A confidence capacity-based power system weak link identification method and system

Abstract

This invention relates to the field of power system weakness identification technology, and in particular to a method and system for identifying power system weaknesses based on confidence capacity. This invention constructs a photovoltaic (PV) output probability prediction model based on quantile regression to achieve probabilistic prediction of PV power, effectively quantifying the uncertainty of distributed renewable energy output. Utilizing the net load definition, it achieves accurate identification of future risk periods. Furthermore, it employs a confidence capacity assessment method based on equivalent reliable capacity, calculating the confidence capacity of renewable energy output to support the system load during risk periods by combining the expectation, variance, and covariance with the load of the PV output probability prediction distribution in distributed PV power plants. Finally, this invention introduces a least-squares support vector machine classifier model to identify weaknesses based on indicators such as confidence capacity, voltage over-limit, and transmission capacity over-limit, improving the accuracy of power system weakness identification.

Description

Technical Field

[0001] This invention relates to the field of power system vulnerability identification technology, and in particular to a method and system for identifying power system vulnerabilities based on confidence capacity. Background Technology

[0002] With the transformation and upgrading of the global energy structure, the utilization rate of distributed solar photovoltaic (PV) power generation in distribution networks is continuously increasing. Its output prediction and system reliability assessment are becoming increasingly important for power grid dispatch and renewable energy consumption. In the field of distributed PV output prediction, existing mainstream technologies typically include physical modeling methods and statistical learning methods. Physical modeling methods rely on the physical characteristic equations of PV cells and numerical weather prediction data, requiring extremely high accuracy of input parameters and involving cumbersome calculations. Statistical learning methods utilize historical data to build regression models, but they have limitations when dealing with nonlinear and highly volatile data. However, distributed PV output is affected by local meteorological factors such as irradiance, temperature, and humidity, as well as multi-scale environmental changes, exhibiting significant randomness and volatility. Traditional physical modeling and statistical learning methods struggle to accurately quantify this uncertainty and cannot effectively capture the spatiotemporal correlation characteristics between different distributed PV sites, resulting in limited accuracy in PV output prediction and failing to meet the practical needs of new power systems for uncertainty quantification and refined dispatch.

[0003] In identifying weak links in the power system, existing technologies largely rely on deterministic photovoltaic (PV) output predictions and simple threshold judgments. This approach assumes a fixed future operating scenario and ignores the impact of the dynamic probabilistic characteristics of renewable energy output changing with weather conditions and system state on system stability. With the large-scale grid connection of solar PV power generation, system operating states are becoming increasingly complex and volatile. Traditional weak link assessment methods fail to fully account for the probabilistic distribution characteristics of PV output, resulting in inaccurate identification of power system weak links. They cannot effectively identify weak links by considering periods of insufficient or drastic fluctuations in renewable energy output, thus hindering the improvement of grid security defense and risk early warning capabilities. Summary of the Invention

[0004] This invention aims to provide a method and system for identifying weak links in a power system based on confidence capacity. It aims to effectively identify weak links in a power system by introducing dynamic evaluation indicators that can take into account risk periods of insufficient or drastic fluctuations in renewable energy output. This addresses the technical problem that existing technologies fail to fully consider the probability distribution characteristics of photovoltaic output, resulting in inaccurate identification of weak links in the power system.

[0005] To achieve the above objectives, the first aspect of the present invention provides a method for identifying weak links in a power system based on confidence capacity, comprising the following steps: Acquire day-ahead power output data and day-ahead meteorological data of distributed photovoltaic power stations in the power system; Construct a photovoltaic power output probability prediction model based on quantile regression; The photovoltaic output probability prediction model is input into the daytime power output data and the daytime meteorological data to obtain the photovoltaic output probability prediction distribution. Obtain the load forecast results of the power system, and then predict the future risk periods of the power system based on the load forecast results; Based on the photovoltaic output probability prediction distribution and the load prediction results, the confidence capacity of the distributed photovoltaic power station during the future risk period is obtained. Based on the photovoltaic output probability prediction distribution, the future operation scenarios of the power system are obtained, and then the vulnerability assessment index of the future operation scenarios is obtained. Construct a least-squares support vector machine model for each node and each line in the power system; The future risk period, the confidence capacity, and the vulnerability assessment index are input into the least squares support vector machine model to obtain the vulnerability identification results for each node and each line in the power system.

[0006] The aforementioned method for identifying weak links in power systems based on confidence capacity firstly constructs a photovoltaic (PV) output probability prediction model based on quantile regression. This model fully leverages the spatiotemporal characteristics of historical data and outputs the probability distribution of PV output, effectively quantifying the randomness and uncertainty of PV output and overcoming the limitations of traditional physical modeling and statistical learning methods in handling nonlinear and highly volatile data. Secondly, by predicting future risk periods and calculating the confidence capacity of distributed PV power stations during those periods, the method combines the probabilistic characteristics of PV output with system load demand. This allows for the assessment of the actual contribution capacity of distributed PV power stations in future operating scenarios, avoiding the evaluation bias caused by relying solely on deterministic predictions in traditional methods. Confidence capacity refers to the capacity that can reliably meet load demand while considering the probabilistic characteristics of PV output, providing boundary conditions that better reflect actual operating conditions for subsequent weak link assessments. Finally, by constructing a least-squares support vector machine model, the complex mapping relationship between future risk periods, confidence capacity, and vulnerability assessment indicators is comprehensively learned. This allows for the accurate identification of vulnerable nodes and lines in the power system under future operating scenarios, achieving rapid and accurate identification of weak links in the power system. This solves the problem of insufficient identification accuracy caused by existing technologies failing to fully consider the probability distribution characteristics of photovoltaic output. Compared to existing assessment methods that rely on deterministic criteria, this invention ensures that the vulnerability identification process fully considers the random fluctuation characteristics of photovoltaic output and the impact of risk periods, significantly improving the power grid's security defense and risk early warning capabilities in the face of new energy fluctuations.

[0007] Furthermore, the construction of the photovoltaic power output probability prediction model based on quantile regression includes: Historical power output data and historical meteorological data of distributed photovoltaic power stations in the power system are obtained, and then a prediction training set and a prediction validation set are constructed based on the historical power output data and the historical meteorological data. Construct a convolutional neural network model so that the convolutional neural network model outputs photovoltaic power output features based on the prediction training set; A long short-term memory neural network model is constructed so that the long short-term memory neural network model can make deterministic predictions of photovoltaic output based on the photovoltaic output characteristics, thereby obtaining the photovoltaic output prediction results; A quantile regression model is constructed so that the quantile regression model obtains several sub-predicted values ​​at different quantile levels based on the photovoltaic power output prediction results, and then obtains the photovoltaic power output probability prediction distribution based on the several sub-predicted values; Obtain the quantile loss function of the photovoltaic output probability prediction distribution, and obtain the continuous hierarchical probability score between the photovoltaic output probability prediction distribution and the prediction validation set; The convolutional neural network model, the long short-term memory neural network model, and the quantile regression model are used to construct the photovoltaic power output probability prediction model, and the probability prediction loss function of the photovoltaic power output probability prediction model is constructed based on the quantile loss function and the continuous hierarchical probability score.

[0008] This implementation proposes a method for constructing a probabilistic prediction model using a hybrid neural network and quantile regression. First, a convolutional neural network (CNN) model is used to extract photovoltaic (PV) output features, leveraging the advantages of CNNs in processing multidimensional data and extracting nonlinear features to effectively capture the complex mapping relationship between meteorological factors and PV output. Then, the advantages of long short-term memory (LSTM) neural networks in processing time-series data are utilized to model the temporal evolution of PV output, obtaining a preliminary deterministic prediction result. Finally, a quantile regression model is introduced. Based on the deterministic prediction result, the quantile regression model obtains predicted values ​​at different quantile levels, extending the deterministic prediction to a probabilistic prediction, thereby obtaining the complete probability distribution of PV output.

[0009] To ensure prediction accuracy, this invention not only employs the traditional quantile loss function to constrain the prediction error at each quantile, but also introduces the Continuous Graded Probability Score (CRPS) as an evaluation index for the overall probability prediction quality. A joint loss function is then constructed based on these two metrics for model training. This process enables the model to simultaneously optimize both the local accuracy of quantile predictions and the global fit of the entire probability density curve, effectively improving the accuracy and reliability of photovoltaic power output probability prediction and solving the technical problem that traditional prediction methods cannot quantify the uncertainty range of predictions.

[0010] Furthermore, the step of obtaining the load forecast results of the power system and then predicting future risk periods of the power system based on the load forecast results includes: To obtain the inherent capacity adequacy level of the power system; Based on the load forecast results and the inherent capacity adequacy level, obtain the net load value for each time node in the preset future period; All time points in the timeframe where the net load value is greater than or equal to zero are considered as the future risk period.

[0011] In this implementation, the net load value is calculated by introducing an inherent capacity adequacy level and combining it with load forecast results. This net load value reflects the load gap that the system still needs to meet by relying on photovoltaic power output after deducting its own reliable power supply capacity. When the net load value is greater than or equal to zero, it means that the system's own power supply capacity is insufficient to meet all load demands, and it must rely on renewable energy output. If the renewable energy output is insufficient or fluctuates too much due to weather conditions at this time, the system is very prone to power supply gaps or operational risks, which can intuitively reflect the supply and demand balance of the system under the support of conventional units. This invention defines the time node when the net load value is greater than or equal to zero as the future risk period, accurately locking in the key time window when the system's power supply capacity may be insufficient and its dependence on photovoltaic output is high. This process focuses the subsequent weak link identification work on the most critical risk operation period, avoiding the waste of resources caused by indiscriminate analysis and identification throughout the entire time period. It not only reduces the amount of data processing for subsequent calculations, but also makes the identification of weak links more targeted, improving the targeting and efficiency of weak link identification.

[0012] Further, obtaining the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results includes: Obtain the expected output value and output variance of the photovoltaic power output probability prediction distribution within the future risk period, and obtain the expected load value of the load prediction result within the future risk period; Obtain the covariance between the expected output value and the expected load value; The confidence capacity of the distributed photovoltaic power station during the future risk period is calculated based on the expected output value, the output variance, the expected load value, and the covariance.

[0013] In this implementation, the present invention proposes a method for calculating dynamic confidence capacity. Traditional confidence capacity is typically a static value based on long-term statistics, while this invention focuses on previously identified future risk periods. By calculating the expected value and variance of photovoltaic (PV) output, as well as the expected value of load, and further solving for their covariance, the correlation between PV output and load demand during risk periods can be quantified. This confidence capacity fully considers the effective support capacity that PV power can provide during the future risk periods when the system most needs electricity. Using this dynamic confidence capacity as input to a subsequent vulnerability identification model allows the model to perceive the changes in the strength of new energy's support capacity for the system during different risk periods, thereby more accurately assessing its impact on the system's vulnerability. This overcomes the limitations of traditional methods that treat PV output as a constant value or simply reduce it, improving the accuracy and scientific rigor of the assessment results.

[0014] Furthermore, based on the photovoltaic output probability prediction distribution, the future operating scenarios of each node and each line of the power system are obtained, and then the vulnerability assessment indicators of the future operating scenarios are obtained, including: Based on the future operating scenario, the number of nodes in the power system that experience voltage overruns during the future risk period is obtained, and the total number of voltage overruns in the power system during the future risk period is obtained. A first evaluation index is obtained based on the number of nodes and the total number of voltage over-limit occurrences. Based on the future operating scenario, the number of power system lines that exceed the transmission capacity limit during the future risk period is obtained, and the total number of power system transmission capacity exceedances during the future risk period is obtained. A second evaluation index is obtained based on the number of lines and the total number of transmission capacity overruns. The first evaluation index and the second evaluation index are used as the weakness evaluation index.

[0015] In this implementation, the system state during future risk periods is simulated within the power system output plan and future operating scenarios generated by the photovoltaic output probability distribution. The first evaluation metric focuses on voltage quality. By statistically analyzing the number of nodes experiencing voltage exceedances and the total number of exceedance events, the system's weakness in voltage stability can be measured. A higher number and more frequent exceedances indicate weaker node voltage support capabilities. The second evaluation metric focuses on transmission capacity. By statistically analyzing the number of lines experiencing transmission capacity exceedances and the total number of exceedance events, the system's weakness in power flow transmission can be measured. Similarly, a higher number indicates a more prominent network transmission bottleneck. Combining these two metrics comprehensively covers the safety constraints of the two key dimensions of node voltage and line power flow, intuitively and quantitatively characterizing the system's weaknesses under different future operating scenarios, and providing representative feature variables for the training of subsequent machine learning models.

[0016] Furthermore, the construction of the least squares support vector machine model for each node and each line of the power system includes: To obtain the failure rate and repair rate of the power system; The normal operation probability density function of the power system is obtained based on the failure rate, and the fault repair probability density function of the power system is obtained based on the repair rate. A first sampling formula for the duration of normal operation of the power system is obtained based on the normal operation probability density function, and a second sampling formula for the fault repair time of the power system is obtained based on the fault repair probability density function. Combining the photovoltaic output probability prediction distribution, the first sampling formula, and the second sampling formula, a power system state sequence is generated using the sequential Monte Carlo simulation method, and then several fault scenarios are generated based on the power system state sequence. The least squares support vector machine model is generated based on several of the aforementioned fault scenarios.

[0017] In this implementation, a probabilistic model describing equipment operation and repair time is established by introducing equipment failure rate and repair rate. Using the first and second sampling formulas, sequential sampling can be performed on the entire future operation-repair process of the equipment, generating a system state sequence containing the time dimension. Combining this sequence with the photovoltaic output probability prediction distribution generates more realistic composite fault scenarios that simultaneously consider the randomness of new energy sources and random equipment failures. Using these highly realistic fault scenarios to train a least-squares support vector machine model allows the model to fully learn which nodes or lines in the system are most likely to become weak links under the superposition of multiple uncertainties. This training method greatly enhances the model's generalization ability and prediction accuracy for complex operating conditions, making the final weakness identification results more practically instructive.

[0018] A second aspect of the present invention provides a power system weak link identification system based on confidence capacity, comprising: The data acquisition module is used to acquire day-ahead power output data and day-ahead meteorological data of distributed photovoltaic power stations in the power system; The model building module is used to build a photovoltaic output probability prediction model based on quantile regression, and to build a least squares support vector machine model for each node and each line in the power system. The photovoltaic output probability prediction module is used to input the daytime output data and the daytime meteorological data into the photovoltaic output probability prediction model to obtain the photovoltaic output probability prediction distribution. The risk period prediction module is used to obtain the load prediction results of the power system, and then predict the future risk periods of the power system based on the load prediction results; The confidence capacity assessment module is used to obtain the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results. The weakness index analysis module is used to obtain the future operation scenario of the power system based on the photovoltaic output probability prediction distribution, and then obtain the weakness assessment index of the future operation scenario. The weak link identification module is used to input the future risk period, the confidence capacity, and the weakness assessment index into the least squares support vector machine model, so as to obtain the weakness identification results of each node and each line in the power system.

[0019] Furthermore, the construction of the photovoltaic power output probability prediction model based on quantile regression includes: Historical power output data and historical meteorological data of distributed photovoltaic power stations in the power system are obtained, and then a prediction training set and a prediction validation set are constructed based on the historical power output data and the historical meteorological data. Construct a convolutional neural network model so that the convolutional neural network model outputs photovoltaic power output features based on the prediction training set; A long short-term memory neural network model is constructed so that the long short-term memory neural network model can make deterministic predictions of photovoltaic output based on the photovoltaic output characteristics, thereby obtaining the photovoltaic output prediction results; A quantile regression model is constructed so that the quantile regression model obtains several sub-predicted values ​​at different quantile levels based on the photovoltaic power output prediction results, and then obtains the photovoltaic power output probability prediction distribution based on the several sub-predicted values; Obtain the quantile loss function of the photovoltaic output probability prediction distribution, and obtain the continuous hierarchical probability score between the photovoltaic output probability prediction distribution and the prediction validation set; The convolutional neural network model, the long short-term memory neural network model, and the quantile regression model are used to construct the photovoltaic power output probability prediction model, and the probability prediction loss function of the photovoltaic power output probability prediction model is constructed based on the quantile loss function and the continuous hierarchical probability score.

[0020] Furthermore, the step of obtaining the load forecast results of the power system and then predicting future risk periods of the power system based on the load forecast results includes: To obtain the inherent capacity adequacy level of the power system; Based on the load forecast results and the inherent capacity adequacy level, obtain the net load value for each time node in the preset future period; All time points in the timeframe where the net load value is greater than or equal to zero are considered as the future risk period.

[0021] Further, obtaining the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results includes: Obtain the expected output value and output variance of the photovoltaic power output probability prediction distribution within the future risk period, and obtain the expected load value of the load prediction result within the future risk period; Obtain the covariance between the expected output value and the expected load value; The confidence capacity of the distributed photovoltaic power station during the future risk period is calculated based on the expected output value, the output variance, the expected load value, and the covariance.

[0022] The method and system for identifying weak links in a power system based on confidence capacity provided by the present invention have at least the following advantages compared to the prior art: This invention constructs a comprehensive evaluation and optimization system for distributed photovoltaic (PV) power generation, encompassing probabilistic prediction and risk identification. First, a quantile regression-convolutional neural network-long short-term memory neural network (QR-CNN-LSTM) deep learning model is used to predict PV power probabilistically, effectively quantifying the uncertainty of distributed renewable energy output and improving prediction accuracy and reliability. Second, using the net load definition, accurate identification of future risk periods is achieved. Based on this, a confidence capacity assessment method based on equivalent reliable capacity is employed. By combining the expectation, variance, and covariance with the load of the renewable energy output probabilistic prediction results, the confidence capacity of renewable energy output to support the system load during risk periods is quantified. Finally, a least squares support vector machine (LSSVM) classifier model is introduced to identify weak links based on indicators such as confidence capacity, voltage overruns, and transmission capacity overruns. This addresses the lack of independence in traditional weighting methods, improving identification accuracy and efficiency. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a method for identifying weak links in a power system based on confidence capacity, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of a power system weak link identification system based on confidence capacity provided in an embodiment of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that the following detailed descriptions are exemplary and intended to provide further detailed explanation of the invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects, not to describe a particular order.

[0025] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0026] Please refer to Figure 1 In order to effectively identify weak links in the power system by introducing dynamic assessment indicators that can take into account risk periods of insufficient or drastic fluctuations in renewable energy output, and to address the technical problem that existing technologies fail to fully consider the probability distribution characteristics of photovoltaic output, resulting in inaccurate identification of weak links in the power system, the first embodiment of this invention provides a method for identifying weak links in the power system based on confidence capacity, comprising the following steps: S1. Obtain day-ahead power output data and day-ahead meteorological data of distributed photovoltaic power stations in the power system.

[0027] Specifically, historical power data and related meteorological characteristics, such as irradiance, temperature, and humidity, of distributed photovoltaic power stations are collected. The data is then preprocessed, including data cleaning, outlier handling, missing value imputation, data normalization, and dataset partitioning.

[0028] S2. Construct a photovoltaic power output probability prediction model based on quantile regression; S3. Input the daytime power output data and the daytime meteorological data into the photovoltaic power output probability prediction model to obtain the photovoltaic power output probability prediction distribution; S4. Obtain the load forecast results of the power system, and then predict the future risk periods of the power system based on the load forecast results; S5. Based on the photovoltaic output probability prediction distribution and the load prediction results, obtain the confidence capacity of the distributed photovoltaic power station during the future risk period; S6. Based on the photovoltaic output probability prediction distribution, obtain the future operation scenario of the power system, and then obtain the vulnerability assessment index of the future operation scenario; S7. Construct a least-squares support vector machine model for each node and each line in the power system; S8. Input the future risk period, the confidence capacity, and the vulnerability assessment index into the least squares support vector machine model to obtain the vulnerability identification results for each node and each line in the power system.

[0029] The aforementioned method for identifying weak links in power systems based on confidence capacity firstly constructs a photovoltaic (PV) output probability prediction model based on quantile regression. This model fully leverages the spatiotemporal characteristics of historical data and outputs the probability distribution of PV output, effectively quantifying the randomness and uncertainty of PV output and overcoming the limitations of traditional physical modeling and statistical learning methods in handling nonlinear and highly volatile data. Secondly, by predicting future risk periods and calculating the confidence capacity of distributed PV power stations during those periods, the method combines the probabilistic characteristics of PV output with system load demand. This allows for the assessment of the actual contribution capacity of distributed PV power stations in future operating scenarios, avoiding the evaluation bias caused by relying solely on deterministic predictions in traditional methods. Confidence capacity refers to the capacity that can reliably meet load demand while considering the probabilistic characteristics of PV output, providing boundary conditions that better reflect actual operating conditions for subsequent weak link assessments. Finally, by constructing a least-squares support vector machine model, the complex mapping relationship between future risk periods, confidence capacity, and vulnerability assessment indicators is comprehensively learned. This allows for the accurate identification of vulnerable nodes and lines in the power system under future operating scenarios, achieving rapid and accurate identification of weak links in the power system. This solves the problem of insufficient identification accuracy caused by existing technologies failing to fully consider the probability distribution characteristics of photovoltaic output. Compared to existing assessment methods that rely on deterministic criteria, this invention ensures that the vulnerability identification process fully considers the random fluctuation characteristics of photovoltaic output and the impact of risk periods, significantly improving the power grid's security defense and risk early warning capabilities in the face of new energy fluctuations.

[0030] Furthermore, the construction of the photovoltaic power output probability prediction model based on quantile regression includes: Historical power output data and historical meteorological data of distributed photovoltaic power stations in the power system are obtained, and then a prediction training set and a prediction validation set are constructed based on the historical power output data and the historical meteorological data. A convolutional neural network model is constructed so that the convolutional neural network model outputs photovoltaic power output features based on the prediction training set.

[0031] Specifically, in time series prediction tasks, convolutional neural networks (CNNs) can effectively capture the local correlations and dynamic changes in time series data through convolutional operations. They are primarily used to extract feature patterns within local time windows, enhancing the model's ability to perceive short-term time series changes. Since the input data is a time series, one-dimensional convolution (1D-CNN) is first used for feature extraction. The input data dimension is (batch size...). × Number of input features × Input time steps), and then gradually scan the time window through sliding convolution kernels. Each convolution kernel corresponds to a different time pattern detector, which can identify patterns such as short-period fluctuations, abrupt changes, and local trends in photovoltaic power. The output dimension is (batch size). × Number of output features (×output time steps), and finally, the nonlinear feature map is introduced by the activation function ReLU (Rectified Linear Unit), which enables the model to learn more complex nonlinear relationships and enhances its ability to model photovoltaic power mutation phenomena.

[0032] A long short-term memory neural network model is constructed so that the long short-term memory neural network model can make deterministic predictions of photovoltaic output based on the photovoltaic output characteristics, thereby obtaining the photovoltaic output prediction results.

[0033] Specifically, Long Short-Term Memory (LSTM) neural networks are an improved model of Recurrent Neural Networks (RNNs). The core components of an LSTM model include a forget gate, an input gate, an output gate, and a memory cell (CellState).

[0034] Among them, the Gate of Oblivion The determination of which information in the previous memory unit state needs to be retained or discarded, thereby regulating the degree of forgetting in long-term memory, is expressed as follows: In the formula, This represents the sigmoid activation function. , It is the forget gate weight matrix. This is the current input. It is the hidden state from the previous moment. It is the forget gate bias term.

[0035] Input gate The degree of updating of the input information at the current moment determines how much new information is written into the state of the memory unit, and its expression is as follows: In the formula, , It is the input gate weight matrix. It is the input gate bias term. It is a candidate memory unit. , It is the candidate memory unit weight matrix. It is a candidate memory unit bias term.

[0036] Memory cell state As a channel for information transmission, it is used to store and accumulate key information to maintain long-term dependency, and its expression is shown in the following formula: In the formula, It represents the state of the memory unit from the previous moment.

[0037] Output gate The expression for determining how much information to output from the current memory cell state to the hidden state for use in the next time step or subsequent layers is as follows: In the formula, , It is the output gate weight matrix. It is the output gate bias term.

[0038] A quantile regression model is constructed so that the quantile regression model obtains several sub-predicted values ​​at different quantile levels based on the photovoltaic power output prediction results, and then obtains the photovoltaic power output probability prediction distribution based on the several sub-predicted values.

[0039] Obtain the quantile loss function of the photovoltaic output probability prediction distribution, and obtain the continuous hierarchical probability score between the photovoltaic output probability prediction distribution and the prediction validation set.

[0040] The convolutional neural network model, the long short-term memory neural network model, and the quantile regression model are used to construct the photovoltaic power output probability prediction model, and the probability prediction loss function of the photovoltaic power output probability prediction model is constructed based on the quantile loss function and the continuous hierarchical probability score.

[0041] It should be noted that quantile regression (QR) is a regression analysis method that differs from traditional least squares regression. It not only estimates the conditional mean of the dependent variable but also describes the overall distribution of the dependent variable.

[0042] Specifically, in this embodiment, it is set that For the variable that needs to be predicted, treat it as a random variable whose distribution is given by the cumulative distribution function. Given: In the formula, It represents probability.

[0043] For quantile regression prediction, at time t, let the true value of the target time series be... Quantile regression models at different quantile levels Sub-predictions at different quantile levels are given. This leads to the probability prediction distribution of photovoltaic power output. The quantile loss function (QS) for evaluating the regression results is as follows: Quantile regression, through the aforementioned asymmetric linear loss function, enables the prediction of a specified quantile of the target variable from a deterministic prediction.

[0044] In probabilistic prediction, the Continuous Ranked Probability Score (CRPS) is used to evaluate the prediction distribution obtained based on the prediction training set. , and the actual observed values ​​in the corresponding prediction validation set That is, the error between historical photovoltaic power output data. Let the continuous graded probability score (CRPS) between y and y be defined as follows: In the formula, x is the integration variable. For the Heaviside function, when The value is 1 if the condition is met, and 0 otherwise. G(x) is the predicted distribution obtained during model training, which uses historical photovoltaic power data and meteorological characteristic data for training.

[0045] In practical applications of photovoltaic forecasting, only discrete time series data can be used for prediction; therefore, it is necessary to derive the discretization of the continuous hierarchical probability score. Assuming that... There is a set of predicted values ​​at any given time. for The set of predicted values Each member represents a set of possible future scenarios. The prediction distribution is transformed into a discrete cumulative distribution function form: In the formula, This represents the weight of each member, which is usually set to be equal, i.e. .

[0046] Using this definition, take We can calculate the continuous graded probability score: In the formula, Representing the quantile level, according to the above expression, a discrete continuous tiered probability score can be obtained by weighting the quantile scores, that is: Summing over the time range yields the final discrete form expression for the continuous hierarchical probability score: In the formula, Indicates the length of the time series. for The elements in the set of true and predicted values ​​of the time series at each moment. That is, time. At that time, the first The predicted values ​​of each quantile. The QR-CNN-LSTM photovoltaic power output probability prediction model minimizes... Using historical photovoltaic (PV) output data and meteorological data as the objective function, the system is trained to predict the probability of PV power generation in the next 24 hours by inputting PV output data and meteorological data from the previous week. This is the basis for the next step of risk period identification.

[0047] This implementation proposes a method for constructing a probabilistic prediction model using a hybrid neural network and quantile regression. First, a convolutional neural network (CNN) model is used to extract photovoltaic (PV) output features, leveraging the advantages of CNNs in processing multidimensional data and extracting nonlinear features to effectively capture the complex mapping relationship between meteorological factors and PV output. Then, the advantages of long short-term memory (LSTM) neural networks in processing time-series data are utilized to model the temporal evolution of PV output, obtaining a preliminary deterministic prediction result. Finally, a quantile regression model is introduced. Based on the deterministic prediction result, the quantile regression model obtains predicted values ​​at different quantile levels, extending the deterministic prediction to a probabilistic prediction, thereby obtaining the complete probability distribution of PV output.

[0048] To ensure prediction accuracy, this invention not only employs the traditional quantile loss function to constrain the prediction error at each quantile, but also introduces the Continuous Graded Probability Score (CRPS) as an evaluation index for the overall probability prediction quality. A joint loss function is then constructed based on these two metrics for model training. This process enables the model to simultaneously optimize both the local accuracy of quantile predictions and the global fit of the entire probability density curve, effectively improving the accuracy and reliability of photovoltaic power output probability prediction and solving the technical problem that traditional prediction methods cannot quantify the uncertainty range of predictions.

[0049] Furthermore, the step of obtaining the load forecast results of the power system and then predicting future risk periods of the power system based on the load forecast results includes: To obtain the inherent capacity adequacy level of the power system; Based on the load forecast results and the inherent capacity adequacy level, obtain the net load value for each time node in the preset future period; All time points in the timeframe where the net load value is greater than or equal to zero are considered as the future risk period.

[0050] In one specific implementation, the future risk period refers to all the times when the system is most vulnerable and most in need of new energy sources to provide capacity support. Therefore, the inherent capacity adequacy level of the system is defined based on the system's own ability to support load demand. The calculation formula is as follows: In the formula, This represents the system's time-series load for the day. For indicator functions, when The value is 1 when it is valid, and 0 otherwise. The expected power shortage allowed by the system is set based on historical system operation data.

[0051] Determine the inherent capacity adequacy level of the system. Then, define the system net load value. To identify future risk periods for the system: According to the definition of net load value, when When the system load exceeds the reliable power supply capacity of existing conventional generating units at that moment, there is a risk of power shortage, requiring supplementation from renewable energy sources; when This indicates that the system load level at that moment is covered by the reliable power supply capacity of the system's conventional generating units, and the risk of power shortage is relatively small.

[0052] definition Net load value Non-negative time period set: That is, the set of all time points where the net load value is greater than or equal to zero. This refers to the period of future risk.

[0053] In this implementation, the net load value is calculated by introducing an inherent capacity adequacy level and combining it with load forecast results. This net load value reflects the load gap that the system still needs to meet by relying on photovoltaic power output after deducting its own reliable power supply capacity. When the net load value is greater than or equal to zero, it means that the system's own power supply capacity is insufficient to meet all load demands, and it must rely on renewable energy output. If the renewable energy output is insufficient or fluctuates too much due to weather conditions at this time, the system is very prone to power supply gaps or operational risks, which can intuitively reflect the supply and demand balance of the system under the support of conventional units. This invention defines the time node when the net load value is greater than or equal to zero as the future risk period, accurately locking in the key time window when the system's power supply capacity may be insufficient and its dependence on photovoltaic output is high. This process focuses the subsequent weak link identification work on the most critical risk operation period, avoiding the waste of resources caused by indiscriminate analysis and identification throughout the entire time period. It not only reduces the amount of data processing for subsequent calculations, but also makes the identification of weak links more targeted, improving the targeting and efficiency of weak link identification.

[0054] Further, obtaining the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results includes: Obtain the expected output value and output variance of the photovoltaic power output probability prediction distribution within the future risk period, and obtain the expected load value of the load prediction result within the future risk period; Obtain the covariance between the expected output value and the expected load value; The confidence capacity of the distributed photovoltaic power station during the future risk period is calculated based on the expected output value, the output variance, the expected load value, and the covariance.

[0055] It should be noted that confidence capacity (CV) is the capacity of a generating unit that is trusted under "equal reliability." That is, while maintaining the existing system's reliable power supply capacity, the additional load that can be met by adding new renewable energy sources can be borne by a conventional generating unit of a certain capacity. The rated capacity of this conventional generating unit is called the confidence capacity. Renewable energy confidence capacity assessment involves calculating the confidence capacity and reliability of a generating unit within a specific grid and time period to facilitate renewable energy integration design and grid planning. In a specific embodiment, based on the definition of confidence capacity (CV), the following equation is established: In the formula, This indicates that the system's total power generation capacity at time t is p, and the load level is... Reliability function at that time; The time-series output of the additional unit being evaluated; G is the set of existing thermal units in the system; T is the number of time cycles; For the system's time-series load; Let g be the capacity of the unit. The confidence capacity (CV) of the additional unit being evaluated is also known as the equivalent reliable capacity (EFC). The system's reliability without this unit is shown on the right side of the equal sign in the above formula.

[0056] Distributed new energy sources during risk periods The calculation can be performed using the following formula: In the formula, Net load value A set of non-negative time periods; Indicates the output of distributed new energy sources Expected contribution during periods of risk; This represents the variance of distributed renewable energy output during periods of high risk. This represents the covariance between the load D and the output of distributed renewable energy during the risk period; This represents the expected load during the risk period. The above values ​​are calculated based on probabilistic forecasts, using the following formulas: In the formula, for and The expected value of the joint distribution; At the quantile level The prediction results at that time; The probability width; The load at time t represents the risk period.

[0057] In this implementation, the present invention proposes a method for calculating dynamic confidence capacity. Traditional confidence capacity is typically a static value based on long-term statistics, while this invention focuses on previously identified future risk periods. By calculating the expected value and variance of photovoltaic (PV) output, as well as the expected value of load, and further solving for their covariance, the correlation between PV output and load demand during risk periods can be quantified. This confidence capacity fully considers the effective support capacity that PV power can provide during the future risk periods when the system most needs electricity. Using this dynamic confidence capacity as input to a subsequent vulnerability identification model allows the model to perceive the changes in the strength of new energy's support capacity for the system during different risk periods, thereby more accurately assessing its impact on the system's vulnerability. This overcomes the limitations of traditional methods that treat PV output as a constant value or simply reduce it, improving the accuracy and scientific rigor of the assessment results.

[0058] Based on the identification of future risk periods through the above steps, from the perspective of power balance, system weaknesses will appear during future risk periods. Therefore, the problem of identifying weaknesses is transformed into a classification problem to be solved. However, there are many factors that may affect weaknesses during future risk periods. If we simply assign weights to each factor and sort them, the indicators will lose some independence after integration. Therefore, this embodiment of the invention proposes to use Least Squares Support Vector Machine (LSSVM) to identify weaknesses during future risk periods.

[0059] Furthermore, based on the photovoltaic output probability prediction distribution, the future operating scenarios of each node and each line of the power system are obtained, and then the vulnerability assessment indicators of the future operating scenarios are obtained, including: Based on the future operating scenario, the number of nodes in the power system that experience voltage overruns during the future risk period is obtained, and the total number of voltage overruns in the power system during the future risk period is obtained. A first evaluation index is obtained based on the number of nodes and the total number of voltage over-limit occurrences. Based on the future operating scenario, the number of power system lines that exceed the transmission capacity limit during the future risk period is obtained, and the total number of power system transmission capacity exceedances during the future risk period is obtained. A second evaluation index is obtained based on the number of lines and the total number of transmission capacity overruns. The first evaluation index and the second evaluation index are used as the weakness evaluation index.

[0060] In one specific embodiment, to measure the degree of weakness of the weak link, that is, the impact of a failure on the stable operation of the system, the following indicators are used: The above formula represents the first evaluation indicator. In the formula, This is the sum of the number of nodes in the power system that experience node voltage exceedances during the future risk period, and the number of such occurrences. Let be the voltage amplitude of the i-th node at time t; and These represent the upper and lower limits of the voltage fluctuation boundary for the i-th node, respectively; N is the total number of nodes in the power system; and T is the future risk period, i.e., the net load value. A set of non-negative time periods.

[0061] The above formula represents the second evaluation indicator. In the formula, This is the sum of the number of power system lines that experience transmission capacity overruns during the future risk period and the number of such occurrences. Let be the transmission capacity of the i-th line at time t; Let N be the upper limit of the transmission capacity of the i-th line at time t; N be the total number of power system lines; and T be the future risk period, i.e., the net load value. A set of non-negative time periods.

[0062] In this implementation, the system state during future risk periods is simulated within the power system output plan and future operating scenarios generated by the photovoltaic output probability distribution. The first evaluation metric focuses on voltage quality. By statistically analyzing the number of nodes experiencing voltage exceedances and the total number of exceedance events, the system's weakness in voltage stability can be measured. A higher number and more frequent exceedances indicate weaker node voltage support capabilities. The second evaluation metric focuses on transmission capacity. By statistically analyzing the number of lines experiencing transmission capacity exceedances and the total number of exceedance events, the system's weakness in power flow transmission can be measured. Similarly, a higher number indicates a more prominent network transmission bottleneck. Combining these two metrics comprehensively covers the safety constraints of the two key dimensions of node voltage and line power flow, intuitively and quantitatively characterizing the system's weaknesses under different future operating scenarios, and providing representative feature variables for the training of subsequent machine learning models.

[0063] Furthermore, the construction of the least squares support vector machine model for each node and each line of the power system includes: To obtain the failure rate and repair rate of the power system; The normal operation probability density function of the power system is obtained based on the failure rate, and the fault repair probability density function of the power system is obtained based on the repair rate. A first sampling formula for the duration of normal operation of the power system is obtained based on the normal operation probability density function, and a second sampling formula for the fault repair time of the power system is obtained based on the fault repair probability density function. Combining the photovoltaic output probability prediction distribution, the first sampling formula, and the second sampling formula, a power system state sequence is generated using the sequential Monte Carlo simulation method, and then several fault scenarios are generated based on the power system state sequence. The least squares support vector machine model is generated based on several of the aforementioned fault scenarios.

[0064] It should be noted that during the risk period, the main states of each component in the system are divided into two categories: normal operation and fault shutdown. Generally speaking, the normal operation duration and fault repair time of system components both follow an exponential distribution, and the probability density function of normal operation is... f ( t） and fault repair probability density function g ( t） They are respectively: In the formula, Failure rate; This refers to the repair rate.

[0065] Integrating the probability density function yields the relationship between probability and time. Then, by generating random numbers between [0,1], the normal operating duration is sampled in reverse. and troubleshooting time The corresponding first sampling formula and the second sampling formula for: In the formula, , It is a random number uniformly distributed between [0,1].

[0066] During the future risk period Ω, by combining the probability prediction results of distributed photovoltaic power output with the system component fault sampling formula, a system state sequence is generated through sequential Monte Carlo simulation, and several fault scenarios of power system components are sampled.

[0067] For any fault scenario, calculate the corresponding first evaluation index for each node and line. Second evaluation indicators Therefore, the future risk period, confidence capacity, and primary assessment metric corresponding to this failure scenario are determined. Second evaluation indicators , as a set of training samples.

[0068] It should be understood that the key to identifying weak links based on the Least Squares Support Vector Machine (LSSVM) model lies in determining the training set. Determining the training dataset... ,in, As input, several sets of confidence capacity, risk period, and primary evaluation indicators are used for several fault scenarios corresponding to the aforementioned power system components. Second evaluation indicators ; The output is the identification and judgment result of each weak link in the power system network structure in the corresponding fault scenario, determining whether it belongs to a weak link or not. An LSSVM model is established. for: In the formula, Let k be the Lagrange multiplier of sample k; b is the total number of samples; b is the bias parameter. The expression for selecting the Gaussian radial basis function (RBF) is shown in the following equation.

[0069] In the formula, It is an exponential function with base e; This represents the RBF core width.

[0070] By constructing the aforementioned least squares support vector machine model and training set, a classification model is finally trained that can identify weak links in the power system grid nodes or lines in the future operation scenario power output plan formulated based on the short-term (24-hour) power output probability prediction results of distributed photovoltaic power generation.

[0071] In this implementation, a probabilistic model describing equipment operation and repair time is established by introducing equipment failure rate and repair rate. Using the first and second sampling formulas, sequential sampling can be performed on the entire future operation-repair process of the equipment, generating a system state sequence containing the time dimension. Combining this sequence with the photovoltaic output probability prediction distribution generates more realistic composite fault scenarios that simultaneously consider the randomness of new energy sources and random equipment failures. Using these highly realistic fault scenarios to train a least-squares support vector machine model allows the model to fully learn which nodes or lines in the system are most likely to become weak links under the superposition of multiple uncertainties. This training method greatly enhances the model's generalization ability and prediction accuracy for complex operating conditions, making the final weakness identification results more practically instructive.

[0072] A second aspect of the present invention provides a power system weak link identification system based on confidence capacity, comprising: The data acquisition module 100 is used to acquire day-ahead power output data and day-ahead meteorological data of distributed photovoltaic power stations in the power system; The model building module 200 is used to build a photovoltaic output probability prediction model based on quantile regression, and to build a least squares support vector machine model for each node and each line of the power system. The photovoltaic output probability prediction module 300 is used to input the daytime output data and the daytime meteorological data into the photovoltaic output probability prediction model to obtain the photovoltaic output probability prediction distribution. The risk period prediction module 400 is used to obtain the load prediction results of the power system and then predict the future risk periods of the power system based on the load prediction results. The confidence capacity assessment module 500 is used to obtain the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results. The weakness index analysis module 600 is used to obtain the future operation scenario of the power system based on the photovoltaic output probability prediction distribution, and then obtain the weakness assessment index of the future operation scenario. The weak link identification module 700 is used to input the future risk period, the confidence capacity and the weakness assessment index into the least squares support vector machine model, so as to obtain the weakness identification results of each node and each line in the power system.

[0073] Furthermore, the construction of the photovoltaic power output probability prediction model based on quantile regression includes: Historical power output data and historical meteorological data of distributed photovoltaic power stations in the power system are obtained, and then a prediction training set and a prediction validation set are constructed based on the historical power output data and the historical meteorological data. Construct a convolutional neural network model so that the convolutional neural network model outputs photovoltaic power output features based on the prediction training set; A long short-term memory neural network model is constructed so that the long short-term memory neural network model can make deterministic predictions of photovoltaic output based on the photovoltaic output characteristics, thereby obtaining the photovoltaic output prediction results; A quantile regression model is constructed so that the quantile regression model obtains several sub-predicted values ​​at different quantile levels based on the photovoltaic power output prediction results, and then obtains the photovoltaic power output probability prediction distribution based on the several sub-predicted values; Obtain the quantile loss function of the photovoltaic output probability prediction distribution, and obtain the continuous hierarchical probability score between the photovoltaic output probability prediction distribution and the prediction validation set; The convolutional neural network model, the long short-term memory neural network model, and the quantile regression model are used to construct the photovoltaic power output probability prediction model, and the probability prediction loss function of the photovoltaic power output probability prediction model is constructed based on the quantile loss function and the continuous hierarchical probability score.

[0074] Furthermore, the step of obtaining the load forecast results of the power system and then predicting future risk periods of the power system based on the load forecast results includes: To obtain the inherent capacity adequacy level of the power system; Based on the load forecast results and the inherent capacity adequacy level, obtain the net load value for each time node in the preset future period; All time points in the timeframe where the net load value is greater than or equal to zero are considered as the future risk period.

[0075] Further, obtaining the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results includes: Obtain the expected output value and output variance of the photovoltaic power output probability prediction distribution within the future risk period, and obtain the expected load value of the load prediction result within the future risk period; Obtain the covariance between the expected output value and the expected load value; The confidence capacity of the distributed photovoltaic power station during the future risk period is calculated based on the expected output value, the output variance, the expected load value, and the covariance.

[0076] The large-scale integration of distributed renewable energy sources brings significant uncertainties to power system operation. Traditional methods for identifying weak links often rely on deterministic predictions or simple weighted assessments, making it difficult to effectively quantify the randomness of renewable energy output and its supporting role in system risk. Existing technologies often lack a complete integration from probabilistic prediction to risk identification, resulting in insufficient prediction accuracy, inaccurate identification of risk periods, and strong subjectivity in weak link classification. Therefore, this solution constructs a comprehensive evaluation and optimization system for identifying weak links in the power system, considering the confidence capacity of distributed renewable energy. First, the uncertainty of renewable energy output is quantified using a photovoltaic power probabilistic prediction method based on CRPS quantile regression. Second, future risk periods of the system are identified based on net load levels, and the renewable energy support capacity during risk periods is assessed using confidence capacity. Finally, weak links in the system are identified and classified based on LSSVM. Compared with existing technologies, the power system weak link identification method and system based on confidence capacity provided by this invention have at least the following advantages: First, this invention presents a photovoltaic power probabilistic prediction method based on CRPS quantile regression: It employs a deep learning model for power prediction based on Continuous Probability Scoring (CRPS) quantile regression QR-CNN-LSTM to achieve probabilistic prediction of distributed photovoltaic power output. This method effectively quantifies output uncertainty through quantile regression prediction, improving prediction accuracy and reliability, and providing a probabilistic basis for subsequent confidence capacity calculation.

[0077] Second, this invention provides a confidence capacity assessment and risk period identification method based on equivalent reliable capacity (EFC): It identifies future risk periods by defining and calculating the system net load value, and calculates the confidence capacity of distributed renewable energy sources during these risk periods based on the Equivalent Reliable Capacity (EFC) principle. This method utilizes the expectation, variance, and covariance with the load of the probability prediction results to quantify the supporting effect of renewable energy output on the system load, thereby accurately identifying vulnerable moments in the system.

[0078] Third, this invention identifies load risk weak links based on LSSVM: it adopts the least squares support vector machine (LSSVM) classifier model and combines it with the fault event probability model for time-series sampling. It identifies and classifies weak links based on indicators such as confidence capacity, voltage limit exceedance, and transmission capacity limit exceedance, avoiding the problem of insufficient independence in traditional weighting methods and improving the accuracy and efficiency of system weak link identification and classification.

[0079] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0080] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described; however, any combination of these technical features that does not contradict each other should be considered within the scope of this specification.

[0081] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the concept of this application, and these improvements and substitutions should also be considered within the scope of protection of this invention. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for identifying weak links in a power system based on confidence capacity, characterized in that, include: Acquire day-ahead power output data and day-ahead meteorological data of distributed photovoltaic power stations in the power system; Construct a photovoltaic power output probability prediction model based on quantile regression; The photovoltaic output probability prediction model is input into the daytime power output data and the daytime meteorological data to obtain the photovoltaic output probability prediction distribution. Obtain the load forecast results of the power system, and then predict the future risk periods of the power system based on the load forecast results; Based on the photovoltaic output probability prediction distribution and the load prediction results, the confidence capacity of the distributed photovoltaic power station during the future risk period is obtained. Based on the photovoltaic output probability prediction distribution, the future operation scenarios of the power system are obtained, and then the vulnerability assessment index of the future operation scenarios is obtained. Construct a least-squares support vector machine model for each node and each line in the power system; The future risk period, the confidence capacity, and the vulnerability assessment index are input into the least squares support vector machine model to obtain the vulnerability identification results for each node and each line in the power system.

2. The method of claim 1, wherein, The construction of the photovoltaic power output probability prediction model based on quantile regression includes: Historical power output data and historical meteorological data of distributed photovoltaic power stations in the power system are obtained, and then a prediction training set and a prediction validation set are constructed based on the historical power output data and the historical meteorological data. Construct a convolutional neural network model so that the convolutional neural network model outputs photovoltaic power output features based on the prediction training set; A long short-term memory neural network model is constructed so that the long short-term memory neural network model can make deterministic predictions of photovoltaic output based on the photovoltaic output characteristics, thereby obtaining the photovoltaic output prediction results; A quantile regression model is constructed so that the quantile regression model obtains several sub-predicted values ​​at different quantile levels based on the photovoltaic power output prediction results, and then obtains the photovoltaic power output probability prediction distribution based on the several sub-predicted values; Obtain the quantile loss function of the photovoltaic output probability prediction distribution, and obtain the continuous hierarchical probability score between the photovoltaic output probability prediction distribution and the prediction validation set; The convolutional neural network model, the long short-term memory neural network model, and the quantile regression model are used to construct the photovoltaic power output probability prediction model, and the probability prediction loss function of the photovoltaic power output probability prediction model is constructed based on the quantile loss function and the continuous hierarchical probability score.

3. The method of claim 1, wherein, The process of obtaining load forecast results for the power system and then predicting future risk periods for the power system based on these load forecast results includes: To obtain the inherent capacity adequacy level of the power system; Based on the load forecast results and the inherent capacity adequacy level, obtain the net load value for each time node in the preset future period; All time points in the timeframe where the net load value is greater than or equal to zero are considered as the future risk period.

4. The method of claim 1, wherein, The process of obtaining the confidence capacity of distributed photovoltaic power stations during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results includes: Obtain the expected output value and output variance of the photovoltaic power output probability prediction distribution within the future risk period, and obtain the expected load value of the load prediction result within the future risk period; Obtain the covariance between the expected output value and the expected load value; The confidence capacity of the distributed photovoltaic power station during the future risk period is calculated based on the expected output value, the output variance, the expected load value, and the covariance.

5. The method of claim 1, wherein, The process of obtaining future operating scenarios of the power system based on the photovoltaic output probability prediction distribution, and then obtaining vulnerability assessment indicators for the future operating scenarios, includes: Based on the future operating scenario, the number of nodes in the power system that experience voltage overruns during the future risk period is obtained, and the total number of voltage overruns in the power system during the future risk period is obtained. A first evaluation index is obtained based on the number of nodes and the total number of voltage over-limit occurrences. Based on the future operating scenario, the number of power system lines that exceed the transmission capacity limit during the future risk period is obtained, and the total number of power system transmission capacity exceedances during the future risk period is obtained. A second evaluation index is obtained based on the number of lines and the total number of transmission capacity overruns. The first evaluation index and the second evaluation index are used as the weakness evaluation index.

6. The method of claim 1, wherein, The construction of the least squares support vector machine model for each node and each line of the power system includes: To obtain the failure rate and repair rate of the power system; The normal operation probability density function of the power system is obtained based on the failure rate, and the fault repair probability density function of the power system is obtained based on the repair rate. A first sampling formula for the duration of normal operation of the power system is obtained based on the normal operation probability density function, and a second sampling formula for the fault repair time of the power system is obtained based on the fault repair probability density function. Combining the photovoltaic output probability prediction distribution, the first sampling formula, and the second sampling formula, a power system state sequence is generated using the sequential Monte Carlo simulation method, and then several fault scenarios are generated based on the power system state sequence. The least squares support vector machine model is generated based on several of the aforementioned fault scenarios.

7. A confidence capacity based power system weak link identification system, characterized by, include: The data acquisition module is used to acquire day-ahead power output data and day-ahead meteorological data of distributed photovoltaic power stations in the power system; The model building module is used to build a photovoltaic output probability prediction model based on quantile regression, and to build a least squares support vector machine model for each node and each line in the power system. The photovoltaic output probability prediction module is used to input the daytime output data and the daytime meteorological data into the photovoltaic output probability prediction model to obtain the photovoltaic output probability prediction distribution. The risk period prediction module is used to obtain the load prediction results of the power system, and then predict the future risk periods of the power system based on the load prediction results; The confidence capacity assessment module is used to obtain the confidence capacity of the distributed photovoltaic power station during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results. The weakness index analysis module is used to obtain the future operation scenario of the power system based on the photovoltaic output probability prediction distribution, and then obtain the weakness assessment index of the future operation scenario. The weak link identification module is used to input the future risk period, the confidence capacity, and the weakness assessment index into the least squares support vector machine model, so as to obtain the weakness identification results of each node and each line in the power system.

8. A power system weak link identification system based on confidence capacity according to claim 7, characterized in that, The construction of the photovoltaic power output probability prediction model based on quantile regression includes: Historical power output data and historical meteorological data of distributed photovoltaic power stations in the power system are obtained, and then a prediction training set and a prediction validation set are constructed based on the historical power output data and the historical meteorological data. Construct a convolutional neural network model so that the convolutional neural network model outputs photovoltaic power output features based on the prediction training set; A long short-term memory neural network model is constructed so that the long short-term memory neural network model can make deterministic predictions of photovoltaic output based on the photovoltaic output characteristics, thereby obtaining the photovoltaic output prediction results; A quantile regression model is constructed so that the quantile regression model obtains several sub-predicted values ​​at different quantile levels based on the photovoltaic power output prediction results, and then obtains the photovoltaic power output probability prediction distribution based on the several sub-predicted values; Obtain the quantile loss function of the photovoltaic output probability prediction distribution, and obtain the continuous hierarchical probability score between the photovoltaic output probability prediction distribution and the prediction validation set; The convolutional neural network model, the long short-term memory neural network model, and the quantile regression model are used to construct the photovoltaic power output probability prediction model, and the probability prediction loss function of the photovoltaic power output probability prediction model is constructed based on the quantile loss function and the continuous hierarchical probability score.

9. A power system weak link identification system based on confidence capacity according to claim 7, characterized in that, The process of obtaining load forecast results for the power system and then predicting future risk periods for the power system based on these load forecast results includes: To obtain the inherent capacity adequacy level of the power system; Based on the load forecast results and the inherent capacity adequacy level, obtain the net load value for each time node in the preset future period; All time points in the timeframe where the net load value is greater than or equal to zero are considered as the future risk period.

10. A power system weak link identification system based on confidence capacity according to claim 7, characterized in that, The process of obtaining the confidence capacity of distributed photovoltaic power stations during the future risk period based on the photovoltaic output probability prediction distribution and the load prediction results includes: Obtain the expected output value and output variance of the photovoltaic power output probability prediction distribution within the future risk period, and obtain the expected load value of the load prediction result within the future risk period; Obtain the covariance between the expected output value and the expected load value; The confidence capacity of the distributed photovoltaic power station during the future risk period is calculated based on the expected output value, the output variance, the expected load value, and the covariance.

Images