Power system short-circuit parameter prediction method, system and device based on space-time relevance and storage medium

Abstract

The invention relates to the technical field of power grid safety evaluation, in particular to a power system short-circuit parameter prediction method, system and device based on time-space relevance and a storage medium. A multi-dimensional state vector including a node voltage vector, a branch current vector, a short-circuit impedance matrix, spatial position information and a time feature vector is constructed, and a space-time correlation feature vector including a time-dimensional feature, a spatial-dimensional feature, a space-time coupling feature and a state feature is extracted. Wavelet transform, Fourier analysis and graph theory analysis are adopted to identify a parameter change rule and a propagation effect. A multi-level prediction model of a statistical prediction layer, a deep learning layer and a fusion layer is constructed, and comprehensive description of dynamic characteristics of short-circuit parameters is realized through spatio-temporal feature fusion. A staged training strategy is adopted, and comprises the steps of establishing a basic model through offline training, updating parameters in real time through online learning, and quantifying uncertainty through a probability prediction framework. And a prediction correction mechanism and a reliability evaluation system are established, and a complete prediction-evaluation-correction closed loop is formed.

Description

Technical Field

[0001] This invention relates to the field of power grid security assessment technology, and in particular to a method, system, device and storage medium for predicting short-circuit parameters of power systems based on spatiotemporal correlation. Background Technology

[0002] Predicting short-circuit parameters in power systems is crucial for fault prevention and system protection. With the continuous expansion of power system scale and the increasing complexity of operating environments, accurate prediction of short-circuit parameters has become a key technology for ensuring the safe and stable operation of power systems. Changes in short-circuit parameters are not only affected by time factors but are also closely related to the spatial topology of the system, exhibiting significant spatiotemporal coupling characteristics.

[0003] Traditional short-circuit parameter prediction methods primarily rely on single time series analysis, using statistical patterns from historical data to predict future parameter changes. While these methods are effective in handling periodic variations, they neglect the spatial correlation of parameter changes and struggle to accurately describe the distribution patterns and propagation characteristics of parameters in complex power networks. When topology changes or local faults occur in the system, traditional methods often fail to capture the impact of these changes on the overall system's short-circuit parameters in a timely manner.

[0004] While existing machine learning methods can handle complex nonlinear relationships and improve prediction accuracy to some extent, most still treat time and space features independently, lacking in-depth consideration of spatiotemporal coupling characteristics. This approach fails to fully explore the inherent patterns of parameter changes in power systems, especially when faced with sudden changes, where the predictive models are clearly inadequately adaptable. Furthermore, existing methods often employ static parameter settings for model optimization, making it difficult to adapt to the dynamic changes in the operating state of power systems.

[0005] In recent years, with the development of smart grid technology, a large amount of multidimensional monitoring data has been accumulated in power systems, providing a data foundation for in-depth analysis of the spatiotemporal variation patterns of short-circuit parameters. However, how to effectively integrate these multi-source heterogeneous data and construct a predictive model that can simultaneously consider temporal evolution and spatial correlation remains a challenging problem in current research. Summary of the Invention

[0006] In view of the problems existing in the prior art, the present invention is proposed.

[0007] Therefore, the problem this invention aims to solve is how to address key issues in short-circuit parameter prediction, such as insufficient spatiotemporal correlation analysis, low prediction accuracy, and poor model adaptability. This invention proposes a prediction method based on spatiotemporal correlation. This method comprehensively describes the system's operating state by constructing a multi-dimensional state vector, employs spatiotemporal feature fusion technology to deeply explore the inherent laws governing parameter changes, establishes a multi-level prediction model to accurately model complex nonlinear relationships, and improves the model's adaptability and prediction accuracy through a dynamic optimization mechanism, thereby providing reliable technical support for the safe and stable operation of power systems.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] In a first aspect, embodiments of the present invention provide a method for predicting short-circuit parameters of a power system based on spatiotemporal correlation, which includes: constructing a power system state vector and extracting a spatiotemporal correlation feature vector based on the state vector;

[0010] The spatiotemporal correlation feature vector is used as input data to construct a multi-level prediction model. The multi-level prediction model achieves a comprehensive characterization of the dynamic characteristics of short-circuit parameters through spatiotemporal feature fusion.

[0011] The multi-level prediction model is trained using a phased training strategy, and a prediction mapping relationship is established based on the historical data of the spatiotemporal correlation feature vector.

[0012] The short-circuit parameter prediction results are output based on the trained prediction model.

[0013] As a preferred embodiment of the power system short-circuit parameter prediction method based on spatiotemporal correlation described in this invention, the spatiotemporal correlation feature vector includes time-dimensional features and spatial-dimensional features. The periodic patterns and trend characteristics of parameter changes are extracted and identified through the time-dimensional features, and the spatial distribution patterns and topological correlation characteristics of the parameters are extracted and identified through the spatial-dimensional features.

[0014] As a preferred embodiment of the power system short-circuit parameter prediction method based on spatiotemporal correlation described in this invention, the extraction of the spatiotemporal correlation feature vector includes: identifying the periodicity of parameter changes through wavelet transform and Fourier analysis; identifying key nodes and weak links through graph theory analysis; and establishing a spatiotemporal mapping relationship to analyze the propagation effect of parameter changes.

[0015] As a preferred embodiment of the power system short-circuit parameter prediction method based on spatiotemporal correlation described in this invention, the multi-level prediction model includes a statistical prediction layer, a deep learning layer, and a fusion layer. The statistical prediction layer captures the basic laws of parameter changes, the deep learning layer extracts complex feature relationships, and the fusion layer realizes the weight combination of different model results.

[0016] The beneficial effects of this preferred technical solution are as follows: By constructing a three-layer architecture design consisting of a statistical prediction layer, a deep learning layer, and a fusion layer, it achieves layered processing and complementary advantages for the short-circuit parameter prediction problem. The initial statistical prediction layer captures the basic laws and periodic characteristics of parameter changes through time series analysis and regression models, providing a stable statistical foundation for prediction. The middle deep learning layer uses multi-layer nonlinear mapping to deeply mine complex nonlinear relationships and latent patterns in spatiotemporal correlation features, effectively handling complex coupling problems that are difficult to solve with traditional statistical methods. The final fusion layer dynamically balances the contributions of the two layers through an adaptive weighting mechanism, ensuring that each layer can achieve optimal performance in different scenarios. Overall, a complete prediction chain is formed from basic law identification to complex relationship extraction to intelligent fusion, achieving the technical effect of organically combining the stability of statistical methods with the nonlinear modeling capabilities of deep learning methods.

[0017] As a preferred embodiment of the power system short-circuit parameter prediction method based on spatiotemporal correlation described in this invention, the construction of the multi-level prediction model includes: a statistical prediction layer receiving spatiotemporal correlation feature vectors, and using time series analysis and regression models to process time-dimensional features and spatial-dimensional features;

[0018] The deep learning layer receives spatiotemporally correlated feature vectors and uses multi-layer nonlinear mapping to process spatiotemporally coupled features.

[0019] The fusion layer receives the outputs from the statistical prediction layer and the deep learning layer, and uses an adaptive weighting mechanism to determine the combined weights of the results from each layer.

[0020] As a preferred embodiment of the power system short-circuit parameter prediction method based on spatiotemporal correlation described in this invention, the phased training strategy includes: using historical spatiotemporal correlation feature vector data to establish a basic model during the offline training phase;

[0021] During the online learning phase, a sliding window mechanism is used to select the latest spatiotemporal correlation feature vector data to update the model parameters in real time;

[0022] A probabilistic prediction framework is introduced, which generates prediction intervals through Monte Carlo simulation based on the uncertainty of spatiotemporal correlation feature vectors.

[0023] The beneficial effects of this preferred technical solution are as follows: By designing a three-stage training strategy of offline training, online learning, and probabilistic prediction, intelligent management of the entire lifecycle of the prediction model is achieved. The initial offline training stage utilizes a large amount of historical spatiotemporal correlation feature vector data to establish a stable and reliable basic model, ensuring the model has a solid predictive foundation and good generalization ability. The intermediate online learning stage continuously receives the latest spatiotemporal correlation feature vector data through a sliding window mechanism, updating model parameters in real time to adapt to the dynamic changes in the power system's operating state, ensuring the timeliness and adaptability of the prediction model. Finally, the introduced probabilistic prediction framework quantifies the uncertainty of the prediction results through Monte Carlo simulation, generating confidence intervals to provide risk assessment information for decision-making. Overall, a complete training system is constructed from establishing a stable foundation to dynamic adaptive updates and uncertainty quantification, achieving continuous optimization and reliability assurance of the prediction model, and realizing a comprehensive technical effect of high prediction accuracy, strong real-time performance, and controllable risk.

[0024] As a preferred embodiment of the power system short-circuit parameter prediction method based on spatiotemporal correlation described in this invention, the output of the prediction result includes generating predicted values ​​and prediction intervals of short-circuit parameters based on the input spatiotemporal correlation feature vector.

[0025] Establish a prediction correction mechanism to compare the actual observed short-circuit parameters with the prediction results and optimize the prediction results through real-time feedback;

[0026] The reliability assessment of prediction results is provided based on the quality of spatiotemporal correlation feature vectors and model performance.

[0027] Secondly, embodiments of the present invention provide a power system short-circuit parameter prediction system based on spatiotemporal correlation, which includes a construction module for constructing a power system state vector and extracting a spatiotemporal correlation feature vector based on the state vector.

[0028] The input module uses the spatiotemporal correlation feature vector as input data to construct a multi-level prediction model. The multi-level prediction model achieves a comprehensive characterization of the dynamic characteristics of short-circuit parameters through spatiotemporal feature fusion.

[0029] The training module employs a phased training strategy to train the multi-level prediction model and establishes a prediction mapping relationship based on historical data of the spatiotemporal correlation feature vectors.

[0030] The output module outputs the short-circuit parameter prediction results based on the trained prediction model.

[0031] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the power system short-circuit parameter prediction method based on spatiotemporal correlation as described in the first aspect of the present invention.

[0032] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the power system short-circuit parameter prediction method based on spatiotemporal correlation as described in the first aspect of the present invention.

[0033] The beneficial effects of this invention are as follows: By constructing a multi-dimensional state vector containing node voltage vectors, branch current vectors, short-circuit impedance matrices, spatial location information, and temporal feature vectors, this invention achieves a comprehensive mathematical description of the operating state of the power system, overcoming the limitations of traditional methods that only focus on single-dimensional information, and providing a complete data foundation for subsequent spatiotemporal correlation analysis. By extracting spatiotemporal correlation feature vectors containing time-dimensional features, spatial-dimensional features, spatiotemporal coupling features, and state features from the state vectors, and using methods such as wavelet transform, Fourier analysis, and graph theory analysis, a deep-level analysis of the variation law of short-circuit parameters is achieved. This allows for the simultaneous capture of the temporal evolution characteristics and spatial propagation characteristics of the parameters, breaking through the technical bottleneck of spatiotemporal separation processing in existing technologies. By constructing a multi-level prediction model with statistical prediction layers, deep learning layers, and fusion layers, and adopting a staged training strategy and probabilistic prediction framework, accurate modeling and uncertainty quantification of complex nonlinear spatiotemporal coupling relationships are achieved, significantly improving prediction accuracy and the model's adaptability. This technical solution forms a complete closed-loop system from data input to result output. Through prediction correction mechanisms and reliability assessments, it achieves high prediction accuracy, strong adaptability, and good real-time performance. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 The flowchart shows a method for predicting short-circuit parameters in power systems based on spatiotemporal correlation.

[0036] Figure 2 A computer equipment diagram for a power system short-circuit parameter prediction method based on spatiotemporal correlation;

[0037] Figure 3Another flowchart for the overall method of predicting short-circuit parameters in power systems based on spatiotemporal correlation. Detailed Implementation

[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0039] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0040] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0041] Example 1

[0042] Reference Figure 1 - Figure 2 This is the first embodiment of the present invention, which provides a method for predicting short-circuit parameters of a power system based on spatiotemporal correlation, including:

[0043] S100: Construct the power system state vector and extract spatiotemporal correlation feature vectors based on the state vector;

[0044] S200: A multi-level prediction model is constructed by using spatiotemporal correlation feature vectors as input data. The multi-level prediction model achieves a comprehensive characterization of the dynamic characteristics of short-circuit parameters through spatiotemporal feature fusion.

[0045] S300: A phased training strategy is used to train a multi-level prediction model, and a prediction mapping relationship is established based on historical data of spatiotemporal correlation feature vectors.

[0046] S400: Outputs short-circuit parameter prediction results based on the trained prediction model.

[0047] S100 establishes a complete mathematical description of the power system's operating state by constructing a multi-dimensional state vector that includes node voltage vectors, branch current vectors, short-circuit impedance matrices, spatial location information, and temporal feature vectors, laying a data foundation for subsequent spatiotemporal correlation analysis. The spatiotemporal correlation feature vectors extracted from the state vectors can simultaneously capture the temporal evolution and spatial propagation characteristics of parameter changes, overcoming the limitations of traditional methods that separate spatiotemporal processing. S200's multi-level prediction model adopts a hierarchical architecture of a statistical prediction layer, a deep learning layer, and a fusion layer. The statistical prediction layer uses time-series analysis methods to handle the periodic and trend changes of parameters; the deep learning layer extracts complex spatiotemporal coupling relationships through nonlinear mapping; and the fusion layer achieves the optimal combination of the results from the two layers through an adaptive weighting mechanism, thereby significantly improving prediction accuracy. S300's phased training strategy ensures the model's practicality and adaptability. Offline training uses historical data to establish a stable basic model, while online learning uses a sliding window mechanism to update model parameters in real time. The probabilistic prediction framework provides a quantitative assessment of prediction uncertainty. The S400 not only outputs accurate short-circuit parameter predictions, but also provides decision support for power grid operation through prediction correction mechanisms and reliability assessments, forming a complete prediction-assessment-correction closed-loop system.

[0048] Example 2

[0049] Reference Figure 2 - Figure 3 This is the second embodiment of the present invention.

[0050] In this embodiment, the detailed process of constructing the power system state vector in step S100 includes the following steps A1-A4:

[0051] A1: The state vector includes node voltage vector, branch current vector, short-circuit impedance matrix, spatial location information, and time feature vector;

[0052] To accurately predict short-circuit parameters, a complete system state model is first required. This model needs to comprehensively reflect the electrical characteristics and spatiotemporal features of the system, laying the foundation for subsequent predictive analysis. The system state vector is defined as:

[0053] S t =V i ,I i Z i ,L i ,T i (1)

[0054] In the formula, V i This is the node voltage vector, containing the voltage magnitude and phase angle information of each node in the system; I i Z is the branch current vector, describing the current distribution in each branch of the network;i L is the short-circuit impedance matrix, characterizing the impedance characteristics of each element in the network; i Spatial location information, including the device's geographic coordinates and topological relationships; T i This is a time feature vector that reflects the time-varying characteristics of the parameters. This state vector describes the system's operating state from multiple dimensions, providing a complete data foundation for short-circuit parameter prediction.

[0055] A2: Spatiotemporal correlation feature vectors include time-dimensional features, spatial-dimensional features, spatiotemporal coupling features, and state features;

[0056] Based on the system state vector, a spatiotemporal correlation feature vector is constructed to deeply analyze the variation patterns of short-circuit parameters. The time-dimensional features are obtained through historical data analysis, including daily variation curves, weekly trends, and seasonal patterns of short-circuit parameters. For example, line impedance increases by approximately 5%–8% during the high temperatures of summer, and decreases accordingly during the low temperatures of winter. The spatial-dimensional features reflect the electrical distance and degree of influence between different nodes, calculated through the network sensitivity matrix. The spatiotemporal coupling features describe the interaction of parameter changes in the spatiotemporal dimensions, such as the diffusion process of the impact range over time when a fault occurs. The state features include information such as the current system operating mode, load level, and equipment commissioning / discharging status.

[0057] A3: By extracting time-dimensional features, we can identify the periodic patterns and trends of parameter changes. By extracting spatial-dimensional features, we can identify the spatial distribution patterns and topological correlation characteristics of parameters.

[0058] Based on the system state vector, a spatiotemporal feature extraction method is constructed to deeply analyze the spatiotemporal variation patterns of short-circuit parameters. This method achieves a comprehensive characterization of the dynamic properties of short-circuit parameters through multi-dimensional feature fusion. The spatiotemporal correlation feature vector is designed as follows:

[0059] F t =X i ,Y i W i ,S i (2)

[0060] Among them, X i The time dimension features include historical trends and periodic characteristics of the parameters; Y i The spatial dimension features reflect the spatial distribution and correlation of the parameters; W i This refers to the spatiotemporal coupling characteristics, describing the interactive effects of parameters in the time and space dimensions; S i These are state characteristics, encompassing system operating states and environmental factors. Feature correlation analysis employs a weighted summation method:

[0061]

[0062] In the formula, α i β represents the weight coefficients for different features, reflecting the importance of each feature; β is the overall correlation coefficient, used to adjust the strength of feature combinations; n is the feature dimension.

[0063] A4: The extraction of spatiotemporal correlation feature vectors includes:

[0064] The periodic patterns of parameter changes are identified through wavelet transform and Fourier analysis; key nodes and weak links are identified through graph theory analysis; and the propagation effect of parameter changes is analyzed by establishing a spatiotemporal mapping relationship.

[0065] For time-dimensional feature extraction, the system employs an improved time-series analysis method, including trend decomposition, periodicity identification, and abrupt change detection. Through wavelet transform and Fourier analysis, the system can accurately identify the periodic patterns of parameter changes, providing a basis for subsequent predictions. For non-periodic changes, the system uses a trend-following algorithm to achieve real-time feature updates.

[0066] Spatial feature extraction focuses on the spatial distribution patterns and topological correlations of parameters. The system establishes a complete spatial feature model, including distance features, connectivity features, and distribution features. Through graph theory analysis, the system can effectively identify key nodes and weak links, providing a basis for network optimization.

[0067] The extraction of spatiotemporal coupling features employs a multidimensional analysis method. By establishing a spatiotemporal mapping relationship, the system achieves a unified description of parameter changes across time and space dimensions. This method pays particular attention to the propagation effect of parameter changes, enabling timely detection and early warning of potential chain reactions. Combined with state features, the system achieves comprehensive perception of the operating environment, ensuring the adaptability of feature extraction. This feature extraction method achieves effective fusion of spatiotemporal information.

[0068] In an optional implementation, the construction of the power system state vector in step S100 can also be based on real-time monitoring data quality assessment and data fusion technology of the power system, combining multi-source heterogeneous data such as SCADA data, PMU data, and meteorological data to establish data quality evaluation indicators and data reliability weight allocation mechanisms. This method, by introducing data integrity, consistency, and timeliness assessments, constructs a state vector that considers data quality, ensuring the reliability of the input data. It is suitable for complex power systems with diverse data sources and varying data quality.

[0069] In another optional implementation, the construction of the power system state vector in step S100 can also be achieved through an automatic feature extraction method using deep feature learning. This method utilizes dimensionality reduction algorithms such as autoencoders and principal component analysis to compress and reconstruct the features of the original power system operating data, automatically identifying and extracting the most representative state features. This method can reduce data redundancy, improve feature representation capabilities, and discover hidden state patterns through unsupervised learning, making it particularly suitable for constructing state vectors from high-dimensional and complex data.

[0070] In this embodiment, the prediction model construction in step S200 includes the following steps B1-B2:

[0071] B1: The multi-level prediction model includes a statistical prediction layer, a deep learning layer, and a fusion layer. The statistical prediction layer captures the basic laws of parameter changes, the deep learning layer extracts complex feature relationships, and the fusion layer realizes the weight combination of different model results.

[0072] The statistical prediction layer captures the fundamental patterns of parameter changes, the deep learning layer extracts complex feature relationships, and the fusion layer combines the weights of different model results. To improve prediction accuracy, the system employs a multi-model fusion strategy. The base layer uses statistical prediction methods, including time series analysis and regression models, to capture the fundamental patterns of parameter changes. The deep learning layer uses an improved deep learning model to extract complex feature relationships through multi-layer nonlinear mapping. The fusion layer achieves the optimal combination of different model results through an adaptive weighting mechanism.

[0073] To address the need for short-circuit parameter prediction, a prediction model based on spatiotemporal features is designed. This model employs a multi-layered architecture, integrating deep learning and statistical analysis methods to achieve accurate prediction of short-circuit parameters. The basic prediction function is expressed as:

[0074]

[0075] Where f(·) is a nonlinear prediction mapping function that establishes the correspondence between input features and prediction results; θ is the model parameter set, obtained through training and optimization; w i The weighting coefficients for different correction terms; δ i The terms represent various correction terms, including environmental impacts, load changes, etc.; m represents the number of correction terms. Considering the spatiotemporal coupling effect, a coupled prediction model is introduced:

[0076] Y t+1 =φ(Y t ,S t )+∈ (5)

[0077] In the formula, φ(·) is the spatiotemporal mapping function, which realizes state transition and prediction; Y t The current system state; S t∈ represents the corresponding spatial features; ∈ represents the error term in the prediction process, which includes model simplification errors and random perturbations.

[0078] In one optional implementation, the statistical prediction layer can also integrate multiple classical time series analysis methods, including the ARIMA model for stationary time series, the SARIMA model for seasonal time series, and the GARCH model for fluctuating and clustered time series. Through an automatic model selection mechanism, the most suitable statistical model is automatically selected based on the statistical characteristics of the time series data, and the prediction results of multiple statistical models are combined using ensemble learning methods, thereby improving the accuracy of the statistical prediction layer.

[0079] In another alternative implementation, the deep learning layer can also employ an attention-enhanced recurrent neural network. This network uses self-attention to identify key information in spatiotemporally related features, increasing the model's focus on important features. The attention weights are calculated based on the similarity and importance scores between features, which can adaptively adjust the contribution of different features to the prediction results. This is particularly effective in mitigating the gradient vanishing problem when processing long-sequence spatiotemporal data.

[0080] B2: The construction of a multi-level prediction model includes:

[0081] The statistical prediction layer receives spatiotemporal correlation feature vectors and uses time series analysis and regression models to process time-dimensional and spatial-dimensional features.

[0082] The deep learning layer receives spatiotemporally correlated feature vectors and uses multi-layer nonlinear mapping to process spatiotemporally coupled features.

[0083] The fusion layer receives the outputs from the statistical prediction layer and the deep learning layer, and uses an adaptive weighting mechanism to determine the combined weights of the results from each layer.

[0084] To improve prediction accuracy, the system employs a multi-model fusion strategy. The base layer uses statistical prediction methods, including time series analysis and regression models, to capture the fundamental patterns of parameter changes. The deep layer uses an improved deep learning model to extract complex feature relationships through multi-layer nonlinear mapping. The fusion layer achieves the optimal combination of results from different models through an adaptive weighting mechanism.

[0085] The training process of the prediction model adopts a phased strategy. First, offline training is performed using historical data to build a basic model. Then, online learning is used to update the model parameters in real time to adapt to changes in system state. A sliding window mechanism is used in the selection of training data to ensure that the model can reflect the latest system characteristics in a timely manner.

[0086] To address uncertainties in the prediction process, the system introduces a probabilistic prediction framework. Multiple sets of prediction results are generated through Monte Carlo simulations, forming prediction intervals to provide a reliability assessment for decision-making. Simultaneously, the system establishes a comprehensive prediction correction mechanism, continuously optimizing prediction results through real-time feedback.

[0087] This prediction model significantly improves the accuracy and reliability of predictions through deep fusion of spatiotemporal features. The model design fully considers the complexity of short-circuit parameter variations, including load fluctuations, equipment switching, and the impact of faults. It has significant engineering application value.

[0088] In an optional implementation, the prediction model construction in step S200 can also be based on the physical constraints and electrical characteristics of the power system, combined with short-circuit current calculation theory and network topology analysis, to establish a short-circuit parameter variation model based on physical mechanisms. This method introduces electrical parameters such as node admittance matrices, network equivalent impedance, and fault boundary conditions to construct a prediction model that considers the physical constraints of the system, ensuring that the prediction results conform to the basic physical laws of the power system. This approach is suitable for application scenarios where the physical rationality of the prediction results is highly critical.

[0089] In another optional implementation, the prediction model construction in step S200 can also be achieved through a multi-model fusion prediction method using ensemble learning. This method utilizes different types of prediction algorithms, such as support vector machines, random forests, and gradient boosting trees, to construct multiple base learners. The prediction results of each base learner are then combined through a voting mechanism or a weighted average. This method fully leverages the advantages of different algorithms, improving the stability and robustness of predictions through model diversity. It is particularly suitable for power system prediction scenarios with inconsistent data quality or complex and variable operating conditions.

[0090] In this embodiment, the phased training strategy in step S300 includes the following steps C1-C3:

[0091] C1: The basic model is built using historical spatiotemporal correlated feature vector data during the offline training phase;

[0092] C2: During the online learning phase, a sliding window mechanism is used to select the latest spatiotemporal correlation feature vector data to update the model parameters in real time;

[0093] In another alternative implementation, the sliding window mechanism can employ a multi-scale window design, simultaneously maintaining data windows at different time scales. Short-term windows capture the latest changes, medium-term windows maintain stability, and long-term windows provide historical benchmarks. By fusing multi-scale information, the model can remain sensitive to the latest changes while maintaining a memory of historical patterns, thus improving the stability and accuracy of predictions.

[0094] C3: Introducing a probabilistic prediction framework based on the uncertainty of spatiotemporal correlation feature vectors to generate prediction intervals through Monte Carlo simulation.

[0095] Probabilistic prediction frameworks are important tools for handling prediction uncertainty, providing risk assessment information for decision-making by quantifying the confidence intervals of prediction results. Monte Carlo simulations consider the uncertainty of input parameters through random sampling, generating multiple sets of possible predictions and forming a prediction distribution. Sources of uncertainty include measurement noise, model errors, and the randomness of future states, requiring the development of corresponding probabilistic models to describe these uncertainties.

[0096] In an optional implementation, the phased training strategy in step S300 can also be based on the seasonal and periodic characteristics of the power system's operating state, employing a layered time window adaptive training method. This method designs training windows at different time scales according to the daily, weekly, and seasonal cycle characteristics of the power system load. Through a multi-time-scale model parameter update strategy, it ensures that the prediction model can simultaneously capture short-term fluctuations and long-term trends, improving the model's adaptability to changes at different time scales.

[0097] In another optional implementation, the phased training strategy in step S300 can also utilize a distributed training method based on federated learning. This method leverages historical monitoring data from multiple power grid regions, such as short-circuit parameters, load characteristics, and equipment status, combined with federated learning algorithms. This allows for collaborative model training and knowledge sharing while protecting the data privacy of each region. This approach fully utilizes data resources from multiple regions, improving the model's generalization ability and prediction accuracy. It is particularly suitable for short-circuit parameter prediction applications in large-scale interconnected power grids.

[0098] In this embodiment, the output of the prediction result in step S400 includes the following steps D1-D3:

[0099] D1: Generate predicted values ​​and prediction intervals for short-circuit parameters based on the input spatiotemporal correlation feature vector;

[0100] To ensure continuous optimization and performance improvement of the prediction model, a dynamic optimization mechanism was designed. This mechanism continuously adjusts the model parameters by optimizing the objective function, thereby improving prediction accuracy. The optimization objective function is designed as follows:

[0101]

[0102] In the formula, y i These are the actual observed values ​​of the short-circuit parameters, reflecting the true state of the system; y i λ represents the predicted output value of the model; k is the number of samples; λ is a coefficient controlling the regularization strength; R(θ) is the regularization term used to prevent overfitting. The parameter update rule based on gradient descent is:

[0103]

[0104] In the formula, θ t and θ t+1 These represent the current and updated model parameters, respectively; α is the learning rate that controls the step size of the parameter update. This is the gradient of the objective function with respect to the parameters.

[0105] The dynamic optimization process employs a layered design, including a parameter optimization layer, a structure optimization layer, and a policy optimization layer. The parameter optimization layer is responsible for real-time adjustment of model parameters, minimizing prediction error through algorithms such as gradient descent. The optimization process uses a mini-batch approach, with each batch containing 100–500 samples to ensure real-time performance and stability. The system incorporates a dynamic learning rate adjustment mechanism, using a larger learning rate in the initial optimization phase to accelerate convergence, and gradually decreasing the learning rate as the optimization progresses to improve accuracy.

[0106] The structural optimization layer focuses on the dynamic adjustment of the model structure, including feature selection, the number of network layers, and the number of nodes. The system dynamically adjusts the combination of input features through feature importance analysis. This adaptive structural optimization ensures that the model complexity matches the problem difficulty.

[0107] The strategy optimization layer enables intelligent selection of optimization strategies, including algorithm selection, hyperparameter tuning, and regularization switching. The system establishes a complete strategy evaluation system, selecting the optimal strategy combination through multi-objective evaluation methods. When a decline in prediction performance is detected, the system automatically triggers strategy adjustments to ensure optimization effectiveness. This optimization mechanism continuously improves model performance through iterative updates.

[0108] D2: Establish a prediction correction mechanism to compare the actual observed short-circuit parameters with the prediction results and optimize the prediction results through real-time feedback;

[0109] This evaluation system combines quantitative indicators with qualitative analysis to achieve a comprehensive assessment of the prediction results. The prediction accuracy is calculated as follows:

[0110]

[0111] In the formula, w i The weighting coefficients for different evaluation indicators reflect the importance of each indicator; P i Here, represents the specific prediction accuracy metric; n represents the number of evaluation metrics. The overall system reliability is expressed as a product:

[0112]

[0113] In the formula, R t,iThe prediction reliability in the time dimension reflects the model's stability over time series; R s,i The predictive reliability for the spatial dimension characterizes the model's ability to grasp spatial distributions; R p,i The overall prediction reliability describes the model's comprehensive predictive capability; m is the number of evaluation periods.

[0114] D3: Provides a reliability assessment of prediction results based on the quality of spatiotemporal correlation feature vectors and model performance.

[0115] The system's operational status is evaluated in real time using multi-dimensional indicators to ensure the efficient and stable operation of the predictive system. The performance evaluation function is designed as follows:

[0116]

[0117] In the formula, λ i The delay factor, μ, describes the system response latency and reflects the time characteristics of the computation process. i The response coefficient characterizes the system's response speed; t is the actual system running time; n is the number of monitoring indicators. The overall system efficiency is evaluated using:

[0118]

[0119] In the formula, C t The computational load required to complete the prediction task includes various arithmetic operations; T t The total computation time is denoted by η; η is a comprehensive factor reflecting computational efficiency. Through real-time monitoring of these indicators, the system can dynamically adjust its computational strategy to ensure the high efficiency of the prediction process. This performance monitoring system provides a quantitative basis for system optimization and effectively guarantees the stable operation of the prediction system.

[0120] In an optional implementation, the prediction result output in step S400 can also be based on the early warning threshold setting of power system security and stability analysis, combined with the safety limits of short-circuit parameters and system protection configuration requirements, to establish a hierarchical early warning mechanism and risk assessment system. This method, by setting different levels of early warning thresholds, achieves intelligent classification of prediction results and risk level division, providing intuitive security assessment information for power grid dispatching and operation and maintenance decisions, and is suitable for power system operation and management with high system security requirements.

[0121] In another optional implementation, the prediction results output in step S400 can also be presented to the user in an intuitive graphical way using visualization analysis and interactive display technologies, such as spatiotemporal data visualization methods like heatmaps, trend charts, and network topology diagrams. This method combines human-computer interaction technology, supports user-defined queries and multi-dimensional data drill-down, and dynamically visualizes the spatiotemporal distribution characteristics and trends of the prediction results. It is particularly suitable for application scenarios that require an intuitive understanding of the prediction results and support for decision analysis.

[0122] In summary, this invention constructs a multi-dimensional state vector comprising node voltage vectors, branch current vectors, short-circuit impedance matrices, spatial location information, and temporal feature vectors, achieving a comprehensive mathematical description of the power system's operating state. This overcomes the limitations of traditional methods that focus only on a single dimension of information, providing a complete data foundation for subsequent spatiotemporal correlation analysis. By extracting spatiotemporal correlation feature vectors containing temporal, spatial, spatiotemporal coupling, and state features from the state vectors, and employing wavelet transform, Fourier analysis, and graph theory analysis, a deep understanding of the short-circuit parameter variation patterns is achieved. This simultaneously captures the temporal evolution and spatial propagation characteristics of the parameters, overcoming the technical bottleneck of spatiotemporal separation processing in existing technologies. By constructing a multi-layered prediction model comprising a statistical prediction layer, a deep learning layer, and a fusion layer, and adopting a phased training strategy and a probabilistic prediction framework, accurate modeling and uncertainty quantification of complex nonlinear spatiotemporal coupling relationships are achieved, significantly improving prediction accuracy and model adaptability. This technical solution forms a complete closed-loop system from data input to result output. Through prediction correction mechanisms and reliability assessments, it achieves high prediction accuracy, strong adaptability, and good real-time performance.

[0123] Example 3

[0124] This invention uses actual operating data from a provincial power grid in 2023 for verification, including short-circuit parameter data from multiple substations. The experiment employs a comparative verification method, selecting three typical methods—time series analysis, spatial correlation, and deep learning—as control groups. To ensure fairness, all methods use the same training and test datasets, divided in a 7:3 ratio. The verification process consists of three stages: first, benchmark performance testing to evaluate the prediction performance of each method under standard conditions; second, scenario adaptability testing to verify the method's performance under different operating scenarios; and finally, robustness testing to evaluate the method's resistance to various disturbances. The test results are as follows:

[0125] Table 1: Performance Comparison of Different Prediction Methods

[0126] method Prediction error (%) Calculation time (s) Stability Indicators Adaptability Indicators Time series analysis 3.25 85.6 0.856 0.823 Spatial association 2.86 92.3 0.883 0.856 Deep learning 2.45 125.8 0.912 0.892 method 1.85 68.5 0.957 0.945

[0127] Table 2: Prediction Performance (Relative Error %) in Different Scenarios

[0128] Scene type Time series analysis Spatial association Deep learning method Normal operation 2.85 2.45 2.12 1.65 Load mutation 3.56 3.12 2.85 1.92 Fault recovery 4.25 3.85 3.45 2.15 Weather changes 3.85 3.56 3.12 2.08

[0129] Table 3: System Robustness Test Results

[0130] Test Project Prediction accuracy (%) Response time (ms) Resource usage (%) recovery ability Parameter mutation 96.5 125 35 powerful noise interference 97.2 98 28 Strong Data missing 95.8 145 42 medium Model drift 96.8 112 38 Strong

[0131] Experimental results demonstrate that the proposed method exhibits significant advantages in prediction performance. The prediction error is reduced to 1.85%, and the computation time is decreased to 68.5 seconds, while maintaining high stability and adaptability. The method demonstrates excellent prediction ability and robustness in various complex scenarios. The entire validation process lasted approximately 100 hours, fully demonstrating the practical engineering value of the proposed method. Comparison with other methods shows significant improvements in prediction accuracy, computational efficiency, and system robustness.

[0132] Example 4

[0133] The above is an illustrative scheme of a power system short-circuit parameter prediction method based on spatiotemporal correlation. It should be noted that the technical solution of this power system short-circuit parameter prediction system based on spatiotemporal correlation belongs to the same concept as the aforementioned power system short-circuit parameter prediction method based on spatiotemporal correlation. Details not described in detail in this embodiment can be found in the description of the aforementioned power system short-circuit parameter prediction method based on spatiotemporal correlation.

[0134] This embodiment also provides a power system short-circuit parameter prediction system based on spatiotemporal correlation, including:

[0135] The module constructs a power system state vector and extracts spatiotemporal correlation feature vectors based on the state vector.

[0136] The input module uses spatiotemporal correlation feature vectors as input data to construct a multi-level prediction model. This model comprehensively characterizes the dynamic properties of short-circuit parameters through spatiotemporal feature fusion. The training module employs a phased training strategy to train the multi-level prediction model, establishing a prediction mapping relationship based on historical data of the spatiotemporal correlation feature vectors. The output module outputs the short-circuit parameter prediction results based on the trained prediction model. This embodiment also provides an electronic device suitable for power system short-circuit parameter prediction based on spatiotemporal correlation, comprising: a memory and a processor; the memory stores computer-executable instructions, and the processor executes these instructions to implement the power system short-circuit parameter prediction method based on spatiotemporal correlation proposed in the above embodiments. This embodiment also provides a storage medium storing a computer program, which, when executed by the processor, implements the power system short-circuit parameter prediction method based on spatiotemporal correlation proposed in the above embodiments. The storage medium proposed in this embodiment and the power system short-circuit parameter prediction method based on spatiotemporal correlation proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0137] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting short-circuit parameters in a power system based on spatiotemporal correlation, characterized in that: This includes constructing a power system state vector and extracting a spatiotemporal correlation feature vector based on the state vector; The spatiotemporal correlation feature vector is used as input data to construct a multi-level prediction model. The multi-level prediction model achieves a comprehensive characterization of the dynamic characteristics of short-circuit parameters through spatiotemporal feature fusion. The multi-level prediction model is trained using a phased training strategy, and a prediction mapping relationship is established based on the historical data of the spatiotemporal correlation feature vector. The short-circuit parameter prediction results are output based on the trained prediction model.

2. The power system short-circuit parameter prediction method based on spatiotemporal correlation as described in claim 1, characterized in that: The spatiotemporal correlation feature vector includes time-dimensional features and spatial-dimensional features. The time-dimensional features are used to extract the periodic patterns and trend characteristics of the changes in the identification parameters, while the spatial-dimensional features are used to extract the spatial distribution patterns and topological correlation characteristics of the identification parameters.

3. The power system short-circuit parameter prediction method based on spatiotemporal correlation as described in claim 2, characterized in that: The extraction of the spatiotemporal correlation feature vector includes identifying the periodic patterns of parameter changes through wavelet transform and Fourier analysis; identifying key nodes and weak links through graph theory analysis; and establishing spatiotemporal mapping relationships to analyze the propagation effect of parameter changes.

4. The power system short-circuit parameter prediction method based on spatiotemporal correlation as described in claim 3, characterized in that: The multi-level prediction model includes a statistical prediction layer, a deep learning layer, and a fusion layer. The statistical prediction layer captures the basic laws of parameter changes, the deep learning layer extracts complex feature relationships, and the fusion layer realizes the weight combination of different model results.

5. The power system short-circuit parameter prediction method based on spatiotemporal correlation as described in claim 4, characterized in that: The construction of the multi-level prediction model includes a statistical prediction layer that receives spatiotemporal correlation feature vectors and uses time series analysis and regression models to process time-dimensional and spatial-dimensional features. The deep learning layer receives spatiotemporally correlated feature vectors and uses multi-layer nonlinear mapping to process spatiotemporally coupled features. The fusion layer receives the outputs from the statistical prediction layer and the deep learning layer, and uses an adaptive weighting mechanism to determine the combined weights of the results from each layer.

6. The power system short-circuit parameter prediction method based on spatiotemporal correlation as described in claim 5, characterized in that: The phased training strategy includes using historical spatiotemporal correlation feature vector data to build a basic model during the offline training phase; During the online learning phase, a sliding window mechanism is used to select the latest spatiotemporal correlation feature vector data to update the model parameters in real time; A probabilistic prediction framework is introduced, which generates prediction intervals through Monte Carlo simulation based on the uncertainty of spatiotemporal correlation feature vectors.

7. The power system short-circuit parameter prediction method based on spatiotemporal correlation as described in claim 6, characterized in that: The output of the prediction results includes generating predicted values ​​and prediction intervals for short-circuit parameters based on the input spatiotemporal correlation feature vector; Establish a prediction correction mechanism to compare the actual observed short-circuit parameters with the prediction results and optimize the prediction results through real-time feedback; The reliability assessment of prediction results is provided based on the quality of spatiotemporal correlation feature vectors and model performance.

8. A power system short-circuit parameter prediction system based on spatiotemporal correlation, based on the power system short-circuit parameter prediction method based on spatiotemporal correlation according to any one of claims 1 to 7, characterized in that: It also includes a construction module for constructing a power system state vector and extracting a spatiotemporal correlation feature vector based on the state vector; The input module uses the spatiotemporal correlation feature vector as input data to construct a multi-level prediction model. The multi-level prediction model achieves a comprehensive characterization of the dynamic characteristics of short-circuit parameters through spatiotemporal feature fusion. The training module employs a phased training strategy to train the multi-level prediction model and establishes a prediction mapping relationship based on historical data of the spatiotemporal correlation feature vectors. The output module outputs the short-circuit parameter prediction results based on the trained prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the power system short-circuit parameter prediction method based on spatiotemporal correlation as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the power system short-circuit parameter prediction method based on spatiotemporal correlation as described in any one of claims 1 to 7.

Images