Distribution method based on AI artificial intelligence and intelligent power distribution cabinet

Abstract

This application relates to the field of data processing technology, and discloses a power distribution method and an intelligent power distribution cabinet based on AI artificial intelligence. The method includes: real-time data collection of key nodes of the power distribution system through an intelligent sensor network to obtain the original power distribution system operation data, and performing cleaning and standardization processing on the data to obtain a standardized data set; analyzing and modeling the standardized data set through multi-level hybrid model training to obtain an AI analysis model of the power distribution system; processing and analyzing the real-time collected power distribution data to obtain a load prediction result and a dynamic balance strategy; performing abnormal pattern recognition and multi-path reasoning on the system operation parameters to obtain an equipment health status assessment and maintenance suggestions; based on the equipment health status assessment and maintenance suggestions, performing multi-objective optimization calculation on the renewable energy supply and electricity demand to obtain an intelligent energy scheduling scheme. This application improves the efficiency and accuracy of the power distribution method based on AI artificial intelligence.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to a power distribution method and an intelligent power distribution cabinet based on AI artificial intelligence. Background Art

[0002] With the rapid development and intelligent transformation of the power system, traditional power distribution methods have been difficult to meet the complex requirements of modern power grids. Currently, many power distribution systems adopt analysis methods based on historical data and static models, and maintain system stability through regular inspections and manual interventions. These methods have improved the reliability and efficiency of the power distribution system to a certain extent, but have obvious limitations in dealing with large-scale, real-time, and multi-variable data. At the same time, some advanced power distribution systems have begun to introduce artificial intelligence technologies, such as machine learning algorithms and neural networks, for load prediction and fault diagnosis, and have achieved certain results.

[0003] However, the existing power distribution methods still face many challenges. First, the real-time performance of data collection and processing is insufficient, resulting in a slow response speed of the system to emergencies. Second, traditional single models or simple AI models are difficult to comprehensively capture the complex dynamic characteristics of the power distribution system, especially in integrating renewable energy and dealing with uncertainties. In addition, existing anomaly detection and fault diagnosis methods are often passive and lagging, and it is difficult to achieve proactive prevention and timely intervention. Finally, energy dispatch optimization usually only considers a single goal, such as economy or reliability, and lacks the ability to comprehensively balance multiple goals. Summary of the Invention

[0004] This application provides a power distribution method and an intelligent power distribution cabinet based on AI artificial intelligence, for the efficiency and accuracy of the power distribution method based on AI artificial intelligence.

[0005] In a first aspect, this application provides a power distribution method based on AI artificial intelligence. The power distribution method based on AI artificial intelligence includes: real-time data collection of key nodes of the power distribution system through an intelligent sensor network to obtain original power distribution system operation data; cleaning and standardizing the original power distribution system operation data through a data preprocessing algorithm to obtain a standardized data set; analyzing and modeling the standardized data set through multi-level hybrid model training to obtain an AI analysis model of the power distribution system; processing and analyzing the power distribution data collected in real time through the AI analysis model to obtain a load prediction result and a dynamic balance strategy; based on the load prediction result and the dynamic balance strategy, performing abnormal pattern recognition and multi-path reasoning on the system operation parameters to obtain an equipment health status assessment and maintenance suggestions; based on the equipment health status assessment and maintenance suggestions, performing multi-objective optimization calculation on renewable energy supply and power consumption demand to obtain an intelligent energy dispatch plan.

[0006] In a second aspect, the present application provides an intelligent power distribution cabinet, which includes:

[0007] An acquisition module, configured to perform real-time data acquisition on key nodes of the power distribution system through an intelligent sensor network to obtain original power distribution system operation data;

[0008] A processing module, configured to perform cleaning and standardization processing on the original power distribution system operation data through a data preprocessing algorithm to obtain a standardized data set;

[0009] A modeling module, configured to perform analysis and modeling on the standardized data set through multi-level hybrid model training to obtain an AI analysis model of the power distribution system;

[0010] An analysis module, configured to perform processing and analysis on real-time acquired power distribution data through the AI analysis model to obtain a load prediction result and a dynamic balance strategy;

[0011] An identification module, configured to perform abnormal mode identification and multi-path reasoning on system operation parameters based on the load prediction result and the dynamic balance strategy to obtain an equipment health status assessment and maintenance suggestions;

[0012] A calculation module, configured to perform multi-objective optimization calculation on renewable energy supply and power consumption demand based on the equipment health status assessment and maintenance suggestions to obtain an intelligent energy scheduling scheme.

[0013] In the technical solution provided by this application, real-time data collection of key nodes in the power distribution system is carried out through an intelligent sensor network, which can comprehensively and timely obtain the system operation status, providing a high-quality and high-timeliness raw data basis for subsequent analysis, and significantly improving the comprehensiveness and timeliness of data collection. The data preprocessing algorithm is used to clean and standardize the original power distribution system operation data, which not only effectively removes the noise and outliers in the data, but also realizes the unified representation of the data, laying a solid data foundation for subsequent model training and analysis, and greatly improving the data quality and usability. The introduction of multi-level hybrid model training enables the system to simultaneously capture the time series characteristics, spatial characteristics and decision-making characteristics of power distribution data, significantly enhancing the expression ability and generalization performance of the model, and being able to more accurately depict the complex dynamic characteristics of the power distribution system. By processing and analyzing the real-time collected power distribution data through the AI analysis model, the system can quickly and accurately generate load prediction results and dynamic balance strategies, greatly improving the prediction accuracy and response speed of the power distribution system, and providing reliable support for real-time scheduling and decision-making. Based on the load prediction results and dynamic balance strategies, abnormal pattern recognition and multi-path reasoning are carried out on the system operation parameters, which can not only timely detect potential equipment failures and system anomalies, but also give accurate equipment health status assessments and maintenance suggestions, significantly improving the reliability and maintenance efficiency of the system. Finally, through multi-objective optimization calculation of renewable energy supply and electricity demand, the obtained intelligent energy scheduling scheme can maximize the utilization rate of renewable energy while ensuring power supply reliability, and achieve the balance of economy and environmental protection, greatly improving the overall operation efficiency and sustainability of the power distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0015] Figure 1 It is a schematic diagram of an embodiment of the power distribution method based on AI artificial intelligence in the embodiment of this application;

[0016] Figure 2 It is a schematic diagram of an embodiment of the intelligent power distribution cabinet in the embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The embodiments of the present application provide a power distribution method and an intelligent power distribution cabinet based on AI artificial intelligence. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 , an embodiment of the power distribution method based on AI artificial intelligence in the embodiments of the present application includes:

[0019] Step S101: Real-time data collection is performed on key nodes of the power distribution system through an intelligent sensor network to obtain original power distribution system operation data;

[0020] Step S102: The original power distribution system operation data is cleaned and standardized through a data preprocessing algorithm to obtain a standardized data set;

[0021] Step S103: The standardized data set is analyzed and modeled through multi-level hybrid model training to obtain an AI analysis model of the power distribution system;

[0022] Step S104: The power distribution data collected in real time is processed and analyzed through the AI analysis model to obtain a load prediction result and a dynamic balance strategy;

[0023] Step S105: Based on the load prediction result and the dynamic balance strategy, abnormal mode recognition and multi-path reasoning are performed on the system operation parameters to obtain an evaluation of the device health status and maintenance suggestions;

[0024] Step S106: Based on the evaluation of the device health status and maintenance suggestions, multi-objective optimization calculations are performed on the renewable energy supply and power consumption demand to obtain an intelligent energy scheduling plan.

[0025] It can be understood that the execution subject of the present application can be an intelligent power distribution cabinet, or a terminal or a server. Specifically, it is not limited here. The embodiments of the present application will be described by taking the server as the execution subject as an example.

[0026] Specifically, real-time data collection is carried out on the key nodes of the distribution system through an intelligent sensor network. These intelligent sensors are arranged at positions such as substations, distribution cabinets, and user ends, collecting parameters such as voltage, current, power factor, harmonic content, etc. The collection frequency can reach the millisecond level to ensure the real-time nature of the data. The collected raw data is cleaned and standardized through a data preprocessing algorithm. During the cleaning process, a median filtering algorithm is used to remove short-term fluctuations and outliers, and interpolation is performed on missing data. The standardization process includes Z-score standardization and one-hot encoding, enabling unified processing of different types of data. The preprocessed data forms a standardized dataset, providing a basis for subsequent modeling. The standardized dataset is analyzed and modeled through multi-level hybrid model training. This process includes constructing various structures such as deep neural networks, long short-term memory networks, attention mechanisms, and convolutional neural networks. Deep neural networks are used to capture non-linear relationships, long short-term memory networks process time-series data, attention mechanisms focus on important features, and convolutional neural networks analyze spatial features. Through iterative training and parameter optimization, these network structures form an AI analysis model that can comprehensively analyze the characteristics of the distribution system.

[0027] The AI analysis model processes and analyzes the real-time collected distribution data to generate load prediction results and dynamic balancing strategies. Load prediction includes short-term (15 minutes to 1 hour), medium-term (1 day to 1 week), and long-term (1 month to 1 year) predictions. During the prediction process, various factors such as historical data patterns, weather forecasts, and social events are considered. The dynamic balancing strategy is based on the load prediction results and automatically adjusts distribution parameters such as transformer tap positions and capacitor bank switching states through reinforcement learning algorithms to optimize power quality and reduce line losses. Based on the load prediction results and dynamic balancing strategies, abnormal pattern recognition and multi-path reasoning are performed on the system operation parameters. Abnormal pattern recognition identifies potential abnormal situations by comparing real-time parameters with historical normal operation data. Multi-path reasoning uses a pre-constructed knowledge graph and combines abnormal patterns to infer possible fault causes. Output device health status assessments and maintenance recommendations, providing a basis for preventive maintenance.

[0028] Finally, based on the device health status assessment and maintenance recommendations, multi-objective optimization calculations are performed on renewable energy supply and electricity demand. This step considers multiple factors such as renewable energy generation forecasts, user electricity demand forecasts, and device health status, and generates an intelligent energy scheduling plan through a multi-objective optimization algorithm. This plan maximizes the utilization rate of renewable energy while meeting electricity demand, and balances multiple objectives such as economy, reliability, and environmental protection.

[0029] For example, in a distribution system in a certain area on a working day in spring, the intelligent sensor network collects data every minute from 8:00 to 9:00 in the morning. In the original data, the voltage of a certain distribution cabinet at 8:30 is 220.5V, the current is 100.2A, and the power factor is 0.92. After data preprocessing, the voltage value is normalized to 0.025 (assuming a mean of 220V and a standard deviation of 20V), and the current value is normalized to -0.01 (assuming a mean of 101A and a standard deviation of 80A). After the AI analysis model processes this data, it predicts that the peak load from 12:00 to 13:00 on that day is 5.2MW, an 8% increase compared to the same period of the previous week. Based on this prediction, the dynamic balancing strategy recommends putting a group of nearby capacitors into use at 11:30 to cope with the upcoming load peak. At the same time, the system detects that the temperature of a transformer is 2°C higher than usual. Through multi-path reasoning, it is inferred that the efficiency of the cooling system may have decreased, and it is recommended to focus on inspection during the routine maintenance next week. Finally, considering the sunny weather on that day and the good expectation of photovoltaic power generation, the intelligent scheduling plan recommends transferring some loads to the photovoltaic power generation system from 13:00 to 15:00 to reduce the use of conventional power sources, and it is expected to reduce carbon emissions by about 2 tons. This series of decisions and operations demonstrate how the AI-based distribution method can achieve comprehensive, real-time, and intelligent power grid management.

[0030] In the embodiments of the present application, by using an intelligent sensor network to collect real-time data of key nodes in the power distribution system, the operating state of the system can be comprehensively and timely obtained, providing a high-quality and highly time-sensitive raw data basis for subsequent analysis, and significantly improving the comprehensiveness and real-time nature of data collection. The use of data preprocessing algorithms to clean and standardize the original operating data of the power distribution system not only effectively removes noise and outliers in the data, but also achieves a unified representation of the data, laying a solid data foundation for subsequent model training and analysis, and greatly improving the data quality and usability. The introduction of multi-level hybrid model training enables the system to simultaneously capture the temporal, spatial, and decision-making characteristics of power distribution data, significantly enhancing the model's expressive ability and generalization performance, and being able to more accurately depict the complex dynamic characteristics of the power distribution system. By using an AI analysis model to process and analyze the real-time collected power distribution data, the system can quickly and accurately generate load prediction results and dynamic balance strategies, greatly improving the prediction accuracy and response speed of the power distribution system, and providing reliable support for real-time scheduling and decision-making. Based on the load prediction results and dynamic balance strategies, anomaly pattern recognition and multi-path reasoning are performed on the system operating parameters, which can not only timely detect potential equipment failures and system anomalies, but also give accurate equipment health status assessments and maintenance recommendations, significantly improving the reliability and maintenance efficiency of the system. Finally, through multi-objective optimization calculations of renewable energy supply and electricity demand, the obtained intelligent energy scheduling plan can maximize the utilization rate of renewable energy while ensuring power supply reliability, and achieve a balance between economy and environmental protection, greatly improving the overall operating efficiency and sustainability of the power distribution system.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] (1) Arrange intelligent sensors for the substations, distribution cabinets, and user terminals of the power distribution system to obtain a multi-level data collection network, and configure communication protocols for the multi-level data collection network to obtain a high-speed real-time data transmission channel;

[0033] (2) Continuously monitor voltage, current, power factor, and harmonic content through the high-speed real-time data transmission channel to obtain multi-dimensional power parameter data, and mark time stamps for the multi-dimensional power parameter data to obtain a time-series associated original data stream;

[0034] (3) Check the data integrity of the time-series associated original data stream to obtain a valid data set, and detect outliers in the valid data set to obtain a preliminarily cleaned data set;

[0035] (4) Smooth the preliminarily cleaned data set through the sliding window algorithm to obtain a denoised data sequence, and perform missing value imputation on the denoised data sequence to obtain a complete data sequence;

[0036] (5) Perform standardization and normalization processing on the complete data sequence to obtain a standard data set with unified dimensions, and perform feature extraction on the standard data set with unified dimensions to obtain a feature vector after dimensionality reduction;

[0037] (6) Identify key features of the feature vector after dimensionality reduction through the principal component analysis algorithm to obtain a core feature set, and perform correlation analysis on the core feature set to obtain a feature correlation matrix;

[0038] (7) Construct a multi-dimensional data cube based on the feature correlation matrix to obtain a multi-angle data analysis view, and perform data clustering on the multi-angle data analysis view to obtain a data distribution pattern.

[0039] Specifically, intelligent sensors are arranged on the substations, distribution cabinets and user terminals of the power distribution system to form a multi-level data acquisition network. These intelligent sensors include voltage sensors, current sensors, power factor measurement devices, harmonic analyzers, etc. By configuring high-speed communication protocols such as 5G or industrial Ethernet, a high-speed real-time data transmission channel is established to ensure the timeliness and integrity of data. This multi-level network structure can comprehensively cover all key nodes of the power distribution system and provide comprehensive and detailed system operation status information. Using these high-speed transmission channels, continuously monitor voltage, current, power factor and harmonic content. The monitoring frequency can reach the millisecond level, forming multi-dimensional power parameter data. Each piece of data is attached with an accurate timestamp to generate a time-sequentially associated original data stream. This time-sequential association ensures the time consistency of the data and provides an important basis for subsequent analysis. The accurate recording of timestamps is crucial for analyzing system dynamic changes and identifying instantaneous anomalies.

[0040] Perform data integrity checks on the original time-series associated data stream, including checking the integrity of data packets, the continuity of timestamps, etc., to obtain a valid data set. Then, perform outlier detection on the valid data set, using statistical methods or machine learning algorithms to identify and mark outlier data points, obtaining a preliminarily cleaned data set. This step can effectively filter out invalid or incorrect data caused by factors such as sensor failures and communication interference. Smooth the preliminarily cleaned data set through a sliding window algorithm. The core idea of the sliding window algorithm is to perform average or median filtering on the data within a fixed-size window, and the window slides along the time axis to achieve smoothing of the entire data sequence. This process effectively removes short-term fluctuations and noise, obtaining a denoised data sequence. For the missing values in the denoised data sequence, interpolation methods such as linear interpolation or spline interpolation are used to complete them, obtaining a complete data sequence. The selection of the sliding window size has an important impact on the results and needs to be adjusted according to data characteristics and application requirements.

[0041] Perform standardization and normalization processing on the complete data sequence to unify data with different dimensions to the same scale. Standardization processing usually uses the Z-score method to convert the data into a distribution with a mean of 0 and a standard deviation of 1. Normalization processing commonly uses the Min-Max method to map the data to the [0,1] interval. After such processing, a standard data set with unified dimensions is obtained, which is convenient for subsequent feature extraction and model training. Then, perform feature extraction on the standard data set, such as extracting time-domain features, frequency-domain features, etc., to obtain a feature vector with reduced dimensions. Identify key features for the feature vector with reduced dimensions through the principal component analysis (PCA) algorithm. The core of PCA is to calculate the eigenvalues and eigenvectors of the covariance matrix, and select the eigenvectors corresponding to the largest several eigenvalues as the principal components. This step can effectively reduce the data dimension while retaining the main information of the data. Select the eigenvectors corresponding to the first k largest eigenvalues to form a core feature set. Perform correlation analysis on the core feature set, calculate the correlation coefficient matrix between features, and obtain a feature association matrix. This matrix reflects the mutual relationship between different features and is of great significance for understanding system behavior and optimizing model structure. Construct a multi-dimensional data cube based on the feature association matrix, with each dimension representing a core feature. This multi-dimensional structure allows data to be analyzed from different perspectives, forming a multi-angle data analysis view. The multi-dimensional data cube facilitates flexible query and in-depth analysis of data. Finally, perform data clustering on the multi-angle data analysis view, such as using the K-means algorithm, to group similar data points into one category, obtaining a data distribution pattern. Such clustering results can help identify the typical operating states and abnormal patterns of the system.

[0042] For example, a certain power distribution system collects data every 5 minutes in a day, and a total of 288 groups of data are collected. Each group of data contains the voltage, current, power factor, and harmonic content of 10 measurement points. In the original data, the voltage at a certain measurement point at 8:00 is 220.5V, the current is 100.2A, the power factor is 0.92, and the total harmonic distortion is 2.5%. Data integrity check finds that 2 groups of data are missing, and outlier detection identifies 5 abnormal points. Smoothing is performed using a 30-minute (6 data points) sliding window, and the abnormal 220.5V is smoothed to 219.8V. After normalization, this voltage value becomes 0.025 (assuming a mean of 220V and a standard deviation of 20V). After PCA analysis, 5 principal components that explain 95% of the variance are retained. K-means clustering (k = 3) divides the data into three categories: low load, medium load, and high load, where the low load accounts for 25%, the medium load accounts for 60%, and the high load accounts for 15%.

[0043] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0044] (1) Align the time series of the original power distribution system operation data to obtain a synchronized data stream, and perform data integrity check on the synchronized data stream to obtain a valid data set;

[0045] (2) Perform outlier detection on the valid data set to obtain a preliminarily cleaned data set, and remove noise from the preliminarily cleaned data set through a median filtering algorithm to obtain a smoothed data sequence;

[0046] (3) Perform missing value imputation on the smoothed data sequence to obtain a complete data sequence, and remove duplicate values from the complete data sequence to obtain a refined data set;

[0047] (4) Perform data type conversion on the refined data set to obtain a data set with a unified format, and perform dimensional normalization on the data set with a unified format to obtain a standardized data set;

[0048] (5) Extract features from the standardized data set to obtain a feature vector set, and perform dimensionality reduction on the feature vector set through principal component analysis to obtain a key feature set;

[0049] (6) Perform correlation analysis on the key feature set to obtain a feature correlation matrix, and perform data clustering based on the feature correlation matrix to obtain a data distribution pattern.

[0050] Specifically, time series alignment is performed on the original distribution system operation data to ensure that data from different measurement points are consistent in time. Time series alignment is achieved by checking the timestamps of each data point and rearranging data from different sources at a unified time interval to obtain a synchronized data stream. Subsequently, data integrity checks are carried out on this synchronized data stream, including checking for missing values, duplicate values, or obvious outliers, thereby obtaining a valid data set. Outlier detection is performed on the valid data set. Outlier detection can use statistical methods such as the three-standard-deviation method or the quartile method, or machine learning algorithms such as Isolation Forest. Detected outliers will be marked or temporarily removed to obtain a preliminarily cleaned data set. Then, noise removal is performed on the preliminarily cleaned data set through the median filtering algorithm. The principle of median filtering is to replace the value at the center of the window with the median of the data within the sliding window. This method is particularly suitable for removing impulse noise while being able to well preserve the edge features of the data. After median filtering, a smoothed data sequence is obtained.

[0051] Missing value imputation for the smoothed data sequence is the next crucial step. Missing value imputation methods include simple linear interpolation, more complex spline interpolation, or machine learning-based methods such as KNN interpolation. The selection of an appropriate imputation method depends on the characteristics of the data and the distribution pattern of the missing values. After imputation, a complete data sequence is obtained. Subsequently, duplicate value removal is carried out on the complete data sequence. This step can reduce data redundancy and improve the efficiency of subsequent processing, ultimately obtaining a refined data set. Data type conversion is performed on the refined data set to ensure that all data are stored in a unified format, such as converting all numerical values to floating-point numbers. This can avoid calculation errors caused by inconsistent data types. Then, dimensional normalization is carried out on the data set in the unified format. Commonly used normalization methods include Min-Max normalization and Z-score standardization. Normalization processing converts data with different dimensions to the same scale, facilitating subsequent feature extraction and model training, and ultimately obtaining a standardized data set.

[0052] Feature extraction from a standardized dataset is an important step to further reduce the data dimension and extract key information. Feature extraction can include time-domain features (such as mean, variance, peak value), frequency-domain features (such as Fourier transform coefficients), etc. The extracted features form a feature vector set. Subsequently, the feature vector set is dimensionally reduced by principal component analysis (PCA). PCA calculates the eigenvalues and eigenvectors of the covariance matrix, selects several principal components with the greatest contribution, realizes the dimensional reduction of the data, and at the same time retains the main information of the data, obtaining a key feature set. Finally, a correlation analysis is performed on the key feature set, the correlation coefficients between the features are calculated, and a feature correlation matrix is obtained. This matrix reflects the strength of the linear relationship between different features. Based on the feature correlation matrix, data clustering can be performed, such as using the K-means or hierarchical clustering algorithm, to group similar data points into one category, obtaining the data distribution pattern. Such clustering results help to understand the internal structure and distribution characteristics of the data.

[0053] For example, a distribution system collects data every 5 minutes, including voltage, current, power factor, and harmonic content at 10 measurement points. In a day's collection, there are 288 groups of original data (24 hours × 12 times / hour). After time series alignment, it is found that the data at 3 time points is out of sync and is corrected by linear interpolation. Data integrity check finds that 5 groups of data are missing and 2 groups of data are repeated. Outlier detection uses the three-standard-deviation method to identify 7 outliers, including 3 voltage outliers and 4 current outliers. Median filtering uses a 5-point window to effectively smooth the data curve. Missing value imputation uses the spline interpolation method to successfully fill in 5 groups of missing data. After removing duplicate values, the data volume is reduced to 285 groups. The data type is uniformly converted to 64-bit floating-point numbers. Normalization processing uses the Min-Max method to map all data to the [0,1] interval. Feature extraction obtains 20 features for each measurement point, totaling 200 features. PCA analysis retains the first 15 principal components that explain 95% of the variance. Correlation analysis finds that there is a strong correlation between voltage and power factor, and between current and harmonic content (correlation coefficient greater than 0.8). Finally, K-means clustering (k = 3) divides the data into three categories: low load, medium load, and high load, accounting for 23%, 62%, and 15% respectively.

[0054] In a specific embodiment, the process of performing step S103 may specifically include the following steps:

[0055] (1) Split the standardized dataset to obtain a training set and a validation set, and perform feature engineering processing on the training set to obtain an enhanced feature set;

[0056] (2)Construct a deep neural network with 5 fully connected layers for the enhanced feature set to obtain the basic network structure, and add batch normalization layers and rectified linear unit activation functions between every two layers in the basic network structure to obtain a multi-layer neural network architecture;

[0057] (3)Add 3 layers of bidirectional long short-term memory layers after the input layer of the multi-layer neural network architecture, with each layer containing 128 neurons, to obtain a time series analysis network, and apply a multi-head self-attention mechanism with the number of heads set to 8 on the output of the long short-term memory layer of the time series analysis network to obtain the key feature weight distribution;

[0058] (4)Input the key feature weight distribution into 3 layers of two-dimensional convolutional layers, with each layer using a 3x3 convolutional kernel and 32 filters, to obtain a spatial feature analysis network, and add a skip connection after each layer of convolution in the spatial feature analysis network to obtain a deep feature representation;

[0059] (5)Input the deep feature representation into a double deep Q-network structure, which contains two identical Q-networks, and each Q-network contains 3 layers of fully connected layers with the number of neurons being 256, 128, and 64 respectively, to obtain a decision optimization network, and connect the pre-trained knowledge graph embedding vector with the output of the second layer of the decision optimization network to obtain a knowledge-enhanced decision representation;

[0060] (6)Add a transfer learning structure containing 2 layers of fully connected layers after the knowledge-enhanced decision representation, with each layer containing 64 neurons, to obtain an adaptive optimization network, and perform parameter tuning on the learning rate, batch size, and dropout rate of the adaptive optimization network through 5-fold cross-validation to obtain the AI analysis model for the power distribution system.

[0061] Specifically, the standardized data set is segmented. Usually, 80% is used as the training set and 20% is used as the validation set. Feature engineering processing is performed on the training set, including feature selection, feature combination, and feature transformation, to obtain an enhanced feature set. Feature engineering aims to improve the expression ability and generalization performance of the model. Based on the enhanced feature set, a deep neural network with 5 fully connected layers is constructed to form the basic network structure. Batch normalization layers and rectified linear unit activation functions are added between every two layers to obtain a multi-layer neural network architecture. Batch normalization helps to alleviate the problem of internal covariate shift and accelerate network convergence, while the rectified linear unit activation function can effectively alleviate the problem of gradient disappearance and improve the non-linear expression ability of the model.

[0062] Add 3 layers of bidirectional long short-term memory (BiLSTM) layers after the input layer of the multi-layer neural network architecture, with each layer containing 128 neurons, to form a time series analysis network. The BiLSTM network can capture the long-term dependencies of data while considering the context information. Apply the multi-head self-attention mechanism to the output of the LSTM layer in the time series analysis network, with the number of heads set to 8, to obtain the key feature weight distribution. The multi-head self-attention mechanism allows the model to focus on the information in different subspaces, enhancing the diversity of feature extraction. Input the key feature weight distribution into 3 layers of two-dimensional convolutional layers, with each layer using a 3x3 convolutional kernel and 32 filters, to construct a spatial feature analysis network. The convolutional operation can effectively capture local features and spatial correlations. Add a skip connection after each layer of convolution in the spatial feature analysis network to obtain the deep feature representation. The skip connection helps to alleviate the vanishing gradient problem in deep networks while retaining the shallow feature information.

[0063] The deep feature representation is input into a double deep Q-network structure, which contains two identical Q-networks. Each Q-network contains 3 layers of fully connected layers with 256, 128, and 64 neurons respectively, to form a decision optimization network. The double deep Q-network structure can improve the stability and accuracy of decision-making. Connect the pre-trained knowledge graph embedding vectors with the output of the second layer of the decision optimization network to obtain the knowledge-enhanced decision representation. The introduction of the knowledge graph provides domain knowledge support for the model, improving the interpretability and accuracy of decision-making. Finally, add a transfer learning structure containing 2 layers of fully connected layers, with each layer containing 64 neurons, after the knowledge-enhanced decision representation to obtain an adaptive optimization network. The transfer learning structure enables the model to adapt to new scenarios and tasks. Tune the parameters of the learning rate, batch size, and dropout rate of the adaptive optimization network through 5-fold cross-validation, and finally obtain the AI analysis model for the power distribution system.

[0064] For example, the dataset of a certain power distribution system contains one year of historical data, a total of 365 days, with 288 time points per day (one sampling point every 5 minutes). Each time point contains data on voltage, current, power factor, and harmonic content at 10 measurement points. After data segmentation, the training set contains data for 292 days, and the validation set contains data for 73 days. After feature engineering, the original features are extended from 40 (10 measurement points × 4 parameters) to 100, including time features (such as hour, day of the week, whether it is a holiday), lag features (data from the previous 1 hour, previous 24 hours), and statistical features (such as moving average, standard deviation).

[0065] The number of neurons in the 5 fully-connected layers of the deep neural network is 512, 256, 128, 64, and 32 respectively. The batch normalization layer standardizes the output of each layer, keeping the mean at 0 and the variance at 1. The bidirectional long short-term memory network processes a 24-hour historical data sequence to capture the intraday load change pattern. The multi-head self-attention mechanism calculates 8 groups of different attention weights to highlight the influence of important time points and features. The convolutional network treats the data of 10 measurement points as a 10×10 two-dimensional grid and extracts local correlations through a 3×3 convolutional kernel. The skip connection directly adds the output of the first-layer convolution to the output of the third-layer convolution to fuse low-level and high-level features. The double deep Q-network structure calculates the Q value of the current state and the target Q value respectively to reduce the overestimation bias. The knowledge graph embedding contains information such as device topology relationships and historical fault records, with a dimension of 128.

[0066] After the model training of the transfer learning structure is completed, it is fine-tuned with a small amount of new scenario data (such as newly added distribution lines) to adapt to the new working environment. During the 5-fold cross-validation process, the learning rate gradually decays from 0.001 to 0.00001, the batch size tries three values of 32, 64, and 128, and the dropout rate is adjusted between 0.1 and 0.5. The final model achieves a 95% load prediction accuracy (error within ±5%) on the validation set, the F1 score of anomaly detection reaches 0.92, the dynamic balancing strategy improves the power quality by 12%, and the overall system efficiency is increased by 8%. This complex AI analysis model can not only accurately predict load changes but also timely identify abnormal conditions.

[0067] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0068] (1) Extract features from the real-time collected distribution data to obtain a real-time feature set, and input the real-time feature set into the AI analysis model to obtain a preliminary prediction result;

[0069] (2) Perform short-term load prediction on the preliminary prediction result to obtain load prediction data from 15 minutes to 1 hour, and perform medium-term load prediction on the load prediction data to obtain the load prediction trend from 1 day to 1 week;

[0070] (3) Perform long-term load prediction based on the load prediction trend to obtain the load prediction plan from 1 month to 1 year, and perform error analysis on the load prediction plan to obtain the prediction credibility assessment;

[0071] (4) Conduct load distribution analysis on the distribution network according to the prediction credibility assessment to obtain the load prediction result, and generate a load balancing plan based on the load prediction result to obtain a preliminary dynamic balancing strategy;

[0072] (5) Perform multi-scenario simulations on the preliminary dynamic balancing strategy to obtain an evaluation of the strategy's effectiveness, and optimize the parameters based on the evaluation of the strategy's effectiveness to obtain the dynamic balancing strategy.

[0073] Specifically, extract features from the real-time collected power distribution data to obtain a real-time feature set. The feature extraction process includes the calculation of time-domain features (such as mean, standard deviation, peak factor), frequency-domain features (such as power spectral density, main frequency components), and statistical features (such as skewness, kurtosis). These features comprehensively describe the current operating state of the power distribution system. Input the extracted real-time feature set into a pre-trained AI analysis model, which may be a deep neural network, a long short-term memory network, or a hybrid structure of both. The model processes these features and outputs preliminary prediction results, including short-term load trends and possible anomalies. Based on the preliminary prediction results, perform short-term load forecasting to obtain load forecasting data for 15 minutes to 1 hour. Short-term forecasting uses time series analysis methods, such as autoregressive integrated moving average model (ARIMA) or exponential smoothing method, considering the recent load change trend and periodic patterns. Subsequently, perform medium-term load forecasting on these short-term load forecasting data to obtain the load forecasting trend for 1 day to 1 week. Medium-term forecasting incorporates more external factors, such as weather forecasts, weekday / weekend patterns, special events, etc., and uses more complex models such as support vector regression (SVR) or random forest algorithms.

[0074] Perform long-term load forecasting based on the medium-term load forecasting trend to obtain a load forecasting plan for 1 month to 1 year. Long-term forecasting needs to consider more macro factors, such as seasonal changes, economic growth trends, population changes, etc. Deep learning models such as long short-term memory network (LSTM) or temporal convolutional network (TCN) may be used here, and these models can capture long-term dependencies. Conduct error analysis on the long-term load forecasting plan, calculate the deviation between the historical forecast and the actual load, and use metrics such as root mean square error (RMSE) or mean absolute percentage error (MAPE) to obtain an evaluation of the prediction credibility. This evaluation reflects the reliability of the prediction results and provides a reference for subsequent decision-making. Based on the evaluation of the prediction credibility, perform load distribution analysis on the power distribution network. This step considers factors such as network topology, substation capacity, line load capacity, etc., and uses power flow calculation methods to analyze the load distribution in the network to obtain detailed load forecasting results. Based on these load forecasting results, generate a load balancing plan. The balancing plan may include measures such as adjusting transformer taps, switching capacitor banks, and dispatching controllable loads, with the goal of minimizing network losses and balancing the load levels of each node. This process forms the preliminary dynamic balancing strategy.

[0075] Finally, multi-scenario simulations are carried out on the preliminary dynamic balancing strategy. The simulation process takes into account various possible situations, such as sudden large loads, fluctuations in renewable energy generation, equipment failures, etc. Using the Monte Carlo simulation method, a large number of random scenarios are generated to evaluate the performance of the strategy under different conditions, and an evaluation of the strategy effect is obtained. According to the evaluation results, the strategy parameters are optimized. Reinforcement learning algorithms such as Deep Q-Network (DQN) or policy gradient methods may be used to continuously adjust and improve the strategy, and finally a dynamic balancing strategy with strong robustness and good adaptability is obtained.

[0076] For example, an intelligent power distribution system covers the commercial and residential areas of a medium-sized city and includes 100 power distribution nodes. Real-time data collection shows that at 10 am on a weekday, the total load is 50 MW, the power factor is 0.92, and the voltage deviation is within the range of ±3%. After feature extraction, the obtained real-time feature set includes the load growth rate (2% every 15 minutes), the voltage fluctuation frequency (10 times per hour), etc. The AI analysis model predicts that the load will continue to increase in the next hour and may reach 55 MW. The ARIMA model is used for short-term load forecasting, and the predicted load after 1 hour is 54.8 MW, with an error range of ±1 MW. For medium-term forecasting, considering a large event three days later, the peak load on that day is predicted to reach 65 MW. Long-term forecasting combines the urban development plan and predicts that the daily average load will increase by 10% after one year and reach 55 MW. Error analysis shows that the MAPE of short-term forecasting is 2%, 5% for medium-term, and 8% for long-term.

[0077] Load distribution analysis finds that the load rate of a substation in the commercial area has reached 85%, posing an overload risk. The preliminary dynamic balancing strategy recommends transferring some loads to a nearby substation with a load rate of only 60%. The multi-scenario simulation considers the situation where the air-conditioning load surges due to sudden high-temperature weather. The simulation results show that the initial strategy may cause local overload in extreme cases. After parameter optimization, the final dynamic balancing strategy reduces the network loss by 7% while ensuring power supply reliability and improves the load balance degree by 15%.

[0078] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0079] (1) Monitor the system operation parameters in real time to obtain a parameter change sequence, and compare the parameter change sequence with historical normal operation data to obtain a set of deviation values;

[0080] (2) Perform clustering analysis on the set of deviation values to obtain a candidate set of abnormal patterns, and extract features from the candidate set of abnormal patterns to obtain abnormal feature vectors;

[0081] (3) Input the abnormal feature vector into the pre-trained anomaly detection model to obtain the abnormal probability distribution, and determine the abnormal level according to the abnormal probability distribution to obtain the evaluation of the device abnormal state;

[0082] (4) Conduct multi-path reasoning in the knowledge graph based on the evaluation of the device abnormal state to obtain the set of fault causes, and perform priority sorting on the set of fault causes to obtain the fault diagnosis result;

[0083] (5) Generate a device health status report according to the fault diagnosis result to obtain the evaluation of the device health status, and formulate a preventive maintenance plan based on the evaluation of the device health status to obtain the maintenance suggestion.

[0084] Specifically, the system operation parameters are monitored in real time to obtain the parameter change sequence. These parameters include key indicators such as voltage, current, power factor, and harmonic content. Through high-precision sensors, data is collected at a sampling frequency of milliseconds to form a continuous parameter change sequence. Subsequently, these parameter change sequences are compared with the historical normal operation data. The comparison process uses the sliding window technique to calculate the parameter statistical features (such as mean, standard deviation, peak value, etc.) within the current time window, and compare them with the statistical features of the historical data in the same period to obtain the deviation value set. Cluster analysis is performed on the deviation value set, using clustering algorithms such as K-means or DBSCAN to classify similar deviation patterns to obtain the abnormal pattern candidate set. The clustering process considers multiple dimensions such as the magnitude, duration, and change trend of the deviation. Feature extraction is performed on the abnormal pattern candidate set, including time-domain features (such as mean, variance, peak factor), frequency-domain features (such as power spectral density, main frequency components), and statistical features (such as skewness, kurtosis) to obtain the abnormal feature vector. These features comprehensively describe all aspects of the abnormal pattern.

[0085] Input the abnormal feature vector into the pre-trained anomaly detection model. This model can be an autoencoder based on deep learning or an isolation forest algorithm based on ensemble learning. The model outputs the abnormal probability distribution, which reflects the degree of deviation of the current operating state from the normal state. According to the abnormal probability distribution, different thresholds are set to divide the abnormal state into different levels such as mild, moderate, and severe to obtain the evaluation of the device abnormal state. This grading method helps to reasonably allocate maintenance resources. Based on the evaluation of the device abnormal state, multi-path reasoning is performed in the pre-constructed knowledge graph. The knowledge graph contains information such as device topology relationships, historical fault records, and expert experience. The multi-path reasoning uses graph traversal algorithms such as depth-first search or breadth-first search to start from the abnormal device node and explore possible fault propagation paths to obtain the set of potential fault causes. Priority sorting is performed on the set of fault causes, considering factors such as the strength of the causal relationship, historical occurrence frequency, and potential impact degree, and using a weighted scoring method to obtain the final fault diagnosis result.

[0086] Generate a device health status report based on the fault diagnosis results. The report includes detailed information such as abnormal parameters, degree of abnormality, possible fault causes and their probabilities. Based on the device health status assessment, formulate a preventive maintenance plan. The maintenance plan takes into account multiple factors such as device importance, fault urgency, maintenance cost and impact on system operation. Through multi-objective optimization algorithms, such as genetic algorithms or particle swarm optimization, obtain maintenance suggestions that balance various requirements.

[0087] For example, an abnormality occurred in a transformer of a distribution substation during continuous monitoring. The real-time monitoring data shows that the oil temperature of this transformer has shown a slow upward trend in the past 24 hours, rising from the normal 65°C to 75°C. Comparing with historical data for the same period, it is found that the temperature deviation exceeds 3 standard deviations. Cluster analysis classifies this temperature anomaly pattern as the "slow temperature rise" type. The abnormal feature vector obtained by feature extraction includes the temperature rise rate (0.42°C / hour), duration (24 hours), intra-day fluctuation range (±1.5°C), etc. The anomaly detection model analyzes these features, outputs an anomaly probability of 0.87, and classifies it as a moderate anomaly. During the knowledge graph reasoning process, considering the location of the transformer, the status of connected devices and historical fault records, three possible fault causes are deduced: decreased cooling system efficiency (probability 0.6), overloading (probability 0.3) and insulation aging (probability 0.1). The health status report details these findings and gives maintenance suggestions: prioritize checking the cooling system, while monitoring the load situation. If the problem persists, arrange for an insulation test during the next planned power outage.

[0088] The preventive maintenance plan recommends arranging a cooling system inspection within 72 hours. Considering that the power supply scope of this transformer includes an important industrial park, it is recommended to carry out maintenance on weekends when the load is low to minimize the impact on users. This example demonstrates how an AI-driven power distribution system can achieve full-process intelligent management from anomaly detection to fault diagnosis and then to maintenance decision-making, greatly improving the reliability and operation and maintenance efficiency of the system.

[0089] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0090] (1) Conduct a health status assessment on the renewable energy power generation equipment to obtain a predicted available power generation capacity, and combine weather forecast data to predict the power generation of renewable energy to obtain a renewable energy supply prediction;

[0091] (2) Conduct an electricity demand analysis based on historical user electricity consumption data and load prediction results to obtain an electricity demand prediction, and perform a matching analysis between the electricity demand prediction and the renewable energy supply prediction to obtain a preliminary supply-demand balance plan;

[0092] (3) Perform multi-objective optimization calculations on the preliminary supply-demand balance plan to obtain a set of candidate scheduling plans, and conduct simulation verification on the set of candidate scheduling plans to obtain a plan feasibility assessment;

[0093] (4) Select the optimal scheduling plan based on the plan feasibility assessment to obtain an intelligent energy scheduling plan.

[0094] Specifically, conduct a health status assessment on renewable energy power generation equipment, which includes a comprehensive analysis of the efficiency, wear degree, and fault conditions of equipment such as photovoltaic panels and wind turbines. By comparing real-time monitoring data with historical performance curves, obtain the current operating efficiency of the equipment. Combining this information, use the following formula to predict the available power generation capacity:

[0095]

[0096] where, C a is the total available power generation capacity, C i is the rated capacity of the i-th device, E i is its current operating efficiency, n is the total number of devices, and combine weather forecast data to predict the power generation of renewable energy. For photovoltaic power generation, use the following model:

[0097] P pv =η×A×G×(1 - 0.005(T - 25))

[0098] Here, P pv is the photovoltaic power generation power, η is the photovoltaic conversion efficiency, A is the area of the photovoltaic panel, G is the solar radiation intensity, and T is the ambient temperature. For wind power generation, use a similar model considering factors such as wind speed and air density. Conduct an electricity demand analysis based on historical user electricity data and load prediction results. Adopt time series analysis methods, such as the ARIMA model or long short-term memory network (LSTM), to predict the electricity demand for the next 24 hours. The prediction model considers variables such as time factors (such as hours, weekdays / weekends), temperature, and special events. Conduct a matching analysis between the electricity demand prediction and the renewable energy supply prediction, calculate the supply-demand difference for each time period, and obtain a preliminary supply-demand balance plan.

[0099] Perform multi-objective optimization calculations on the preliminary supply-demand balance plan, considering objectives such as economy, reliability, and environmental protection. The optimization problem can be expressed as:

[0100] min f(x) = [f1(x), f2(x), f3(x)]

[0101] subject to: g i (x) ≤ 0, i = 1,..., m; hj(x) = 0, j = 1,..., n

[0102] Among them, f1(x) represents the operating cost, f2(x) represents the power supply reliability index, and f3(x) represents the carbon emissions. g i (x) and h j (x) are inequality and equality constraints respectively, representing various limiting conditions of the system operation. Solve this problem using a multi-objective genetic algorithm or a particle swarm optimization algorithm to obtain a series of non-dominated solutions, forming a candidate scheduling solution set. Conduct simulation verification on the candidate scheduling solution set, use power system simulation software to simulate the system operation status under different solutions, and evaluate the feasibility and stability of the solutions. The simulation process considers abnormal conditions such as load fluctuations and equipment failures to comprehensively evaluate the robustness of the solutions. According to the simulation results, calculate the comprehensive score of each solution, and select the solution with the highest score as the final intelligent energy scheduling solution.

[0103] For example, an intelligent microgrid system includes 10 MW of photovoltaic power generation, 5 MW of wind power generation, and a 20 MWh energy storage system. The health status assessment shows that the efficiency of the photovoltaic system is 92%, and the efficiency of the wind power system is 95%. The weather forecast shows that the sun will be sufficient the next day, and the predicted photovoltaic power generation is 80 MWh, and the wind power generation is 40 MWh. The electricity demand forecast results show that the demand during the peak electricity consumption period (10:00 - 14:00) is 15 MW, and the demand during the low valley period (02:00 - 06:00) is 5 MW. The preliminary supply-demand balance plan suggests storing excess power in the energy storage system during the low valley period and releasing the energy storage to supplement power supply during the peak period. The multi-objective optimization calculation considers three objectives: power cost, reliability, and carbon emissions, and generates 10 candidate solutions. After simulation verification, the optimal solution releases 10 MWh of power from the energy storage system during the peak period, while maintaining a 2 MW reserve capacity to cope with sudden demands, and is expected to reduce the operating cost by 15%, improve the power supply reliability by 8%, and reduce carbon emissions by 20 tons. This example demonstrates how an AI-driven intelligent power distribution system optimizes the utilization of renewable energy, balances the supply-demand relationship, and achieves multiple goals of economy, reliability, and environmental protection.

[0104] The above describes the power distribution method based on AI artificial intelligence in the embodiments of the present application. Next, the intelligent power distribution cabinet in the embodiments of the present application will be described. Please refer to Figure 2 One embodiment of the intelligent power distribution cabinet in the embodiments of the present application includes:

[0105] An acquisition module 201, configured to perform real-time data acquisition on key nodes of the power distribution system through an intelligent sensor network to obtain original power distribution system operation data;

[0106] A processing module 202, configured to perform cleaning and standardization processing on the original power distribution system operation data through a data preprocessing algorithm to obtain a standardized data set;

[0107] The modeling module 203 is used to analyze and model the standardized data set through multi-level hybrid model training to obtain an AI analysis model for the power distribution system;

[0108] The analysis module 204 is used to process and analyze the power distribution data collected in real time through the AI analysis model to obtain a load prediction result and a dynamic balance strategy;

[0109] The identification module 205 is used to perform abnormal pattern recognition and multi-path reasoning on the system operation parameters based on the load prediction result and the dynamic balance strategy to obtain an equipment health status assessment and maintenance suggestions;

[0110] The calculation module 206 is used to perform multi-objective optimization calculation on the renewable energy supply and power consumption demand based on the equipment health status assessment and maintenance suggestions to obtain an intelligent energy scheduling plan.

[0111] Through the collaborative cooperation of the above-mentioned various components, real-time data collection of key nodes of the power distribution system is carried out through the intelligent sensor network, and the system operation status can be comprehensively and timely obtained, providing a high-quality and high-timeliness raw data basis for subsequent analysis, and significantly improving the comprehensiveness and real-time of data collection. The data preprocessing algorithm is used to clean and standardize the original power distribution system operation data, which not only effectively removes the noise and outliers in the data, but also realizes the unified representation of the data, laying a solid data foundation for subsequent model training and analysis, and greatly improving the data quality and usability. The introduction of multi-level hybrid model training enables the system to capture the time series characteristics, spatial characteristics and decision-making characteristics of power distribution data at the same time, significantly enhancing the expression ability and generalization performance of the model, and being able to more accurately describe the complex dynamic characteristics of the power distribution system. Through the AI analysis model to process and analyze the power distribution data collected in real time, the system can quickly and accurately generate the load prediction result and the dynamic balance strategy, greatly improving the prediction accuracy and response speed of the power distribution system, and providing reliable support for real-time scheduling and decision-making. Based on the load prediction result and the dynamic balance strategy, abnormal pattern recognition and multi-path reasoning are performed on the system operation parameters, which can not only timely detect potential equipment failures and system abnormalities, but also give accurate equipment health status assessments and maintenance suggestions, significantly improving the reliability and maintenance efficiency of the system. Finally, through the multi-objective optimization calculation of the renewable energy supply and power consumption demand, the obtained intelligent energy scheduling plan can maximize the utilization rate of renewable energy while ensuring power supply reliability, and achieve the balance of economy and environmental protection, greatly improving the overall operation efficiency and sustainability of the power distribution system.

[0112] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A power distribution method based on AI artificial intelligence, characterized in that: The power distribution method based on AI artificial intelligence includes: Real-time data collection of key nodes of the power distribution system is carried out through the intelligent sensor network to obtain the original power distribution system operation data; Cleaning and standardizing the original power distribution system operation data using a data preprocessing algorithm to obtain a standardized data set; The standardized data set is analyzed and modeled through multi-level hybrid model training to obtain an AI analysis model for the distribution system, including: dividing the standardized data set to obtain a training set and a validation set, performing feature engineering on the training set to obtain an enhanced feature set; constructing a deep neural network containing 5 layers of fully connected layers for the enhanced feature set to obtain a basic network structure and adding a batch normalization layer and a rectified linear unit activation function between each two layers to obtain a multi-layer neural network architecture; adding 3 layers of bidirectional long and short-term memory layers after the input layer of the multi-layer neural network architecture to obtain a time series analysis network and applying multi-head self-injection on the long and short-term memory layer. The intention mechanism is used to obtain the weight distribution of key features; the key feature weight distribution is input into three layers of two-dimensional convolutional layers to obtain a spatial feature analysis network. Skip connections are added after each convolution layer to obtain a deep feature representation and input into a dual-depth Q network structure to obtain a decision optimization network. The pre-trained knowledge graph embedding vector is connected to the second-layer output of the decision optimization network to obtain a decision representation; a transfer learning structure is added after the decision representation to obtain an adaptive optimization network. The learning rate, batch size, and random deactivation rate parameters of the adaptive optimization network are tuned through 5-fold cross-validation to obtain an AI analysis model for the distribution system; The AI analysis model processes and analyzes the real-time collected power distribution data to obtain load prediction results and dynamic balancing strategies; Based on the load prediction results and dynamic balancing strategy, abnormal pattern recognition and multi-path reasoning are performed on system operating parameters to obtain equipment health status assessment and maintenance recommendations; Based on the equipment health status assessment and maintenance recommendations, a multi-objective optimization calculation is performed on renewable energy supply and electricity demand to obtain an intelligent energy scheduling solution.

2. The power distribution method based on AI artificial intelligence according to claim 1, characterized in that: The real-time data collection of key nodes of the power distribution system through the intelligent sensor network is used to obtain the original power distribution system operation data, including: Arrange intelligent sensors on the substations, distribution cabinets and user terminals of the power distribution system to obtain a multi-level data acquisition network, and configure the communication protocol on the multi-level data acquisition network to obtain a high-speed, real-time data transmission channel; Continuously monitoring voltage, current, power factor, and harmonic content through the high-speed, real-time data transmission channel to obtain multi-dimensional power parameter data, and timestamping the multi-dimensional power parameter data to obtain a time-series-associated original data stream; Performing a data integrity check on the original data stream associated with the time series to obtain a valid data set, and performing an outlier detection on the valid data set to obtain a preliminarily cleaned data set; Smoothing the preliminarily cleaned data set using a sliding window algorithm to obtain a denoised data sequence, and interpolating missing values in the denoised data sequence to obtain a complete data sequence; Standardizing and normalizing the complete data sequence to obtain a standard data set with unified dimensions, and performing feature extraction on the standard data set with unified dimensions to obtain a feature vector after dimensionality reduction; Performing key feature identification on the eigenvector after dimensionality reduction by a principal component analysis algorithm to obtain a core feature set, and performing correlation analysis on the core feature set to obtain a feature correlation matrix; A multidimensional data cube is constructed based on the feature association matrix to obtain multi-angle data analysis views, and data clustering is performed on the multi-angle data analysis views to obtain a data distribution pattern.

3. The power distribution method based on AI artificial intelligence according to claim 1, characterized in that: The raw power distribution system operation data is cleaned and standardized by a data preprocessing algorithm to obtain a standardized data set, including: Performing time series alignment on the original power distribution system operation data to obtain a synchronized data stream, and performing data integrity check on the synchronized data stream to obtain a valid data set; Performing outlier detection on the valid data set to obtain a preliminary cleaned data set, and removing noise from the preliminary cleaned data set using a median filtering algorithm to obtain a smoothed data sequence; Interpolating missing values on the smoothed data sequence to obtain a complete data sequence, and removing duplicate values from the complete data sequence to obtain a streamlined data set; Performing data type conversion on the streamlined data set to obtain a data set in a unified format, and performing dimension normalization processing on the data set in the unified format to obtain a standardized data set; Performing feature extraction on the standardized data set to obtain a feature vector set, and performing dimensionality reduction processing on the feature vector set through principal component analysis to obtain a key feature set; Correlation analysis is performed on the key feature set to obtain a feature association matrix, and data clustering is performed based on the feature association matrix to obtain a data distribution pattern.

4. The power distribution method based on AI artificial intelligence according to claim 1, characterized in that: The AI analysis model processes and analyzes the real-time collected power distribution data to obtain load prediction results and dynamic balancing strategies, including: Extracting features from the real-time collected power distribution data to obtain a real-time feature set, and inputting the real-time feature set into the AI analysis model to obtain a preliminary prediction result; Performing short-term load forecasting on the preliminary forecast results to obtain load forecast data for 15 minutes to 1 hour, and performing medium-term load forecasting on the load forecast data to obtain load forecast trends for 1 day to 1 week; Performing long-term load forecasting based on the load forecast trend to obtain a load forecast plan for 1 month to 1 year, and performing error analysis on the load forecast plan to obtain a forecast credibility assessment; Performing load distribution analysis on the power distribution network according to the prediction credibility assessment to obtain a load prediction result, and generating a load balancing solution based on the load prediction result to obtain a preliminary dynamic balancing strategy; The preliminary dynamic balance strategy is simulated in multiple scenarios to obtain a strategy effect evaluation, and parameters are optimized based on the strategy effect evaluation to obtain a dynamic balance strategy.

5. The power distribution method based on AI artificial intelligence according to claim 1, characterized in that: Based on the load prediction results and dynamic balancing strategy, abnormal pattern recognition and multi-path reasoning are performed on system operating parameters to obtain equipment health status assessment and maintenance recommendations, including: Monitor system operating parameters in real time to obtain a parameter change sequence, and compare the parameter change sequence with historical normal operating data to obtain a set of deviation values; Performing cluster analysis on the deviation value set to obtain an abnormal pattern candidate set, and performing feature extraction on the abnormal pattern candidate set to obtain an abnormal feature vector; Inputting the abnormal feature vector into a pre-trained anomaly detection model to obtain an abnormal probability distribution, and determining an abnormality level based on the abnormal probability distribution to obtain an abnormal state assessment of the device; Based on the abnormal state evaluation of the equipment, multi-path reasoning is performed in the knowledge graph to obtain a set of fault causes, and the set of fault causes is prioritized to obtain a fault diagnosis result; An equipment health status report is generated based on the fault diagnosis results to obtain an equipment health status assessment, and a preventive maintenance plan is formulated based on the equipment health status assessment to obtain maintenance recommendations.

6. The power distribution method based on AI artificial intelligence according to claim 1, characterized in that: Based on the equipment health status assessment and maintenance recommendations, a multi-objective optimization calculation is performed on renewable energy supply and electricity demand to obtain an intelligent energy scheduling solution, including: Conduct health status assessments on renewable energy power generation equipment to obtain available power generation capacity forecasts, and combine weather forecast data to forecast renewable energy power generation and obtain renewable energy supply forecasts; Performing electricity demand analysis based on historical user electricity consumption data and load forecast results to obtain an electricity demand forecast, and performing a matching analysis between the electricity demand forecast and the renewable energy supply forecast to obtain a preliminary supply and demand balance solution; Performing multi-objective optimization calculations on the preliminary supply and demand balance plan to obtain a set of candidate scheduling plans, and performing simulation verification on the set of candidate scheduling plans to obtain a feasibility assessment of the plans; The optimal scheduling scheme is selected according to the feasibility evaluation of the scheme to obtain the intelligent energy scheduling scheme.

7. An intelligent power distribution cabinet, used to implement the power distribution method based on AI artificial intelligence as described in any one of claims 1 to 6, characterized in that: The intelligent power distribution cabinet includes: The acquisition module is used to collect real-time data from key nodes of the power distribution system through the intelligent sensor network to obtain the original power distribution system operation data; A processing module, configured to clean and standardize the raw power distribution system operation data using a data preprocessing algorithm to obtain a standardized data set; The modeling module is used to analyze and model the standardized data set through multi-level hybrid model training to obtain an AI analysis model of the power distribution system, including: dividing the standardized data set to obtain a training set and a validation set, performing feature engineering on the training set to obtain an enhanced feature set; constructing a deep neural network with 5 fully connected layers for the enhanced feature set to obtain a basic network structure and adding a batch normalization layer and a rectified linear unit activation function between each two layers to obtain a multi-layer neural network architecture; adding 3 layers of bidirectional long and short-term memory layers after the input layer of the multi-layer neural network architecture to obtain a time series analysis network and applying multi-layer long and short-term memory layers to the long and short-term memory layers. The head self-attention mechanism is used to obtain the weight distribution of key features; the key feature weight distribution is input into three layers of two-dimensional convolutional layers to obtain a spatial feature analysis network. Skip connections are added after each convolution layer to obtain a deep feature representation and input into a dual-depth Q network structure to obtain a decision optimization network. The pre-trained knowledge graph embedding vector is connected to the second-layer output of the decision optimization network to obtain a decision representation; a transfer learning structure is added after the decision representation to obtain an adaptive optimization network. The learning rate, batch size, and random deactivation rate parameters of the adaptive optimization network are tuned through 5-fold cross-validation to obtain an AI analysis model for the distribution system; An analysis module, configured to process and analyze the real-time collected power distribution data using the AI analysis model to obtain load prediction results and dynamic balancing strategies; an identification module for performing abnormal pattern recognition and multi-path reasoning on system operating parameters based on the load prediction results and dynamic balancing strategy to obtain equipment health status assessment and maintenance recommendations; The calculation module is used to perform multi-objective optimization calculations on renewable energy supply and electricity demand based on the equipment health status assessment and maintenance recommendations to obtain an intelligent energy scheduling solution.

Images