Load prediction method, system and device based on situation of power distribution network, and storage medium

Abstract

The invention discloses a load prediction method, system and device based on the situation of a power distribution network and a storage medium, and the method comprises the steps: collecting historical load data and real-time monitoring data of the power distribution network, and carrying out the ultra-short-term load prediction; calculating a key load index of the power distribution network based on the real-time monitoring data, evaluating the safety state of the key load index, and generating a preliminary safety situation evaluation result; and fusing the preliminary security situation assessment result and the ultra-short-term load prediction result, and predicting the final regional power distribution network security situation. And optimizing operation parameters of the power distribution network according to the final load prediction result. According to the method, the defect that the non-linear characteristic processing capacity of an error correction technology is insufficient is overcome, and the dynamic correlation analysis and time sequence modeling capacity and the multi-dimensional data integration and application capacity of an existing situation assessment technology are improved.

Description

Technical Field

[0001] This invention relates to the technical field of power grid forecasting and adjustment, and in particular to a load forecasting method, system, device and storage medium based on the distribution network status. Background Technology

[0002] Currently, with the rapid development of global energy structure adjustment and smart grid technology, the distribution network, as an important component of the power system, undertakes the important functions of power transmission, dispatch, and distribution. The construction of intelligent distribution networks has gradually become a key goal of power system modernization. The safety, reliability, and economy of its operation directly affect the stable operation of the entire power grid. As an important component of the smart grid, load forecasting technology has received widespread attention in recent years. Traditional load forecasting methods mainly include time series analysis, regression analysis, and exponential smoothing models. These methods were widely used in the early stages of power grid development due to their simple calculation and convenient application. With the large-scale development of modern distribution networks and the increasing complexity of load demand, load data exhibits significant nonlinear dynamic characteristics, including periodic fluctuations, random disturbances, and sudden changes. With the development of science and technology, artificial intelligence technology has gradually been introduced into the field of load forecasting. Emerging deep learning algorithms such as Least Squares Support Vector Machine (LS-SVM), Generative Adversarial Network (GAN), and Long Short-Term Memory (LSTM) have been widely studied and applied. These methods have shown significant advantages in handling complex load characteristics through high-dimensional feature mapping, nonlinear modeling, and time series dependency modeling.

[0003] Despite significant progress in load forecasting and situation assessment technologies in recent years, several problems remain to be addressed, particularly in handling dynamic load characteristics, leading to large prediction deviations in special scenarios. Furthermore, existing error correction methods rely on simple linear compensation models, lacking multidimensional optimization and dynamic adjustment capabilities, thus affecting the reliability of prediction results. Situation assessment technologies have limitations in dynamic correlation analysis and time series modeling of key load indicators, failing to fully capture the spatiotemporal evolution characteristics of distribution network operation, resulting in insufficient accuracy in safety assessments. Operation parameter optimization methods are mostly based on static scheduling models, failing to fully consider real-time distribution network operation data and the dynamic characteristics of the State of Charge (SOC) of energy storage devices, especially in energy storage scheduling, where dynamic adjustment and efficient matching are difficult to achieve. Summary of the Invention

[0004] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a load forecasting method, system, device, and storage medium based on distribution network conditions to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, embodiments of the present invention provide a load forecasting method based on the status of a distribution network, comprising: collecting historical load data and real-time monitoring data of the distribution network, and performing ultra-short-term load forecasting;

[0007] Based on the real-time monitoring data, the key load indicators of the distribution network are calculated, and the safety status of the key load indicators is assessed to generate preliminary safety situation assessment results.

[0008] By combining the preliminary security situation assessment results with the ultra-short-term load forecast results, the final regional distribution network security situation is predicted.

[0009] Optimize the distribution network operating parameters based on the final load forecast results.

[0010] As a preferred embodiment of the load forecasting method based on distribution network status described in this invention, the following steps are included: collecting historical load data and real-time monitoring data of the distribution network to perform ultra-short-term load forecasting:

[0011] Extract periodic features from historical load data, identify daily fluctuation patterns and seasonal trends in load, and extract long-term trends from the data as input features for ultra-short-term load forecasting models.

[0012] The periodic characteristics include daily periodic load demand fluctuations, weekly periodic load demand fluctuations, and annual periodic load demand fluctuations.

[0013] The objective function of the ultra-short-term load forecasting model is characterized as the error between the predicted load value and the actual load value, and the model parameters are trained iteratively.

[0014] Based on the trained ultra-short-term load prediction model, the ultra-short-term load of the distribution network is predicted by combining the real-time input features extracted from real-time monitoring data.

[0015] As a preferred embodiment of the load forecasting method based on distribution network status described in this invention, the method further includes: calculating the error between the actual load value and the ultra-short-term load forecast value obtained from real-time monitoring data, obtaining the difference between the ultra-short-term load forecast value and the actual load value, and correcting the forecast error.

[0016] An error dataset is constructed using the initial prediction error. The initial prediction error and random noise are concatenated to form the input vector of the generative adversarial network model. The generative adversarial network model is constructed using the input vector as the input data of the generator. The prediction error is optimized once according to the activation function.

[0017] The initial prediction error from the error dataset and the first-order optimized prediction error generated by the generator are used as input data to the discriminator to calculate the probability that a sample is a true error sample; the training strategies of the discriminator and the generator are optimized until the loss function converges, and the first-order optimized prediction error is output based on the input vector of the trained generator.

[0018] The beneficial effects of this preferred technical solution are that by combining the generative adversarial network (GAN) to optimize the error correction process, it significantly enhances the ability to handle nonlinear characteristics, makes up for the limitations of traditional error correction methods, and ensures that the prediction results are more accurate and reliable.

[0019] As a preferred embodiment of the load forecasting method based on distribution network status described in this invention, the method further includes: adding the day before and the day after the holiday into the holiday, calculating the average load of weekdays and holidays, defining a holiday adjustment factor based on the relationship between the average load of weekdays and holidays, and dynamically adjusting the error value based on the holiday adjustment factor.

[0020] The adjusted initial prediction error is combined with the first optimized prediction error to update the error dataset;

[0021] Based on the features extracted from historical load data as input to the BP neural network model, a BP neural network model is constructed, and the loss function of the BP neural network model is defined.

[0022] The gradient is calculated and the weights of the BP neural network model are updated through backpropagation until the loss function converges. The updated error dataset is then input into the trained BP neural network model to obtain the error correction value predicted by the BP neural network model, and the final ultra-short-term load prediction is calculated.

[0023] As a preferred embodiment of the load forecasting method based on distribution network status described in this invention, the following steps are included: calculating key load indicators of the distribution network based on real-time monitoring data, assessing the security status of the key load indicators, and generating preliminary security status assessment results:

[0024] Collect key load indicators, including load rate, voltage deviation rate, voltage stability index, overload prediction index, and equipment health assessment index, and process the collected key load indicator data.

[0025] A comprehensive security situation assessment index is generated by comprehensively evaluating key load indicators; a preliminary security situation assessment report is generated based on the key load indicators and the comprehensive security situation assessment index.

[0026] As a preferred embodiment of the load forecasting method based on distribution network status described in this invention, the method for forecasting the final regional distribution network security status by integrating the preliminary security status assessment results and the ultra-short-term load forecast results includes: integrating the data from the preliminary security status assessment report and the ultra-short-term load forecast.

[0027] The fused data is input into the CNN-LSTM hybrid network. Spatial features are extracted through the convolutional operation of the generative adversarial network module of the CNN-LSTM hybrid network, and temporal features are extracted through the memory state update and hidden state update operations of the long short-term memory module.

[0028] Spatial and temporal features are extracted and concatenated into comprehensive features. These comprehensive features are then input into the fully connected layer of a CNN-LSTM hybrid network to output a regional distribution network security situation prediction. Based on historical load data and expert experience, risk thresholds are set, and the regional distribution network security situation prediction is used to distinguish the regional distribution network risk levels according to these risk thresholds. Finally, the regional distribution network security situation prediction is output.

[0029] The beneficial effects of this preferred technical solution are that it extracts the spatiotemporal features of the security situation through a CNN-LSTM hybrid network, optimizes dynamic correlation analysis and time series modeling techniques, thereby comprehensively improving the accuracy and predictive ability of distribution network situation assessment; and achieves intelligent optimization of key operating parameters by dynamically adjusting node voltage, line power flow and energy storage scheduling strategies.

[0030] As a preferred embodiment of the load forecasting method based on distribution network status described in this invention, optimizing the distribution network operating parameters according to the final load forecasting results includes:

[0031] Based on the periodic fluctuations of the distribution network load, the characteristics of rapid load changes, and the fluctuation period in the load data, a low-frequency threshold is set; based on the final prediction of the ultra-short-term load and the set low-frequency threshold, the ultra-short-term load prediction data is frequency decomposed to obtain the long-term load and short-term load.

[0032] Based on the scheduling system's operating parameters, the quantum particle swarm optimization algorithm is used to set the number of particles, initialize the parameters of each particle, set the objective function, and calculate the objective function value of each particle, including voltage deviation minimization, power flow optimization, and energy storage scheduling optimization; real-time state of charge data is collected from energy storage devices, and the energy storage scheduling coefficient is optimized.

[0033] The interfaces connecting electrical equipment in the distribution network are used as nodes in the voltage deviation minimization part, and the physical lines in the distribution network responsible for transmitting and distributing electrical energy are used as lines in the power flow optimization part; the rated values ​​of the distribution network operating parameters are used as constraints on the particle objective function value, and the particle fitness value is updated when the particle position violates the constraints.

[0034] The updated particle fitness value is compared with the historical particle fitness value, and the global and local optimal positions of the particles are updated according to the comparison results. The global and local optimal positions are taken as the global optimal solution and the local optimal solution, respectively. The positions of the particles and the objective function value are dynamically adjusted by updating the particle swarm fitness value until the fitness value converges. The optimized running parameters are then output and input into the scheduling system for scheduling.

[0035] Secondly, the present invention provides a load forecasting system based on the distribution network status, comprising:

[0036] The data acquisition module is used to collect historical load data and real-time monitoring data of the distribution network for ultra-short-term load forecasting.

[0037] The security assessment module is used to calculate key load indicators of the distribution network based on the real-time monitoring data, assess the security status of the key load indicators, and generate preliminary security situation assessment results.

[0038] The situation prediction module is used to integrate the preliminary security situation assessment results with the ultra-short-term load forecast results to predict the final regional distribution network security situation.

[0039] The optimization module is used to optimize the operating parameters of the distribution network based on the final load forecast results.

[0040] Thirdly, the present invention provides an electronic device, comprising:

[0041] Memory and processor;

[0042] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the load forecasting method based on the distribution network situation.

[0043] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the load forecasting method based on the distribution network situation.

[0044] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention effectively addresses the problem of insufficient modeling of load dynamic characteristics by real-time monitoring of key load indicators in the distribution network and dynamic assessment of its safety status. Especially in holiday and load surge scenarios, it significantly improves the accuracy and robustness of load forecasting. Secondly, by combining generative adversarial networks to optimize the error correction process, it significantly enhances the ability to handle nonlinear characteristics, makes up for the limitations of traditional error correction methods, and ensures more accurate and reliable prediction results. By extracting the spatiotemporal features of the safety situation through a CNN-LSTM hybrid network, it optimizes dynamic correlation analysis and time series modeling techniques, thereby comprehensively improving the accuracy and predictive ability of distribution network situation assessment. By dynamically adjusting node voltage, line power flow, and energy storage scheduling strategies, it achieves intelligent optimization of key operating parameters. In particular, based on the real-time state of charge of energy storage devices, it dynamically optimizes the energy storage scheduling coefficient, thereby achieving the effects of load peak shaving and valley filling, reducing equipment overload risk, and extending equipment life. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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. Wherein:

[0046] Figure 1 This is a schematic diagram of the method flow for a load forecasting method based on the distribution network status according to an embodiment of the present invention. Detailed Implementation

[0047] 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.

[0048] Example 1, referring to Figure 1 As one embodiment of the present invention, this embodiment provides a load forecasting method based on distribution network status, including:

[0049] S100: Collects historical load data and real-time monitoring data of the distribution network to perform ultra-short-term load forecasting;

[0050] S200: Calculates key load indicators of the distribution network based on real-time monitoring data, assesses the security status of key load indicators, and generates preliminary security situation assessment results.

[0051] S300: Integrates preliminary security situation assessment results with ultra-short-term load forecast results to predict the final regional distribution network security situation;

[0052] S400: Optimize the operating parameters of the distribution network based on the final load forecast results.

[0053] It should be noted that existing methods cannot effectively handle dynamic load characteristics, especially during holidays and periods of sudden load changes, where prediction accuracy drops significantly and traditional energy storage scheduling schemes lack dynamic adjustment and efficient matching. This invention collects historical load data and real-time monitoring data from the distribution network to perform ultra-short-term load forecasting, and corrects forecast errors to perform final ultra-short-term load forecasting. Based on real-time monitoring data of the distribution network, it collects key load indicators of the distribution network and assesses the safety status of these indicators, generating a preliminary safety situation assessment report. Based on the preliminary safety situation assessment report and the ultra-short-term load forecast, it generates the final regional distribution network safety situation forecast. It optimizes the operating parameters of the distribution network based on the final ultra-short-term load forecast. The historical load data, real-time monitoring data, and regional distribution network safety situation forecast are encrypted and transmitted to a database for storage and backup. This overcomes the shortcomings of error correction technology in handling nonlinear characteristics, improves the ability of existing situation assessment technology to perform dynamic correlation analysis and time series modeling, and also enhances the ability to integrate and apply multi-dimensional data.

[0054] In this embodiment of the invention, step S100, which involves collecting historical load data and real-time monitoring data of the distribution network and performing ultra-short-term load forecasting, includes:

[0055] Extract periodic features from historical load data, identify daily fluctuation patterns and seasonal trends in load, and extract long-term trends from the data as input features for ultra-short-term load forecasting models.

[0056] Specifically, historical load data and real-time monitoring data are collected and preprocessed through the phased array units and data acquisition and monitoring control system of the distribution network;

[0057] Historical load data includes load values, timestamps, area identifiers, and environmental data; real-time monitoring data includes voltage, current, power, frequency, phase angle, temperature, and equipment status.

[0058] Furthermore, the collected historical load data and real-time monitoring data are denoised and outlier detected, and Z-score standardized to transform the data into a distribution with a mean of 0 and a standard deviation of 1. The Min-Max method is then used to normalize the data, and the load values ​​are scaled to the range of [0, 1].

[0059] In an optional embodiment, in step S100, the periodic features in the historical load data can be extracted by the Fourier transform method, the daily fluctuation pattern and seasonal change trend of the load can be identified, and the long-term trend in the data can be extracted by regression analysis as the input features of the prediction model.

[0060] Furthermore, the periodic characteristics include daily periodic load demand fluctuations, weekly periodic load demand fluctuations, and annual periodic load demand fluctuations.

[0061] Specifically, the periodic characteristics include daily periodic load demand fluctuations, weekly periodic load demand fluctuations, and annual periodic load demand fluctuations. Daily fluctuation patterns include peak and off-peak electricity consumption periods. Seasonal change trends include summer and winter. Long-term trends include population growth, economic development, technological changes, and policy adjustments.

[0062] Furthermore, the objective function of the ultra-short-term load forecasting model is characterized as the error between the predicted load value and the actual load value, and the model parameters are trained iteratively.

[0063] Specifically, the Least Squares Support Vector Machine (LS-SVM) model is used for preliminary ultra-short-term load prediction. Based on the input features as input to the LS-SVM model, the LS-SVM model is represented as follows:

[0064]

[0065] in, Indicates input features, This represents a kernel function mapping used to map input features. Mapping from the original space to a higher-dimensional feature space Represents the weights of the model. This represents the transpose of the model's weight vector. This represents the bias term, which is an adjustable parameter in the LS-SVM model.

[0066] Define the objective function as:

[0067]

[0068] in, This represents the objective function of the LS-SVM model. Represents the regularization term, It is a regularization parameter. It is the first The error term between the predicted load value and the actual load value after model training. This indicates the number of input features.

[0069] Furthermore, based on the trained ultra-short-term load prediction model, the ultra-short-term load of the distribution network is predicted by combining the real-time input features extracted from real-time monitoring data.

[0070] Specifically, by optimizing the objective function, limiting model complexity, and iteratively adjusting model parameters, the objective function is optimized until it reaches its optimal level. Convergence occurred in the most recent training sessions;

[0071] Based on the real-time input features extracted from real-time monitoring data, a pre-trained LS-SVM model is used to predict future ultra-short-term loads. :

[0072]

[0073] in, Indicates time Ultra-short-term load forecasting Represents the Lagrange factor. Indicates real-time input features, This represents the input features for model training. Represents the kernel function. Indicates the bias term. This indicates the number of input features in real time.

[0074] It should be noted that by extracting periodic features from historical load data using Fourier transform and combining them with regression analysis to extract long-term trends, rich feature inputs are provided for the load forecasting model, enhancing its adaptability to complex load changes and improving its robustness and stability. This solves the problem of low model applicability and prediction accuracy caused by insufficient input features in traditional load forecasting. Furthermore, by using the least squares support vector machine (LS-SVM) model and mapping input features to a high-dimensional space based on the radial basis function (RBF), the model's ability to fit nonlinear characteristics is improved. By optimizing the objective function to limit model complexity and iteratively training until convergence, the accuracy and robustness of ultra-short-term load forecasting are significantly improved, especially in special scenarios such as holidays or large load fluctuations.

[0075] In this embodiment of the invention, step S100 further includes: performing error calculation based on the actual load value and the ultra-short-term load forecast value obtained from real-time monitoring data, obtaining the difference between the ultra-short-term load forecast value and the actual load value, and correcting the forecast error;

[0076] Specifically, by substituting the actual load value and the ultra-short-term load forecast value obtained from real-time monitoring data into the error calculation formula, the difference between the ultra-short-term load forecast value and the actual load value is obtained:

[0077]

[0078] in, This represents the actual load value. This represents the predicted load value. Indicates time The initial prediction error;

[0079] Furthermore, an error dataset is constructed using the initial prediction error. The initial prediction error and random noise are concatenated to form the input vector of the generative adversarial network model. The generative adversarial network model is constructed using the input vector as the input data of the generator, and the prediction error is optimized once according to the activation function.

[0080] Specifically, according to The activation function outputs the first-order optimization prediction error. :

[0081]

[0082] in, Represents the input vector. It is random noise. This is the initial prediction error based on historical load data. This represents the weight matrix from the generator input layer to the intermediate layers. This represents the bias term of the generator's hidden layer. This represents the activation function of the intermediate layer of the generator. This represents the weight matrix from the intermediate layers of the generator to the output layer. This represents the bias term of the generator output layer. This represents the activation function of the generator output layer;

[0083] Furthermore, the initial prediction error in the error dataset and the first-order optimized prediction error generated by the generator are used as input data to the discriminator to calculate the probability that a sample is a true error sample; the training strategies of the discriminator and the generator are optimized until the loss function converges, and the first-order optimized prediction error is output based on the input vector of the trained generator.

[0084] Specifically, the initial prediction error in the error dataset And the first-order optimized prediction error generated by the generator The data is fed into the discriminator to calculate the probability that a sample is a true error sample:

[0085]

[0086] in, This represents the probability that the discriminator outputs a true error sample, and its value ranges from [0, 1]. This represents the input data of the discriminator. Representing a non-linear activation function enhances the expressive power of the decision maker. This represents the weight matrix of the discriminator output layer. This represents the bias term of the discriminator output layer. This indicates that the Sigmoid activation function is used to map the output to a range of values ​​[0, 1].

[0087] Calculate the generator loss and discriminator loss based on the loss function:

[0088] Generator loss function:

[0089]

[0090] in, Indicates generator loss. Indicates random noise follows a distribution, This represents the prediction error for a single optimization. The judgment value;

[0091] Discriminator loss function:

[0092]

[0093] in, Indicates discriminator loss. Indicates the initial prediction error The judgment value;

[0094] The training strategy of fixing the generator to optimize the discriminator and fixing the discriminator to optimize the generator is repeated to improve the realism of the generated samples until the loss function converges. The prediction error is then optimized once based on the trained generator input vector into the GAN.

[0095] In this embodiment of the invention, step S100 further includes: adding the day before and the day after the holiday into the holiday, calculating the average load of weekdays and holidays, defining a holiday adjustment factor based on the average load of weekdays and holidays, and dynamically adjusting the error value based on the holiday adjustment factor.

[0096] Specifically, the day before and the day after the holiday are added to the holiday calculation, and the average load for weekdays and holidays is calculated:

[0097]

[0098]

[0099] Holidays include the day before and the day after the holiday. This represents the average workload on weekdays. This represents the average load during holidays. This indicates the total number of working days. This indicates the total number of holidays. This represents the actual load value;

[0100] Based on the average load relationship between weekdays and holidays, a holiday adjustment factor is defined. :

[0101]

[0102] in, >1 indicates that the workload is higher during holidays than during normal times. <1 indicates that the workload during holidays is lower than during normal times;

[0103] The error value is dynamically adjusted based on the holiday adjustment factor.

[0104]

[0105] in, This represents the initial prediction error. Indicates the holiday adjustment factor. This indicates a marker; if it's a holiday, then... It is 1 if it is true, otherwise it is 0.

[0106] Furthermore, the adjusted initial prediction error is combined with the first-order optimized prediction error to update the error dataset, as shown below:

[0107]

[0108] Furthermore, based on the features extracted from the historical load data as input to the BP neural network model, a BP neural network model is constructed, and the loss function of the BP neural network model is defined.

[0109] Specifically, features from historical load data are extracted as input features for the BP neural network model. Multiple hidden layers are added, each using the ReLU non-linear activation function, and the mean squared error (MSE) is used as the loss function to construct the BP neural network model.

[0110]

[0111] in, The weight matrix of the BP neural network is updated using the backpropagation algorithm. This represents the input features of the BP neural network model. This represents the activation function of the output layer. This represents the bias term of the BP neural network model;

[0112] Define the loss function as follows:

[0113]

[0114] in, Indicates the loss value. Indicates the number of samples. This represents the error correction value predicted by the BP neural network model.

[0115] The gradient is calculated and the weights of the BP neural network model are updated through backpropagation until the loss function converges. The updated error dataset is then input into the trained BP neural network model to obtain the error correction value predicted by the BP neural network model, and the final ultra-short-term load prediction is calculated.

[0116] Specifically, the gradient is calculated and the weights of the BP neural network model are updated through backpropagation:

[0117]

[0118] in, Indicates the weight update amount. Indicates the learning rate. This represents the gradient of the loss function with respect to the weights.

[0119] The forward and backward propagation operations are iterated until the loss function converges. During the iteration, a sliding window method is used to obtain the latest load data from the real-time monitoring system, update the error dataset, train the GAN model to output the latest error samples, update the error dataset again, and input the updated error dataset into the trained BP neural network model to obtain the result. And calculate the final ultra-short-term load forecast. :

[0120]

[0121] in, This represents the ultra-short-term load forecast after error correction. This represents the initial ultra-short-term load forecast. This indicates the error correction for the predictions made by the BP neural network model.

[0122] It should be noted that by optimizing the prediction error using a Generative Adversarial Network (GAN) model, the initial prediction error is combined with random noise to generate an error dataset. High-quality error samples are generated through alternating optimization of the generator and discriminator. A backpropagation (BP) neural network is used to correct the error, effectively solving the problem of insufficient handling of nonlinear dynamic characteristics by traditional linear error correction methods. This significantly improves the reliability and accuracy of the prediction results. Based on the difference between the average load on holidays and weekdays, a holiday adjustment factor is dynamically set to correct the error value, more accurately reflecting the load characteristics under special time nodes and adapting to scenarios with drastic changes in load characteristics, thus significantly improving the adaptability of the prediction. By continuously acquiring real-time monitoring data using the sliding window method and dynamically updating the error dataset, and through iterative training of the GAN model and BP neural network, the latest error samples and corrected ultra-short-term load prediction values ​​are output in real time, ensuring that the prediction model can quickly respond to real-time load changes and improving the real-time performance and timeliness of the prediction.

[0123] In this embodiment of the invention, step S200, which involves calculating key load indicators of the distribution network based on real-time monitoring data, assessing the safety status of these key load indicators, and generating preliminary safety situation assessment results, includes:

[0124] Collect key load indicators, including load rate, voltage deviation rate, voltage stability index, overload prediction index, and equipment health assessment index, and process the collected key load indicator data.

[0125] Specifically, the collected key load indicator data are standardized and normalized.

[0126] The load factor is calculated as the ratio of the current load value of the equipment to its maximum load capacity.

[0127]

[0128] in, Indicates load rate, Indicates the current time The load value, The equipment's maximum load capacity;

[0129] The voltage deviation rate is calculated as follows:

[0130]

[0131] in, Indicates voltage deviation rate. This is represented as the voltage value at the current moment. This is the standard voltage value for the power distribution network;

[0132] By introducing a graph neural network (GCN) model and combining it with the topology information of the distribution network, a voltage stability index is generated. The topology information of the distribution network refers to modeling the distribution network as an undirected weighted graph based on real-time monitoring data.

[0133]

[0134] in, This represents the set of nodes, containing all power grid nodes. This represents the set of edges, which symbolize the power line connections between nodes. This represents an adjacency matrix, describing the connection relationships and weights between nodes;

[0135] Adjacency Matrix The construction rules are expressed as follows: if the node and nodes If there are power lines, then If the value is greater than 0, the weight can be calculated using the line admittance and line impedance.

[0136]

[0137] in, This indicates whether a path exists between nodes. The impedance of the line between nodes. For resistance, For reactance;

[0138] By constructing a node feature matrix using a distribution network model, the node characteristics are output, and the voltage stability score of each node is further calculated. as follows:

[0139]

[0140] in, The number of nodes in the distribution network. Indicates the first Nodes Voltage at time, Indicates the first Nodes Active power at any given time Indicates the first Nodes Reactive power at any given moment Indicates the first Nodes Temperature at any moment;

[0141] Information is aggregated using graph convolution operations in a graph neural network (Graph GCN) model, and the features of each node are updated.

[0142]

[0143] in, It is the first The node feature matrix of a layer represents the node features of the current layer. Initially... = , The representation is a normalized adjacency matrix, describing the connection relationships between nodes. Indicates the first The layer's weight matrix is ​​used to perform linear transformations on the node features.

[0144] Furthermore, key load indicators are comprehensively evaluated to generate a comprehensive security situation assessment index; based on the key load indicators and the comprehensive security situation assessment index, a preliminary security situation assessment report is generated.

[0145] Specifically, voltage stability scores are calculated based on the distribution network topology information and the graph neural network (GCN) model. :

[0146]

[0147] in Indicates voltage stability score. This represents the final output node feature matrix. Features in the last layer of the feature matrix;

[0148] Overload prediction metrics refer to the prediction of future load based on load rate data using a Long Short-Term Memory (LSTM) network. By setting an overload risk threshold, it determines whether the equipment is in an overload risk state. The load prediction value is expressed as:

[0149]

[0150] in, Indicates the first Future moments of each node The load forecast value, Indicates the first Future moments of each node Historical load values, Indicates the first Future moments of each node eigenvector values;

[0151] Based on the set overload risk threshold Calculate the overload risk value:

[0152]

[0153] in Indicates device In time Overload risk value, This is the set overload risk threshold, typically 0.8 or 0.9. This indicates that the load rate exceeds the overload risk threshold. This indicates that the load rate exceeds the overload risk threshold but is less than the overload risk threshold. Different threshold ranges also represent different risk levels. Indicates low risk. , Indicates high risk;

[0154] The equipment health assessment indicators are calculated using the Random Forest (RF) model to determine the equipment health status. :

[0155]

[0156] in, Indicates the device at time health status, It is the number of trees in the random forest. Is the equipment in Data running in real time It is the first The prediction results of the decision tree on the health status of the equipment;

[0157] Based on the clustering algorithm in the anomaly detection algorithm and the set device health status threshold, the difference between the device health status score and the set device health status score is output. :

[0158]

[0159] in Indicates the device at time Health status score, This indicates the set device health status score. This indicates the difference between the current health status score of the device and the set health status score. The larger the difference, the worse the health status of the device.

[0160] Risk assessment based on equipment health status score: This indicates that the equipment is in normal health condition. This indicates that the equipment is slightly damaged. This indicates that the equipment's health condition is severely damaged. This indicates that the equipment's health status is at risk of failure.

[0161] By using a weighted fusion approach, key load indicators are comprehensively evaluated to generate a comprehensive security situation assessment index. :

[0162]

[0163] in, This represents the comprehensive security situation assessment index. Indicates the number of key indicators. Indicates the first A standardized key load metric Indicates the first The weights of key load indicators are dynamically adjusted based on the actual application needs of the distribution network.

[0164] Based on key load indicators and a comprehensive safety situation assessment index, a preliminary safety situation assessment report is generated. The report includes a list of equipment numbers, voltage stability analysis results, overload risk trend charts, equipment health status, and maintenance recommendations.

[0165] It should be noted that by collecting key indicators such as load factor and voltage deviation rate, the operating status of the distribution network can be dynamically monitored, the equipment load can be accurately assessed, and the operating pressure of the distribution network can be effectively perceived. Combined with the LSTM model, the future load status can be predicted, the overload risk can be assessed and graded warnings can be issued, which significantly improves the safety and foresight of the distribution network operation. The voltage deviation rate quantifies the degree of voltage anomaly, provides a basis for optimized control, improves the power supply quality and stability of the power grid, and uses a weighted fusion method to comprehensively evaluate the status of multiple indicators, generate a unified safety situation index, meet the needs of different operating scenarios, improve the accuracy and flexibility of situation assessment, and generate an assessment report that includes risk trends, equipment status and maintenance recommendations, providing comprehensive support for power grid optimization decisions and avoiding chain reactions caused by potential faults.

[0166] In this embodiment of the invention, step S300, which integrates the preliminary security situation assessment results and the ultra-short-term load forecast results to predict the final regional distribution network security situation, includes: integrating the data from the preliminary security situation assessment report and the ultra-short-term load forecast.

[0167] The fused data is input into the CNN-LSTM hybrid network. Spatial features are extracted through the convolutional operation of the generative adversarial network module of the CNN-LSTM hybrid network, and temporal features are extracted through the memory state update and hidden state update operations of the long short-term memory module.

[0168] Spatial and temporal features are extracted and concatenated into comprehensive features. These comprehensive features are then input into the fully connected layer of a CNN-LSTM hybrid network to output a regional distribution network security situation prediction. Based on historical load data and expert experience, risk thresholds are set, and the regional distribution network security situation prediction is used to distinguish the regional distribution network risk levels according to these risk thresholds. Finally, the regional distribution network security situation prediction is output.

[0169] Specifically, a weighted fusion method is used to merge the data into a unified dataset:

[0170]

[0171] in, It is the weighted and merged data. , This indicates the weighting coefficients set according to the actual situation;

[0172] Weighted and fused data is input into a CNN-LSTM hybrid network. Spatial features are extracted through the convolutional operations of the CNN module, and temporal features are extracted through the memory state update and hidden state update operations of the LSTM module. The extracted spatial and temporal features are concatenated into a comprehensive feature, which is then input into the fully connected layer of the CNN-LSTM hybrid network to output a regional distribution network security situation prediction value. Based on historical load data and expert experience, risk thresholds are set, and the regional distribution network security situation prediction value is classified into regional distribution network risk levels according to the risk thresholds. The final regional distribution network security situation prediction is then output. By combining real-time data and periodically updating the model data, the CNN-LSTM model training is re-triggered using the threshold as a trigger condition, continuously optimizing the final output regional distribution network security situation prediction result.

[0173] It should be noted that by using a weighted fusion method, multi-dimensional target data is unified into comprehensive features, taking into account the principles of safety, stability, economy, and responsiveness. By extracting spatial features through a CNN module and temporal features through an LSTM module, the spatiotemporal variation patterns of the distribution network are effectively integrated, significantly improving the predictive model's ability to model complex nonlinear data. It can comprehensively reflect the dynamic evolution characteristics of the distribution network. By combining historical load data and expert experience to set risk thresholds, different risk levels are distinguished based on predicted values, generating a dynamic security situation of the regional distribution network. This provides power grid operators with the ability to identify high-risk areas and time points in advance, significantly enhancing the foresight of control strategies.

[0174] In this embodiment of the invention, step S400, which optimizes the distribution network operating parameters based on the final load forecast results, includes:

[0175] Based on the periodic fluctuations of the distribution network load, the characteristics of rapid load changes, and the fluctuation period in the load data, a low-frequency threshold is set; based on the final prediction of the ultra-short-term load and the set low-frequency threshold, the ultra-short-term load prediction data is frequency decomposed to obtain the long-term load and short-term load.

[0176] Based on the scheduling system's operating parameters, the quantum particle swarm optimization algorithm is used to set the number of particles, initialize the parameters of each particle, set the objective function, and calculate the objective function value of each particle, including voltage deviation minimization, power flow optimization, and energy storage scheduling optimization; real-time state of charge data is collected from energy storage devices, and the energy storage scheduling coefficient is optimized.

[0177] The interfaces connecting electrical equipment in the distribution network are used as nodes in the voltage deviation minimization part, and the physical lines in the distribution network responsible for transmitting and distributing electrical energy are used as lines in the power flow optimization part; the rated values ​​of the distribution network operating parameters are used as constraints on the particle objective function value, and the particle fitness value is updated when the particle position violates the constraints.

[0178] The updated particle fitness value is compared with the historical particle fitness value, and the global and local optimal positions of the particles are updated according to the comparison results. The global and local optimal positions are taken as the global optimal solution and the local optimal solution, respectively. The positions of the particles and the objective function value are dynamically adjusted by updating the particle swarm fitness value until the fitness value converges. The optimized running parameters are then output and input into the scheduling system for scheduling.

[0179] Specifically, based on the periodic fluctuations of the distribution network load, the characteristics of rapid load changes, and the fluctuation period in the load data, a low-frequency threshold is set:

[0180]

[0181] in, This indicates the fluctuation period of the load data;

[0182] Based on the final prediction of ultra-short-term load and the set low-frequency threshold, the VMD variational mode decomposition method is used to decompose the ultra-short-term load prediction data into frequency to obtain long-term load and short-term load. The t-SNE algorithm is then used to denoise and remove redundant information from the long-term load and short-term load.

[0183] Based on the scheduling system's operating parameters, the quantum particle swarm optimization algorithm is used to set the number of particles, initialize the parameters of each particle, set the objective function, and calculate the objective function value of each particle, including voltage deviation minimization, power flow optimization, and energy storage scheduling optimization.

[0184] Based on collecting real-time state of charge data from energy storage devices and optimizing energy storage dispatch coefficients:

[0185]

[0186] in, Indicates the scheduling coefficient. Indicates the maximum power limit of the energy storage system. Indicates long-term load. Indicates the weighting factor. Indicates the time of energy storage devices Real-time state of charge, Indicates the target state of charge of the energy storage device;

[0187] The energy storage dispatch optimization part is represented as:

[0188]

[0189] in, This represents the objective function for energy storage scheduling. Indicates the time of the energy storage system The charging power, Indicates the scheduling coefficient. Indicates the time step;

[0190] The interfaces connecting electrical equipment in the distribution network are regarded as nodes in the voltage deviation minimization part, and the physical lines in the distribution network responsible for transmitting and distributing electrical energy are regarded as lines in the power flow optimization part.

[0191] The portion of the voltage deviation that is minimized is represented as:

[0192]

[0193] in, This represents the objective function for minimizing voltage deviation. Indicates the total number of electrical equipment nodes. Represents the voltage at the node. Indicates the node reference voltage;

[0194] The power flow optimization part is represented as:

[0195]

[0196] in, Describe the objective function of power flow. Represents the set of all lines. Indicates the line The actual power flow, Indicates the line Maximum permissible power flow;

[0197] Define the fitness value for each particle. express:

[0198]

[0199] in, This represents the value of the function that minimizes voltage deviation. This represents the power flow optimization function value;

[0200] The rated values ​​of the distribution network operating parameters are used as constraints on the particle objective function value. When a particle position violates the constraints, a penalty term is triggered, and the particle fitness value is updated. :

[0201] Penalty items for:

[0202]

[0203]

[0204] in, This represents the updated particle fitness value. Indicates the penalty coefficient. Indicates the current particle, Represents particles The One variable, and Indicates the upper and lower limits of the current particle variable. Indicates a penalty marker. Violating constraints triggers penalties. No penalty is triggered;

[0205] The updated particle fitness value is compared with the historical particle fitness value, and the global and local optimal positions of the particle are updated based on the comparison results. The global and local optimal positions are then used as the global optimal solution and the local optimal solution, respectively, to update the particle position. The update formula is as follows:

[0206] Updated reference points:

[0207]

[0208] Update location:

[0209]

[0210] in, Indicates the first Each particle reference point This indicates the location of the global optimal solution. Indicates the location of the local optimum. Represents the weighting factor, ranging from [0, 1]. Indicates the first Particles in time Location, Indicates the contractility factor. This represents a random number generated by a random function, ranging from [0, 1]. Indicates time Down and The distance;

[0211] By updating the particle swarm fitness value, the position of the particles and the objective function value are dynamically adjusted until the fitness value converges. The optimized running parameters are then output and input into the scheduling system for scheduling.

[0212] Operating parameters include node voltage, line power flow, and energy storage power dispatch.

[0213] By leveraging the periodic fluctuation characteristics of distribution network loads, a low-frequency threshold is set, and the VMD variational mode decomposition method is used to decompose the load data into two parts: long-term trend and short-term fluctuation. This accurately extracts the dynamic characteristics of the load, making distribution network operation planning and real-time scheduling more reliable and accurate. The t-SNE algorithm is applied to the long-term load and short-term load respectively for noise reduction and to remove redundant information, which significantly optimizes the quality of data input, improves the efficiency and accuracy of the prediction model, and ensures that the model focuses on key load characteristics.

[0214] By employing the quantum particle swarm optimization algorithm, multi-objective optimization of voltage deviation, power flow, and energy storage scheduling is performed, quickly finding the global optimal solution that meets the multi-objective requirements. This significantly improves the adaptability and reliability of the distribution network scheduling system. Based on the dynamic difference between the real-time state of charge (SOC) and the target SOC of energy storage devices, the scheduling coefficient is adjusted to enhance the response capability of energy storage devices to load fluctuations, improve scheduling efficiency, and optimize the stability and economy of distribution network operation. Combining the constraints of node voltage and line power flow, the optimization algorithm iteratively minimizes the deviation, ensuring that system parameters operate within a reasonable range, reducing the risk of line overload and voltage instability, reducing losses, and improving the safety and stability of system operation.

[0215] By penalizing the positions of particles that violate constraints and dynamically updating the fitness values ​​of particles, the optimization process is ensured to strictly comply with the physical constraints of the distribution network, thereby improving the practical feasibility of the optimization results. After convergence of the quantum particle swarm optimization algorithm, the optimized node voltage, line power flow, and energy storage scheduling parameters are output, providing accurate operating parameters that meet the requirements of multi-objective optimization and providing reliable operating guidance for the scheduling system.

[0216] In this embodiment of the invention, the historical load data, real-time monitoring data, and regional distribution network security status prediction are also transmitted to a database via encryption for storage and backup.

[0217] Specifically, based on distributed data storage technology and end-to-end encrypted transmission mechanism, data encryption during transmission is achieved through the TLS protocol. The AES-256 algorithm is used to encrypt historical load data, real-time monitoring data, and regional distribution network security situation prediction. Blockchain technology is used to record data storage and backup operation logs. The scalability and reliability of data storage are improved through the distributed storage system HDFS. At the same time, an off-site backup mechanism is enabled to regularly perform multi-node storage and verification of data.

[0218] The system ensures data transmission security through the TLS protocol, enhances data storage security and integrity through the AES-256 algorithm, improves the scalability and reliability of the storage system through the distributed storage system HDFS, and enhances data backup and recovery capabilities by enabling off-site backup mechanisms.

[0219] Example 2: The above example is an illustrative scheme of a load forecasting method based on distribution network status. It should be noted that the technical solution of this load forecasting system based on distribution network status belongs to the same concept as the technical solution of the load forecasting method based on distribution network status described above. Details not described in detail in the technical solution of the load forecasting system based on distribution network status in this example can be found in the description of the technical solution of the load forecasting method based on distribution network status described above.

[0220] This embodiment provides a load forecasting system based on distribution network conditions, comprising:

[0221] The data acquisition module is used to collect historical load data and real-time monitoring data of the distribution network for ultra-short-term load forecasting.

[0222] The security assessment module is used to calculate key load indicators of the distribution network based on real-time monitoring data, assess the security status of key load indicators, and generate preliminary security situation assessment results.

[0223] The situation prediction module is used to integrate the preliminary security situation assessment results with the ultra-short-term load forecast results to predict the final regional distribution network security situation.

[0224] The optimization module is used to optimize the operating parameters of the distribution network based on the final load forecast results.

[0225] This embodiment also provides an electronic device applicable to load forecasting methods based on distribution network conditions, including:

[0226] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the load forecasting method based on the distribution network situation as proposed in the above embodiments.

[0227] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the load forecasting method based on the distribution network status as proposed in the above embodiments.

[0228] The storage medium proposed in this embodiment and the load forecasting method based on the distribution network situation 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.

[0229] 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.

[0230] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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 load forecasting method based on power distribution network situation, characterized in that, The method comprises the following steps: Collecting historical load data and real-time monitoring data of the power distribution network to perform ultra-short-term load forecasting; Based on the real-time monitoring data, the key load indicators of the power distribution network are calculated, and the safety state of the key load indicators is evaluated to generate a preliminary safety situation assessment result; Fusing the preliminary safety situation assessment result and the ultra-short-term load forecasting result to predict the final regional power distribution network safety situation; Optimizing the operation parameters of the power distribution network according to the final load forecasting result.

2. The power distribution grid posture based load forecasting method as claimed in claim 1, wherein, The method of collecting historical load data and real-time monitoring data of the power distribution network to perform ultra-short-term load forecasting comprises the following steps: Extracting periodic features in historical load data, identifying daily fluctuation patterns and seasonal trends of load, and extracting long-term trends in data as input features of the ultra-short-term load forecasting model; The periodic features include daily periodic load demand fluctuations, weekly periodic load demand fluctuations, and annual periodic load demand fluctuations; The objective function of the ultra-short-term load forecasting model is characterized as the error between the predicted load value and the actual load value, and the model parameters are iteratively trained; Based on the trained ultra-short-term load forecasting model, the real-time input features extracted from the real-time monitoring data are combined to predict the ultra-short-term load of the power distribution network.

3. The power distribution grid posture based load forecasting method as claimed in claim 2, wherein, Further comprising: Based on the real load value obtained from the real-time monitoring data and the ultra-short-term load forecasting value, the error is calculated to obtain the difference between the ultra-short-term load forecasting value and the actual load value, and the prediction error is corrected; An error data set is constructed based on the initial prediction error, and a generator is constructed based on the initial prediction error and random noise to form an input vector of the generator; The error data set is input into the discriminator as input data, and the probability that the sample is a real error sample is calculated; The training strategy of the discriminator and the generator is optimized until the loss function converges, and the one-time optimized prediction error is output based on the input vector of the trained generator.

4. The power distribution grid posture based load forecasting method as claimed in claim 3, wherein, Further comprising: Adding the day before and the day after the holiday into the holiday, calculating the load average of weekdays and holidays, defining a holiday adjustment factor based on the relationship between the load average of weekdays and holidays, and dynamically adjusting the error value based on the holiday adjustment factor; The updated initial prediction error and the one-time optimized prediction error are combined to update the error data set; Based on the extracted features in the historical load data as the input of the BP neural network model, a BP neural network model is constructed, and the loss function of the BP neural network model is defined; The gradient is calculated by back propagation and the BP neural network model weight is updated until the loss function converges, the updated error data set is input into the trained BP neural network model to obtain the error correction value predicted by the BP neural network model, and the final ultra-short-term load forecasting is calculated.

5. The power distribution grid posture based load forecasting method as claimed in claim 4, wherein, The method of calculating the key load indicators of the power distribution network based on the real-time monitoring data and evaluating the safety state of the key load indicators to generate a preliminary safety situation assessment result comprises the following steps: The key load indicators include a load rate, a voltage deviation rate, a voltage stability index, an overload prediction indicator, and a device health evaluation indicator, and the collected key load indicator data is processed; The key load indicators are comprehensively evaluated to generate a comprehensive security situation evaluation index; and based on the key load indicators and the comprehensive security situation evaluation index, a preliminary security situation evaluation report is generated.

6. The power distribution grid posture based load forecasting method as claimed in claim 5, wherein, The preliminary security situation evaluation result and the ultra-short-term load prediction result are fused to predict a final regional power distribution network security situation, including fusing data in the preliminary security situation evaluation report and the ultra-short-term load prediction; The fused data is input into a CNN-LSTM hybrid network, and a convolution operation of an adversarial network module of the CNN-LSTM hybrid network extracts spatial features, and a memory state update and a hidden state update operation of a long short-term memory module extract time features; The spatial features and the time features are extracted and spliced into comprehensive features, the comprehensive features are input into a full connection layer of the CNN-LSTM hybrid network, a regional power distribution network security situation prediction value is output, a risk threshold value is set according to historical load data and expert experience, the regional power distribution network security situation prediction value is divided into regional power distribution network risk levels according to the risk threshold value, and a final regional power distribution network security situation prediction is output.

7. The power distribution grid posture based load forecasting method as claimed in claim 6, wherein, According to the final load prediction result, the power distribution network operation parameters are optimized, including: According to the periodic fluctuation of the power distribution network load, the characteristics of the rapid change of the load, and the fluctuation period in the load data, a low-frequency threshold value is set; based on the ultra-short-term load final prediction and the set low-frequency threshold value, the ultra-short-term load prediction data is frequency-decomposed to obtain long-term load and short-term load; Based on the dispatching system operation parameters, the number of particle swarms is set by a quantum particle swarm optimization algorithm, each particle parameter is initialized, a target function is set and the target function value of each particle is calculated, including voltage deviation minimization, power flow optimization, and energy storage scheduling optimization; real-time state of charge data of the energy storage device is collected and the energy storage scheduling coefficient is optimized; The interface connected by the electrical equipment in the power distribution network is taken as a node in the voltage deviation minimization part, and the physical line responsible for transmitting and distributing electric energy in the power distribution network is taken as a line in the power flow optimization part; the rated value of the power distribution network operation parameter is taken as a constraint condition of the particle target function value, and the particle position violating the constraint condition triggers a penalty term, and the particle fitness value is updated; The updated particle fitness value and the particle historical fitness value are compared, and the global and local optimal positions are updated according to the comparison result, and the global and local optimal positions are taken as the global optimal solution and the local optimal solution; the particle position and the target function value are dynamically adjusted through the particle swarm fitness value update until the fitness value converges, and the optimized operation parameters are output; the optimized operation parameters are input into the dispatching system for dispatching.

8. A load forecasting system based on power distribution network situation, applied to the method of any one of claims 1-7, characterized in that, The system includes: a data acquisition module, configured to collect historical load data and real-time monitoring data of the power distribution network, and perform ultra-short-term load prediction; a security evaluation module, configured to calculate key load indicators of the power distribution network based on the real-time monitoring data, evaluate security states of the key load indicators, and generate a preliminary security situation evaluation result; and a prediction module, configured to fuse the preliminary security situation evaluation result and an ultra-short-term load prediction result to predict a final regional power distribution network security situation. a situation prediction module, configured to fuse the preliminary security situation assessment result and the ultra-short-term load prediction result to predict a final security situation of the regional power distribution network; an optimization module, configured to optimize power distribution network operation parameters according to the final load prediction result. 9.An electronic device, comprising: a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, which, when executed by the processor, implement the steps of the load prediction method based on the situation of the power distribution network according to any one of claims 1 to 7. 10.A computer readable storage medium storing computer executable instructions, which, when executed by a processor, implement the steps of the load prediction method based on the situation of the power distribution network according to any one of claims 1 to 7.

Images