A method and device for calculating ground voltage of a low-voltage distribution network

Abstract

The application discloses a low-voltage distribution network ground voltage calculation method and device, relates to the distribution network voltage prediction technical field, and mainly aims at improving the determination efficiency of the ground voltage. The main technical scheme of the application is as follows: an input vector is acquired, including the low-voltage side first-end three-phase phase voltage amplitude of a distribution transformer, three-phase active power and reactive power of each load node and photovoltaic injection power; a weight adjacency matrix between each load node in the distribution network is determined according to a preset method, and each element represents the impedance value of a connected node line; a response vector is obtained according to the weight adjacency matrix and a node injection complex power vector, a topology vector is extracted from non-zero elements, and the node injection complex power vector is obtained from the photovoltaic injection power; an enhanced input vector is obtained by splicing the topology vector and the input vector, and is input into a trained voltage prediction model to obtain three-phase voltage prediction results of each node; and the neutral line voltage drop is calculated according to the three-phase voltage prediction results, so that the ground voltage is obtained by correcting the three-phase voltage prediction results.

Description

Technical Field

[0001] This invention relates to the field of distribution network voltage prediction technology, and in particular to a method and apparatus for calculating the voltage to ground in a low-voltage distribution network. Background Technology

[0002] In low-voltage distribution networks, accurately calculating the voltage to ground at each node (i.e., the actual voltage between the phase line and the ground) is particularly important. This is because critical operation and maintenance decisions, such as the operational safety of user equipment and low-voltage alarms, are directly based on the voltage to ground.

[0003] In a typical three-phase four-wire low-voltage distribution network, in addition to the three-phase lines, a neutral line is also provided to provide a current loop for a large number of single-phase loads. Since three-phase loads are often unbalanced, the sum of the three-phase currents is not zero. When the unbalanced current flows through the neutral line, it creates a voltage drop across its impedance, causing the neutral point potential to deviate from the ground potential. As a result, the phase-neutral point voltage (commonly referred to as the phase voltage) is no longer equivalent to the actual voltage to ground. If this deviation is ignored and the phase voltage is directly used to replace the voltage to ground, significant errors will occur under unbalanced operating conditions.

[0004] To reflect this phenomenon, existing power flow calculation methods typically treat the neutral point voltage offset and three-phase node voltages as unknown variables, incorporating them into a three-phase four-wire power flow model and employing forward-backward substitution or the Newton-Raphson method for coupled iterative solutions. However, a strong coupling exists between the neutral point potential and the three-phase voltages: the three-phase imbalance determines the neutral current, and the neutral voltage drop, in turn, alters the phase-to-ground voltages. Therefore, each iteration must simultaneously update both the three-phase voltages and the neutral point state, making decoupling impossible. This necessitates an additional global scan of the neutral current and voltage drop in each iteration of the forward-backward substitution method, while the Newton-Raphson method requires expanding the dimensions of the Jacobian matrix to encompass the neutral point variables, significantly increasing the computational load for each iteration. The entire solution process requires multiple round trip corrections, ultimately leading to low overall solution efficiency. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method and apparatus for calculating the voltage to ground in a low-voltage distribution network, the main purpose of which is to improve the efficiency of determining the voltage to ground.

[0006] To solve the above-mentioned technical problems, the present invention proposes the following solution: In a first aspect, the present invention provides a method for calculating the voltage to ground in a low-voltage distribution network, the method comprising: Obtain a predefined input vector, which includes the amplitude of the three-phase phase voltage at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injected power; The weighted adjacency matrix between each load node in the distribution network is determined according to a preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connecting nodes. The response vector is obtained by using the weighted adjacency matrix and the node-injected complex power vector of each load node, and the non-zero element extraction operation is performed on the response vector to obtain the topology vector. The node-injected complex power vector is obtained based on the photovoltaic injection power. The topology vector is concatenated with the input vector to obtain the enhanced input vector, and the enhanced input vector is input into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node; The neutral voltage drop is calculated based on the predicted three-phase voltage, and the predicted three-phase voltage is corrected based on the neutral voltage drop to obtain the ground voltage of each node.

[0007] In a second aspect, the present invention provides a device for calculating the voltage to ground in a low-voltage distribution network, the device comprising: The vector acquisition unit is used to acquire a predefined input vector, which includes the three-phase phase voltage amplitude at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injection power. The matrix determination unit is used to determine the weighted adjacency matrix between each load node in the distribution network according to a preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connecting nodes. The vector determination unit is used to obtain a response vector based on the weighted adjacency matrix determined by the matrix determination unit and the node injection complex power vector of each load node, and to perform a non-zero element extraction operation on the response vector to obtain a topology vector. The node injection complex power vector is obtained based on the photovoltaic injection power. The vector concatenation unit is used to concatenate the topological vector determined by the vector determination unit with the input vector obtained by the vector acquisition unit to obtain an enhanced input vector, and input the enhanced input vector into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node; The voltage correction unit is used to calculate the neutral line voltage drop based on the three-phase voltage prediction results obtained by the vector splicing unit, and to correct the three-phase voltage prediction results based on the neutral line voltage drop to obtain the ground voltage of each node.

[0008] To achieve the above objectives, according to a third aspect of the present invention, a storage medium is provided, the storage medium including a stored program, wherein, when the program is executed, the device where the storage medium is located is controlled to execute the low-voltage distribution network voltage calculation method of the first aspect.

[0009] To achieve the above objectives, according to a fourth aspect of the present invention, a processor is provided for running a program, wherein the program executes the method for calculating the voltage to ground of a low-voltage distribution network as described in the first aspect.

[0010] By means of the above technical solution, the present invention provides a method and apparatus for calculating the voltage to ground in a low-voltage distribution network. First, a predefined input vector is obtained, which includes the amplitude of the three-phase phase voltage at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injected power. Second, a weighted adjacency matrix between each load node in the distribution network is determined according to a preset method, wherein each element represents the impedance value of the line between the corresponding phase-connected nodes, thereby expressing the electrical structure of the network in a quantitative form. Then, a response vector is generated based on the weighted adjacency matrix and the node injected complex power vector of each load node, and a non-zero element extraction operation is performed on the response vector to obtain a topology vector. Subsequently, the topology vector is concatenated with the original input vector to form an enhanced input vector, which is then input into a pre-trained voltage prediction model to directly output the three-phase phase voltage prediction results of each node. This process does not require the introduction of the neutral point potential as a solution variable inside the model, thus avoiding the complexity caused by the mutual coupling of three-phase voltage and neutral point state in traditional three-phase four-wire power flow. Finally, based on the obtained three-phase voltage prediction results, the neutral voltage drop is calculated independently, and the phase voltage is corrected in one step to obtain the true ground voltage of each node. Since the entire process divides phase voltage prediction and neutral voltage drop correction into two sequential, feedback-free iteration stages, computational efficiency is significantly improved. At the same time, it ensures that the ground voltage results accurately reflect the actual potential difference between the phase line and the ground, providing a reliable basis for operation and maintenance decisions such as low-voltage alarms and equipment safety assessments.

[0011] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0012] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a method for calculating the voltage to ground in a low-voltage distribution network according to an embodiment of the present invention is shown. Figure 2This invention provides a flowchart of another method for calculating the voltage to ground in a low-voltage distribution network according to an embodiment of the present invention. Figure 3 This diagram shows a block diagram of a low-voltage distribution network ground voltage calculation device provided by an embodiment of the present invention; Figure 4 This invention provides a block diagram of another low-voltage distribution network ground voltage calculation device according to an embodiment of the present invention. Detailed Implementation

[0013] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0014] Currently, in the calculation of voltage to ground in low-voltage distribution networks, the neutral point potential is usually treated as an additional state variable and incorporated into the forward-backward substitution method or the Newton-Raphson method for coupled iterative solution along with the three-phase node voltages. However, in practical applications, it has been found that there is a strong coupling relationship between the neutral point and the three-phase voltages. That is, the imbalance of the three-phase load determines the magnitude of the neutral line current, and the voltage drop caused by the neutral line current flowing through the line impedance changes the actual voltage to ground of each phase, thus affecting the current calculation in the next round. This interdependence means that the three-phase voltages and the neutral point state must be updated synchronously in each iteration, making decoupling impossible and significantly increasing the computational complexity.

[0015] To address the above issues, we considered whether we could directly use data-driven models to predict the voltage to ground end-to-end. However, if we completely ignore the physical mechanism of neutral point drift, the model's generalization ability is insufficient in scenarios such as drastic fluctuations in photovoltaic power output or sudden load changes, resulting in poor stability.

[0016] Ultimately, through continuous creative work, it was realized that the key was not in choosing a purely physical model or a purely data model, but in how to reasonably decouple the two originally closely coupled physical processes of "phase voltage distribution" and "neutral point offset".

[0017] Based on this understanding, this invention proposes a method for calculating the voltage to ground in low-voltage distribution networks. This method decomposes the originally coupled solution process into two sequential steps that do not require iterative feedback, fundamentally avoiding computational redundancy caused by repeated iterations. While ensuring physical rationality, it significantly improves computational efficiency.

[0018] This invention is mainly applicable to low-voltage distribution network scenarios. Low-voltage distribution network refers to the power network that distributes electrical energy to end users starting from the secondary side (i.e., the low-voltage side) of the distribution transformer. Its nominal voltage level is usually 380V / 220V (three-phase four-wire system), where 380V is the line voltage (the voltage between any two phases) and 220V is the phase voltage (the voltage between any phase line and the neutral line). The actual power supply targets are mainly residential, commercial and small industrial and commercial users.

[0019] In a three-phase four-wire system, user equipment (such as sockets and appliances) actually experiences the phase-to-ground voltage, not the phase-to-neutral voltage. When the three phases are unbalanced, current flows through the neutral line, creating a voltage drop across its impedance, causing the neutral point potential to deviate from the ground potential. In this case, the phase-to-neutral voltage ≠ phase-to-ground voltage. If the maintenance system still uses phase voltage as the basis for judgment, it may misjudge low or overvoltage conditions, affecting the accuracy of power quality assessment, equipment safety protection, and fault alarms. Therefore, accurately calculating the true phase-to-ground voltage at each node is fundamental to the observability, measurability, and controllability of low-voltage distribution networks.

[0020] Next, we will combine Figure 1 This invention describes a method for calculating the voltage to ground in a low-voltage distribution network, and its specific execution steps are as follows: Figure 1 As shown, it includes: 101. Obtain the predefined input vector, which includes the three-phase voltage amplitude at the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injection power.

[0021] In this step, the effective values ​​of the three-phase voltages at the beginning of the low-voltage side of the distribution transformer (i.e., the voltage amplitudes of phases A, B, and C at the current moment) are first collected, serving as the initial reference for the system's operating status. The voltage at this location reflects the power supply input status of the entire low-voltage line and forms the basis for subsequent analysis of voltage changes at each node.

[0022] Simultaneously, three-phase active and reactive power data are collected from each load node. This data is typically uploaded in real time by smart meters or terminal devices installed on-site, and can be used to accurately grasp the actual power consumption of each node and its impact on the system's reactive power demand.

[0023] For load nodes connected to distributed photovoltaic (PV) systems, it is also necessary to collect their injected active and reactive power (collectively referred to as PV injected power). It is important to note that not all nodes have PV equipment installed; there may be situations where only some nodes are simultaneously connected to both the load and the PV system. Modern PV inverters can not only supply active power to the grid but also generate or absorb reactive power according to dispatch instructions or local control strategies; therefore, their active and reactive power outputs must be recorded synchronously.

[0024] The above parameters were chosen because they collectively determine the current operating state of the distribution network: the head-end voltage provides a reference, load power characterizes electricity demand, and photovoltaic injection reflects the supporting role of distributed power sources. These three types of data, combined, can comprehensively describe the power flow and voltage level of the low-voltage distribution network, providing reliable input for subsequent analysis, control, or optimization. Ultimately, these data will be organized into a unified input vector according to a predefined format.

[0025] 102. Determine the weighted adjacency matrix between each load node in the distribution network according to the preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connected nodes.

[0026] 103. Obtain the response vector based on the weighted adjacency matrix and the node injection complex power vector of each load node, and perform non-zero element extraction operation on the response vector to obtain the topology vector.

[0027] In step 102, the physical connections between load nodes are first determined based on the distribution network topology, i.e., which nodes are directly connected by lines. Then, for each line segment connecting two load nodes, the corresponding three-phase impedance data is determined, including the self-impedances of phases A, B, and C (i.e., the resistance and reactance corresponding to A–A, B–B, and C–C), and the mutual impedances between phases (e.g., A–B, A–C, B–C and their symmetrical terms B–A, C–A, and C–B). All impedances are represented in complex form R+jX, reflecting the resistive and reactive characteristics of the line.

[0028] For any pair of directly connected load nodes, their electrical coupling relationships are organized into a 3×3 structure according to phase sequence: the first row corresponds to the coupling of phase A with the other three phases, in the order A–A, A–B, A–C; the second row starts from phase B, with couplings of B–A, B–B, B–C; the third row starts from phase C, with couplings of C–A, C–B, C–C. If there is no direct electrical connection between two nodes, the corresponding impedance value of that node is set to zero. Through this method, a weighted adjacency matrix is ​​finally constructed.

[0029] For example, the weighted adjacency matrix is ​​as follows: (Formula 1) (Formula 2) (Formula 3) In the formula: and They are nodes , Between The phase resistance and reactance; and They are nodes No. Phase and node No. The mutual resistance and mutual reactance between phases, if the node , There is no line connection between them or When the time is 0, its value is 0.

[0030] After proceeding to step 103, the injected complex power vector of each load node can be obtained. Then, the response vector is obtained based on the weighted adjacency matrix and the node injected complex power vector of each load node. The non-zero element extraction operation is then performed on the response vector to obtain the topology vector.

[0031] Specifically, the node-injected complex power vector is represented as S = P + jQ, where: The real part P represents the net injected active power of the node, which is the algebraic sum of the active power injected by the photovoltaic system (taking a positive value) and the active power consumed by the load (taking a negative value); the imaginary part Q represents the net injected reactive power of the node, which is the algebraic sum of the reactive power output by the photovoltaic inverter (taking a positive value) and the reactive power consumed by the load (taking a negative value).

[0032] In this invention, the response vector generation process is as follows: The weighted adjacency matrix constructed in step 102 is multiplied by the node injection complex power vector of each load node to obtain a complex response vector. The process is as follows: (Formula 4) 104. Concatenate the topology vector with the input vector to obtain the enhanced input vector, and input the enhanced input vector into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node.

[0033] After obtaining the topology vector in step 103, the topology vector and the input vector can be concatenated to form an enhanced input vector. This enhanced input vector is then fed into a pre-trained voltage prediction model to predict the three-phase voltage amplitude of each load node.

[0034] The voltage prediction model used in this invention can be a recurrent neural network (DA-RNN) based on a two-stage attention mechanism. This model first automatically identifies key variables that have a significant impact on voltage (such as high-penetration photovoltaic access nodes) through an input attention mechanism, and then uses a time-series attention mechanism to focus on the time period that is most relevant to the current moment in the historical operating state, thereby improving the accuracy and robustness of three-phase voltage prediction.

[0035] Furthermore, the model can also employ other structures, such as graph neural networks (GNNs), which can directly utilize the node-connection relationships of the distribution network to transmit feature information between electrically adjacent nodes, fitting the natural graph structure of the distribution network; or spatiotemporal graph convolutional networks (ST-GCNs), which can simultaneously characterize the electrical coupling relationships in the spatial dimension and the dynamic changes in load and photovoltaic output in the temporal dimension. In general, as long as the model can receive the enhanced input vector and output the predicted three-phase voltage of each load node, it can be applied to this invention and is not limited to a specific network architecture.

[0036] 105. Calculate the neutral voltage drop based on the three-phase voltage prediction results, and correct the three-phase voltage prediction results based on the neutral voltage drop to obtain the ground voltage of each node.

[0037] After obtaining the predicted three-phase voltages for each load node in step 104, since low-voltage distribution networks generally adopt a three-phase four-wire structure, current will flow through the neutral line when the three-phase load is unbalanced or when there is a single-phase distributed power source (such as residential photovoltaic), resulting in a voltage drop. This voltage drop will cause the neutral point potential to shift, causing the actual phase-to-ground voltage to deviate from the predicted phase voltage value. Therefore, this step calculates or estimates the neutral line voltage drop and corrects the predicted results accordingly, ultimately obtaining a more accurate representation of the phase-to-ground voltage for each node under actual operating conditions.

[0038] Specifically, in this process, one can either use the basic principles of circuitry to estimate the neutral voltage drop based on the three-phase voltage prediction results, and then correct the three-phase voltage prediction results based on the neutral voltage drop to obtain the voltage to ground of each node; or one can use a machine learning model to predict the neutral voltage drop based on the three-phase voltage prediction results, and then correct the three-phase voltage prediction results based on the neutral voltage drop to obtain the voltage to ground of each node.

[0039] As can be seen, in this invention, the influence of the neutral line is completely ignored, and the ideal three-phase phase voltage is directly predicted by the model. Based on this result, only one independent neutral line voltage drop estimation is performed, and the voltage to ground is obtained in one correction. The entire process does not feed the corrected voltage to ground back to the prediction model, nor does it require recalculating the current or updating the voltage drop again. Therefore, there is no circular dependency and no need for iteration.

[0040] This serial structure, which predicts a single correction, cuts off the bidirectional coupling link between the voltage and the neutral point, thus avoiding the problems of convergence difficulties and computational time caused by coupling in traditional methods.

[0041] In practical applications, any of the above correction methods can be flexibly selected based on the completeness of the data, the computing resources on the edge side, and the accuracy requirements. The two methods can also be used in combination; for example, preliminary corrections can be made based on the physical model first, followed by data-driven modeling to compensate for the residuals; or the outputs of the two methods can be weighted and fused according to their confidence levels. Such combined strategies can further improve the accuracy and robustness of ground voltage estimation while maintaining physical rationality.

[0042] Based on the above Figure 1 As can be seen from the implementation method, the method for calculating the voltage to ground in a low-voltage distribution network provided by this invention first obtains a predefined input vector, which includes the amplitude of the three-phase phase voltage at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injected power. Second, a weighted adjacency matrix between each load node in the distribution network is determined according to a preset method, where each element represents the impedance value of the line between the corresponding phase-connected nodes, thereby expressing the electrical structure of the network in a quantitative form. Then, a response vector is generated based on the weighted adjacency matrix and the node injected complex power vector of each load node, and a non-zero element extraction operation is performed on the response vector to obtain a topology vector. Subsequently, the topology vector is concatenated with the original input vector to form an enhanced input vector, which is then input into a pre-trained voltage prediction model to directly output the three-phase phase voltage prediction results of each node. This process does not require the introduction of the neutral point potential as a solution variable inside the model, thus avoiding the complexity brought about by the mutual coupling of three-phase voltage and neutral point state in traditional three-phase four-wire power flow. Finally, based on the obtained three-phase voltage prediction results, the neutral voltage drop is calculated independently, and the phase voltage is corrected in one step to obtain the true ground voltage of each node. Since the entire process divides phase voltage prediction and neutral voltage drop correction into two sequential, feedback-free iteration stages, computational efficiency is significantly improved. At the same time, it ensures that the ground voltage results accurately reflect the actual potential difference between the phase line and the ground, providing a reliable basis for operation and maintenance decisions such as low-voltage alarms and equipment safety assessments.

[0043] Furthermore, as a response to Figure 1 Further refinement and extension of the illustrated embodiment, this invention also provides another method for calculating the voltage to ground in a low-voltage distribution network, such as... Figure 2 As shown, the specific steps are as follows: 201. Obtain the predefined input vector, which includes the three-phase voltage amplitude at the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injection power.

[0044] 202. Determine the weighted adjacency matrix between each load node in the distribution network according to the preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connected nodes.

[0045] The implementation methods of steps 201-202 are the same as those of steps 101-102, and can achieve the same technical effect and solve the same technical problem, so they will not be repeated here.

[0046] 203. Obtain the response vector based on the weighted adjacency matrix and the node injection complex power vector of each load node, and perform non-zero element extraction operation on the response vector to obtain the topology vector.

[0047] In this step, zero elements in the response vector can be removed, and non-zero elements can be retained. Then, the non-zero elements are decomposed into real and imaginary parts. The real part can be determined as the resistor topology feature vector, and the imaginary part can be determined as the reactance topology feature vector. Finally, the resistor topology feature vector and the reactance topology feature vector are determined as the topology vector.

[0048] 204. Concatenate the topology vector with the input vector to obtain the enhanced input vector, and input the enhanced input vector into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node.

[0049] In this embodiment, it should be noted that the training steps of the pre-trained voltage prediction model are as follows: Obtain the training dataset (the data dimension of the training dataset is consistent with the augmented input vector), and then input the training samples in the training dataset into the recurrent neural network model based on the two-stage attention mechanism to obtain the three-phase voltage prediction results of each node (including the three-phase voltage amplitude and phase angle).

[0050] Subsequently, the predicted voltage result is substituted into the three-phase power flow imbalance equation to calculate the total injected active and reactive power (i.e., theoretical total injected power) that should theoretically be satisfied under this voltage condition. Here, "total injected power" refers to the algebraic sum of the net injected power of all load nodes in the three phases A, B, and C of the entire network.

[0051] Next, two types of losses are constructed: Data-driven loss: measures the error between the predicted three-phase voltage and the measured voltage, usually expressed as mean square error.

[0052] Physical constraint loss: measures the deviation between the theoretical total injected power derived from the predicted voltage and the actual total injected power.

[0053] Finally, the weighted sum of data-driven loss and physical constraint loss is used as the total loss function. With the goal of minimizing this total loss function, the parameters of the DA-RNN model are iteratively adjusted through the backpropagation algorithm until the total loss meets the convergence condition or reaches the preset training rounds. Finally, a well-trained voltage prediction model is obtained. In this way, through the synergistic effect of physical constraint loss and data-driven loss, the model output not only closely approximates the measured voltage but also conforms to the basic physical laws of power grid operation.

[0054] The model structure includes: Input attention encoder: It encodes the input feature sequence based on a long short-term memory network and dynamically calculates the importance weight of each input variable at the current time to highlight key perturbation sources.

[0055] Temporal attention decoder: During the decoding stage, the temporal attention mechanism adaptively aggregates historical encoded states to generate the context vector most relevant to the current prediction.

[0056] Fully connected output layer: maps the context vector to the predicted three-phase voltage values ​​of each node, including voltage magnitude and phase angle.

[0057] Furthermore, it should be noted that the collected data is susceptible to factors such as communication interference, measurement equipment errors, or sudden load changes, which may lead to abnormal or abrupt values ​​in the enhanced input vector, thus affecting the stability of voltage prediction. Therefore, in the practical application stage of the model, the enhanced input vector can be normalized first, scaling its features across all dimensions to a uniform numerical range. Subsequently, a sliding window algorithm is used to scan the normalized vector sequence to identify outliers that deviate from the normal fluctuation range. For the detected outliers, they are corrected by combining valid data from adjacent time periods using linear interpolation or historical mean filling, generating a smoother and more reliable corrected enhanced input vector. Finally, the corrected vector is input into the trained voltage prediction model to improve the robustness and practicality of the prediction results.

[0058] 205. Calculate the neutral voltage drop based on the three-phase voltage prediction results, and correct the three-phase voltage prediction results based on the neutral voltage drop to obtain the ground voltage of each node.

[0059] In this step, firstly, the predicted three-phase voltages of each node are output from the voltage prediction model. Combined with the load impedance corresponding to each load node, the injection current for each phase is calculated using the forward method. Then, the three-phase currents are vector-summed to obtain the current flowing through the neutral line of that node. Next, the neutral line voltage drop is calculated based on the neutral line current (the far-end node of the neutral line is treated as a concentrated loop, and the voltage drop on the neutral line is calculated). Finally, the neutral line voltage drops are superimposed onto the three-phase voltage values ​​in the predicted three-phase voltage results to obtain the voltage to ground for each node.

[0060] It should be noted that the load impedance used here refers to the equivalent impedance of the load under the current operating conditions, and its value changes dynamically with the load power and voltage; while the "line impedance" mentioned in the previous steps refers to the impedance of the conductor itself connecting the two nodes, which is a fixed parameter of the power grid and is used to describe the electrical characteristics of the network topology.

[0061] It should be explained that the forward push method used in this step is fundamentally different from the forward push-back substitution method in traditional power flow calculation: the traditional method requires repeated iterations between voltage and current because the two are strongly coupled and must be solved simultaneously to obtain the overall network state; while the forward push process of this invention is based on the premise that the voltage prediction model has already output the three-phase phase voltage, and only uses this as a fixed input to calculate the neutral current and voltage drop in one direction, and corrects the voltage to ground in one go. The whole process has no feedback, no iteration, and does not update the original voltage again.

[0062] The reason for this decoupling is that voltage prediction itself implicitly satisfies the basic power balance during the training phase through physical constraint loss. The correction here is only a post-processing step, designed to compensate for neutral point shift, rather than a re-solution. This decoupling design avoids convergence problems and computational overhead.

[0063] Among them, based on the neutral line voltage drop correction of the three-phase phase voltage prediction results, the formula for the ground voltage of each node is obtained as follows: (Formula 5) That is, the voltage to ground at each node is the sum of the neutral line voltage drop and the three-phase voltage amplitude in the predicted three-phase voltage results.

[0064] 206. Adjustment recommendations based on the voltage output to ground.

[0065] To ensure stable power grid operation and protect users' electrical equipment, it is necessary to determine whether the voltage to ground at each node is within the preset operating range after obtaining the voltage. If the voltage to ground at a node exceeds this range, the direction (too high or too low) and magnitude of the deviation should be further analyzed to provide a basis for subsequent regulation.

[0066] Subsequently, based on the aforementioned deviation direction and magnitude, load regulation can be used as a control measure. Specifically: If the voltage to ground of a node exceeds the preset operating range, an adjustment command is output to increase the controllable load in the preset area corresponding to the node (such as starting an electric water heater, charging energy storage, or increasing the power of an air conditioner); if the voltage to ground of a node is lower than the preset operating range, an adjustment command is output to reduce the non-critical load in the preset area corresponding to the node. The adjustment range in the adjustment command is related to the magnitude of the voltage deviation from the preset operating range.

[0067] Through the above process, the system not only achieves full coverage perception of the voltage status of all nodes in the network, but also generates control suggestions in advance based on the prediction results, effectively improving the power supply quality and operational reliability of the low-voltage distribution network.

[0068] Furthermore, as a response to the above Figure 1 In addition to the implementation of the method shown, this embodiment of the invention also provides a low-voltage distribution network ground voltage calculation device for calculating the above-mentioned voltage. Figure 1 The method shown is implemented accordingly. This device embodiment corresponds to the foregoing method embodiment. For ease of reading, this device embodiment will not repeat the details of the foregoing method embodiment, but it should be clear that the device in this embodiment can implement all the contents of the foregoing method embodiment. Figure 3 As shown, the device includes: The vector acquisition unit 301 is used to acquire a predefined input vector, which includes the amplitude of the three-phase phase voltage at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injection power. The matrix determination unit 302 is used to determine the weighted adjacency matrix between each load node in the distribution network according to a preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connecting nodes. The vector determination unit 303 is used to obtain a response vector based on the weighted adjacency matrix determined by the matrix determination unit 302 and the node injection complex power vector of each load node, and to perform a non-zero element extraction operation on the response vector to obtain a topology vector. The node injection complex power vector is obtained based on the photovoltaic injection power. The vector splicing unit 304 is used to splice the topological vector determined by the vector determination unit 303 with the input vector obtained by the vector acquisition unit 301 to obtain an enhanced input vector, and input the enhanced input vector into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node. The voltage correction unit 305 is used to calculate the neutral line voltage drop based on the three-phase voltage prediction results obtained by the vector splicing unit 304, and correct the three-phase voltage prediction results based on the neutral line voltage drop to obtain the ground voltage of each node.

[0069] Furthermore, as a response to the above Figure 2 In addition to the implementation of the method shown, this embodiment of the invention also provides another low-voltage distribution network ground voltage calculation device for calculating the above-mentioned... Figure 2 The method shown is implemented accordingly. This device embodiment corresponds to the foregoing method embodiment. For ease of reading, this device embodiment will not repeat the details of the foregoing method embodiment, but it should be clear that the device in this embodiment can implement all the contents of the foregoing method embodiment. Figure 4 As shown, the device includes: The vector acquisition unit 301 is used to acquire a predefined input vector, which includes the amplitude of the three-phase phase voltage at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injection power. The matrix determination unit 302 is used to determine the weighted adjacency matrix between each load node in the distribution network according to a preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connecting nodes. The vector determination unit 303 is used to obtain a response vector based on the weighted adjacency matrix determined by the matrix determination unit 302 and the node injection complex power vector of each load node, and to perform a non-zero element extraction operation on the response vector to obtain a topology vector. The node injection complex power vector is obtained based on the photovoltaic injection power. The vector splicing unit 304 is used to splice the topological vector determined by the vector determination unit 303 with the input vector obtained by the vector acquisition unit 301 to obtain an enhanced input vector, and input the enhanced input vector into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node. The voltage correction unit 305 is used to calculate the neutral line voltage drop based on the three-phase voltage prediction results obtained by the vector splicing unit 304, and correct the three-phase voltage prediction results based on the neutral line voltage drop to obtain the ground voltage of each node.

[0070] In one optional implementation, the voltage correction unit 305 is specifically used for: Obtain the load impedance of each load node; Based on the three-phase voltage values ​​and load impedance in the predicted three-phase voltage results, the injection current of each phase of each load node is calculated, and the injection current of each phase of each load node is added together to obtain the neutral current of each load node. Calculate the neutral line voltage drop based on the neutral line current; The neutral line voltage drop is superimposed onto the three-phase voltage values ​​in the three-phase voltage prediction results to obtain the ground voltage of each node.

[0071] In one optional implementation, the training steps of the pre-trained voltage prediction model in the vector splicing unit 304 are as follows: Obtain the training dataset; The training samples in the training dataset are input into a recurrent neural network model based on a two-stage attention mechanism to obtain the three-phase voltage prediction results of each node; Based on the three-phase phase voltage prediction results, the theoretical total injected power is calculated using the three-phase power flow imbalance equation. The physical constraint loss is obtained based on the theoretical total injection power and the actual total injection power, and the data-driven loss is obtained based on the three-phase phase voltage values ​​in the three-phase phase voltage prediction results and the actual three-phase phase voltage values. The physical constraint loss term and the data-driven loss term are weighted and combined to form the total loss function; With the goal of minimizing the total loss function, the parameters of the recurrent neural network model based on the two-stage attention mechanism are iteratively adjusted until the convergence condition is met, thus obtaining a trained voltage prediction model.

[0072] In one optional implementation, when the vector determination unit 303 performs a non-zero element extraction operation on the response vector to obtain a topological vector, it is specifically used for: Remove zero elements from the response vector and retain non-zero elements; The non-zero element is decomposed into its real and imaginary parts; The real part is defined as the topological characteristic vector of the resistor, and the imaginary part is defined as the topological characteristic vector of the reactance. The resistor topology feature vector and the reactance topology feature vector are determined as the topology vector.

[0073] In one optional implementation, after obtaining the ground voltage of each node, the device further includes a voltage determination unit 306, which is specifically used for: Determine whether the voltage to ground of each node is within the preset operating range; If not, a voltage adjustment suggestion is output based on the difference between the voltage to ground of each node and the preset operating range, and the voltage adjustment suggestion includes load regulation.

[0074] In another optional implementation, when the voltage determination unit 306 outputs a voltage adjustment suggestion based on the difference between the voltage to ground of each node and the preset operating range, it is specifically used for: If the voltage to ground of a node exceeds the preset operating range, an adjustment command is output to increase the controllable load within the preset area corresponding to the node. If the voltage to ground of a node is lower than the preset operating range, an adjustment command is output to reduce non-critical loads in the preset area corresponding to the node. The adjustment range in the adjustment command is related to the magnitude of the voltage deviation from the preset operating range.

[0075] In one optional implementation, before inputting the enhanced input vector into the pre-trained voltage prediction model, the device further includes a data processing unit 307, which is specifically used for: The enhanced input vector is normalized to obtain a normalized enhanced input vector; The sliding window algorithm is used to detect outliers in the normalized enhanced input vector, and the outliers are interpolated and corrected to obtain the corrected enhanced input vector. The corrected enhanced input vector is then input into the voltage prediction model.

[0076] Furthermore, embodiments of the present invention also provide a storage medium for storing a computer program, wherein the computer program, when running, controls the device where the storage medium is located to execute the above-described... Figure 1-2 The method for calculating the voltage to ground in low-voltage distribution networks described in the document.

[0077] Furthermore, embodiments of the present invention also provide a processor for running a program, wherein the program executes the above-described... Figure 1-2 The method for calculating the voltage to ground in low-voltage distribution networks described in the document.

[0078] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0079] It is understood that the relevant features in the above methods and apparatus can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between embodiments and do not represent the superiority or inferiority of any particular embodiment.

[0080] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0081] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0082] In addition, the memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0083] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0085] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0087] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0088] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0089] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0090] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0091] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for calculating the voltage to ground in a low-voltage distribution network, characterized in that, The method includes: Obtain a predefined input vector, which includes the amplitude of the three-phase phase voltage at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injected power; The weighted adjacency matrix between each load node in the distribution network is determined according to a preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connecting nodes. The response vector is obtained by using the weighted adjacency matrix and the node-injected complex power vector of each load node, and the non-zero element extraction operation is performed on the response vector to obtain the topology vector. The node-injected complex power vector is obtained based on the photovoltaic injection power. The topology vector is concatenated with the input vector to obtain the enhanced input vector, and the enhanced input vector is input into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node; The neutral voltage drop is calculated based on the predicted three-phase voltage, and the predicted three-phase voltage is corrected based on the neutral voltage drop to obtain the ground voltage of each node.

2. The method according to claim 1, characterized in that, The neutral voltage drop is calculated based on the predicted three-phase voltage, and the predicted three-phase voltage is corrected based on the neutral voltage drop to obtain the ground voltage of each node, including: Obtain the load impedance of each load node; Based on the three-phase voltage values ​​and load impedance in the predicted three-phase voltage results, the injection current of each phase of each load node is calculated, and the injection current of each phase of each load node is added together to obtain the neutral current of each load node. Calculate the neutral line voltage drop based on the neutral line current; The neutral line voltage drop is superimposed onto the three-phase voltage values ​​in the three-phase voltage prediction results to obtain the ground voltage of each node.

3. The method according to claim 1, characterized in that, The training steps for a pre-trained voltage prediction model are as follows: Obtain the training dataset; The training samples in the training dataset are input into a recurrent neural network model based on a two-stage attention mechanism to obtain the three-phase voltage prediction results of each node; Based on the three-phase phase voltage prediction results, the theoretical total injected power is calculated using the three-phase power flow imbalance equation. The physical constraint loss is obtained based on the theoretical total injection power and the actual total injection power, and the data-driven loss is obtained based on the three-phase phase voltage values ​​in the three-phase phase voltage prediction results and the actual three-phase phase voltage values. The physical constraint loss term and the data-driven loss term are weighted and combined to form the total loss function; With the goal of minimizing the total loss function, the parameters of the recurrent neural network model based on the two-stage attention mechanism are iteratively adjusted until the convergence condition is met, thus obtaining a trained voltage prediction model.

4. The method according to claim 1, characterized in that, Performing a non-zero element extraction operation on the response vector to obtain a topological vector includes: Remove zero elements from the response vector and retain non-zero elements; The non-zero element is decomposed into its real and imaginary parts; The real part is defined as the topological characteristic vector of the resistor, and the imaginary part is defined as the topological characteristic vector of the reactance. The resistor topology feature vector and the reactance topology feature vector are determined as the topology vector.

5. The method according to claim 1, characterized in that, After obtaining the ground voltage of each node, the method further includes: Determine whether the voltage to ground of each node is within the preset operating range; If not, a voltage adjustment suggestion is output based on the difference between the voltage to ground of each node and the preset operating range, and the voltage adjustment suggestion includes load regulation.

6. The method according to claim 5, characterized in that, Based on the difference between the voltage to ground at each node and the preset operating range, output voltage adjustment suggestions are provided, including: If the voltage to ground of a node exceeds the preset operating range, an adjustment command is output to increase the controllable load within the preset area corresponding to the node. If the voltage to ground of a node is lower than the preset operating range, an adjustment command is output to reduce non-critical loads in the preset area corresponding to the node. The adjustment range in the adjustment command is related to the magnitude of the voltage deviation from the preset operating range.

7. The method according to claim 1, characterized in that, Before inputting the enhanced input vector into the pre-trained voltage prediction model, the method further includes: The enhanced input vector is normalized to obtain a normalized enhanced input vector; The sliding window algorithm is used to detect outliers in the normalized enhanced input vector, and the outliers are interpolated and corrected to obtain the corrected enhanced input vector. The corrected enhanced input vector is then input into the voltage prediction model.

8. A device for calculating the voltage to ground in a low-voltage distribution network, characterized in that, The device includes: The vector acquisition unit is used to acquire a predefined input vector, which includes the three-phase phase voltage amplitude at the head end of the low-voltage side of the distribution transformer, the three-phase active power and three-phase reactive power of each load node, and the photovoltaic injection power. The matrix determination unit is used to determine the weighted adjacency matrix between each load node in the distribution network according to a preset determination method. Each element in the weighted adjacency matrix represents the impedance value of the line between the connecting nodes. The vector determination unit is used to obtain a response vector based on the weighted adjacency matrix determined by the matrix determination unit and the node injection complex power vector of each load node, and to perform a non-zero element extraction operation on the response vector to obtain a topology vector. The node injection complex power vector is obtained based on the photovoltaic injection power. The vector concatenation unit is used to concatenate the topological vector determined by the vector determination unit with the input vector obtained by the vector acquisition unit to obtain an enhanced input vector, and input the enhanced input vector into the pre-trained voltage prediction model to obtain the three-phase voltage prediction results of each node; The voltage correction unit is used to calculate the neutral line voltage drop based on the three-phase voltage prediction results obtained by the vector splicing unit, and to correct the three-phase voltage prediction results based on the neutral line voltage drop to obtain the ground voltage of each node.

9. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the storage medium to perform the method for calculating the voltage to ground of a low-voltage distribution network as described in any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes the method for calculating the voltage to ground of a low-voltage distribution network as described in any one of claims 1 to 7.

Images