Method and system for evaluating photovoltaic bearing capacity of power distribution network building

Abstract

The invention discloses a power distribution network building photovoltaic bearing capacity assessment method and system, and the method comprises the steps: obtaining predicted meteorological covariable data of a power distribution network connected with building photovoltaic, inputting the predicted meteorological covariable data into a trained condition diffusion model, and outputting photovoltaic output distribution. The method comprises the following steps: defining Renyi divergence based on photovoltaic output distribution and real probability distribution corresponding to the photovoltaic output distribution, establishing a dynamic distribution robust fuzzy set constrained by a dynamic radius based on the Renyi divergence, and establishing a two-stage bearing capacity distribution robust optimization model; and solving the two-stage bearing capacity distribution robust optimization model by using a column and constraint generation algorithm based on the dynamic distribution robust fuzzy set to obtain an optimization result, and obtaining the maximum photovoltaic bearing capacity based on the optimization result, thereby effectively improving the accuracy, flexibility and robustness of power distribution network photovoltaic bearing capacity evaluation.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic carrying capacity assessment technology, and in particular to a method and system for assessing the photovoltaic carrying capacity of building distribution networks. Background Technology

[0002] Building-integrated photovoltaics (BIPV), as an important form of distributed photovoltaic (PV) power generation, is rapidly gaining popularity in urban power distribution networks. BIPV refers to photovoltaic power generation systems installed on building rooftops, facades, and other components. It offers advantages such as local energy consumption, reduced transmission losses, and improved land use efficiency, making it a key technology for achieving building energy self-sufficiency and promoting zero-carbon building development. BIPV has become a major component of new power sources for power distribution networks. However, the large-scale integration of BIPV has brought new challenges to power distribution networks: on the one hand, BIPV power output exhibits significant randomness, fluctuation, and intermittency, influenced by various factors such as weather conditions, building orientation, and obstructions, making its uncertainty characteristics more complex than those of ground-mounted centralized PV power plants; on the other hand, BIPV typically connects to low-voltage power distribution networks, with dispersed connection points and large capacity differences, placing higher demands on the voltage quality, line current-carrying capacity, and transformer capacity of the power distribution network.

[0003] Against this backdrop, assessing the building-integrated photovoltaic (BIPV) carrying capacity of distribution networks has become a core issue in power system planning and operation. BIPV carrying capacity refers to the maximum installed capacity of building-integrated photovoltaics (BIPV) that a distribution network can accommodate under operational constraints such as voltage deviation, line current, transformer capacity, and power balance. Accurately assessing BIPV carrying capacity is crucial for guiding the rational layout of BIPV, optimizing distribution network planning and renovation schemes, and ensuring the safe and economical operation of the power grid. Due to the high degree of uncertainty in BIPV output, how to fully consider this uncertainty in the assessment, ensuring both the robustness and security of the assessment results while avoiding resource waste due to excessive conservatism, is a pressing technical challenge that needs to be addressed.

[0004] Existing methods for assessing the carrying capacity of building-integrated photovoltaics (BIPV) systems in power distribution networks are mainly divided into two categories: deterministic methods and uncertainty optimization methods. Deterministic methods typically select photovoltaic output data from several typical days (e.g., sunny, cloudy, rainy days) or typical times, and determine the maximum connectable capacity of each node without violating operational constraints through power flow calculations. These methods are computationally simple and easy to implement in engineering, but they are essentially based on sensitivity analysis of limited scenarios and cannot fully reflect the random fluctuation characteristics of building-integrated photovoltaic output. Because the selection of typical scenarios is subjective and cannot cover all possible operating conditions, the assessment results of deterministic methods often lack reliability guarantees and may underestimate the operational risks under extreme scenarios.

[0005] Uncertainty optimization methods address the uncertainty of building-integrated photovoltaic (BIPV) output from a probabilistic or robust optimization perspective, primarily including three categories: stochastic optimization, traditional robust optimization, and degenerate robust optimization. Stochastic optimization methods assume that BIPV output follows a known probability distribution, determining carrying capacity by minimizing expected costs or maximizing expected benefits. However, this method requires high accuracy in its distribution assumptions. In practical applications, due to the complex influence of the building microenvironment (such as shading from surrounding buildings and reflected light), the output distribution of BIPV is often difficult to accurately describe using standard parameter distributions, limiting the reliability of stochastic optimization results. Traditional robust optimization methods, on the other hand, do not rely on probability distribution assumptions, only requiring uncertain variables to vary within a given deterministic set (such as box uncertainty or ellipsoidal uncertainty), optimizing decisions under worst-case conditions. While traditional robust optimization can guarantee the robustness of the solution, it often leads to overly conservative assessments and underestimation of carrying capacity due to the lack of utilization of historical statistical information, resulting in insufficient utilization of BIPV resources.

[0006] The Kullback-Leibler (KL) divergence optimization method has attracted widespread attention in recent years. This method lies between stochastic optimization and traditional robust optimization. By constructing a fuzzy set containing the true distribution, it optimizes decisions in the worst-case scenario. It utilizes statistical information from historical data while avoiding strong assumptions about the precise distribution, theoretically achieving a good balance between robustness and economy. Within the Kullback-Leibler optimization framework, existing technologies generally use fuzzy sets based on divergence metrics to characterize the uncertainty distribution of building-integrated photovoltaic (BIPV) output. Among these, the Kullback-Leibler (KL) divergence is one of the most commonly used measurement tools. For example, existing schemes construct a fuzzy set centered on the historical empirical distribution and with the KL divergence as the radius. Within this fuzzy set, they search for the probability distribution that minimizes the expected value of the objective function, and then evaluate the carrying capacity under this worst-case distribution.

[0007] The basic framework of the existing technology is as follows: First, a power flow constraint model for the distribution network is established. This typically employs a linearized DistFlow model (a mathematical model used to describe power flow in the distribution network) or a DC power flow model to simplify calculations. These models transform the original non-convex nonlinear optimization problem into a linear programming or quadratic programming problem by ignoring the impact of active power on voltage or linearizing the nonlinear power flow equations. Then, a scenario generation method (Monte Carlo sampling) is used to obtain building-integrated photovoltaic (BIPV) output samples, constructing an empirical probability distribution. Next, a static fuzzy set is constructed centered on the empirical distribution and combined with KL divergence. This fuzzy set contains all probability distributions whose KL divergence with the empirical distribution does not exceed a given radius. Finally, a robust optimization model is established, aiming to maximize the total BIPV capacity of the distribution network. Constraints include worst-case power flow constraints, voltage constraints, and line capacity constraints, ultimately yielding a robust estimate of the BIPV carrying capacity of the distribution network.

[0008] However, existing technologies have the following shortcomings: 1) Regarding robustness metrics: KL divergence, as a special case of the f-divergence family, is overly sensitive to changes in the distribution tails, easily leading to overly conservative evaluation results. Mathematically, KL divergence requires absolute continuity between the candidate distribution and the reference distribution; that is, the support set of the candidate distribution must be contained within the support set of the reference distribution. Otherwise, the KL divergence value becomes infinite. This strict requirement means that when observational data is limited, the constructed fuzzy set may be too small to cover the reasonable range of variation in the true distribution; or, to avoid the fuzzy set being too small, artificially increasing the radius parameter may make the fuzzy set too large, including unreasonable extreme distributions, leading to overly conservative evaluation results. Especially when photovoltaic output is affected by extreme weather and tail events occur, KL divergence will react significantly to small changes in the probability of the tails, causing the optimization model to overemphasize low-probability extreme scenarios while neglecting the economic efficiency of high-probability normal operation scenarios. Furthermore, KL divergence cannot flexibly adjust the degree of attention to the distribution center and tails; its logarithmic form determines that its sensitivity to the entire distribution is fixed, lacking a parameterized adjustment mechanism to balance robustness and economy. This inflexibility limits decision-makers' ability to adjust the level of conservatism based on risk appetite and operational needs in practical applications, affecting the accuracy and practicality of carrying capacity assessments. Therefore, it is necessary to study more practical sub-Bruker optimization methods.

[0009] 2) Regarding the construction of fuzzy sets for distributed photovoltaic (PV) power: Existing methods using static fuzzy sets have significant limitations. Static fuzzy sets use a uniform radius parameter for all time periods and weather conditions. The theoretical basis for this approach is the assumption that the uncertainty of PV power output is statistically stationary and does not change with external conditions. However, this assumption contradicts the physical mechanism of PV power generation. In reality, the uncertainty level of PV power output is closely related to meteorological conditions: under clear weather conditions, with less cloud cover, PV power output is relatively stable with less fluctuation; while under cloudy or rainy weather conditions, rapid cloud movement leads to drastic fluctuations in irradiance, significantly increasing the uncertainty of PV power output. The statistical distribution characteristics of PV power output also differ significantly across seasons due to variations in solar altitude angle, atmospheric transparency, and cloud characteristics. Furthermore, during sunrise and sunset, PV power output is in an upward or downward phase, and its rate of change and uncertainty differ fundamentally from those at noon. Static fuzzy sets cannot capture this uncertainty characteristic that dynamically changes with meteorological covariates (such as irradiance, cloud cover, temperature, humidity, and wind speed). This leads to a situation where, in practical applications, static fuzzy sets can be overly conservative under certain conditions (such as sunny days), overestimating the degree of uncertainty and resulting in an underestimation of the carrying capacity, leading to insufficient utilization of photovoltaic resources. Conversely, under other conditions (such as extreme weather), they may be insufficiently robust, underestimating the degree of uncertainty and causing safety hazards in the assessment results. This "one-size-fits-all" approach neither accurately reflects the true distribution of uncertainty nor provides refined decision support for different operating scenarios. Therefore, it is necessary to study methods for constructing dynamic radii of fuzzy sets that vary with the environment.

[0010] 3) In terms of scenario generation for distributed spectral optimization, traditional Monte Carlo methods rely on parameterized assumptions about the photovoltaic (PV) output distribution. These assumptions are often based on mathematical simplicity rather than physical reality. For example, assuming that PV output follows a normal or Beta distribution facilitates sampling calculations. However, actual PV output is influenced by a combination of meteorological factors, and its distribution often exhibits complex multimodal characteristics, asymmetry, and heavy tails, making it difficult to accurately characterize with a single parameter distribution. Even with mixed distributions or piecewise modeling, the accuracy of parameter estimation still highly depends on the sufficiency of historical data. In practical engineering, the number of samples under specific meteorological conditions is often limited, leading to significant parameter estimation bias. Furthermore, traditional Monte Carlo methods suffer from the curse of dimensionality in high-dimensional scenarios (such as joint distributions with multiple nodes and time periods), resulting in slow convergence speeds, requiring the generation of a large number of samples to ensure estimation accuracy, and thus low computational efficiency. Although Generative Adversarial Networks (GANs), as a deep generative model, can learn complex high-dimensional data distributions without parameterized assumptions, GANs have an inherent pattern collapse problem during training. This means that the generator tends to generate some high-quality samples while ignoring other patterns in the data distribution, resulting in insufficient diversity of generated samples and failing to fully cover all possibilities of the real distribution.

[0011] The aforementioned three shortcomings are interconnected and collectively constrain the performance of existing methods for assessing the photovoltaic carrying capacity of distribution networks. The inflexibility of the measurement tools limits the accurate characterization of uncertainty by fuzzy sets; the coarseness of static fuzzy sets prevents the full utilization of meteorological information to improve assessment accuracy; and the deficiencies in scene generation methods fundamentally affect the quality and representativeness of the empirical distribution. Therefore, there is an urgent need to develop a more precise and flexible measurement tool, a dynamically adjustable fuzzy set construction method, and a more accurate and reliable scene generation technology to systematically improve the accuracy, flexibility, and robustness of photovoltaic carrying capacity assessment for distribution networks. Summary of the Invention

[0012] The technical problem to be solved by the present invention is to provide a method and system for assessing the photovoltaic carrying capacity of a power distribution network, which can effectively improve the accuracy, flexibility and robustness of the assessment of the photovoltaic carrying capacity of the power distribution network.

[0013] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for assessing the photovoltaic carrying capacity of a power distribution network building includes the following steps: Obtain forecast meteorological covariate data for the power distribution network connected to building-integrated photovoltaics; The predicted meteorological covariate data are input into the trained conditional diffusion model, which outputs the photovoltaic power output distribution. The Rényi divergence is defined based on the photovoltaic power output distribution and its corresponding true probability distribution. A dynamic sub-Brussels bar fuzzy set constrained by dynamic radius is established based on the Rényi divergence; Establish a two-stage load-bearing capacity sub-Bluerge optimization model for the aforementioned distribution network; Based on the dynamic sub-bar fuzzy set, the column and constraint generation algorithm is used to solve the two-stage bearing capacity sub-bar optimization model to obtain the optimization result; The maximum photovoltaic carrying capacity is obtained based on the optimization results.

[0014] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A building photovoltaic carrying capacity assessment system for a power distribution network includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps in the aforementioned building photovoltaic carrying capacity assessment method for a power distribution network.

[0015] The beneficial effects of this invention are as follows: Predicted meteorological covariate data of the distribution network connected to building-integrated photovoltaics (BIPV) are obtained; this data is input into a trained conditional diffusion model to output BIPV power output distribution; Rényi divergence is defined based on the BIPV power output distribution and its corresponding true probability distribution; a dynamic fractional Brussels bar fuzzy set constrained by dynamic radius is established based on the Rényi divergence; a two-stage carrying capacity fractional Brussels bar optimization model is established; the two-stage carrying capacity fractional Brussels bar optimization model is solved using a column and constraint generation algorithm based on the dynamic fractional Brussels bar fuzzy set to obtain the optimization result; and the maximum BIPV carrying capacity is obtained based on the optimization result. Rényi divergence is thus introduced as a degree... A fuzzy set is constructed using quantitative tools to avoid oversensitivity to the distribution tails and lack of flexible adjustment mechanisms. A conditional diffusion model is used as a scene generation tool, deeply integrating meteorological covariate information to accurately reproduce the differentiated distribution characteristics of building photovoltaic (PV) output under different meteorological modes, while ensuring the diversity and representativeness of the generated samples. Based on this, a dynamic sub-Bruker fuzzy set is constructed, which considers the essential characteristics of the dynamic change of PV output uncertainty with meteorological covariates. This can finely characterize the spatiotemporal variation law of PV output uncertainty. Finally, a two-stage carrying capacity sub-Bruker optimization model is established and solved, thereby effectively improving the accuracy, flexibility and robustness of the PV carrying capacity assessment of the distribution network. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for assessing the photovoltaic carrying capacity of a power distribution network building, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a photovoltaic carrying capacity assessment system for a power distribution network building, according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the training and generation process of the conditional diffusion model in a method for assessing the photovoltaic carrying capacity of a power distribution network building according to an embodiment of the present invention. Figure 4 This is a flowchart of the model solution in a method for assessing the photovoltaic carrying capacity of a power distribution network building according to an embodiment of the present invention; Figure 5 This is a framework diagram for assessing the carrying capacity of building-integrated photovoltaics in a power distribution network according to an embodiment of the present invention. Figure 6 This is a schematic diagram of a power distribution network calculation example in a method for assessing the photovoltaic carrying capacity of a power distribution network building according to an embodiment of the present invention. Detailed Implementation

[0017] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0018] Before detailing the embodiments of this application, some related concepts will first be explained: Conditional diffusion model: A type of diffusion model that introduces additional conditional information to guide the generation process, thereby generating data samples that meet specific requirements; Rényi divergence: A generalized tool for measuring the difference between two probability distributions, determined by parameters. (Often referred to as the order or exponent) controls its sensitivity; Column-and-Constraint Generation (C&CG) is an exact decomposition-based algorithm that breaks down the original problem into a main problem and subproblems. It is particularly suitable for handling nonconvexity caused by integer tracing variables or complex uncertain structures.

[0019] In existing technologies, the inflexibility of measurement tools limits the accurate characterization of uncertainty by fuzzy sets; the coarseness of static fuzzy sets makes it impossible to fully utilize meteorological information to improve assessment accuracy; and the defects of scene generation methods affect the quality and representativeness of experience distributions from the source.

[0020] To at least address the aforementioned problems, the present invention provides a method and system for assessing the carrying capacity of building-integrated photovoltaic (BIPV) power distribution networks. This method and system are applicable to distribution networks with BIPV systems that require capacity assessment. The specific implementation methods are described below: Please refer to Figure 1 One embodiment of the present invention is as follows: A method for assessing the photovoltaic carrying capacity of a power distribution network building includes the following steps: S1. Obtain historical operating data of the distribution network connected to building photovoltaics, wherein the historical operating data includes the historical output sequence of distributed photovoltaics at each node and its corresponding historical meteorological covariate data.

[0021] In one alternative implementation, the historical meteorological covariate data includes, but is not limited to, irradiance, temperature, humidity, and cloud cover.

[0022] S2. Normalize the historical operation data to obtain the processed historical operation data.

[0023] In this way, the historical operating data is normalized and then used to train the conditional diffusion model, resulting in a trained conditional diffusion model. This ensures the reliability of the model training and guarantees that the conditional diffusion model generates an accurate and reliable photovoltaic power output distribution.

[0024] S3. Use the processed historical running data to train the conditional diffusion model to obtain the trained conditional diffusion model.

[0025] Specifically, the forward noise-adding process is defined as follows: A diffusion process is constructed to gradually transform real photovoltaic power output data into Gaussian noise, which is described by the following Itô stochastic differential equation (Itô SDE): ; In the formula, express t Noisy photovoltaic power output data at any given time. This represents the preset noise variance scheduling function (Noise Schedule), which varies with time. t Monotonically increasing, w t This represents standard Brownian motion; as... t T Data distribution p t (x) approaches the standard normal distribution In this formula t The positive time step; Define the reverse denoising generation process: To recover the photovoltaic power output distribution that conforms to specific meteorological conditions from noise, a reverse time SDE corresponding to the forward process is defined: ; In the formula, Indicates inverse Brownian motion. The Conditional Score Function, which is the core objective of this invention, represents the score under given meteorological conditions, i.e., meteorological covariate data. Below, the gradient field of the logarithm of the data probability density indicates the direction of denoising. In this formula... t This is the reverse time step; Training a score network and its loss function: Build a deep neural network (i.e., the score network) fits the above conditional score function by minimizing the following weighted score matching loss function. Training network parameters : ; In the formula, Represents the transition probability The true score function; due to the transition probability It is Gaussian, and its true score function is calculated analytically, thus avoiding direct differentiation for an unknown true distribution.

[0026] Once the conditional diffusion model is trained, it can be updated quarterly or semi-annually with the latest data to adapt to seasonal changes and new power grid equipment.

[0027] S4. Obtain the forecast meteorological covariate data of the power distribution network connected to the building photovoltaic system.

[0028] S5. Input the predicted meteorological covariate data into the trained conditional diffusion model and output the photovoltaic power output distribution.

[0029] Specifically, by using a numerical solver (such as the Euler-Maruyama method) to solve the inverse SDE, the corresponding photovoltaic power output distribution can be generated based on the input forecast meteorological covariate data.

[0030] like Figure 3 As shown, Figure 3 The working principle of the conditional diffusion model used to generate photovoltaic power output distribution is demonstrated. Part (a) illustrates the basic logic of training the conditional diffusion model, that is, the data structure is destroyed by a forward noise addition process, and the intrinsic distribution of the data is learned by a reverse noise reduction process through iterative training. Part (b) shows the detailed process of the model, including the training stage of training the model using historical multi-source data, and the generation stage of generating the photovoltaic power output distribution under a specific scenario by numerically solving the inverse stochastic differential equation (SDE) after giving meteorological forecast conditions (i.e. forecast meteorological covariate data).

[0031] S6. Define the Rényi divergence based on the photovoltaic power output distribution and its corresponding true probability distribution, specifically including S61-S62: S61. Discretize the photovoltaic power output distribution into a reference probability distribution that includes multiple typical scenarios.

[0032] S62. Define the Rényi divergence between the reference probability distribution and its corresponding true probability distribution, specifically as follows: ; In the formula, Represents the reference probability distribution and its corresponding true probability distribution Rényi divergence between them , Indicates the order parameter, S This represents the total number of typical scenarios. p s Describes the true probability distribution at the th... s Probabilities in a typical scenario p 0,s Indicates the reference probability distribution at the th s The probability under a typical scenario, and satisfying Selecting the order parameter >1 to control sensitivity to extreme tail scenarios.

[0033] In this way, the continuous photovoltaic power output distribution generated by the conditional diffusion model is discretized into a reference probability distribution containing multiple typical scenarios. Rényi divergence is used as a statistical distance metric to quantify the difference between the true probability distribution and the reference probability distribution. Rényi divergence achieves flexible adjustment of the sensitivity of different parts of the distribution by introducing an order parameter.

[0034] S7. Based on the Rényi divergence, establish a dynamic sub-Bruker fuzzy set constrained by the dynamic radius, specifically as follows: ; In the formula, Represents a dynamic fuzzy set of Brussels bars. express S The space of nonnegative real numbers is used to limit the number of nonnegative real numbers. P The range of values ​​for indicates that P It is a by S A vector consisting of n real numbers, and this S All real numbers must be non-negative.

[0035] In this way, the dynamic fuzzy set is a set of all probability distributions that satisfy the Rényi divergence constraint. By dynamically adjusting the radius, this fuzzy set mathematically forms a confidence region centered on the reference probability distribution and covering the actual photovoltaic power output distribution. This ensures the adaptive coverage capability of the optimization model for photovoltaic power output fluctuations under different meteorological confidence levels, and can accurately capture the dynamic changes in building photovoltaic power output uncertainty with meteorological conditions.

[0036] Among these, the reliability of the currently input meteorological covariate data is evaluated using a conditional diffusion model and a fractional network trained based on historical data. Since the training objective of the conditional diffusion model is to minimize the fractional matching loss, its goal is to train the neural network to approximate the logarithmic gradient (i.e., the fractional function) of the data distribution after Gaussian noise perturbation. For given meteorological covariate data, its loss function is mathematically defined as the weighted Euclidean distance between the model-estimated fractional field and the actual perturbation distribution fractional field. Therefore, before S7, it also includes: The loss function for the predicted meteorological covariate data is established as follows: ; In the formula, Represents forecast weather covariate data loss function, This represents a positive weighting function. This represents the set of all learnable weights and bias parameters in a neural network. Indicates by parameters The fitted conditional fractional function, Indicates the diffusion time t The state after adding noise, Represents the true fractional function, This represents the transfer nucleus in the forward diffusion process. This represents a sample of actual photovoltaic power output; the loss value statistically reflects the model's ability to reconstruct noisy data into true data, when the input... To train a fractional network to accurately predict gradient directions during common, stable weather conditions, The model approaches zero; however, when rare or extreme weather conditions are input (i.e., out-of-distribution samples, OOD), the model is difficult to reconstruct accurately, resulting in a significant increase in estimation error and a corresponding increase in loss.

[0037] The dynamic radius is calculated based on the aforementioned loss function, specifically as follows: ; In the formula, Indicates data of covariates related to weather forecasting The dynamic radius changes in real time. r base This represents the preset basic safety radius constant. This represents the sensitivity adjustment coefficient.

[0038] In this way, for a given meteorological condition Its loss function is mathematically defined as the weighted Euclidean distance between the model-estimated fractional field and the true perturbation distribution fractional field. The loss value statistically reflects the model's ability to reconstruct noisy data into true data. When the input is predicted meteorological covariate data... To train a fractional network to accurately predict gradient directions during common, stable weather conditions, The loss value approaches zero. However, when rare or extreme weather conditions (i.e., out-of-distribution samples, OOD) are input, the model is difficult to reconstruct accurately, resulting in a significant increase in estimation error and corresponding increase in loss. This loss value is used to construct a dynamic radius calculation model that changes dynamically with the environment. This model consists of two parts: a basic safety radius and an environmental adaptation increment. This dynamic adjustment mechanism effectively solves the problem of the traditional fixed radius method in balancing conservatism and robustness.

[0039] By employing a dynamic radius mechanism based on fractional matching loss, when weather forecasts indicate a stable environment and high model confidence, Smaller, dynamic radius Automatic shrinkage makes the optimization results closer to the deterministic solution, thereby improving economic efficiency; when extreme weather causes increased uncertainty in model predictions, Increase, driving dynamic radius Expansion, thereby expanding the fuzzy set The coverage range ensures the operational safety of the optimization results under extreme scenarios. This dynamic adjustment mechanism effectively solves the problem of the traditional fixed radius method's difficulty in balancing conservatism and robustness.

[0040] S8. Establish a two-stage load-bearing capacity sub-Brow bar optimization model for the aforementioned distribution network, specifically including S81-S82: S81. Establish a three-level objective function with the goal of minimizing the sum of the investment cost of the static var compensator, the capacitor switching penalty cost, and the expected operating penalty cost under the worst-case distribution. Specifically: ; In the formula, n and Q SVC For the first level of decision variables, n Indicates the number of capacitor banks put into operation. Q SVC This indicates the planned capacity of the static var compensator. This represents the set of nodes equipped with static var compensators (SVCs). This represents the unit capacity investment cost of a static var compensator. Represents a node j The planned static var compensator capacity is as follows: T Indicates a time period. This represents the set of nodes where capacitor banks are installed. c sw This represents the penalty coefficient for a single capacitor switching operation, used to prevent unnecessary redundant switching. express t Time Node j The number of capacitor banks installed at the location express t -1 time node j The number of capacitor banks installed at the location P For the second level of decision variables, y s These are the third-level decision variables, containing all voltages and currents at the nodes. Indicates network loss weight. express t Time of the first s Total network loss in a typical scenario Indicates the weight of the light-abandoning penalty. This represents the set of nodes equipped with photovoltaic panels. express t Time Node j In the s Wasted light in a typical scenario.

[0041] The model is logically divided into three layers: the first layer is the decision-making process for capacitor switching and static var compensator installation before the occurrence of uncertainty; the second layer is to find the worst-case photovoltaic output probability distribution in the constructed dynamic fuzzy set; and the third layer is to minimize the operating cost by adjusting the grid power flow and curtailment under given capacitor state and photovoltaic scenario.

[0042] S82. Establish first-stage constraints and second-stage constraints corresponding to the three-layer objective function. The first-stage constraints include explicit capacity and operation constraints of the grouped switching capacitor banks and capacity planning constraints of the static var compensator. The second-stage constraints include two-stage connection constraints, power flow model of second-order cone branch of distribution network, system safe operation and feeder capacity constraints, and building photovoltaic constraints.

[0043] In this way, a three-level objective function is established with the goal of minimizing the sum of the investment cost of the static var compensator, the capacitor switching penalty cost, and the expected operating penalty cost under the worst-case distribution. The first-stage constraints and the second-stage constraints corresponding to the three-level objective function are also established. Based on this, a three-level two-stage sub-Bruker optimization model based on the min-max-min structure is established. The reactive power optimization in the first stage provides voltage support for photovoltaic absorption in the second stage, thereby maximizing the photovoltaic absorption capacity under the worst-case probability distribution.

[0044] The first phase aims to determine the planning and dispatching strategies for the distribution network before uncertainties occur. This phase includes two types of decision variables: one is the switching state variable of the grouped capacitor banks (CBs). Secondly, the planned capacity variable of the Static Var Compensator (SVC). .

[0045] Decision variables Not only must integer constraints and upper limit constraints on installed capacity be met, but the maximum number of switching operations per day must also be strictly observed to prevent frequent operations from shortening the equipment life. Therefore, the explicit capacity and operating constraints of the grouped switching capacitor banks include group status constraints, capacitor bank capacity constraints, and daily switching frequency constraints.

[0046] The group number state constraint is: ; In the formula, Represents a node j The maximum number of capacitor banks that can be switched on. Represents the set of integers.

[0047] The capacity constraint of the capacitor bank is: ; In the formula, Represents a node j At the capacitor bank t The reactive power constantly injected into the power grid, This indicates the capacitance of a single capacitor bank.

[0048] The daily number of cuts is constrained as follows: ; In the formula, This represents the sum of absolute values ​​of capacitor switching state changes, with a preset threshold that limits the total number of actions throughout the day.

[0049] For the continuously variable transmission (SVC) device, the first stage treats it as a planning object, determines its maximum installed capacity, and thus limits the dynamic adjustment range in the third stage. Therefore, the capacity planning constraint of the static var compensator is: ; In the formula, Represents a node j The maximum capacity of the static var compensator to be configured in the area.

[0050] Through the above constraints, the model not only formulates a day-ahead scheduling table for CB that satisfies the explicit capacity definition in the first stage, but also establishes the capacity planning boundary of SVC. These variables will be passed as fixed parameters to the second stage. This will directly lock the CB output in each scenario of the second phase, and This will serve as the upper limit boundary for the second-stage SVC dynamic adjustment.

[0051] The second-stage constraints aim to describe the physical boundaries of the distribution network to maintain power balance and safe operation through continuous adjustment, given that the planning decisions in the first stage have been determined and the specific photovoltaic output scenarios have been revealed. All decision variables in this stage are continuous variables.

[0052] Although the first phase of decision Regardless of the scenario, it physically determines the upper limit of reactive power injection in all scenarios during the second phase. Therefore, for any scenario... s , t Time Node j reactive power injected by CB Since the number of groups determined in the first stage is strictly locked, the two-stage relationship constraint connecting the first-stage planning solution and the second-stage running variables is as follows: ; ; In the formula, Representing a scene s , t Time Node j The planned capacity of static var compensators; This constraint indicates that regardless of fluctuations in photovoltaic output, the capacitor bank at any given time... t The switching state is fixed and cannot be specially adjusted for specific scenarios.

[0053] To ensure the convexity of the innermost optimization problem and the efficiency of solving subproblems in the C&CG algorithm, a second-order cone relaxation technique is used to describe the physical power flow of the distribution network. A second-order cone relaxation power flow model in polar coordinates is used to describe the physical characteristics of the distribution network. Therefore, the second-order cone branch power flow model of the distribution network includes node power balance constraints, branch power flow definition constraints, and second-order cone relaxation constraints.

[0054] The node power balance constraint is: ; ; ; ; In the formula, express t Time of the first s Nodes in a typical scenario i Total injected active power at the location, express t Time of the first s Nodes in a typical scenario i Total injected reactive power at the location, express t Time of the first s Nodes in a typical scenario i Active load at the location, express t Time of the first s Nodes in a typical scenario i reactive load at the location, P ij,t,s Let (i,j) represent the active power flow of branch (i,j). Q ij,t,s This represents the reactive power flow of branch (i,j). Represents nodes i The set of adjacent nodes, express t Time of the first s Active power input at the substation gate in a typical scenario express t Time of the firsts In a typical scenario, the reactive power input at the substation gate... express t Time of the first s Nodes in a typical scenario i The active power output of the photovoltaic inverter at the location, express t Time of the first s Nodes in a typical scenario i The reactive power output of the photovoltaic inverter at the location, express t Time of the first s Nodes in a typical scenario i The reactive power injected into the capacitor bank at that location, Representing a scene s , t Time Node i The planned configuration of static var compensators.

[0055] The branch power flow definition constraint is as follows: For branch roads l =(i,j), and its head and tail power are defined as follows: ; ; ; ; ; ; ; ; In the formula, g ij Indicates the conductance of branch (i,j). b ij Indicates the susceptance of branch (i,j). u i,t,s Indicates the first auxiliary variable. R l,t,s Indicates a branch l The product of the real parts of the voltage phasors at both ends. T l,t,s Indicates a branch l The imaginary product of the voltage phasors at both ends P ji,t,s Let (j,i) represent the active power flow of branch (j,i). Q ji,t,s This represents the reactive power flow of branch (j,i). u j,t,s Indicates the second auxiliary variable.V i,t,s express t Time of the first s Nodes in a typical scenario i voltage amplitude, V j,t,s express t Time of the first s Nodes in a typical scenario j voltage amplitude, l,t,s express t Time of the first s Branch paths in a typical scenario l The voltage phase angle difference at both ends.

[0056] For auxiliary variables u i,t,s , u j,t,s , R l,t,s as well as T l,t,s A second-order cone relaxation is performed to ensure the convexity of the model; therefore, the second-order cone relaxation constraint is: .

[0057] In one alternative implementation, the method further includes: in order to optimize network loss, the system needs to be explicitly defined in... t Typical scenarios s Total active power loss In the branch injection model, the loss of a single branch is equal to the sum of the active power flowing into the beginning and end of that branch. Therefore, the total network loss is: ; In the formula, E It represents the set of all branches in the power grid.

[0058] The system's safe operation and feeder capacity constraints include voltage safety constraints and feeder capacity constraints.

[0059] The voltage safety constraint is: ; In the formula, Represents a node j Minimum permissible voltage amplitude Represents a node j The maximum permissible voltage amplitude.

[0060] The feeder capacity constraint requires that the apparent power of the branch circuits meet the thermal stability limit. S max The quadratic constraint, after being linearized by a second-order cone relaxation, becomes: .

[0061] The building photovoltaic constraints are: The installed capacity of photovoltaic (PV) systems must be within the allowable minimum and maximum planning range, and the actual output of PV systems must not exceed the product of their installed capacity and the power conversion factor at the current moment. Therefore, there are upper limits on PV capacity and output. ; ; In the formula, Indicates the first s Nodes in a typical scenario j Photovoltaic installation capacity at the location, This indicates the lower limit of photovoltaic installation capacity. Indicates the upper limit of photovoltaic installation capacity. express t Time of the first s Nodes in a typical scenario j The active power output of the photovoltaic inverter at the location, express t Photovoltaic power conversion coefficient at any given time; The total power generation of photovoltaic (PV) systems during the dispatch cycle must meet a certain minimum output level to ensure the effective utilization of PV resources. ; In the formula, x PV Indicates the minimum output level; The reactive power output of a photovoltaic inverter is limited by the minimum / maximum power factor angle, thus subject to power factor constraints. ; In the formula, This represents the angle indicating the minimum power factor. Indicates the maximum power factor angle. express t Time of the first s Nodes in a typical scenario j The reactive power output of the photovoltaic inverter at the location; definition Scene generated for diffusion model s Maximum available photovoltaic power output, actual grid-connected power Continuous downward adjustments are allowed, therefore the actual photovoltaic output is constrained as follows: .

[0062] S9. Based on the dynamic sub-bar fuzzy set, the column and constraint generation algorithm is used to solve the two-stage bearing capacity sub-bar optimization model to obtain the optimization results, specifically including S91-S95: S91. The two-stage load-bearing capacity split-bar optimization model is decomposed into a main problem and sub-problems. The main problem serves as the planning layer, optimizing the capacitor switching strategy and static var compensator planning capacity in the first stage, and using auxiliary variables to approximate the worst-case operating cost in the second stage. Its goal is to provide a lower bound solution that satisfies the constraints of the currently known worst-case scenario set. The sub-problems serve as the attack layer, with the goal of finding a new worst-case probability distribution in the dynamic split-bar fuzzy set, given the first-stage decision of the main problem, so as to maximize the expected operating cost of the system and provide an upper bound and a new cutting plane constraint for the main problem.

[0063] Specifically, the main problem is constructed, which aims to minimize the investment and operational costs of the first stage and the estimated costs of the second stage, while satisfying the constraints of the first stage. Its mathematical model is as follows: ; In the formula, As the third auxiliary variable, it represents the lower bound of the worst-case expected operating cost in the second stage. k Indicates the number of iterations. Indicates the first k The worst-case probability distribution returned by the subproblem in the next iteration. x This represents the vector of decision variables in the first stage, specifically containing the number of capacitor switching groups at each node at all time points. SVC planning capacity of each node That is, x={ , }, Q s ( x () indicates that a corresponding scenario has been introduced. s The operating cost function following the second-stage variables (i.e., the second-stage constraints and cost calculation logic); The subproblem is further broken down into two independent steps for solving: Step a: Fix the first-stage decision variables of the main problem propagation. x * Under the premise of each photovoltaic power output scenario s The second-stage constraint model of the distribution network's second-order cone branch power flow is solved independently. Since this model is a standard second-order cone programming (SOCP) problem, the minimum operating cost value for each scenario is quickly calculated using a commercial solver. h s * : ; Step b: Based on the cost constant vectors for each scenario calculated in step a A convex optimization model is established with the objective of maximizing the expected cost. This model seeks the worst-case photovoltaic output probability distribution while satisfying the probability normalization constraint and the Rényi divergence constraint defined above. : .

[0064] S92, Initialize lower bound LB= Upper Boundary UB=+ and number of iterations k =0.

[0065] S93. Solve the main problem to obtain the optimal planning scheme. x * Update the lower bound to the objective function value of the principal problem.

[0066] S94. The optimal planning scheme is... x * The subproblem is input, and the subproblem is solved to obtain the worst-case probability distribution. P (k+1) The objective function value of the subproblem Update the upper bound UB = min(UB, first stage cost + ... Z sub ).

[0067] Solving the subproblem involves sequentially executing steps a and b.

[0068] S95. Determine whether the difference between the upper bound and the lower bound is less than the preset convergence tolerance. If yes, stop the iteration and output the optimization result; otherwise, generate a new optimal cutting plane and add it to the main problem, increment the iteration count by one, and return to execute S93.

[0069] In this way, the column and constraint generation algorithm decomposes the original problem into a master problem (MP) and a sub-problem (SP) and solves them alternately and iteratively, which can efficiently and accurately obtain the optimal optimization result.

[0070] like Figure 4 As shown, Figure 4 This demonstrates the specific workflow of the column and constraint generation algorithm, which takes into account distribution network parameters, photovoltaic prediction data, and Rényi divergence parameters. The algorithm decomposes the original three-layer, two-stage load-bearing capacity sub-Blule bar optimization model into a main problem and sub-problems, which are then iteratively combined. The main problem is responsible for optimizing the planning decisions in the first stage, namely capacitor switching and SVC capacity, and provides the lower bound LB for system cost. The sub-problems, using a decoupled solution strategy, first calculate the minimum operating cost for each photovoltaic scenario in parallel, and then find the worst-case probability distribution within the dynamic sub-Blule bar fuzzy set based on this cost. This allows the calculation of the upper bound UB for system cost, and the generation of a new cutting plane that is fed back to the main problem. The algorithm iteratively tightens the upper and lower bounds until the convergence condition is met, at which point the optimization result is output.

[0071] S10. Based on the optimization results, the maximum photovoltaic carrying capacity is obtained.

[0072] The optimization result corresponds to the maximum accessible photovoltaic capacity under given uncertainties and security constraints.

[0073] like Figure 5 As shown, Figure 5 This paper presents a framework for the carrying capacity of building-integrated photovoltaic (PV) distribution networks based on Rényi divergence and dynamic fractional-barred fuzzy sets. The framework consists of three stages: a data-driven perception stage, where a conditional diffusion model is trained using the historical output sequences of distributed PV at each node and their corresponding historical meteorological covariate data to extract the probabilistic characteristics of PV output as a reference distribution (S1-S5); a dynamic uncertainty modeling stage, where a dynamic fractional-barred fuzzy set that changes in real-time with the environment is constructed based on Rényi divergence and fractional matching loss (S6-S7); and a robust carrying capacity assessment stage, where the maximum PV carrying capacity of the distribution network is evaluated by constructing and solving a two-stage robust optimization model for source-grid collaboration (S8-S10). The outputs of these three stages are ultimately converged to form a comprehensive assessment scheme that balances economic efficiency and system security.

[0074] Select as Figure 6 The simulation example uses a real-world distribution network with a voltage level of 10kV and a typical radial grid structure. It includes 15 building-integrated photovoltaic (BIPV) nodes. The model was trained using 300 days of historical operating data and tested with 100 days of new data. Evaluation metrics include load-bearing capacity and voltage exceedance probability.

[0075] As shown in Table 1, the Rényi divergence used in this invention has three technical advantages: (1) Flexible robustness adjustment capability: through parameters It can provide differentiated assessment results from 8.2MW to 10.5MW, while KL divergence can only give a single fixed result; (2) Reduce oversensitivity to the tail of the distribution: in When the value is 1.2, the conservative bias is only 6.7%, while the KL divergence reaches 19.0%, and excessive conservatism leads to a waste of resources; (3) Improved out-of-sample robustness: Rényi divergence ( The out-of-sample voltage exceedance rate of (=1.1) is 1.8%, which is 43.8% lower than the KL divergence of 3.2% and 78.8% lower than the traditional robust optimization of 8.5%.

[0076] Table 1. Comparison of the overall performance of different robust optimization methods

[0077] As shown in Table 2, the conditional diffusion model adopted in this invention significantly outperforms existing methods in terms of distribution fitting accuracy, sample diversity, conditional response accuracy, and computational efficiency. Compared with the GAN method, the Wasserstein distance (bulldozer distance) is reduced by an average of 64% under various meteorological scenarios, and the sample coverage rate is increased from 67.3% to 94.6% (an improvement of 40.6%). It can capture 17 out of 18 real modes, while GAN can only capture 9, thus effectively avoiding the mode collapse problem. The average RMSE under different irradiation conditions is reduced from 13.8% to 4.6% (a reduction of 66.7%), accurately reflecting the complex influence of meteorological conditions on photovoltaic power output distribution. At the same time, the computation time is reduced from 1850 seconds to 285 seconds (an improvement of 84.6%), and the training process is stable without repeated hyperparameter tuning. Compared with the traditional Monte Carlo method, the conditional diffusion model can learn complex multimodal and asymmetric distribution characteristics without parameterized distribution assumptions, and the required number of samples is reduced from 50,000 to 5,000, with comprehensive superior performance.

[0078] Table 2. Comparison of overall performance of different scene generation methods

[0079] As shown in Table 3, the dynamic sub-Brubar fuzzy set adjustment method proposed in this invention generates corresponding conditional scenarios based on daily weather forecasts and dynamically determines the fuzzy set radius for that day. This effectively eliminates extreme distributions in the static fuzzy set that do not conform to the weather conditions of the day, significantly improving the economic efficiency of the assessment. Under sunny conditions, the dynamic sub-Brubar fuzzy set radius shrinks from 0.55 to 0.32, the carrying capacity assessment increases from 9.1MW to 11.3MW, and the resource utilization rate increases by 24.2%. Under cloudy and rainy weather, the dynamic sub-Brubar fuzzy set radius expands to 0.73 to ensure safety, avoiding the overload risk that the static fuzzy set may cause under these conditions. The weighted average resource utilization rate increases by 12.1%, fully demonstrating the technical advantage of the dynamic sub-Brubar fuzzy set in improving the economic efficiency of the assessment by eliminating irrelevant extreme scenarios.

[0080] Table 3. Comparison of comprehensive performance between dynamic and static fuzzy sets.

[0081] As shown in Table 4, this invention, through the systematic integration of three innovative technologies—Rényi divergence, conditional diffusion model, and dynamic KL divergence fuzzy set—achieves a 10.8% increase in carrying capacity (from 9.1MW to 10.3MW), a 52.0% reduction in voltage exceedance rate (from 2.9% to 1.2%), a 10.7 percentage point increase in resource utilization (from 85.0% to 96.2%), a 35.7% improvement in out-of-sample robustness, and a 65.2% improvement in scene generation quality compared to the existing best method (KL divergence + static fuzzy set + Monte Carlo). It also possesses a parameterized flexible adjustment capability not found in existing technologies (through…). It balances robustness and economy in parameters and has strong weather condition response capabilities (by finely characterizing time-varying uncertainties through dynamic fuzzy sets). In terms of computational efficiency, it improves by 21.6% compared to the Monte Carlo method and is far superior to the GAN method. Its comprehensive technical performance is superior to existing technologies, providing an accurate, flexible and robust technical solution for assessing the photovoltaic carrying capacity of building distribution networks.

[0082] Table 4. Comprehensive comparison between the present invention and the prior art

[0083] In summary, the above-mentioned method for assessing the photovoltaic carrying capacity of a distribution network addresses the problem that KL divergence is overly sensitive to the distribution tails and lacks a flexible adjustment mechanism by introducing Rényi divergence as a metric to construct a fuzzy set. Rényi divergence incorporates an order parameter... To achieve flexible adjustment of sensitivity to different parts of the distribution: when When the value is greater than 1, the sensitivity to the distribution tails decreases, and the evaluation results are more economical; when When the value is less than 1, more attention is paid to the tail, and the assessment is more conservative; when When the value approaches 1, it degenerates into KL divergence. This invention constructs a parameterized dynamic sub-Bruker fuzzy set centered on the historical data empirical distribution and with the Rényi divergence as the radius. Through parameters... The choice achieves a flexible balance between robustness and economy, and further derives the dual form of the sub-Brutal optimization problem based on Rényi divergence fuzzy sets, transforming the infinite-dimensional probability distribution optimization problem into a finite-dimensional convex optimization problem. This problem is efficiently solved using a quadratic cone programming solver, realizing a parameterized robust assessment of the building photovoltaic (PV) carrying capacity. Addressing the problem that static fuzzy sets cannot capture the dynamic changes in PV output uncertainty with meteorological conditions, this invention employs a conditional diffusion model as a scene generation tool. This model adds noise to the data through a forward diffusion process and then restores the data through a reverse denoising process. During the denoising process, meteorological covariate information is deeply integrated. Compared with traditional Monte Carlo methods, the conditional diffusion model does not require parameterized distribution assumptions and can learn complex multimodal and asymmetric distributions. Compared with GANs, the diffusion model is more stable in training, does not suffer from mode collapse, and can more effectively integrate meteorological condition information. Specifically, it integrates solar irradiance, cloud cover, temperature, and other factors. Meteorological variables such as humidity and wind speed are input as conditional vectors into a conditional denoising neural network. The model is trained by minimizing the denoising loss function, enabling it to learn the true distribution characteristics of building photovoltaic power output under different meteorological conditions. Given specific meteorological conditions, the model generates multiple scenario samples under those conditions. These samples accurately reflect the statistical characteristics of that meteorological mode. Based on the generated conditional scenario samples, the fuzzy set radius of each operating segment is dynamically determined: under stable sunny conditions, the sample set has small fluctuations, and the radius automatically shrinks; under cloudy or extreme weather conditions, the sample set has large fluctuations, and the radius automatically expands, thus achieving differentiated risk management for different operating scenarios. Furthermore, by combining the two innovative technologies mentioned above with a second-order cone power flow model, a complete evaluation framework is formed. A physical constraint model of the distribution network is established, and the non-convex AC power flow equations are accurately transformed into second-order cone constraints using a second-order cone relaxation technique. This model has higher accuracy than the linearized model and can accurately characterize the coupling relationship between voltage, active power, and reactive power. Simultaneously, a two-stage capacity-carrying capacity sub-Bruker optimization model is constructed, aiming to maximize the total grid-connected capacity of building-integrated photovoltaics (BIPV). This problem is transformed into a deterministic optimization problem using duality theory and solved using a commercial solver, obtaining capacity estimates and optimal grid-connected capacity allocation schemes under different risk preference parameters. This method systematically addresses the shortcomings of existing technologies in terms of measurement tools, fuzzy set construction, scene generation, and physical modeling, achieving an accurate, flexible, and robust comprehensive evaluation of the BIPV capacity of distribution networks.

[0084] According to another aspect of the invention, Figure 2 This is a schematic diagram illustrating a building-based photovoltaic (PV) carrying capacity assessment system for a power distribution network according to an embodiment of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the building-based PV carrying capacity assessment method for a power distribution network as described above.

[0085] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for assessing the photovoltaic carrying capacity of building distribution networks, characterized in that, Including the following steps: Obtain forecast meteorological covariate data for the power distribution network connected to building-integrated photovoltaics; The predicted meteorological covariate data are input into the trained conditional diffusion model, which outputs the photovoltaic power output distribution. The Rényi divergence is defined based on the photovoltaic power output distribution and its corresponding true probability distribution. A dynamic sub-Brussels bar fuzzy set constrained by dynamic radius is established based on the Rényi divergence; Establish a two-stage load-bearing capacity sub-Bluerge optimization model for the aforementioned distribution network; Based on the dynamic sub-bar fuzzy set, the column and constraint generation algorithm is used to solve the two-stage bearing capacity sub-bar optimization model to obtain the optimization result; The maximum photovoltaic carrying capacity is obtained based on the optimization results.

2. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 1, characterized in that, Before obtaining the forecast meteorological covariate data for the distribution network connected to building-integrated photovoltaics, the following steps are also included: Obtain historical operating data of the distribution network connected to building photovoltaics, including the historical output sequence of distributed photovoltaics at each node and its corresponding historical meteorological covariate data; The historical operation data is normalized to obtain the processed historical operation data; The conditional diffusion model is trained using the processed historical operating data to obtain the trained conditional diffusion model.

3. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 1, characterized in that, Based on the photovoltaic power output distribution and its corresponding true probability distribution, the Rényi divergence is defined as follows: The photovoltaic power output distribution is discretized into a reference probability distribution that includes multiple typical scenarios; The Rényi divergence between the reference probability distribution and its corresponding true probability distribution is defined as follows: ； In the formula, Represents the reference probability distribution and its corresponding true probability distribution Rényi divergence between them Indicates the order parameter, S This represents the total number of typical scenarios. p s Describes the true probability distribution at the th... s Probabilities in a typical scenario p 0,s Indicates the reference probability distribution at the th s The probability under a typical scenario, and satisfying .

4. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 3, characterized in that, Before establishing the dynamic sub-Bruker fuzzy set constrained by the dynamic radius based on the Rényi divergence, the following steps are also included: Establish the loss function for the predicted meteorological covariate data; The dynamic radius is calculated based on the loss function.

5. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 4, characterized in that, The loss function for the predicted meteorological covariate data is established as follows: ； In the formula, Represents forecast weather covariate data loss function, This represents a positive weighting function. This represents the set of all learnable weights and bias parameters in a neural network. Indicates by parameters The fitted conditional fractional function, Indicates the diffusion time t The state after adding noise, Represents the true fractional function, This represents the transfer nucleus in the forward diffusion process. This represents a sample of actual photovoltaic power output; The dynamic radius is calculated based on the aforementioned loss function, specifically as follows: ； In the formula, Indicates data of covariates related to weather forecasting The dynamic radius changes in real time. r base This represents the preset basic safety radius constant. This represents the sensitivity adjustment coefficient.

6. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 5, characterized in that, Based on the Rényi divergence, a dynamic sub-Bruker fuzzy set constrained by the dynamic radius is established, specifically as follows: ； In the formula, Represents a dynamic fuzzy set of Brussels bars. express S The space of nonnegative real numbers.

7. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 6, characterized in that, The two-stage carrying capacity sub-Blu-ray bar optimization model for the aforementioned distribution network includes: A three-level objective function is established with the goal of minimizing the sum of the investment cost of the static var compensator, the capacitor switching penalty cost, and the expected operating penalty cost under the worst-case distribution. Establish first-stage constraints and second-stage constraints corresponding to the three-layer objective function. The first-stage constraints include explicit capacity and operation constraints of grouped switching capacitor banks and capacity planning constraints of static var compensators. The second-stage constraints include two-stage connection constraints, power flow model of second-order cone branch of distribution network, system safe operation and feeder capacity constraints, and building photovoltaic constraints.

8. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 7, characterized in that, A three-level objective function is established to minimize the sum of the investment cost of the static var compensator, the capacitor switching penalty cost, and the expected operating penalty cost under the worst-case distribution. Specifically: ； In the formula, n and Q SVC For the first level of decision variables, n Indicates the number of capacitor banks put into operation. Q SVC This indicates the planned capacity of the static var compensator. This represents the set of nodes equipped with static var compensators (SVCs). This represents the unit capacity investment cost of a static var compensator. Represents a node j The planned static var compensator capacity is as follows: T Indicates a time period. This represents the set of nodes where capacitor banks are installed. c sw This represents the penalty coefficient for a single capacitor switching operation, used to prevent unnecessary redundant switching. express t Time Node j The number of capacitor banks installed at the location express t -1 time node j The number of capacitor banks installed at the location P For the second level of decision variables, y s These are the third-level decision variables, containing all voltages and currents at the nodes. Indicates network loss weight. express t Time of the first s Total network loss in a typical scenario Indicates the weight of the light-abandoning penalty. This represents the set of nodes equipped with photovoltaic panels. express t Time Node j In the s Wasted light in a typical scenario.

9. The method for assessing the photovoltaic carrying capacity of a power distribution network building according to claim 1, characterized in that, Based on the dynamic sub-bar fuzzy set, the column and constraint generation algorithm is used to solve the two-stage bearing capacity sub-bar optimization model, and the optimization results include: The two-stage load-bearing capacity split-bar optimization model is decomposed into a main problem and sub-problems. The main problem optimizes the capacitor switching strategy and static var compensator planning capacity in the first stage, and uses auxiliary variables to approximate the worst-case operating cost in the second stage. Its goal is to provide a lower bound solution that satisfies the constraints of the currently known worst-case scenario set. The goal of the sub-problems is to find a new worst-case probability distribution in the dynamic split-bar fuzzy set, given the first-stage decision of the main problem, so as to maximize the expected operating cost of the system and provide an upper bound and a new cutting plane constraint for the main problem. Initialize the lower bound, upper bound, and number of iterations; Solve the main problem to obtain the optimal planning scheme, and update the lower bound to the objective function value of the main problem; The optimal planning scheme is passed into the subproblem, the subproblem is solved to obtain the worst-case probability distribution and the objective function value of the subproblem, and the upper bound is updated. Determine whether the difference between the upper bound and the lower bound is less than the preset convergence tolerance. If yes, stop the iteration and output the optimization result. If no, generate a new optimal cutting plane and add it to the main problem, increment the iteration count by one, and return to execute the steps of solving the main problem.

10. A building photovoltaic carrying capacity assessment system for a power distribution network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the method for assessing the photovoltaic carrying capacity of a power distribution network building as described in any one of claims 1 to 9.

Images