A power distribution network dynamic risk assessment method and device based on scenario reduction, a terminal device, and a storage medium

Abstract

The application discloses a power distribution network dynamic risk assessment method and device based on scene reduction, a terminal device and a storage medium, and belongs to the field of power system monitoring. The method comprises the following steps: constructing an enhanced scene set according to source and load historical time sequence data and a preset source and load space-time constraint; taking minimizing the probability distribution deviation of the reduced enhanced scene set as the target, recursively removing redundant scenes based on the probability distribution distance between scenes, distributing the occurrence probability of the redundant scenes to the corresponding nearest neighbor reserved scenes to obtain a statistical typical scene set; selecting scenes with a scene risk measure index greater than a preset scene threshold from the enhanced scene set as extreme risk scenes, combining the statistical typical scene set to calculate a union to obtain a risk assessment scene set; constructing a sensitivity correlation model according to the operation state of the power distribution network determined by the configuration parameters of the power distribution network, and performing weighting on the time sequence risk value obtained by model solving to obtain a comprehensive risk score for evaluation. Through the application, the problem of low accuracy of existing evaluation is solved.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring technology, and in particular to a method, device, terminal equipment and storage medium for dynamic risk assessment of distribution networks based on scenario reduction. Background Technology

[0002] In recent years, global climate change and energy security issues have become increasingly severe, prompting the world to accelerate energy structure transformation and actively promote green and low-carbon development strategies. With distributed photovoltaic, wind power, and other renewable energy sources being integrated into distribution networks at high penetration rates, the distribution network exhibits strong randomness, volatility, and nonlinearity. Therefore, scenario-based operational risk assessment has become a key technology for ensuring the safe and stable operation of distribution networks.

[0003] Existing scene reduction techniques generally prioritize statistical distribution fidelity as their core optimization objective. They eliminate redundant scenes using clustering algorithms or probabilistic distance metrics to ensure that the reduced scene set closely approximates the overall probability distribution characteristics of the original scene. However, these methods often exclude extreme operating conditions with extremely low probability of occurrence but serious consequences for system safety, as these conditions constitute a very small proportion of the probability distribution and are frequently judged as redundant noise. This leads to an overall optimistic risk assessment, failing to identify potential extreme system operating risks and ultimately resulting in low accuracy of the risk assessment results. Summary of the Invention

[0004] This invention provides a method, device, terminal equipment, and storage medium for dynamic risk assessment of power distribution networks based on scenario reduction, which can effectively solve the problem that existing technologies cannot identify potential extreme system operation risks, ultimately leading to low accuracy of risk assessment results.

[0005] One embodiment of the present invention provides a method for dynamic risk assessment of distribution networks based on scenario reduction, comprising: Obtain the distribution network configuration parameters and historical time-series data of the source and load of the distribution network to be evaluated; An enhanced scene set is obtained by constructing a scene set based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load; With the goal of minimizing the probability distribution deviation of the enhanced scene set, redundant scenes are recursively eliminated based on the probability distribution distance between each enhanced scene in the enhanced scene set. The occurrence probability of the redundant scene is then assigned to the nearest neighboring retained scene to obtain a statistical typical scene set. Based on the preset scenario risk measurement index, scenarios with scenario risk measurement indices greater than the preset scenario threshold are selected from the enhanced scenario set as extreme risk scenarios. Then, the extreme risk scenarios and the statistical typical scenario set are combined to obtain the risk assessment scenario set. The distribution network operation status is determined based on the distribution network configuration parameters. A sensitivity correlation model is constructed based on the distribution network operation status to characterize the relationship between node power, node voltage, and branch power flow. The time-series risk value of each risk assessment scenario in the risk assessment scenario set is solved based on the sensitivity correlation model. The time-series risk values ​​are weighted and fused to obtain a comprehensive risk score, and a risk assessment is performed based on the comprehensive risk score.

[0006] Furthermore, the source-load historical time-series data includes: historical output time-series data of distributed power sources and historical operation time-series data of loads; An enhanced scene set is constructed based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load, including: A first marginal probability distribution model of distributed power source power fluctuation is constructed based on the historical output time series data, and a second marginal probability distribution model of load power fluctuation is constructed based on the historical operating time series data. Based on the first marginal probability distribution model, the second marginal probability distribution model, and the preset source load spatiotemporal constraints, a time-series initial scenario set is generated; Extreme operating features are extracted based on the historical output time series data and the historical operating time series data. Extreme operating condition samples are generated based on the extreme operating features. The extreme operating condition samples are added to the initial time series scenario set to obtain the enhanced scenario set.

[0007] Further, based on the first marginal probability distribution model, the second marginal probability distribution model, and the preset source load spatiotemporal constraints, a time-series initial scene set is generated, including: Based on the preset source-load spatiotemporal constraints, the spatiotemporal correlation matrix between distributed power sources and loads is determined; Based on the Nataf transform, the non-normal random variables corresponding to the first marginal probability distribution model and the second marginal probability distribution model are mapped to the standard normal space. At the same time, the spatiotemporal correlation matrix is ​​positive definite to obtain the effective correlation matrix in the standard normal space. Under the constraint of the effective correlation matrix, random samples in the standard normal space are generated by Latin hypercube sampling. The random samples are then inversely transformed and mapped back to the original power space to obtain the initial time-series scene set.

[0008] Furthermore, with the goal of minimizing the probability distribution deviation of the enhanced scene set, redundant scenes are recursively eliminated based on the probability distribution distance between each enhanced scene in the enhanced scene set. The occurrence probability corresponding to the redundant scene is then assigned to the nearest neighbor retained scene, resulting in a statistically typical scene set, including: Calculate the spatiotemporal geometric distance between each enhanced scene in the enhanced scene set, and calculate the probability distribution distance based on the preset occurrence probability and the spatiotemporal geometric distance; With the goal of minimizing the probability distribution deviation of the reduced augmented scene set, the augmented scene with the smallest probability distribution distance is recursively removed by synchronous backtracking method. The augmented scene to be removed is regarded as a redundant scene, and the augmented scene that is not removed is regarded as a retained scene. The occurrence probabilities of redundant scenarios are redistributed to the retained scenarios with the closest spatiotemporal geometric distance to obtain a set of statistically typical scenarios. The probability distribution bias is determined based on the difference in probability distribution between the enhanced scene set before reduction and the statistically typical scene set after reduction.

[0009] Furthermore, based on preset scenario risk measurement indicators, scenarios with scenario risk measurement indicators greater than preset scenario thresholds are selected from the enhanced scenario set as extreme risk scenarios, including: Using the preset distribution network node voltage safety threshold and branch rated current carrying capacity threshold as safety constraint boundaries, a scenario risk measurement index is constructed to characterize the severity of scenario operation exceeding limits. For each enhanced scenario in the enhanced scenario set, the degree to which the node voltage and branch power flow deviate from the safety constraint boundary corresponding to the scenario source load power are calculated at each time step to obtain the risk measurement value at each time step, and the maximum risk measurement value is used as the scenario risk measurement index value. Enhanced scenarios where the scenario risk measurement index value is greater than a preset scenario threshold are classified as extreme risk scenarios.

[0010] Furthermore, the power distribution network configuration parameters include: network topology data, branch electrical data, and node source-load configuration data; The distribution network operating status is determined based on the distribution network configuration parameters. A sensitivity correlation model is constructed based on the distribution network operating status to characterize the relationship between node power, node voltage, and branch power flow, including: Based on the network topology data, the branch electrical data, and the node source-load configuration data, the operating status of the distribution network is determined, and the corresponding node voltage reference value and branch power flow reference value are determined based on the operating status of the distribution network. The distribution network Jacobian matrix is ​​obtained by performing a first-order Taylor expansion of the preset power balance equations of the distribution network nodes at the operating state of the distribution network. By inverting the Jacobian matrix of the distribution network, and based on the node voltage reference value and the branch power flow reference value, the first linear correlation between the node power injection change and the node voltage change, and the second linear correlation between the node power injection change and the branch power flow change are derived. A sensitivity correlation model is constructed based on the first linear correlation and the second linear correlation.

[0011] Furthermore, based on the sensitivity correlation model, the time-series risk values ​​for each risk assessment scenario in the risk assessment scenario set are calculated, including: For each risk assessment scenario in the risk assessment scenario set, the source load power value at the corresponding time is extracted, and the source load power value is subtracted from the source load power benchmark value corresponding to the power grid operating state to obtain the source load power disturbance value. The source-load power disturbance value is input into the sensitivity correlation model, and the node voltage operation value and branch power flow operation value at each time are obtained based on the first linear correlation relationship and the second linear correlation relationship. Based on the node voltage operating value and the preset distribution network node voltage safety threshold, the voltage over-limit risk value is calculated; based on the branch power flow operating value and the preset branch rated current carrying capacity threshold, the branch power flow overload risk value is calculated; based on the node voltage over-limit state corresponding to the voltage over-limit risk value, the branch power flow over-limit state corresponding to the branch power flow overload risk value, and the preset resource regulation margin, the regulation capacity shortage risk value is determined. The voltage over-limit risk value, branch power flow overload risk value, and regulation capacity shortage risk value are used as time-series risk values.

[0012] As an improvement to the above solution, another embodiment of the present invention provides a dynamic risk assessment device for distribution networks based on scenario reduction, comprising: The data acquisition module is used to acquire the distribution network configuration parameters and source-load historical time-series data of the distribution network to be evaluated. The enhanced scene set construction module is used to construct a scene set based on the source load historical time series data and preset source load spatiotemporal constraints, thereby obtaining an enhanced scene set. The scene reduction module is used to minimize the probability distribution deviation of the reduction enhancement scene set. Based on the probability distribution distance between each enhancement scene in the enhancement scene set, redundant scenes are recursively eliminated. The occurrence probability of the redundant scene is assigned to the corresponding nearest neighbor retained scene to obtain a statistical typical scene set. The risk assessment scenario set construction module is used to select scenarios whose scenario risk measurement indicators are greater than preset scenario thresholds from the enhanced scenario set based on preset scenario risk measurement indicators, as extreme risk scenarios, and perform union calculation based on the extreme risk scenarios and the statistical typical scenario set to obtain the risk assessment scenario set. The time-series risk value solving module is used to determine the operating status of the distribution network based on the distribution network configuration parameters, construct a sensitivity correlation model to characterize the relationship between node power and node voltage and branch power flow based on the distribution network operating status, and solve the time-series risk value of each risk assessment scenario in the risk assessment scenario set based on the sensitivity correlation model. The power grid risk assessment module is used to perform weighted fusion based on the time-series risk values ​​to obtain a comprehensive risk score, and to conduct risk assessment based on the comprehensive risk score.

[0013] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a dynamic risk assessment method for distribution networks based on scenario reduction as described in the above embodiments.

[0014] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the dynamic risk assessment method for distribution networks based on scenario reduction described in the above embodiment.

[0015] By implementing this invention, at least the following beneficial effects are achieved: This invention provides a method, device, terminal equipment, and storage medium for dynamic risk assessment of distribution networks based on scenario reduction. The method first constructs an enhanced scenario set based on historical time-series data of source and load and preset spatiotemporal constraints. It then expands the sample size of extreme scenarios based on historical time-series data of source and load, increasing the cardinality of extreme scenarios in the overall scenario set and alleviating the problem of extreme sample scarcity from the data source level. Next, a hybrid scenario reduction strategy is adopted to minimize the probability distribution deviation of the enhanced scenario set. Redundant scenarios are recursively eliminated based on the probability distribution distance between each enhanced scenario in the enhanced scenario set. The occurrence probability of redundant scenarios is assigned to the nearest neighbor retained scenario, resulting in a statistically typical scenario set. Unlike traditional methods that rely solely on a single probability distribution distance to filter and delete samples, this method does not use probability percentage as the sole criterion for scenario selection, ensuring that low-probability, high-risk extreme scenarios are not directly eliminated due to low distribution weight. Next, based on preset scenario risk measurement indicators, scenarios with scenario risk measurement indicators greater than preset scenario thresholds are selected from the enhanced scenario set as extreme risk scenarios. A union calculation is then performed on these extreme risk scenarios and the statistical typical scenario set to obtain a risk assessment scenario set. This risk assessment scenario set integrates extreme risk scenarios and the statistical typical scenario set, accurately capturing potential extreme operational risks in the distribution network and addressing the industry pain point of overly optimistic assessment results, fundamentally improving the accuracy of dynamic risk assessment results. Furthermore, in conjunction with a sensitivity correlation model and the distribution network operating status, time-series risk values ​​are calculated. From expanding extreme samples during the generation stage of the statistical typical scenario set, to locking in extreme risk scenarios during the reduction stage, and then to the integration of scenario risk measurement indicators, the risk information corresponding to extreme operating conditions is fully transmitted to the final comprehensive risk score. This allows the risk assessment to objectively reflect all operational risks of the distribution network, including extreme disturbances, improving the accuracy of risk assessment results and providing accurate risk decision-making information for distribution network dispatching and operation. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a dynamic risk assessment method for power distribution networks based on scenario reduction, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the specific architecture of a WGAN-GP network provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a dynamic risk assessment device for power distribution networks based on scenario reduction, provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 To address the problem that existing technologies cannot identify potential extreme system operation risks, ultimately leading to low accuracy in risk assessment results, an embodiment of the present invention provides a flowchart illustrating a dynamic risk assessment method for distribution networks based on scenario reduction, comprising: S1. Obtain the distribution network configuration parameters and source-load historical time-series data of the distribution network to be evaluated; Specifically, distribution network configuration parameters are the physical basis for risk assessment, including three categories: network topology data, branch electrical data, and node source-load configuration data. Network topology data refers to the connection relationships between nodes and branches; branch electrical data refers to parameters such as line impedance and rated current carrying capacity; and node source-load configuration data refers to configuration information such as the capacity of distributed power sources connected to each node and the load baseline capacity. Historical time-series source-load data records the historical output of distributed power sources and the historical operation data of loads in chronological order, reflecting the historical fluctuation patterns of source-load power, and serves as the data foundation for constructing uncertainty models and generating operational scenarios.

[0019] To illustrate, the distribution network configuration parameters and source-load historical time-series data of the distribution network to be evaluated are collected. The distribution network configuration parameters are the static physical basis of the power grid, and the source-load historical time-series data are the data carrier of the source-load fluctuation pattern. Together, they constitute the data input layer of the entire evaluation method.

[0020] S2. Construct a scene set based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load to obtain an enhanced scene set; Specifically, the enhanced scenario set is a set of scenarios that supplements the regular operating scenarios with extreme operating condition samples. It covers both the regular operating range and extreme boundary conditions, thus solving the problem of insufficient historical extreme samples.

[0021] To illustrate, based on historical time-series data of source loads, fluctuation patterns are mined, and conventional operating scenarios are generated by combining preset spatiotemporal constraints of source loads. Then, extreme operating condition samples are added, ultimately resulting in an enhanced scenario set covering both conventional and extreme operating conditions. The aim is to transform the uncertainty of source loads into computable discrete scenarios, while supplementing the missing extreme samples in historical data.

[0022] Preferably, the source-load historical time-series data includes: historical output time-series data of distributed power sources and historical operation time-series data of loads; An enhanced scene set is constructed based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load, including: A first marginal probability distribution model of distributed power source power fluctuation is constructed based on the historical output time series data, and a second marginal probability distribution model of load power fluctuation is constructed based on the historical operating time series data. Based on the first marginal probability distribution model, the second marginal probability distribution model, and the preset source load spatiotemporal constraints, a time-series initial scenario set is generated; Extreme operating features are extracted based on the historical output time series data and the historical operating time series data. Extreme operating condition samples are generated based on the extreme operating features. The extreme operating condition samples are added to the initial time series scenario set to obtain the enhanced scenario set.

[0023] Specifically, historical output time-series data refers to the historical output power time-series data of distributed power sources, including time-period output records of different types of power sources such as distributed photovoltaic and distributed wind power. Historical operating time-series data refers to the historical power consumption time-series data of electricity load, including time-period electricity consumption records of residential load, industrial load, and electric vehicle charging load. The first marginal probability distribution model describes the random distribution law of distributed power source output, with different distribution forms corresponding to different types of power sources. The second marginal probability distribution model describes the random distribution law of load power. The initial time-series scenario set represents a set of scenarios generated based solely on the conventional probability distribution, covering the conventional operating range, and is a fundamental component of the enhanced scenario set. Extreme value operating characteristics are the statistical characteristics corresponding to extreme operating conditions in the historical source-load data, such as output extremes, load extremes, and extreme fluctuations of multi-source coupling, which serve as the basis for generating extreme operating condition samples. Extreme operating condition samples represent source-load time-series scenario samples generated by simulating extreme operating conditions, used to supplement the shortcomings of insufficient extreme samples in historical data.

[0024] Schematic, for historical output time-series data of distributed power sources, a first marginal probability distribution model is constructed by matching the probability distributions of different types of power sources. Distributed photovoltaic output follows a Beta distribution, with its shape parameter fitted by the mean and variance of historical irradiance data; distributed wind power output follows a two-parameter Weibull distribution, with the shape parameter reflecting the concentration of power fluctuations and the proportion parameter reflecting the overall power level. For historical load operation time-series data, a second marginal probability distribution model is constructed using a Gaussian mixture model. This model uses a weighted combination of multiple Gaussian distributions to characterize the multi-peak fluctuation characteristics of the load, with each Gaussian component corresponding to a typical load operating state, such as off-peak, off-peak, and peak modes.

[0025] By combining the first and second marginal probability distribution models, along with preset source-load spatiotemporal constraints, a time-series initial scenario set covering the normal operating range is generated. The source-load spatiotemporal constraints mainly refer to the statistical correlation between the power of source loads at different nodes and at different time periods. This correlation must be satisfied when generating scenarios to conform to actual operating patterns.

[0026] Then, extreme value operating features are extracted from historical output time-series data and historical operating time-series data. Based on these extreme value operating features, extreme operating condition samples that conform to the extreme distribution are generated. These extreme operating condition samples are then added to the initial time-series scenario set to obtain the enhanced scenario set. In specific implementation, a Wasserstein generative adversarial network with gradient penalty is used to learn the extreme value features in the historical data. Through iterative training, the generator outputs time-series samples that conform to the extreme operating condition distribution. At the same time, a shift factor is set to shift the generation target towards the tail of the distribution, selecting samples with higher extremeness to add to ensure coverage of extreme operating conditions.

[0027] In a preferred embodiment of the present invention, for distributed photovoltaic power, its per-unit output value per unit time is... It follows a Beta distribution, and its marginal probability distribution model is: ,in, This represents the normalized output or per-unit output value of distributed photovoltaic power per unit time, with a value range of [0,1]. , The shape parameter of the Beta distribution is determined by the mean and variance of historical irradiance data; This is a Beta function.

[0028] For distributed wind power, the wind speed and corresponding output It usually follows a two-parameter Weibull distribution, and its marginal probability distribution model is: In the formula, k is a shape parameter that reflects the degree of concentration of power fluctuations; This is a proportional parameter.

[0029] For electric vehicles and conventional loads, a Gaussian mixture model is used to characterize their multimodal characteristics, and its marginal probability distribution model is as follows: In the formula, The combined load power variable, which consists of electric vehicle charging load and conventional load, can be the node load power or its normalized value; M is the number of mixed components, each of which corresponds to a typical load operating state or electric vehicle charging behavior mode, such as off-peak load state, off-peak load state, peak load state or concentrated charging state. Let be the weight of the i-th component; The mean is Covariance is The normal distribution , These are the mean vector and covariance matrix of the historical comprehensive load sample under the m-th load operation mode, respectively.

[0030] Most existing scenario generation methods only sample from conventional distributions, relying entirely on historical data for extreme samples. However, extreme conditions are rare in actual operation, leading to insufficient boundary coverage of the scenario set and an inability to capture extreme risks in subsequent assessments. While some methods perform data augmentation, the generated extreme samples do not conform to real statistical patterns, resulting in low practical value. This embodiment matches the most suitable probability distribution for different types of source loads: a Beta distribution for photovoltaics, a Weibull distribution for wind power, and a Gaussian mixture model for loads. This approach better reflects actual fluctuation characteristics than single-distribution modeling, resulting in higher confidence levels for the generated conventional scenarios. This embodiment uses generative adversarial networks to learn the extreme value features of real historical data, ensuring that the generated extreme samples conform to the inherent patterns of source load fluctuations and are not fabricated invalid samples. Combined with tail offset filtering, it can accurately locate high-risk extreme conditions, effectively compensating for the lack of historical extreme samples.

[0031] Preferably, a time-series initial scene set is generated based on the first marginal probability distribution model, the second marginal probability distribution model, and preset source load spatiotemporal constraints, including: Based on the preset source-load spatiotemporal constraints, the spatiotemporal correlation matrix between distributed power sources and loads is determined; Based on the Nataf transform, the non-normal random variables corresponding to the first marginal probability distribution model and the second marginal probability distribution model are mapped to the standard normal space. At the same time, the spatiotemporal correlation matrix is ​​positive definite to obtain the effective correlation matrix in the standard normal space. Under the constraint of the effective correlation matrix, random samples in the standard normal space are generated by Latin hypercube sampling. The random samples are then inversely transformed and mapped back to the original power space to obtain the initial time-series scene set.

[0032] Specifically, the spatiotemporal correlation matrix is ​​a matrix that quantifies the statistical correlation between random variables of source and load at different nodes and between power at different time periods. Its elements are the correlation coefficients between variables, representing the mathematical expression of spatiotemporal constraints on source and load. The Natav transform is a classic transformation method for handling non-normally correlated random variables. Its function is to map correlated random variables following arbitrary marginal distributions to a standard normal space, facilitating subsequent correlation constraint sampling. The standard normal space is the variable space where random variables follow a standard normal distribution with a mean of 0 and a variance of 1. The effective correlation matrix represents the standard normal space correlation matrix that meets the positive definite requirement after positive definite verification and correction, serving as the constraint condition for subsequent sampling. The inverse transform maps the sampling results from the standard normal space back to the original power variable space, yielding the actual source and load power scenario.

[0033] Schematic, based on preset source-load spatiotemporal constraints, Pearson correlation coefficients are calculated using historical time-series data of source and load to construct a spatiotemporal correlation matrix of distributed power sources and load random variables. Each element in the spatiotemporal correlation matrix corresponds to the statistical correlation between two source-load variables, including both the spatial correlation of different nodes at the same time and the temporal correlation of the same node at different times, thus fully characterizing the spatiotemporal correlation characteristics of source-load fluctuations.

[0034] Then, through the Natav transformation, the random variables corresponding to the first and second marginal probability distribution models, which follow non-normal distributions, are mapped to the standard normal space. Simultaneously, the original spatiotemporal correlation matrix is ​​transformed to obtain the initial correlation matrix in the standard normal space. Since the transformed initial correlation matrix may not satisfy positive definiteness and cannot be directly used for sampling, it needs to be positive definite. The alternating projection method is used to iteratively correct the non-positive definite initial correlation matrix, ultimately obtaining an effective correlation matrix that meets the positive definiteness requirement.

[0035] Under the constraint of the effective correlation matrix, the Latin hypercube sampling method is used to generate random samples that uniformly cover the probability space in the standard normal space. Then, through the Natav inverse transform, the samples in the standard normal space are mapped back to the original source-charge power space, finally obtaining a time-series initial scene set that conforms to the marginal distribution and spatiotemporal correlation. At the same time, each initial scene is assigned an equal initial occurrence probability, and the sum of the probabilities of all scenes is 1, satisfying the probability normalization requirement.

[0036] In a preferred embodiment of the present invention, to characterize the statistical correlation between source-source and source-load relationships among different nodes, a Nataf transformation is introduced to map the non-normal variable space to a standard normal space. First, variable transformation is performed. For the correlated random variable vector X, which includes the distributed photovoltaic power output, distributed wind power output, electric vehicle charging load, and conventional load power of each node, i.e. Through the marginal cumulative distribution function Convert it to a standard normal variable , In the formula, It is the inverse cumulative distribution function of the standard normal distribution.

[0037] Then, correlation coefficient adjustments are performed. The initial correlation matrix of the original variables. The initial correlation matrix of the original variables was determined using Pearson correlation coefficients based on historical time series data of distributed photovoltaic power output, distributed wind power output, electric vehicle load, and conventional load. It needs to be mapped to an effective correlation matrix in standard normal space. Its elements and The conversion relationship is as follows: In the formula, , These are the i-th original source load random variables. The mean and standard deviation; It is a two-dimensional standard normal probability density function.

[0038] Using improved Latin hypercube sampling Random samples are generated under constraints. The L-Mapm alternating projection method is used to correct the non-positive definiteness of the correlation matrix, ensuring uniform coverage of the sampling points in the n-dimensional probability space. Each sample corresponds to a complete 24-hour source-load power trajectory, including photovoltaic output, wind power output, electric vehicle load, and conventional load data for each node. Multiple random samples are arranged according to scenario number to form the initial time-series scenario set. .

[0039] In a preferred embodiment of the present invention, to address the problem of scarce extreme samples in historical data, a Wasserstein generative adversarial network with gradient penalty is used for data augmentation. The specific architecture of the WGAN-GP network is as follows: Figure 2 As shown, the loss function of WGAN-GP measures the true distribution of source loads in a temporal scene using the Wasserstein distance. With the generated distribution The difference is added, and a gradient penalty term is added to satisfy Lipschitz continuity: In the formula, L is the loss function; This is the gradient penalty coefficient, used to satisfy the 1-Lipschitz continuity constraint, and is usually set to 10. Represents the mathematical expectation; This represents a source payload time-series scene sample generated by the generator; This represents a real-world scene sample obtained from the real source-load temporal scene distribution; This represents the gradient-penalized sample obtained by linear interpolation of the real sample and the generated sample.

[0040] Finally, set the offset factor. =0.1, during the K iterations, the generator training target is shifted towards the tail of the distribution, and extreme condition samples that meet the extreme feature measure are selected to supplement the initial time series scene set: In the formula, The sample set after the k-th iteration. The top-ranked sets retained from the previous set based on the extrema metric A real sample, For generated from the generator Among the candidate samples, the top-ranked samples were selected using a metric function. An extreme sample.

[0041] Through this iterative process, an enhanced scenario set containing a high proportion of extreme operating conditions is ultimately generated.

[0042] Existing scenario generation methods either ignore the spatiotemporal correlation of source and load, resulting in scenarios that do not conform to actual operating patterns; or they are crude in handling the correlation of non-normal variables, leading to non-positive definite problems in the correlation matrix and distorted sampling results. This embodiment uses Natav transform to handle the correlation problem of non-normal variables, a classic and mature method for uncertainty analysis of power systems. It can accurately restore the spatiotemporal correlation characteristics between source and load, and the generated scenario is more in line with actual operating conditions. The alternating projection method is used to correct the non-positive definite problem of the correlation matrix. The algorithm converges and is stable. The corrected matrix not only meets the positive definite requirement, but also retains the original relevant information to the greatest extent, without causing correlation distortion due to correction. Latin hypercube sampling is a stratified sampling method, which has higher spatial coverage than traditional Monte Carlo sampling, better accuracy with the same sample size, or in other words, requires less sample size for the same accuracy, and can effectively control the scale of the initial scenario.

[0043] S3. With the goal of minimizing the probability distribution deviation of the enhanced scene set, redundant scenes are recursively eliminated based on the probability distribution distance between each enhanced scene in the enhanced scene set. The occurrence probability of the redundant scene is assigned to the corresponding nearest neighbor retained scene to obtain a statistical typical scene set. Specifically, the statistical typical scene set represents the set of typical scenes obtained by statistically faithfully reducing the enhanced scene set. The number of scenes is greatly reduced, but the overall probability distribution characteristics are very close to the original set, which is used to compress the amount of subsequent calculations.

[0044] To illustrate, with the goal of minimizing the probability distribution deviation before and after reduction, the probability distribution distance between each scene in the enhanced scene set is first calculated. Then, redundant scenes are recursively removed one by one, while the occurrence probability of the deleted scenes is transferred to the nearest retained scene. Finally, a statistically typical scene set with a reduced number of scenes and preserved statistical characteristics is obtained. The purpose of this step is to compress the scene scale, reduce the subsequent computational burden, and at the same time preserve the probability distribution characteristics of the original scenes as much as possible.

[0045] Preferably, with the goal of minimizing the probability distribution deviation of the enhanced scene set, redundant scenes are recursively eliminated based on the probability distribution distance between each enhanced scene in the enhanced scene set. The occurrence probability corresponding to the redundant scene is then assigned to the nearest neighbor retained scene to obtain a statistically typical scene set, including: Calculate the spatiotemporal geometric distance between each enhanced scene in the enhanced scene set, and calculate the probability distribution distance based on the preset occurrence probability and the spatiotemporal geometric distance; With the goal of minimizing the probability distribution deviation of the reduced augmented scene set, the augmented scene with the smallest probability distribution distance is recursively removed by synchronous backtracking method. The augmented scene to be removed is regarded as a redundant scene, and the augmented scene that is not removed is regarded as a retained scene. The occurrence probabilities of redundant scenarios are redistributed to the retained scenarios with the closest spatiotemporal geometric distance to obtain a set of statistically typical scenarios. The probability distribution bias is determined based on the difference in probability distribution between the enhanced scene set before reduction and the statistically typical scene set after reduction.

[0046] Specifically, spatiotemporal geometric distance is an indicator of the degree of difference between two temporal scenarios. It compares the source load power differences of each node at each time step, and the sum of these differences yields the overall distance; the closer the distance, the more similar the two scenarios. Occurrence probability reflects the likelihood of a scenario occurring and is determined by the sampling method. Probability distribution distance is a quantitative indicator of the degree of difference between two probability distributions; this invention uses the Kantorovich distance. The synchronous backtracking method compresses the scenario set by successively eliminating scenarios with the least impact on the overall distribution, while ensuring minimal probability distribution deviation. Probability redistribution means that the probability of eliminated scenarios cannot be directly discarded but transferred to the most similar retained scenarios to ensure the conservation of total probability. Probability distribution deviation is the degree of difference between the probability distribution of the retained scenarios after reduction and the original scenario probability distribution; the smaller the deviation, the higher the statistical fidelity after reduction.

[0047] Schematic, the spatiotemporal geometric distance between any two temporal scenes in the augmented scene set is first calculated. This is done by calculating the Euclidean norm of the source-load power difference time-by-time and node-by-node, and then summing over the entire time period to obtain the overall geometric distance between the two scenes. Based on this, and considering the occurrence probability of each scene, the Kantorovich distance is used as the probability distribution distance to quantify the overall probability distribution difference of the scene set. The physical meaning of the Kantorovich distance is the minimum total cost of probability relocation required to adjust the probability distribution of the original scene to the probability distribution of the preserved scene; the unit cost of relocation is the spatiotemporal geometric distance between scenes.

[0048] Then, with minimizing the probability distribution deviation after reduction as the optimization objective, a synchronous backtracking method is used for recursive reduction: in each round, all remaining scenes are traversed, the increment of the overall Kantorovich distance after removing the scene is calculated, and the scene with the smallest increment is selected for removal, which is determined to be a redundant scene; this process is repeated until the number of retained scenes reaches the preset requirement. Each time a scene is removed, the probability distribution of the remaining scenes is updated synchronously, ensuring that each reduction step proceeds in the currently optimal direction. For each redundant scene removed, all occurrence probabilities corresponding to that scene are completely redistributed to the retained scene with the closest spatiotemporal geometric distance to it. After all redundant scenes are removed, the remaining retained scenes constitute a statistically typical scene set. The final probability of each retained scene is equal to its initial probability plus the sum of all allocated probabilities, ensuring the overall probability of the scene set is conserved throughout the process.

[0049] In a preferred embodiment of the present invention, a hybrid scene reduction strategy is employed. First, statistical fidelity reduction is performed based on Kantorovich distance, and then a problem-driven scene reduction (PDSR) mechanism is introduced. First, the calculation of any two augmented scenes in the augmented scene set is performed. and The spatiotemporal geometric distance between : In the formula, T is the duration of a single augmentation scene, and K is the total number of source payload nodes in each augmentation scene. It is the Euclidean norm, used to quantify the differences in physical quantities in different scenarios at the same moment.

[0050] Secondly, we define the index set of the deleted redundant scenes as J, and recursively remove scenes using a synchronous backtracking method, such that the Kantorovich distance between the original scene probability distribution P of the enhanced scene set and the probability distribution Q of the retained scene set after statistical reduction is calculated. Minimum: In the formula, Let be the original probability of scenario i; This represents the redundant scene i that was deleted and its place in the retained set. The distance to the nearest neighbor scene j. This step is used to progressively remove redundant scenes that have the least impact on the Kantorovich distance.

[0051] To maintain the overall probability conservation, each redundant scenario that is removed is... The probability compensation is given to the nearest retained scene j: In the formula, The final probability of retaining scene j after reduction; This represents the set of all scenes that were removed because scene j was the nearest neighbor. Through this reduction layer, we obtain... A typical statistical scenario.

[0052] Building upon the aforementioned statistical reduction, a Problem-Driven Scenario Reduction (PDSR) mechanism is introduced to prevent the accidental removal of scenarios with severe consequences but low probability. The physical power scenario is projected as a system state deviation index using the node voltage / current sensitivity matrix L. A scenario risk operation measure for the i-th augmented scenario at time t is defined. : In the formula, , To quickly estimate node voltage and branch current values ​​using power flow sensitivity. , These are the upper and lower safety constraint thresholds for voltage operation. This represents the maximum allowable steady-state current carrying capacity of the branch.

[0053] Subsequently, the maximum scene risk measure for each enhanced scene was calculated over the entire time period. In the enhanced scene set Selected from ( All scenarios (with a preset risk trigger threshold) are defined as critically rare extreme risk scenarios.

[0054] Finally, the cut obtained A set of typical statistical scenarios and the selected Perform union processing on each extreme risk scenario: For extreme risk scenarios where mandatory retention is enforced, the probability... Revised to ,in This is a risk preference adjustment coefficient. The original probabilities of extreme risk scenarios are forcibly retained to ensure that their impact is explicitly captured in subsequent risk indicator calculations, rather than being overwhelmed by high-probability regular scenarios. Finally, the probabilities of all retained scenarios are normalized to meet the following conditions: .

[0055] The above-mentioned hybrid scene reduction strategy achieves an optimal balance between probability distribution fidelity and physical security boundary control for scene sets.

[0056] This embodiment aims to minimize probability distribution deviation, using the Kantorovich distance as a quantification metric. Each reduction step moves towards the optimal direction, ensuring that the reduced scene set retains the statistical distribution characteristics of the original scenes to the greatest extent possible. A nearest neighbor probability redistribution rule is employed, transferring all probabilities of deleted scenes to the most similar retained scenes, maintaining a total probability of 1 throughout the process. This prevents probability loss due to reduction, resulting in more accurate subsequent probability-based risk calculations.

[0057] S4. Based on the preset scenario risk measurement index, select scenarios whose scenario risk measurement index is greater than the preset scenario threshold from the enhanced scenario set as extreme risk scenarios, and perform union calculation based on the extreme risk scenarios and the statistical typical scenario set to obtain the risk assessment scenario set. Specifically, extreme risk scenarios refer to scenarios selected from the enhanced scenario set whose operational risk level exceeds a preset scenario threshold. These scenarios are usually not very probable, but once they occur, they will have a serious impact on power grid security and belong to a critical few scenarios that must be given special attention. The risk assessment scenario set represents the final set of assessment scenarios obtained by taking the union of the statistical typical scenario set and the extreme risk scenario set, while taking into account both statistical representativeness and coverage of extreme risks. It is the direct input for subsequent risk calculations.

[0058] To illustrate, firstly, based on preset scenario risk measurement indicators, extreme risk scenarios with excessive risk levels are selected from the original enhanced scenario set. Then, these extreme risk scenarios are combined with the statistical typical scenario set obtained above to form the final risk assessment scenario set. The reason for this is that pure statistical reduction may easily miss low-probability, high-impact extreme scenarios. By separately screening and forcibly retaining them, we can ensure that extreme risks are not overlooked.

[0059] Preferably, based on a preset scenario risk measurement index, scenarios with scenario risk measurement indices greater than a preset scenario threshold are selected from the enhanced scenario set as extreme risk scenarios, including: Using the preset distribution network node voltage safety threshold and branch rated current carrying capacity threshold as safety constraint boundaries, a scenario risk measurement index is constructed to characterize the severity of scenario operation exceeding limits. For each enhanced scenario in the enhanced scenario set, the degree to which the node voltage and branch power flow deviate from the safety constraint boundary corresponding to the scenario source load power are calculated at each time step to obtain the risk measurement value at each time step, and the maximum risk measurement value is used as the scenario risk measurement index value. Enhanced scenarios where the scenario risk measurement index value is greater than a preset scenario threshold are classified as extreme risk scenarios.

[0060] Specifically, safety constraint boundaries are critical thresholds for the safe operation of the distribution network. Exceeding these thresholds constitutes out-of-limit operation and poses safety risks. These mainly include two categories: node voltage safety thresholds and branch rated current carrying capacity thresholds. The node voltage safety threshold represents the allowable fluctuation range of node voltage as specified in the power grid operation standards, typically including upper and lower limits; exceeding this range constitutes a voltage over-limit. The branch rated current carrying capacity threshold represents the maximum current value that a branch line is allowed to carry over a long period; exceeding this value can lead to line overheating, accelerated aging, or even faults. The scenario risk measurement index is a comprehensive indicator of the severity of operational risks in a single scenario, calculated based on the degree of voltage and power flow over-limits. A higher value indicates a higher scenario risk. The risk measurement value represents the severity of risk in a single scenario at a specific moment; the maximum value over the entire period is the overall risk measurement value for that scenario. Scenario thresholds are the critical values ​​for distinguishing between normal and extreme risk scenarios, set by operators based on the power grid's risk tolerance. Scenarios exceeding these thresholds are classified as extreme risk scenarios.

[0061] Schematic, using the distribution network node voltage safety threshold and branch rated current carrying capacity threshold as safety constraint boundaries, a scenario risk measurement index capable of quantifying the severity of operational limits exceeding limits is constructed. The calculation logic of the scenario risk measurement index is as follows: define voltage limit exceeding severity function and power flow overload severity function respectively; the further the voltage deviates from the safe range and the more the power flow exceeds the rated value, the higher the value of the severity function; combine the two to form a risk measurement value at a single moment, comprehensively reflecting the operational risk level at that moment.

[0062] For each scenario in the enhanced scenario set, the node voltage deviation and branch power flow deviation corresponding to the source load power at each time step are calculated to obtain the risk measurement value at each time step. Then, the maximum value of the risk measurement values ​​at all times throughout the entire period is taken as the overall scenario risk measurement index value for this scenario. The reason for taking the maximum value is that extreme risks are often determined by the most severe time step; as long as there is one time step with extremely high risk, the scenario is considered a high-risk scenario. Finally, the scenario risk measurement index value of each scenario is compared with a preset scenario threshold. Scenarios with index values ​​greater than the scenario threshold are identified as extreme risk scenarios and are all filtered out and retained separately.

[0063] Traditional purely statistical scenario reduction methods only consider the similarity of scenario probability distributions, neglecting the actual operational risks of the scenarios. Many low-probability but high-risk extreme scenarios are discarded as redundancy due to their significant differences from regular scenarios and their small number, leading to the omission of major risks in subsequent assessments. Other solutions, while retaining extreme scenarios, use singular screening indicators, such as voltage or power flow, resulting in incomplete risk assessment. This embodiment constructs risk measurement indicators based on the actual safety constraints of the power grid, selecting only truly high-risk scenarios that threaten grid safety, rather than simply statistically abnormal samples, making the screening results more practically valuable for engineering applications. Furthermore, it uses the maximum value across all time periods as the overall risk value for each scenario, capturing the characteristics of single-point breakthroughs and overall high risk in extreme scenarios, preventing the omission of extreme risks at a particular moment simply because the scenario is normal for most of the time, thus making the assessment more rigorous. Moreover, the scenario threshold can be flexibly adjusted according to the risk tolerance of different power grids; a higher threshold is set for high-risk scenarios, and a lower threshold is set for safety-oriented scenarios, adapting to different operational management needs.

[0064] S5. Determine the operating status of the distribution network based on the distribution network configuration parameters, construct a sensitivity correlation model to characterize the relationship between node power and node voltage and branch power flow based on the distribution network operating status, and solve the time-series risk value of each risk assessment scenario in the risk assessment scenario set based on the sensitivity correlation model. Specifically, the distribution network operating state represents the steady-state operating point of the distribution network under typical operating conditions, corresponding to the benchmark node voltage, branch power flow, and source-load power levels, serving as a baseline reference for constructing the sensitivity model. The sensitivity correlation model represents a linear model derived from the benchmark operating state, characterizing the quantitative correlation between changes in node power injection and changes in node voltage and branch power flow. It replaces nonlinear power flow iteration with linear calculations, significantly improving calculation speed. The time-series risk value represents the quantitative value of the operating risk of a single scenario over the entire time period, covering multiple dimensions such as voltage exceedance, power flow overload, and insufficient regulation capacity, reflecting the risk level of the scenario evolving over time.

[0065] To illustrate, the operating status of the distribution network is first determined based on the network configuration parameters. Then, based on this, a sensitivity correlation model between node power and node voltage, and branch power flow is derived. Finally, the sensitivity correlation model is used to quickly calculate the time-series risk value for each scenario in the risk assessment scenario set. Replacing traditional nonlinear power flow calculations with sensitivity correlation models is the core step in improving computational efficiency.

[0066] Preferably, the power distribution network configuration parameters include: network topology data, branch electrical data, and node source-load configuration data; The distribution network operating status is determined based on the distribution network configuration parameters. A sensitivity correlation model is constructed based on the distribution network operating status to characterize the relationship between node power, node voltage, and branch power flow, including: Based on the network topology data, the branch electrical data, and the node source-load configuration data, the operating status of the distribution network is determined, and the corresponding node voltage reference value and branch power flow reference value are determined based on the operating status of the distribution network. The distribution network Jacobian matrix is ​​obtained by performing a first-order Taylor expansion of the preset power balance equations of the distribution network nodes at the operating state of the distribution network. By inverting the Jacobian matrix of the distribution network, and based on the node voltage reference value and the branch power flow reference value, the first linear correlation between the node power injection change and the node voltage change, and the second linear correlation between the node power injection change and the branch power flow change are derived. A sensitivity correlation model is constructed based on the first linear correlation and the second linear correlation.

[0067] Specifically, the node voltage reference values ​​are the voltage amplitude and phase angle of each node in the distribution network under reference operating conditions, serving as the reference benchmark for sensitivity calculation. The branch power flow reference values ​​are the power and current values ​​of each branch in the distribution network under reference operating conditions, also serving as the reference benchmark for sensitivity calculation. The node power balance equations are a set of nonlinear equations relating the active and reactive power balance of distribution network nodes, forming the fundamental equations for power flow calculation. The Jacobian matrix reflects the linear mapping relationship between power changes and voltage changes, determined by the network topology and branch parameters. The first linear correlation represents the linear correspondence between changes in node power injection and changes in node voltage, obtained by inverting the Jacobian matrix. The second linear correlation represents the linear correspondence between changes in node power injection and changes in branch power flow, derived from the first linear correlation combined with the branch power flow equations.

[0068] Schematic, based on the distribution network topology data, branch electrical data, and node source-load configuration data, the operating state of the distribution network, i.e., the steady-state power flow solution under typical operating conditions, is calculated. Based on the distribution network operating state, the reference voltage values ​​for each node, the reference power flow values ​​for each branch, and the source-load power values ​​for each node are extracted. Using the node power balance equation as a basis, a first-order Taylor expansion of the power equation is performed at the distribution network operating state point. After ignoring higher-order small terms, the linear relationship between power disturbance and voltage disturbance is obtained, and the coefficient matrix is ​​the distribution network Jacobian matrix. The Jacobian matrix typically consists of four submatrices, reflecting the partial derivatives of active power with respect to phase angle, active power with respect to voltage magnitude, reactive power with respect to phase angle, and reactive power with respect to voltage magnitude, respectively.

[0069] Then, by inverting the Jacobian matrix, we obtain the linear mapping relationship between the changes in nodal power injection and the changes in nodal voltage, which is the first linear correlation. Based on this, and combining the equations relating branch power flow and nodal voltage, we further derive the linear mapping relationship between the changes in nodal power injection and the changes in branch power flow, which is the second linear correlation. Integrating the first and second linear correlations forms a complete sensitivity correlation model. By simply inputting the change in source load power, this model can quickly calculate the corresponding changes in nodal voltage and branch power flow.

[0070] In a preferred embodiment of the present invention, the sensitivity coefficient is derived using the Jacobian matrix under the operating conditions of the distribution network, and a linear response model of node voltage and branch power flow to source load power fluctuations is established. A first-order Taylor expansion of the distribution network power balance equation is performed at this reference point under the operating conditions of the distribution network: In the formula, and It represents the disturbance vector of active and reactive power of the node, corresponding to the fluctuation of distributed power output and load in a typical scenario. and is the vector of changes in the phase angle and magnitude of the node voltage; J is the Jacobian matrix, composed of submatrices {H,N,K,L}, which reflects the network topology and branch parameter characteristics.

[0071] By inverting the Jacobian matrix, the first linear relationship between voltage and power change can be obtained: In the formula, and These are the sensitivity matrices of voltage to active and reactive power, respectively. Using this linear correlation, scenario prediction can be performed quickly. The voltage evolution trajectory at different time periods avoids the computational overhead of multiple nonlinear power flow iterations in a single scenario.

[0072] Calculate the branch current using the branch power flow equations. Distribution factor for node power injection: In the formula, Let be the sensitivity coefficient of the branch current ij to the active power injection at node n.

[0073] Traditional risk assessments mostly employ nonlinear power flow iteration to calculate the operational status of each scenario, resulting in large computational loads and long processing times. When dealing with numerous scenarios, this approach fails to meet the timeliness requirements of online assessments. This embodiment replaces the nonlinear power flow iteration with a linear sensitivity correlation model. The computational load for a single scenario is reduced from iteratively solving a system of equations to simple matrix multiplication, increasing the computational speed by tens of times. This makes it particularly suitable for batch calculations across multiple scenarios. Since most source-load fluctuations in the distribution network are small disturbances near the baseline state, the linear approximation of the first-order Taylor expansion is sufficiently accurate, with errors within an acceptable engineering range, and will not affect the accuracy of the risk assessment.

[0074] Preferably, the time-series risk value of each risk assessment scenario in the risk assessment scenario set is solved based on the sensitivity correlation model, including: For each risk assessment scenario in the risk assessment scenario set, the source load power value at the corresponding time is extracted, and the source load power value is subtracted from the source load power benchmark value corresponding to the power grid operating state to obtain the source load power disturbance value. The source-load power disturbance value is input into the sensitivity correlation model, and the node voltage operation value and branch power flow operation value at each time are obtained based on the first linear correlation relationship and the second linear correlation relationship. Based on the node voltage operating value and the preset distribution network node voltage safety threshold, the voltage over-limit risk value is calculated; based on the branch power flow operating value and the preset branch rated current carrying capacity threshold, the branch power flow overload risk value is calculated; based on the node voltage over-limit state corresponding to the voltage over-limit risk value, the branch power flow over-limit state corresponding to the branch power flow overload risk value, and the preset resource regulation margin, the regulation capacity shortage risk value is determined. The voltage over-limit risk value, branch power flow overload risk value, and regulation capacity shortage risk value are used as time-series risk values.

[0075] Specifically, the source-load power baseline value is the output and load power of distributed generation at each node under baseline operating conditions, serving as a reference benchmark for calculating power disturbances. The source-load power disturbance value is the difference between the source-load power in the risk assessment scenario and the baseline power, i.e., the power fluctuation. The node voltage operating value is the actual value of the node voltage after the disturbance. The branch power flow operating value is the actual value of the branch current / power after the disturbance. The voltage over-limit risk value represents a risk indicator of the probability and severity of node voltage deviating from the safe range, and is a component of the time-series risk value. The branch power flow overload risk value represents a risk indicator of the probability and severity of branch current exceeding the rated current carrying capacity, and is a component of the time-series risk value. The resource regulation margin represents the maximum regulation capacity of all adjustable resources in the distribution network, such as on-load tap-changing transformers, distributed generation reactive power regulation, capacitor banks, etc., and is the basis for judging whether the regulation capacity is insufficient. The regulation capacity shortage risk value represents a risk indicator of the probability and severity of the power system's adjustable resources being unable to eliminate operating over-limits, reflecting the risk of system runaway under extreme disturbances.

[0076] Schematic, for each risk assessment scenario in the risk assessment scenario set, the source load power value at each moment of that scenario is extracted, and the difference between this value and the source load power benchmark value corresponding to the baseline operating state is calculated to obtain the source load power disturbance value at each moment. The source load power disturbance value at each moment is input into the sensitivity correlation model, and the node voltage change is calculated through the first linear correlation relationship. After superimposing the benchmark value, the node voltage operating value at each moment is obtained. The branch power flow change is calculated through the second linear correlation relationship, and after superimposing the benchmark value, the branch power flow operating value at each moment is obtained.

[0077] Based on the obtained node voltage operating values ​​and node voltage safety thresholds, the voltage over-limit risk value for each scenario throughout the entire time period is calculated, comprehensively reflecting the probability and severity of over-limit. Based on the branch power flow operating values ​​and branch rated current carrying capacity thresholds, the branch power flow overload risk value is calculated. Then, combined with the voltage over-limit status, power flow over-limit status, and the system active and reactive power resource regulation margins determined by the distribution network configuration parameters, the regulation capacity shortage risk value is calculated.

[0078] By integrating the voltage over-limit risk value, branch power flow overload risk value, and regulation capacity shortage risk value corresponding to the entire time period of a single scenario, a complete time-series risk value for that scenario is formed, which fully covers the three dimensions of voltage safety, power flow safety, and regulation capacity.

[0079] In a preferred embodiment of the present invention, the voltage exceedance risk value is used to assess the probability and severity of node voltage deviations from the legally permissible safety range under distributed generation output fluctuations in the power system. Its secondary indicators are as follows: ① Nodal voltage exceedance severity function: In the formula, Let be the voltage value of node i at time t in scenario s; , These are the upper and lower safety constraint thresholds for voltage operation.

[0080] ② Dynamic voltage risk indicators: In the formula, Let be the probability of scenario s; This is a 0-1 flag; if an out-of-limit event occurs, the value is 1; otherwise, it is 0. For the set of simulation time periods; N represents the total number of simulation periods; N represents the total number of system nodes.

[0081] The branch power flow overload risk value addresses line overload issues caused by reverse power flow from distributed generation sources. Its secondary indicators are as follows: ① Power flow overload severity function: In the formula, Let be the current amplitude of branch j in scenario s at time t; This is the maximum allowable steady-state current carrying capacity of this branch.

[0082] ②Comprehensive tidal current overload risk In the formula, L is the set of system branches; This is a flag for determining branch circuit overload.

[0083] The risk value of insufficient regulation capacity represents the risk of insufficient response margin of the system's flexible resources in response to sudden power fluctuations from distributed sources. Its expression is as follows: In the formula, This indicates that in scenario s, even after using all active / reactive power regulation resources across the entire network, the total number of lines or nodes that exceed voltage or power flow limits cannot be eliminated. This represents the total number of distribution network lines. This indicator reflects the probability of the system running out of control under extreme disturbances.

[0084] In a preferred embodiment of the present invention, to balance the differences between subjective experience and objective data, a game theory-based weighting method is used to determine the weights of each indicator. A discriminant matrix for pairwise comparisons of the indicators is constructed using the Analytic Hierarchy Process (AHP), and the importance weights of each indicator are determined based on expert experience. Principal component analysis (PCA) is used to reduce the dimensionality of multi-scenario simulation data, and objective weights are automatically calculated based on the variance contribution rate of each indicator. .

[0085] Seeking the optimal weight vector , making The sum of squared deviations from each basic weight is minimized: Solve the corresponding system of linear equations using the Lagrange multiplier method: In the formula, , These are the combination coefficients after the game. Normalizing the coefficients yields... The final comprehensive risk score is calculated as follows: In the formula, These are the risk values ​​for each dimension calculated above.

[0086] This embodiment covers three dimensions: voltage over-limit, power flow overload, and insufficient regulation capacity. It not only examines whether there are over-limit issues, but also whether the over-limit issues can be absorbed through regulation. This is more in line with the actual management needs of the distribution network operation, and the evaluation results are more valuable.

[0087] S6. Perform weighted fusion based on the time-series risk values ​​to obtain a comprehensive risk score, and conduct a risk assessment based on the comprehensive risk score.

[0088] Specifically, the comprehensive risk score represents the overall risk quantification result obtained by weighting and fusing the time-series risk values ​​of each scenario, and is used to comprehensively evaluate the overall operational risk level of the distribution network.

[0089] Schematic, the risk weights are first determined using a game theory-based combined weighting method: subjective weights are determined using the analytic hierarchy process (AHP) combined with dispatching expert experience, and objective weights are determined using principal component analysis (PCA) based on scenario simulation data. Then, using these two types of weights as the game players, the Nash equilibrium is solved to obtain a combined weight vector that minimizes the sum of squared deviations. Finally, combining the occurrence probability of each risk assessment scenario with the combined weight vector, the time-series risk values ​​of all scenarios are weighted and fused to calculate the comprehensive risk score of the distribution network. Based on this comprehensive risk score, the dynamic risk assessment and level determination of the distribution network are completed.

[0090] Existing technologies either prioritize computational efficiency at the expense of extreme risk identification capabilities, or retain a large number of samples to cover extreme scenarios, resulting in low computational efficiency. Furthermore, risk modeling and solution methods are relatively crude. This invention adopts a reduction approach, preserving the statistical distribution characteristics of normal operating scenarios while avoiding the accidental deletion of low-probability, high-impact extreme conditions. This solves the problem of overly optimistic evaluation results in traditional methods, significantly improving the accuracy of extreme risk identification. By drastically compressing the scenario scale through scenario reduction, and combining it with a sensitivity correlation model to replace power flow iteration, costs are reduced simultaneously from both the perspectives of scenario scale and computational load per scenario. The computational efficiency is an order of magnitude higher than traditional full-scenario power flow simulation, meeting the timeliness requirements of online early warning. A complete chain is formed from data input to scenario generation, scenario reduction, risk calculation, and result output. Combined with game theory-based weighting, subjective experience and objective data are balanced, resulting in higher confidence in the evaluation results and providing more reliable decision-making basis for dispatching and operation personnel.

[0091] By implementing this embodiment, a scenario set is first constructed based on historical time-series data of the source load and preset spatiotemporal constraints to obtain an enhanced scenario set. The sample size of extreme scenarios is then expanded based on historical time-series data of the source load, increasing the cardinality of extreme scenarios in the overall scenario set and alleviating the problem of extreme sample scarcity from the data source level. Then, a hybrid scenario reduction strategy is adopted, aiming to minimize the probability distribution deviation of the enhanced scenario set. Redundant scenarios are recursively eliminated based on the probability distribution distance between each enhanced scenario in the enhanced scenario set. The occurrence probability corresponding to the redundant scenarios is allocated to the nearest neighbor retained scenario, resulting in a statistically typical scenario set. Unlike traditional methods that rely solely on a single probability distribution distance to filter and delete samples, this method no longer uses probability percentage as the sole criterion for scenario selection, ensuring that low-probability, high-risk extreme scenarios are not directly eliminated due to low distribution weights. Next, based on preset scenario risk measurement indicators, scenarios with scenario risk measurement indicators greater than preset scenario thresholds are selected from the enhanced scenario set as extreme risk scenarios. A union calculation is then performed on these extreme risk scenarios and the statistical typical scenario set to obtain a risk assessment scenario set. This risk assessment scenario set integrates extreme risk scenarios and the statistical typical scenario set, accurately capturing potential extreme operational risks in the distribution network and addressing the industry pain point of overly optimistic assessment results, fundamentally improving the accuracy of dynamic risk assessment results. Furthermore, in conjunction with a sensitivity correlation model and the distribution network operating status, time-series risk values ​​are calculated. From expanding extreme samples during the generation stage of the statistical typical scenario set, to locking in extreme risk scenarios during the reduction stage, and then to the integration of scenario risk measurement indicators, the risk information corresponding to extreme operating conditions is fully transmitted to the final comprehensive risk score. This allows the risk assessment to objectively reflect all operational risks of the distribution network, including extreme disturbances, improving the accuracy of risk assessment results and providing accurate risk decision-making information for distribution network dispatching and operation.

[0092] See Figure 3This is a schematic diagram of the structure of a dynamic risk assessment device for distribution networks based on scenario reduction, provided in an embodiment of the present invention, comprising: The data acquisition module is used to acquire the distribution network configuration parameters and source-load historical time-series data of the distribution network to be evaluated. The enhanced scene set construction module is used to construct a scene set based on the source load historical time series data and preset source load spatiotemporal constraints, thereby obtaining an enhanced scene set. The scene reduction module is used to minimize the probability distribution deviation of the reduction enhancement scene set. Based on the probability distribution distance between each enhancement scene in the enhancement scene set, redundant scenes are recursively eliminated. The occurrence probability of the redundant scene is assigned to the corresponding nearest neighbor retained scene to obtain a statistical typical scene set. The risk assessment scenario set construction module is used to select scenarios whose scenario risk measurement indicators are greater than preset scenario thresholds from the enhanced scenario set based on preset scenario risk measurement indicators, as extreme risk scenarios, and perform union calculation based on the extreme risk scenarios and the statistical typical scenario set to obtain the risk assessment scenario set. The time-series risk value solving module is used to determine the operating status of the distribution network based on the distribution network configuration parameters, construct a sensitivity correlation model to characterize the relationship between node power and node voltage and branch power flow based on the distribution network operating status, and solve the time-series risk value of each risk assessment scenario in the risk assessment scenario set based on the sensitivity correlation model. The power grid risk assessment module is used to perform weighted fusion based on the time-series risk values ​​to obtain a comprehensive risk score, and to conduct risk assessment based on the comprehensive risk score.

[0093] This invention provides a dynamic risk assessment device for distribution networks based on scenario reduction. First, a scenario set is constructed based on historical time-series data of source and load and preset spatiotemporal constraints, resulting in an enhanced scenario set. The sample size of extreme scenarios is expanded based on historical time-series data of source and load, increasing the cardinality of extreme scenarios in the overall scenario set and alleviating the problem of extreme sample scarcity from the data source level. Then, a hybrid scenario reduction strategy is adopted, aiming to minimize the probability distribution deviation of the enhanced scenario set. Redundant scenarios are recursively eliminated based on the probability distribution distance between each enhanced scenario in the enhanced scenario set. The occurrence probability corresponding to the redundant scenario is allocated to the corresponding nearest neighbor retained scenario, resulting in a statistically typical scenario set. Unlike traditional methods that rely solely on a single probability distribution distance to filter and delete samples, this method no longer uses probability proportion as the sole criterion for scenario selection, ensuring that low-probability, high-risk extreme scenarios are not directly eliminated due to low distribution weight. Next, based on preset scenario risk measurement indicators, scenarios with scenario risk measurement indicators greater than preset scenario thresholds are selected from the enhanced scenario set as extreme risk scenarios. A union calculation is then performed on these extreme risk scenarios and the statistical typical scenario set to obtain a risk assessment scenario set. This risk assessment scenario set integrates extreme risk scenarios and the statistical typical scenario set, accurately capturing potential extreme operational risks in the distribution network and addressing the industry pain point of overly optimistic assessment results, fundamentally improving the accuracy of dynamic risk assessment results. Furthermore, in conjunction with a sensitivity correlation model and the distribution network operating status, time-series risk values ​​are calculated. From expanding extreme samples during the generation stage of the statistical typical scenario set, to locking in extreme risk scenarios during the reduction stage, and then to the integration of scenario risk measurement indicators, the risk information corresponding to extreme operating conditions is fully transmitted to the final comprehensive risk score. This allows the risk assessment to objectively reflect all operational risks of the distribution network, including extreme disturbances, improving the accuracy of risk assessment results and providing accurate risk decision-making information for distribution network dispatching and operation.

[0094] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0095] Those skilled in the art will understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0096] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a dynamic risk assessment method for power distribution networks based on scenario reduction as described in the above embodiments. The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0097] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0098] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0099] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the dynamic risk assessment method for distribution networks based on scenario reduction described in the above embodiment.

[0100] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0101] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A dynamic risk assessment method for distribution networks based on scenario reduction, characterized in that, include: Obtain the distribution network configuration parameters and historical time-series data of the source and load of the distribution network to be evaluated; An enhanced scene set is obtained by constructing a scene set based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load; With the goal of minimizing the probability distribution deviation of the enhanced scene set, redundant scenes are recursively eliminated based on the probability distribution distance between each enhanced scene in the enhanced scene set. The occurrence probability of the redundant scene is then assigned to the nearest neighboring retained scene to obtain a statistical typical scene set. Based on the preset scenario risk measurement index, scenarios with scenario risk measurement indices greater than the preset scenario threshold are selected from the enhanced scenario set as extreme risk scenarios. Then, the extreme risk scenarios and the statistical typical scenario set are combined to obtain the risk assessment scenario set. The distribution network operation status is determined based on the distribution network configuration parameters. A sensitivity correlation model is constructed based on the distribution network operation status to characterize the relationship between node power, node voltage, and branch power flow. The time-series risk value of each risk assessment scenario in the risk assessment scenario set is solved based on the sensitivity correlation model. The time-series risk values ​​are weighted and fused to obtain a comprehensive risk score, and a risk assessment is performed based on the comprehensive risk score.

2. The method for dynamic risk assessment of distribution networks based on scenario reduction as described in claim 1, characterized in that, The historical time-series data of the source and load includes: historical output time-series data of distributed power sources and historical operation time-series data of the loads; An enhanced scene set is constructed based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load, including: A first marginal probability distribution model of distributed power source power fluctuation is constructed based on the historical output time series data, and a second marginal probability distribution model of load power fluctuation is constructed based on the historical operating time series data. Based on the first marginal probability distribution model, the second marginal probability distribution model, and the preset source load spatiotemporal constraints, a time-series initial scenario set is generated; Extreme operating features are extracted based on the historical output time series data and the historical operating time series data. Extreme operating condition samples are generated based on the extreme operating features. The extreme operating condition samples are added to the initial time series scenario set to obtain the enhanced scenario set.

3. The method for dynamic risk assessment of distribution networks based on scenario reduction as described in claim 2, characterized in that, Based on the first marginal probability distribution model, the second marginal probability distribution model, and the preset source load spatiotemporal constraints, a time-series initial scene set is generated, including: Based on the preset source-load spatiotemporal constraints, the spatiotemporal correlation matrix between distributed power sources and loads is determined; Based on the Nataf transform, the non-normal random variables corresponding to the first marginal probability distribution model and the second marginal probability distribution model are mapped to the standard normal space. At the same time, the spatiotemporal correlation matrix is ​​positive definite to obtain the effective correlation matrix in the standard normal space. Under the constraint of the effective correlation matrix, random samples in the standard normal space are generated by Latin hypercube sampling. The random samples are then inversely transformed and mapped back to the original power space to obtain the initial time-series scene set.

4. The method for dynamic risk assessment of distribution networks based on scenario reduction as described in claim 1, characterized in that, With the goal of minimizing the probability distribution deviation of the enhanced scene set, redundant scenes are recursively eliminated based on the probability distribution distance between each enhanced scene in the enhanced scene set. The occurrence probability of the redundant scenes is then assigned to the corresponding nearest neighbor retained scenes, resulting in a statistically typical scene set, including: Calculate the spatiotemporal geometric distance between each enhanced scene in the enhanced scene set, and calculate the probability distribution distance based on the preset occurrence probability and the spatiotemporal geometric distance; With the goal of minimizing the probability distribution deviation of the reduced augmented scene set, the augmented scene with the smallest probability distribution distance is recursively removed by synchronous backtracking method. The augmented scene to be removed is regarded as a redundant scene, and the augmented scene that is not removed is regarded as a retained scene. The occurrence probabilities of redundant scenarios are redistributed to the retained scenarios with the closest spatiotemporal geometric distance to obtain a set of statistically typical scenarios. The probability distribution bias is determined based on the difference in probability distribution between the enhanced scene set before reduction and the statistically typical scene set after reduction.

5. The method for dynamic risk assessment of distribution networks based on scenario reduction as described in claim 1, characterized in that, Based on preset scenario risk measurement indicators, scenarios with scenario risk measurement indicators greater than preset scenario thresholds are selected from the enhanced scenario set as extreme risk scenarios, including: Using the preset distribution network node voltage safety threshold and branch rated current carrying capacity threshold as safety constraint boundaries, a scenario risk measurement index is constructed to characterize the severity of scenario operation exceeding limits. For each enhanced scenario in the enhanced scenario set, the degree to which the node voltage and branch power flow deviate from the safety constraint boundary corresponding to the scenario source load power are calculated at each time step to obtain the risk measurement value at each time step, and the maximum risk measurement value is used as the scenario risk measurement index value. Enhanced scenarios where the scenario risk measurement index value is greater than a preset scenario threshold are classified as extreme risk scenarios.

6. The method for dynamic risk assessment of distribution networks based on scenario reduction as described in claim 1, characterized in that, The power distribution network configuration parameters include: network topology data, branch electrical data, and node source-load configuration data; The distribution network operating status is determined based on the distribution network configuration parameters. A sensitivity correlation model is constructed based on the distribution network operating status to characterize the relationship between node power, node voltage, and branch power flow, including: Based on the network topology data, the branch electrical data, and the node source-load configuration data, the operating status of the distribution network is determined, and the corresponding node voltage reference value and branch power flow reference value are determined based on the operating status of the distribution network. The distribution network Jacobian matrix is ​​obtained by performing a first-order Taylor expansion of the preset power balance equations of the distribution network nodes at the operating state of the distribution network. By inverting the Jacobian matrix of the distribution network, and based on the node voltage reference value and the branch power flow reference value, the first linear correlation between the node power injection change and the node voltage change, and the second linear correlation between the node power injection change and the branch power flow change are derived. A sensitivity correlation model is constructed based on the first linear correlation and the second linear correlation.

7. The method for dynamic risk assessment of distribution networks based on scenario reduction as described in claim 6, characterized in that, The time-series risk values ​​for each risk assessment scenario in the risk assessment scenario set are calculated based on the sensitivity correlation model, including: For each risk assessment scenario in the risk assessment scenario set, the source load power value at the corresponding time is extracted, and the source load power value is subtracted from the source load power benchmark value corresponding to the power grid operating state to obtain the source load power disturbance value. The source-load power disturbance value is input into the sensitivity correlation model, and the node voltage operation value and branch power flow operation value at each time are obtained based on the first linear correlation relationship and the second linear correlation relationship. Based on the node voltage operating value and the preset distribution network node voltage safety threshold, the voltage over-limit risk value is calculated; based on the branch power flow operating value and the preset branch rated current carrying capacity threshold, the branch power flow overload risk value is calculated; based on the node voltage over-limit state corresponding to the voltage over-limit risk value, the branch power flow over-limit state corresponding to the branch power flow overload risk value, and the preset resource regulation margin, the regulation capacity shortage risk value is determined. The voltage over-limit risk value, branch power flow overload risk value, and regulation capacity shortage risk value are used as time-series risk values.

8. A dynamic risk assessment device for distribution networks based on scenario reduction, characterized in that, include: The data acquisition module is used to acquire the distribution network configuration parameters and historical time-series data of the source and load of the distribution network to be evaluated. The enhanced scene set construction module is used to construct a scene set based on the historical time-series data of the source load and the preset spatiotemporal constraints of the source load, thereby obtaining an enhanced scene set. The scene reduction module is used to minimize the probability distribution deviation of the reduction enhancement scene set. Based on the probability distribution distance between each enhancement scene in the enhancement scene set, redundant scenes are recursively eliminated. The occurrence probability of the redundant scene is assigned to the corresponding nearest neighbor retained scene to obtain a statistical typical scene set. The risk assessment scenario set construction module is used to select scenarios whose scenario risk measurement indicators are greater than preset scenario thresholds from the enhanced scenario set based on preset scenario risk measurement indicators, as extreme risk scenarios, and perform union calculation based on the extreme risk scenarios and the statistical typical scenario set to obtain the risk assessment scenario set. The time-series risk value solving module is used to determine the operating status of the distribution network based on the distribution network configuration parameters, construct a sensitivity correlation model to characterize the relationship between node power and node voltage and branch power flow based on the distribution network operating status, and solve the time-series risk value of each risk assessment scenario in the risk assessment scenario set based on the sensitivity correlation model. The power grid risk assessment module is used to perform weighted fusion based on the time-series risk values ​​to obtain a comprehensive risk score, and to conduct risk assessment based on the comprehensive risk score.

9. A terminal device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a dynamic risk assessment method for distribution networks based on scenario reduction as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a dynamic risk assessment method for distribution networks based on scenario reduction as described in any one of claims 1 to 7.

Images