Power distribution network voltage control method and related device

Abstract

The invention discloses a power distribution network voltage control method and a related device, and the method comprises the steps: carrying out the precise sensing of a multi-scale voltage situation in a space-time and physical mixed manner, namely predicting the future node voltage and a corresponding multi-scale node voltage out-of-limit risk index, obtaining a basic control strategy through a main channel, and carrying out the prediction of a multi-scale node voltage. A corresponding robust compensation strategy is obtained by adopting an anti-interference channel, the high-dimensional problem solving efficiency is improved in combination with a Red-beak-Baijie optimization algorithm, all distributed energy sources are equivalent to a quantized virtual impedance network in dual-channel MPC optimization and power distribution network node voltage control, the dimension optimization bottleneck is broken through, and the power distribution network node voltage control efficiency is improved. Finally, a global prediction-hierarchical decision-edge execution full-chain control system can be formed, and the method is suitable for voltage stability control in a high-proportion distributed energy access scene.

Description

Technical Field

[0001] This invention relates to a voltage control method and related devices for power distribution networks, belonging to the field of power distribution network control technology. Background Technology

[0002] With the rapid development of new power systems and the large-scale integration of flexible resources such as distributed photovoltaics, energy storage, and electric vehicles, distribution networks are facing increasingly severe power quality problems such as voltage fluctuations and exceeding limits. Traditional distribution network voltage control methods rely on fixed regulation modes, which are difficult to cope with voltage stability control in scenarios with a high proportion of distributed energy integration. Summary of the Invention

[0003] This invention provides a voltage control method and related device for power distribution networks, which solves the problems in the background art.

[0004] According to one aspect of this application, a power distribution network voltage control method is provided, comprising: Input the time series data of node voltage and topology data of the distribution network into the prediction model to obtain the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales. Based on the predicted node voltage values ​​and the actual node voltage values, an error assessment of the predicted node voltage values ​​is performed. Based on the error assessment results and the node voltage over-limit risk index, dual-channel MPC optimization is performed to obtain a set of node voltage control strategies. In the dual-channel MPC optimization, the main channel and the disturbance rejection channel generate the basic control strategy and the robust compensation strategy, respectively. The node voltage control strategy is the strategy obtained by weighted fusion of the basic control strategy and the corresponding robust compensation strategy. The Red-billed Blue Magpie optimization algorithm is used to obtain the optimal node voltage control strategy from the set of node voltage control strategies. In the Red-billed Blue Magpie optimization algorithm, the difference between the objective function of the current solution and the global optimal solution is used to simulate the energy barrier in quantum mechanics, and the probability of an individual in the optimization solution passing through the energy barrier is calculated. Based on the optimal node voltage control strategy, the distribution network node voltage is controlled; in the dual-channel MPC optimization and distribution network node voltage control, all distributed energy resources are equivalent to a quantized virtual impedance network.

[0005] Furthermore, the process by which the prediction model processes node voltage time-series data and topology data includes: Based on the topology data, a dynamic adjacency matrix of the distribution network is constructed; where the elements in the dynamic adjacency matrix are the dynamic adjacency weights between nodes, and the dynamic adjacency weights between nodes represent the electrical association strength. Wavelet scattering feature extraction is performed on node voltage time series data to obtain node voltage features at multiple scales; Based on the dynamic adjacency matrix and node voltage characteristics, a physical constraint attention mechanism is used to obtain the node voltage characteristics of the fused topology data. By inputting the node voltage characteristics of the fused topology data into the encoder, the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators at multiple scales are obtained.

[0006] Furthermore, the dynamic adjacency weight formula is as follows: ; In the formula, for t Time of the first i The node and the first j The dynamic adjacency weight of each node. σ This is the normalization function. V i and Q i For the first i Node voltage and reactive power, V j and P j For the first j Node voltage and active power, For the first i The voltage of the first node is related to the first node voltage. j Sensitivity of active power at each node For the first i The reactive power of the node affects the first j Sensitivity of node voltage, || z ij ||For the first i The node and the first j The impedance amplitude of the line between nodes. I ij For the first i The node and the first j The current in the lines between nodes; The node voltage characteristics at multiple scales include low-frequency trend components, mid-frequency fluctuation components, and high-frequency abrupt change components; the wavelet scattering feature extraction formula is: ; In the formula, W V This is the multi-scale node voltage feature vector extracted by wavelet scattering. These are the wavelet basis functions used to extract low-frequency trend components. The Mexican hat wavelet function is used to extract frequency fluctuation components. Here is the Haar wavelet function used for high-frequency abrupt changes, and * denotes convolution. VThis represents the node voltage vector.

[0007] Furthermore, the optimization of the main channel takes the node voltage prediction value as the scenario and aims to minimize the sum of voltage tracking deviation penalty, control cost, and control increment smoothing penalty; the objective function for the main channel optimization is: ; In the formula, V t for t The node voltage vector at time t. V ref This is a vector of node voltage reference values. W v The voltage tracking weight matrix is ​​used to reflect the importance of the voltage at each node. The weighted square norm is a dimensional indicator used to quantify voltage deviation or control costs. The weighted square norm is used to measure the cost of control. u t and u t-1 They are respectively t Time and t The control variable at time -1 is the flexible regulation resource of the distribution network. R The control cost weight matrix is ​​used to suppress sudden changes in control variables. ρ For smoothing coefficients, N To optimize the prediction time domain length, u To optimize variables, Represents the norm.

[0008] Furthermore, considering the uncertainties of distributed energy output and load fluctuations, the optimization of the disturbance rejection channel generates a robust compensation strategy corresponding to the basic control strategy; the objective function for the disturbance rejection channel optimization is: ; ; In the formula, This is the set of distributions of uncertainty parameters that describe the statistical characteristics of prediction errors in the error evaluation results. P for The probability distribution of uncertainty for P The mathematical expectation operator below, For disturbance in P The expected value under, for t The prediction deviation of the node voltage at time t. Q The node voltage deviation penalty matrix is ​​represented by the superscript. T Indicates transpose. Let be the perturbation vector. To predict the mean of the disturbance, For Wasserstein distance, To perturb the upper bound of expectation, The Wasserstein radius is... N For optimized prediction time domain length.

[0009] Furthermore, during weighted fusion, the weights of the basic control strategy are the main channel weights, and the weights of the robust compensation strategy are the disturbance rejection channel weights. The formula for determining the weights is: ; ; ; In the formula, for t Based on the error intensity in the error assessment results at all times E t and node voltage over-limit risk indicators R t Determined initial weights for the main channel Baseline value, Control E t and R t The sensitivity coefficient, for t The weight of the main channel at any given time. for t The weights of the time-sensitive channel, where clip is the clip function. σ This is the normalization function. These are the lower and upper limits of the main channel weight, respectively.

[0010] Furthermore, the process of obtaining a quantized virtual impedance network includes: Distributed resources are represented as quantized virtual impedances, and a game equilibrium model between distributed energy resources and the power grid is constructed. The ADMM algorithm accelerated by Nesterov is used to iteratively solve the game equilibrium model between distributed energy resources and the power grid to obtain the quantized virtual impedance network. In the iterative solution, the weights of the quantized virtual impedances are updated based on sensitivity analysis. The quantized virtual impedances are used to describe the equivalent impedance characteristics of distributed energy resources to the power grid.

[0011] Furthermore, the distributed resources are equivalent to quantized virtual impedance, as shown in the formula: ; In the formula, For quantized virtual impedance, nThe total number of distributed resources participating in the aggregation. For the first k The ground-state impedance of a distributed resource. To entangle the weighting coefficients, To reflect the first k The phase angle of dynamic response latency of a distributed resource. For the first k Device operating status characteristic information of a distributed resource It is a tensor product operator.

[0012] Furthermore, the game equilibrium model between distributed energy resources and the power grid is as follows: ; In the formula, Z v For virtual impedance matrix, V ref This is a vector of node voltage reference values. V For node voltage vectors, λ This is a regularization coefficient used to balance the accuracy of voltage tracking at the balancing node with the cost of voltage regulation. Represents the norm, superscript T Indicates transpose. Let Δ be the trace norm of the virtual impedance. P This represents the power regulation vector of distributed energy resources. μ As a weighting factor, V nom The local nominal voltage for distributed energy resources. P The actual power vector at the nodes is determined by the power flow equations of the distribution network. Virtual impedance and node voltage deviation The real part of the mapping relation, This is a Gaussian noise term.

[0013] According to another aspect of this application, a power distribution network voltage control device is provided, comprising: The prediction module inputs the time-series data of node voltage and topology data of the distribution network into the prediction model to obtain the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales. The evaluation module assesses the error of the node voltage prediction based on the predicted node voltage and the actual node voltage. The dual-channel MPC optimization module performs dual-channel MPC optimization based on error assessment results and node voltage over-limit risk indicators to obtain a set of node voltage control strategies. In the dual-channel MPC optimization, the main channel and the disturbance rejection channel generate basic control strategies and robust compensation strategies, respectively. The node voltage control strategy is a strategy obtained by weighted fusion of the basic control strategy and the corresponding robust compensation strategy. The strategy optimization module uses the Red-billed Blue Magpie optimization algorithm to obtain the optimal node voltage control strategy from the set of node voltage control strategies. In the Red-billed Blue Magpie optimization algorithm, the difference between the objective function of the current solution and the global optimal solution is used to simulate the energy barrier in quantum mechanics, and the probability of an individual in the optimization solution passing through the energy barrier is optimized. The voltage control module performs distribution network node voltage control according to the optimal node voltage control strategy; in the dual-channel MPC optimization and distribution network node voltage control, all distributed energy resources are equivalent to a quantized virtual impedance network.

[0014] According to another aspect of this application, a computer-readable storage medium is provided that stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform a power distribution network voltage control method.

[0015] According to another aspect of this application, a computer device is provided, including one or more processors and one or more memories, wherein one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing a power distribution network voltage control method.

[0016] The beneficial effects achieved by this invention are as follows: This invention adopts a spatiotemporal and physical hybrid approach for accurate perception of multi-scale voltage status, that is, predicting future node voltages and corresponding multi-scale node voltage exceedance risk indicators. It uses the main channel to obtain the basic control strategy and the anti-disturbance channel to obtain the corresponding robust compensation strategy. Combined with the Red-beaked Blue Magpie optimization algorithm, it improves the efficiency of solving high-dimensional problems. In dual-channel MPC optimization and distribution network node voltage control, it equates all distributed energy sources to a quantized virtual impedance network, breaking through the optimization dimension bottleneck. Finally, it can form a full-chain control system of "global prediction - hierarchical decision-making - edge execution", which is suitable for voltage stability control in scenarios with a high proportion of distributed energy access. Attached Figure Description

[0017] Figure 1 A flowchart of a power distribution network voltage control method; Figure 2 A flowchart illustrating the processing of node voltage time-series data and topology data for the prediction model; Figure 3 This is a block diagram of a power distribution network voltage control device. Detailed Implementation

[0018] 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 embodiments of this application, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0020] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0021] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0022] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0023] It should be noted that similar symbols and letters in the following figures represent similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0024] GTPT (Graph Temporal Physics-informed Transformer) is a deep learning architecture that integrates graph neural networks, time series modeling, and physical constraints. It aims to address the challenges of modeling spatiotemporal physical systems. This model captures complex dependencies in graph-structured data through an attention mechanism and embeds physical knowledge such as partial differential equations (PDEs) to improve prediction accuracy, generalization ability, and data efficiency. Dual-channel MPC optimization is an extension of Model Predictive Control (MPC). It improves system performance by decomposing the control task into two parallel or cooperative channels. This design is typically used to handle multi-timescale dynamics, separate priorities, or enhance robustness. The Red-billed Blue Magpie Optimization Algorithm is a novel metaheuristic optimization algorithm. By simulating the foraging behavior of red-billed blue magpies, it provides a new approach to solving complex optimization problems. Its innovation lies in its dynamic cooperation mechanism, which automatically adjusts the search strategy according to the problem complexity, compensating for the shortcomings of traditional algorithms in modeling collaborative group behavior.

[0025] This application provides a power distribution network voltage control method based on GTPT, dual-channel MPC optimization, and the Red-beaked Blue Magpie optimization algorithm. This control method can be executed by a control device, which can be a terminal device or a server. The terminal device can include, but is not limited to, mobile phones, computers, etc. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, big data, and artificial intelligence platforms, etc. Optionally, this control method can also be executed collaboratively by multiple electronic devices with computing power. For ease of explanation, subsequent embodiments will be described as being executed by a control device.

[0026] See Figure 1 , Figure 1 This is a flowchart of a power distribution network voltage control method provided in an embodiment of this application. The control method can be executed by a control device and may include at least the following steps: Step 1: Input the node voltage time-series data and topology data of the distribution network into the prediction model (i.e., Figure 1 The spatiotemporal-physical fusion prediction model in the data center obtains the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales.

[0027] It should be noted that the node voltage time series data and topology data of the distribution network are... Figure 1 Medium voltage state monitoring data, step 1 specifically involves using GTPT to achieve multi-scale voltage situation awareness, see [link / reference]. Figure 2 In some embodiments, the specific process by which the prediction model processes node voltage time-series data and topology data may include: 11) Based on the topology data, construct the dynamic adjacency matrix of the distribution network, extract wavelet scattering features from the node voltage time series data, and obtain the node voltage features at multiple scales; where the elements in the dynamic adjacency matrix are the dynamic adjacency weights between nodes, and the dynamic adjacency weights between nodes characterize the electrical correlation strength.

[0028] It should be noted that, in order to capture the impact of changes in the distribution network topology on voltage propagation in real time, quantify the electrical correlation strength between nodes, and solve the problem that traditional static adjacency matrices cannot adapt to dynamic topology changes, a dynamic topology representation is adopted here. In the dynamic adjacency matrix, the dynamic adjacency weight can be expressed by the formula: ; In the formula, for t Time of the first i The node and the firstj The dynamic adjacency weight of each node characterizes the strength of electrical association. σ This is a normalization function, which normalizes the weights to the [0,1] interval. Specifically, the Sigmoid function can be used. V i and Q i For the first i Node voltage and reactive power, V j and P j For the first j Node voltage and active power, For the first i The voltage of the first node is related to the first node voltage. j The sensitivity of active power at each node reflects the impact of changes in active power on voltage propagation. For the first i The reactive power of the node affects the first j The sensitivity of the node voltage characterizes the reactive power regulation capability. z ij ||For the first i The node and the first j The impedance magnitude of the lines between nodes; the smaller the impedance, the stronger the electrical connection. I ij For the first i The node and the first j The current in the line between nodes, tanh( I ij )for I ij The hyperbolic tangent transform suppresses weight saturation when the current is too large.

[0029] Dynamic topology representation provides a spatial correlation basis for subsequent physical constraint attention mechanisms. By quantifying the electrical coupling relationships between nodes, a graph-time-series transformation network of the power grid is constructed.

[0030] It should be noted that multi-scale fluctuation features are extracted from the node voltage time series data, and low-frequency trend components (hourly-level changes), medium-frequency fluctuation components (minute-level fluctuations) and high-frequency abrupt change components (second-level perturbations) are separated to enhance the model's ability to capture voltage changes on fast / slow time scales.

[0031] The wavelet scattering feature extraction formula can be expressed as: ; In the formula, W V This is the multi-scale node voltage feature vector extracted by wavelet scattering, which integrates low-frequency trend components, mid-frequency fluctuation components, and high-frequency abrupt change components. These are the wavelet basis functions used to extract low-frequency trend components. The Mexican hat wavelet function is used to extract frequency fluctuation components. Here, * represents the Haar wavelet function used for high-frequency abrupt changes, and * denotes convolution, achieving time-frequency decomposition of the signal and the wavelet basis. V For node voltage vectors, They are respectively V The first and second derivatives.

[0032] Low-frequency trend components are used to correct the long-term weight allocation of the dynamic adjacency matrix, mid-frequency fluctuation components guide the physical attention mechanism to model minute-level adjustment actions, and high-frequency mutation components trigger the fast response strategy of the local control layer.

[0033] 12) Based on the dynamic adjacency matrix and node voltage characteristics, a physical constraint attention mechanism is used to obtain the node voltage characteristics of the fused topology data.

[0034] It should be noted that 12) uses a dynamic adjacency matrix to characterize the electrical coupling strength and embeds the power flow equation into the feature evolution process through a physical constraint attention mechanism.

[0035] It should be noted that the core of the physics-constrained attention mechanism lies in using the dynamic adjacency matrix as prior physical knowledge to modify the weight distribution of the traditional attention mechanism, forcing the model to focus on nodes with close electrical connections; the specific formula is as follows: ; ; In the formula, H To integrate the node voltage characteristics of the topology data (i.e., the updated node state representation). These are the query matrix, key matrix, and value matrix, respectively. They are respectively the corresponding The learnable linear mapping weight matrix, superscript T Indicates transpose. d k Let be the dimension of the key vector. Taking the square root is to prevent the gradient from vanishing due to excessively large dot product values. This is the physical constraint coefficient, initially set to 1.0. A This is the dynamic adjacency weight.

[0036] Will W V Linear mapping is performed to generate query vectors, key vectors, and value vectors. Then, a physical bias term is constructed using a dynamic adjacency matrix and injected into the calculation of attention scores to enhance the model's focus on nodes with strong electrical connections. Finally, the node state representation, which integrates physical information, is obtained through normalization.

[0037] It should be noted that the above dynamic topology representation constructs the spatiotemporal graph structure of the power grid, quantifies the dynamic electrical connections between nodes, and the physical constraint attention mechanism is based on topological connections and power flow equations, allocating attention weights that conform to physical laws; wavelet scattering features provide multi-scale voltage fluctuation characteristics, driving the model to distinguish between long-term / medium-term / short-term control needs.

[0038] 13) Input the node voltage characteristics of the fused topology data into the encoder to obtain the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales.

[0039] It should be noted that the encoder adopts a multi-task decoding head structure and performs the following steps in sequence: (1) Feature decoding: The high-dimensional feature vector output by the encoder enters two parallel fully connected layers, which are used for regression tasks and classification / evaluation tasks, respectively; (2) Voltage prediction branch: The first branch directly outputs the node voltage prediction values ​​at multiple future times through a linear activation function; (3) Voltage prediction branch: The second branch combines the voltage safety threshold to output the corresponding node voltage over-limit risk index, probability index based on prediction distribution, etc.

[0040] Step 2: Based on the predicted node voltage values ​​and the actual node voltage values, perform an error assessment of the predicted node voltage values.

[0041] Step 3: Based on the error assessment results and the node voltage over-limit risk index, perform dual-channel MPC optimization to obtain a set of node voltage control strategies. In the dual-channel MPC optimization, the main channel and the disturbance rejection channel generate the basic control strategy and the robust compensation strategy, respectively. The node voltage control strategy is the strategy obtained by weighted fusion of the basic control strategy and the corresponding robust compensation strategy.

[0042] It should be noted that traditional MPC has a contradiction between conservatism and robustness. To ensure the robustness of voltage regulation while avoiding the loss of regulation margin caused by excessive conservatism, a dual-channel logic is designed here.

[0043] The main channel optimizes the nominal scenario (undisturbed) based on the prediction model to quickly generate a basic control strategy. Specifically, it uses the node voltage prediction value as the scenario and aims to minimize the sum of voltage tracking deviation penalty, control cost, and control increment smoothing penalty to generate the basic control strategy. The disturbance rejection channel considers the uncertainty of distributed energy output and load fluctuations to ensure the feasibility of the basic control strategy under disturbance scenarios. Specifically, it generates a robust compensation strategy corresponding to the basic control strategy under the consideration of the uncertainty of distributed energy output and load fluctuations. Through decoupled optimization of the two channels, the main channel is responsible for efficiency, while the disturbance rejection channel ensures robustness, avoiding the trade-offs of a single model. The weights of the two channels are adaptively adjusted according to the real-time prediction error to balance optimality and conservatism.

[0044] The objective function for optimizing the main channel can be expressed as: ; In the formula, V t for t The node voltage vector at time t. V ref This is a vector of node voltage reference values, typically expressed as a per-unit value of 1.0. W v The voltage tracking weight matrix (diagonal matrix) reflects the importance of the voltage at each node. The weighted square norm, internally representing the voltage deviation, is a dimensional indicator used to quantify voltage deviation or control costs. Its physical meaning is to measure voltage tracking error. The weighted square norm is used to measure the cost of control, i.e., the price. u t and u t-1 They are respectively t Time and t The control variables at time -1 (such as photovoltaic reactive power output and energy storage charging and discharging power) are the flexible regulation resources of the distribution network. R The control cost weight matrix is ​​used to suppress sudden changes in control variables. ρ To smooth out the coefficients and avoid drastic fluctuations in control commands, N To optimize the prediction time domain length, u (i.e., the subscript of min) represents the optimization variable, namely voltage. Represents the norm.

[0045] The objective function for optimizing the anti-interference channel can be expressed as: ; ; In the formula, This is the set of distributions of uncertainty parameters that describe the statistical characteristics of prediction errors in the error evaluation results. P for The probability distribution of uncertainty for P The mathematical expectation operator below, For disturbance in P The expected value under, for t The prediction deviation of the node voltage at time t. Q The node voltage deviation penalty matrix is ​​typically a diagonal matrix with superscripts. T Indicates transpose. This is a disturbance vector (such as photovoltaic output deviation, load fluctuation). To predict the mean of the disturbance, The Wasserstein distance controls the "width" of the distribution set. To constrain the upper bound of the expected perturbation, the intensity of uncertainty is limited. The Wasserstein radius is... N For optimized prediction time domain length.

[0046] During weighted fusion, the weights of the basic control strategy are the main channel weights, and the weights of the robust compensation strategy are the disturbance rejection channel weights. The formula for determining the weights is as follows: ; ; ; In the formula, for t Based on the error intensity in the error assessment results at all times E t and node voltage over-limit risk indicators R t Determined initial weights for the main channel Baseline value, Control E t and R t The sensitivity coefficient, for t The weight of the main channel at any given time. for t The weights of the time-sensitive channel, where clip is the clip function. σ This is the normalization function. These are the lower and upper limits of the main channel weight, respectively.

[0047] See Figure 1 The main process of dual-channel MPC optimization is as follows: first, determine the weight of the main channel and the weight of the anti-interference channel; then, perform nominal optimization on the main channel to obtain the nominal control sequence, i.e. the basic control strategy; perform robust optimization on the anti-interference channel to obtain the robust compensation strategy; and finally, perform weighted fusion of the two.

[0048] Step 4: The Red-billed Blue Magpie optimization algorithm is used to obtain the optimal node voltage control strategy from the set of node voltage control strategies. In the Red-billed Blue Magpie optimization algorithm, the difference between the objective function of the current solution and the global optimal solution is used to simulate the energy barrier in quantum mechanics, and the probability of an individual in the optimization solution passing through the energy barrier is calculated.

[0049] It should be noted that the Red-billed Blue Magpie optimization algorithm is a mature method. It mimics the visual foraging and cooperative vigilance behaviors of Red-billed Blue Magpie flocks, combining this with the "quantum tunneling" theory of quantum computing. This allows individuals in the optimization process to cross the "potential barrier" (i.e., escape local extrema) with a certain probability, thus solving high-dimensional non-convex optimization problems. The optimization process mainly includes, in sequence, visual foraging search, quantum tunneling escape, energy state assessment, and determination of optimal control measurements.

[0050] The visual foraging search formula can be expressed as: ; In the formula, The first k 1+1 generation and the first k 1st generation i The position (solution vector) of an individual. For the first k 1st generation j The position of an individual, S This represents the set of high-quality solutions that rank in the top 30% of fitness within an individual's visual range. The learning rate controls the speed at which the solution approaches a high-quality solution; its initial value is 8, and it decreases with each iteration. This is the distance attenuation coefficient. Indicates the first j 1 individual and the first i The Euclidean distance of an individual This is the Lévy flight coefficient, typically taken as 0.01 to 0.1. For Levi's flight random item, obey , Let be a random variable that follows a normal distribution.

[0051] The quantum tunneling escape mechanism is that when an individual gets trapped in a local optimum, it escapes with probability. Triggering dimensional travel; the probability can be expressed by the formula:

[0052]

[0053] In the formula, The difference between the objective function of the current solution and the global optimal solution. This is the tunneling sensitivity coefficient. For temperature parameters, as k 1. Exponential decay, To control the decay rate, the activity level that escapes local optima is... T 0 represents the initial temperature parameter, which increases with each iteration ( kThe increase in temperature parameter leads to an exponential decay, resulting in a higher quantum tunneling probability in the early stages (allowing birds to explore space and find food as much as possible) and a lower probability in the later stages (facilitating convergence).

[0054] The energy state adaptive strategy can be expressed as: ; In the formula, For learning rate, They are respectively The upper and lower limits, For the first i The energy state (inverse of fitness) of an individual. These represent the population average, optimal, and worst energy values, respectively.

[0055] The optimization process described above can be specifically described as follows: during the iteration process, the objective function value corresponding to the current individual position is monitored in real time; the difference between the objective function of the current solution and the global optimal solution is calculated. This difference simulates the energy barrier in quantum mechanics; it is calculated using the formula. Generate a random number between [0, 1]. If the random number is less than 1, then... If the current position is changed, it triggers dimensional crossing, which forcibly changes the current position of the individual, causing it to jump out of the current search area; otherwise, it continues to perform the normal visual foraging search.

[0056] Step 5: Perform distribution network node voltage control based on the optimal node voltage control strategy.

[0057] It should be noted that there are numerous and heterogeneous distributed resources (photovoltaics, energy storage, etc.) in the distribution network. Directly controlling all devices would lead to an explosion of optimization dimensions. Therefore, they can be equivalent to a unified "virtual impedance network", which can transform the coordinated control of large-scale devices into an optimization problem of a small number of circuit parameters, thus greatly reducing the amount of computation.

[0058] Traditional virtual impedance networks tend to overlook the dynamic characteristics of devices and the synergistic effect of quantum computing, leading to computational inefficiencies as the resource aggregation dimension increases quadratically with the number of nodes. Static impedance matching fails to adapt to rapid fluctuations in renewable energy output. Furthermore, the lack of quantum entanglement in multi-device collaboration limits adjustment precision. Therefore, in some embodiments, a quantized virtual impedance network is proposed, mapping physical devices to a quantum Hilbert space differential game optimization model. This establishes a device-grid game equilibrium model and updates impedance weights online based on sensitivity analysis.

[0059] Specifically, in dual-channel MPC optimization and distribution network node voltage control, all distributed energy resources are equivalent to a quantized virtual impedance network. In dual-channel MPC optimization, the virtual impedance network serves as an equivalent aggregation model of the controlled object, used to predict and calculate control strategies. In distribution network node voltage control, specifically at the edge execution end, after receiving the virtual impedance command, the device controller adjusts its output according to the impedance value to support the power grid.

[0060] The equivalent process can be represented as: 1) Equivalent distributed resources to quantized virtual impedance and construct a game equilibrium model between distributed energy and the power grid; whereby the quantized virtual impedance is used to describe the equivalent impedance characteristics of distributed energy to the power grid.

[0061] Distributed resources can be equated to quantized virtual impedance, which can be expressed by the following formula: ; In the formula, For quantized virtual impedance, n The total number of distributed resources participating in the aggregation. For the first k The ground-state impedance of a distributed resource. To entangle the weighting coefficients, , To reflect the first k The phase angle of the dynamic response latency of a distributed resource reflects the dynamic response latency of the device. For the first k The device operating status characteristics of a distributed resource, namely device state qubits, encode information such as SOC and output mode. It is a tensor product operator.

[0062] The game equilibrium model between distributed energy resources and the power grid can be expressed as: ; In the formula, Z v This is a virtual impedance matrix that describes the equivalent impedance characteristics of a device to the power grid. V ref This is a vector of node voltage reference values, typically expressed as a per-unit value of 1.0. V For node voltage vectors, λ The regularization coefficient, which balances the voltage tracking accuracy and voltage regulation cost of the balancing node, is indicated by the superscript. T Indicates transpose. Δ is the trace norm of the virtual impedance, reflecting the total cost of the adjustment action. P This refers to the power regulation vector of distributed energy resources, including photovoltaic reactive power output and energy storage charging and discharging. μAs a weighting factor, the control equipment's regulation benefits are weighted relative to voltage deviation. V nom This refers to the local nominal voltage of distributed energy resources, reflecting their expected operating state. P The actual power vector at the nodes is determined by the power flow equations of the distribution network. Virtual impedance and node voltage deviation The real part of the mapping relation, The term represents Gaussian noise, and the standard deviation represents the statistical characteristics of the prediction error.

[0063] 2) The Nesterov-accelerated ADMM algorithm is used to iteratively solve the game equilibrium model between distributed energy resources and the power grid to obtain a quantized virtual impedance network; in the iterative solution, the weights of the quantized virtual impedance are updated based on sensitivity analysis.

[0064] The iterative formula for the ADMM algorithm accelerated by Nesterov can be expressed as: ; In the formula, For the first The virtual impedance matrix of the next iteration To augment the Lagrange function, the parameters Located between 0.1 and 1.0, The first Second and third The power regulation vector of distributed energy in the next iteration. The first Second and third The dual variable or Lagrange multiplier in the ADMM (Alternating Direction Multiplier) algorithm in the next iteration.

[0065] After every 5 iterations, a momentum term is applied to accelerate the process. ,parameter .

[0066] The following convergence conditions must be met: Original residuals: ; Dual residuals: ; Maximum number of iterations: .

[0067] It should be noted that the traditional impedance weight is a static weight. Static weights cannot adapt to the rapid fluctuations in new energy output and load, which can lead to local optima. Furthermore, they do not consider the real-time sensitivity changes of voltage to impedance. Therefore, this paper updates the quantized virtual impedance weights based on sensitivity analysis. The core of this approach includes: sensitivity analysis, which calculates the partial derivative of voltage with respect to virtual impedance; random perturbation injection, which prevents the weight update from getting stuck in local optima; and hyperbolic tangent constraints, which limit the reactive power adjustment range.

[0068] It should be noted that updating the quantized virtual impedance weights based on sensitivity analysis can include the following specific steps: S1) Initialization parameters and disturbance generation.

[0069] The impedance weight vector α1 and the reactive power adjustment coefficient vector β1 are initialized as zero vectors. In order to avoid the optimization getting trapped in local optima, a random disturbance factor δ that follows a uniform distribution U(-ε1,ε1) is generated; where ε1 is a preset disturbance coefficient of 0.05.

[0070] S2) Based on the sensitivity update of node traversal, perform the following calculations for each node in the distribution network: Calculate the overall sensitivity index: Based on the voltage-impedance sensitivity matrix, calculate the first... i The average sensitivity S_avg of each node is given by the formula: S_avg = (1 / I )×∑S( i , k );in, I Let S be the total number of nodes. i , k ) represents the sensitivity matrix to voltage-impedance. i The sensitivity of each row is summed, and in the sensitivity matrix, i Representative node, k Representing distributed resources, element S( i , k ) can represent the first k The impedance change of the first distributed resource affects the second... i The influence of node voltage, i.e., sensitivity.

[0071] Update impedance weights: Combine the sigmoid activation function and the random perturbation factor to calculate the initial impedance weights α1_i, with the formula α1_i=σ(S_avg)×(1+δ).

[0072] Update reactive power regulation coefficient: Based on the deviation between the reactive power reference value Q_ref and the actual value Q_act, update the reactive power regulation coefficient β1_i using the hyperbolic tangent function. The formula is: β1_i=tanh(Q_ref / (Q_act+ε1))×η1; where η1 is the learning rate.

[0073] S3) Global normalization processing: In order to meet the physical constraints, the vectors calculated above are normalized.

[0074] Specifically, impedance weight normalization: The impedance weight vector is normalized using the L1 norm to ensure that the sum of the impedance weights of all distributed resources is 1. Reactive power coefficient normalization: The reactive power regulation coefficient vector is normalized using the infinity norm to ensure that the maximum regulation coefficient does not exceed 1.

[0075] S4) Outputs the final normalized impedance weight vector and reactive power adjustment coefficient vector, which are used to update the quantized virtual impedance network.

[0076] The above method can realize a multi-timescale hierarchical control system, specifically including three levels: 1. Global optimization layer, with a timescale of 5 minutes to 5 seconds, responsible for the medium- and long-term voltage planning of the distribution network. The technical implementation is to use the CTPT model for voltage situation prediction and combine it with the Red-beaked Blue Magpie optimization algorithm for global strategy optimization; 2. Regional regulation layer, with a timescale of 1 minute to 5 minutes, responsible for the coordinated scheduling and regional balancing of distributed resources. The technical implementation is to aggregate resources based on a quantum virtual impedance network and use the distributed ADMM algorithm to solve the game equilibrium; 3. Local control layer, with a timescale of seconds or minutes, responsible for rapid dynamic compensation to deal with high-frequency voltage fluctuations. The technical implementation is to achieve real-time response at the edge based on a reinforcement learning agent.

[0077] The above method employs a spatiotemporal and physical hybrid approach for precise multi-scale voltage situation perception, predicting future node voltages and corresponding multi-scale node voltage exceedance risk indicators. It uses the main channel to acquire basic control strategies and the anti-disturbance channel to acquire corresponding robust compensation strategies. Combined with the Red-beaked Blue Magpie optimization algorithm, it improves the efficiency of solving high-dimensional problems. In dual-channel MPC optimization and distribution network node voltage control, it equates all distributed energy sources to a quantized virtual impedance network, breaking through the optimization dimension bottleneck. Ultimately, it can form a full-chain control system of "global prediction - hierarchical decision-making - edge execution", which is suitable for voltage stability control in scenarios with a high proportion of distributed energy access.

[0078] See Figure 3 , Figure 3 This is a block diagram of a power distribution network voltage control device provided in an embodiment of this application. This device is a virtual device that can be loaded and executed by a computer device, which may include the aforementioned control equipment. Figure 3 The device may include a prediction module, an evaluation module, a dual-channel MPC optimization module, a strategy optimization module, and a voltage control module, which, when used to execute the above control method, can... The prediction module inputs the time-series data of node voltages and topology data of the distribution network into the prediction model to obtain the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales.

[0079] The evaluation module assesses the error of the node voltage prediction based on the predicted node voltage and the actual node voltage.

[0080] The dual-channel MPC optimization module performs dual-channel MPC optimization based on error assessment results and node voltage over-limit risk indicators to obtain a set of node voltage control strategies. In the dual-channel MPC optimization, the main channel and the disturbance rejection channel generate a basic control strategy and a robust compensation strategy, respectively. The node voltage control strategy is a weighted fusion of the basic control strategy and the corresponding robust compensation strategy.

[0081] The strategy optimization module uses the Red-beaked Blue Magpie optimization algorithm to obtain the optimal node voltage control strategy from the set of node voltage control strategies.

[0082] The voltage control module performs distribution network node voltage control according to the optimal node voltage control strategy; in the dual-channel MPC optimization and distribution network node voltage control, all distributed energy resources are equivalent to a quantized virtual impedance network.

[0083] The aforementioned device employs a spatiotemporal and physical hybrid approach for precise multi-scale voltage situation perception, predicting future node voltages and corresponding multi-scale node voltage exceedance risk indicators. It uses a main channel to acquire basic control strategies and an anti-disturbance channel to acquire corresponding robust compensation strategies. Combined with the Red-beaked Blue Magpie optimization algorithm, it improves the efficiency of solving high-dimensional problems. In dual-channel MPC optimization and distribution network node voltage control, it treats all distributed energy sources as quantized virtual impedance networks, breaking through the bottleneck of optimization dimensions. Ultimately, it can form a full-chain control system of "global prediction - hierarchical decision-making - edge execution", which is suitable for voltage stability control in scenarios with a high proportion of distributed energy access.

[0084] This application also relates to a computer-readable storage medium that stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform a power distribution network voltage control method.

[0085] This application also relates to a computer device including one or more processors and one or more memories, wherein one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing a power distribution network voltage control method.

[0086] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.

[0087] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] 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.

[0089] 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.

[0090] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A voltage control method for a power distribution network, characterized in that, include: Input the time series data of node voltage and topology data of the distribution network into the prediction model to obtain the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales. Based on the predicted node voltage values ​​and the actual node voltage values, an error assessment of the predicted node voltage values ​​is performed. Based on the error assessment results and the node voltage over-limit risk index, dual-channel MPC optimization is performed to obtain a set of node voltage control strategies. In the dual-channel MPC optimization, the main channel and the disturbance rejection channel generate the basic control strategy and the robust compensation strategy, respectively. The node voltage control strategy is the strategy obtained by weighted fusion of the basic control strategy and the corresponding robust compensation strategy. The Red-billed Blue Magpie optimization algorithm is used to obtain the optimal node voltage control strategy from the set of node voltage control strategies. In the Red-billed Blue Magpie optimization algorithm, the difference between the objective function of the current solution and the global optimal solution is used to simulate the energy barrier in quantum mechanics, and the probability of an individual in the optimization solution passing through the energy barrier is calculated. Based on the optimal node voltage control strategy, the distribution network node voltage is controlled; in the dual-channel MPC optimization and distribution network node voltage control, all distributed energy resources are equivalent to a quantized virtual impedance network.

2. The method according to claim 1, characterized in that, The process by which the prediction model processes node voltage time-series data and topology data includes: Based on the topology data, a dynamic adjacency matrix of the distribution network is constructed; where the elements in the dynamic adjacency matrix are the dynamic adjacency weights between nodes, and the dynamic adjacency weights between nodes represent the electrical association strength. Wavelet scattering feature extraction is performed on node voltage time series data to obtain node voltage features at multiple scales; Based on the dynamic adjacency matrix and node voltage characteristics, a physical constraint attention mechanism is used to obtain the node voltage characteristics of the fused topology data. By inputting the node voltage characteristics of the fused topology data into the encoder, the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators at multiple scales are obtained.

3. The method according to claim 2, characterized in that, The formula for dynamic adjacency weight is: ； In the formula, Let V be the dynamic adjacency weight between the i-th node and the j-th node at time t, σ be the normalization function, and V be the dynamic adjacency weight between the i-th node and the j-th node at time t. i and Q i Let V be the voltage and reactive power of the i-th node. j and P j Let the voltage and active power of the j-th node be denoted as . Let be the sensitivity of the voltage at node i to the active power at node j. Let |z| be the sensitivity of the reactive power at node i to the voltage at node j. ij || represents the impedance magnitude of the line between the i-th node and the j-th node, I ij Let be the current in the line between the i-th node and the j-th node; The node voltage characteristics at multiple scales include low-frequency trend components, mid-frequency fluctuation components, and high-frequency abrupt change components; the wavelet scattering feature extraction formula is: ； In the formula, W V This is the multi-scale node voltage feature vector extracted by wavelet scattering. These are the wavelet basis functions used to extract low-frequency trend components. The Mexican hat wavelet function is used to extract frequency fluctuation components. Here, represents the Haar wavelet function used for high-frequency abrupt changes, * represents convolution, and V represents the node voltage vector.

4. The method according to claim 1, characterized in that, The optimization of the main channel is based on the node voltage prediction value as the scenario, and aims to minimize the sum of voltage tracking deviation penalty, control cost, and control increment smoothing penalty; the objective function for the main channel optimization is: ； In the formula, V t Let V be the node voltage vector at time t. ref W is the node voltage reference vector. v The voltage tracking weight matrix is ​​used to reflect the importance of the voltage at each node. The weighted square norm is a dimensional indicator used to quantify voltage deviation or control costs. The weighted square norm, u, is used to measure the cost of control. t and u t-1 Let be the control variables at time t and time t-1, respectively. The control variables are the flexible regulation resources of the distribution network. R is the control cost weight matrix for suppressing abrupt changes in the control variables. ρ is the smoothing coefficient. N is the optimized prediction time domain length. u is the optimization variable. Represents the norm.

5. The method according to claim 1, characterized in that, The optimization of the disturbance rejection channel, considering the uncertainties of distributed energy output and load fluctuations, generates a robust compensation strategy corresponding to the basic control strategy; the objective function for the disturbance rejection channel optimization is: ； ； In the formula, Let P be the set of distributions of uncertainty parameters describing the statistical characteristics of the prediction error in the error evaluation results. The probability distribution of uncertainty Let P be the mathematical expectation operator. Let P be the expected value of the perturbation. Let be the prediction deviation of the node voltage at time t, Q be the node voltage deviation penalty matrix, and the superscript T denotes transpose. Let be the perturbation vector. To predict the mean of the disturbance, For Wasserstein distance, To perturb the upper bound of expectation, Where is the Wasserstein radius, and N is the optimized prediction time domain length.

6. The method according to claim 1, characterized in that, During weighted fusion, the weights of the basic control strategy are the main channel weights, and the weights of the robust compensation strategy are the disturbance rejection channel weights. The formula for determining the weights is as follows: ； ； ； In the formula, Let E be the error intensity based on the error assessment results at time t. t and node voltage over-limit risk index R t Determined initial weights for the main channel Baseline value, Control E t and R t Sensitivity coefficient, Let be the weight of the main channel at time t. Let be the weight of the anti-interference channel at time t, where clip is the clip function, and σ is the normalization function. These are the lower and upper limits of the main channel weight, respectively.

7. The method according to claim 1, characterized in that, The process of obtaining a quantized virtual impedance network includes: Distributed resources are represented as quantized virtual impedances, and a game equilibrium model between distributed energy resources and the power grid is constructed. The ADMM algorithm accelerated by Nesterov is used to iteratively solve the game equilibrium model between distributed energy resources and the power grid to obtain the quantized virtual impedance network. In the iterative solution, the weights of the quantized virtual impedances are updated based on sensitivity analysis. The quantized virtual impedances are used to describe the equivalent impedance characteristics of distributed energy resources to the power grid.

8. The method according to claim 7, characterized in that, Distributed resources can be equivalently represented as quantized virtual impedance, as shown in the formula: ； In the formula, Let n be the quantized virtual impedance, and n be the total number of distributed resources participating in the aggregation. Let the ground-state impedance be the k-th distributed resource. To entangle the weighting coefficients, To reflect the phase angle of the dynamic response delay of the k-th distributed resource, This refers to the device operating status characteristic information of the k-th distributed resource. It is a tensor product operator.

9. The method according to claim 7, characterized in that, The game equilibrium model between distributed energy resources and the power grid is as follows: ； In the formula, Z v V is a virtual impedance matrix. ref Let V be the node voltage reference vector, and λ be the regularization coefficient balancing node voltage tracking accuracy and voltage regulation cost. The norm is indicated by the superscript T, which indicates transpose. Let V be the trace norm of the virtual impedance, ΔP be the power regulation vector of the distributed energy source, μ be the tradeoff coefficient, and V be the trace norm of the virtual impedance. nom Let P be the local nominal voltage of the distributed energy source, and let P be the actual power vector of the node determined by the power flow equations of the distribution network. Virtual impedance and node voltage deviation The real part of the mapping relation, This is a Gaussian noise term.

10. A power distribution network voltage control device, characterized in that, include: The prediction module inputs the time-series data of node voltage and topology data of the distribution network into the prediction model to obtain the future node voltage prediction values ​​of each node in the distribution network and the corresponding node voltage over-limit risk indicators under multiple scales. The evaluation module assesses the error of the node voltage prediction based on the predicted node voltage and the actual node voltage. The dual-channel MPC optimization module performs dual-channel MPC optimization based on error assessment results and node voltage over-limit risk indicators to obtain a set of node voltage control strategies. In the dual-channel MPC optimization, the main channel and the disturbance rejection channel generate basic control strategies and robust compensation strategies, respectively. The node voltage control strategy is a strategy obtained by weighted fusion of the basic control strategy and the corresponding robust compensation strategy. The strategy optimization module uses the Red-billed Blue Magpie optimization algorithm to obtain the optimal node voltage control strategy from the set of node voltage control strategies. In the Red-billed Blue Magpie optimization algorithm, the difference between the objective function of the current solution and the global optimal solution is used to simulate the energy barrier in quantum mechanics, and the probability of an individual in the optimization solution passing through the energy barrier is optimized. The voltage control module performs distribution network node voltage control according to the optimal node voltage control strategy; in the dual-channel MPC optimization and distribution network node voltage control, all distributed energy resources are equivalent to a quantized virtual impedance network.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of any one of claims 1 to 9.

12. A computer device, characterized in that, include: One or more processors and one or more memories, one or more programs stored in one or more memories and configured to be executed by one or more processors, the one or more programs including instructions for performing the method of any one of claims 1 to 9.

Images