Method, system and equipment for predicting short-circuit capacity of power distribution network based on artificial intelligence, and medium

Abstract

The invention discloses a power distribution network short-circuit capacity prediction method, system and device based on artificial intelligence and a medium, and belongs to the technical field of power systems and automation thereof, and the method comprises the steps: carrying out the preprocessing of original data of the power distribution network short-circuit capacity, defining a power distribution network node state vector, and carrying out the feature extraction based on the state vector; time sequence modeling is carried out on the extracted features, the feature extraction capability is enhanced through an attention mechanism, and a prediction model is constructed based on the extracted features to map the features into a short-circuit capacity prediction value; and meanwhile, an optimal prediction effect is obtained through a loss function and optimized, and meanwhile, a multi-dimensional evaluation system is constructed to evaluate the prediction performance of the model. Through organic combination of multi-dimensional state characterization, time sequence-attention collaborative modeling and multi-target optimization evaluation, comprehensive improvement of the short-circuit capacity prediction of the power distribution network in accuracy, robustness and engineering practicability is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power systems and their automation, in particular to a power distribution network short-circuit capacity prediction method, system, device and medium based on artificial intelligence. BACKGROUND

[0002] With the continuous expansion of the scale and the increasing complexity of the structure of the power distribution network, accurate prediction of short-circuit capacity is crucial to ensure safe operation of the system. Traditional short-circuit capacity calculation methods mainly rely on network topology structure and device parameters, and the calculation process is complex and difficult to adapt to the dynamic changes of the power distribution network. Although there have been studies using machine learning methods for short-circuit capacity prediction, the prediction accuracy and real-time performance still need to be improved when facing large-scale data and complex scenarios. SUMMARY

[0003] In view of the above problems, the present application is proposed.

[0004] Therefore, the present application aims to solve the core technical problems of low calculation efficiency, poor adaptability and insufficient prediction accuracy of traditional power distribution network short-circuit capacity calculation methods in dynamic and complex operating environments To solve the above technical problems, the present application provides the following technical solutions: a power distribution network short-circuit capacity prediction method based on artificial intelligence, comprising, The original data of the power distribution network short-circuit capacity is preprocessed, the node state vector of the power distribution network is defined, and feature extraction is performed based on the state vector; the extracted features are time-series modeled, and the feature extraction capability is enhanced through an attention mechanism; based on the extracted features, a prediction model is constructed to map the features to short-circuit capacity prediction values; at the same time, the optimal prediction effect is obtained through a loss function and optimized, and a multi-dimensional evaluation system is constructed to evaluate the prediction performance of the model.

[0005] As a preferred scheme of the power distribution network short-circuit capacity prediction method based on artificial intelligence, the preprocessing includes standardizing the original data; An accurate system state description is established, and a node state vector of the power distribution network is defined; Based on the node state vector of the power distribution network, an improved multi-layer perception mechanism is used for feature extraction.

[0006] As a preferred scheme of the power distribution network short-circuit capacity prediction method based on artificial intelligence, the time-series modeling of the extracted features includes time-series modeling of the extracted features through an LSTM network; The discarded proportion is determined to determine the discarded historical information; The update degree of the current input information and the output of the management information are controlled.

[0007] As a preferred scheme of the power distribution network short-circuit capacity prediction method based on artificial intelligence, the enhanced feature extraction capability comprises enhancing the feature extraction capability through a multi-head self-attention mechanism on the basis of feature extraction and time series modeling. The multiple groups of independent attention results are calculated in parallel, and are fused through a connection operation and a linear transformation.

[0008] The preferred technical scheme in the embodiment of the application has the beneficial effects that the LSTM network and the multi-head self-attention mechanism are used to enhance the feature extraction capability, the time dependence is effectively captured, and the prediction accuracy is improved.

[0009] As a preferred scheme of the power distribution network short-circuit capacity prediction method based on artificial intelligence, the constructing the prediction model to map the features to the short-circuit capacity prediction values comprises, after the feature extraction and the attention enhancement are completed, constructing the prediction model to map the features to the short-circuit capacity prediction values, based on the extracted features, constructing the prediction model to be , is the predicted short-circuit capacity, is the time series feature, and g is an output layer activation function. Meanwhile, a batch training strategy is used, the number of samples in each batch is 64, the number of training rounds is 100, the initial value of the learning rate is set to 0.001, and a cosine annealing strategy is used to dynamically adjust the learning rate: wherein T is the total number of training rounds, t is the current round number, is the learning rate value used by the model in the t-th training round, is the maximum learning rate, is the minimum learning rate. A cross-validation method is used to evaluate the model performance, the proportion of the validation set is 20%, the early stopping strategy is used to prevent overfitting, and the training is stopped when the validation loss does not improve for 10 consecutive rounds.

[0010] As a preferred scheme of the power distribution network short-circuit capacity prediction method based on artificial intelligence, the obtaining the optimal prediction effect through the loss function comprises designing a comprehensive loss function, considering the prediction error and a regularization term: wherein, , , is a weight coefficient, is a true value, is the predicted value, MSE is the mean squared error, MAE is the mean absolute error, L2 is the regularization term, n is the total number of samples, and i is the variable index.

[0011] The preferred technical solution in the embodiments of the present invention has the following advantages: designing a loss function that combines prediction error and regularization term, optimizing the model training process, preventing overfitting, and thus obtaining the best prediction effect.

[0012] As a preferred embodiment of the artificial intelligence-based distribution network short-circuit capacity prediction method described in this invention, the prediction performance evaluation model constructed using a multi-dimensional evaluation system includes optimizing the model using an improved Adam optimizer, combining momentum and adaptive learning rate. The improved Adam optimizer is as follows: in, Let be the first-order momentum at the current time t. For the previous moment The first momentum at time, Let be the second momentum at the current time t. For the previous moment Second momentum at time This is the current gradient; The current parameter value. The parameter value is from the previous time step; It is the first-order momentum decay factor. It is the second-order momentum decay factor. For learning rate, It is the numerical stability constant; Constructing a multi-dimensional evaluation indicator system: Where RMSE is the root mean square error and MAPE is the mean absolute percentage error. The coefficient of determination.

[0013] The preferred technical solution in the embodiments of the present invention has the following advantages: by adopting an improved Adam optimizer and a multi-index evaluation system, the learning rate is dynamically adjusted and the model performance is comprehensively evaluated, ensuring the robustness and generalization ability of the model.

[0014] Another objective of this invention is to provide an artificial intelligence-based distribution network short-circuit capacity prediction system.

[0015] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an artificial intelligence-based distribution network short-circuit capacity prediction system, comprising: a feature extraction module, a time series modeling module, and a prediction output module; The feature extraction module preprocesses the raw data of the distribution network short-circuit capacity, defines the distribution network node state vector, and performs feature extraction based on the state vector. The temporal modeling module performs temporal modeling on the extracted features and enhances the feature extraction capability through an attention mechanism. Based on the extracted features, a prediction model is constructed to map the features to short-circuit capacity prediction values. The prediction output module obtains and optimizes the optimal prediction effect through a loss function, and at the same time constructs a multi-dimensional evaluation system to evaluate the prediction performance of the model.

[0016] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the artificial intelligence-based distribution network short-circuit capacity prediction method.

[0017] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the artificial intelligence-based distribution network short-circuit capacity prediction method.

[0018] The beneficial effects of this invention are as follows: By standardizing the original data and constructing a multi-dimensional node state vector, followed by nonlinear feature extraction using an improved multilayer perceptron, this invention achieves a comprehensive and abstract feature representation of the electrical state of the distribution network, laying a high-quality data foundation for prediction. Furthermore, by introducing an LSTM network to capture temporal dependencies, combining it with a multi-head self-attention mechanism to enhance key feature recognition, and employing a dynamic learning rate strategy for model training, this invention achieves accurate capture and efficient learning of the dynamic evolution of the power grid's operating state, significantly improving the model's adaptability to complex operating conditions. Finally, by designing a loss function that integrates MSE, MAE, and regularization, and using an improved Adam optimizer for collaborative optimization, and constructing a multi-dimensional evaluation system for full-process monitoring, this invention achieves the optimal balance between prediction accuracy, model stability, and generalization ability. These interconnected and synergistic steps collectively achieve the beneficial effect of comprehensively improving the accuracy, robustness, and engineering practical value of short-circuit capacity prediction, effectively meeting the high standards of operational safety and reliability required by smart grids. Attached Figure Description

[0019] 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.

[0020] Figure 1 The present invention provides an overall flowchart of an artificial intelligence-based method for predicting the short-circuit capacity of a distribution network, which is an embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides an artificial intelligence-based method for predicting the short-circuit capacity of a distribution network, comprising: S100. Preprocess the raw data of the distribution network short-circuit capacity, define the distribution network node state vector, and extract features based on the state vector; S200. Perform time-series modeling on the extracted features and enhance the feature extraction capability through an attention mechanism. Based on the extracted features, construct a prediction model to map the features to the predicted short-circuit capacity value. S300, at the same time, obtain the optimal prediction effect through the loss function and optimize it, and at the same time construct a multi-dimensional evaluation system to evaluate the prediction performance of the model; It should be noted that existing technologies have the following limitations in predicting distribution network short-circuit capacity: First, traditional methods often rely on a single physical model or a simple data-driven model, failing to effectively integrate temporal characteristics and key state information, resulting in limited prediction accuracy; second, these methods usually ignore the dynamic temporal characteristics of power grid state data or lack the ability to focus on key features, making it difficult for the model to capture complex change patterns; third, existing methods are prone to overfitting during model training and lack a comprehensive evaluation system to fully guarantee the model's generalization ability and robustness.

[0023] Therefore, to address the aforementioned problems, a neural network structure comprising three modules—feature extraction, temporal modeling, and prediction output—is constructed through steps S100–S300. By fusing feature extraction, temporal modeling, and an attention mechanism in these three stages, the spatiotemporal dynamic characteristics of the data are effectively captured. Furthermore, optimization and validation are performed using a comprehensive loss function and a multi-dimensional evaluation system, thereby achieving a synergistic improvement in both prediction accuracy and model generalization ability.

[0024] Example 2, refer to Figure 1 This is one embodiment of the present invention, which provides an artificial intelligence-based method for predicting the short-circuit capacity of a distribution network, comprising: In this embodiment of the invention, step S100 involves preprocessing the raw data of the distribution network short-circuit capacity, defining the distribution network node state vector, and extracting features based on the state vector, including the following steps S101-S102: In an embodiment of the present invention, S101, preprocessing the raw data of the distribution network short-circuit capacity includes the following step A1: A1. Standardize the raw data: in, The mean, The standard deviation is denoted as .

[0025] In an optional implementation, the standardization process in S101 can be as follows: calculate the maximum absolute value of all values ​​in the original data to determine its order of magnitude (i.e., a power of 10); then divide each original data value by this power of 10 to scale the data range to the interval [-1, 1].

[0026] In another optional implementation, the standardization process in S101 can also be as follows: calculate the median and quartile range (i.e., the difference between the upper quartile and the lower quartile) of the original data; subtract the median from each original data value and then divide by the quartile range, thereby reducing the impact of outliers on the standardization results.

[0027] In an embodiment of the present invention, S102, an accurate system state description is established, and the distribution network node state vector is defined; based on the distribution network node state vector, an improved multi-layer sensing mechanism is used for feature extraction, including the following steps B1-B2: B1. In distribution network short-circuit capacity prediction, the primary task is to establish an accurate system state description. This involves defining the distribution network node state vector. ,in, , These are the node voltage and current, respectively. For equivalent impedance, , Active and reactive power It is the phase angle.

[0028] By constructing a state vector containing six key parameters Among them, node voltage and current Reflects the basic electrical characteristics of the node; equivalent impedance Characterizes the response characteristics of nodes to short-circuit faults; active power and reactive power Describes the load state of the node; phase angle This reflects the angular relationship between the node voltage and the reference phase. This multi-dimensional state representation method lays the foundation for subsequent feature extraction and predictive modeling.

[0029] B2. Based on the system state vector, a powerful feature extraction mechanism is needed to capture the nonlinear relationships between state parameters. An improved multilayer perceptron is used for feature extraction: in, This is the output of layer l. This is the weight matrix. For bias vectors, This is the activation function.

[0030] In each layer of the network, the input features Through the weight matrix Linear transformation and bias vector The translation, followed by the activation function The nonlinear mapping yields higher-level feature representations. Through multi-level cascading, the model can progressively extract more abstract and effective feature representations.

[0031] In an optional implementation, feature extraction in S102 can be based on convolutional neural network (CNN) feature extraction, which reorganizes the distribution network node state vector into a two-dimensional grid structure, such as by time series or node topology arrangement, to simulate image data. Multiple convolutional layers are applied for feature extraction, with each convolutional layer using a small filter to scan the grid and capture local feature patterns (such as short-term fluctuations in voltage and current). Pooling layers (such as max pooling) are used to reduce the feature dimensionality and retain salient features. The extracted features are then mapped to a high-level representation through a fully connected layer for subsequent prediction. However, this approach is not suitable for non-image-like data of distribution network state vectors, resulting in insufficient capture of sequence dependencies in feature extraction.

[0032] In another alternative implementation, feature extraction in S102 can also be based on autoencoder feature extraction, constructing a symmetric autoencoder network comprising an encoder and a decoder. The encoder consists of multiple fully connected layers, progressively compressing the state vector into the latent space (low-dimensional representation), introducing nonlinearity through activation functions to extract key features (such as the implicit relationship between impedance and power). The decoder reconstructs the original state vector from the latent features, minimizing reconstruction errors during training to ensure that the latent features retain key information. The encoder's output is used as the extracted features for short-circuit capacity prediction; however, its specificity for short-circuit capacity prediction is relatively weak.

[0033] In this embodiment of the invention, S200 performs temporal modeling on the extracted features and enhances the feature extraction capability through an attention mechanism. Based on the extracted features, a prediction model is constructed to map the features to short-circuit capacity prediction values, including the following steps S201-S203: S201. The extracted features are modeled temporally using an LSTM network to determine the discard ratio, which determines the historical information to be discarded, controls the update level of the current input information, and manages the output of the information. Considering the significant temporal correlations in the operating states of the distribution network, an LSTM network is introduced to perform temporal modeling of the extracted features. Through a carefully designed gating mechanism, the LSTM network can effectively capture both long-term and short-term temporal dependencies. The temporal features are modeled using an LSTM network as follows: in, The "Forget Gate" determines the proportion of information to be discarded; As an input gate, it controls the degree to which new information is updated; This represents the current state of the cell. Candidate cell state; This represents the cell state at the previous moment; This is the hidden state from the previous moment; This is the input for the current moment; , , This is the corresponding weight matrix; , , For the corresponding bias vector; It is the sigmoid activation function; It is the hyperbolic tangent activation function; This represents the Hadamard (element-by-element) product.

[0034] The network contains three key gating units: the forget gate. Decide which historical information to discard, input gate Controls the update level of the current input information, output gate (through cell state) (Reflects) the output of management information.

[0035] in, This represents the sigmoid activation function, used to normalize the gated signal to the [0,1] interval; The function is used to generate candidate state values; symbol This indicates an element-wise multiplication operation.

[0036] With the help of this gating mechanism, the LSTM network can adaptively maintain and update the timing information, providing a more reliable feature representation for short-circuit capacity prediction.

[0037] S202. Based on feature extraction and temporal modeling, feature extraction capability is enhanced through a multi-head self-attention mechanism; multiple independent attention results are computed in parallel and fused through connection operations and linear transformations. Building upon feature extraction and temporal modeling, to further enhance the model's ability to identify key features, a multi-head self-attention mechanism is introduced to strengthen feature extraction capabilities: Where Q, K, and V are the query, key, and value matrices, respectively. For feature dimensions.

[0038] Basic attention calculations are performed by calculating the similarity between the query matrix Q and the key matrix K, combined with a scaling factor. Normalization is performed, and finally, the weighted feature representation is obtained by multiplying it with the value matrix V. This mechanism can adaptively assign importance weights to different features. Moreover, the multi-head attention mechanism computes multiple independent attention results in parallel and uses concatenation operations and linear transformations. By fusing features, the model can simultaneously focus on different feature patterns from multiple perspectives. This design significantly enhances the model's ability to capture complex feature relationships.

[0039] S203. After completing feature extraction and attention enhancement, a prediction model is constructed that maps features to predicted short-circuit capacity values. Based on the extracted features, a prediction model is constructed as follows: ,in, For the predicted short-circuit capacity, For temporal features, g is the activation function of the output layer; The prediction model employs an improved linear mapping structure, where It includes feature representations that have undergone temporal modeling and attention processing, and is expressed through a weight matrix. and bias vector The transformation is performed, and finally the short-circuit capacity prediction value is obtained through the activation function g. To ensure the accuracy and stability of the prediction results, the activation function g is carefully designed to both guarantee the nonlinear expressiveness of the output and avoid the gradient vanishing problem.

[0040] Simultaneously, a batch training strategy is adopted, with 64 samples per batch, 100 training rounds, and an initial learning rate of 0.001. Cosine annealing is used for dynamic adjustment. Where T is the total number of training rounds, and t is the current round number. To represent the learning rate value used by the model in the t-th training round, To maximize the learning rate, Minimum learning rate; Cross-validation was used to evaluate model performance, with a validation set ratio of 20%. An early stopping strategy was employed to prevent overfitting, and training was stopped when the validation loss showed no improvement for 10 consecutive rounds.

[0041] In an embodiment of the present invention, step S300 simultaneously obtains and optimizes the optimal prediction effect through a loss function, and constructs a multi-dimensional evaluation system to evaluate the prediction performance of the model, including the following steps S301-S302: In an embodiment of the present invention, S301, to ensure the effectiveness of model training, a comprehensive loss function is designed, including the following step C1: C1. Considering prediction error and regularization term: in, , , These are the weighting coefficients. For the true value, is the predicted value, MSE is the mean squared error, MAE is the mean absolute error, L2 is the regularization term, n is the total number of samples, and i is the variable index.

[0042] The loss function L comprises three key components: mean squared error (MSE), mean absolute error (MAE), and L2 regularization. MSE is more sensitive to larger prediction biases, helping to suppress outliers; MAE focuses more on the absolute magnitude of the overall prediction bias, contributing to improved prediction stability; and the L2 regularization term prevents overfitting. These are determined by the weighting coefficients. , , The dynamic adjustment can balance these three objectives during training to achieve optimal prediction results.

[0043] In an optional implementation, achieving optimal prediction performance in S301 can be achieved by using Huber loss combined with L2 regularization. During model training, the Huber loss function replaces the MSE and MAE combination in the original scheme. The Huber loss balances the behavior of squared error and absolute error through a threshold parameter: when the prediction error is less than the threshold, squared error is used to maintain smoothness; when the prediction error is greater than the threshold, absolute error is used to reduce the impact of outliers. Simultaneously, the L2 regularization term is retained to prevent overfitting, and the regularization strength is adjusted through weight coefficients. During training, the threshold parameter is typically set to a fixed value and fine-tuned through cross-validation; however, it cannot optimally balance error sensitivity when the data distribution is uneven.

[0044] In another optional implementation, achieving optimal prediction performance in S301 can be achieved by replacing the MSE and MAE combination in the original scheme with a quantile loss function during model training. Quantile loss quantifies the asymmetry of prediction error by setting quantiles (such as the median quantile), enabling the model to better handle biases in different directions. Simultaneously, the L2 regularization term is retained to control model complexity. During training, the quantile is typically chosen to be 0.5 (corresponding to median regression), and the quantile positions are optimized through grid search to adapt to specific scenarios. However, quantile loss is sensitive to the assumption of symmetry in the error distribution, and its efficiency is lower than MSE in symmetric error scenarios, with higher computational costs.

[0045] S302. The training performance of the model largely depends on the choice of optimization algorithm. An improved Adam optimizer, combining momentum and adaptive learning rate, is used for optimization. The improved Adam optimizer is as follows: in, Let be the first-order momentum at the current time t. For the previous moment The first momentum at time, Let be the second momentum at the current time t. For the previous moment Second momentum at time This is the current gradient; The current parameter value. The parameter value is from the previous time step; The first-order momentum decay factor is set to 0.9; The second-order momentum decay factor is set to 0.999; The learning rate is set to 0.999. Let be the numerical stability constant, and take 1e. -8 ; First-order momentum Through parameters Controlling the decay of historical gradient information to smooth the gradient update path; second-order momentum. Then through parameters The squares of accumulated historical gradients are used to adaptively adjust the learning step size of each parameter. The final parameter update rule is as follows: ,in Based on the learning rate, A small constant is used for numerical stability. This optimization strategy not only accelerates the convergence speed but also effectively avoids local optima. Moreover, this invention improves the traditional Adam algorithm by introducing a dynamic decay factor and gradient pruning mechanism, further enhancing the stability and efficiency of the optimization process.

[0046] In an embodiment of the present invention, constructing a multi-dimensional evaluation system to evaluate the predictive performance of the model includes the following steps D1: D1. To comprehensively evaluate the predictive performance of the model, a multi-dimensional evaluation index system is constructed: Where RMSE is the root mean square error and MAPE is the mean absolute percentage error. The coefficient of determination.

[0047] The root mean square error (RMSE) reflects the overall deviation between predicted and actual values. It is more sensitive to larger errors and helps to identify significant prediction biases in the model. The mean absolute percentage error (MAPE) provides a measure of relative error, facilitating performance comparisons across prediction tasks of different magnitudes. The coefficient of determination... These three metrics measure how well the model's predicted values ​​explain changes in the actual values; the closer the value is to 1, the better the model's predictive performance. They together form a complete evaluation system, assessing model performance from different perspectives.

[0048] These metrics are not only used to evaluate the performance of the final model, but also play a crucial guiding role in the training process. By monitoring changes in these metrics in real time, problems in model training, such as overfitting or underfitting, can be identified promptly, and corresponding adjustments can be made to ensure the effectiveness and reliability of model training. This dynamic evaluation and optimization mechanism enables the method of this invention to maintain optimal predictive performance at all times.

[0049] In an optional implementation, the predictive performance of the model in S302 can be evaluated by using mean absolute error (MAE) and mean squared error (MSE) as evaluation metrics. The mean absolute error (MAE) is calculated by averaging the absolute values ​​of the differences between the predicted and actual values ​​of all samples, thus assessing the overall level of prediction error and reflecting the stability and consistency of the model. The mean squared error (MSE) is calculated by averaging the squares of the differences between the predicted and actual values ​​of all samples, emphasizing the impact of larger errors and helping to identify significant biases in the model. Combining the MAE and MSE metrics comprehensively evaluates the model's predictive performance from both the absolute magnitude of the error and the squared amplification effect, ensuring the model's reliability under both normal and abnormal conditions.

[0050] In another optional implementation, the predictive performance of the model in S302 can also be evaluated using the Median Absolute Error (MedAE) and the Explained Variance Score as evaluation metrics. The Median Absolute Error (MedAE) is calculated by taking the median of the absolute values ​​of the differences between the predicted values ​​and the true values ​​of all samples, which reduces the impact of outliers or extreme cases on the evaluation results and provides a more robust error assessment. The Explained Variance Score is calculated by taking the degree to which the model's predicted values ​​explain the variance of the true values, which evaluates the model's goodness of fit and generalization ability, and reflects the model's ability to capture data changes. Combining the MedAE and the Explained Variance Score evaluates the model's performance from two dimensions: anti-interference and variance explanation, which is suitable for scenarios where the data is noisy or unevenly distributed.

[0051] Example 3 is an embodiment of the present invention, which provides a method for predicting the short-circuit capacity of a distribution network based on artificial intelligence. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0052] This study uses actual operating data from a provincial power company for 2023-2024 for verification, including historical operating data from 10 substations. The experimental results are as follows: Table 1 Comparison of prediction performance of different methods

[0053] Table 2. Forecasting Error Analysis at Different Load Levels (MAPE%)

[0054] Table 3 Model Stability Analysis

[0055] Experimental results show that the method of this invention has significant advantages over traditional methods and other machine learning methods: the prediction accuracy is significantly improved, the RMSE is reduced to 75.3 MVA, and the MAPE is reduced to 4.86%; the computational efficiency is greatly improved, with a prediction time of only 12.8 seconds; it exhibits good prediction performance under different load levels; and the model has good stability with small fluctuations in various indicators.

[0056] Example 4 is an embodiment of the present invention. The above is an illustrative scheme of a distribution network short-circuit capacity prediction method based on artificial intelligence. It should be noted that the technical solution of a distribution network short-circuit capacity prediction system based on artificial intelligence and the technical solution of the distribution network short-circuit capacity prediction method based on artificial intelligence described above belong to the same concept. For details not described in detail in the technical solution of the distribution network short-circuit capacity prediction system based on artificial intelligence in this embodiment, please refer to the description of the technical solution of the distribution network short-circuit capacity prediction method based on artificial intelligence described above.

[0057] This embodiment provides an artificial intelligence-based distribution network short-circuit capacity prediction system, including: a feature extraction module, a time series modeling module, and a prediction output module; The feature extraction module preprocesses the raw data of the distribution network short-circuit capacity, defines the distribution network node state vector, and performs feature extraction based on the state vector. The temporal modeling module performs temporal modeling on the extracted features and enhances the feature extraction capability through an attention mechanism. Based on the extracted features, a prediction model is constructed to map the features to short-circuit capacity prediction values. The prediction output module obtains and optimizes the optimal prediction effect through a loss function, and at the same time constructs a multi-dimensional evaluation system to evaluate the prediction performance of the model.

[0058] This embodiment also provides an electronic device applicable to an artificial intelligence-based distribution network short-circuit capacity prediction method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the artificial intelligence-based distribution network short-circuit capacity prediction method proposed in the above embodiment.

[0059] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements an artificial intelligence-based method for predicting the short-circuit capacity of a distribution network as proposed in the above embodiments.

[0060] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for predicting the short-circuit capacity of a distribution network based on artificial intelligence proposed in the above embodiments. 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.

[0061] Based on the above description of the implementation methods, those skilled in the art can 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, but in many cases the former is a better implementation method. 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.

[0062] 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. An artificial intelligence-based short circuit capacity prediction method for a power distribution network, characterized by: The method comprises the following steps: The pre-processing comprises standardizing the original data; An accurate system state description is established, and a distribution network node state vector is defined; Based on the distribution network node state vector, an improved multi-layer perception mechanism is used for feature extraction.

2. The method of claim 1, wherein the method is based on artificial intelligence. The time series modeling of the extracted features comprises time series modeling of the extracted features through an LSTM network; The discarded proportion is determined to determine the discarded historical information; The update degree of the current input information and the output of the management information are controlled.

3. The method of claim 2, wherein the method is based on artificial intelligence. The enhanced feature extraction capability comprises enhancing the feature extraction capability through a multi-head self-attention mechanism based on feature extraction and time series modeling; Parallel computing of multiple independent attention results is performed, and fusion is performed through connection operation and linear transformation. A batch training strategy is used, the number of samples per batch is 64, the number of training rounds is 100, the initial value of the learning rate is set to 0.001, and a cosine annealing strategy is used for dynamic adjustment:

4. The method of claim 3, wherein the method is based on artificial intelligence. A cross-validation method is used to evaluate the model performance, the verification set proportion is 20%, the early stopping strategy is used to prevent overfitting, and the training is stopped when the verification loss does not improve for 10 consecutive rounds. The optimal prediction effect obtained through the loss function comprises designing a comprehensive loss function considering the prediction error and the regularization term:

5. The method of claim 4, wherein the method is based on artificial intelligence. The constructing the prediction model to map the features to the short-circuit capacity prediction value includes, after the feature extraction and attention enhancement are completed, constructing the prediction model to map the features to the short-circuit capacity prediction value, based on the extracted features, the prediction model is constructed to wherein, is the predicted short-circuit capacity, is the time sequence feature, and g is an output layer activation function. The improved Adam optimizer is used for optimization combined with momentum and adaptive learning rate, and the improved Adam optimizer is: wherein T is the total number of training rounds, t is the current round number, is a learning rate value used by the model at the tth training round, is the maximum learning rate, is the minimum learning rate; A multi-dimensional evaluation index system is constructed to evaluate the prediction performance of the model:

6. The method of claim 5, wherein the method is based on artificial intelligence. The method comprises the following steps: wherein, , , is a weight coefficient, is a true value, is a predicted value, MSE is a mean squared error, MAE is a mean absolute error, L2 is a regularization term, n is a total number of samples, and i is a variable index.

7. The method of claim 6, wherein the method is based on artificial intelligence. The feature extraction module, the time series modeling module, and the prediction output module are used to perform the pre-processing on the original data of the distribution network short-circuit capacity, define the distribution network node state vector, and perform feature extraction based on the state vector. wherein, is the first order momentum at the current time t, is the first order momentum at the previous time is the first order momentum at the previous time is the second order momentum at the current time t, is the second order momentum at the previous time is the second order momentum at the previous time is the current gradient; is the current parameter value, is the previous parameter value; is the first order momentum decay factor, is the second order momentum decay factor, is the learning rate, is a numerical stability constant; The time series modeling module performs time series modeling on the extracted features, enhances the feature extraction capability through the attention mechanism, and constructs a prediction model to map the features to short-circuit capacity prediction values based on the extracted features. where RMSE is the root mean square error and MAPE is the mean absolute percentage error, R2is the coefficient of determination.

8. An artificial intelligence-based short circuit capacity prediction system for power distribution networks, applying an artificial intelligence-based short circuit capacity prediction method according to any one of claims 1 to 7, characterized in that, The prediction output module obtains the optimal prediction effect through the loss function and optimizes it, and constructs a multi-dimensional evaluation system to evaluate the prediction performance of the model. The processor executes the computer program to realize the steps of the short-circuit capacity prediction method of the distribution network based on artificial intelligence in any one of claims 1 to 7. The computer program is executed by the processor to realize the steps of the short-circuit capacity prediction method of the distribution network based on artificial intelligence in any one of claims 1 to 7. ​ ​ 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. ​ 10. A computer-readable storage medium having stored thereon a computer program, characterized in that, ​

Images