

[0047] Based on the training set in the key indicator data set, the base learner layer of the line loss prediction model is trained, and the optimal parameter combination of each sub-model is determined to obtain the training result of the base learner layer;

[0048] Based on the training result of the base learner layer, determining the weight distribution of each sub-model of the base learner layer to obtain a trained base learner layer;

[0049] The training result set of the base learner is used as input to train the meta-learner layer of the line loss prediction model to obtain a trained meta-learner layer;

[0050] Based on the trained base learner layer and the trained meta learner layer, a trained line loss prediction model is obtained;

[0051] The test set in the key indicator data set is input into the trained base learner layer, and each sub-model is combined with the weight to obtain the prediction result set of the base learner layer;

[0052] The prediction result set of the base learner layer is input into the trained meta-learner layer to perform a preliminary evaluation of the line loss prediction model performance, thereby obtaining a preliminary evaluation result of the line loss prediction model performance;

[0053] According to the preliminary evaluation results of the line loss prediction model performance, the parameters of the line loss prediction model are adjusted to obtain the line loss prediction model.

[0054] Optionally, based on the Stacking algorithm model, a random forest model, a support vector machine model, a K nearest neighbor model, a gradient boosting decision tree model, and a multi-layer neural network prediction model are selected as sub-models of the base learner layer of the line loss prediction model; and a linear regression model is selected as a sub-model of the meta-learner layer.

[0055] Optionally, the method for determining the optimal parameter combination includes:

[0056] Obtain the parameters that need to be optimized for each sub-model of the base learner layer, and set a finite set of parameter optimization ranges;

[0057] Exhaustively traverse the finite set and obtain the set evaluation indicators that can characterize the prediction effect of the model, the evaluation indicators include: root mean square error, mean absolute error;

[0058] Based on the evaluation index, the parameter combination that achieves the highest prediction accuracy is selected.

[0059] Optionally, the method for determining the weight distribution of each sub-model of the base learner layer includes:

[0060] Based on the K-fold cross-validation method, the training set is divided into K parts, and each sub-model in the base learner layer is trained K times to obtain K trained models;

[0061] The training set is used to perform label prediction output on the trained K models. The accuracy or error of the output results is calculated with the true value label data in the training set to obtain the accuracy or error of each of the K models.

[0062] Determine the weight distribution of the K models in the sub-model according to the proportion of the accuracy or error of each of the K models in the sub-model;

[0063] The K-fold cross-validation method and the following steps are used to determine the weight distribution of each sub-model in the base learner layer.

[0064] In a second aspect, the present invention provides an active distribution network line loss situation awareness device, comprising:

[0065] A data acquisition module is used to obtain the operating data of the active distribution network;

[0066] A global topology identification module, used to perform preliminary identification processing based on the operating data and the global topology identification model to obtain a preliminary topology identification result of the active distribution network;

[0067] A partition topology identification module is used to perform partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active power distribution network; wherein the final topology identification result includes the topology identification results of each partition;

[0068] A key indicator determination module is used to determine the operating factors of each grid-connected entity that affect the line loss of the active distribution network based on the final topology identification result; and based on the operating factors, obtain key indicators for line loss prediction;

[0069] The line loss situation prediction module is used to input the key line loss prediction indicators into the line loss prediction model for prediction, and obtain the prediction result of the active distribution network line loss situation.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] The active distribution network line loss situation perception method proposed in the present invention obtains the final topology identification result (topology space) of the active distribution network through two topology identification processes, and determines the key indicators for line loss prediction based on the final topology identification result; combines the key indicators for line loss prediction and the line loss prediction model to predict the line loss of the active distribution network, and obtains the prediction result of the line loss situation of the active distribution network; realizes the effective perception and prediction of the line loss of the active distribution network, and improves the generalization ability of the distribution network line loss situation perception under the access of multiple grid-connected entities.

[0072] The present invention utilizes the difference characteristics between distribution network areas and combines the global topology identification model with the partitioned topology identification model to improve the topology identification accuracy and speed.

[0073] The present invention establishes a line loss prediction index system through the correlation characteristics between the access of the grid-connected entity and the line loss of the distribution network, uses the improved grey correlation analysis method to extract the key indicators of line loss prediction, and uses the Stacking algorithm improved by parameter optimization and weight allocation to predict the line loss of the active distribution network, and obtains the prediction result of the line loss situation of the active distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor, among which:

[0075] Figure 1 The flowchart of the active distribution network line loss situation awareness method in the embodiment of the present invention is shown;

[0076] Figure 2 The flowchart of active power distribution network topology identification in the embodiment of the present invention is shown;

[0077] Figure 3 A schematic diagram of a mechanism model of a grid-connected entity connected to a distribution network in an embodiment of the present invention is shown;

[0078] Figure 4 The active distribution network line loss prediction model diagram based on the Stacking algorithm in the embodiment of the present invention is shown;

[0079] Figure 5 The schematic diagram of the active distribution network line loss situation awareness structure in an embodiment of the present invention is shown;

[0080] Figure 6 The figure shows the initial topology structure diagram of the IEEE 33-node distribution network in an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0082] The application principle of the present invention is described in detail below in conjunction with the accompanying drawings.

[0083] Example 1

[0084] like Figure 1 As shown, an embodiment of the present invention provides a method for actively sensing line loss situation in a power distribution network, comprising the following steps:

[0085] S01: Obtaining the operating data of the active distribution network;

[0086] S02: performing preliminary identification processing based on the operating data and the global topology identification model to obtain a preliminary topology identification result of the active distribution network;

[0087] S03: performing partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active distribution network; wherein the final topology identification result includes topology identification results of each partition;

[0088] S04: Based on the final topology identification result, determine the operating factors of each grid-connected entity that affect the line loss of the active distribution network, and obtain key indicators for line loss prediction based on the operating factors;

[0089] S05: Inputting the key indicators for line loss prediction into a line loss prediction model for prediction, and obtaining a prediction result of the line loss situation of the active distribution network.

[0090] The active distribution network line loss situation awareness method proposed in the present invention obtains the final topology identification result of the active distribution network based on global topology identification and partitioned topology identification, and obtains key indicators for line loss prediction; based on the key indicators, the line loss of the active distribution network is predicted using the constructed line loss prediction model, thereby achieving effective perception and prediction of the active distribution network line loss, and improving the generalization capability of the distribution network line loss situation awareness under the access of multiple grid-connected entities.

[0091] Example 2

[0092] The embodiment of the present invention provides a method for actively sensing line loss in a power distribution network, comprising the following steps:

[0093] S01: Obtaining the operating data of the active distribution network;

[0094] S02: performing preliminary identification processing based on the operating data and the global topology identification model to obtain a preliminary topology identification result of the active distribution network;

[0095] S03: performing partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active distribution network; wherein the final topology identification result includes topology identification results of each partition;

[0096] S04: Based on the final topology identification result, determine the operating factors of each grid-connected entity that affect the line loss of the active distribution network, and obtain key indicators for line loss prediction based on the operating factors;

[0097] S05: Inputting the key indicators for line loss prediction into a line loss prediction model for prediction, and obtaining a prediction result of the line loss situation of the active distribution network.

[0098] With the gradual access of distributed power sources, electric vehicle charging piles, energy storage devices and other grid-connected entities to the distribution network, the increasingly complex network structure and operation mode of the distribution network have made it difficult for the line loss analysis method suitable for the traditional distribution network to cope with the current demand for refined line loss management. It is necessary to propose a method that can effectively perceive and analyze the line loss situation in a complex environment based on the operating characteristics of the active distribution network. Since the line loss analysis of the distribution network needs to rely on a real-time and accurate topological structure, it is necessary to realize the active distribution network topological structure identification as the basis for the subsequent in-depth mining of line loss information. By utilizing the difference characteristics between distribution network regions and combining the distribution network partitioning method that is conducive to topological identification, the accuracy and speed can be improved through partitioned topological identification.

[0099] The data collected from synchronous phasor measurement devices, smart meters and other measuring devices for distribution network topology identification are mainly node voltage amplitude, active injection power and reactive injection power (operation data). In order to eliminate redundant features, reduce the spatial dimension of the input parameters of the topology identification model, speed up the topology identification speed and improve the accuracy, feature selection is required to obtain the optimal feature subset.

[0100] In this embodiment, the method for constructing the global topology identification model in step S02 includes the following specific steps:

[0101] Acquire a training data set, the training data set comprising: active distribution network operation data under different topological structures, and topological labels corresponding to the operation data;

[0102] Process the data in the training data set to obtain the original feature set;

[0103] Perform topology recognition feature selection based on the original feature set to obtain the optimal topology recognition feature subset;

[0104] The SVM (support vector machine) model is trained using the optimal topology recognition feature subset to construct a global topology recognition model.

[0105] Specifically, the known topological structures of the distribution network during normal operation are recorded as 1, 2, 3..., N respectively; in the offline training stage of the global topology identification model based on SVM, the distribution network operation data under different topological structures collected by the measuring device and their corresponding topological labels together constitute the original training data set. After data processing, the topological identification contribution of each type of data is analyzed, and the optimal topological identification feature subset is screened out. The SVM algorithm is used to train N classification models corresponding to each topological structure; in the online identification stage, the distribution network operation data obtains prediction scores in each classification model, and the label of the topological identification result is output according to the maximum value of the score, corresponding to the preliminary topological identification result of the distribution network as a whole.

[0106] Among them, the topology recognition feature selection is performed based on the original feature set to obtain the optimal topology recognition feature subset, including:

[0107] The mutual information theory is used to calculate the topological recognition contribution of different features of the original feature set. The formula is as follows:

[0108]

[0109]

[0110] in, is the contribution of different features to topological identification, I i is the mutual information value of the i-th feature, m is the total number of features, x is the measurement data of the i-th feature, X i is the set of the i-th feature data, y is the topological label corresponding to the feature, Y is the set of variables y, p(x,y) is the joint probability distribution function of variables x and y, and p(x) and p(y) are the marginal probability distribution functions of variables x and y respectively.

[0111] The topology recognition contributions of different features are ranked from large to small, and the features ranked at the top are selected in turn as the optimal topology recognition feature subset.

[0112] In this embodiment, the steps of obtaining the final topology identification result of the active power distribution network in step S03 are as follows:

[0113] S031: performing regional division on the preliminary topological identification result obtained in S02 to obtain a regional division result;

[0114] S032: inputting the region division result into the constructed partition topology identification model for identification, and obtaining the partition topology identification result;

[0115] S033: If the partition topology identification result is the same as the topological structure represented by the preliminary topology identification result, the partition topology identification result is output as the final topology identification result;

[0116] If the topological structure represented by the partition topology identification result is different from that represented by the preliminary topology identification result, the following steps S0331-S0332 are repeated until the preset condition is met, and the partition topology identification result is output as the final topology identification result; the preset condition is: the topological structure represented by the partition topology identification result is the same as that represented by the partition topology identification result;

[0117] S0331: After dividing the partition topology identification result into regions, the result is input into the partition topology identification model to obtain the partition topology identification result;

[0118] S0332: Use the current partition topology identification result to update the partition topology identification result.

[0119] Specifically, Figure 2 As shown in the figure, for the collected real-time measurement data of the active distribution network, after missing value filling and normalized data preprocessing and topology identification feature selection, the global topology preliminary identification stage is entered as follows:

[0120] The global topology identification model based on SVM is used to obtain the preliminary topology identification results of the distribution network.

[0121] After obtaining the preliminary topology identification results, the SVM partition topology preliminary identification stage is entered as follows:

[0122] Perform regional division (grid division) on the preliminary topology identification results of the distribution network to obtain the area after the initial division;

[0123] The area obtained after the initial division is input into the partition topology identification model based on SVM for identification, and the partition topology identification result is obtained. It is judged whether the partition topology identification result is the same as the preliminary topology identification result. If they are the same, the final topology identification result is directly output. If they are different, the iteration is repeated until the new partition topology identification result is the same as the previous partition identification result, and the final topology identification result is output.

[0124] Specifically, the topological structures during normal operation of the distribution network are numbered 1, 2, ..., N respectively.

[0125] The preliminary topology identification result of the active distribution network is obtained through the global topology identification model, that is, the bit topology number T is identified, T∈{1,2,…,N}.

[0126] Performing regional division on the preliminary topology identification results of the active distribution network to obtain regional division results;

[0127] According to the regional division result, the constructed partition topology identification model is used to perform identification, and the partition topology identification result is obtained, that is, the bit topology number V is identified;

[0128] If T=V, it means that the partition topology identification result is the same as the preliminary global topology identification result, and the output is the final topology identification result;

[0129] If T≠V, it means that the partition topology identification result is different from the preliminary global topology identification result. Then the topology identification result V is used to perform regional division and partition topology identification again, and the cycle is repeated until the final topology identification result can be output.

[0130] In this embodiment, the topology identification result of the active power distribution network is divided into regions, including:

[0131] Based on the topology identification results of the active distribution network, the topology structure of the active distribution network is determined and the power flow calculation is performed to obtain the power correction equation;

[0132] The sensitivity matrix of the active distribution network is obtained according to the power correction equation, and the electrical distance between each node in the active distribution network and each distributed generation access node is calculated;

[0133] Combining the calculation result of the electrical distance with the hierarchical clustering algorithm, each distributed power source connected to the active distribution network is taken as the cluster center of the sub-partition, and the area is divided based on the partition restriction conditions; the partition restriction conditions include: the number of nodes in each sub-partition is at least 2 and cannot exceed half of the total number of nodes, the nodes in each sub-partition cannot be repeated, and all nodes in the active distribution network are included in each sub-partition.

[0134] Specifically, based on the topology identification results of the active distribution network, the topology structure of the active distribution network is determined and the power flow calculation is performed to obtain the power correction equation:

[0135]

[0136] in, , They represent the changes in active power and reactive power of nodes in the active distribution network respectively; , They represent the changes in node voltage phase angle and voltage amplitude respectively; J is the Jacobian matrix, which represents the partial derivative relationship; , They represent the partial derivative relationship between node active power, reactive power and node voltage phase angle respectively; , They respectively represent the partial derivative relationship between node active power, reactive power and node voltage amplitude.

[0137] According to the power correction equation, the sensitivity matrix S of the system is:

[0138]

[0139] Among them, S PU, S QU They represent active voltage sensitivity factor and reactive voltage sensitivity factor respectively, and are used to indicate the degree of voltage change caused by node power change; , Both represent the degree of voltage phase angle change caused by node power change.

[0140] When the influence of voltage phase angle and node active power is ignored, the reactive voltage sensitivity matrix is:

[0141]

[0142] The electrical distance matrix can be calculated as:

[0143]

[0144]

[0145] in, is the ratio of the voltage change at node u to that at node v;

[0146] Based on the electrical distance matrix, the electrical distance between each node and each distributed generation access node s in the distribution network is calculated. , node electrical distance set for:

[0147]

[0148] The distributed power sources connected to the distribution network are used as the cluster centers of the sub-partitions. Let G be the region number and the cluster center node of the region be S. G , the sum of squared errors within the cluster of nodes except the cluster center for:

[0149]

[0150] Where l is the number of nodes in region G, is the cluster center node s of node u and region G G electrical distance.

[0151] According to the calculation results of the sum of squared errors within the cluster, find the most similar cluster for each node:

[0152]

[0153] In this embodiment, the method for constructing the partition topology identification model in step S032 includes:

[0154] Acquire a partition training data set, wherein the partition training data set includes: active distribution network regional operation data after completing regional division of a known active distribution network topology structure, and the number of the active distribution network regional setting after the regional division;

[0155] The optimal topology recognition feature subset is used to train the SVM model and construct a partition topology recognition model.

[0156] Specifically, after completing the regional division of the known distribution network topology, different regions are numbered as 1, 2, .., M, and various topological structures of region G are recorded as G 1 ,G 2 ,...,G c , c is the topological structure type of the region, ; In the offline training stage, the distribution network area measurement data collected by the measuring device and the set area number together constitute the original training data set. After obtaining the optimal topology identification feature subset, the SVM algorithm is used to train the classification model of each topological structure in each area; in the online recognition stage, the global topology identification model based on SVM is first used to obtain the preliminary topological identification structure of the entire network of the data set. After regional division, the topological multi-classification model suitable for the area is matched according to the area number, and the measurement data of the area is used for more detailed topological structure identification.

[0157] Example 3

[0158] Based on the active distribution network topology obtained by the final topology identification result, the line loss change law caused by the access of distributed power sources, electric vehicle charging piles, energy storage devices and other grid-connected entities to the distribution network is analyzed: the access of distributed power sources, electric vehicle charging piles, energy storage devices and other grid-connected entities in the active distribution network has changed the operating characteristics of the distribution network, making the line loss more complex and changeable. Their different output models and operating characteristics lead to different influencing mechanisms on the line loss. Therefore, it is necessary to analyze the operating factors affecting the line loss and the line loss change law when distributed power sources, electric vehicle charging piles and energy storage devices are connected to the distribution network.

[0159] In this embodiment, based on the final topology identification result, the operating factors of each grid-connected entity that affect the line loss of the active distribution network are determined, including:

[0160] Based on the final topology identification results of the active distribution network, the line loss changes before and after the multiple grid-connected entities are connected to the active distribution network are calculated by theoretical formula, and the line loss changes before and after the grid-connected entities are connected to the active distribution network are obtained;

[0161] Combined with the line loss variation, the line loss of each grid-connected entity connected to the active distribution network is simulated to determine the line loss variation law caused by each grid-connected entity connected to the active distribution network;

[0162] According to the change law, and in combination with the actual topological structure and operating status of the active distribution network, the operating factors of each grid-connected entity that affect the line loss of the active distribution network are determined; the operating factors include: access location, grid-connected capacity, operating mode, and power factor.

[0163] In this embodiment, the mechanism model of distributed generation access to the distribution network is as follows: Figure 3 As shown, the power connected to node u is A theoretical analysis is conducted on the line loss changes before and after the distributed generation is connected to the distribution network.

[0164] The line loss of the distribution network before the distributed generation is connected is:

[0165]

[0166] Where R is the resistance per unit length of the line; l i is the length of the i-th section of the distribution network; z is the number of load nodes on the distribution network line; U i is the voltage of the i-th node of the line; P i is the active power of the ith node of the line; Q i is the reactive power of the ith node of the line.

[0167] After the distributed generation is connected at node u, the line loss of the distribution network is:

[0168]

[0169] The line loss change is:

[0170]

[0171] It can be obtained from the formula that the different grid-connected capacity and access location of distributed power sources will lead to different impacts of their access on the line loss of the distribution network. When the grid-connected capacity of distributed power sources is within a certain range, the access of distributed power sources will reduce the line loss of the distribution point network. When it exceeds a certain value, the access of distributed power sources will increase the line loss of the distribution network. This range depends on the total load power of the distribution network. The specific value of the line loss change caused by the access of distributed power sources is jointly determined by the grid-connected capacity and the access location. The specific impact mode needs to be further analyzed in combination with the actual topological structure and operating status of the distribution network.

[0172] Therefore, based on the theoretical analysis, a simulation model is established to connect distributed power sources to the distribution network with different access points and different grid-connected powers, calculate the line loss of the distribution network, and compare and analyze the changing rules of the line loss.

[0173] In this embodiment, the mechanism model of electric vehicle charging piles connected to the distribution network is as follows Figure 3As shown in the figure, a theoretical analysis is conducted on the line loss changes before and after the electric vehicle charging pile is connected to the distribution network;

[0174] The line loss of the distribution network before the electric vehicle charging pile is connected is:

[0175]

[0176] When the electric vehicle charging pile is in the charging stage, it acts as a load with power After node u is connected, the line loss of the distribution network is:

[0177]

[0178] The line loss change is:

[0179]

[0180] It can be obtained from the formula that when electric vehicle charging piles are used as charging loads, they will increase the total load power of the distribution network and increase the line loss of the distribution network.

[0181] When the electric vehicle is in the V2G discharge stage, it acts as a power source with power After node u is connected, the line loss of the distribution network is:

[0182]

[0183] The line loss change is:

[0184]

[0185] It can be seen from the formula that the impact mechanism of electric vehicle charging piles on distribution network line loss is the same as that of distributed power supply access, and further analysis is needed in combination with the actual topological structure and operating status of the distribution network.

[0186] A simulation model is established, considering the charging stage and V2G discharging stage of electric vehicle charging pile operation. The electric vehicle charging pile is connected to the distribution network with different access points and different grid-connected powers, the line loss of the distribution network is calculated, and the change law of line loss is compared and analyzed.

[0187] In this embodiment, the mechanism model of energy storage device connected to the distribution network is as follows: Figure 3 As shown, a theoretical analysis is conducted on the line loss changes before and after the energy storage device is connected to the distribution network;

[0188] The line loss of the distribution network before the energy storage device is connected is:

[0189]

[0190] When the energy storage device is in the charging stage, it acts as a load with power After node u is connected, the line loss of the distribution network is:

[0191]

[0192] The line loss change is:

[0193]

[0194] The reactive power of the energy storage device is:

[0195]

[0196] Wherein, λ is the power factor of the energy storage device, when the χ value is 1, reactive power is emitted, and when the χ value is 0, reactive power is absorbed.

[0197] It can be seen from the formula that when the energy storage device absorbs reactive power, the total load power of the distribution network is increased, causing the line loss to increase; when the energy storage device generates reactive power, the impact on the line loss needs to be judged based on the actual operating conditions.

[0198] When the energy storage device is in the discharge stage, it acts as a power source with power After node u is connected, the line loss of the distribution network is:

[0199]

[0200] The line loss change is:

[0201]

[0202] The reactive power of the energy storage device is:

[0203]

[0204] It can be obtained from the formula that when the energy storage device generates reactive power, the impact mechanism on the line loss of the distribution network is the same as that of the distributed power source; when the energy storage device absorbs reactive power, if there is a reactive power shortage near the access point, the reactive power burden of the distribution network will be increased.

[0205] A simulation model is established, taking into account the charging and discharging stages as well as the leading and lagging power factors of the energy storage device. The energy storage device is connected to the distribution network with different access points, different grid-connected powers, and different power factors. The line loss of the distribution network is calculated, and the change law of the line loss is compared and analyzed.

[0206] Example 4

[0207] In order to predict the line loss situation when distributed power sources, electric vehicle charging piles, energy storage devices and other grid-connected entities are connected to the distribution network, an active distribution network line loss prediction index system with multi-subject access influencing indicators is established, and the key indicators are extracted using the improved grey correlation analysis method. A line loss prediction model based on the Stacking algorithm is established, and the accuracy of line loss prediction is improved by optimizing the internal parameters of the model and allocating weights, thereby improving the generalization ability of the model.

[0208] In this embodiment, based on the operating factors of each grid-connected entity that affect the line loss of the active distribution network, key indicators for line loss prediction are obtained. The specific steps are as follows:

[0209] Combined with the operational factors of each grid-connected entity that affect the line loss of the active distribution network, the correlation characteristics between the grid-connected entity output data, load data and active distribution network flow data are analyzed;

[0210] According to the correlation characteristics, an active distribution network line loss prediction index system is established;

[0211] The key indicators for line loss prediction are extracted by using the active distribution network line loss prediction index system and the improved grey correlation analysis method.

[0212] Specifically, the active distribution network line loss prediction index system includes:

[0213] Line attribute indicators are the basic parameters of the active distribution network, including: trunk line length, total line length, number of distribution transformers, and distribution transformer models;

[0214] Basic operation characteristic indicators are daily operation parameters of active distribution network, including active power supply, reactive power supply, and historical line loss;

[0215] Management factor indicators are non-technical factors that cause line losses, including: equipment aging, meter reading accuracy, and operating years;

[0216] Photovoltaic characteristic indicators are the operating factors that affect line loss after distributed photovoltaics are connected to the distribution network, including grid-connected capacity and access location.

[0217] Energy storage device indicators are the operating factors that affect line loss after the energy storage device is connected to the distribution network, including grid-connected capacity, access location, operating mode, and power factor.

[0218] The electric vehicle charging pile indicators are the operating factors that affect the line loss after the electric vehicle charging piles are connected to the distribution network, including grid-connected capacity, access location, and operating mode.

[0219] In this embodiment, the improved grey relational analysis method includes:

[0220] The line loss of the distribution network is taken as the reference sequence b, and the line loss indicators in the active distribution network line loss prediction index system are taken as the comparison sequence a. After normalizing the data, the grey correlation coefficient is calculated as follows:

[0221]

[0222] in, is the correlation coefficient between the line loss index t in the comparison sequence a and the reference sequence at the kth element, is the resolution coefficient;

[0223] Based on the grey correlation coefficient, the grey correlation of each line loss index is obtained as follows:

[0224]

[0225] Among them, n is the number of samples, that is, the number of elements contained in each line loss index, r t is the grey relational degree of line loss index t;

[0226] The values ​​of each line loss index are converted into probability distribution, and the entropy values ​​of each line loss index are weighted averaged to obtain the weight coefficient;

[0227] Combining the grey correlation degree of each line loss index and the weight coefficient, a weighted correlation degree is obtained, which is shown as follows:

[0228] ;

[0229] ;

[0230] ;

[0231] in: is the weighted correlation, p tk is the proportion of the kth element of line loss index t to the index, n is the number of elements contained in the line loss index, d is the number of indicators, E t is the entropy value of line loss index t, is the weight coefficient of line loss index t;

[0232] The indicators are sorted according to the weighted correlation to determine the influence of each line loss indicator and extract the key indicators for line loss prediction.

[0233] In this embodiment, the method for obtaining the constructed line loss prediction model includes:

[0234] Using the key indicators of active distribution network line loss prediction, the first layer (base learner layer) of the line loss prediction model selects the random forest algorithm, support vector machine algorithm, K nearest neighbor algorithm, gradient boosting decision tree algorithm and multi-layer neural network prediction algorithm with high complexity and strong learning ability as the base learner (sub-model), and the second layer (meta-learner layer) selects the linear regression model as the meta-learner, and uses the linear regression model to fuse the prediction results of multiple base learner models to improve the generalization ability of the line loss prediction model; in order to reduce the risk of model overfitting, the K-fold cross-validation method is used in the model training stage to divide the training set into multiple sub-training sets, so that the trained model results are more stable. Figure 4 As shown in the figure, the first-layer prediction model includes KNN model, SVM model, RF model, GBDT model and MLP model, which respectively represent sub-models based on the above-mentioned K nearest neighbor algorithm, support vector machine algorithm, random forest algorithm, gradient boosting decision tree algorithm and multi-layer neural network prediction algorithm; the second-layer prediction model includes linear regression model.

[0235] Specifically, parameter optimization in the line loss prediction model includes:

[0236] The grid search method is used to optimize the parameters of the sub-models. First, the parameters that need to be optimized for each sub-model are listed, and a finite set of parameter optimization ranges is set. Then, an exhaustive traversal is performed in the finite set, and the parameter combination that achieves the highest prediction accuracy is selected by calculating the evaluation indicators that can characterize the model prediction effect, such as the root mean square error, mean absolute error, etc.

[0237] Specifically, the weight distribution in the line loss prediction model includes:

[0238] The accuracy of the learners trained by different training samples under the same base learner model is different. If the test set of the meta-learner is obtained by adding and averaging, the superiority of the meta-learner test set obtained by the base learner model with higher accuracy will be concealed. Therefore, the test set results are weighted according to the test accuracy obtained by the training set of each model of the base learner to obtain the label output of the final test set on the base learner. The steps are as follows:

[0239] For each base learner (sub-model) in the online loss prediction model, after using the K-fold cross-validation method, the training set will be divided into K parts, and each base learner will be trained K times to obtain K models:

[0240] The training set data is used to perform label prediction output on the K trained learner models. The accuracy or error of the output results is calculated with the true value label data in the training set to obtain the accuracy or error of each of the K models. ;

[0241] According to the accuracy of the model exist The proportion in determines the weight of the model ;

[0242] Input the test set to the K models trained by the learner respectively, and use the weights to output the results The final output label of the test set data on the trained base learner model is obtained by weighting.

[0243] Example 5

[0244] like Figure 5 As shown, this embodiment also includes an active distribution network line loss situation awareness device, including:

[0245] A data acquisition module is used to obtain the operating data of the active distribution network;

[0246] A global topology identification module, used to perform preliminary identification processing based on the operating data and the global topology identification model to obtain a preliminary topology identification result of the active distribution network;

[0247] A partition topology identification module is used to perform partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active power distribution network; wherein the final topology identification result includes the topology identification results of each partition;

[0248] A key indicator determination module is used to determine the operating factors of each grid-connected entity that affect the line loss of the active distribution network based on the final topology identification result; and based on the operating factors, obtain key indicators for line loss prediction;

[0249] The line loss situation prediction module is used to input the key line loss prediction indicators into the line loss prediction model for prediction, and obtain the prediction result of the active distribution network line loss situation.

[0250] Example 6

[0251] This embodiment is described using an IEEE 33-node distribution network system;

[0252] (1) Data acquisition module:

[0253] Based on the original IEEE 33-node distribution network system, photovoltaic generators are connected to nodes 9 and 19, and wind turbines are connected to nodes 24 and 30;

[0254] Collect measurement data of the system running in a typical topology.

[0255] (2) Global topology identification module

[0256] After calculating the contribution of topology identification features, the voltage amplitude features are selected as a more effective subset of topology identification features. The global topology identification model is trained using the support vector machine algorithm, and preliminary topology identification results are obtained based on the global topology identification model.

[0257] (3) Partition topology identification module

[0258] The preliminary topology identification results are divided into regions, and the regional division results of the initial topology structure of IEEE33 nodes are shown as follows: Figure 6 As shown, a partition topology recognition model under each topological structure in each region is obtained by training in each region.

[0259] The performance of the partition topology identification method proposed in the present invention is compared with the performance of only performing global topology identification, as shown in Table 1. The topology identification method proposed in the present invention can effectively improve the accuracy of topology identification.

[0260] Table 1 Performance of topology identification methods

[0261]

[0262] (4) Key indicator determination module

[0263] After the partition topology identification module identifies the operating topology of the active distribution network, the topology is combined with the simulation model to analyze the line loss changes caused by the connection of distributed power sources, electric vehicle charging piles, energy storage devices and other grid-connected entities to the distribution network.

[0264] This embodiment uses the initial topology of the IEEE 33-node distribution network system to analyze the operating factors that affect the line loss and the line loss variation law when each grid-connected entity is connected.

[0265] From the analysis of the method proposed in Example 3, it can be obtained that when the grid-connected power of the distributed power source is less than 20% of the total load power, access from any node will reduce the line loss of the distribution network. When the access point is close to the end of the line, that is, nodes 18, 22, 25, and 33, the line loss of the distribution network can be reduced to a greater extent. When the output of the distributed power source gradually increases, the access of some nodes will increase the line loss of the distribution network, and the access of some nodes will reduce the line loss. When the output of the distributed power source is larger than the total load power of the system, the line loss of the distribution network will be increased to a greater extent when the access location is close to the end of the system line. Distributed power sources affect the line loss of the distribution network through the access location and grid-connected capacity. If the distributed power source can be consumed locally at a close distance, the line loss of the distribution network can be reduced.

[0266] From the analysis of the method proposed in Example 3, it can be concluded that during the charging stage, the access of electric vehicle charging piles will definitely increase the line loss of the distribution network. When the access location is far away from the power node 1, that is, close to nodes 19 and 33, the line loss increases more. At the same access location, when the grid-connected capacity is larger, the increased distribution network line loss increases more. During the V2G discharge stage, the electric vehicle charging pile can reversely send power to the distribution network to reduce the line loss of the distribution network. The operating mechanism at this time is similar to the access of distributed power sources. When the grid-connected capacity is small and the access location is close to the end of the line, the line loss is reduced to a greater extent.

[0267] From the analysis of the method proposed in Example 3, it can be obtained that in the charging stage, it is simulated that in this system, whether the energy storage device operates under the leading power factor or the lagging power factor, the line loss of the distribution network will increase, and the greater the grid-connected power, the greater the increase in line loss. In comparison, when the same grid-connected power and the same access location are used, the increase in line loss when operating with the lagging power factor is less than that when operating with the leading power. In the discharge stage, when the energy storage device operates with the lagging power factor and emits reactive power, the impact mechanism is the same as that of the distributed power source; when the energy storage device operates with the leading power factor and absorbs reactive power, the impact on the line loss of the distribution network needs to be combined with the reactive power flow near the access location. If there is excess reactive power near the access location, it is beneficial to reduce the line loss of the distribution network.

[0268] After obtaining the operating factors that affect the line loss of the distribution network due to the different operating characteristics of distributed power sources, electric vehicle charging piles, and energy storage devices, a line loss prediction index system was established, and the key indicators for line loss prediction were extracted using an improved grey correlation analysis method.

[0269] (5) Line loss situation prediction module

[0270] The Stacking algorithm improved by parameter optimization and weight allocation is used to predict the line loss of the active distribution network, and the prediction results of the line loss situation of the active distribution network are obtained.

[0271] Among them, the parameters that need to be optimized for each sub-model are shown in Table 2.

[0272] Table 2 Model optimization parameters

[0273]

[0274] By flexibly adjusting the parameters of the model under different distribution network operating environments, the optimal parameter combination of each model is obtained to maximize the model performance and improve the generalization ability of the model.

[0275] In summary, based on the data acquisition module, global topology identification module, partitioned topology identification module, key indicator determination module and line loss situation prediction module, a line loss situation awareness method suitable for active distribution network is finally proposed to achieve effective perception of distribution network line loss, improve the generalization ability of line loss situation awareness when each subject is connected to the distribution network.

[0276] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for actively sensing line loss in a distribution network, characterized in that: include: Obtaining the operating data of active distribution networks; Performing preliminary identification processing based on the operating data and the global topology identification model to obtain preliminary topology identification results of the active distribution network; Performing partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active distribution network; wherein the final topology identification result includes the topology identification results of each partition; Based on the final topology identification result, determining the operating factors of each grid-connected entity that affect the line loss of the active distribution network, and based on the operating factors, obtaining key indicators for line loss prediction; The key indicators for line loss prediction are input into a line loss prediction model for prediction, and a prediction result of the line loss situation of the active distribution network is obtained.

2. The active distribution network line loss situation awareness method according to claim 1 is characterized in that: The preliminary topology identification result is subjected to a partition identification process to obtain a final topology identification result of the active distribution network, including: After dividing the preliminary topology identification result into regions, the regions are input into a partition topology identification model to obtain a partition topology identification result; If the partition topology identification result is the same as the preliminary topology identification result, the output partition topology identification result is the final topology identification result of the active distribution network; If the partition topology identification result is different from the preliminary topology identification result, the loop steps are iteratively executed until the preset conditions are met to obtain the final topology identification result; wherein the loop steps include: after dividing the partition topology identification result into regions, inputting it into the partition topology identification model to obtain the current partition topology identification result; using the current partition topology identification result to update the partition topology identification result; the preset condition is: the current partition topology identification result is the same as the partition topology identification result.

3. The active distribution network line loss situation awareness method according to claim 2 is characterized in that: The region division of the topology identification result includes: Based on the topology identification results, the active distribution network topology is determined and the power flow calculation is performed to obtain the power correction equation; The sensitivity matrix is ​​obtained according to the power correction equation, and the electrical distance between each node in the active distribution network and each distributed generation access node is calculated; Combining the calculation result of the electrical distance with the hierarchical clustering algorithm, each distributed power source connected to the active distribution network is taken as the cluster center of the sub-partition, and the area is divided based on the partition restriction conditions; the partition restriction conditions include: the number of nodes in each sub-partition is at least 2 and cannot exceed half of the total number of nodes, the nodes in each sub-partition cannot be repeated, and all nodes in the active distribution network are included in each sub-partition.

4. The active distribution network line loss situation awareness method according to claim 1 is characterized in that: Based on the final topology identification result, the operating factors of each grid-connected entity that affect the line loss of the active distribution network are determined, including: Based on the final topology identification results of the active distribution network, the line loss changes before and after the multiple grid-connected entities are connected to the active distribution network are calculated by theoretical formula, and the line loss changes before and after the grid-connected entities are connected to the active distribution network are obtained; Combined with the line loss variation, the line loss of each grid-connected entity connected to the active distribution network is simulated to determine the line loss variation law caused by each grid-connected entity connected to the active distribution network; According to the change law, and in combination with the actual topological structure and operating status of the active distribution network, the operating factors of each grid-connected entity that affect the line loss of the active distribution network are determined; the operating factors include: access location, grid-connected capacity, operating mode, and power factor.

5. The active distribution network line loss situation awareness method according to claim 4 is characterized in that: Based on the above operating factors, key indicators for line loss prediction are obtained, including: Combined with the operational factors of each grid-connected entity that affect the line loss of the active distribution network, the correlation characteristics between the grid-connected entity output data, load data and the active distribution network flow data are determined; According to the correlation characteristics, an active distribution network line loss prediction index system is established; The key indicators for line loss prediction are extracted by using the active distribution network line loss prediction index system and the improved grey correlation analysis method.

6. The active distribution network line loss situation awareness method according to claim 5 is characterized in that: The line loss prediction index system includes: Line attribute indicators are the basic parameters of the active distribution network, including: trunk line length, total line length, number of distribution transformers, and distribution transformer models; Basic operation characteristic indicators are daily operation parameters of active distribution network, including active power supply, reactive power supply, and historical line loss; Management factor indicators are non-technical factors that cause line losses, including: equipment aging, meter reading accuracy, and operating years; The subject influencing indicators are the operating factors that affect the line loss after each grid-connected subject is connected to the active distribution network, including: grid-connected capacity, access location, operating mode, and power factor.

7. The active distribution network line loss situation awareness method according to claim 6 is characterized in that: The improved grey relational analysis method comprises: The line loss of active distribution network is taken as the reference sequence, and each line loss index in the line loss prediction index system is taken as the comparison sequence. After normalizing the data, the grey correlation coefficient is obtained. Based on the grey correlation coefficient, the grey correlation of each line loss index is obtained; The values ​​of each line loss index are converted into probability distribution, and the entropy values ​​of each line loss index are weighted averaged to obtain the weight coefficient; Combining the grey correlation degree of each line loss index and the weight coefficient, a weighted correlation degree is obtained; The indicators are sorted according to the weighted correlation to determine the influence of each line loss indicator and extract the key indicators for line loss prediction.

8. The active distribution network line loss situation awareness method according to claim 7 is characterized in that: The formula of the weighted association degree is as follows: ； ； ； in: is the weighted correlation, P tk is the proportion of the kth element of line loss index t to the index, n is the number of elements contained in the line loss index, d is the number of indicators, E t is the entropy value of line loss index t, is the weight coefficient of line loss index t; r t is the grey relational degree of line loss index t.

9. The active distribution network line loss situation awareness method according to claim 8 is characterized in that: The method for obtaining the line loss prediction model includes: Acquire a key indicator data set, wherein the key indicator data set includes a training set and a test set; Based on the training set in the key indicator data set, the base learner layer of the line loss prediction model is trained, and the optimal parameter combination of each sub-model is determined to obtain the training result of the base learner layer; Based on the training result of the base learner layer, determining the weight distribution of each sub-model of the base learner layer to obtain a trained base learner layer; The training result set of the base learner is used as input to train the meta-learner layer of the line loss prediction model to obtain a trained meta-learner layer; Based on the trained base learner layer and the trained meta learner layer, a trained line loss prediction model is obtained; The test set in the key indicator data set is input into the trained base learner layer, and each sub-model is combined with the weight to obtain the prediction result set of the base learner layer; The prediction result set of the base learner layer is input into the trained meta-learner layer to perform a preliminary evaluation of the line loss prediction model performance, thereby obtaining a preliminary evaluation result of the line loss prediction model performance; According to the preliminary evaluation results of the line loss prediction model performance, the parameters of the line loss prediction model are adjusted to obtain the line loss prediction model.

10. The active distribution network line loss situation awareness method according to claim 9, characterized in that: Based on the Stacking algorithm model, the random forest model, support vector machine model, K nearest neighbor model, gradient boosting decision tree model and multi-layer neural network prediction model are selected as sub-models of the base learner layer of the line loss prediction model; the linear regression model is selected as the sub-model of the meta-learner layer.

11. The active distribution network line loss situation awareness method according to claim 10, characterized in that: The method for determining the optimal parameter combination includes: Obtain the parameters that need to be optimized for each sub-model of the base learner layer, and set a finite set of parameter optimization ranges; Exhaustively traverse the finite set and obtain the set evaluation indicators that can characterize the prediction effect of the model, the evaluation indicators include: root mean square error, mean absolute error; Based on the evaluation index, the parameter combination that achieves the highest prediction accuracy is selected.

12. The active distribution network line loss situation awareness method according to claim 10, characterized in that: The method for determining the weight distribution of each sub-model of the base learner layer includes: Based on the K-fold cross-validation method, the training set is divided into K parts, and each sub-model in the base learner layer is trained K times to obtain K trained models; The training set is used to perform label prediction output on the trained K models. The accuracy or error of the output results is calculated with the true value label data in the training set to obtain the accuracy or error of each of the K models. Determine the weight distribution of the K models in the sub-model according to the proportion of the accuracy or error of each of the K models in the sub-model; The K-fold cross-validation method and the following steps are used to determine the weight distribution of each sub-model in the base learner layer.

13. An active distribution network line loss situation awareness device, characterized in that: include: A data acquisition module is used to obtain the operating data of the active distribution network; A global topology identification module, used to perform preliminary identification processing based on the operating data and the global topology identification model to obtain a preliminary topology identification result of the active distribution network; A partition topology identification module is used to perform partition identification processing on the preliminary topology identification result to obtain a final topology identification result of the active distribution network; wherein the final topology identification result includes the topology identification results of each partition; A key indicator determination module is used to determine the operating factors of each grid-connected entity that affect the line loss of the active distribution network based on the final topology identification result; and based on the operating factors, obtain key indicators for line loss prediction; The line loss situation prediction module is used to input the key indicators of line loss prediction into the line loss prediction model for prediction, so as to obtain the prediction result of the line loss situation of the active distribution network.

Images