A line loss management method and system based on low-voltage power grid multi-source data fusion

Abstract

The application discloses a line loss treatment method based on low-voltage power grid multi-source data fusion, and belongs to the information technology field.The method is as follows: based on the power grid topological structure and the changed power supply access point, the power supply path is reconstructed; the reverse power flow characteristics of the frequency point are extracted to correct the node impedance matrix of the current power supply path, and the power supply access point impedance parameter set is obtained; the power transmission path is updated through the reverse power flow characteristics, and the power distribution characteristics are obtained; the line loss distribution prediction model is constructed by fusing the power grid topological structure, the current power supply path, the power supply access point impedance parameter set and the power distribution characteristics, and the line loss abnormal area is identified and treated. Through the implementation of the application, the influence of the factors such as the change of the transformer area topological structure, the change of the power supply access point and the load fluctuation on the line loss is not considered in the prior art, so that the predicted line loss distribution has a large error compared with the actual situation, and the line loss treatment effect is poor.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of information technology, and particularly relates to a line loss treatment method and system based on low-voltage power grid multi-source data fusion. BACKGROUND

[0002] The low-voltage power grid is an important part of the power system, and its operation efficiency is directly related to energy utilization and power supply reliability. In the low-voltage power grid, line loss is a key indicator for measuring the operation efficiency of the power grid, which is used to describe the power loss in the process of power transmission. Therefore, line loss treatment is one of the common means to improve the utilization of power grid resources. Optimizing line loss treatment not only reduces energy waste, but also has important significance for improving the intelligent level of the power grid and supporting new energy consumption.

[0003] However, the existing line loss treatment method mainly depends on static line loss calculation model and offline analysis, and can only calculate the line loss distribution through the set power supply path and operation condition to obtain the corresponding line loss treatment strategy. However, in fact, the line loss is easily affected by the complex factors such as the change of the transformer area topology structure, the change of the power access point and the switching of the power supply mode, so that the distribution changes. Therefore, when the above changes occur in the low-voltage power grid, since the static line loss calculation model is difficult to capture the dynamic characteristics such as the power supply path reconstruction, the change of the node impedance characteristics and the fluctuation of the load rate of the key equipment in real time, the prediction accuracy of the line loss distribution is greatly reduced, and the line loss treatment measures obtained are difficult to accurately match the actual operation state of the low-voltage power grid, so that the line loss treatment effect is greatly reduced. SUMMARY

[0004] The application provides a line loss treatment method and system based on low-voltage power grid multi-source data fusion, which can solve the problem that the existing technology does not consider the complex influence of factors such as the change of the transformer area topology structure, the change of the power access point and the fluctuation of the load on the line loss, resulting in a large error between the predicted line loss distribution and the actual situation, and causing poor energy utilization and power supply reliability of the power grid.

[0005] The first aspect of the application provides a line loss treatment method based on low-voltage power grid multi-source data fusion, which comprises:

[0006] Based on the topology structure of the low-voltage power grid and the change information of the power access point, the initial power supply path is reconstructed by adjusting the switching priority of the power access point, and the current power supply path is generated;

[0007] The voltage response value of the power access point at each frequency point is collected, the reverse power flow characteristics of the frequency point are extracted through the voltage response value, the node impedance matrix of the current power supply path is corrected using the reverse power flow characteristics, and the impedance parameter set of the power access point is obtained;

[0008] based on the reverse power flow feature, the current power supply path and the preset load balancing factor, the power transmission path is updated by calculating the load rate of the power equipment, and a power distribution feature is obtained;

[0009] The topology structure, the current power supply path, the power supply access point impedance parameter set and the power distribution feature are fused to construct a line loss distribution prediction model.

[0010] The real-time operation data of the low-voltage power grid is input into the line loss distribution prediction model, the line loss abnormal area is identified and the area priority is sorted, and the line loss treatment priority sequence is determined.

[0011] According to the line loss treatment priority sequence, the node injection power of the line loss abnormal area is adjusted.

[0012] The above scheme first reconstructs the power supply path by collecting the change of the power supply access point, obtains the current power supply path to real-time display the operation state of each node. Then the impedance characteristic change caused by the change of the power supply access point is measured by collecting the voltage response value, and the reverse power flow data at the moment when the voltage response value exceeds the preset voltage deviation range is collected to determine the influence of the reverse power flow in the power supply line on the voltage rise. In this way, the node impedance matrix is corrected, the power supply access point impedance parameter set after the change of the impedance characteristic is obtained, and the change characteristic of the node impedance in the low-voltage power grid is accurately obtained. Then it is calculated whether there is a situation that the load rate is too high in each power supply area on the current power supply path, and the power transmission path is updated based on the demand of load transfer, the power distribution strategy is reconstructed for each power supply area, and the power distribution feature is extracted. The line loss treatment strategy constructed subsequently considers the load balancing of the low-voltage power grid to provide data support. The line loss distribution prediction model constructed considers the influence of the change of the power supply path caused by the topology change of the power grid, the change of the impedance characteristic and the load fluctuation on the line loss, so the predicted line loss distribution is more in line with the actual situation. The output line loss treatment priority sequence is reasonable, and the line loss treatment and resource allocation can be realized reasonably and accurately, and the energy utilization rate and power supply reliability of the power grid are improved.

[0013] In a possible implementation method of the first aspect, based on the topology structure of the low-voltage power grid and the change information of the power supply access point, the initial power supply path is reconstructed by adjusting the switching priority of the power supply access point to generate the current power supply path, specifically:

[0014] According to the power supply connection state between each node in the low-voltage power grid, the topology structure is partitioned to obtain the initial power supply path;

[0015] Based on the topology structure, the power flow direction of each power supply access point is tracked by a shortest path search algorithm to collect the change information, and the power supply access point whose node position or operation state is changed is identified as a changed access point.

[0016] The power supply and demand power balance value of the changed access point is calculated, and the initial power supply path is reconstructed by adjusting the switching priority of the changed access point according to the power supply and demand power balance value, to obtain a current power supply path.

[0017] The above scheme obtains an initial power supply path according to the actual node power supply connection state after the change of the power grid topology structure, then locates the power supply access point whose node position or operating state is changed by tracking the power flow direction of each power supply access point, to accurately obtain the current power supply path that meets the actual situation by reconstructing the power flow distribution of the initial power supply path according to the change of the power flow of the power grid, and more accurately express the influence of the change of the power grid topology structure on the power flow change, thereby providing data support for the construction of a subsequent line loss distribution prediction model.

[0018] In a possible implementation method of the first aspect, the voltage response value of the power supply access point at each frequency point is collected, and the reverse power flow feature of the frequency point is extracted through the voltage response value, specifically as follows:

[0019] The electrical data of each power supply access point under the current power supply path is collected, and a node impedance matrix of the current power supply path is constructed;

[0020] A harmonic injection device is used to apply an excitation signal to each power supply access point, and the voltage response value of each power supply access point at each frequency point is recorded;

[0021] If there is a frequency point whose voltage response value exceeds a preset voltage deviation range, the reverse power flow feature of the frequency point is extracted through wavelet transform; wherein the reverse power flow feature includes reverse power flow duration, reverse power flow peak value and reverse power flow frequency statistics.

[0022] The above scheme first constructs a node impedance matrix of the current power supply path, then quickly obtains the transient impedance characteristics of the node by collecting the voltage response value through applying an excitation signal to the power supply access point; if the voltage response value at a certain frequency point exceeds the preset range, the reverse power flow feature of the frequency point is collected to determine the influence of the reverse power flow in the power supply line on the voltage rise, thereby providing data support for accurately measuring the change of the node impedance characteristics.

[0023] In a possible implementation method of the first aspect, the reverse power flow feature is used to correct the node impedance matrix of the current power supply path to obtain a power supply access point impedance parameter set, specifically as follows:

[0024] The node impedance matrix is corrected according to the reverse power flow frequency statistics;

[0025] The node steady state of the power access point is detected according to the corrected node impedance matrix, and the impedance parameter set of the power access point is generated by calculating the impedance amplitude, impedance phase angle and node stability margin of the power access point.

[0026] The above scheme amplifies the voltage problem of the weak link of the power grid due to the reverse power flow of the power grid, thereby highlighting the impedance parameter error, and therefore needs to correct the node impedance matrix through the reverse power flow data. Then, the impedance amplitude, impedance phase angle and node stability margin of the power access point are calculated by detecting the stability state of the node, and the impedance parameter set of the power access point after the topology of the power grid changes is obtained.

[0027] In a possible implementation method of the first aspect, the power transmission path is updated based on the reverse power flow feature, the current power supply path and the preset load balancing factor by calculating the load rate of the power equipment, and the power distribution feature is obtained, specifically as follows:

[0028] According to the reverse power flow feature, the power equipment on the current power supply path is classified;

[0029] According to the classification result and the load balancing factor of each power supply area in the low-voltage power grid, the load rate of each power equipment is obtained by weighting the power distribution of the power equipment;

[0030] The power mutual aid amount between each power supply area is calculated through the load rate, and the power distribution coefficient of each power supply area is updated based on the power mutual aid amount;

[0031] According to the power distribution coefficient, each power transmission path is re-planned, the impedance and loss of each power transmission path are calculated, and the power distribution feature is obtained.

[0032] The above scheme classifies the power equipment first, and then calculates the load rate of the power equipment through the load balancing factor of each power supply area. Since there is power transmission between each power supply area to achieve load balancing, after the power flow distribution of the power supply path changes, the load rate of each power equipment needs to be adjusted based on the principle of load balancing, and therefore the power transmission path is re-planned through the power mutual aid amount between each power supply area, so as to measure the influence of the change of the power grid topology structure on the load fluctuation of the key equipment.

[0033] In a possible implementation method of the first aspect, the power mutual aid amount between each power supply area is calculated through the load rate, and the power distribution coefficient of each power supply area is updated based on the power mutual aid amount, specifically as follows:

[0034] The load transfer demand amount of the power equipment whose load rate exceeds the rated value is obtained, and the power mutual aid amount between each power supply area is calculated through the load transfer demand amount.

[0035] updating the power distribution coefficient of each power supply area according to the power mutual assistance amount, the power supply level of each power supply area and the user capacity.

[0036] In a possible implementation method of the first aspect, the topology structure, the current power supply path, the set of power source access point impedance parameters and the power distribution feature are fused to construct a line loss distribution prediction model, specifically:

[0037] The line length and the wire cross section are determined according to the topology structure and the current power supply path, a node connection weight matrix is generated, and the electrical characteristics of the nodes in the node connection weight matrix are described by the set of power source access point impedance parameters to obtain an electrical feature matrix;

[0038] The electrical feature matrix and the power distribution feature are decomposed and time series feature analyzed by wavelet transform to obtain a space-time correlation feature vector;

[0039] The sample data set based on the space-time correlation feature vector is labeled by using the real-time collected line loss data to obtain a standardized training data set;

[0040] The line loss distribution prediction model is obtained by training the residual neural network by using the standardized training data set.

[0041] The above scheme provides the training data set for the model by using the current power supply path, the set of power source access point impedance parameters and the power distribution feature, so that when the model predicts the line loss distribution, not only the influence of the topology structure change is considered, but also the impedance characteristic transformation, the load fluctuation and the power supply path change caused by the topology structure change are considered, and the line loss distribution is accurately predicted.

[0042] In a possible implementation method of the first aspect, real-time operation data of the low-voltage power grid is input to the line loss distribution prediction model, line loss abnormal areas are identified and prioritized, and a line loss treatment priority sequence is determined, specifically:

[0043] The real-time operation data and environmental data of the low-voltage power grid are obtained;

[0044] The real-time operation data, the environmental data and a preset load scheduling rule are input to the line loss distribution prediction model, the line loss rate of each node in the current power supply path is calculated, and a real-time line loss distribution table is generated;

[0045] The voltage qualification rate of each power supply area in the low-voltage power grid is calculated according to the voltage deviation value of each node in the current power supply path; wherein the voltage deviation value is extracted according to the real-time line loss distribution table;

[0046] The power supply area with the voltage qualification rate exceeding a preset voltage qualification rate threshold is regarded as a line loss abnormal area.

[0047] According to the user power supply level and the electricity demand of the line loss abnormal area, the line loss abnormal area is prioritized to obtain a line loss treatment priority sequence.

[0048] The above scheme calculates the line loss rate of each node through real-time operation data and environmental data, obtains real-time line loss distribution, and accurately identifies the line loss abnormal area. Then, considering the user power supply level and the electricity demand of each area, the order of line loss treatment is reasonably arranged to optimize the line loss treatment effect.

[0049] In a possible implementation method of the first aspect, according to the line loss treatment priority sequence, the node injection power of the line loss abnormal area is adjusted, specifically:

[0050] Read the electricity load data of each line loss abnormal area and the node injection power upper limit value of each period;

[0051] Combine the electricity load data, time-of-use electricity price, and preset load scheduling rules to obtain the electricity characteristics of each line loss abnormal area;

[0052] Based on the electricity characteristics, the node injection power of the nodes in each line loss abnormal area is adjusted in turn through the node injection power upper limit value; wherein the order of adjusting each line loss abnormal area is determined by the line loss treatment priority sequence.

[0053] The above scheme adjusts the node injection power of the line loss abnormal area based on the electricity characteristics of each area to compensate for power and control the node voltage within a reasonable range to reduce the loss of electric energy in the transportation and distribution process.

[0054] The second aspect of the present application provides a line loss treatment system based on low-voltage power grid multi-source data fusion, the system comprises: a power supply path reconstruction module, an impedance characteristic extraction module, a load fluctuation feature extraction module, a model construction module, a line loss abnormal area identification module and a line loss treatment module;

[0055] The power supply path reconstruction module is used to reconstruct the initial power supply path by adjusting the switching priority of the power supply access point based on the topology of the low-voltage power grid and the change information of the power supply access point, and generate the current power supply path.

[0056] The impedance characteristic extraction module is used to collect the voltage response value of the power supply access point at each frequency point, extract the reverse power flow characteristics of the frequency point through the voltage response value, correct the node impedance matrix of the current power supply path using the reverse power flow characteristics, and obtain the impedance parameter set of the power supply access point.

[0057] The load fluctuation feature extraction module is configured to update the power transmission path by calculating the load rate of the power equipment based on the reverse power flow feature, the current power supply path, and a preset load balancing factor, to obtain a power distribution feature.

[0058] The model construction module is configured to fuse the topology structure, the current power supply path, the power source access point impedance parameter set, and the power distribution feature to construct a line loss distribution prediction model.

[0059] The line loss abnormal area identification module is configured to input real-time operation data of the low-voltage power grid into the line loss distribution prediction model, identify a line loss abnormal area, and perform area priority sorting to determine a line loss treatment priority sequence.

[0060] The line loss treatment module is configured to adjust the node injection power of the low-voltage power grid according to the line loss treatment priority sequence. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some of the embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0062] Figure 1 is a specific flowchart of a line loss treatment method based on low-voltage power grid multi-source data fusion provided by an embodiment of the present application;

[0063] Figure 2 is a structural diagram of a line loss treatment system based on low-voltage power grid multi-source data fusion provided by an embodiment of the present application. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0065] It should be understood that the step numbers used herein are only for the convenience of description, and are not intended to limit the execution sequence of the steps.

[0066] First Embodiment

[0067] The existing line loss distribution prediction model mainly considers the influence of the change of the topology structure of the power grid station area, but ignores the additional influence of the change of the topology structure. Embodiments of the present application consider factors such as the change of the power supply path, the change of the node impedance, and the load fluctuation caused by the change of the topology structure, construct a model that can more accurately predict the line loss distribution after the change of the topology structure, and the predicted line loss distribution is more in line with the actual situation, thereby providing effective data support for line loss management, and greatly improving the energy utilization rate and power supply reliability of the low-voltage power grid.

[0068] As shown in Figure 1 To solve the problem that the existing technology does not consider the complex influence of factors such as the change of the topology structure of the station area, the change of the power supply access point, and the load fluctuation on the line loss, resulting in a large error between the predicted line loss distribution and the actual situation, and causing poor energy utilization rate and power supply reliability of the power grid, a first embodiment of the present application provides a specific flowchart of a line loss management method based on low-voltage power grid multi-source data fusion. The line loss management method based on low-voltage power grid multi-source data fusion of the present embodiment includes steps S1 to S6, which are described in detail as follows:

[0069] Step S1, based on the topology structure of the low-voltage power grid and the change information of the power supply access point, the initial power supply path is reconstructed by adjusting the switching priority of the power supply access point, and the current power supply path is generated.

[0070] In the embodiments of the present application, the topology structure of the changed low-voltage power grid is first obtained, and the power supply path is reconstructed based on the topology structure. After the topology structure of the low-voltage power grid changes, the endpoint positions of each power supply line are obtained to construct an initial tree-shaped data structure, and then a corresponding node index matrix is constructed based on the tree-shaped data structure according to the coordinates of the power supply access points.

[0071] The voltage data of each node in the low-voltage power grid is collected in real time, and the voltage data is preprocessed. The preprocessing result is superimposed with the node index matrix to obtain a voltage data matrix. The power flow relationship between nodes in the voltage data matrix is tracked by using a recursive depth-first search algorithm, and then the topology structure of the current low-voltage power grid is generated in combination with the power supply connection state between nodes. The power flow is used to describe the flow rule of electric energy in the power grid, and can be used to evaluate the operation state of the power grid.

[0072] Specifically, the three standard deviation method is used to identify abnormal values of the voltage data of the nodes, and then the time series adjacent value average value replacement method is used to process missing values in the voltage data.

[0073] Optionally, in the embodiments of the present application, the power supply access point is a distributed photovoltaic power supply access point.

[0074] Then the Euclidean distance between nodes is calculated to obtain a node space distance matrix. Based on the node space distance matrix and the power supply connection state, a density-based spatial clustering algorithm is used to partition the nodes in the topology to divide a plurality of power supply areas. Then the power supply access points are classified according to the power supply areas, and the path from each power supply access point to the substation to which it belongs is calculated by a minimum spanning tree algorithm, so as to obtain the initial power supply path corresponding to each power supply access point.

[0075] As an improvement of the above scheme, the embodiments of the present application further segment the voltage data of each node in the power grid by the sliding time window method, establish a node voltage prediction model by the support vector regression algorithm to predict the voltage change trend of the node, and generate a node voltage prediction matrix. Meanwhile, a power supply path index table is constructed based on the distributed photovoltaic power supply access point, and the topology structure of the initial power supply path is adjusted by combining the node voltage prediction matrix and the power supply path index table. Further, in the process of adjusting the topology structure of the initial power supply path, a weight index is also used to evaluate the adjustment scheme, which includes the power supply path length, the voltage deviation value of the node, and the equipment load rate, etc.

[0076] However, due to the change of the topology structure of the low-voltage power grid, the node position and operating state of the power supply access point will also change, which will further affect the line loss distribution. Therefore, the embodiments of the present application reconstruct the power supply path of the low-voltage power grid for the power supply access point whose node position or operating state has changed, so as to more accurately reflect the actual line loss level.

[0077] Before reconstructing the power supply path of the low-voltage power grid, it is necessary to identify which power supply access points have changed, i.e. to identify the changed access points. The power flow is traced from each power supply access point, and the power direction of each power supply access point is determined by the node voltage amplitude and phase difference in the tracing process, and the feeder attribution of each power supply access point can be determined by the power direction, i.e. which feeder is responsible for transmitting electric energy from the substation to the user end. Then the feature vectors such as node voltage, active power and reactive power of each power supply access point are collected to construct the operating state identification matrix of each power supply access point. The operating state identification matrix is compared in time sequence, and the power supply access points whose positions and operating states have changed are determined by calculating the difference value of the operating state identification matrix of adjacent time sequences.

[0078] After determining the changed access point, the power supply and power consumption balance value of the changed access point is calculated to obtain a power mutual aid coefficient between nodes. The node importance of the power supply access point is calculated according to the power mutual aid coefficient between nodes, and the node importance is normalized by combining the maximum capacity of the feeder and the upper limit value of the power supply radius to obtain a node switching priority sequence for the power supply access point. The maximum capacity of the feeder can be obtained through the feeder attribution relationship of the voltage access point. The upper limit value of the power supply radius is the line length from the power supply access point to the farthest load point. The power mutual aid coefficient between nodes reflects the power supply and power consumption capacity between nodes, and the power generation capacity and load demand of the node itself are considered in the calculation process. Under the constraint of the maximum capacity of the feeder, the node switching priority is determined by the node importance.

[0079] The node switching priority is the priority of the feeder to be preferentially switched when the power supply access point needs to switch the power supply path. For example, when the rated capacity of a feeder is 400A and the current load rate of the feeder reaches 85%, if a power supply access point on the feeder needs to switch the power supply path, the adjacent feeder with a load rate lower than 60% is preferentially selected.

[0080] Based on the node switching priority sequence, the switching priority of the changed access point can be adjusted to reconstruct the power flow distribution of the initial power supply path, and the current power supply path is determined. The switching priority is the sequence of the power supply path to be preferentially switched when the power supply access point searches for a feasible path.

[0081] For example, in a power distribution network of an industrial park, the industrial park contains 12 10kV feeder lines, and a total of 36 voltage access points are provided. When the phase difference between nodes exceeds 15 degrees, it is determined that there is a power flow transmission relationship. When the voltage amplitude difference between adjacent nodes is less than 3%, it is determined that they are on the same feeder line. When determining the power flow direction, a power flow direction determination result matrix representing the power flow direction between nodes is constructed. In the matrix, 1 represents positive power flow, -1 represents negative power flow, and 0 represents no power flow. The node voltage of each power access point in the operating state identification matrix is the average value in the sampling period, and the active power and reactive power are the maximum values in the sampling period. In the power distribution network of the industrial park, the operating mode of the power access point can be divided into three categories: grid-connected power generation, off-grid standby, and fault shutdown. The operating state identification matrix uses one-hot encoding, and the dimension is the product of the number of voltage access points and the number of operating modes. When a power access point P15 changes from grid-connected power generation mode to off-grid standby mode, the power mutual aid coefficients of its adjacent nodes P14 and P16 are 0.8 and 0.6 respectively, indicating that node P14 has stronger support capability. When the power supply radius of a voltage access point exceeds 800 meters, the node importance degree decays to 50% of the original value. When the rated capacity of a feeder line is 400 amperes and the current load rate of the feeder line reaches 85%, if the power access point P15 on the feeder line needs to switch the power supply path, the adjacent feeder line with a load rate lower than 60% is preferentially selected.

[0082] In the power distribution network of the industrial park, taking the switching of the power access point as an example, the voltage gradually converges from 385V to 382V, the voltage variation of the adjacent 5 nodes is less than 1V, and the branch impedance parameter deviation is less than 0.5%. Finally, it is confirmed that the switching scheme is feasible, and the initial power supply path is reconstructed. In this way, the continuous grid-connected operation of the distributed power supply is ensured, and the stability of various operating indicators of the power distribution network is maintained. In practical applications, power flow reconstruction often involves the coordinated switching of multiple power access points. When power access points P15 and P18 change simultaneously, the switching operation is performed twice based on the node importance ranking results. After each switching, the power flow solution convergence needs to be verified to ensure the safety and controllability of the entire reconstruction process.

[0083] In step S2, the voltage response values of the power access points at each frequency point are collected, the inverse power flow characteristics of the frequency points are extracted from the voltage response values, the node impedance matrix of the current power supply path is corrected using the inverse power flow characteristics, and the power access point impedance parameter set is obtained.

[0084] The node voltage data and branch current data and other electrical data at each voltage access point of the current power supply path are collected, the node impedance matrix is calculated by the voltage and current phasor ratio, and the voltage and current measurement noise is filtered out by an adaptive Kalman filter during the construction of the matrix.

[0085] The harmonic injection device is used to apply an excitation signal to each power access point, and the voltage response value and current response value of each power access point at each frequency point are recorded. Then the voltage response value and current response value are processed by Fourier transform to obtain the impedance frequency characteristic curve of each power access point. The impedance frequency characteristic curve is determined using a preset voltage deviation range, and if there is a frequency point where the voltage response value exceeds the preset voltage deviation range, the reverse power flow data of the frequency point is extracted and wavelet transform is used to analyze the time-frequency characteristics of the data to obtain reverse power flow characteristics including reverse power flow duration, reverse power flow peak value and reverse power flow frequency statistics. The excitation signal applied can be of different frequencies.

[0086] According to the reverse power flow frequency statistics, the node impedance correction vector is estimated by the recursive least squares method, and then the node impedance correction vector is used to correct the node impedance matrix. Based on the corrected node impedance matrix, the impedance stability margin index is calculated, and the impedance stability margin index corresponding to each node is compared with the set margin threshold to determine the node stability state of the power access point, and a power access point impedance parameter set containing the impedance amplitude, impedance phase angle and node stability margin of the power access point is generated.

[0087] Further, the data in the power access point impedance parameter set is stored in time sequence, and a sliding time window method is used to establish an impedance change trend curve to reflect the dynamic impedance characteristics of the power input node. The node impedance characteristics can reflect the dynamic relationship between voltage and current, and are of great significance for the stability evaluation of the power access point.

[0088] For example, a 10kV distribution line is taken as an example, and 15 measurement points are set on the line, of which 5 are voltage access points, the node voltage sampling frequency is 10kHz, and the current sampling accuracy is 0.1A. When filtering the voltage and current measurement noise of the electrical data of a certain power access point, an adaptive Kalman filter is used to reduce the measurement noise from 0.5Ω to 0.05Ω. In the process of constructing the impedance frequency characteristic curve, the harmonic injection device generates excitation signals of 2 to 13 times, and the signal amplitude is 5% of the fundamental wave. The power access point P3 is tested, and the impedance value is 2.5Ω and the phase angle is 30 degrees at 50 Hz fundamental frequency, and it shows inductive characteristics as the frequency increases. When the frequency reaches 550 Hz, the impedance amplitude rises to 12Ω and the phase angle increases to 75 degrees, at which time the node voltage distortion rate exceeds 3%, which is beyond the voltage deviation range.

[0089] Optionally, in the above example, the threshold of the node stability margin is set to 2.5, when the maximum reverse power flow occurs at a certain power access point, the node stability margin decreases to 1.8, which is lower than the threshold, indicating that the node is in a potentially unstable state. The node impedance values are recorded at intervals of 1 minute, including impedance amplitude, impedance phase angle, node stability margin, and the impedance change trend of the node is analyzed by using a 60-minute sliding time window. Finally, by analyzing the impedance change trend curve, it is found that the node stability margin of a certain power access point is lower than the threshold for 4 times in a day, and the cumulative duration reaches 75 minutes. This continuous low-margin operation state reflects the potential stability risk of the point under high-power reverse power flow conditions, which needs to be adjusted to maintain the stability of the power grid.

[0090] Step S3, based on the reverse power flow characteristics, the current power supply path and the preset load balancing factor, the power transmission path is updated by calculating the load rate of the power equipment, and the power distribution characteristics are obtained.

[0091] In the embodiments of the present application, the equipment on the power supply path is classified by the reverse power flow duration, the reverse power flow peak value and the reverse power flow frequency statistics, and then the real-time node injection power value of each equipment is calculated. The power distribution is weighted in combination with the load balancing factor of each power supply area in the low-voltage power grid, and a load rate matrix is constructed to describe the load rate of each power equipment.

[0092] Further, the load rate matrix is normalized.

[0093] Then use the equipment rated capacity multiple as a reference value, and mark the power equipment whose load rate exceeds the reference value with an overload identifier. Based on the overload identifier, the load transfer demand of the corresponding power equipment is extracted, and the power mutual aid amount between each power supply area is calculated through the load transfer demand and the power supply area of the power equipment. According to the pre-constructed power transfer channel between the power supply areas, in combination with the power mutual aid amount, a load balancing optimization sequence for the power supply areas is obtained. Then, through the load balancing optimization sequence, the power supply level and the user capacity of each power supply area, the power distribution coefficient of each power supply area is updated. Wherein, the equipment rated capacity multiple is obtained by classifying the running state of the power equipment by support vector machine;

[0094] Optionally, the area with high importance gets a larger power distribution coefficient. In the embodiments of the present application, if the importance of a certain area is 0.9, which is 0.3 higher than that of an ordinary area, the corresponding power distribution ratio is increased by 15%.

[0095] According to the set power equipment power injection limit value and the power distribution coefficient, a power transmission path is planned by a minimum spanning tree algorithm, impedance and loss of each power transmission path are calculated, and power distribution characteristics are obtained. The power equipment power injection limit value is a threshold value of node injection power defined in load control.

[0096] Generally, when planning a power transmission path, the scheme with the minimum line loss is selected, and the load rate of each circuit breaker on the power supply path is controlled below 80%.

[0097] As an improvement of the above scheme, the embodiment of the application also uses the planned power transmission path and the calculated loss data to construct a power distribution strategy, and then verifies the power distribution strategy by device capacity margin and transmission efficiency double constraints. In actual situations, the power distribution characteristics are updated every 15 minutes to reflect the load change and power output fluctuation in time. When the power output rapidly changes to cause a 50% power drop of the device, the system needs to complete power redistribution within 1 minute to avoid device overload.

[0098] In addition, the power distribution strategy needs to be verified by device capacity and transmission efficiency double constraints. In the embodiment of the application, the capacity constraint requires that the load rate of each device does not exceed 90% of the rated value, and the transmission efficiency constraint limits the line loss rate to be lower than 3%.

[0099] Taking a 10kV distribution line as an example, the line includes 3 power supply areas, and 4 power source access points are arranged in each area. Through analysis of the reverse power flow characteristics of the line, it is found that the power source access point appears a continuous reverse power flow phenomenon for more than 30 minutes in the period of sufficient sunlight, the peak power reaches 80% of the grid capacity, and the average occurs 3 to 4 times per day. Moreover, under the typical summer sunlight conditions, the injection power of the power source access point presents obvious time sequence characteristics. Taking area A as an example, the real-time value of the injection power of the four power source access points reaches the maximum from 11 o'clock to 14 o'clock, and the cumulative injection power is 2000kW, accounting for 75% of the rated capacity of the transformer of area A. Assuming that the load balancing factor of area A is 0.85, the corresponding load rate matrix after weighting shows that the transformer of area A is in an overload state.

[0100] In the above example, the temperature, load rate and running time of the power equipment are classified by the support vector machine, and 1.2 times of the rated capacity of the device is determined as the reference value. When the load rate of the transformer of area A reaches 125%, an overload alarm is triggered, and 400kW load needs to be transferred to the adjacent area B.

[0101] For example, through the power distribution strategy of the embodiment of the present application, the transformer load rate of the area A is reduced to 85%, the power factor is increased to 0.95, and the line loss rate is maintained at about 2.5%. It is illustrated that through the reasonable power distribution strategy, the power supply demand of the important user is guaranteed, and the balanced utilization of the equipment capacity is realized.

[0102] In step S4, the topology structure, the current power supply path, the impedance parameter set of the power supply access point, and the power distribution feature are fused to construct a line loss distribution prediction model.

[0103] In the embodiment of the present application, the line length and the wire cross section of the line on which the node is located are determined through the topology structure and the current power supply path of the low-voltage power grid, and are taken as the power supply path weight between nodes to generate a node connection weight matrix. Then, the electrical characteristics of the nodes in the node connection weight matrix are described through the impedance parameter set of the power supply access point, and the electrical feature matrix is obtained through the normalization processing of the power distribution data.

[0104] Then, the topology structure feature vector is extracted by using the graph embedding algorithm according to the correlation strength between the nodes in the electrical feature matrix, the spatial relationship between the nodes is modeled by performing multi-layer convolution operation on the topology structure feature vector, and the topology space feature matrix is generated. The spatio-temporal correlation feature vector is obtained by performing time sequence sampling on the topology space feature matrix, and performing multi-scale decomposition on the node impedance and the power distribution feature through wavelet transform and performing time sequence correlation analysis on the decomposed feature sequence.

[0105] For example, the graph embedding algorithm is used in the embodiment of the present application to map the 25-dimensional node feature vector to an 8-dimensional feature space, and the topological correlation information between the nodes is retained. The multi-layer convolution operation uses a 3x3 convolution kernel, and the spatial relationship features between the nodes are extracted through 3 layers of convolution. Batch normalization processing is used after each layer of convolution, which effectively improves the stability of feature extraction.

[0106] In addition, in the embodiment of the present application, the db4 wavelet basis function is selected to perform 8-layer wavelet decomposition on the node impedance and the power distribution feature, and the feature sequence of different frequency bands is obtained. It can be seen from the feature sequence that, due to the power fluctuation of the distributed photovoltaic power supply access point under typical sunlight conditions, the impedance change caused by the power fluctuation mainly concentrates in the 3rd to 5th layer wavelet coefficients, which is highly consistent with the actual line loss change rule.

[0107] According to the above operation, the obtained spatio-temporal correlation feature vector contains factors such as power supply path change, impedance characteristic change, and load fluctuation caused by the topological change of the power grid, which can assist the line loss distribution prediction model constructed subsequently to more accurately perform line loss distribution prediction.

[0108] Further, based on the space-time correlation feature vector, a sample data set is constructed, then the sample data set is labeled by using the measured line loss data, and the training sample size is expanded by using a data enhancement method to obtain a standardized training data set.

[0109] In the embodiment of the application, the construction of the training sample adopts 30-day historical operation data, containing 36000 data points. Through time translation and noise superposition for data enhancement, the expanded training set reaches 108000 samples. The samples contain different weather conditions, load levels and photovoltaic output states, and have strong representativeness.

[0110] Then, the model based on the residual neural network is trained by using the standardized training data set, the network parameters are optimized by using the cross-validation method in the training process, and a line loss distribution prediction model is constructed. The residual neural network adopts a 16-layer structure, each residual block contains 2 convolution layers, and the network parameters are optimized by five-fold cross-validation.

[0111] Taking a 10kV distribution network as an example, the distribution network contains 25 distribution nodes, of which 8 are distributed photovoltaic power access points, and the dynamic impedance parameters of the power access points fluctuate in the range of 0.5 to 2Ω. When the solar intensity increases from 200 watts per square meter to 800 watts per square meter, the power output of a power access point increases from 50 kilowatts to 200 kilowatts. The line loss distribution prediction model can accurately capture the trend of the line loss rate, which increases from the initial 2.8% to 3.5%. Moreover, through the online correction mechanism of the model, the prediction error can be controlled within 0.2%. According to the prediction results output by the line loss distribution prediction model, it can be known that during the period of intense photovoltaic output fluctuation, the line loss distribution presents obvious space-time correlation. From 9 o'clock to 11 o'clock in the morning, with the increase of photovoltaic output, the line loss rate at the end of the feeder decreases significantly, while the line loss rate of the section close to the substation changes little. This non-uniform distribution characteristic is highly consistent with the measured data, verifying the adaptability of the prediction model to complex working conditions. Continuous prediction tracking shows that the model can adapt to the changes of photovoltaic output characteristics in different seasons, and the prediction accuracy remains stable.

[0112] Step S5, inputting real-time operation data of the low-voltage power grid into the line loss distribution prediction model, identifying line loss abnormal areas and performing regional priority sorting to determine a line loss treatment priority sequence.

[0113] In the embodiment of the present application, real-time operation data and environmental data of the power distribution network are collected, including temperature, humidity, meteorological data and equipment operation state data, and then combined with time-of-use electricity price and load scheduling rules, the line loss rate of each node in the current power supply path is calculated through a line loss distribution prediction model to generate a real-time line loss distribution table. According to the real-time line loss distribution data table, the voltage deviation value of each node is extracted, and then the voltage qualified rate of each power supply area in the low-voltage power grid is calculated through the voltage deviation value, and the power supply area with the voltage qualified rate exceeding the preset voltage qualified rate threshold is regarded as a line loss abnormal area.

[0114] The power consumption load curve of each node in the line loss abnormal area is obtained from the power distribution automation master station, the node voltage out-of-limit degree is calculated according to the power supply voltage deviation limit value, the power consumption demand in each period is counted through the power consumption load curve to generate a regional power consumption feature table. Based on the regional power consumption feature table, users are classified, a user load characteristic classification tree is constructed by using a hierarchical clustering algorithm, user power supply is graded according to the power supply guarantee level, and a user power supply level grading result table is output. The data in the user power supply level grading result table is standardized, a comprehensive score of voltage deviation improvement and power consumption demand satisfaction is calculated by using a multi-objective optimization algorithm, and a regional governance scoring table is generated. According to the regional governance scoring table, a priority evaluation function is constructed, the weight coefficients of the three dimensions of voltage deviation improvement, power consumption demand satisfaction and user power supply level are calculated by using the analytic hierarchy process, and a line loss governance priority sequence is output.

[0115] Among them, the user power consumption characteristic is reflected by the power consumption load curve. The hospital power consumption load fluctuation rate is within 10%, and the whole day maintains a high load level. The commercial area power consumption load presents obvious peak-valley characteristics, and the maximum load appears at 19:00-21:00, which is 2.5 times the valley load. The industrial park power consumption load shows strong periodicity, and the load change law of the morning shift, the middle shift and the evening shift is obvious. Moreover, the voltage out-of-limit situation of the commercial area is 7% lower than the voltage during the power consumption peak period, which exceeds the limit value requirement of ±5%.

[0116] In addition, the user type can be roughly divided into three layers according to the regional power consumption feature table classification. The first layer is divided into three types of industry, commerce and civil according to the power consumption property, the second layer is divided into large, medium and small users according to the power consumption scale, and the third layer is divided into special, first-level and second-level users according to the power supply guarantee requirement. The user power supply level specifically includes: important users such as hospitals, communications and finance are classified into the special guarantee category, and the power supply reliability requirement reaches 99.9%. Commercial complexes are classified into the first-level guarantee category, and the power supply reliability requirement is 99.5%. Ordinary civil load is classified into the second-level guarantee category, and the power supply reliability requirement is 99%.

[0117] Taking a 10kV distribution network as an example, the collected environmental data covers meteorological parameters such as temperature, humidity, and sunshine intensity, and the collected real-time operation data of the distribution network includes transformer temperature, switch state and other information. When the summer temperature reaches 35 degrees Celsius, the transformer temperature rises to 65 degrees Celsius, and the line loss rate increases to 4.5%, which is 0.8 percentage points higher than that at normal temperature. Combined with time-of-use electricity price, the peak-to-valley electricity price ratio is 3:1, and the electricity load reaches the maximum value from 14:00 to 16:00. When the voltage qualification rate is lower than 95%, the region is marked as a line loss abnormal region, and the voltage deviation value of the power supply region can be calculated. Among the 15 monitored power supply regions, 3 line loss abnormal regions are located at the network end with a power supply radius exceeding 2 kilometers.

[0118] For example, in the process of outputting the line loss treatment priority sequence, the voltage deviation improvement amount and the electricity demand satisfaction degree are normalized to achieve multi-objective optimization. If the target value of the voltage deviation improvement amount of a line loss abnormal region is set to 3%, and the target value of the electricity demand satisfaction degree is set to 98%, the weight coefficients of the three dimensions of voltage deviation improvement amount, electricity demand satisfaction degree, and user power supply level are calculated as 0.4, 0.3, and 0.3, respectively. Based on the weight coefficients, the line loss treatment priority sequence is obtained, in which the hospital power supply region ranks the highest with a comprehensive score of 0.92, followed by the industrial park with a score of 0.85, and then the commercial area with a score of 0.78. This multi-dimensional priority ranking ensures the rational allocation of power supply resources and the quality of electricity for important loads.

[0119] Step S6, adjusting the node injection power of the line loss abnormal region according to the line loss treatment priority sequence.

[0120] According to the line loss treatment priority sequence, read the electricity load data of each line loss abnormal region, calculate the upper limit value of the node injection power of each period through the regional electricity characteristic table and the electricity load data, and generate an initial power distribution scheme. The regional electricity characteristic table is obtained through time-of-use electricity price and a preset load scheduling rule.

[0121] Then, the initial power distribution scheme is corrected in combination with the load balancing factor of the line loss abnormal region, the initial power distribution scheme is optimized by calculating the power mutual aid capacity of each region, and an optimized power distribution scheme is obtained.

[0122] Based on the optimized power distribution scheme, execute the power adjustment instruction, and based on the electricity characteristics, adjust the node injection power of each node in each line loss abnormal region through the upper limit value of the node injection power.

[0123] As an improvement of the above scheme, the embodiment of the present application considers the power loss in the transmission process when constructing the power distribution scheme, and therefore constructs a power transmission cost function, in which the line loss rate weight is 0.4 and the voltage deviation weight is 0.6.

[0124] While adjusting the node injection power of the node, real-time values of voltage and current are obtained from the power distribution automation master station, the change of switch state and the switching time stamp in the power supply path switching process are recorded, and a mode switching log is generated, which records the switch action sequence and the like, including the voltage and current values, power factor and other key parameters at the switching time. The execution result is evaluated by using a line loss distribution prediction model, the voltage deviation rate, power factor and load rate of each node are recorded through a time sequence database, and an operating state feature set is constructed. The operating state feature set and the mode switching log are input into the line loss distribution prediction model for iterative training, the prediction parameters are updated by using a sliding time window, and the corrected line loss distribution prediction value is obtained. Based on the corrected line loss distribution prediction value, the switching priority sequence is updated, the power supply path is re-planned by using the shortest path algorithm, the power transmission feasibility is verified by using the maximum flow algorithm, and the continuously optimized line loss treatment scheme is output.

[0125] In the operating state feature set, the early warning threshold of the voltage deviation rate of the node is set to ±5%, the lower limit value of the power factor is 0.9, and the upper limit value of the load rate is 85%.

[0126] For example, in the process of one-time power supply line switching in the embodiment of the present application, the tie switch SC1 acts at 10:30, and the corresponding voltage rises from 378V to 382V and the power factor increases from 0.92 to 0.95.

[0127] In the embodiment of the present application, the length of the sliding time window is 15 minutes, and the parameters of the line loss distribution prediction model need to be updated every 15 minutes in the period of intense fluctuation of photovoltaic output, so as to ensure the real-time of the prediction result. In other embodiments, the length of the sliding time window can also be 30 minutes.

[0128] Taking a 10kV distribution network as an example, the distribution network is provided with three distributed photovoltaic power supply access points, and the total installed capacity is 1500kW, and the daily power generation fluctuation range is between 6000 and 8000kWh. One day is divided into 96 time periods, and the upper limit value of the node injection power of each time period is determined according to the power consumption characteristics. In the peak load period of 14:00-16:00, the upper limit of the node injection power of the photovoltaic power generation is set to 90% of the rated capacity, and in the load valley period, the upper limit of the node injection power is reduced to 60%. In the distribution network, if the load balancing factor of a power supply area A is 0.85, and the power interconnection capacity with the adjacent area B is 300kW. When the photovoltaic output of the power supply area A reaches 500kW, and the load of the power supply area B is only 300kW, 150kW is transmitted to the power supply area B through power interconnection, so that the load rate of the two areas tends to be balanced, and load balancing is realized. When it is detected that the photovoltaic output of the power supply area A decreases by 50% within 30 minutes, the line loss distribution prediction model timely adjusts the power distribution scheme of the area, and the priority sequence is updated accordingly. Through the shortest path algorithm, the power supply path is re-planned, part of the load of the power supply area A is transferred to the adjacent area, and the transmission capacity of the new path is verified through the maximum flow algorithm. In addition, during the adjustment process, the voltage of each node is kept within the qualified range, and the line loss rate is controlled within the expected level.

[0129] The embodiments of the present application have the following beneficial effects:

[0130] The embodiments of the present application first reconstruct the power supply path by collecting the change of the power supply access point, obtain the current power supply path to real-time display the running state of each node. Then, the impedance characteristic change brought by the change of the power supply access point is measured by collecting the voltage response value of the power supply access point, and the influence of the reverse power flow in the power supply line on the voltage rise is determined by collecting the reverse power flow data at the moment when the voltage response value exceeds the preset voltage deviation range, so as to correct the node impedance matrix and obtain the impedance parameter set of the power supply access point after the change of the impedance characteristic. The change characteristics of the node impedance in the low-voltage power grid are accurately obtained. Then, whether there is a situation that the load rate is too high in each power supply area on the current power supply path is calculated, and the power transmission path is updated based on the demand of load transfer, the power distribution strategy is reconstructed for each power supply area, and the power distribution characteristics are extracted, which provides data support for the load balancing of the low-voltage power grid for the subsequent line loss management strategy. The line loss distribution prediction model constructed considers the influence of the change of the power supply path, the change of the impedance characteristic and the load fluctuation on the line loss caused by the change of the topology of the power grid, so that the predicted line loss distribution is more in line with the actual situation, the output line loss management priority sequence is reasonable, and the line loss management and resource allocation can be reasonably and accurately realized, and the energy utilization rate and power supply reliability of the power grid are improved.

[0131] Second embodiment

[0132] Further, in order to implement the low-voltage power grid multi-source data fusion based line loss management system corresponding to the above-mentioned method embodiment, the corresponding functions and technical effects are realized, Figure 2 A structural diagram of the low-voltage power grid multi-source data fusion based line loss management system is provided. For ease of illustration, only the part related to the present embodiment is shown. The low-voltage power grid multi-source data fusion based line loss management system provided by the present embodiment comprises:

[0133] The power supply path reconstruction module 201 is configured to reconstruct the initial power supply path by adjusting the switching priority of the power supply access point based on the topology of the low-voltage power grid and the change information of the power supply access point, and generate the current power supply path.

[0134] In the present embodiment, the topology is processed by partitioning according to the power supply connection state between nodes in the low-voltage power grid, and the initial power supply path is obtained;

[0135] Based on the topology, the power flow direction of each power supply access point is tracked by a shortest path search algorithm to collect the change information, and the power supply access point whose node position or running state is changed is identified as a changed access point;

[0136] The power supply and reception power balance value of the changed access point is calculated, and the initial power supply path is reconstructed by adjusting the switching priority of the changed access point according to the power supply and reception power balance value, and the current power supply path is obtained.

[0137] The impedance characteristic extraction module 202 is configured to collect the voltage response value of the power supply access point at each frequency point, extract the reverse power flow characteristic of the frequency point by the voltage response value, correct the node impedance matrix of the current power supply path using the reverse power flow characteristic, and obtain the impedance parameter set of the power supply access point.

[0138] In the present embodiment, the electrical data of each power supply access point under the current power supply path is collected, and the node impedance matrix of the current power supply path is constructed;

[0139] A harmonic injection device is used to apply an excitation signal to each power supply access point, and the voltage response value of each power supply access point at each frequency point is recorded.

[0140] If there is a frequency point whose voltage response value exceeds the preset voltage deviation range, the reverse power flow characteristic of the frequency point is extracted by wavelet transform; wherein the reverse power flow characteristic includes reverse power flow duration, reverse power flow peak value and reverse power flow frequency statistics.

[0141] The load fluctuation feature extraction module 203 is configured to update the power transmission path by calculating the load rate of the power equipment based on the reverse power flow feature, the current power supply path and the preset load balancing factor, and obtain a power distribution feature.

[0142] In the embodiment of the present application, the equipment on the power supply path is classified by the reverse power flow duration, the reverse power flow peak value and the reverse power flow frequency statistics, and then the real-time node injection power value of each equipment is calculated. The power distribution is weighted in combination with the load balancing factor of each power supply area in the low-voltage power grid, a load rate matrix is constructed to describe the load rate of each power equipment.

[0143] Further, the load rate matrix is normalized.

[0144] Then, the equipment rated capacity multiple is used as a reference value, and an overload identifier is marked to the power equipment whose load rate exceeds the reference value. The load transfer demand of the corresponding power equipment is extracted based on the overload identifier, and the power mutual aid amount between each power supply area is calculated through the load transfer demand and the power supply area of the power equipment. According to the pre-constructed power transfer channel between the power supply areas, the power mutual aid amount is combined to obtain a load balancing optimization sequence for the power supply areas. Then, the power distribution coefficient of each power supply area is updated through the load balancing optimization sequence, the power supply level and the user capacity of each power supply area.

[0145] According to the set power equipment power injection limit value and the power distribution coefficient, the power transmission path is planned through the minimum spanning tree algorithm, the impedance and loss of each power transmission path are calculated, and a power distribution feature is obtained. The power equipment power injection limit value is a threshold value of the node injection power in the load control.

[0146] As an improvement of the above scheme, the embodiment of the present application also uses the planned power transmission path and the calculated loss data to construct a power distribution strategy, and then verifies the power distribution strategy through the dual constraints of equipment capacity margin and transmission efficiency.

[0147] The model construction module 204 is configured to fuse the topology structure, the current power supply path, the power source access point impedance parameter set and the power distribution feature to construct a line loss distribution prediction model.

[0148] In the embodiment of the present application, the line length and the conductor cross section are determined through the topology structure and the current power supply path of the low-voltage power grid, and a node connection weight matrix is generated. Then, the node electrical characteristics in the node connection weight matrix are described through the power source access point impedance parameter set, and the power distribution data is normalized to obtain an electrical feature matrix.

[0149] Then, according to the correlation strength between nodes in the electrical characteristic matrix, a topology structure feature vector is extracted by using a graph embedding algorithm, the spatial relationship between nodes is modeled by performing multi-layer convolution operation on the topology structure feature vector, and a topology space feature matrix is generated. The topology space feature matrix is then time-series sampled, the node impedance and power distribution features are multi-scale decomposed by wavelet transform, and time-series correlation analysis is performed on the decomposed feature sequence to obtain a space-time correlation feature vector.

[0150] According to the above operation, the obtained space-time correlation feature vector contains factors such as power supply path change, impedance characteristic change and load fluctuation caused by topology change of the power grid, which can assist the subsequent line loss distribution prediction model to more accurately perform line loss distribution prediction.

[0151] Further, a sample data set is constructed based on the space-time correlation feature vector, then the sample data set is labeled by using the measured line loss data, and the training sample size is expanded by using a data enhancement method to obtain a standardized training data set.

[0152] Then, a model based on a residual neural network is trained by using the standardized training data set, the network parameters are optimized by using a cross-validation method during the training process, and a line loss distribution prediction model is constructed.

[0153] The line loss abnormal area identification module 205 is configured to input the real-time operation data of the low-voltage power grid into the line loss distribution prediction model, identify line loss abnormal areas and perform regional priority sorting, and determine a line loss treatment priority sequence.

[0154] In the embodiments of the present application, real-time operation data and environmental data of the low-voltage power grid are obtained.

[0155] The real-time operation data, the environmental data and a preset load scheduling rule are input into the line loss distribution prediction model, the line loss rate of each node in the current power supply path is calculated, and a real-time line loss distribution table is generated.

[0156] The voltage qualified rate of each power supply area in the low-voltage power grid is calculated according to the voltage deviation value of each node in the current power supply path, wherein the voltage deviation value is extracted according to the real-time line loss distribution table.

[0157] The power supply area with the voltage qualified rate exceeding a preset voltage qualified rate threshold is regarded as a line loss abnormal area.

[0158] The line loss abnormal areas are prioritized according to the user power supply level and the electricity demand of the line loss abnormal areas, and a line loss treatment priority sequence is obtained.

[0159] The line loss treatment module 206 is configured to adjust the node injection power of the low-voltage power grid according to the line loss treatment priority sequence.

[0160] In the embodiment of the present application, according to the line loss management priority sequence, the electricity load data of each line loss abnormal area is read, the node injection power upper limit value of each period is calculated through the area electricity consumption characteristic table and the electricity load data, and an initial power distribution scheme is generated. The area electricity consumption characteristic table is obtained through the time-of-use electricity price and the preset load scheduling rule.

[0161] Then the initial power distribution scheme is corrected in combination with the load balancing factor of the line loss abnormal area, the initial power distribution scheme is optimized through the calculation of the power mutual aid capacity of each area, and an optimized power distribution scheme is obtained.

[0162] Based on the optimized power distribution scheme, a power adjustment instruction is executed, and based on the electricity consumption characteristics, the node injection power of each node in the line loss abnormal area is adjusted in turn through the node injection power upper limit value.

[0163] While adjusting the node injection power of the node, the voltage and current real-time values are obtained from the power distribution automation master station, the switch state change and switching time stamp in the power supply path switching process are recorded, and a mode switching log is generated. The execution result is evaluated by using a line loss distribution prediction model, the voltage deviation rate, power factor and load rate of each node are recorded through a time series database, and an operating state feature set is constructed. The operating state feature set and the mode switching log are input into the line loss distribution prediction model for iterative training, the prediction parameters are updated by using a sliding time window, and a corrected line loss distribution prediction value is obtained. Based on the corrected line loss distribution prediction value, the switching priority sequence is updated, the power supply path is re-planned by using the shortest path algorithm, the power transmission feasibility is verified by using the maximum flow algorithm, and a continuously optimized line loss management scheme is output.

[0164] In the embodiment of the present application, the length of the sliding time window is 15 minutes.

[0165] In some embodiments, the power supply path reconstruction module 201, in particular:

[0166] First, the topology structure of the changed low-voltage power grid is obtained to reconstruct the power supply path. After the topology structure of the low-voltage power grid changes, the endpoint positions of each power supply line are obtained to construct an initial tree-shaped data structure, and then a corresponding node index matrix is constructed according to the coordinates of the power supply access points based on the tree-shaped data structure.

[0167] The voltage data of each node in the low-voltage power grid is collected in real time, and the voltage data is preprocessed. The preprocessing result is superimposed with the node index matrix to obtain a voltage data matrix. A recursive depth-first search algorithm is used to track the power flow relationship between nodes in the voltage data matrix, and then the topology structure of the current low-voltage power grid is generated in combination with the power supply connection state between nodes. The power flow is used to describe the flow rule of electric energy in the power grid, and can be used to evaluate the operation state of the power grid.

[0168] Specifically, the three standard deviation method is used to identify abnormal values of the voltage data of the nodes, and then the time series adjacent value average value replacement method is used to process missing values in the voltage data.

[0169] Optionally, in the embodiment of the application, the power supply access point is a distributed photovoltaic power supply access point.

[0170] Then, the Euclidean distance between nodes is calculated to obtain a node space distance matrix. Based on the node space distance matrix and the power supply connection state, a density-based spatial clustering algorithm is used to partition the nodes in the topology structure to divide a plurality of power supply areas. Then, the power supply access points are classified according to the power supply areas, and the minimum spanning tree algorithm is used to calculate the path from each power supply access point to the substation to which it belongs, so as to obtain the initial power supply path corresponding to each power supply access point.

[0171] As an improvement of the above scheme, the voltage data of each node in the power grid is segmented by the sliding time window method, and a node voltage prediction model is established by the support vector regression algorithm to predict the voltage trend of the node and generate a node voltage prediction matrix. Meanwhile, a power supply path index table is constructed based on the distributed photovoltaic power supply access point, and the topology structure of the initial power supply path is adjusted by combining the node voltage prediction matrix and the power supply path index table. Further, in the process of adjusting the topology structure of the initial power supply path, a weight index is also used to evaluate the adjustment scheme, and the weight index includes the power supply path length, the voltage deviation value of the node, and the equipment load rate, etc.

[0172] However, due to the change of the topology structure of the low-voltage power grid, the node position and operation state of the power supply access point also change, which further affects the line loss distribution. Therefore, the low-voltage power grid power supply path is reconstructed for the power supply access point whose node position or operation state has been changed, so as to more accurately reflect the actual line loss level.

[0173] Before reconstructing the power supply path of the low-voltage power grid, it is necessary to identify which power supply access points have changed, i.e., to identify the changed access points. The power flow is traced from each power supply access point, and the power direction of each power supply access point is determined by the node voltage amplitude and phase difference during the tracing process. The feeder attribution of each power supply access point can be determined by the power direction, i.e., which feeder is responsible for transmitting power from the transformer substation to the user end. Then, the node voltage, active power, and reactive power, etc. of each power supply access point are collected to construct the operating state identification matrix of each power supply access point. The operating state identification matrix is compared in time sequence, and the power supply access points with changes in location and operating state are determined by calculating the difference between the operating state identification matrices of adjacent time sequences.

[0174] After determining the changed access points, the power supply and power consumption balance values of the changed access points are calculated to obtain the power mutual aid coefficient between nodes. The node importance of the power supply access point is calculated according to the power mutual aid coefficient between nodes, and the node importance is normalized by combining the feeder maximum capacity and the upper limit value of the power supply radius to obtain the node switching priority sequence for the power supply access point. The feeder maximum capacity can be obtained through the feeder attribution of the voltage access point. The upper limit value of the power supply radius is the line length from the power supply access point to the farthest load point.

[0175] Based on the node switching priority sequence, the switching priority of the changed access points can be adjusted to reconstruct the power flow distribution of the initial power supply path and determine the current power supply path.

[0176] In some embodiments, the impedance characteristic extraction module 202, in particular:

[0177] The node voltage data and branch current data, etc. are collected at each voltage access point of the current power supply path, the node impedance matrix is calculated by the voltage and current phasor ratio, and the voltage and current measurement noise is filtered out by the adaptive Kalman filter during the matrix construction process.

[0178] A harmonic injection device is used to apply an excitation signal to each power supply access point, and the voltage response value and current response value of each power supply access point at each frequency point are recorded. Then, the voltage response value and current response value are processed by Fourier transform to obtain the impedance frequency characteristic curve of each power supply access point. The impedance frequency characteristic curve is determined using a preset voltage deviation range. If there is a frequency point where the voltage response value exceeds the preset voltage deviation range, the reverse power flow data of the frequency point is extracted and wavelet transform is used to analyze the time-frequency characteristics of the data to obtain the reverse power flow characteristics including the reverse power flow duration, reverse power flow peak value, and reverse power flow frequency statistics. The excitation signal applied can be of different frequencies.

[0179] According to the reverse power flow frequency statistics, the node impedance correction vector is estimated by the recursive least square method, and then the node impedance correction vector is used to correct the node impedance matrix. Based on the corrected node impedance matrix, the impedance stability margin index is calculated, and the impedance stability margin index corresponding to each node is compared with the set margin threshold to judge the node stability state of the power access point, and the impedance parameter set of the power access point containing the impedance amplitude, impedance phase angle and node stability margin of the power access point is generated.

[0180] Further, the data in the power access point impedance parameter set is stored in time sequence, and the sliding time window method is used to establish the impedance change trend curve to reflect the dynamic impedance characteristics of the power input node. The node impedance characteristics can reflect the dynamic relationship between voltage and current, and are of great significance to the stability evaluation of the power access point.

[0181] For example, taking a 10kV distribution line as an example, 15 measurement points are set on the line, of which 5 are voltage access points, the node voltage sampling frequency is 10kHz, and the current sampling accuracy is 0.1A. When filtering the voltage and current measurement noise of the electrical data of a certain power access point, the adaptive Kalman filter is used to reduce the measurement noise from 0.5Ω to 0.05Ω. In the process of constructing the impedance frequency characteristic curve, the harmonic injection device generates excitation signals of 2 to 13 times, and the signal amplitude is 5% of the fundamental wave. The power access point P3 is tested, and the impedance value is 2.5Ω and the phase angle is 30 degrees at 50 Hz fundamental frequency. With the increase of frequency, it shows inductive characteristics. When the frequency reaches 550 Hz, the impedance amplitude rises to 12Ω, and the phase angle increases to 75 degrees. At this time, the node voltage distortion rate exceeds 3%, which is beyond the voltage deviation range.

[0182] Optionally, in the above example, the threshold value of the node stability margin is set to 2.5, and when a certain power access point appears the maximum reverse power flow, the node stability margin decreases to 1.8, which is lower than the threshold value, indicating that the node is in a potential unstable state. The node impedance value is recorded at an interval of 1 minute, including the data of impedance amplitude, impedance phase angle and node stability margin, and the sliding time window of 60 minutes is used to analyze the impedance change trend of the node. Finally, through the analysis of the impedance change trend curve, it is found that a certain power access point has 4 times of node stability margin lower than the threshold value in a day, and the cumulative duration reaches 75 minutes. This continuous low margin operation state reflects the potential stability risk of the point under high-power reverse power flow conditions, which needs to be adjusted to maintain the stability of the power grid.

[0183] In some embodiments, the line loss anomaly area identification module 205, in particular:

[0184] The real-time operation data and environmental data of the power distribution network are collected, including temperature, humidity, weather data and equipment operation state data, and then combined with time-of-use electricity price and load scheduling rules, the line loss rate of each node in the current power supply path is calculated through the line loss distribution prediction model to generate a real-time line loss distribution table. According to the real-time line loss distribution data table, the voltage deviation value of each node is extracted, and then the voltage qualified rate of each power supply area in the low-voltage power grid is calculated through the voltage deviation value, and the power supply area with the voltage qualified rate exceeding the preset voltage qualified rate threshold is regarded as the line loss abnormal area.

[0185] The power consumption load curve of each node in the line loss abnormal area is obtained from the power distribution automation master station, the node voltage overrun degree is calculated according to the power supply voltage deviation limit value, the power consumption demand in each period is counted through the power consumption load curve, and a regional power consumption feature table is generated. Based on the regional power consumption feature table, the users are classified, a user load characteristic classification tree is constructed by using a hierarchical clustering algorithm, the user power supply is graded according to the power supply guarantee level, and a user power supply grade classification result table is output. The data in the user power supply grade classification result table is standardized, a comprehensive score of voltage deviation improvement and power consumption demand satisfaction is calculated by using a multi-objective optimization algorithm, and a regional governance scoring table is generated. According to the regional governance scoring table, a priority evaluation function is constructed, the weight coefficients of the three dimensions of voltage deviation improvement, power consumption demand satisfaction and user power supply level are calculated by using the analytic hierarchy process, and a line loss governance priority sequence is output.

[0186] Among them, the user power consumption characteristic is reflected by the power consumption load curve. The hospital power consumption load fluctuation rate is within 10%, and the whole day maintains a high load level. The commercial area power consumption load presents obvious peak-valley characteristics, and the maximum load appears from 19:00 to 21:00, which is 2.5 times the valley load. The industrial park power consumption load shows strong periodicity, and the load change law of the morning shift, the middle shift and the evening shift is obvious. Moreover, the voltage overrun of the commercial area is as low as 7% during the power consumption peak period, which exceeds the limit value requirement of ±5%.

[0187] In addition, the user types can be roughly divided into three layers according to the regional power consumption feature table classification. The first layer is divided into industrial, commercial and civil three categories according to the power consumption property, the second layer is divided into large, medium and small users according to the power consumption scale, and the third layer is divided into special, first and second level users according to the power supply guarantee requirement. The user power supply level specifically includes: important users such as hospitals, communications and finance are classified into special guarantee category, and the power supply reliability requirement reaches 99.9%. Commercial complexes are classified into first-class guarantee category, and the power supply reliability requirement is 99.5%. Ordinary civil load is classified into second-class guarantee category, and the power supply reliability requirement is 99%.

[0188] Taking a 10kV power distribution network as an example, the collected environmental data covers meteorological parameters such as air temperature, humidity, and sunshine intensity, and the collected real-time operation data of the power distribution network includes transformer temperature, switch state and other information. When the summer temperature reaches 35 degrees Celsius, the transformer temperature rises to 65 degrees Celsius, and the line loss rate increases to 4.5%, which is 0.8 percentage points higher than that at normal temperature. Combined with time-of-use electricity price, the peak-to-valley electricity price ratio is 3:1, and the electricity load reaches the maximum value from 14:00 to 16:00. When the voltage qualified rate is lower than 95%, the region is marked as a line loss abnormal region, and the voltage deviation value of the power supply region can be calculated. Among the 15 monitored power supply regions, 3 line loss abnormal regions are located at the network end with a power supply radius exceeding 2 kilometers.

[0189] For example, in the process of outputting the line loss treatment priority sequence, the voltage deviation improvement amount and the electricity demand satisfaction degree are normalized to realize multi-objective optimization. If the target value of the voltage deviation improvement amount of a line loss abnormal region is set to 3%, and the target value of the electricity demand satisfaction degree is set to 98%, the weight coefficients of the three dimensions of voltage deviation improvement amount, electricity demand satisfaction degree and user power supply level are calculated as 0.4, 0.3 and 0.3, respectively. Based on the weight coefficients, the line loss treatment priority sequence is obtained, in which the hospital power supply region ranks the highest with a comprehensive score of 0.92, followed by the industrial park with a score of 0.85, and then the commercial area with a score of 0.78. This multi-dimensional priority ranking ensures the reasonable allocation of power supply resources and the electricity quality of important loads.

[0190] Implementing the embodiments of the present application has the following beneficial effects:

[0191] The embodiments of the present application reconstruct the power supply path by collecting the change of the power supply access point first, obtain the current power supply path to show the running state of each node in real time. Then the voltage response value of the power supply access point is collected to measure the impedance characteristic change caused by the change of the access point, and the reverse power flow data at the moment when the voltage response value exceeds the preset voltage deviation range is collected to determine the influence of the reverse power flow in the power supply line on the voltage rise, so as to correct the node impedance matrix, obtain the impedance parameter set of the power supply access point after the impedance characteristic changes, and accurately obtain the change characteristics of the node impedance in the low-voltage power grid. Then it is calculated whether there is a situation that the load rate is too high in each power supply area on the current power supply path, and the power transmission path is updated based on the demand of load transfer, the power distribution strategy is reconstructed for each power supply area, and the power distribution characteristics are extracted, which provides data support for the load balancing of the low-voltage power grid for the subsequent line loss management strategy. The line loss distribution prediction model constructed then considers the influence of the change of the power supply path caused by the topology change of the power grid, the change of the impedance characteristic and the load fluctuation on the line loss, so the predicted line loss distribution is more in line with the actual situation, the output line loss management priority sequence is reasonable, and the line loss management and resource allocation can be realized reasonably and accurately, and the energy utilization rate and power supply reliability of the power grid are improved.

[0192] The above specific embodiments further illustrate the purpose, technical solutions and advantages of the present application. It should be understood that the above description is only for specific embodiments of the present application and does not limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A line loss management method based on low-voltage power grid multi-source data fusion, characterized in that, The method comprises the following steps: Based on the topology structure of the low-voltage power grid and the change information of the power access point, the initial power supply path is reconstructed by adjusting the switching priority of the power access point to generate the current power supply path; Collect the voltage response value of the power access point at each frequency point, extract the reverse power flow characteristics of the frequency point through the voltage response value, and correct the node impedance matrix of the current power supply path using the reverse power flow characteristics to obtain the impedance parameter set of the power access point; Based on the reverse power flow characteristics, the current power supply path and the preset load balancing factor, the power transmission path is updated by calculating the load rate of the power equipment to obtain the power distribution characteristics; Fuse the topology structure, the current power supply path, the impedance parameter set of the power access point and the power distribution characteristics to construct a line loss distribution prediction model, specifically: according to the topology structure and the current power supply path, determine the line length and the conductor cross section, generate the node connection weight matrix, and describe the node electrical characteristics in the node connection weight matrix through the impedance parameter set of the power access point to obtain the electrical characteristic matrix; Through wavelet transform, the electrical characteristic matrix and the power distribution characteristics are decomposed and time sequence characteristic analyzed to obtain the space-time correlation feature vector; The sample data set based on the space-time correlation feature vector is labeled using the real-time collected line loss data to obtain the standardized training data set; The line loss distribution prediction model is obtained by training the residual neural network through the standardized training data set; Input the real-time running data of the low-voltage power grid into the line loss distribution prediction model to identify the line loss abnormal area and perform regional priority sorting to determine the line loss treatment priority sequence; According to the line loss treatment priority sequence, adjust the node injection power of the line loss abnormal area.

2. The line loss governance method based on low-voltage power grid multi-source data fusion according to claim 1, characterized in that, The topology structure of the low-voltage power grid and the change information of the power access point are used to reconstruct the initial power supply path by adjusting the switching priority of the power access point to generate the current power supply path, specifically: According to the power supply connection state between each node in the low-voltage power grid, the topology structure is partitioned to obtain the initial power supply path; Based on the topology structure, the power flow direction of each power access point is tracked by using the shortest path search algorithm to collect the change information and identify the power access point whose node position or running state has changed, which is called changed access point; The power supply and reception power balance value of the changed access point is calculated, and the initial power supply path is reconstructed by adjusting the switching priority of the changed access point according to the power supply and reception power balance value to obtain the current power supply path. 3.The line loss governance method based on low-voltage power grid multi-source data fusion of claim 1, characterized in that, The voltage response value of the power access point at each frequency point is collected, and the reverse power flow characteristics of the frequency point are extracted through the voltage response value, specifically: Collect the electrical data of each power access point under the current power supply path to construct the node impedance matrix of the current power supply path; Use the harmonic injection device to apply an excitation signal to each power access point, and record the voltage response value of each power access point at each frequency point; If the frequency point where the voltage response value exceeds the preset voltage deviation range exists, an inverse power flow feature of the frequency point is extracted by wavelet transform; wherein the inverse power flow feature includes an inverse power flow duration, an inverse power flow peak value and an inverse power flow frequency statistical quantity.

4. The line loss governance method based on low-voltage power grid multi-source data fusion according to any one of claims 1 to 3, characterized in that, The node impedance matrix of the current power supply path is corrected using the inverse power flow feature to obtain a power source access point impedance parameter set, specifically: The node impedance matrix is corrected according to the inverse power flow frequency statistical quantity; The node stable state of the power source access point is detected according to the corrected node impedance matrix, and the impedance amplitude, impedance phase angle and node stability margin of the power source access point are calculated to generate the power source access point impedance parameter set.

5. The line loss governance method based on low-voltage power grid multi-source data fusion according to claim 1, characterized in that, The power transmission path is updated by calculating the load rate of the power equipment based on the inverse power flow feature, the current power supply path and the preset load balancing factor to obtain a power distribution feature, specifically: The power equipment on the current power supply path is classified according to the inverse power flow feature; The load rate of each power equipment is obtained by weighting the power distribution of the power equipment according to the classification result and the load balancing factor of each power supply area in the low-voltage power grid; The power mutual aid amount between each power supply area is calculated by the load rate, and the power distribution coefficient of each power supply area is updated based on the power mutual aid amount; Each power transmission path is re-planned according to the power distribution coefficient, and the impedance and loss of each power transmission path are calculated to obtain the power distribution feature.

6. The line loss governance method based on low-voltage power grid multi-source data fusion according to claim 5, characterized in that, The power mutual aid amount between each power supply area is calculated by the load rate, and the power distribution coefficient of each power supply area is updated based on the power mutual aid amount, specifically: The load transfer demand amount of the power equipment whose load rate exceeds the rated value is obtained, and the power mutual aid amount between each power supply area is calculated by the load transfer demand amount; The power distribution coefficient of each power supply area is updated according to the power mutual aid amount, the power supply level and the user capacity of each power supply area.

7. The line loss governance method based on low-voltage power grid multi-source data fusion according to claim 1, characterized in that, The real-time operation data of the low-voltage power grid is input into the line loss distribution prediction model to identify the line loss abnormal area and perform regional priority sorting, and determine the line loss treatment priority sequence, specifically: Real-time operation data and environmental data of the low-voltage power grid are obtained; The real-time operation data, the environmental data and the preset load scheduling rule are input into the line loss distribution prediction model to calculate the line loss rate of each node in the current power supply path, and a real-time line loss distribution table is generated; The voltage qualified rate of each power supply area in the low-voltage power grid is calculated according to the voltage deviation value of each node in the current power supply path; wherein the voltage deviation value is extracted from the real-time line loss distribution table; The power supply area whose voltage qualified rate exceeds the preset voltage qualified rate threshold is regarded as a line loss abnormal area; The line loss abnormal area is prioritized according to the user power supply level and the electricity demand of the line loss abnormal area to obtain the line loss treatment priority sequence. 8.The line loss governance method based on low-voltage power grid multi-source data fusion of claim 1, wherein, The node injection power of the line loss abnormal area is adjusted according to the line loss treatment priority sequence, specifically: The electricity load data of each line loss abnormal area and the node injection power upper limit value of each period are read; Combine the power consumption data, time-of-use electricity price and preset load scheduling rules to obtain power consumption characteristics of each line loss abnormal area; Based on the power consumption characteristics, adjust the node injection power of each line loss abnormal area in sequence by the upper limit of the node injection power; wherein the sequence of adjusting each line loss abnormal area is determined by the line loss treatment priority sequence.

9. A line loss management system based on low-voltage power grid multi-source data fusion, characterized in that, It comprises: A power supply path reconstruction module, an impedance characteristic extraction module, a load fluctuation feature extraction module, a model construction module, a line loss abnormal area identification module and a line loss treatment module; The power supply path reconstruction module is used to reconstruct the initial power supply path by adjusting the switching priority of the power supply access point based on the topology structure of the low-voltage power grid and the change information of the power supply access point, and generate the current power supply path; The impedance characteristic extraction module is used to collect the voltage response value of the power supply access point at each frequency point, extract the reverse power flow characteristics of the frequency point through the voltage response value, correct the node impedance matrix of the current power supply path using the reverse power flow characteristics, and obtain the impedance parameter set of the power supply access point; The load fluctuation feature extraction module is used to update the power transmission path by calculating the load rate of the power equipment based on the reverse power flow characteristics, the current power supply path and the preset load balancing factor, and obtain the power distribution characteristics; The model construction module is used to fuse the topology structure, the current power supply path, the impedance parameter set of the power supply access point and the power distribution characteristics to construct a line loss distribution prediction model, specifically: determine the line length and conductor cross section according to the topology structure and the current power supply path, generate a node connection weight matrix, and describe the node electrical characteristics in the node connection weight matrix through the impedance parameter set of the power supply access point to obtain an electrical characteristic matrix; decompose and analyze the time sequence characteristics of the electrical characteristic matrix and the power distribution characteristics through wavelet transform to obtain a space-time correlation feature vector; label the sample data set based on the space-time correlation feature vector using real-time collected line loss data to obtain a standardized training data set; train a residual neural network through the standardized training data set to obtain a line loss distribution prediction model; The line loss abnormal area identification module is used to input the real-time running data of the low-voltage power grid into the line loss distribution prediction model, identify the line loss abnormal area and perform regional priority sorting, and determine the line loss treatment priority sequence; The line loss treatment module is used to adjust the node injection power of the low-voltage power grid according to the line loss treatment priority sequence.

Images