A power grid transient stability analysis and control method, system and computer readable storage medium
Patent Information
- Application Number
- CN202611023769.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-10
- Publication Date
- 2026-10-27
AI Technical Summary
但现有研究多集中于单一模型或简单组合，缺乏针对电网暂态稳定问题的系统化建模方法，尤其是在多节点耦合、扰动传播以及实时在线预测方面仍存在不足
[0035]本发明基于相量测量单元同步测量数据构建电网运行数据模型，能够获取高时间分辨率和高精度的电网动态信息，为稳定性分析提供可靠的数据基础，解决了传统方法依赖精确系统模型参数、适应性差的问题；
Smart Images
Figure CN122992724A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method, system, and computer-readable storage medium for power grid transient stability analysis and control. Background Technology
[0002] With the continuous expansion of power system scale and the sustained increase in the proportion of renewable energy grid connection, power grid operation is becoming increasingly complex, and the dynamic characteristics and uncertainties of the system are significantly enhanced. The volatility and intermittency of renewable energy sources such as wind and solar power make the power grid more susceptible to transient instability when subjected to disturbances or faults, seriously threatening the safe and stable operation of the power system. Therefore, how to achieve rapid and accurate assessment of the transient stability state of the power grid has become an important research direction in the field of power system operation and control.
[0003] Traditional methods for analyzing power grid transient stability mainly include numerical integration methods based on time-domain simulation and analytical methods based on energy functions. While time-domain simulation methods offer high accuracy, they are computationally intensive and lack real-time performance, making them unsuitable for online analysis. Energy function methods, while improving computational efficiency to some extent, have limited applicability under complex power grid structures and multiple disturbance conditions. Furthermore, these methods typically rely on accurate system model parameters; changes in the power grid's operating state or inaccurate models can affect the reliability of the analysis results.
[0004] In recent years, with the development of Wide Area Measurement Systems (WAMS), Phasor Measurement Units (PMUs) have been able to acquire power grid operating status information with high sampling rates and high time synchronization accuracy, providing an important data foundation for real-time monitoring and analysis of power grid dynamic behavior. However, existing analysis methods based on PMU data mostly employ traditional statistical analysis or shallow machine learning methods, which are insufficient to fully exploit the complex coupling characteristics of power grid operating data in the spatial and temporal dimensions, resulting in limited accuracy and generalization ability in stability assessment.
[0005] Meanwhile, deep learning methods have demonstrated significant advantages in time-series data modeling and feature extraction. For example, Graph Convolutional Networks (GCNs) are strong in spatial feature extraction, while Long Short-Term Memory Networks (LSTMs) perform well in handling dynamic changes in time series data. However, existing research largely focuses on single models or simple combinations, lacking systematic modeling methods for power grid transient stability problems, particularly in areas such as multi-node coupling, disturbance propagation, and real-time online prediction. Furthermore, most methods only address stability assessment and lack strategy generation mechanisms integrated with actual power grid operation and control, making it difficult to directly guide engineering applications.
[0006] Therefore, the existing technology still has the following shortcomings: First, it is insufficient to model the spatiotemporal coupling characteristics of the power grid operation state, making it difficult to accurately reflect the dynamic behavior of complex power grids; second, it is difficult to balance the real-time performance and accuracy of stability prediction; and third, it lacks a closed-loop control strategy based on the prediction results, which cannot effectively support the decision-making for safe operation of the power grid. Summary of the Invention
[0007] The present invention provides a method, system, and computer-readable storage medium for power grid transient stability analysis and control in order to solve the problems existing in the prior art.
[0008] The technical solutions adopted in this invention are as follows:
[0009] A method for power grid transient stability analysis and control includes the following steps:
[0010] S1: Deploy phasor measurement units at key nodes of the power grid to collect power grid operation status data in real time and perform unified time synchronization processing on the power grid operation status data of multiple nodes;
[0011] S2: Preprocess and feature-construct the power grid operation status data, and construct a multi-node time-series feature dataset in combination with the power grid topology;
[0012] S3: Construct a power grid transient stability prediction model. The model includes a graph convolutional network, a long short-term memory network, and an attention mechanism module. The graph convolutional network performs spatial aggregation on the node features in the multi-node temporal feature dataset based on the power grid topology adjacency matrix to extract spatial features. The long short-term memory network dynamically models the time series features in the multi-node temporal feature dataset to extract temporal features. The attention mechanism module adaptively weights the spatial features and temporal features respectively, and then fuses the weighted spatial features and the weighted temporal features to obtain fused spatiotemporal features.
[0013] S4: Use historical operating data and simulation data to train and optimize the parameters of the power grid transient stability prediction model to obtain the trained power grid transient stability prediction model;
[0014] S5: The real-time collected power grid operation status data is preprocessed and feature constructed in S2 and then input into the trained power grid transient stability prediction model to predict the power grid transient stability online. The model outputs stability / instability classification results, stability margin prediction values, and instability type classification results. The instability type classification results include power angle instability trend, voltage stability decline, and severe instability risk.
[0015] S6: Generate corresponding safety control strategies based on the stability / instability classification results, stability margin prediction values, and instability type classification results.
[0016] Furthermore, the power grid operating status data includes voltage phasors, current phasors, frequency, active power, and reactive power;
[0017] The preprocessing includes outlier removal, missing data repair, noise filtering, and normalization.
[0018] Furthermore, the feature construction adopts a sliding time window method to construct the power grid operation status data into a three-dimensional feature tensor. The dimensions of the three-dimensional feature tensor include time dimension, node dimension and feature dimension.
[0019] Furthermore, the graph convolutional network performs multi-layer graph convolution operations based on the power grid topology adjacency matrix to extract the electrical coupling relationships and spatial distribution characteristics between different nodes.
[0020] Furthermore, the Long Short-Term Memory (LSTM) network models the long-term dependencies in the dynamic process of the power grid through forget gates, input gates, output gates, and cell states.
[0021] Furthermore, the attention mechanism module includes a spatial attention mechanism and a temporal attention mechanism. The spatial attention mechanism is used to weight the spatial features, and the temporal attention mechanism is used to weight the temporal features.
[0022] Furthermore, in S4, a joint loss function is used to train the power grid transient stability prediction model. The joint loss function includes stability / instability classification loss, stability margin prediction loss, and instability type classification loss.
[0023] Furthermore, the safety control strategy is generated by constructing a control optimization objective function, and satisfies power balance constraints, voltage constraints, and generator output constraints;
[0024] When the stability / instability classification result is stable, maintain the current operating mode;
[0025] When the stability / instability classification result is instability, corresponding control measures are generated based on the instability type classification result: when the instability type classification result is power angle instability trend, generator output adjustment is executed;
[0026] When the instability type classification result is a decrease in voltage stability, the reactive power compensation device is switched on and off and the voltage is regulated.
[0027] When the instability type classification result is a severe instability risk, load reduction control is implemented.
[0028] This invention also discloses a power grid transient stability analysis and control system, including...
[0029] The data acquisition module is used to deploy phasor measurement units at key nodes of the power grid and collect power grid operation status data in real time, as well as to perform unified time synchronization processing on power grid operation status data from multiple nodes.
[0030] The data processing module is used to preprocess and construct features from power grid operation status data, and to construct a multi-node time-series feature dataset in combination with the power grid topology.
[0031] The stability prediction module includes a power grid transient stability prediction model, which is used to predict the transient stability state of the power grid based on the multi-node time-series feature dataset, and outputs stability / instability classification results, stability margin prediction values, and instability type classification results.
[0032] The control decision module is used to generate corresponding safety control strategies based on the stability / instability classification results, stability margin prediction values, and instability type classification results.
[0033] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any one of claims 1 to 8.
[0034] The present invention has the following beneficial effects:
[0035] This invention constructs a power grid operation data model based on synchronous measurement data from phasor measurement units, which can acquire high temporal resolution and high precision power grid dynamic information, providing a reliable data foundation for stability analysis and solving the problems of traditional methods relying on precise system model parameters and having poor adaptability.
[0036] This invention achieves joint modeling of the power grid's operating state in both spatial and temporal dimensions by fusing graph convolutional networks and long short-term memory networks and introducing an attention mechanism module. This fully explores the coupling relationships between data from multiple nodes, thereby significantly improving the accuracy of power grid transient stability prediction and solving the problem of insufficient modeling of the spatiotemporal coupling characteristics of the power grid in existing technologies.
[0037] This invention uses a data-driven approach to replace the traditional model-based analysis method, avoiding dependence on precise system parameters and exhibiting better adaptability and generalization ability.
[0038] This invention enables rapid online prediction of the transient stability state of the power grid, has high real-time performance, and is suitable for dynamic security assessment in complex power grid environments. It solves the problems of large computational load and poor real-time performance of traditional time-domain simulation methods.
[0039] This invention introduces a safety control strategy generation mechanism based on stability prediction, realizing a closed-loop process of "prediction-decision-control". Based on the stability / instability classification results, stability margin prediction values and instability type classification results, corresponding safety control strategies are generated, which effectively improves the safety and reliability of power grid operation and solves the problem that existing methods only stay at the discrimination level and lack a mechanism to combine with actual control. Attached Figure Description
[0040] Figure 1 This is a schematic diagram of the topology of the IEEE-39 node standard test system.
[0041] Figure 2 This is a schematic diagram of the power grid transient stability prediction model.
[0042] Figure 3(a) shows the curves of accuracy and precision as a function of prediction time.
[0043] Figure 3(b) shows the curves of false positive rate and false discovery rate as a function of prediction time.
[0044] Figure 3(c) shows the box plots of the performance indicators of each model.
[0045] Figure 4(a) shows the accuracy variation curve.
[0046] Figure 4(b) shows the accuracy rate variation curve.
[0047] Figure 4(c) shows the curve of the false positive rate.
[0048] Figure 4(d) shows the curve of false detection rate. Detailed Implementation
[0049] The invention will now be further described with reference to the accompanying drawings.
[0050] This embodiment provides a detailed description of the technical solution of the present invention based on the IEEE-39 node standard test system. The structure of the test system is as follows: Figure 1 As shown, it includes multiple generator nodes, load nodes, and transmission lines, which can completely simulate the operating characteristics and transient disturbance processes of a real power grid. The specific implementation process of the power grid transient stability prediction and security control method based on phasor measurement units and deep learning described in this embodiment is as follows.
[0051] The first step is to collect and synchronize power grid operation data.
[0052] Phasor measurement units are deployed at key bus nodes, generator output nodes, and both ends of important transmission channels in the power grid to collect multi-source heterogeneous operation data of the power grid, including phasor measurement unit measurement data, monitoring and data acquisition system data, power grid topology data, generator operating parameters, and load data.
[0053] The phasor measurement unit collects grid operation status data such as voltage phasors, current phasors, system frequency, active power and reactive power of each node in real time, and performs time synchronization processing on the grid operation status data of multiple nodes through a unified high-precision time reference to form a multi-node time-series measurement dataset.
[0054] This method consists of two phases: online operation and offline training. In the online operation phase, the power grid operation status data collected by the phasor measurement unit is acquired for real-time transient stability prediction. In the offline training phase, a set of key anticipated faults can be constructed by combining historical operation records and simulation analysis results for training sample generation, label construction, and control strategy verification.
[0055] For any node in the power grid At any given moment The voltage phasor is represented as:
[0056] ,
[0057] in, Represents a node At any moment The voltage phasor, Indicates voltage amplitude. Indicates the voltage phase angle. It represents the imaginary unit.
[0058] Correspondingly, the current phasor is represented as:
[0059] ,
[0060] in, Represents current phasor. Indicates the current amplitude. This indicates the phase angle of the current.
[0061] Based on the above phasor relationship, the active and reactive power at the nodes can be calculated:
[0062] ,
[0063] ,
[0064] in, Indicates the active power of the node. This represents the reactive power at the nodes. The system frequency is expressed as the rate of change of the voltage phase angle:
[0065] ,
[0066] in, Indicates the node frequency.
[0067] At discrete time Next, construct the power grid state vector:
[0068] ,
[0069] Where N represents the number of power grid nodes.
[0070] Further construct the time series data matrix:
[0071] ,
[0072] in, Representing feature dimension, Indicates the length of the time series.
[0073] To ensure consistency of multi-source data over time, time synchronization is performed on the data at each node to meet the following requirements:
[0074] ,
[0075] When communication delays or data gaps exist, time alignment is performed on the data:
[0076] ,
[0077] in, This represents the aligned data.
[0078] By combining historical operating data and simulation analysis results, a fault feature vector is constructed:
[0079] ,
[0080] in, Indicates the fault type. Indicates the location of the fault. Indicates the time when the fault occurred. Indicates the time when the fault is cleared.
[0081] By combining historical operational data and simulation analysis results, a fault feature vector is constructed during the offline training phase:
[0082] ,
[0083] in, This represents the fault type of the r-th fault scenario. Indicates the location of the fault. Indicates the time when the fault occurred. Indicates the time when the fault is cleared.
[0084] Further construct a set of key anticipated faults:
[0085] ,
[0086] in, Indicates the number of fault scenarios.
[0087] It should be noted that the fault feature vector and key anticipated failure set It is primarily used for offline simulation sample generation, training label construction, and control strategy verification, and is not a required input for the deep learning model in the online prediction phase. In the online prediction phase, the input to the deep learning model is the spatiotemporal characteristics constructed based on phasor measurement unit (PMU) measurement data and the power grid topology.
[0088] Based on a set of key anticipated faults, power grid operation state sequences under different fault scenarios are generated through offline simulation or historical data playback. Based on the simulation results, corresponding stability labels, stability margin labels, and instability type labels are constructed to form a training dataset.
[0089] ,
[0090] in, This represents the r-th running state time series sample. Indicates stable / unstable classification labels. Indicates stability margin label, Indicates the type of instability label. Indicates the number of training samples. Fault feature vector. Used for offline sample generation and label construction, not as input variables for the online prediction stage model.
[0091] This embodiment selects the IEEE-39 node standard test system as the research object, and its structure is as follows: Figure 1 As shown, the system includes generator nodes, load nodes, and transmission lines, and is used to simulate the actual operation of the power grid.
[0092] The second step is data preprocessing and feature construction.
[0093] The collected phasor measurement data are preprocessed and feature constructed, including outlier removal, missing data repair, noise filtering and normalization. A multi-node time-series feature dataset is constructed in combination with the power grid topology. This dataset is organized in the form of a three-dimensional feature tensor with dimensions including time dimension, node dimension and feature dimension, and is used for model training and online prediction.
[0094] Outlier detection employs a statistical threshold-based method. For any feature, the discrimination criterion is:
[0095] ,
[0096] in, and Let these represent the mean and standard deviation of this feature, respectively. This is the threshold coefficient. For detected outliers, replacement is performed using the neighborhood mean or interpolation method.
[0097] To address the missing data issue, interpolation methods were used to repair it, resulting in a complete data sequence.
[0098] ,
[0099] in, This indicates the data after repair.
[0100] To reduce the impact of measurement noise on data quality, the data is filtered to obtain a smoothed sequence:
[0101] ,
[0102] in, This represents the filtered data. This represents a filtering operator.
[0103] After data cleaning, each feature is normalized to eliminate the influence of units and improve the stability of model training. This is expressed as:
[0104] ,
[0105] in, and These represent the minimum and maximum values of the feature, respectively.
[0106] In the time dimension, a sliding time window is used to construct temporal features. For a window length of... Time series data, constructing input samples:
[0107] ,
[0108] in, Indicates the first A time window sample.
[0109] Furthermore, spatial features are constructed by combining power grid topology information. Let the power grid topology adjacency matrix be:
[0110] ,
[0111] Among them, if node With nodes If there is an electrical connection, then ,otherwise .
[0112] Based on adjacency relationships, spatial aggregation of node features is performed:
[0113] ,
[0114] in, Represents a node Spatial characteristics.
[0115] Finally, by fusing temporal and spatial features, a spatiotemporal representation of the power grid is obtained:
[0116] ,
[0117] in, This represents the spatiotemporal characteristics after fusion. This represents the feature fusion function.
[0118] In this embodiment, the corresponding adjacency matrix is constructed based on the IEEE-39 node system:
[0119] ,
[0120] Based on this topology, spatial modeling of node features is performed, and sample data is generated using a sliding time window, thereby forming a multi-node temporal feature dataset suitable for subsequent deep learning model input.
[0121] The third step is to construct a power grid transient stability prediction model.
[0122] The structure of the constructed power grid transient stability assessment model is as follows: Figure 2 As shown, named GCN-LSTM, it includes a spatial feature extraction module and a time series modeling module, specifically composed of a graph convolutional network, a long short-term memory network, and an attention mechanism module. It should be noted that the fault feature vectors constructed in the preceding steps are only used for sample generation and label construction during the offline training phase, and are not used as input features for the deep learning model in this step. The model only completes transient stability prediction based on a multi-node time series feature dataset.
[0123] The mapping relationship between model input and output is represented as follows:
[0124] ,
[0125] in, The parameter is Deep learning prediction models, This represents the spatiotemporal characteristics of the power grid at time k. Indicates the stable / instable classification result. This represents the predicted stability margin. This indicates the classification result of the instability type.
[0126] First, based on the power grid topology, a graph structure model is performed on the node features to obtain a spatial feature representation:
[0127] ,
[0128] Among them, it means No. Layer node feature matrix, Indicates trainable weight parameters. Represents a non-linear activation function, representing The normalized adjacency matrix is defined as follows:
[0129] ,
[0130] in, Degree matrix, Represents the identity matrix.
[0131] After obtaining the spatial features, the dynamic characteristics in the time dimension are further modeled. For time series inputs, a recurrent neural network is used to model them, resulting in a temporal feature representation:
[0132] ,
[0133] in, Indicates time The hidden state, Indicates input features, This represents a loop unit function.
[0134] To improve the model's ability to model long-term dependencies, a Long Short-Term Memory (LSTM) network is used to model time-series features. For time intervals... Input features LSTM models the long-term dependencies in the propagation of power grid disturbances through forget gates, input gates, output gates, and cell states. Its computational process is expressed as follows:
[0135] ,
[0136] ,
[0137] ,
[0138] ,
[0139] ,
[0140] ,
[0141] in, , and These represent the forget gate, input gate, and output gate, respectively. Indicates the state of candidate cells. This indicates the current state of the cell. This indicates the hidden state at the current moment. , , , as well as , , , This represents a trainable weight matrix. This represents the Sigmoid activation function. This represents the hyperbolic tangent activation function. This indicates element-wise multiplication.
[0142] After obtaining the spatial features extracted by the graph convolutional neural network and the temporal features extracted by the long short-term memory network, an attention mechanism module is introduced to perform weighted processing on the features.
[0143] The time-series hidden states output by the Long Short-Term Memory (LSTM) network are weighted and expressed as follows:
[0144] ,
[0145] in, This indicates that the Long Short-Term Memory network at time t The hidden state, This represents the corresponding attention weight.
[0146] The spatial features extracted by the graph convolutional neural network are weighted:
[0147] ,
[0148] in, The graph convolutional neural network extracts the first... Spatial features, This represents the corresponding spatial attention weights. This is the feature scoring function.
[0149] After extracting spatial and temporal features, they are fused to obtain a unified feature representation:
[0150] ,
[0151] Finally, the fused features are input into the stability classification output layer, the stability margin output layer, and the instability type classification output layer, respectively, to obtain the stability / instability classification results, stability margin prediction values, and instability type classification results:
[0152] ,
[0153] in, Indicates the output result. and These represent the output layer weights and biases, respectively.
[0154] The fourth step is model training and parameter optimization.
[0155] Based on the offline training dataset constructed in the first step and the power grid transient stability prediction model established in the third step, the model is trained and its parameters are optimized. Specifically:
[0156] For the The training samples are inputs consisting of spatiotemporal features derived from PMU measurement data and power grid topology. Its monitoring labels include stability / instability classification labels. Stability margin label and instability type labels .
[0157] It should be noted that the fault feature vector During the training phase, it is used to generate the sequence of operating states and corresponding labels under different disturbance scenarios, but it is not used as a direct input variable during model training. In other words, the model training process learns the mapping relationship between the spatiotemporal features of the phasor measurement unit and the transient steady state, rather than the mapping relationship between the fault feature vector and the steady state.
[0158] Therefore, the model training samples can be represented as:
[0159] ,
[0160] The model prediction results are expressed as follows:
[0161] ,
[0162] To simultaneously optimize the results of stability classification, stability margin prediction, and instability type classification, a joint loss function is constructed:
[0163] ,
[0164] in, Indicates stable / instable classification loss. Indicates the stability margin for predicting loss. Indicates the loss by type of instability. , and These are the corresponding weighting coefficients.
[0165] The stable / instability classification loss can be expressed as:
[0166] ,
[0167] The stability margin prediction loss can be expressed as:
[0168] ,
[0169] The loss for instability type classification can be expressed as:
[0170] ,
[0171] Based on the joint loss function described above, the model parameters are updated using the backpropagation algorithm and the adaptive optimization algorithm:
[0172] ,
[0173] in, Represents the set of model parameters. This represents the learning rate.
[0174] Through the training process described above, the model can learn the mapping relationship between the spatiotemporal characteristics of phasor measurement units and the transient stable state of the power grid. After training is completed, the online prediction stage only requires inputting the spatiotemporal characteristics constructed from real-time phasor measurement unit data to output stability / instability classification results, stability margin prediction values, and instability type classification results.
[0175] This embodiment trains the model based on the sample dataset generated by the IEEE-39-bus system. By updating the model parameters through multiple rounds of iteration, the loss function gradually converges, thereby obtaining the optimal model parameters for power grid transient stability prediction.
[0176] The fifth step is online prediction of power grid transient stability.
[0177] The power grid transient stability assessment model trained in the fourth step is deployed in the power grid online operation environment. Real-time collected power grid operation data is processed to achieve online prediction of the power grid transient stability state. Specifically:
[0178] The real-time phasor measurement data is converted into power grid spatiotemporal features according to the preprocessing and feature construction method in the second step, and then input into the trained power grid transient stability assessment model. The model outputs three types of results:
[0179] ,
[0180] ,
[0181] ,
[0182] in, Indicates the stable / instable classification result. This represents the predicted stability margin. This indicates the classification results of the instability type. , , These represent the weight matrices of the corresponding output layer. , , These represent the bias terms of the corresponding output layer.
[0183] Instability type classification results The outputs include three categories: power angle instability trend, voltage stability degradation, and severe instability risk, which are represented as follows:
[0184] ,
[0185] in, This represents the probability that the system has a tendency to become unstable at the power angle. This indicates the probability that the system's voltage stability will decrease. This indicates the probability that the system has a serious risk of instability.
[0186] Determine the instability type at the current moment based on the instability type classification results:
[0187] ,
[0188] in, This indicates the type of instability predicted at the current moment. At its maximum, the system is determined to have a tendency for power angle instability; when At its maximum, the system is deemed to have experienced a decrease in voltage stability; when... At its maximum, the system is deemed to have a serious risk of instability.
[0189] In this way, this step not only outputs the stability / instability classification results and stability margin predictions, but also the instability type classification results, providing a direct basis for the generation of differentiated control strategies in the sixth step.
[0190] Compared with GCN and LSTM models, the transient stability prediction performance of the proposed method is shown in Figures 3(a) to (c). The results show that the GCN-LSTM model exhibits a significant performance improvement.
[0191] The prediction results under noise conditions are shown in Figures 4(a) to 4(d). The results show that the method of the present invention still has good robustness under noise interference.
[0192] Step 6: Generation of power grid security control strategies.
[0193] Based on the power grid transient stability prediction results obtained in step 5, a power grid security control strategy is generated to suppress system instability and improve the safety and stability of power grid operation.
[0194] Specifically, according to the power grid at time Based on the stability judgment results, when the system is in a stable state, the current operating mode is maintained; when the system is at risk of instability, a control strategy optimization model is constructed to adjust the power grid operating state.
[0195] Let the power grid control variable vector be:
[0196] ,
[0197] in, This indicates the active power regulation of the generator. Indicates reactive power regulation. Indicates the amount of load removed. This indicates the amount of voltage regulation.
[0198] With the goal of improving system stability, an optimization objective function is constructed:
[0199] ,
[0200] in, This represents the stability prediction value in step five. and These are the weighting coefficients.
[0201] Simultaneously satisfying grid operation constraints, including power balance constraints:
[0202] ,
[0203] Voltage constraint:
[0204] ,
[0205] Generator output constraints:
[0206] ,
[0207] After solving the above optimization problem, the optimal control strategy is obtained. And it acts on the power grid system.
[0208] Furthermore, based on the instability type classification results output in step five, corresponding control measures are generated. The instability type classification results include power angle instability trend, voltage stability decline, and severe instability risk. Let the power grid control variable vector be:
[0209] ,
[0210] in, This indicates the generator's active power output regulation. This indicates the reactive power compensation adjustment amount. Indicates the amount of load reduction. This indicates the amount of voltage regulation.
[0211] Based on the instability type obtained in step 5 Select the appropriate control strategy. When When this occurs, it indicates that the system has a tendency to become unstable in terms of power angle. A generator active power output adjustment strategy is adopted, and its control variables are expressed as follows:
[0212] ,
[0213] when When this occurs, it indicates a decrease in system voltage stability. A reactive power compensation and voltage regulation strategy is employed, with the control variables expressed as follows:
[0214] ,
[0215] when When this occurs, it indicates a serious risk of system instability, and a comprehensive emergency control strategy is adopted, with the control variables represented as follows:
[0216] ,
[0217] in, This indicates a tendency for the angle of attack to become unstable. This indicates a decrease in voltage stability. This indicates a serious risk of instability. , and These represent the optimal control strategies for the three types of instability.
[0218] Furthermore, with the goal of improving system stability margin and reducing control cost, a control optimization objective function is constructed:
[0219] ,
[0220] ,
[0221] in, This represents the stability margin prediction value output in step five. Represents the control variable vector. and These are the weighting coefficients.
[0222] In the above manner, the generator output adjustment, reactive power compensation and voltage regulation, and emergency load reduction control in step six are all directly triggered by the instability type classification results output in step five, thereby establishing the correspondence between the prediction results and the differentiated control strategy.
[0223] A control simulation environment was built based on the IEEE-39-bus system. After detecting the instability trend of the system, control measures such as generator regulation and load shedding were automatically executed to verify the effectiveness and feasibility of the method in the actual power grid.
[0224] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.
Claims
1. A method for transient stability analysis and control of a power grid, characterized in that: Includes the following steps: S1: Deploy phasor measurement units at key nodes of the power grid to collect power grid operation status data in real time and perform unified time synchronization processing on the power grid operation status data of multiple nodes; S2: Preprocess and feature-construct the power grid operation status data, and construct a multi-node time-series feature dataset in combination with the power grid topology; S3: Construct a power grid transient stability prediction model. The model includes a graph convolutional network, a long short-term memory network, and an attention mechanism module. The graph convolutional network performs spatial aggregation on the node features in the multi-node temporal feature dataset based on the power grid topology adjacency matrix to extract spatial features. The long short-term memory network dynamically models the time series features in the multi-node temporal feature dataset to extract temporal features. The attention mechanism module adaptively weights the spatial features and temporal features respectively, and then fuses the weighted spatial features and the weighted temporal features to obtain fused spatiotemporal features. S4: Use historical operating data and simulation data to train and optimize the parameters of the power grid transient stability prediction model to obtain the trained power grid transient stability prediction model; S5: The real-time collected power grid operation status data is preprocessed and feature constructed in S2 and then input into the trained power grid transient stability prediction model to predict the power grid transient stability online. The model outputs stability / instability classification results, stability margin prediction values, and instability type classification results. The instability type classification results include power angle instability trend, voltage stability decline, and severe instability risk. S6: Generate corresponding safety control strategies based on the stability / instability classification results, stability margin prediction values, and instability type classification results.
2. The power grid transient stability analysis and control method as described in claim 1, characterized in that: The power grid operating status data includes voltage phasors, current phasors, frequency, active power, and reactive power. The preprocessing includes outlier removal, missing data repair, noise filtering, and normalization.
3. The power grid transient stability analysis and control method as described in claim 1, characterized in that: The feature construction adopts a sliding time window method to construct the power grid operation status data into a three-dimensional feature tensor. The dimensions of the three-dimensional feature tensor include time dimension, node dimension and feature dimension.
4. The power grid transient stability analysis and control method as described in claim 1, characterized in that: The graph convolutional network performs multi-layer graph convolution operations based on the power grid topology adjacency matrix to extract the electrical coupling relationships and spatial distribution characteristics between different nodes.
5. The power grid transient stability analysis and control method as described in claim 1, characterized in that: Long Short-Term Memory (LSTM) networks model long-term dependencies in the dynamic process of power grids through forget gates, input gates, output gates, and cell states.
6. The power grid transient stability analysis and control method as described in claim 1, characterized in that: The attention mechanism module includes a spatial attention mechanism and a temporal attention mechanism. The spatial attention mechanism is used to weight the spatial features, and the temporal attention mechanism is used to weight the temporal features.
7. The power grid transient stability analysis and control method as described in claim 1, characterized in that: In S4, a joint loss function is used to train the power grid transient stability prediction model. The joint loss function includes stability / instability classification loss, stability margin prediction loss, and instability type classification loss.
8. The power grid transient stability analysis and control method as described in claim 1, characterized in that: The safety control strategy is generated by constructing a control optimization objective function and satisfies power balance constraints, voltage constraints, and generator output constraints. When the stability / instability classification result is stable, maintain the current operating mode; When the stability / instability classification result is instability, corresponding control measures are generated based on the instability type classification result: when the instability type classification result is power angle instability trend, generator output adjustment is executed; When the instability type classification result is a decrease in voltage stability, the reactive power compensation device is switched on and off and the voltage is regulated. When the instability type classification result is a severe instability risk, load reduction control is implemented.
9. A power grid transient stability analysis and control system, characterized in that: include The data acquisition module is used to deploy phasor measurement units at key nodes of the power grid and collect power grid operation status data in real time, as well as to perform unified time synchronization processing on power grid operation status data from multiple nodes. The data processing module is used to preprocess and construct features from power grid operation status data, and to construct a multi-node time-series feature dataset in combination with the power grid topology. The stability prediction module includes a power grid transient stability prediction model, which is used to predict the transient stability state of the power grid based on the multi-node time-series feature dataset, and outputs stability / instability classification results, stability margin prediction values, and instability type classification results. The control decision module is used to generate corresponding safety control strategies based on the stability / instability classification results, stability margin prediction values, and instability type classification results.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the method according to any one of claims 1 to 8.