Method for identifying weak link of future-state power system based on wind and light output

Abstract

The invention discloses a method for identifying a weak link of a future-state power system based on wind and light output, and relates to the technical field of power processing. Comprising the following steps: determining a node graph of the power system, meteorological prediction data of an area where the power system is located in a future duration, a prediction load level and historical power operation characteristic data of the power system in a historical operation process; inputting the meteorological prediction data into the target wind and light output prediction model to obtain a wind and light output prediction value of the power system; determining a typical operation scene set according to the power system node graph and historical power flow data in the historical power operation characteristic data; and determining weak links of the power system in the future operation scene according to the predicted load level, the typical operation scene set and the wind and light output predicted value. According to the technical scheme, the accuracy and precision of prediction of the future weak link can be improved, and the authenticity of the future weak link can be improved.

Description

Technical Field

[0001] This invention relates to the field of power processing technology, specifically to a method for identifying weak links in future power systems based on wind and solar power output. Background Technology

[0002] With the development of new energy power systems, the penetration rate of renewable energy sources such as wind power and photovoltaics in the power system continues to increase. However, the randomness, volatility, and intermittency of wind and solar power output have greatly altered the power flow distribution and dynamic characteristics of the power system, causing the system's operating state to shift from a traditional deterministic mode to a highly uncertain mode. Against this backdrop, identifying the weak links in the future power system at a certain future period (such as the planning year or a typical operating day) is of vital importance for guiding grid planning, optimizing operation modes, and preventing the risk of large-scale power outages.

[0003] Traditional methods for identifying weaknesses in power systems are mostly based on deterministic scenarios or historical typical operating modes. For example, they assess the system's safety level at preset operating points through static security analysis, short-circuit current calculation, and transient stability analysis. While these methods are mature, they have significant limitations: First, they typically fail to fully consider the inherent uncertainties in wind and solar power output, often using typical scenario methods or probabilistic approaches for rough estimations. This fails to accurately depict the dynamic evolution of wind and solar power output over time and its coupling relationship with loads and grid structure, potentially leading to identification results that deviate from the actual risks faced by the system. Second, with the development of high-precision numerical weather prediction and artificial intelligence forecasting technologies, the accuracy of short-term and even ultra-short-term forecasts of wind and solar power output has significantly improved. However, existing weakness identification methods fail to effectively utilize this high-precision forecast information and fail to effectively combine prediction and identification, limiting the foresight and accuracy of the identification methods. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying weak links in future power systems based on wind and solar power output, so as to solve the technical problems existing in related technologies.

[0005] To achieve the above objectives, the present invention provides a method for identifying weak links in a future power system based on wind and solar power output, comprising:

[0006] The node diagram of the power system, the meteorological forecast data of the area where the power system is located in the future, the future load level, and the historical power operation characteristics data of the power system in the historical operation process are determined. The power system is a multi-energy power system including wind and solar power output.

[0007] Meteorological forecast data is input into the target wind and solar power output prediction model to obtain the wind and solar power output prediction value of the power system. The target wind and solar power output prediction model is trained by historical wind and solar power output values ​​and corresponding historical meteorological data from historical power operation characteristic data.

[0008] Based on the power system node diagram and historical power flow data in historical power operation characteristic data, a set of typical operating scenarios is determined. Each typical operating scenario in the set of typical operating scenarios includes a corresponding weak link.

[0009] Based on future load levels, typical operating scenarios, and wind and solar power output forecasts, weak links in the power system in future operating scenarios are identified.

[0010] Optionally, the target wind and solar power output prediction model is obtained through the following method:

[0011] Determine the historical wind and solar power output values ​​in the historical power operation characteristic data, as well as the historical meteorological data corresponding to the historical wind and solar power output values;

[0012] The target wind and solar power output prediction model is obtained by training the wind and solar power output prediction model with historical wind and solar power output values ​​and historical meteorological data.

[0013] Optionally, the historical power operation characteristic data is obtained through the following method:

[0014] Acquire historical power data and target files of the power system during its historical operation, wherein the target files are used to configure the power operation feature data to be extracted;

[0015] Extract key power feature information from the target file, and determine historical power operation feature data based on the key power feature information and historical power operation data. The key power feature information is obtained by processing the historical power operation report through a key feature extraction model.

[0016] Optionally, determining historical power operation characteristic data based on key power characteristic information and historical power operation data includes:

[0017] Based on the power station affiliation information of key power characteristics, the power stations where historical operating power data is located are divided into regions, and the load data of each region is calculated.

[0018] Based on the cross-sectional lines with key power characteristics information, determine the AC line power flow data in the historical power operation data, and calculate the power flow value corresponding to the cross-sectional lines based on the AC line power flow data.

[0019] Based on the key power characteristics information of the DC converter station, the target lines flowing to the DC converter station in the historical power data are determined, and the power values ​​of the target lines are summarized. The total power of each DC transmission channel is calculated based on the power values ​​of the target lines, wherein each DC transmission channel includes a pair of DC converter stations.

[0020] By integrating the load data of each region, the power flow value of the corresponding cross-section line, and the total power of each DC transmission channel into the target file, historical power operation characteristic data is obtained.

[0021] Optionally, the key feature extraction model includes a keyword extraction model, a high-frequency information deep indexing and feature association model, and an expert experience fusion and key feature optimization model; the key power feature information is obtained through the following methods:

[0022] Obtain historical power operation reports generated by the power system during its historical operation;

[0023] The key power feature words that are related to the spatiotemporal distribution characteristics of power grid flow are extracted from the historical power operation report by the keyword extraction model. The location and contextual information of the key power feature words in the historical power operation report are determined. The key power feature words, their location and contextual information are stored in a structured manner to obtain the first structured feature text.

[0024] The target words in the first structured feature text are extracted by high-frequency information deep indexing and feature association model. The target words are then processed by dynamic inverted indexing mechanism and feature association mining algorithm to obtain the second structured feature text. The target words are words that appear more than or equal to a preset number of times in the first structured feature text.

[0025] By integrating expert experience and using a key feature optimization model to perform collaborative analysis on the second structured feature text and processing the second structured feature text using preset rules, key power feature information is obtained.

[0026] Optionally, determining the set of typical operating scenarios based on the power system node diagram and historical power flow data from historical power operation characteristic data includes:

[0027] The vulnerability index corresponding to the historical power flow data of each transmission line in the historical power operation characteristic data is calculated by the HITS algorithm. Multiple vulnerability indices are obtained, and based on the multiple vulnerability indices and the power system node diagram, the weak links of sub-targets in the historical operation scenario are determined.

[0028] Cluster analysis was performed on historical power operation characteristic data to obtain Typical operating scenarios, wherein, Each typical operating scenario includes its corresponding weak points.

[0029] Weak links of sub-targets in historical operational scenarios and Typical operating scenarios and their corresponding weaknesses are identified as a set of typical operating scenarios.

[0030] Optionally, the step of determining the weak links of sub-objectives under historical operating scenarios based on multiple vulnerability indicators and the power system node diagram includes:

[0031] Multiple vulnerability indicators are sorted from largest to smallest, and the transmission lines corresponding to the first preset number of vulnerability indicators are determined as the target lines in the historical operation scenario.

[0032] The power system node graph is processed by a depth-first search algorithm to generate the minimum cut set of the power system. The minimum cut set is then searched by recursive enumeration to generate candidate weak links.

[0033] Candidate weak links, including those along the target route, are identified as weak links in sub-targets under historical operational scenarios.

[0034] Optionally, determining the weak links in the power system in future operating scenarios based on future load levels, a set of typical operating scenarios, and predicted wind and solar power output includes:

[0035] The first feature vector is determined by identifying the future load level and the predicted wind and solar power output, and the second feature vector is determined by identifying the Euclidean distance between each second feature vector and the first feature vector in the typical operation scenario set. The typical operation scenario corresponding to the Euclidean distance that is less than or equal to the preset distance is identified as the target typical operation scenario.

[0036] Multiple first-selected weak links in the typical operating scenario of the target are identified, and static security and stability verification is performed on each first-selected weak link based on the future load level and the predicted wind and solar power output. The first-selected weak links that pass the verification are identified as weak links in the power system in the future operating scenario.

[0037] Optionally, the step of performing static security and stability verification on each first pre-selected weak link based on future load levels and predicted wind and solar power output, and determining the first pre-selected weak links that pass the verification as weak links in the power system in future operating scenarios, includes:

[0038] Based on the predicted wind and solar power output and the future load level, power flow calculations are performed on each first pre-selected weak link to obtain the first total active power of the transmission section of each first pre-selected weak link, and the first pre-selected weak link corresponding to the first total active power that is greater than or equal to the preset first total active power is determined as the second pre-selected weak link.

[0039] For each second pre-selected weak link, a stability check is performed, and the second total active power and total long-term allowable current carrying capacity of each second pre-selected weak link are obtained after the stability check.

[0040] Based on the second total active power and the total long-term allowable current carrying capacity, weak links in the power system in future operating scenarios are identified.

[0041] Optionally, determining the weak links in the power system in future operating scenarios based on the second total active power and the total long-term allowable current carrying capacity includes:

[0042] The second total active power is divided by the total long-term allowable current carrying capacity to obtain the quotient value. The second pre-selected weak link corresponding to the quotient value greater than the preset quotient value is identified as the weak link in the future operation scenario.

[0043] Optionally, the node diagram of the power system is obtained by the following method:

[0044] By treating parallel buses in a power system as a single bus, treating multiple lines between two nodes as a single line, and removing independent nodes, a node diagram of the power system is obtained.

[0045] The above technical solution involves inputting meteorological forecast data into a target wind and solar power output prediction model to obtain predicted wind and solar power output values ​​for the power system. Based on the power system node diagram and historical power flow data from historical power operation characteristic data, a set of typical operating scenarios is determined. Then, based on future load levels, the set of typical operating scenarios, and the predicted wind and solar power output values, weak links in the power system within these future operating scenarios are identified. The target wind and solar power output prediction model is trained using historical wind and solar power output values ​​and corresponding historical meteorological data from historical power operation characteristic data. By incorporating the inherent uncertainty of wind and solar power output into the identification of weak links in future operating scenarios, and by determining these weak links based on predicted wind and solar power output values, future load levels, and the set of typical operating scenarios, the accuracy and precision of predicting weak links in future operating scenarios can be improved, as well as the realism of identifying these weak links.

[0046] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0047] Figure 1 This is a schematic diagram illustrating a method for identifying weak links in a future power system based on wind and solar power output according to an exemplary embodiment of the present invention.

[0048] Figure 2This is a step framework diagram illustrating a method for identifying weak links in a future power system based on wind and solar power output according to an exemplary embodiment of the present invention.

[0049] Figure 3 This is a schematic diagram illustrating the basic architecture of the AlexNet model according to an exemplary embodiment of the present invention.

[0050] Figure 4 This invention presents a comparison of wind power prediction results based on the AlexNet model in different seasons, according to an exemplary embodiment of the present invention.

[0051] Figure 5 This invention presents a comparison of photovoltaic power prediction results based on the AlexNet model, according to an exemplary embodiment of the present invention.

[0052] Figure 6 This is a schematic diagram illustrating the criticality of a critical path in a standard IEEE 39-node system according to an exemplary embodiment of the present invention.

[0053] Figure 7 This is a schematic diagram of an IEEE 39-node test system for new energy access according to an exemplary embodiment of the present invention.

[0054] Figure 8 This is a schematic diagram illustrating the comparison and analysis of matching operation scenarios and initial samples under different schemes according to an exemplary embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, so as to provide a better understanding of the concept of the present invention, the technical problem solved, the technical features constituting the technical solution, and the technical effects brought about.

[0056] like Figure 1 As shown, Figure 1 This is a schematic diagram illustrating a method for identifying weak links in a future power system based on wind and solar power output according to an exemplary embodiment of the present invention. (Refer to...) Figure 1 The method includes;

[0057] S101: Determine the node diagram of the power system, the meteorological forecast data of the area where the power system is located in the future, the future load level, and the historical power operation characteristic data of the power system in the historical operation process, wherein the power system is a multi-energy power system including wind and solar power output;

[0058] S102: Input meteorological forecast data into the target wind and solar power output prediction model to obtain the wind and solar power output prediction value of the power system. The target wind and solar power output prediction model is trained by historical wind and solar power output values ​​and corresponding historical meteorological data in historical power operation characteristic data.

[0059] S103: Based on the power system node diagram and historical power flow data in the historical power operation characteristic data, determine the set of typical operating scenarios, wherein each typical operating scenario in the set of typical operating scenarios includes the corresponding weak link;

[0060] S104: Based on future load levels, typical operating scenarios, and wind and solar power output forecasts, identify the weak links in the power system during future operating scenarios.

[0061] The above technical solution involves inputting meteorological forecast data into a target wind and solar power output prediction model to obtain predicted wind and solar power output values ​​for the power system. Based on the power system node diagram and historical power flow data from historical power operation characteristic data, a set of typical operating scenarios is determined. Then, based on future load levels, the set of typical operating scenarios, and the predicted wind and solar power output values, weak links in the power system within these future operating scenarios are identified. The target wind and solar power output prediction model is trained using historical wind and solar power output values ​​and corresponding historical meteorological data from historical power operation characteristic data. By incorporating the inherent uncertainty of wind and solar power output into the identification of weak links in future operating scenarios, and by determining these weak links based on predicted wind and solar power output values, future load levels, and the set of typical operating scenarios, the accuracy and precision of predicting weak links in future operating scenarios can be improved, as well as the realism of identifying these weak links.

[0062] To enable those skilled in the art to better understand the method for identifying weak links in future power systems based on wind and solar power output provided by this invention, the above steps are illustrated in detail below.

[0063] For example, a node diagram of a power system can be a schematic diagram that abstracts the power system as connected nodes. A power system can be a multi-energy generation system, which may include wind, solar, hydro, and thermal energy, etc. Weather forecast data can be the predicted data for the power system over a future time period obtained through weather forecasts. Future load levels can be the future load levels assessed based on historical power operation data obtained during the power system's historical operation. Historical power operation characteristic data can be data used to characterize the power system's power characteristics during its historical operation, including historical wind and solar power output values, historical power flow data, and other power-related data.

[0064] In one possible manner, the node diagram of the power system is obtained by the following method:

[0065] By treating parallel buses in a power system as a single bus, treating multiple lines between two nodes as a single line, and removing independent nodes, a node diagram of the power system is obtained.

[0066] It should be understood that when constructing a power system node graph, a graph-theoretic adjacency matrix can be used to store and analyze the relationships between power system network nodes. This allows for the merging of multiple transmission lines on the same tower, the consolidation of multiple lines between two nodes into a generalized loop, and the removal of independent nodes, thus obtaining the power system node graph and simplifying the computational complexity of subsequent analyses. By simplifying the topology of complex power grids, the analytical dimensionality and complexity of subsequent calculations are significantly reduced, laying the foundation for handling large-scale systems.

[0067] For example, a target wind and solar power output prediction model can be used to predict the wind and solar power output of the power system over a future period. In embodiments of the invention, such as... Figure 2 As shown, meteorological forecast data can be input into the target wind and solar power output prediction model to obtain the predicted wind and solar power output value.

[0068] In one possible manner, the target wind and solar power output prediction model is obtained by the following method:

[0069] Determine the historical wind and solar power output values ​​in the historical power operation characteristic data, as well as the historical meteorological data corresponding to the historical wind and solar power output values;

[0070] The target wind and solar power output prediction model is obtained by training the wind and solar power output prediction model with historical wind and solar power output values ​​and historical meteorological data.

[0071] It should be understood that the target wind and solar power output prediction model is trained based on historical wind and solar power output values ​​and corresponding historical meteorological data. In this embodiment of the invention, historical power operation characteristic data includes historical wind and solar power output values. Then, based on the time corresponding to the acquisition of the historical wind and solar power output values, the corresponding historical meteorological data can be obtained. The wind and solar power output prediction model is then trained based on the historical meteorological data and the historical wind and solar power output values ​​to obtain the target wind and solar power output prediction model. This can improve the accuracy of the predicted wind and solar power output values.

[0072] In this embodiment of the invention, the target wind and solar power output prediction model can be a short-term wind and solar power output prediction model based on AlexNet for multiple scenarios. Specifically, as shown below... Figure 3As shown, params represents the parameter size of the AlexNet model, and FLOPs represents the computational cost of the AlexNet model. The design of the AlexNet model in the LRN (Local Response Normalization) layer and the Dropout layer is as follows.

[0073] LRN layer: By suppressing feedback values ​​from neurons with small local responses while amplifying those with relatively larger local responses, the local response values ​​are normalized, improving the network's generalization ability. Dropout layer practical application: AlexNet successfully applied Dropout layers to actual model training and validated its results. By randomly ignoring some neurons in the fully connected layers, the network structure changes with each training iteration, effectively combining multiple different sub-networks for training, thus reducing overfitting.

[0074] The AlexNet model's main network consists of 8 layers: the first 5 are convolutional layers, followed by max-pooling layers, and the last 3 are fully connected layers. The convolutional layers are the key layers in AlexNet. The computation process of the convolutional kernel begins with an initial region of the image and moves in a specific direction with a certain stride to capture information from the entire image until the entire image is processed.

[0075] The computation process of the convolution kernel in each local region can be viewed as the sum of the products of the corresponding numbers in the overlapping part of the convolution kernel and the input sample.

[0076] To maintain the feature map size and dimensions, a padding method can be chosen, namely, zero-padding at the boundaries. As seen in the convolution operation process, stacking more convolutional layers allows for the extraction of relatively higher-level and deeper features from basic features, thereby capturing richer image information.

[0077] Pooling utilizes a sliding window, which can be viewed as a special type of parameterless convolutional kernel sliding across the input feature map, taking the maximum or mean value within the window as the output. The specific calculation process of max pooling is shown below:

[0078] ;

[0079] in, In the input feature map before pooling, the i-th element at a certain spatial location within the pooling window coverage area... The characteristic values ​​of each channel, The height of the pooled window, The width of the pooled window. The number of channels in the feature map. For the network layer index, In the output feature map after pooling, the first... The first in the spatial position of the layer The characteristic values ​​of each channel.

[0080] After image features are fully extracted through convolutional and pooling layers, fully connected layers are needed to combine the various local features to form global features. This invention uses the Softmax function for the fully connected layer, and the specific calculation is as follows:

[0081] ;

[0082] In the formula, and Let represent the input and weight matrices of the d-th neuron in the (l+1)-th layer, respectively. This represents the output of the l-th layer. This represents the number of neurons.

[0083] Furthermore, to enhance the network's ability to model nonlinear mappings, activation functions are often introduced and applied to the outputs of each layer. This invention uses the ReLU function, specifically calculated as follows:

[0084] ;

[0085] Meanwhile, in order to evaluate the overall predictive performance of the model from multiple perspectives as comprehensively as possible, the root mean square error was selected. Mean absolute percentage error And the coefficient of determination R, which can more intuitively highlight the robustness of the model. 2 As a comprehensive evaluation index to verify the predictive performance of the model, the mathematical expressions of the above three are as follows.

[0086] ;

[0087] ;

[0088] ;

[0089] In the above three equations, y(i) represents the actual power, y'(i) represents the model prediction value, represents the average value of the actual power, and N is the number of y(i). This represents the average value of the predicted results.

[0090] in, It is used to measure the deviation between the predicted value and the actual value. The larger the value, the greater the prediction error. The value ranges from [0, +∞]. A smaller value indicates a better predictive performance; a value of 0 indicates a perfect model, while a value greater than 100% indicates a poor model. R 2The theoretical range is [0, 1]. The closer the result is to 1, the better the model fit; conversely, the closer the result is to 0, the worse the model fit.

[0091] This step combines meteorological data with historical power output data and uses the AlexNet model to make short-term predictions of wind and solar power output.

[0092] In one possible manner, the historical power operation characteristic data is obtained through the following method:

[0093] Acquire historical power data and target files of the power system during its historical operation, wherein the target files are used to configure the power operation feature data to be extracted;

[0094] Extract key power feature information from the target file, and determine historical power operation feature data based on the key power feature information and historical power operation data. The key power feature information is obtained by processing the historical power operation report through a key feature extraction model.

[0095] It should be understood that the target file can be a file that configures power operation characteristic data. The target file may include the mapping table filename and key power characteristic information. Then, based on the key power characteristic information, the power operation characteristic data can be searched in historical power operation data to obtain historical power operation characteristic data.

[0096] In one possible manner, determining historical power operation characteristic data based on key power characteristic information and historical power operation data includes:

[0097] Based on the power station affiliation information of key power characteristics, the power stations where historical operating power data is located are divided into regions, and the load data of each region is calculated.

[0098] Based on the cross-sectional lines with key power characteristics information, determine the AC line power flow data in the historical power operation data, and calculate the power flow value corresponding to the cross-sectional lines based on the AC line power flow data.

[0099] Based on the key power characteristics information of the DC converter station, the target lines flowing to the DC converter station in the historical power data are determined, and the power values ​​of the target lines are summarized. The total power of each DC transmission channel is calculated based on the power values ​​of the target lines, wherein each DC transmission channel includes a pair of DC converter stations.

[0100] By integrating the load data of each region, the power flow value of the corresponding cross-section line, and the total power of each DC transmission channel into the target file, historical power operation characteristic data is obtained.

[0101] It should be understood that, in the specific processing, the power plant affiliation information in the key power characteristic information can be used to divide the power plants where historical operating power data is located into regions, and various load data can be summarized and statistically analyzed by region to obtain the load data for each region. For the calculation of cross-sectional power flow, the program combines the AC line power flow data in the historical operating power data with the cross-sectional lines in the key power characteristic information, retrieves the corresponding AC lines in the historical operating power data, and obtains the power flow value for each cross-section. Regarding the power statistics of DC transmission channels, the program first extracts the names of DC converter stations listed in the key power characteristic information, then retrieves all line information flowing to these converter stations in the historical operating power data, summarizes the power values ​​of the corresponding lines, and thus calculates the total power of each DC transmission channel. Afterwards, the load data for each region, the power flow values ​​corresponding to the cross-sectional lines, and the total power of each DC transmission channel are integrated into the target file to obtain historical power operation characteristic data.

[0102] The above technical solution applies text representation learning technology from the field of natural language processing to the extraction of power grid operation features, enabling the automated and intelligent extraction of key information affecting the system status from massive, unstructured scheduling data.

[0103] Among possible approaches, the key feature extraction model includes a keyword extraction model, a high-frequency information deep indexing and feature association model, and an expert experience fusion and key feature optimization model; the key power feature information is obtained through the following methods:

[0104] Obtain historical power operation reports generated by the power system during its historical operation;

[0105] The key power feature words that are related to the spatiotemporal distribution characteristics of power grid flow are extracted from the historical power operation report by the keyword extraction model. The location and contextual information of the key power feature words in the historical power operation report are determined. The key power feature words, their location and contextual information are stored in a structured manner to obtain the first structured feature text.

[0106] The target words in the first structured feature text are extracted by high-frequency information deep indexing and feature association model. The target words are then processed by dynamic inverted indexing mechanism and feature association mining algorithm to obtain the second structured feature text. The target words are words that appear more than or equal to a preset number of times in the first structured feature text.

[0107] By integrating expert experience and using a key feature optimization model to perform collaborative analysis on the second structured feature text and processing the second structured feature text using preset rules, key power feature information is obtained.

[0108] It should be understood that historical power operation reports can include power data generated by the power system during its historical operation. Obtaining key power feature information from historical power operation reports using a key feature extraction model can involve the following steps.

[0109] The keyword extraction model is a core component of the intelligent analysis architecture for power grid mode calculations. This model is used to extract key features that influence the spatiotemporal distribution characteristics of power grid flow. It enables precise location, context-dependent extraction, and structured storage of key feature terms within relevant technical reports. Its core functions include multi-level keyword indexing and management, precise chapter / section location, intelligent context extraction and correlation analysis, and multi-dimensional output and integrated applications.

[0110] Multi-level keyword indexing and management: Based on the "Power Grid Dispatch Standard Terminology" DL / T 961-2020, it adopts a three-level classification system of equipment / operation / risk, supports user-defined keyword library, such as: operation mode, section, DC, etc., and automatically expands synonyms, such as "main transformer" can match "main transformer", etc., and the keyword library supports dynamic updates.

[0111] Precise chapter / section location: Based on document structure parsing technology, the report content is broken down into four logical units: "chapter → section → paragraph → sentence," ensuring location accuracy down to the individual paragraph level. It supports physical location marking and logical location mapping, facilitating subsequent manual review and system referencing.

[0112] Contextual intelligent extraction and association analysis: Automatically extracts the section containing the keyword, analyzes other text information in the section containing the keyword, and extracts frequently occurring terms and high-risk terms in the section. For example, frequently occurring terms can be cross-section and short-circuit current limiting measures, while high-risk terms can be power flow exceeding limits and voltage instability.

[0113] Multi-dimensional output and integrated application: Structured feature storage results are generated, written into the power grid knowledge base, and a multi-dimensional index is established, including the file source, location information, and contextual text of the keywords. The feature storage results are then manually verified using expert experience to remove irrelevant textual information.

[0114] Through the above steps, key power feature words that are related to the spatiotemporal distribution characteristics of power grid flow are extracted from historical power operation reports. The key power feature words, their locations, and contextual information are then stored in a structured manner to obtain the first structured feature text.

[0115] The first structured feature text can then be processed using a high-frequency information deep indexing and feature association model. Building upon the original keyword extraction model, the high-frequency information deep indexing and feature association model supplements the existing model with full-text secondary indexing of related information and feature association mining functions, forming a multi-level analysis architecture. First, the high-frequency information deep indexing and feature association model extracts target words from the located keywords and their context. Then, based on the inverted index mechanism, it records the specific location of each high-frequency word in the document and related terms. Based on this index, the model further mines feature information strongly associated with high-frequency words, and finally stores the structured data in the power grid knowledge base, obtaining the second structured feature text. Its core functions include:

[0116] Dynamic inverted indexing mechanism: The model employs a hybrid indexing strategy, combining textual location information and semantic associations. For example, for high-frequency words such as "transmission channel / section," it not only records the document location where they appear but also associates them with features such as the transmission channel to which they belong and stability limits, forming multi-dimensional index entries. The index data includes word frequency weights, contextual fragments, and association evidence.

[0117] Feature association mining algorithm: First, electrical topology expansion is implemented. If high-frequency words involve power grid equipment, the topology adjacent sites are automatically queried in the mode calculation report. If they do not exist, the step is skipped. Then, the names of short-circuit current limiting measures are extracted. For terms containing "measures", specific operations are extracted through relevant rules.

[0118] Subsequently, the second structured feature text can be collaboratively analyzed using expert experience fusion and key feature optimization models, and processed using preset rules to obtain key power feature information. A closed-loop evaluation process of "machine initial screening - manual verification - feature optimization" can be formed by introducing a manual verification step involving dispatching operation experts and mode calculation engineers. Its core functions include:

[0119] Human-machine collaborative analysis mechanism: The model automatically generates a candidate set of key features, and the confidence level of the samples is marked by the frequency of keyword occurrence. An expert intervention is introduced to verify the physical rationality of strongly correlated electrical features and supplement implicit features that are not clearly defined but are actually important. Finally, the feature relationships confirmed by experts are written into the knowledge graph to optimize the subsequent automatic analysis model and realize the human-machine collaborative analysis mechanism.

[0120] Expert Rule Base Construction: The expert rule template base adopts a "three-layer, four-dimensional" systematic architecture, integrating various types of power grid operation knowledge in a structured manner. A dynamic adjustment mechanism for feature weights is set up, defining domain-specific weight formulas based on expert experience. This rule base includes three core modules: basic electrical rules, operational experience rules, and special scenario rules. Basic electrical rules cover mandatory standards such as equipment safety operation and power grid stability constraints; operational experience rules integrate typical handling cases of power grids in various regions; and special scenario rules address special operating conditions such as extreme weather and major incidents.

[0121] For example, after obtaining historical power operation characteristic data, historical power flow data can be extracted from this data, and a set of typical operating scenarios can be determined based on the power system node diagram and power and operation characteristic data. Mechanism analysis and knowledge mining under historical operating scenarios can then be achieved.

[0122] In one possible approach, determining a set of typical operating scenarios based on the power system node diagram and historical power flow data from historical power operation characteristic data includes:

[0123] The vulnerability index corresponding to the historical power flow data of each transmission line in the historical power operation feature data is calculated by the HITS algorithm (Hyperlink-Induced Topic Search). Multiple vulnerability indicators are obtained, and based on the multiple vulnerability indicators and the power system node diagram, the weak links of sub-targets in the historical operation scenario are determined.

[0124] Cluster analysis was performed on historical power operation characteristic data to obtain Typical operating scenarios, wherein, Each typical operating scenario includes its corresponding weak points.

[0125] Weak links of sub-targets in historical operational scenarios and Typical operating scenarios and their corresponding weaknesses are identified as a set of typical operating scenarios.

[0126] It should be understood that vulnerability indicators are core quantitative parameters used to quantify the ability of a system or model to withstand disturbances, faults, and attacks, as well as the degree of risk exposure. The HITS algorithm can measure the importance of a node in the network by analyzing the link relationships between nodes and calculating two key indicators for each node—hub value and authority value. In this embodiment of the invention, the vulnerability indicators corresponding to the historical power flow data of each transmission line can be calculated according to the HITS algorithm, and the weak links of sub-targets under historical operating scenarios can be determined based on multiple vulnerability indicators and the power system node diagram. One historical operating scenario can correspond to one weak link of a sub-target, or multiple weak links of a sub-target. A historical operating scenario can be a spatiotemporal snapshot constructed based on the actual operation records of the power grid, which can completely reflect the operating status of all elements of the power grid—source, grid, and load—within a specific time period. In this embodiment of the invention, the historical operating scenario can be obtained based on historical power flow data. Then, the historical power operation feature data can be clustered using the K-medoids clustering algorithm to obtain... Typical operating scenarios. The principle of the K-medoids clustering algorithm is as follows.

[0127] Because the K-medoids algorithm is quite sensitive to the selection of initial cluster centers, an unreasonable selection can easily lead to the clustering results getting trapped in local optima. Therefore, to improve the effectiveness of the clustering method, this invention modifies the selection of initial cluster centers. The specific steps for selecting initial cluster centers in the K-medoids algorithm are as follows:

[0128] Step 1: For a sample containing m elements, calculate the Euclidean distance between the samples to form the sample distance matrix D. m×m :

[0129] ;

[0130] in, For the feature vector x i To x j European distance, For the feature vector x i The k-th sample in the dataset, For the feature vector x j The k-th sample in the dataset, This represents the number of samples.

[0131] Step 2: Based on the distance matrix D, calculate the sum of distances between samples, select the sample point with the smallest sum of distances as the first cluster center, and denote it as c1;

[0132] Step 3: Calculate the average distance between samples :

[0133] ;

[0134] Step 4: Delete clusters smaller than the distance to the cluster center Sample points;

[0135] Step 5: Calculate the sum of distances between each unselected sample point and the determined cluster centers, and select the sample point with the largest sum of distances to the cluster centers as the next cluster center;

[0136] Step 6: Repeat steps 4-6 until k cluster centers are found.

[0137] Using the initial cluster center selection method, k initial cluster centers of the system are obtained, c = [c1, c2, ..., c...]. k After that, the steps for clustering the samples based on the K-medoids algorithm are as follows:

[0138] Calculate the Euclidean distance between each sample point and the cluster center, and assign the sample to the cluster with the nearest sample center;

[0139] Calculate the sum of squared deviations of each point within the cluster. Choose the point that minimizes the sum of squared deviations within the cluster as the new center for each cluster:

[0140] ;

[0141] in, Let y be the Euclidean distance from a sample y within a cluster to the cluster center. Let k be the set of samples contained in the k-th cluster.

[0142] Repeat the above steps until the center of the cluster no longer changes, or until the predetermined number of iterations is reached.

[0143] Based on the above improved strategy, using key features affecting the power grid operation status as input, an improved K-medoids algorithm is used to perform cluster analysis on historical operation scenarios to obtain... Typical operating scenarios.

[0144] Then, the weak points of sub-targets in historical operational scenarios and Typical operating scenarios and their corresponding weaknesses are identified as a set of typical operating scenarios.

[0145] In one possible approach, the identification of weak links in sub-targets under historical operating scenarios based on multiple vulnerability indicators and a power system node diagram includes:

[0146] Multiple vulnerability indicators are sorted from largest to smallest, and the transmission lines corresponding to the first preset number of vulnerability indicators are determined as the target lines in the historical operation scenario.

[0147] The power system node graph is processed by a depth-first search algorithm to generate the minimum cut set of the power system. The minimum cut set is then searched by recursive enumeration to generate candidate weak links.

[0148] Candidate weak links, including those along the target route, are identified as weak links in sub-targets under historical operational scenarios.

[0149] It should be understood that the specific implementation principle of calculating the vulnerability index corresponding to the historical power flow data of each transmission line in the historical power operation characteristic data using the HITS algorithm is as follows. The HITS algorithm reflects the importance of a node by indicating whether it points to a high-quality information node or is pointed to by a high-quality information node in the network. It sets the authority value and hub value of all nodes to 1, and calculates the authority value and hub value of each node iteratively based on the network link relationship. Specifically, the authority value of a node is the sum of the hub values ​​of all nodes linked to that node; the hub value of a node is the sum of the authority values ​​of all nodes linked to that node, as shown in the following calculation:

[0150] ;

[0151] ;

[0152] In the formula: and Let a represent the authority value and the hub value of node i, respectively; ij This represents the node link relationship. When node i points to node j, a ij =1.

[0153] After establishing the fault link matrix Z of the network, the links are abstracted as nodes in network theory, and the relationships between links are abstracted as connections between nodes. Furthermore, the authority value of the links is calculated based on the HITS algorithm. and hub value These represent the ability to be faulted by other lines and the ability of a line to cause faults in other lines, respectively. The iterative calculation process is shown below:

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] In the formula: and This represents the authority value and hub value of line i after the k-th iteration; and The result is normalized.

[0159] Finally, this invention proposes a line vulnerability index based on the network fault transfer matrix and the HITS algorithm. The top 10 routes based on criticality indicators were selected as the target routes for the system, and their calculation is shown below:

[0160] ;

[0161] In the formula: and These represent the final authority value and hub value of line i, respectively.

[0162] The power system node graph can then be processed using a depth-first search algorithm to generate the minimum cut set of the power system. A recursive enumeration method is then used to search for the minimum cut set to generate candidate weak links. The specific implementation process is as follows.

[0163] The power system node graph is processed using a depth-first search algorithm to generate the minimum cut sets of the power system. The depth-first search algorithm can use a stack as the core data structure and follow a "last-in, first-out" traversal rule. Then, a recursive enumeration method is used to search for candidate weak links based on the minimum cut set strategy. These candidate weak links are then matched according to the vulnerability ranking of important lines to obtain the weak links in historical operating scenarios. The specific steps are as follows:

[0164] Multiple vulnerability indicators are sorted from largest to smallest, and the transmission lines corresponding to the first preset number of vulnerability indicators are determined as the target lines in the historical operation scenario.

[0165] A recursive enumeration method is used to pre-search and screen the minimum cut sets of the power system as candidate weak links;

[0166] Candidate weak links, including those along the target route, are identified as weak links in sub-targets under historical operational scenarios.

[0167] For example, the weak links in the power system in future operating scenarios can then be identified based on future load levels, a set of typical operating scenarios, and predicted wind and solar power output.

[0168] In one possible approach, identifying the weak links in the power system under future operating scenarios based on future load levels, a set of typical operating scenarios, and predicted wind and solar power output includes:

[0169] The first feature vector is determined by identifying the future load level and the predicted wind and solar power output, and the second feature vector is determined by identifying the Euclidean distance between each second feature vector and the first feature vector in the typical operation scenario set. The typical operation scenario corresponding to the Euclidean distance that is less than or equal to the preset distance is identified as the target typical operation scenario.

[0170] Determine multiple first preselected weak links for the target typical operating scenarios, perform static security and stability verification on each first preselected weak link according to the future state load level and the predicted value of wind and light output, and determine the first preselected weak links that pass the verification as the weak links of the power system in the future operating scenarios.

[0171] It should be understood that the first eigenvector can be represented by the calculation formula F future = , , , and the second eigenvector can be F true = , , , where n represents the serial number of the clustering center of the typical operating scenarios, and 0 < n ≤ k. By calculating the Euclidean distance between F future and F true , form the sample distance matrix of various operating scenarios, and determine the typical operating scenarios corresponding to the preset distance less than or equal to or the typical operating scenario with the smallest distance as the target typical operating scenario, and the weak links under this target typical operating scenario are the first preselected weak links.

[0172] After that, static security and stability verification can be performed on each first preselected weak link according to the future state load level and the predicted value of wind and light output, and the first preselected weak links that pass the verification are determined as the weak links of the power system in the future operating scenarios. Among them, the static security and stability verification can be an analysis process of verifying whether the system can maintain the voltage, frequency, and power flow within the safety constraints and does not occur accidents such as voltage collapse and power flow overlimit after a preset disturbance under the current or预想 static operating state of the power grid.

[0173] In a possible way, the step of performing static security and stability verification on each first preselected weak link according to the future state load level and the predicted value of wind and light output, and determining the first preselected weak links that pass the verification as the weak links of the power system in the future operating scenarios includes:

[0174] Perform power flow calculation on each first preselected weak link according to the predicted value of wind and light output and the future state load level, obtain the first total active power of the transmission section of each first preselected weak link, and determine the first preselected weak links corresponding to the first total active power greater than or equal to the preset value as the second preselected weak links;

[0175] Perform stability verification on each second preselected weak link, and obtain the second total active power and the total long-term allowable current-carrying capacity of each second preselected weak link after the stability verification;

[0176] Based on the second total active power and the total long-term allowable current carrying capacity, weak links in the power system in future operating scenarios are identified.

[0177] It should be understood that, in the specific calculation process, the vulnerability indicators corresponding to each first pre-selected weak link are ranked, and the first total active power of each first pre-selected weak link is calculated. The first total active power is the sum of the active power on each transmission line. The thermal stability safety constraint for each transmission line can be expressed by the following formula:

[0178] ;

[0179] ;

[0180] in, U represents the active power of transmission section g, where g belongs to the pre-selected weak link set C; N Indicates the rated voltage; I max Indicates the current limit of the circuit; T c This represents the temperature correction factor, which is affected by both the ambient temperature and the conductor's operating temperature. Typically, the ambient temperature is taken as 25℃ and the conductor temperature as 80℃. Under these conditions, T... c =1; This represents the power factor, typically taken as 0.9 or 0.95. The heavy load rate threshold can be set according to actual needs. This allows for the establishment of cross-sectional heavy load rate evaluation indicators. :

[0181] ;

[0182] in, The weak link g represents the sum of the long-term allowable current carrying capacity of all lines except the most critical transmission line.

[0183] Next, the first pre-selected weak links corresponding to a first total active power greater than or equal to a preset first weak link are identified as second pre-selected weak links. For each second pre-selected weak link, stability verification is performed, and the second total active power and total long-term allowable current carrying capacity of each second pre-selected weak link are obtained after stability verification. Based on the second total active power and total long-term allowable current carrying capacity, weak links in the power system in future operating scenarios are determined. Specifically, key weak links are identified in a refined manner based on the "N-1" stability verification principle and the "N-1" principle for planned maintenance; otherwise, they are abandoned. Then, a safety margin is set. Taking the weak link g as an example, we formulate its "N-1" stability constraints and planned maintenance "N-1" constraints:

[0184] ;

[0185] in, This indicates the active power of the weak link g under the "N-1" or planned maintenance "N-1" mode; Set according to actual needs.

[0186] ;

[0187] in, This indicates the calculation margin for weak points under planned maintenance methods. The value range is [0, +∞], usually A value of ≥1 indicates that the weak link has exceeded the limit and has failed the static security and stability verification; otherwise, the verification is considered successful. The second pre-selected weak link corresponding to the one that passes the verification is identified as a weak link in the future operation scenario.

[0188] In the specific implementation and comparison process, in order to verify the effectiveness and adaptability of the proposed AlexNet-based model in actual power grid applications, actual wind power and photovoltaic power output data of a certain province were selected as verification objects.

[0189] First, wind power and photovoltaic power output values ​​from January to December 2018 were collected from a wind farm in a certain province. This sample dataset also included eight meteorological characteristic indicators, such as wind direction, wind speed, ambient temperature, atmospheric pressure, and atmospheric humidity, for the collection dates. The data resolution was 5 minutes, consistent with the quantification standard for ultra-short-term forecasts. To ensure the rationality of the simulation results, the sample set was divided into four time scales, with the 8th, 15th, 22nd, and 29th of each month (February 28th) totaling 48 days, 288 samples, and 13,824 sample points selected as the test set to verify the model's performance. The remaining data points were used as the model training set. All sample sets were cleaned before simulation analysis.

[0190] Secondly, photovoltaic power output values ​​from a photovoltaic power station in a certain province, collected from January to March 2018, were selected. This sample dataset also included eight meteorological characteristic indicators for the collection dates, such as direct solar irradiance, diffuse solar irradiance, wind direction, wind speed, ambient temperature, and relative humidity. The data collection resolution was 5 minutes, consistent with the quantification standard for ultra-short-term forecasts. To enhance the readability of the experiment, data nodes with zero solar irradiance in the morning and evening were removed, retaining only 48 sample points from 7:00 to 16:55 each day. The 12 sets of samples (576 sample points) from March 8th, 15th, 22nd, and 29th were selected as the test set to verify the model's performance, while the remaining data points served as the model training set. All sample sets underwent data cleaning before simulation analysis.

[0191] Specific analysis results: Based on a data-driven approach, historical meteorological data features were used as input features for the model. The TCN-LSTM algorithm in deep learning was selected as a comparison algorithm against the AlexNet model. Four-hour periods were extracted from the test set on January 8th, April 8th, July 8th, and October 8th respectively as test targets to verify the prediction performance of the AlexNet model for ultra-short-term wind power prediction in different seasons. The wind power prediction results for the target area are shown in Table 1. For details, please refer to... Figure 4 .

[0192] Table 1. Prediction accuracy of wind power clusters in different seasons based on the AlexNet model. .

[0193] Compared to TCN-LSTM in machine learning, the AlexNet model shows improvements in all three evaluation metrics across different seasons, demonstrating significantly superior predictive performance. The σ values ​​for the test targets in different seasons... RMSE None of the indicators exceeded 0.7, with the lowest being only 0.29, resulting in an overall improvement of 33.09% compared to the TCN-LSTM method; its average R² reached 0.80, an improvement of 8.11% compared to the SDA method; σ MAPE The main reason for the high index is that it is easily affected by near-zero values, so the index fluctuates relatively much, but it can still remain within 0.50. Based on current research results, after excluding the influence of objective factors such as data differences, the obtained evaluation index fully verifies the effectiveness of the model.

[0194] Data-driven approaches can effectively reflect the impact of input features on output results and make it easier to build spatial mapping models with higher adaptability, which is something that model-driven approaches cannot match.

[0195] This example uses two deep learning models, AlexNet and TCN-LSTM, for modeling. XgBoost regression, a high-performing classic regression model, is chosen as the comparison. The study aims to predict photovoltaic output power from 48 sample points between 7:00 and 10:55 on March 8th, testing whether data-driven and the proposed AlexNet model can effectively handle the ultra-short-term photovoltaic power prediction problem. The simulated ultra-short-term photovoltaic power prediction curve is shown below. Figure 5 As shown in Table 2, the specific results of the predictive indicator analysis can be found therein.

[0196] Table 2. Prediction accuracy of photovoltaic power based on the AlexNet model .

[0197] As shown in Table 2, all prediction methods employ a data-driven approach. A relatively mature deep learning algorithm is compared with an ensemble learning algorithm. The AlexNet model proposed in this embodiment significantly outperforms the two classic algorithms compared. Among the three evaluation metrics, There is at least a 19.89% improvement. It also increased by at least 1.46%, R 2 There was also an improvement of 0.19-0.22. Although the complexity of the proposed AlexNet model is more severe than the other two methods, its prediction performance and stability have achieved significant breakthroughs.

[0198] As stated above, the AlexNet algorithm demonstrates superior prediction performance in wind power and photovoltaic power prediction problems compared to the XgBoost and TCN-LSTM algorithms, which have been validated by multiple sources. Furthermore, it ensures model stability while maintaining a data-driven foundation, resulting in a comprehensive performance improvement for the proposed model. Therefore, this invention selects the AlexNet method as the primary approach for addressing the ultra-short-term prediction of renewable energy output. Analysis of the above calculation results shows that the method proposed in this invention can guarantee the feasibility and rationality of renewable energy output prediction.

[0199] To verify the effectiveness and adaptability of the target line and weak link identification method in this embodiment, the standard IEEE 39-bus system was selected as the test target. The future load level, unit combination status, and topology of the system were adjusted to construct different operating scenarios. The load was set to fluctuate between 70% and 130% of the standard load, and topology changes were achieved by line disconnection. Meanwhile, since the generator output channel is the most important line in the system, it was not considered during the line identification process.

[0200] Specific analysis results: Vulnerability indices for each transmission line based on historical power flow data calculated using the HITS algorithm can be found in [link to analysis]. Figure 6 Furthermore, the transaction volume between generators and load nodes is calculated, and the top ten generator-load node pairs with the highest transaction volume are selected to filter the search range of the transmission channel. The specific transaction volume ranking is shown in Table 3.

[0201] Table 3 Ranking of Active Power Transactions between Generator and Load Nodes in the IEEE-39 System .

[0202] Depend on Figure 6As shown in Table 3, the identified target lines all correspond to the critical transmission channels between the respective generator-load node pairs. For each generator-load node pair, the matched line set almost covers its most important transmission paths; once these lines are disconnected, the load's power supply capacity will be significantly limited. In particular, lines carrying large power flows not only determine the power transmission within their respective sets but also significantly affect the power flow distribution of other lines in the entire network. For example, after lines 16-19 are disconnected, the power flow of the system will no longer converge.

[0203] The vulnerability index of the transmission lines in this embodiment is established based on the importance of the lines and the power flow changes caused by line faults. Therefore, this embodiment verifies the effectiveness of the proposed critical line identification method by analyzing the identification results. The specific results are as follows: Line 6-11 is the most important line because its base state load rate is 79.8%. At the same time, line 6-11 is the transmission channel of generator 32. When lines 10-13 and 13-14 fail, line 6-11 will carry the transferred power flow, and its load rate will reach 99.8%, causing the line transmission power to approach its limit. If line 6-11 fails at this time, generator 32 will be disconnected from the grid, causing the system to lose a large amount of power. Meanwhile, line 6-11, together with lines 4-14, 15-16, and 16-17, constitutes the transmission channel for generators 33, 34, 35, and 36 to meet the central load demand of the system. For example, when line 4-14 fails, to meet the load demand of node 4, the power flow will shift to lines 6-11 and 16-17. At this time, the power flow of line 6-11 will be close to its limit. If line 6-11 fails at this time, line 16-17 will bear the transferred power flow and reach its transmission limit, causing a cascading failure in the system and resulting in system disconnection. Generators 32, 33, 34, 35, and 36 will form islands, causing system load loss. Therefore, line 6-11 not only has a high failure frequency, but the load loss caused by its failure is also relatively severe, so line 6-11 has a high criticality. At the same time, the base state load rate of line 3-18 is only 0.36%, and there are multiple power flow transfer lines such as lines 25-26, 17-18, and 4-14. Moreover, line 3-18 is not a generator output channel, so it is not likely to cause generator disconnection accidents. Therefore, its criticality is low. In summary, the method disclosed in this embodiment can effectively identify the critical lines that are prone to causing system failure.

[0204] The preliminary search for weak links revealed 33 combinations of 2 lines, 90 combinations of 3 lines, 197 combinations of 4 lines, and 300 combinations of 5 lines. When matching candidate weak links, they were sorted by the transaction volume between generators and load nodes from highest to lowest, prioritizing weak links with higher vulnerability indices and fewer candidate paths. Of the 10 identified critical transmission lines, 6 corresponding cut sets were matched. Specific results are shown in Table 4.

[0205] Table 4 Weaknesses of Sub-targets at IEEE-39 Node .

[0206] To verify the rationality of the partitioning method, the quality of partitioning is usually evaluated by whether the weak link lines between two partitions carry a large power flow and whether the lines are closely connected. Simulation analysis shows that the power flow transfer coefficients between the weak link lines after partitioning using the method disclosed in this embodiment are all greater than 0.2. Furthermore, during the N-1 safety check analysis, the weak links were found to have a risk of thermal instability. Therefore, the partitioning method disclosed in this embodiment is relatively reasonable, and the resulting weak link lines are closely connected.

[0207] To verify the effectiveness and adaptability of the power system operation scenario clustering and scenario matching method and the weak link identification method in future operation scenarios disclosed in this embodiment, the IEEE 39-node system with renewable energy integration was selected as the test target. The traditional generating units connected to nodes 35 and 37 were replaced with wind turbines (W) and photovoltaic units (P), respectively, to test the proposed method and verify the comprehensive performance of the proposed critical section identification method. See the diagram of the IEEE 39-node system considering renewable energy integration. Figure 7 Among them, G1 to G8 are thermal power units, and GB is a balanced unit. The system load level was set to vary within the range of 70% to 130% of the rated operating condition, and a total of 1958 convergent samples were generated as the test set.

[0208] This invention presents a power system operation scenario clustering and scenario matching method based on the K-medoids clustering algorithm. First, based on the key power feature information obtained using a key feature extraction model, features related to unit start-up / shutdown and system topology linkage are excluded. Considering the discriminative power and effectiveness of the features, preliminary screening yields 21-dimensional features representing node active and reactive power; 37-dimensional features representing node voltage; 29-dimensional features representing node phase angle; 7-dimensional features representing generator active power output; and 8-dimensional features representing generator reactive power output. Principal components that fully represent the sample features are further selected. The dimensionality of these 8-dimensional features is reduced, and the principal components of all features are combined for further clustering.

[0209] After extracting key power feature information, historical power operation feature data is obtained through the key power feature information. For the historical power flow data in the historical power operation feature data, the number of cluster centers is set from 2 to 70, and the average intra-class distance of samples under different k values ​​is calculated. The result k=23 is taken as the optimal number of clusters.

[0210] To verify the effectiveness of the proposed methods for selecting the optimal number of clusters and the initial cluster center selection, the following four schemes were set up: Scheme 1: Selecting the optimal number of clusters as... (n is the number of samples, and the initial cluster centers are random); Scheme 2: Select the optimal number of clusters as... And improve the initial cluster centers; Scheme 3: Select the optimal number of clusters as 23, and the initial cluster centers are random; Scheme 4: Select the optimal number of clusters as 23 and improve the cluster centers.

[0211] Select an initial sample S0, and compare it with Scheme 1 and Scheme 2 (cluster number) respectively. Similarity matching is performed on the running scenarios corresponding to the cluster centers of Scheme 3 (optimal number of clusters, random initial centers) and Scheme 4 (double optimization). Figure 8 As shown, Scheme 1 is not compared because its results show a large deviation.

[0212] The absolute value curves of the load active power change rate of different schemes show that the load distribution corresponding to the operation scenario of scheme 4 is most similar to the initial sample, while the operation scenarios corresponding to schemes 2 and 3 have obvious deviations in the peak and valley sections.

[0213] Based on the simulated wind and solar forecasts for nodes 35 and 37, and assuming the future load level of the power system, a comprehensive thermal stability analysis was conducted on potential weak points. The identification results of weak points in the future operating scenario are shown in Table 5.

[0214] Table 5. Identification Results of Weak Links and Critical Paths in the Improved IEEE 39-Node System .

[0215] The test platform verified that the simulation of weak link identification using the enumeration method with 200 sets of samples took 1889 minutes (approximately 31.5 hours), with an average time of 566.7 seconds per set of samples. In contrast, the proposed method, through the scenario pattern matching mechanism, can complete the identification of critical lines and weak links in only 2.1 seconds. Experimental data confirms that the method reduces the processing latency to the second level while ensuring identification accuracy, meeting the timeliness requirements of real-time safety assessment of power systems and providing effective technical support for online pre-decision making.

[0216] The above technical solutions significantly improve the accuracy and interpretability of the identification results. By utilizing representation learning to automatically extract key features of power grid operation and leveraging clustering algorithms to mine historical operational patterns, the adaptive and intelligent advantages of the algorithm are demonstrated; it can accurately locate structural weaknesses in the system. This fusion strategy makes the identification process both data-driven and forward-looking, while maintaining the rigor of the physical model. The results are more easily accepted and trusted by domain experts, facilitating their application in actual dispatch and control; it effectively addresses the assessment challenges arising from the uncertainty on both the source and load sides under high-proportion renewable energy access. By determining the future operational boundary through high-precision prediction, similar cases are quickly matched in historical operational scenarios, using their set of weak points as initial references, and then refined power flow and thermal stability verification is performed only on the critical path. This method avoids the complex stability calculations of the entire network under massive scenarios in traditional enumeration methods, greatly improving analysis efficiency while ensuring the forward-looking and accurate results, and meeting the actual operational needs of online simulation and rapid decision-making in the new power system environment.

[0217] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying weak links of a future state power system based on wind and solar power output, characterized in that, include: The node diagram of the power system, the meteorological forecast data of the area where the power system is located in the future, the future load level, and the historical power operation characteristics data of the power system in the historical operation process are determined. The power system is a multi-energy power system including wind and solar power output. Meteorological forecast data is input into the target wind and solar power output prediction model to obtain the wind and solar power output prediction value of the power system. The target wind and solar power output prediction model is trained by historical wind and solar power output values ​​and corresponding historical meteorological data from historical power operation characteristic data. Based on the power system node diagram and historical power flow data in historical power operation characteristic data, a set of typical operating scenarios is determined. Each typical operating scenario in the set of typical operating scenarios includes a corresponding weak link. Based on future load levels, typical operating scenarios, and wind and solar power output forecasts, weak links in the power system in future operating scenarios are identified.

2. The method for identifying weak links in a future power system based on wind and solar power output according to claim 1, characterized in that, The target wind and solar power output prediction model was obtained through the following method: Determine the historical wind and solar power output values ​​in the historical power operation characteristic data, as well as the historical meteorological data corresponding to the historical wind and solar power output values; The target wind and solar power output prediction model is obtained by training the wind and solar power output prediction model with historical wind and solar power output values ​​and historical meteorological data.

3. The method for identifying weak links in a future power system based on wind and solar power output according to claim 1, characterized in that, The historical power operation characteristic data was obtained through the following method: Acquire historical power data and target files of the power system during its historical operation, wherein the target files are used to configure the power operation feature data to be extracted; Extract key power feature information from the target file, and determine historical power operation feature data based on the key power feature information and historical power operation data. The key power feature information is obtained by processing the historical power operation report through a key feature extraction model.

4. The method for identifying weak links in a future power system based on wind and solar power output according to claim 3, characterized in that, The process of determining historical power operation characteristic data based on key power characteristic information and historical power operation data includes: Based on the power station affiliation information of key power characteristics, the power stations where historical operating power data is located are divided into regions, and the load data of each region is calculated. Based on the cross-sectional lines with key power characteristics information, determine the AC line power flow data in the historical power operation data, and calculate the power flow value corresponding to the cross-sectional lines based on the AC line power flow data. Based on the key power characteristics information of the DC converter station, the target lines flowing to the DC converter station in the historical power data are determined, and the power values ​​of the target lines are summarized. The total power of each DC transmission channel is calculated based on the power values ​​of the target lines, wherein each DC transmission channel includes a pair of DC converter stations. By integrating the load data of each region, the power flow value of the corresponding cross-section line, and the total power of each DC transmission channel into the target file, historical power operation characteristic data is obtained.

5. The method for identifying weak links in a future power system based on wind and solar power output according to claim 3, characterized in that, The key feature extraction model includes a keyword extraction model, a high-frequency information deep indexing and feature association model, and an expert experience fusion and key feature optimization model; the key power feature information is obtained through the following methods: Obtain historical power operation reports generated by the power system during its historical operation; The key power feature words that are related to the spatiotemporal distribution characteristics of power grid flow are extracted from the historical power operation report by the keyword extraction model. The location and contextual information of the key power feature words in the historical power operation report are determined. The key power feature words, their location and contextual information are stored in a structured manner to obtain the first structured feature text. The target words in the first structured feature text are extracted by high-frequency information deep indexing and feature association model. The target words are then processed by dynamic inverted indexing mechanism and feature association mining algorithm to obtain the second structured feature text. The target words are words that appear more than or equal to a preset number of times in the first structured feature text. By integrating expert experience and using a key feature optimization model to perform collaborative analysis on the second structured feature text and processing the second structured feature text using preset rules, key power feature information is obtained.

6. The method for identifying weak links in a future power system based on wind and solar power output according to claim 1, characterized in that, The process involves determining a set of typical operating scenarios based on the power system node diagram and historical power flow data from historical power operation characteristic data, including: The vulnerability index corresponding to the historical power flow data of each transmission line in the historical power operation characteristic data is calculated by the HITS algorithm. Multiple vulnerability indices are obtained, and based on the multiple vulnerability indices and the power system node diagram, the weak links of sub-targets in the historical operation scenario are determined. Cluster analysis was performed on historical power operation characteristic data to obtain Typical operating scenarios, wherein, Each typical operating scenario includes its corresponding weak points. Weak links of sub-targets in historical operational scenarios and Typical operating scenarios and their corresponding weaknesses are identified as a set of typical operating scenarios.

7. The method for identifying weak links in a future power system based on wind and solar power output according to claim 6, characterized in that, The process of identifying weak links in sub-targets under historical operating scenarios based on multiple vulnerability indicators and power system node diagrams includes: Multiple vulnerability indicators are sorted from largest to smallest, and the transmission lines corresponding to the first preset number of vulnerability indicators are determined as the target lines in the historical operation scenario. The power system node graph is processed by a depth-first search algorithm to generate the minimum cut set of the power system. The minimum cut set is then searched by recursive enumeration to generate candidate weak links. Candidate weak links, including those along the target route, are identified as weak links in sub-targets under historical operational scenarios.

8. The method for identifying weak links in a future power system based on wind and solar power output according to claim 1, characterized in that, The process of identifying weak links in the power system in future operating scenarios based on future load levels, typical operating scenario sets, and wind and solar power output forecasts includes: The first feature vector is determined by identifying the future load level and the predicted wind and solar power output, and the second feature vector is determined by identifying the typical operation scenario in the typical operation scenario set. The Euclidean distance between each second feature vector and the first feature vector is calculated, and the typical operation scenario corresponding to the Euclidean distance that is less than or equal to the preset distance is identified as the target typical operation scenario. Multiple first-selected weak links in the typical operating scenario of the target are identified, and static security and stability verification is performed on each first-selected weak link based on the future load level and the predicted wind and solar power output. The first-selected weak links that pass the verification are identified as weak links in the power system in the future operating scenario.

9. The method for identifying weak links in a future power system based on wind and solar power output according to claim 8, characterized in that, The process of performing static security and stability verification on each first pre-selected weak link based on future load levels and predicted wind and solar power output, and identifying the first pre-selected weak links that pass the verification as weak links in the power system in future operating scenarios, includes: Based on the predicted wind and solar power output and the future load level, power flow calculations are performed on each first pre-selected weak link to obtain the first total active power of the transmission section of each first pre-selected weak link, and the first pre-selected weak link corresponding to the first total active power that is greater than or equal to the preset first total active power is determined as the second pre-selected weak link. For each second pre-selected weak link, a stability check is performed, and the second total active power and total long-term allowable current carrying capacity of each second pre-selected weak link are obtained after the stability check. Based on the second total active power and the total long-term allowable current carrying capacity, weak links in the power system in future operating scenarios are identified.

10. The method for identifying weak links in a future power system based on wind and solar power output according to claim 9, characterized in that, The determination of weak links in the power system in future operating scenarios based on the second total active power and the total long-term allowable current carrying capacity includes: The second total active power is divided by the total long-term allowable current carrying capacity to obtain the quotient value. The second pre-selected weak link corresponding to the quotient value greater than the preset quotient value is identified as the weak link in the future operation scenario.

11. The method for identifying weak links in a future power system based on wind and solar power output according to any one of claims 1-10, characterized in that, The node diagram of the power system was obtained through the following method: By treating parallel buses in a power system as a single bus, treating multiple lines between two nodes as a single line, and removing independent nodes, a node diagram of the power system is obtained.

Images