Transformer load distribution methods, systems, and computer program products

Abstract

This application discloses a transformer load allocation method, system, and computer program product, belonging to the field of smart grid technology. The method includes: using actual measurement data from the low-voltage side of transformers at the end of a distribution substation to perform forward and reverse deductions of Kirchhoff's laws to complete the electrical characteristics and true loss values ​​of all transformers' high-voltage side; based on the completed data, eliminating redundant features, constructing and training a transformer loss prediction model; traversing the boundaries of each substation in the entire distribution substation area, and reconstructing the transfer ratio decision variables into a re-transfer electrical feature vector dataset based on real-time data from all transformers; calling the transformer loss prediction model to construct an objective function for optimizing the total transformer loss in the substation; and using a genetic algorithm iteratively to obtain the optimal total loss and optimal transformer load allocation scheme based on the re-transfer electrical feature vector dataset. This application provides a high-quality data foundation for intelligent management and control of distribution substations, greatly improving the accuracy and computational efficiency of loss prediction.

Description

Technical Field

[0001] This application relates to the field of smart grids, and in particular to a transformer load distribution method, system, and computer program product. Background Technology

[0002] With the advancement of new power system construction, the lean management of distribution substations, as the power supply range or area of ​​distribution transformers (substations), has become a key link in reducing line losses and improving energy efficiency. In actual distribution network operation, a distribution substation includes multiple substations, and multiple transformers often operate in parallel within the same substation. Due to the randomness and tidal nature of user electricity consumption, uneven load distribution often occurs among the transformers, resulting in high overall losses in the substation. Therefore, optimizing load distribution by adjusting tie switches has become an effective means to achieve economical operation of the substation.

[0003] Existing load allocation technologies typically rely on real-time data collected by distribution automation systems for decision-making, generally employing the economic operating range method or the mathematical programming method that minimizes total network losses. These schemes usually directly assume the high-voltage side voltage of the transformer is at its rated value and ignore line impedance voltage drop. They construct a static quadratic function model based on the transformer's rated parameters to estimate losses. After obtaining low-voltage side load data, they use enumeration methods or traditional heuristic algorithms to find the load switching scheme that minimizes total losses. Some schemes also attempt to introduce state estimation techniques to use low-voltage side meter data to infer node states as the basis for optimized scheduling.

[0004] However, existing technologies suffer from significant drawbacks in practical engineering applications, including distorted basic data, low accuracy in quantitative assessment, and difficulty in safely implementing optimal decision-making. Due to the extremely high insulation requirements and retrofitting costs of 10kV high-voltage side equipment, most public transformers do not have measuring equipment installed on their high-voltage sides. Existing technologies directly assume that the high-voltage side voltage is the nominal voltage of the substation outgoing lines, completely ignoring the impedance voltage drop and power flow losses in the actual line topology. This results in inaccurate subsequent condition assessments and loss calculations based on the source data. Furthermore, the static empirical formulas used in traditional loss assessment methods completely disregard the temporal characteristics of dynamic load fluctuations, failing to fit the nonlinear loss changes caused by drastic fluctuations in load rate and power deviation rate under complex operating conditions. Consequently, the predicted results deviate significantly from the actual operating values. Furthermore, existing optimization decisions based on traditional heuristic algorithms often only perform simple additions and subtractions of active load when generating allocation schemes. They lack physical reconstruction and multiple hard constraint verification of the full-dimensional electrical characteristics such as current, voltage, and power factor on the high and low voltage sides after load transfer. This makes the algorithm prone to getting trapped in local optima, and the output scheme often faces engineering risks such as transformer over-limit or voltage drop during physical execution, and cannot be directly converted into reliable automated hardware control instructions. Summary of the Invention

[0005] This application aims to provide a transformer load allocation method, system, and computer program product that, based on data completion and loss prediction, provides the optimal real-time load allocation for public transformers in the distribution area, solving the problems of missing high-voltage side data, distorted dynamic loss estimation, and difficulty in directly implementing load transfer decisions in engineering.

[0006] To achieve the above objectives, the technical solution of this application is as follows: A transformer load distribution method, comprising, By using actual measurement data of the low-voltage side of the transformer at the end of the distribution substation, Kirchhoff's laws are deduced in both forward and reverse directions to complete the electrical characteristics and true values ​​of losses of all transformers on the high-voltage side. Based on the historical electrical characteristics and true loss values ​​of all transformer high-voltage and low-voltage sides, a feature filtering algorithm based on the correlation of sample graph similarity is used to remove redundant features to obtain an electrical feature vector dataset, and a transformer loss prediction model is constructed and trained. Traverse the boundaries of each distribution room in the entire distribution area, and based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, reconstruct the transfer ratio decision variables into a re-transferred electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop. The transformer loss prediction model is used to construct the objective function for optimizing the total transformer loss in the substation. Based on the reprinted electrical feature vector dataset, a genetic algorithm is used to iteratively obtain the current optimal total loss of the substation. After traversing all substations in the entire substation area, the optimal transformer load allocation scheme with the lowest total loss is obtained.

[0007] Optionally, the step of using actual measurement data from the low-voltage side of the transformer at the end of the distribution substation to perform forward and reverse derivations of Kirchhoff's laws to complete the electrical characteristics and true values ​​of losses on the high-voltage side of all transformers specifically includes: The distribution substation topology network node parameters are obtained, and the initial data of the outgoing line side of the transformer at the beginning of the distribution substation and the actual measured data of the low-voltage side of the transformer at the end of the distribution substation are read synchronously. The initial data of the outgoing line side of the transformer at the beginning of the distribution substation includes: the initial value of the three-phase voltage of the substation outgoing line, that is, the initial value of the three-phase voltage of the transformer high-voltage side. The actual measured data of the low-voltage side of the transformer at the end of the distribution substation includes: the actual measured voltage, the actual measured current, the actual measured output active power and power factor of the low-voltage side. The distribution substation topology network node parameters include: the mapping relationship of the high-voltage side access nodes of the transformer, the set of downstream sub-nodes of each node, the branch resistance, the branch reactance and the rated transformation ratio of each transformer. A unified transformer loss calculation model is established for the forward and reverse derivation of Kirchhoff's laws for distribution network line current conversion and topology reverse convergence calculation, voltage forward approximation calculation, and transformer active power loss calculation. The electrical characteristics and true values ​​of losses on the high-voltage side of all transformers are calculated based on the transformer loss calculation model. The electrical characteristics on the high-voltage side of the transformer include: the complete three-phase voltage, the complete three-phase current, the complete input active power, and the true values ​​of active power loss on the high-voltage side of the k-th transformer. The transformer loss calculation model is expressed as follows:

[0008] in, Transformer serial number; The topology node number; For nodes The upstream parent node; For nodes Downstream child nodes; For nodes The set of all downstream child nodes; For the first The topology node number connected to the high-voltage side of the transformer; The actual measured current phasor on the low-voltage side of the k-th transformer; The rated turns ratio of the kth transformer being called; The actual measured power factor on the low-voltage side of the k-th transformer; For the obtained first The equivalent injected current phasor is completed on the high-voltage side of the transformer. For nodes The equivalent injection current at the location; For nodes To the node branch road ( u , v The current phasor; For nodes to downstream child nodes branch road ( v c) Current phasors; and Branch roads ( u , v The resistance and reactance of ) Indicates from the root node to the... High-voltage side connection node of transformer The set of branches traversed between them; For branch roads ( u , v The voltage drop of ) For nodes Voltage phasors; For nodes Voltage phasors; For the first Voltage phasors of the topology nodes connected to the high-voltage side of the transformer; The voltage phasor of the root node; For the first Complete the voltage phasor on the high-voltage side of the transformer; For the first The high-voltage side of the transformer is used to supplement the input active power. For the first Actual measured active power output on the low-voltage side of the transformer; For the first True value of active power loss of a transformer; Indicates conjugate operation; represents the real part; j represents the imaginary unit.

[0009] Optionally, the step of constructing and training a transformer loss prediction model based on the historical electrical characteristics and true loss values ​​of all transformer high-voltage and low-voltage sides, using a feature filtering algorithm based on sample graph similarity to remove redundant features and obtain an electrical feature vector dataset, specifically includes: Based on the historical electrical characteristics and loss true values ​​of all transformer high-voltage and low-voltage sides after completion, a 17-dimensional electrical feature vector is constructed as the first training set and test set. Read the first training set, construct the sample graph distance matrix, use sparse regularization to remove redundant features, and automatically select the electrical feature vector dataset that is highly correlated with transformer line loss as the second training set; A multi-head self-attention neural network is constructed. The second training set is mapped to the hidden layer vectors of the attention mechanism neural network for training. The dependencies within the second training set are dynamically calculated, and the gradient is continuously calculated using an adaptive optimizer. The weights and bias parameters of each layer of the network are iteratively updated until convergence, thus obtaining the transformer loss prediction model. The 17-dimensional electrical feature vector includes: the transformer high-voltage side complete three-phase voltage, the high-voltage side complete three-phase current, the high-voltage side complete input active power, and the low-voltage side actual measured three-phase voltage, the low-voltage side actual measured three-phase current, the low-voltage side actual measured output active power, and derived features; the derived features include: active load rate, apparent power load rate, and power deviation rate.

[0010] Optionally, the construction of the sample graph distance matrix, using sparse regularization to remove redundant features, and automatically selecting the electrical feature vector dataset highly correlated with transformer line loss as the second training set, specifically includes: calculating the norm of each row of the feature weight matrix in the optimization result of the sample graph distance matrix, wherein the norm is a quantitative index of the correlation between the corresponding feature and the output in the first training set, and selecting the electrical feature vector dataset correlated with transformer line loss as the second training set based on the score of the quantitative index.

[0011] Optionally, the traversal of the boundaries of each distribution room in the entire distribution area includes: constructing an incremental outer loop index model to initialize the outer traversal sequence number of the distribution room, establishing a hierarchical mapping of the distribution area, distribution room and transformer, and ensuring that the subsequent load transfer optimization calculation is restricted to be executed sequentially within the physical boundaries of each independent distribution room.

[0012] Optionally, the step of reconstructing the transfer ratio decision variables into a post-transfer electrical feature vector dataset based on the decoupling of active and reactive power between the high-voltage side shunt current and the line voltage drop includes: obtaining the actual measured three-phase voltage, actual measured three-phase current, and actual measured output active power on the low-voltage side of all transformers; reconstructing the post-transfer electrical characteristics of each transformer based on the load transfer ratio decision variables; obtaining the post-transfer low-voltage side three-phase voltage and low-voltage side three-phase current through back-calculation of low-voltage side output active power-current and voltage disturbance correction method based on equivalent impedance; completing the post-transfer high-voltage side three-phase voltage, high-voltage side three-phase current, and high-voltage side input active power; and finally splicing them to form a 17-dimensional post-transfer electrical feature vector with parameter types consistent with the 17-dimensional electrical feature vector input to the training transformer loss prediction model, as shown below:

[0013]

[0014] Where K is the iteration number of the outer layer of the power distribution room currently performing the optimization calculation; For the first The serial number of the target transformer in each power distribution room; For the first The power distribution room faces the first The set of source transformers from which the load is transferred by the transformer, r being the th transformer. The power distribution room faces the first The serial number of the source transformer to which the load is transferred from the transformer; For the first The first power distribution room is controlled by the first The set of target transformers from which the load is transferred out by the transformer, s is the set of transformers that are the first transformers. The first power distribution room is controlled by the first The serial number of the target transformer from which the load is transferred out by the transformer; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; and The first Before the transfer in the power distribution room The actual measured output active and reactive power of the transformer on the low-voltage side; For the first The first power distribution room The low-voltage side power factor of the transformer before reflow, i.e., the first The first power distribution room The actual measured low-voltage side power factor of the transformer; and The first The first power distribution room The active and reactive power on the low-voltage side of the transformer after reflow; For the first The first power distribution room The low-voltage side of the transformer is viewed as having the power output after reflow. For the first The first power distribution room Power factor of the transformer after low-voltage side transfer; and The first The first power distribution room Active load rate and apparent power load rate after the transformer is reloaded; For the first The first power distribution room Power deviation rate after transformer reflow; For the first The first power distribution room Rated capacity of the transformer; and The first The first power distribution room Measured voltage and current phasors on the low-voltage side of the transformer before reflow. For the first The first power distribution room The initial value of the low-voltage side current of the transformer is obtained based on the output active power of the low-voltage side after transfer and the low-voltage side voltage before transfer. For the first The first power distribution room Equivalent branch impedance on the low-voltage side of the transformer; and The first The first power distribution room Voltage and current phasors on the low-voltage side of the transformer after reflow; For the first The first power distribution room Rated turns ratio of the transformer; For the first The first power distribution room The equivalent injected current phasor on the high-voltage side obtained after the transformer is transferred; For the first The first power distribution room The topology node number connected to the high-voltage side of the transformer; For the first Nodes within a power distribution room The equivalent injected current after reprinting; For nodes The set of downstream child nodes; For the first Branch circuits within a power distribution room ( u , v The current phasor after the transfer; For the first Branch circuits within a power distribution room ( v , c The current phasor after the transfer; For the first Branch circuits within a power distribution room ( u , v The high-voltage side complex impedance; For the first Branch circuits within a power distribution room ( u , v Voltage drop after retransmission; For the first After transfer within a power distribution room, the node Voltage phasor; For the first The voltage phasor of node u after transfer within a power distribution room; For the first The first power distribution room The voltage phasor on the high-voltage side of the transformer after transfer; For the first After being transferred inside the power distribution room, the first Voltage phasors of the topology nodes connected to the high-voltage side of the transformer; For the first The first power distribution room The active power input to the high-voltage side of the transformer after transfer; For the first The first power distribution room The active power loss of the transformer after the transfer is completed; For the first The first power distribution room The 17-dimensional electrical characteristic vector of the transformer after transfer, where subscript A represents phase A, subscript B represents phase B, and subscript C represents phase C; To prevent extremely small positive numbers with a denominator of zero; Indicates conjugate operation; This indicates taking the real part; j represents the imaginary unit; The 17-dimensional transduced electrical feature vector includes: the three-phase voltage, three-phase current, and input active power of the high-voltage side transduced by the transformer, as well as the three-phase voltage, three-phase current, output active power, and transduced derivative features of the low-voltage side transduced by the transformer.

[0015] Optionally, the step of using a genetic algorithm to iteratively obtain the optimal total loss of the current substation based on the re-transfer electrical feature vector dataset includes: constructing a transformer load allocation model using a genetic algorithm; in each iteration evaluation of population crossover and mutation, using the load transfer ratio decision variable generated in the current iteration and the 17-dimensional re-transfer electrical feature vector as input parameters; taking the minimum sum of the predicted losses of the transformers participating in the optimization in the current substation as the optimization objective; calling the transformer loss prediction model to construct the objective function for optimizing the total transformer loss of the substation for forward reasoning; obtaining the optimal total loss of the current substation and the corresponding optimal transformer load allocation scheme; and after traversing all substations in the entire substation area, obtaining the optimal transformer load allocation scheme with the lowest total loss among all substations in the entire substation area. Specifically, a genetic algorithm-based transformer load allocation model is constructed to obtain the current optimal total loss of the power distribution room and the corresponding optimal transformer load allocation scheme, as shown below:

[0016] in, For the first The objective function for optimizing the total transformer loss in a power distribution room; For the first The decision set consisting of decision variables for the proportion of all load transfers within a single power distribution room; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; For the first The total number of transformers participating in the optimization within each power distribution room; For the first The first power distribution room The 17-dimensional electrical feature vector of the transformer after relocation under the current candidate relocation scheme; The reconfiguration mapping function for electrical characteristics after reloading, as defined in step 2, represents the reconfiguration of the first electrical characteristic based on the current load reloading decision variables. The electrical quantities on the low-voltage side after the transformer is transferred are obtained, and the electrical characteristics on the high-voltage side are completed to obtain the corresponding 17-dimensional electrical feature vector after transfer. For step S2, the optimal parameter set Solidified transformer loss prediction model; For the first The apparent power on the low-voltage side after the transformer is transferred; For the first The maximum allowable safe apparent power of a transformer; For the first The first power distribution room The transformer and the first The power transfer connection indicator between transformers is set to 1 if there is an actual interconnection switch or an executable power transfer channel between them, otherwise it is set to 0. For the first The constraint set consisting of all constraints within a power distribution room; For solving operators in genetic algorithms; The first one obtained by iterative solution using a genetic algorithm Optimal transformer load distribution scheme for each power distribution room.

[0017] Optionally, obtaining the optimal transformer load allocation scheme with the lowest total loss after traversing all distribution rooms in the entire distribution area includes: performing a loop condition judgment on the total number of distribution rooms in the entire distribution area to determine whether the termination condition is met. If yes, extract the independent sub-schemes of all distribution rooms and perform full set splicing to output the optimal transformer load allocation scheme covering all distribution rooms in the distribution area; if no, continue the calculation for the next distribution room until the termination condition is met; wherein, the termination condition includes: the current distribution room number is the total number of distribution rooms in the distribution area; After obtaining the optimal transformer load allocation scheme, the process also includes: parsing the optimal transformer load allocation scheme into an optimal load allocation control command and sending it to each transformer.

[0018] A transformer load distribution system is provided for performing a transformer load distribution method as described above, the transformer load distribution system comprising: The communication unit is used to receive external data in real time and send the calculated optimal load distribution control command to each transformer. The storage unit is used to cache the historical / real-time operating data of the entire radio station, the trained attention mechanism neural network model file and instruction log, and access the data through a bidirectional high-speed data bus; The computing unit is used to read and write data from the storage unit according to the instruction cycle, execute a transformer load allocation method, and output the optimal transformer load allocation scheme. The display unit is used to obtain the optimal transformer load allocation scheme output by the calculation unit in one direction, and provides human-computer interaction and visualization rendering; The computing unit includes: The transformer high-voltage side data completion module is used to perform forward and reverse deduction of Kirchhoff's laws using actual measurement data of the low-voltage side of the transformer at the end of the distribution substation, complete the electrical characteristics and loss true values ​​of all transformer high-voltage side, and write the real-time electrical characteristics and loss true values ​​of all transformer high-voltage side and low-voltage side as a standardized dataset into the storage unit. The transformer loss prediction module is used to construct and train a transformer loss prediction model based on the historical electrical characteristics and loss true values ​​of all transformers after the high-voltage and low-voltage sides are completed. It uses a feature filtering algorithm based on the correlation of sample graph similarity to remove redundant features to obtain an electrical feature vector dataset. The transformer load transfer module is used to traverse the boundaries of each substation in the entire distribution area. Based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, the transfer ratio decision variables are reconstructed into a transferred electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop. The transformer load allocation module is used to call the transformer loss prediction model to construct the objective function for optimizing the total loss of the transformer in the distribution room. Based on the reprinted electrical feature vector dataset, a genetic algorithm is used to iteratively obtain the optimal total loss of the current distribution room. After traversing all distribution rooms in the entire distribution area, the optimal transformer load allocation scheme with the lowest total loss is obtained. This scheme is then parsed into an optimal load allocation control command and output to the communication unit for distribution. Finally, the optimal transformer load allocation scheme and the corresponding predicted loss true value are output to the display unit for rendering.

[0019] A computer program product includes a computer program that, when executed by a processor, implements the steps of a transformer load distribution method as described above.

[0020] The transformer load allocation method, system, and computer program product provided in this application, in terms of observation capabilities, eliminates the need to install expensive measurement and sensing terminals on the 10kV high-voltage side of the transformer. By utilizing embedded current reverse convergence and voltage forward approximation algorithms, it can accurately reconstruct the complete electrical characteristics of the high-voltage side using real-time low-voltage side data, breaking through the blind spots of physical measurement at zero cost and providing a high-quality data foundation for lean management of distribution areas. In terms of evaluation accuracy, it uses graph similarity regularization to screen highly correlated features and relies on an embedded attention mechanism neural network to replace the traditional static loss formula, accurately capturing the nonlinear impact of reactive power fluctuations, load rate, and other comprehensive factors on line losses. By permanently storing the trained model parameters in the storage unit, it achieves microsecond-level ultra-fast prediction across modules, greatly improving the absolute accuracy and computational efficiency of loss prediction. In terms of decision control, this application overcomes the crude approach of traditional algorithms that only calculate active power addition and subtraction. It incorporates active and reactive power decoupling and full-dimensional electrical characteristic reconstruction calculations in each optimization iteration, taking into account real physical reactions such as voltage dips and power factor changes after power transfer, and strictly applying multiple operational constraints. This not only avoids the trap of heuristic algorithms falling into solutions without physical meaning, but also ensures that every hardware control command output by the system fully complies with circuit safety laws, directly driving the low-voltage tie switch to perform actions, truly realizing unmanned and intelligent line loss management in distribution substations. The system provided in this application, as a highly integrated hardware and software collaborative allocation system, possesses significant advantages in dimensionality reduction and extremely high industrial practical value.

[0021] To make the above-mentioned features and advantages of the application more apparent and understandable, specific embodiments are provided below, and detailed descriptions are given in conjunction with the accompanying drawings. Attached Figure Description

[0022] Figure 1 A flowchart of the transformer load distribution method provided in this application.

[0023] Figure 2 A block diagram of the transformer load distribution system provided in this application. Detailed Implementation

[0024] To make the objectives and technical solutions of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.

[0025] In a specific embodiment of this application, please refer to Figure 1 , Figure 1The flowchart of the transformer load allocation method provided in this application is as follows: The transformer load allocation method provided in this application realizes the optimal real-time load allocation of 10KV public transformer in the distribution area based on data completion and loss prediction, including: steps S1 to S4.

[0026] Step S1: Use the actual measurement data of the low-voltage side of the transformer at the end of the distribution substation to perform forward and reverse derivation of Kirchhoff's laws to complete the electrical characteristics and true values ​​of losses of all transformers on the high-voltage side; Step S2: Based on the historical electrical characteristics and true loss values ​​of all transformer high-voltage and low-voltage sides after completion, redundant features are removed using a feature filtering algorithm based on the correlation of sample graph similarity to obtain an electrical feature vector dataset, and a transformer loss prediction model is constructed and trained. Step S3: Traverse the boundaries of each distribution room in the entire distribution area. Based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, reconstruct the transfer ratio decision variables into a reconstructed electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop. Step S4: Call the transformer loss prediction model to construct the objective function for optimizing the total transformer loss in the distribution room. Based on the reprinted electrical feature vector dataset, use a genetic algorithm to iteratively obtain the optimal total loss of the current distribution room. After traversing all distribution rooms in the entire distribution area, obtain the optimal transformer load allocation scheme with the lowest total loss.

[0027] Specifically, all transformers mentioned in this application are public transformers, and all power distribution rooms are public transformer power distribution rooms.

[0028] The transformer load allocation method provided in this application eliminates the need for expensive measurement and sensing terminals on the 10kV high-voltage side of the transformer in terms of observation capabilities. It utilizes embedded current reverse convergence and voltage forward approximation algorithms to accurately reconstruct the complete electrical characteristics of the high-voltage side using real-time low-voltage side data, breaking through the blind spots of physical measurement at zero cost and providing a high-quality data foundation for lean management of distribution areas. Regarding assessment accuracy, it uses graph similarity regularization to screen highly correlated features and relies on an embedded attention mechanism neural network to replace the traditional static loss formula, accurately capturing the nonlinear impact of reactive power fluctuations, load rate, and other comprehensive factors on line losses. By permanently storing the trained model parameters in the storage unit, it achieves microsecond-level ultra-fast prediction across modules, greatly improving the absolute accuracy and computational efficiency of loss prediction. In terms of decision control, this application overcomes the crude mode of traditional algorithms that only calculate active power addition and subtraction. In each optimization iteration, it incorporates active and reactive power decoupling and full-dimensional reconstruction calculation of electrical characteristics, taking into account real physical reactions such as voltage drop and power factor changes after power transfer, and strictly applying multiple operational constraints. This not only avoids the trap of heuristic algorithms falling into solutions without physical meaning, but also ensures that every hardware control command output by the system complies with circuit safety laws 100%, and can directly drive the low-voltage tie switch to perform actions, truly realizing the unmanned and intelligent line loss management of the distribution transformer area.

[0029] In step S1, please refer to Figure 1 In step S1, Kirchhoff's laws are derived in both forward and reverse directions using actual measurement data from the low-voltage side of the transformer at the end of the distribution substation, thus completing the electrical characteristics and true values ​​of losses on the high-voltage side of all transformers.

[0030] As an example, step S1 specifically includes steps S11 to S13.

[0031] Step S11: Obtain the node parameters of the distribution transformer area topology network, and synchronously read the initial data of the outgoing line side of the transformer at the beginning of the distribution transformer area and the actual measurement data of the low-voltage side of the transformer at the end of the distribution transformer area. Step S12: Establish a unified transformer loss calculation model by performing forward and reverse derivations of Kirchhoff's laws for distribution network line current conversion and topology reverse convergence calculation, voltage forward approximation calculation, and transformer active power loss calculation. Step S13: Calculate the electrical characteristics and true values ​​of losses on the high-voltage side of all transformers based on the transformer loss calculation model.

[0032] As an example, in step S11, the first end of the distribution transformer area is the outgoing end of the 10kV outgoing substation, which is the root node of the distribution transformer area topology network, and inputs three-phase voltage to the transformer; the last end of the distribution transformer area is the low-voltage side of the transformer. In the distribution transformer area, the high-voltage side of the transformer is connected to the 10kV outgoing line, and the low-voltage side is connected to the 400V power supply user, i.e., the load.

[0033] Specifically, the initial data for the outgoing line side of the transformer at the head of the distribution substation includes: the initial value of the three-phase voltage of the outgoing line of the substation, that is, the initial value of the three-phase voltage on the high-voltage side of the transformer.

[0034] The actual measured data of the low-voltage side of the transformer at the end of the distribution area include: actual measured voltage, actual measured current, actual measured output active power, and power factor of the low-voltage side.

[0035] The topology network node parameters of the distribution substation include: the mapping relationship of the high-voltage side access nodes of the transformer, the set of downstream sub-nodes of each node, the branch resistance, the branch reactance, and the rated transformation ratio of each transformer.

[0036] As an example, in step S12, the transformer loss calculation model is represented as follows: (1) in, Transformer serial number; The topology node number; For nodes The upstream parent node; For nodes Downstream child nodes; For nodes The set of all downstream child nodes; For the first The topology node number connected to the high-voltage side of the transformer; The actual measured current phasor on the low-voltage side of the k-th transformer; The rated turns ratio of the kth transformer being called; The actual measured power factor on the low-voltage side of the k-th transformer; For the obtained first The equivalent injected current phasor is completed on the high-voltage side of the transformer. For nodes The equivalent injection current at the location; For nodes To the node branch road ( u , v The current phasor; For nodes to downstream child nodes branch road ( v c) Current phasors; and Branch roads ( u , v The resistance and reactance of ) Indicates from the root node to the... High-voltage side connection node of transformer The set of branches traversed between them; For branch roads ( u , v The voltage drop of ) For nodes Voltage phasors; For nodes Voltage phasors; For the first Voltage phasors of the topology nodes connected to the high-voltage side of the transformer; The voltage phasor of the root node; For the first Complete the voltage phasor on the high-voltage side of the transformer; For the first The high-voltage side of the transformer is used to supplement the input active power. For the first Actual measured active power output on the low-voltage side of the transformer; For the first True value of active power loss of a transformer; Indicates conjugate operation; represents the real part; j represents the imaginary unit.

[0037] Specifically, Formula (1) is a single-phase representative expression given for ease of explanation. In the actual calculation process, if the distribution area is in a three-phase unbalanced operation scenario, the low-voltage side current conversion, high-voltage side equivalent injection current construction, topology reverse convergence, voltage forward recursion, and loss calculation are performed independently for phases A, B, and C, respectively, to obtain the high-voltage side complete three-phase voltage, high-voltage side complete three-phase current, and corresponding complete power. If the distribution area meets the three-phase approximate balance condition, the three-phase equivalent model can be used for approximate solution. Furthermore, all the transformer high-voltage side complete three-phase voltage, three-phase current, and power characteristics used in step S2 are obtained by uniformly organizing the above phase-by-phase calculation results or equivalent calculation results.

[0038] As an example, in step S13, the electrical characteristics of the high-voltage side of the transformer include: the complete three-phase voltage, the complete three-phase current, the complete input active power, and the true value of active power loss on the high-voltage side of the k-th transformer.

[0039] Furthermore, the complete electrical characteristics of the high-voltage side of all transformers in the entire distribution substation and the true values ​​of the losses of all transformers in the entire distribution substation are stored as the underlying physical data source to provide targeted support for the training of the transformer loss prediction model in subsequent steps, thus solving the observation problem of missing data on the high-voltage side of the distribution substation.

[0040] In step S2, please refer to Figure 1In step S2, based on the historical electrical characteristics and true loss values ​​of all transformer high-voltage and low-voltage sides after completion, a feature filtering algorithm based on the correlation of sample graph similarity is used to remove redundant features to obtain an electrical feature vector dataset, and then a transformer loss prediction model is constructed and trained.

[0041] As an example, step S2 specifically includes steps S21 to S23.

[0042] Step S21: Based on the historical electrical characteristics and loss true values ​​of all transformer high-voltage and low-voltage sides after completion, construct a 17-dimensional electrical feature vector as the first training set and test set; Step S22: Read the first training set, construct the sample graph distance matrix, use sparse regularization to remove redundant features, and automatically select the electrical feature vector dataset that is highly correlated with transformer line loss as the second training set; Step S23: Construct a multi-head self-attention mechanism (Transformer) neural network, map the second training set to the hidden layer vectors of the attention mechanism neural network for training, dynamically calculate the dependencies within the second training set, and continuously calculate the gradient using an adaptive optimizer, iteratively updating the weights and bias parameters of each layer of the network until convergence, to obtain the transformer loss prediction model.

[0043] As an example, in step S21, based on the electrical characteristics and true loss values ​​of the high-voltage and low-voltage sides of all transformers extracted in step S1 for the past year, derived features are calculated to construct a 17-dimensional electrical feature vector. The derived features include: active power load rate, apparent power load rate, and power deviation rate. The 17-dimensional electrical feature vector includes: 3D complete three-phase voltage on the high-voltage side, 3D complete three-phase current on the high-voltage side, 1D complete input active power on the high-voltage side, and 3D actual measured three-phase voltage, 3D actual measured three-phase current, 1D actual measured output active power on the low-voltage side, and 3D derived features. In sample processing, each sample is... Taking the transformer as the basic object, the first The 17-dimensional electrical feature vector corresponding to each transformer is used as the input feature vector for each sample. First, the training set and test set are divided according to a preset ratio. Then, based on the minimum and maximum values ​​of each dimension of the features in the first training set, a consistent min-max normalization mapping is performed on the training set and test set to eliminate dimensional differences and ensure consistent model evaluation. Specifically, the 17-dimensional electrical feature vector corresponding to each transformer is used as the input feature vector for each sample. The parameters extracted from the transformer include: the complete three-phase voltage on the high-voltage side, the complete three-phase current on the high-voltage side, the complete input active power on the high-voltage side, the actual measured three-phase voltage on the low-voltage side, the actual measured three-phase current on the low-voltage side, and the actual measured output active power on the low-voltage side, with the true value of active power loss used as a monitoring label.

[0044] Specifically, the first training set and test set are constructed as follows: (2) in, Let t be the active load rate of the k-th transformer. Let be the apparent power load rate of the k-th transformer at time t; Let be the power deviation rate of the k-th transformer at time t; Let t be the actual measured power factor on the low-voltage side of the kth transformer at time t; The actual measured power factor on the low-voltage side of the k-th transformer at time t-1; The actual measured total apparent power of the low-voltage side outgoing lines of the k-th transformer at time t; This represents the rated capacity of the k-th transformer. To prevent extremely small positive numbers with a denominator of zero; Let be the normalized input feature vector of the k-th transformer at time t; Let be the input feature vector of the k-th transformer at time t; and These are the minimum and maximum values ​​of each feature in the first training set and the test set, respectively.

[0045] As an example, in step S22, a sample graph distance matrix is ​​constructed, and sparse regularization is used to remove redundant features, automatically selecting a dataset of electrical feature vectors highly correlated with transformer line losses, as shown below: (3) in, The objective function for optimizing the graph similarity matrix and the equivalent linear feature weight matrix between input and output is defined by three terms: the first term ensures that the sample graph structure for correlation learning retains the structural information of the original similarity; the second term uses the sample graph similarity weights to weight the similarity of the linearly equivalent outputs; and the third term is the feature weight matrix W. Norm. A The sample map distance matrix is... A Each element in the table represents the radial basis function value between two samples in the first training set. S This is a graph similarity matrix used to characterize the adjacency relationships between samples while preserving the data topology. W This is the equivalent linear feature weight matrix between input and output, used to weight the original features and filter important features; The element is the feature weight matrix W, i.e., the element in the first training set. Sample and the bth The weights between them; αThe weight coefficients for the second term of the optimization objective function are used to adjust the importance of structure preservation, and are manually adjusted based on experimental results; β The weighting coefficients for the third term of the objective function are manually adjusted based on experimental results. For the elements in the graph similarity matrix, i.e., the elements in the first training set... Sample and the bth The similarity between them; n Let I be the number of samples in the first training set; let I be the identity matrix; and let C be the desired number of clusters, such that the graph similarity matrix... S C clusters are formed; is the square of the Frobenius norm of the matrix; the superscript T is the matrix transpose symbol; Graph similarity matrix S Rank.

[0046] Furthermore, the optimization objective function includes four constraints, specifically including: the first constraint is to avoid the graph similarity matrix. S A row containing all zeros appears; the second constraint prevents the graph similarity matrix from being filled with zeros. S The simulation of the radial basis function shows a small amount of negative bias; the third constraint ensures the normalization and mapping uniformity of the feature weight matrix; the fourth constraint guarantees the block clustering of the graph similarity matrix S during the optimization process.

[0047] After completing the optimization solution of equation (3), the feature weight matrix in the optimization result is... W Calculate each row The norm is a quantitative indicator of the correlation between the corresponding features and the output in the first training set. Based on the score of the quantitative indicator, a matrix is ​​formed from the electrical feature vector dataset that is highly correlated with transformer line loss, which is used as the second training set and serves as the high-correlation feature matrix for subsequent training of the neural network. .

[0048] As an example, in step S23, the transformer loss prediction model is represented as follows: (4) in, , , The input high correlation feature matrix is ​​respectively The resulting query matrix, key matrix, and value matrix are mapped. , , These are the linear mapping weight matrices for the query matrix, key matrix, and value matrix, respectively. The dimension of the key vector is used to scale the dot product result; The output of the self-attention mechanism serves as the forward attention mapping model; This is the normalization function; For input high correlation feature matrix The Middle The predicted loss value corresponding to each sample; For input high correlation feature matrix The Middle The actual loss label corresponding to each sample; For input high correlation feature matrix The Middle Each normalized sample inputs a feature vector; The parameter set is A transformer loss prediction model employing a self-attention mechanism; is the training loss function; N is the total number of samples extracted from the second training set.

[0049] Furthermore, the transformer loss prediction model can automatically assign different weights to different features under different operating conditions, and perform nonlinear mapping through a feedforward neural network, ultimately regressing and predicting the current state's first loss at the output layer. A transformer in The predicted loss value at any given time solves the technical problem that the dynamic loss of transformers under complex physical conditions is difficult to accurately fit using static formulas.

[0050] Furthermore, once the transformer loss prediction model reaches a preset accuracy on the test set, a parameter solidification and output logic model is constructed to encapsulate and store the learnable parameters of the transformer loss prediction model. This stored parameter is a trained prediction model file, serving as the core for rapid inference during subsequent large-scale power transfer simulation calculations when generating transformer load transfer schemes. The parameter solidification and output logic model is represented as follows: (5) in, This refers to the set of all learnable parameters that are continuously calculated and updated during the training of the attention mechanism neural network. The optimal set of parameters that minimizes the loss function after the attention mechanism neural network has converged during training is permanently frozen and written into storage for evaluation and retrieval when generating transformer load transfer schemes later. To use the optimal parameter set Solidified transformer loss prediction model; This is a transformer loss prediction model after parameters are frozen.

[0051] In step S3, please refer to Figure 1In step S3, the boundaries of each distribution room in the entire distribution area are traversed. Based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, the transfer ratio decision variables are reconstructed into a re-transferred electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop.

[0052] As an example, traversing the boundaries of all substations in a full distribution substation area includes: constructing an incremental outer loop index model to initialize the outer traversal sequence number of the substations, establishing a hierarchical mapping of "distribution substation area - substation - transformer," and ensuring that subsequent load transfer optimization calculations are strictly limited to sequential execution within the mutually independent physical boundaries of each substation. The incremental outer loop index model is represented as follows: (6) Where K is the iteration number of the outer layer of the power distribution room currently performing the optimization calculation, and its initial value is set to 0; This represents the total number of substations within the entire substation area. After implementing this incremental model, an independent optimization process is then performed for the Kth substation.

[0053] As an example, the transfer ratio decision variables are reconstructed into a post-transfer electrical feature vector dataset by decoupling the high-voltage side shunt current and line voltage drop through active and reactive power. This includes: obtaining the actual measured three-phase voltage, actual measured three-phase current, and actual measured output active power on the low-voltage side of all transformers; reconstructing the post-transfer electrical characteristics of each transformer based on the load transfer ratio decision variables; obtaining the post-transfer low-voltage side three-phase voltage and three-phase current through back-calculation of low-voltage side output active power and voltage disturbance correction based on equivalent impedance; completing the post-transfer high-voltage side three-phase voltage, high-voltage side three-phase current, and high-voltage side input active power; and finally splicing them together to form a 17-dimensional post-transfer electrical feature vector with parameter types consistent with the 17-dimensional electrical feature vector input to the training transformer loss prediction model, as shown below: (7) (8) Where K is the iteration number of the outer layer of the power distribution room currently performing the optimization calculation; For the first The serial number of the target transformer in each power distribution room; For the first The power distribution room faces the first The set of source transformers from which the load is transferred by the transformer, r being the th transformer. The power distribution room faces the first The serial number of the source transformer to which the load is transferred from the transformer; For the first The first power distribution room is controlled by the first The set of target transformers from which the load is transferred out by the transformer, s is the set of transformers that are the first transformers. The first power distribution room is controlled by the first The serial number of the target transformer from which the load is transferred out by the transformer; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; and The first Before the transfer in the power distribution room The actual measured output active and reactive power of the transformer on the low-voltage side; For the first The first power distribution room The low-voltage side power factor of the transformer before reflow, i.e., the first The first power distribution room The actual measured low-voltage side power factor of the transformer; and The first The first power distribution room The active and reactive power on the low-voltage side of the transformer after reflow; For the first The first power distribution room The low-voltage side of the transformer is viewed as having the power output after reflow. For the first The first power distribution room Power factor of the transformer after low-voltage side transfer; and The first The first power distribution room Active load rate and apparent power load rate after the transformer is reloaded; For the first The first power distribution room Power deviation rate after transformer reflow; For the first The first power distribution room Rated capacity of the transformer; and The first The first power distribution room Measured voltage and current phasors on the low-voltage side of the transformer before reflow. For the first The first power distribution room The initial value of the low-voltage side current of the transformer is obtained based on the output active power of the low-voltage side after transfer and the low-voltage side voltage before transfer. For the first The first power distribution room Equivalent branch impedance on the low-voltage side of the transformer; and The first The first power distribution room Voltage and current phasors on the low-voltage side of the transformer after reflow; For the first The first power distribution room Rated turns ratio of the transformer; For the first The first power distribution room The equivalent injected current phasor on the high-voltage side obtained after the transformer is transferred; For the first The first power distribution room The topology node number connected to the high-voltage side of the transformer; For the first Nodes within a power distribution room The equivalent injected current after reprinting; For nodes The set of downstream child nodes; For the first Branch circuits within a power distribution room ( u , v The current phasor after the transfer; For the first Branch circuits within a power distribution room ( v , c The current phasor after the transfer; For the first Branch circuits within a power distribution room ( u , v The high-voltage side complex impedance; For the first Branch circuits within a power distribution room ( u , v Voltage drop after retransmission; For the first After transfer within a power distribution room, the node Voltage phasor; For the first The voltage phasor of node u after transfer within a power distribution room; For the first The first power distribution room The voltage phasor on the high-voltage side of the transformer after transfer; For the first After being transferred inside the power distribution room, the first Voltage phasors of the topology nodes connected to the high-voltage side of the transformer; For the first The first power distribution room The active power input to the high-voltage side of the transformer after transfer; For the first The first power distribution room The active power loss of the transformer after the transfer is completed; For the first The first power distribution room The 17-dimensional electrical characteristic vector of the transformer after transfer, where subscript A represents phase A, subscript B represents phase B, and subscript C represents phase C; To prevent extremely small positive numbers with a denominator of zero; Indicates conjugate operation; represents the real part; j represents the imaginary unit.

[0054] Specifically, Formula (7) is a single-phase representative expression given for ease of explanation. In the actual calculation process, if the distribution station is in a three-phase unbalanced operation scenario, phases A, B, and C are reconstructed separately; if the distribution station meets the three-phase approximate balance condition, a three-phase equivalent model can be used for reconstruction and uniformly mapped to the required three-phase input characteristic form.

[0055] Specifically, based on the principles of power component migration and high-voltage side load shunting, an analytical mapping is established from the transfer ratio decision variable to the 17-dimensional electrical feature vector after transfer. The 17-dimensional electrical feature vector after transfer is generated by splicing together the following: 3-dimensional high-voltage side transfer three-phase voltage, 3-dimensional high-voltage side transfer three-phase current, 1-dimensional high-voltage side transfer input active power, 3-dimensional low-voltage side transfer three-phase voltage, 3-dimensional low-voltage side transfer three-phase current, 1-dimensional low-voltage side transfer output active power, and 3-dimensional transfer derived features.

[0056] In step S4, please refer to Figure 1 In step S4, the transformer loss prediction model is called to construct the objective function for optimizing the total transformer loss in the distribution room. Based on the reprinted electrical feature vector dataset, a genetic algorithm is used to iteratively obtain the current optimal total loss of the distribution room. After traversing all distribution rooms in the entire distribution area, the optimal transformer load allocation scheme with the lowest total loss is obtained.

[0057] As an example, the optimal total loss of the current substation is obtained by using a genetic algorithm iteratively based on the re-transfer electrical feature vector dataset. Specifically, this involves: constructing a transformer load allocation model using a genetic algorithm; in each iteration evaluation of population crossover and mutation, using the load transfer ratio decision variable generated in the current iteration and the 17-dimensional re-transfer electrical feature vector as input parameters; taking the minimum sum of the predicted re-transfer losses of the transformers participating in the optimization in the current substation as the optimization objective; calling the transformer loss prediction model to construct the objective function for optimizing the total transformer loss of the substation for forward reasoning; and obtaining the optimal total loss of the current substation and the corresponding optimal transformer load allocation scheme.

[0058] Furthermore, when using a genetic algorithm to iteratively obtain the optimal total loss of the current power distribution room based on the transferred electrical feature vector dataset, in addition to the load transfer ratio range constraint, the upper limit constraint of the receiving end capacity, and the constraint of the transferable connection relationship, the constraints also include: For any source transformer, the total proportion of loads transferred out shall not exceed 1; The incremental power at the receiving end and the reduced power at the source end remain consistent under the principle of power conservation for any retransmission relationship. When there is no actual interconnection switch or transfer channel between transformers, the corresponding load transfer ratio is 0.

[0059] Specifically, a genetic algorithm-based transformer load allocation model is constructed to obtain the current optimal total loss of the power distribution room and the corresponding optimal transformer load allocation scheme, as shown below: (9) in, For the first The objective function for optimizing the total transformer loss in a power distribution room; For the first The decision set consisting of decision variables for the proportion of all load transfers within a single power distribution room; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; For the first The total number of transformers participating in the optimization within each power distribution room; For the first The first power distribution room The 17-dimensional electrical feature vector of the transformer after relocation under the current candidate relocation scheme; The reconfiguration mapping function for electrical characteristics after reloading, as defined in step 2, represents the reconfiguration of the first electrical characteristic based on the current load reloading decision variables. The electrical quantities on the low-voltage side after the transformer is transferred are obtained, and the electrical characteristics on the high-voltage side are completed to obtain the corresponding 17-dimensional electrical feature vector after transfer. For step S2, the optimal parameter set Solidified transformer loss prediction model; For the first The apparent power on the low-voltage side after the transformer is transferred; For the first The maximum allowable safe apparent power of a transformer; For the first The first power distribution room The transformer and the first The power transfer connection indicator between transformers is set to 1 if there is an actual interconnection switch or an executable power transfer channel between them, otherwise it is set to 0. For the first The constraint set consisting of all constraints within a power distribution room; For solving operators in genetic algorithms; The first one obtained by iterative solution using a genetic algorithm Optimal transformer load distribution scheme for each power distribution room.

[0060] Specifically, after locking the physical boundaries of each substation, the attribution status of the low-voltage branch is used as a 0-1 Boolean decision variable. Under the premise of satisfying physical constraints such as node power conservation and equipment capacity, the transformer loss prediction model is called to perform forward reasoning evaluation. After iterative optimization, the scheme for the Kth substation is completed, and the optimal transformer load allocation scheme for the Kth substation is output.

[0061] As an example, since there are no cross-substation transfer channels between different substations, and the corresponding objective functions and constraint sets are independent of each other, the optimization problem of the entire distribution substation area can be decomposed into multiple independent substation problems. After traversing all substations in the entire distribution substation area, the optimal transformer load allocation scheme with the lowest total loss for all substations in the entire distribution substation area will be obtained. This includes: constructing a loop-based decision and global scheme concatenation model, performing a loop condition check on the total number of substations in the entire distribution substation area, and determining whether the termination condition is met. If yes, the independent sub-schemes of all substations are extracted and concatenated to output the optimal transformer load allocation scheme covering all substations in the substation area; if no, the calculation continues for the next substation until the termination condition is met. The loop-based decision and global scheme concatenation model is represented as follows: (10) in, The total number of distribution rooms within the distribution area; symbol " " indicates the update operation of the outer loop number by the computing unit; For the first Optimal transformer load allocation scheme for each power distribution room; This indicates a union and aggregation operation on the optimal transformer load allocation schemes for each power distribution room; The final output is the optimal transformer load allocation scheme with the lowest total loss for all substations in the entire distribution area.

[0062] As an example, based solely on the real-time measured voltage, current, output active power, and power factor of the low-voltage side, the optimal transformer load allocation scheme with the lowest total loss for all distribution rooms can be obtained through data completion, neural network model training, and iterative prediction optimization.

[0063] Furthermore, in generating the optimal transformer load allocation scheme Then, the optimal transformer load allocation scheme will be determined. The solution involves issuing optimal load allocation control commands to each transformer, transforming continuous load transfer into a discrete branch switching problem. This addresses the limitation that the physical actuators of the distribution network can only perform discrete topology switching actions. Simultaneously, the optimal transformer load allocation scheme is... Output and display the corresponding true values ​​of the predicted loss.

[0064] This application also provides a transformer load distribution system; please refer to [link / reference]. Figure 2 , Figure 2 This is a block diagram of the transformer load distribution system provided in this application. The transformer load distribution system is used to execute the above-described transformer load distribution method, including: Communication unit 1 is used to receive external data in real time and send the calculated optimal load distribution control command to each transformer; Storage unit 2 is used to cache historical / real-time operating data of the full distribution radio area, the trained attention mechanism neural network model file and instruction log, and access the data through a bidirectional high-speed data bus; The calculation unit 3 is used to read and write data from the storage unit 2 according to the instruction cycle, execute the above-mentioned transformer load allocation method, and output the optimal transformer load allocation scheme. Display unit 4 is used to obtain the optimal transformer load allocation scheme output by calculation unit 3 in one direction, and provides human-computer interaction and visualization rendering.

[0065] As an example, communication unit 1 is used to input actual measurement data such as the three-phase voltage, current, and total active power data of each transformer's low-voltage side, as well as distribution network topology information, from an external data source, and to issue optimal load allocation control commands through an external actuator. Storage unit 2 has an embedded database used to persistently store historical / real-time operating data and physical line impedance parameters of the entire distribution substation, and to store the trained attention mechanism neural network model file and command logs. Calculation unit 3 has an embedded multi-core algorithm stream that executes one of the transformer load allocation methods described above, realizing transformer high-voltage side data completion, neural network model training, and load allocation scheme optimization. Display unit 4 calls the calculation results from calculation unit 3 to intuitively output the real-time loss values ​​of each transformer and the power transfer schemes of all transformers in the entire distribution substation.

[0066] Specifically, external data sources include: metering systems, marketing systems, and intelligent monitoring systems in each power distribution room.

[0067] Specifically, the external actuators include: the low-voltage interconnection cabinets in each power distribution room.

[0068] As an example, computing unit 3 includes: The transformer high-voltage side data completion module 31 is used to perform forward and reverse deduction of Kirchhoff's laws using the actual measurement data of the low-voltage side of the transformer at the end of the distribution substation, complete the electrical characteristics and loss true values ​​of all transformer high-voltage side, and write the real-time electrical characteristics and loss true values ​​of all transformer high-voltage side and low-voltage side as a standardized dataset into the storage unit 2. The transformer loss prediction module 32 is used to construct and train a transformer loss prediction model by using a feature filtering algorithm based on the correlation of sample graph similarity to remove redundant features and obtain an electrical feature vector dataset based on the historical electrical characteristics and loss true values ​​of all transformer high-voltage and low-voltage sides after completion. The transformer load transfer module 33 is used to traverse the boundaries of each substation in the entire distribution area. Based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, the transfer ratio decision variables are reconstructed into a transferred electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop. The transformer load allocation module 34 is used to call the transformer loss prediction model to construct the objective function for optimizing the total loss of the transformer in the distribution room. Based on the reprinted electrical feature vector dataset, the optimal total loss of the current distribution room is obtained by using a genetic algorithm. After traversing all distribution rooms in the entire distribution area, the optimal transformer load allocation scheme with the lowest total loss is obtained. This scheme is then parsed into an optimal load allocation control command and output to the communication unit for distribution. Finally, the optimal transformer load allocation scheme and the corresponding predicted loss true value are output to the display unit for rendering.

[0069] Specifically, the computing unit 3, as the decision-making core of the transformer load distribution system, completes the entire computing process from incomplete measurement data to electrical characteristic reconstruction and multi-dimensional space optimization in a closed loop within the system. Finally, it autonomously generates executable load distribution physical action commands, thereby empowering the edge power distribution terminal equipment, i.e., the transformer, with intelligent decision-making and local control capabilities.

[0070] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the transformer load distribution method described above.

[0071] The transformer load allocation method, system, and computer program product provided in this application, in terms of observation capabilities, eliminates the need to install expensive measurement and sensing terminals on the 10kV high-voltage side of the transformer. By utilizing embedded current reverse convergence and voltage forward approximation algorithms, it can accurately reconstruct the complete electrical characteristics of the high-voltage side using real-time low-voltage side data, breaking through the blind spots of physical measurement at zero cost and providing a high-quality data foundation for lean management of distribution areas. In terms of evaluation accuracy, it uses graph similarity regularization to screen highly correlated features and relies on an embedded attention mechanism neural network to replace the traditional static loss formula, accurately capturing the nonlinear impact of reactive power fluctuations, load rate, and other comprehensive factors on line losses. By permanently storing the trained model parameters in the storage unit, it achieves microsecond-level ultra-fast prediction across modules, greatly improving the absolute accuracy and computational efficiency of loss prediction. In terms of decision control, this application overcomes the crude approach of traditional algorithms that only calculate active power addition and subtraction. It incorporates active and reactive power decoupling and full-dimensional electrical characteristic reconstruction calculations in each optimization iteration, taking into account real physical reactions such as voltage dips and power factor changes after power transfer, and strictly applying multiple operational constraints. This not only avoids the trap of heuristic algorithms falling into solutions without physical meaning, but also ensures that every hardware control command output by the system fully complies with circuit safety laws, directly driving the low-voltage tie switch to perform actions, truly realizing unmanned and intelligent line loss management in distribution substations. The system provided in this application, as a highly integrated hardware and software collaborative allocation system, possesses significant advantages in dimensionality reduction and extremely high industrial practical value.

[0072] Although this application has been disclosed above with reference to embodiments, it is not intended to limit this application. Anyone skilled in the art may make some modifications and refinements without departing from the spirit and scope of this application. Therefore, the scope of protection of this application shall be determined by the appended claims.

Claims

1. A transformer load distribution method, characterized in that, include, By using actual measurement data of the low-voltage side of the transformer at the end of the distribution substation, Kirchhoff's laws are deduced in both forward and reverse directions to complete the electrical characteristics and true values ​​of losses of all transformers on the high-voltage side. Based on the historical electrical characteristics and true loss values ​​of all transformer high-voltage and low-voltage sides, a feature filtering algorithm based on the correlation of sample graph similarity is used to remove redundant features to obtain an electrical feature vector dataset, and a transformer loss prediction model is constructed and trained. Traverse the boundaries of each distribution room in the entire distribution area, and based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, reconstruct the transfer ratio decision variables into a re-transferred electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop. The transformer loss prediction model is used to construct the objective function for optimizing the total transformer loss in the substation. Based on the reprinted electrical feature vector dataset, a genetic algorithm is used to iteratively obtain the current optimal total loss of the substation. After traversing all substations in the entire substation area, the optimal transformer load allocation scheme with the lowest total loss is obtained.

2. The transformer load distribution method as described in claim 1, characterized in that, The process of using actual measurement data from the low-voltage side of the transformer at the end of the distribution substation to perform forward and reverse derivations of Kirchhoff's laws to complete the electrical characteristics and true values ​​of losses on the high-voltage side of all transformers specifically includes: The distribution substation topology network node parameters are obtained, and the initial data of the outgoing line side of the transformer at the beginning of the distribution substation and the actual measured data of the low-voltage side of the transformer at the end of the distribution substation are read synchronously. The initial data of the outgoing line side of the transformer at the beginning of the distribution substation includes: the initial value of the three-phase voltage of the substation outgoing line, that is, the initial value of the three-phase voltage of the transformer high-voltage side. The actual measured data of the low-voltage side of the transformer at the end of the distribution substation includes: the actual measured voltage, the actual measured current, the actual measured output active power and power factor of the low-voltage side. The distribution substation topology network node parameters include: the mapping relationship of the high-voltage side access nodes of the transformer, the set of downstream sub-nodes of each node, the branch resistance, the branch reactance and the rated transformation ratio of each transformer. A unified transformer loss calculation model is established for the forward and reverse derivation of Kirchhoff's laws for distribution network line current conversion and topology reverse convergence calculation, voltage forward approximation calculation, and transformer active power loss calculation. The electrical characteristics and true values ​​of losses on the high-voltage side of all transformers are calculated based on the transformer loss calculation model. The electrical characteristics on the high-voltage side of the transformer include: the complete three-phase voltage, the complete three-phase current, the complete input active power, and the true values ​​of active power loss on the high-voltage side of the k-th transformer. The transformer loss calculation model is expressed as follows: in, Transformer serial number; The topology node number; For nodes The upstream parent node; For nodes Downstream child nodes; For nodes The set of all downstream child nodes; For the first The topology node number connected to the high-voltage side of the transformer; The actual measured current phasor on the low-voltage side of the k-th transformer; The rated turns ratio of the kth transformer being called; The actual measured power factor on the low-voltage side of the k-th transformer; For the obtained first The equivalent injected current phasor is completed on the high-voltage side of the transformer. For nodes The equivalent injection current at the location; For nodes To the node The current phasor of the branch (u,v); For nodes to downstream child nodes The current phasor of the branch (v,c); and These are the resistance and reactance of the branch (u,v), respectively; Indicates from the root node to the... High-voltage side connection node of transformer The set of branches traversed between them; Let (u,v) be the voltage drop across the branch. For nodes Voltage phasors; For nodes Voltage phasors; For the first Voltage phasors of the topology nodes connected to the high-voltage side of the transformer; The voltage phasor of the root node; For the first Complete the voltage phasor on the high-voltage side of the transformer; For the first The high-voltage side of the transformer is used to supplement the input active power. For the first Actual measured active power output on the low-voltage side of the transformer; For the first True value of active power loss of a transformer; Indicates conjugate operation; represents the real part; j represents the imaginary unit.

3. The transformer load distribution method as described in claim 2, characterized in that, The process involves using historical electrical characteristics and true loss values ​​from both the high-voltage and low-voltage sides of all transformers, employing a feature filtering algorithm based on sample graph similarity to remove redundant features and obtain an electrical feature vector dataset. This dataset is then used to construct and train a transformer loss prediction model, specifically including: Based on the historical electrical characteristics and loss true values ​​of all transformer high-voltage and low-voltage sides after completion, a 17-dimensional electrical feature vector is constructed as the first training set and test set. Read the first training set, construct the sample graph distance matrix, use sparse regularization to remove redundant features, and automatically select the electrical feature vector dataset that is highly correlated with transformer line loss as the second training set; A multi-head self-attention neural network is constructed. The second training set is mapped to the hidden layer vectors of the attention mechanism neural network for training. The dependencies within the second training set are dynamically calculated, and the gradient is continuously calculated using an adaptive optimizer. The weights and bias parameters of each layer of the network are iteratively updated until convergence, thus obtaining the transformer loss prediction model. The 17-dimensional electrical feature vector includes: the transformer high-voltage side complete three-phase voltage, the high-voltage side complete three-phase current, the high-voltage side complete input active power, and the low-voltage side actual measured three-phase voltage, the low-voltage side actual measured three-phase current, the low-voltage side actual measured output active power, and derived features; the derived features include: active load rate, apparent power load rate, and power deviation rate.

4. The transformer load distribution method as described in claim 3, characterized in that, The construction of the sample graph distance matrix involves using sparse regularization to remove redundant features and automatically selecting an electrical feature vector dataset highly correlated with transformer line loss as the second training set. Specifically, this includes calculating the norm of each row of the feature weight matrix in the optimization result of the sample graph distance matrix. The norm is a quantitative indicator of the correlation between the corresponding feature and the output in the first training set. Based on the score of the quantitative indicator, the electrical feature vector dataset correlated with transformer line loss is selected as the second training set.

5. The transformer load distribution method as described in claim 1, characterized in that, The process of traversing the boundaries of each distribution room in the entire distribution area includes: constructing an incremental outer loop index model to initialize the outer traversal sequence number of the distribution room, establishing a hierarchical mapping of the distribution area, distribution room, and transformer, and ensuring that subsequent load transfer optimization calculations are restricted to sequential execution within the mutually independent physical boundaries of each distribution room.

6. The transformer load distribution method as described in claim 3, characterized in that, The process of reconstructing the transfer ratio decision variables into a post-transfer electrical feature vector dataset based on the decoupling of active and reactive power between the high-voltage side shunt current and the line voltage drop includes: acquiring the actual measured three-phase voltage, actual measured three-phase current, and actual measured output active power on the low-voltage side of all transformers; reconstructing the post-transfer electrical characteristics of each transformer based on the load transfer ratio decision variables; obtaining the post-transfer low-voltage side three-phase voltage and low-voltage side three-phase current through back-calculation of low-voltage side output active power and voltage disturbance correction based on equivalent impedance; completing the post-transfer high-voltage side three-phase voltage, high-voltage side three-phase current, and high-voltage side input active power; and finally splicing them together to form a 17-dimensional post-transfer electrical feature vector with parameter types consistent with the 17-dimensional electrical feature vector input to the training transformer loss prediction model, as shown below: in, This is the iteration number of the outer layer of the power distribution room currently undergoing optimization calculation; For the first The serial number of the target transformer in each power distribution room; For the first The power distribution room faces the first The set of source transformers from which the load is transferred by the transformer, r being the th transformer. The power distribution room faces the first The serial number of the source transformer to which the load is transferred from the transformer; For the first The first power distribution room is controlled by the first The set of target transformers from which the load is transferred out by the transformer, s is the set of transformers that are the first transformers. The first power distribution room is controlled by the first The serial number of the target transformer from which the load is transferred out by the transformer; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; and The first Before the transfer in the power distribution room The actual measured output active and reactive power of the transformer on the low-voltage side; For the first The first power distribution room The low-voltage side power factor of the transformer before reflow, i.e., the first The first power distribution room The actual measured low-voltage side power factor of the transformer; and The first The first power distribution room The active and reactive power on the low-voltage side of the transformer after reflow; For the first The first power distribution room The low-voltage side of the transformer is viewed as having the power output after reflow. For the first The first power distribution room Power factor of the transformer after low-voltage side transfer; and The first The first power distribution room Active load rate and apparent power load rate after the transformer is reloaded; For the first The first power distribution room Power deviation rate after transformer reflow; For the first The first power distribution room Rated capacity of the transformer; and The first The first power distribution room Measured voltage and current phasors on the low-voltage side of the transformer before reflow. For the first The first power distribution room The initial value of the low-voltage side current of the transformer is obtained based on the output active power of the low-voltage side after transfer and the low-voltage side voltage before transfer. For the first The first power distribution room Equivalent branch impedance on the low-voltage side of the transformer; and The first The first power distribution room The voltage phasor and current phasor on the low-voltage side of the transformer after transfer; For the first The first power distribution room Rated turns ratio of the transformer; For the first The first power distribution room The equivalent injected current phasor on the high-voltage side obtained after the transformer is transferred; For the first The first power distribution room The topology node number connected to the high-voltage side of the transformer; For the first Nodes within a power distribution room The equivalent injected current after reprinting; For nodes The set of downstream child nodes; For the first The current phasor after transfer in the branch circuit (u,v) within the power distribution room; For the first The current phasor after transfer in the branch circuit (v,c) within the power distribution room; For the first The high-voltage side complex impedance of the branch circuit (u,v) in each power distribution room; For the first Voltage drop after transfer in a branch circuit (u,v) within a power distribution room; For the first After transfer within a power distribution room, the node Voltage phasor; For the first The voltage phasor of node u after transfer within a power distribution room; For the first The first power distribution room The voltage phasor on the high-voltage side of the transformer after transfer; For the first After being transferred inside the power distribution room, the first Voltage phasors of the topology nodes connected to the high-voltage side of the transformer; For the first The first power distribution room The active power input to the high-voltage side of the transformer after transfer; For the first The first power distribution room The active power loss of the transformer after the transfer is completed; For the first The first power distribution room The 17-dimensional electrical characteristic vector of the transformer after transfer, where subscript A represents phase A, subscript B represents phase B, and subscript C represents phase C; To prevent extremely small positive numbers with a denominator of zero; Indicates conjugate operation; This indicates taking the real part; j represents the imaginary unit; The 17-dimensional transduced electrical feature vector includes: the three-phase voltage, three-phase current, and input active power of the high-voltage side transduced by the transformer, as well as the three-phase voltage, three-phase current, output active power, and transduced derivative features of the low-voltage side transduced by the transformer.

7. The transformer load distribution method as described in claim 6, characterized in that, The step of obtaining the optimal total loss of the current substation using a genetic algorithm based on the re-transfer electrical feature vector dataset includes: constructing a transformer load allocation model using a genetic algorithm; in each iteration evaluation of population crossover and mutation, using the load transfer ratio decision variable generated in the current iteration and the 17-dimensional re-transfer electrical feature vector as input parameters; taking the minimum sum of the predicted losses of the transformers participating in the optimization in the current substation as the optimization objective; calling the transformer loss prediction model to construct the objective function for optimizing the total transformer loss of the substation for forward reasoning; obtaining the optimal total loss of the current substation and the corresponding optimal transformer load allocation scheme; and after traversing all substations in the entire substation area, obtaining the optimal transformer load allocation scheme with the lowest total loss among all substations in the entire substation area. Specifically, a genetic algorithm-based transformer load allocation model is constructed to obtain the current optimal total loss of the power distribution room and the corresponding optimal transformer load allocation scheme, as shown below: in, For the first The objective function for optimizing the total transformer loss in a power distribution room; For the first The decision set consisting of decision variables for the proportion of all load transfers within a single power distribution room; For the first The first power distribution room is controlled by the first Transformer to the first The proportion of load transferred by the transformer; For the first The total number of transformers participating in the optimization within each power distribution room; For the first The first power distribution room The 17-dimensional electrical feature vector of the transformer after relocation under the current candidate relocation scheme; The reconfiguration mapping function for electrical characteristics after reloading, as defined in step 2, represents the reconfiguration of the first electrical characteristic based on the current load reloading decision variables. The electrical quantities on the low-voltage side after the transformer is transferred are obtained, and the electrical characteristics on the high-voltage side are completed to obtain the corresponding 17-dimensional electrical feature vector after transfer. For step S2, the optimal parameter set Solidified transformer loss prediction model; For the first The apparent power on the low-voltage side after the transformer is transferred; For the first The maximum allowable safe apparent power of a transformer; For the first The first power distribution room The transformer and the first The power transfer connection indicator between transformers is set to 1 if there is an actual interconnection switch or an executable power transfer channel between them, otherwise it is set to 0. For the first The constraint set consisting of all constraints within a power distribution room; For solving operators in genetic algorithms; The first one obtained by iterative solution using a genetic algorithm Optimal transformer load distribution scheme for each power distribution room.

8. The transformer load distribution method as described in claim 7, characterized in that, The process of obtaining the optimal transformer load allocation scheme with the lowest total loss after traversing all substations in the entire distribution area includes: performing a loop condition check on the total number of substations in the entire distribution area to determine whether the termination condition is met; if so, extracting the independent sub-schemes of all substations and splicing them together to output the optimal transformer load allocation scheme covering all substations in the distribution area; if not, continuing the calculation for the next substation until the termination condition is met; wherein, the termination condition includes: the current substation number is equal to the total number of substations in the distribution area; After obtaining the optimal transformer load allocation scheme, the process also includes: parsing the optimal transformer load allocation scheme into an optimal load allocation control command and sending it to each transformer.

9. A transformer load distribution system for executing a transformer load distribution method as described in any one of claims 1 to 8, characterized in that, The transformer load distribution system includes: The communication unit is used to receive external data in real time and send the calculated optimal load distribution control command to each transformer. The storage unit is used to cache the historical / real-time operating data of the entire radio station, the trained attention mechanism neural network model file and instruction log, and access the data through a bidirectional high-speed data bus; The computing unit is used to read and write data from the storage unit according to the instruction cycle, execute a transformer load allocation method, and output the optimal transformer load allocation scheme. The display unit is used to obtain the optimal transformer load allocation scheme output by the calculation unit in one direction, and provides human-computer interaction and visualization rendering; The computing unit includes: The transformer high-voltage side data completion module is used to perform forward and reverse deduction of Kirchhoff's laws using actual measurement data of the low-voltage side of the transformer at the end of the distribution substation, complete the electrical characteristics and loss true values ​​of all transformer high-voltage side, and write the real-time electrical characteristics and loss true values ​​of all transformer high-voltage side and low-voltage side as a standardized dataset into the storage unit. The transformer loss prediction module is used to construct and train a transformer loss prediction model based on the historical electrical characteristics and loss true values ​​of all transformers after the high-voltage and low-voltage sides are completed. It uses a feature filtering algorithm based on the correlation of sample graph similarity to remove redundant features to obtain an electrical feature vector dataset. The transformer load transfer module is used to traverse the boundaries of each substation in the entire distribution area. Based on the complete electrical characteristics and loss true values ​​of the high-voltage and low-voltage sides of all transformers, the transfer ratio decision variables are reconstructed into a transferred electrical feature vector dataset by decoupling active power and reactive power according to the high-voltage side shunt current and line voltage drop. The transformer load allocation module is used to call the transformer loss prediction model to construct the objective function for optimizing the total loss of the transformer in the distribution room. Based on the reprinted electrical feature vector dataset, a genetic algorithm is used to iteratively obtain the optimal total loss of the current distribution room. After traversing all distribution rooms in the entire distribution area, the optimal transformer load allocation scheme with the lowest total loss is obtained. This scheme is then parsed into an optimal load allocation control command and output to the communication unit for distribution. Finally, the optimal transformer load allocation scheme and the corresponding predicted loss true value are output to the display unit for rendering.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of a transformer load distribution method as described in any one of claims 1 to 8.

Images