A method and system for comprehensive management of power distribution communication network

Through the combination of One-Class SVM and LSTM models, abnormal patterns are identified and resource allocation is optimized, and the problem of low resource allocation efficiency in traditional network management in 5G distribution communication networks is solved, and efficient resource utilization and power system stability are achieved. .

CN118647092BActive Publication Date: 2024-12-03STATE GRID INFO TELECOM GREAT POWER SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411099471.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-12-03
Estimated Expiration
2044-08-12

AI Technical Summary

Technical Problem

When traditional network management solutions deal with high-dimensional and dynamic 5G distribution communication network needs, they are difficult to meet the level of reliability, flexibility and intelligence, resulting in low resource allocation efficiency and affecting the stable operation and user experience of the power system.

Method used

The One-Class SVM model is used to detect abnormalities, combine the LSTM model to predict network status, identify abnormal patterns through clustering algorithms and feature importance analysis, and use SD-WAN's dynamic resource scheduling strategy to optimize 5G slice resource allocation, adjust bandwidth and Priority, predict and prevent potential congestion in advance.

Benefits of technology

It effectively improves the resource utilization rate and system performance of the 5G network, improves the detection accuracy and prediction accuracy of the network, realizes real-time response and potential congestion prevention, and improves the stability and user experience of the power system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118647092B_ABST
    Figure CN118647092B_ABST
Patent Text Reader

Abstract

The present invention relates to a method and system for integrated management of power distribution communication network, comprising the following steps: S1: obtaining 5G network slice data, card connection data and power terminal collection data opened by public network operators, and preprocessing; S2: constructing an anomaly detection model based on the One-Class SVM model, identifying abnormal data points based on the anomaly detection model, clustering and classifying abnormal data points using clustering algorithms and feature importance analysis, identifying different types of abnormal patterns, and combining network and performance characteristics to determine specific abnormal patterns; S3: constructing a prediction model based on the LSTM model to predict future network status; S4: combining the anomaly detection results with the prediction results, using the dynamic resource scheduling strategy SD-WAN to optimize 5G slice resource allocation and improve network utilization; S5: performing data visualization, creating a dashboard, and displaying network status, traffic, and performance information. The present invention effectively optimizes the resource allocation of 5G networks and improves network utilization and system performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of power communication management, and in particular to a method and system for comprehensive management of a power distribution communication network. Background Art

[0002] With the continuous development of power systems and the deepening of intelligentization, traditional power distribution networks are gradually shifting towards intelligence and automation, giving rise to the rise of smart grids. Smart grids not only achieve automated control and real-time monitoring in power generation, transmission and distribution, but also closely connect various power equipment, sensors and control systems through advanced communication technologies, achieving real-time data transmission and processing, thereby improving the efficiency and reliability of power systems.

[0003] In the construction and operation of smart grids, distribution communication networks play a key role. The distribution communication network is the neural network of the power system, responsible for transmitting a large amount of real-time data, including power terminal collection data, equipment status information, network control instructions, and various distribution system parameters. The accurate and fast transmission and processing of these data have a vital impact on the stable operation, real-time control and fault diagnosis of the power system. With the in-depth application of 5G networks, distribution communication networks will play a more important role in smart grids. Through 5G empowerment, distribution communication networks can not only improve data transmission rate and security, but also realize more intelligent and automated network management and resource scheduling, further improving the overall performance and user experience of the power system. However, the widespread application of 5G networks requires distribution communication networks to have higher reliability, flexibility and intelligence. However, traditional network management solutions seem to be a little powerless when dealing with such high-dimensional and dynamic network requirements. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a comprehensive management method and system for distribution communication networks, so as to effectively optimize the resource allocation of 5G networks and improve network utilization and system performance.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for integrated management of a power distribution communication network comprises the following steps:

[0007] S1: Obtain 5G network slicing data, card connection data, and power terminal collection data opened by public network operators, and perform data cleaning, deduplication, and abnormal data processing, convert heterogeneous data into a unified format, and store it;

[0008] S2: Build an anomaly detection model based on the One-Class SVM model, identify abnormal data points based on the anomaly detection model, use clustering algorithm and feature importance analysis to cluster and classify abnormal data points, identify different types of abnormal patterns, and determine specific abnormal patterns based on network and performance characteristics;

[0009] S3: Build a prediction model based on the LSTM model to predict the future network status;

[0010] S4: Combining the anomaly detection results with the prediction results, the dynamic resource scheduling strategy SD-WAN is used to optimize the 5G slice resource allocation and improve network utilization, as follows:

[0011] Based on the anomaly detection results, the centralized control function of SD-WAN is used to globally schedule network resources to improve resource utilization. The resource allocation of each slice is dynamically adjusted according to the business importance and real-time network status, including bandwidth and priority adjustments:

[0012] Bandwidth adjustment:

[0013]

[0014] Among them, B new is the adjusted bandwidth, B current is the current bandwidth; ΔB is the bandwidth adjustment, calculated based on the abnormal mode;

[0015] Priority Adjustment:

[0016] ;

[0017] Among them, P new is the adjusted priority, P current is the current priority; ΔP is the adjustment of the priority, calculated according to the abnormal pattern;

[0018] According to the prediction results of the prediction model, adjust network resources in advance to prevent potential congestion:

[0019] ;

[0020] in, For the future resource allocation after time units; According to the prediction results Function for resource adjustment; t represents the current time;

[0021] S5: Perform data visualization and create dashboards to display network status, traffic, and performance information.

[0022] Furthermore, S1 is specifically:

[0023] S11: Obtain the 5G network slice configuration, performance indicators and usage data opened by the operator, and obtain the connection status and traffic data of the SIM card from the operator's connection management system;

[0024] S12: Perform data cleaning, deduplication and abnormal data processing;

[0025] S13: Perform field mapping and format conversion on the processed data from different sources to make them conform to the unified data format specification.

[0026] Furthermore, a prediction model is constructed based on the LSTM model, as follows:

[0027] A multi-layer LSTM network is used, and the specific structure includes an input layer, several LSTM layers and a Dense layer; a dropout layer is provided after each LSTM layer;

[0028] The LSTM layer is as follows:

[0029] ;

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] in, is the input feature vector of time step i; is the hidden state at time step i-1; For the Gate of Oblivion; is the input gate; is the candidate value, i.e. the candidate memory cell at the current moment; is the cell state, indicating the memory cell state at the current moment; is the state of the memory cell at the previous moment; is the output gate; is the hidden state at the current moment; , , , They are the weight matrices of the forget gate, input gate, candidate value, and output gate respectively; , , , They are the bias vectors of the forget gate, input gate, candidate value, and output gate respectively; is the activation function;

[0036] Define hyperparameters, including the number of LSTM layers, number of hidden units, learning rate, and batch size;

[0037] During the training process, grid search is used to select the best hyperparameter combination, and K-fold cross validation is used to evaluate the performance of different hyperparameter combinations to obtain the optimal hyperparameter combination and the optimal prediction model.

[0038] Furthermore, the prediction model dynamically adjusts the learning rate during the training process, as follows:

[0039] According to the optimal hyperparameter combination, the initial learning rate is obtained η initial ; Define the learning rate scheduling parameters, including the initial stage length t initial , the epoch number t at the end of the mid-term mid ; Total number of training epochs t final, Decay rate constant λ, minimum learning rate η min;

[0040] During the training process, the learning rate is dynamically adjusted according to the following formula:

[0041] ;

[0042] in, is the learning rate at the end of the mid-term;

[0043] In the early stage, a learning rate greater than the threshold is used to converge quickly, so a relatively high learning rate is maintained within the initial epochs:

[0044] ;

[0045] In the medium term, according to the exponential decay method, the learning rate is gradually reduced to achieve stable convergence, and the learning rate gradually decreases in an exponential decay manner:

[0046] ;

[0047] In the later stage, fine-tune with a learning rate less than the threshold to achieve the best performance. Gradually reduce the learning rate and use cosine annealing to further reduce the learning rate;

[0048] .

[0049] Furthermore, an anomaly detection model is constructed based on the One-Class SVM model, as follows:

[0050] Collect network performance indicators and performance indicator data, where network performance indicators include latency, throughput, and packet loss rate; system performance indicators include CPU load, memory usage, and disk I / O; and use SMOTE for oversampling and undersampling to balance data categories;

[0051] Convert time series data into feature vectors and construct training data sets;

[0052] The optimization goal of the One-Class SVM model is:

[0053] ;

[0054] Where W is the weight vector of the model, is the bias term, is a hyperparameter that controls the anomaly rate, is the feature mapping function, It is input feature vectors, n is the number of input feature vectors;

[0055] The decision function is:

[0056] ;

[0057] Solve the optimization objective by Lagrange multiplier method:

[0058] ;

[0059] ;

[0060] in, and They are and Lagrange multipliers, is the kernel function; It is input feature vector;

[0061] A One-Class SVM model is trained based on the training dataset, semi-supervised learning is used to generate pseudo labels and update the model, and Bayesian optimization is used to adjust the model parameters.

[0062] Furthermore, clustering algorithms and feature importance analysis are used to cluster and classify abnormal data points, identify different types of abnormal patterns, and combine network and performance characteristics to determine specific abnormal patterns, as follows:

[0063] Perform feature analysis on historical abnormal samples to determine their original feature data, use K-Means to classify abnormal samples, and identify different types of abnormal patterns:

[0064] ① Initialize k cluster centers {u1,u2,...,u k};

[0065] ②Assign each abnormal sample to the nearest cluster center:

[0066] ;

[0067] in, is the jth abnormal sample identified by the One-Class SVM model, is the cluster center containing abnormal samples; For the Cluster centers;

[0068] ③ Update cluster center:

[0069] ;

[0070] ④ Repeat steps ②-③ until the clustering no longer changes;

[0071] Use the tree model to perform feature importance analysis to determine which original features contribute most to the classification of abnormal samples;

[0072] Cluster analysis and feature importance analysis are combined to determine different types of abnormal patterns. Each abnormal pattern defines a set of feature thresholds and matching conditions, and matches are performed based on the abnormal samples detected by the anomaly detection model.

[0073] Furthermore, the tree model is used to perform feature importance analysis to determine which original features contribute most to the classification of abnormal samples, as follows:

[0074] Train a random forest model:

[0075] ;

[0076] Where T is the number of decision trees in the random forest, Indicates The predicted value of the decision tree for the input sample x;

[0077] For each node in the decision tree , through the relevant features Perform segmentation and calculate the change in Gini impurity before and after segmentation. Calculate the Gini importance of each feature. :

[0078] For each feature , accumulate the Gini impurity reduction on all nodes, add up the accumulated values ​​on all trees, and get the feature Total Gini importance of:

[0079] ;

[0080] in, is a node in the tree; Is with the node Related features, It is through the characteristics Split Node After the reduction of impurities; is the first Features; N t’ is the set of all nodes in the t'th tree;

[0081] Furthermore, step S5 is specifically as follows:

[0082] The dashboard includes a three-layer structure. The top layer displays the network health status, including key performance indicators and an overview of overall anomalies. The middle layer displays key indicators and forecast data, including traffic and latency in time series.

[0083] The bottom layer displays specific abnormal events and scheduling results, including detailed abnormal causes and scheduling strategies;

[0084] The display of abnormal detection results lists all abnormal events in a table, including time, node, type, severity, and adds hyperlinks. Click to view detailed abnormal reports. A heat map is used to display the distribution of abnormal events within a certain period of time. The number of abnormal events is aggregated by abnormal type and time period through a bar chart.

[0085] The forecast results are displayed by comparing the historical data of traffic and delay indicators with the forecast data based on the time series graph; the area graph shows the difference between the forecast data and the actual data;

[0086] The changes in resource allocation before and after scheduling are displayed through the network topology diagram, color coding is used to indicate the changes in resource utilization of different nodes, and radar charts are used to compare the comprehensive changes in key performance indicators before and after scheduling.

[0087] A distribution communication network integrated management system comprises a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the distribution communication network integrated management method as described above.

[0088] The present invention has the following beneficial effects:

[0089] 1. The present invention effectively optimizes the resource allocation of 5G networks, improves network utilization and system performance;

[0090] 2. The present invention uses the One-Class SVM model for anomaly detection, effectively identifies abnormal data points outside normal data, and improves the detection accuracy of network anomalies. Clustering algorithm and feature importance analysis: further clustering and feature analysis are performed on the detected abnormal data points;

[0091] 3. The present invention predicts the future network state based on the long short-term memory (LSTM) model, realizes accurate prediction of future network load and performance, and dynamically adjusts the learning rate at different training stages, thereby improving the training efficiency and effect of the model;

[0092] 4. The present invention adopts dynamic 5G slice resource adjustment based on anomaly detection and prediction results, which can not only respond to the current abnormal situation in real time and adjust the bandwidth and priority, but also make resource scheduling plans in advance according to the prediction results, prevent potential network congestion, and improve network performance and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0094] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0095] refer to Figure 1 In this embodiment, a method for integrated management of a power distribution communication network is provided, comprising the following steps:

[0096] S1: Obtain 5G network slicing data, card connection data, and power terminal collection data opened by public network operators, and perform data cleaning, deduplication, and abnormal data processing, convert heterogeneous data into a unified format, and store it;

[0097] S2: Build an anomaly detection model based on the One-Class SVM model, identify abnormal data points based on the anomaly detection model, use clustering algorithms and feature importance analysis to cluster and classify abnormal data points, identify different types of abnormal patterns, and combine network and performance characteristics to determine specific abnormal patterns; network anomalies and performance bottlenecks;

[0098] S3: Build a prediction model based on the LSTM model to predict the future network status;

[0099] S4: Combining the anomaly detection results with the prediction results, the dynamic resource scheduling strategy SD-WAN is used to optimize the 5G slice resource allocation and improve network utilization, as follows:

[0100] Based on the anomaly detection results, the centralized control function of SD-WAN is used to globally schedule network resources to improve resource utilization. The resource allocation of each slice is dynamically adjusted according to the business importance and real-time network status, including bandwidth and priority adjustments:

[0101] Bandwidth adjustment:

[0102]

[0103] Among them, B new is the adjusted bandwidth, B current is the current bandwidth; ΔB is the bandwidth adjustment, calculated based on the abnormal mode;

[0104] Priority Adjustment:

[0105] ;

[0106] Among them, P new is the adjusted priority, P current is the current priority; ΔP is the adjustment of the priority, calculated according to the abnormal pattern;

[0107] According to the prediction results of the prediction model, adjust network resources in advance to prevent potential congestion:

[0108] ;

[0109] in, For the future resource allocation after time units; According to the prediction results Function for resource adjustment; t represents the current time;

[0110] S5: Perform data visualization and create dashboards to display network status, traffic, and performance information.

[0111] In this embodiment, S1 is specifically:

[0112] S11: Obtain the 5G network slice configuration, performance indicators and usage data opened by the operator, and obtain the connection status and traffic data of the SIM card from the operator's connection management system;

[0113] S12: Perform data cleaning, deduplication and abnormal data processing;

[0114] S13: Perform field mapping and format conversion on the processed data from different sources to make them conform to the unified data format specification.

[0115] In this embodiment, a prediction model is constructed based on the LSTM model, as follows:

[0116] A multi-layer LSTM network is used, and the specific structure includes an input layer, several LSTM layers and a Dense layer; a dropout layer is provided after each LSTM layer;

[0117] The LSTM layer is as follows:

[0118] ;

[0119] ;

[0120] ;

[0121] ;

[0122] ;

[0123] ;

[0124] in, is the input feature vector of time step i; is the hidden state at time step i-1; For the Gate of Oblivion; is the input gate; is the candidate value, i.e. the candidate memory cell at the current moment; is the cell state, indicating the memory cell state at the current moment; is the state of the memory cell at the previous moment; is the output gate; is the hidden state at the current moment; , , , They are the weight matrices of the forget gate, input gate, candidate value, and output gate respectively; , , , They are the bias vectors of the forget gate, input gate, candidate value, and output gate respectively; is the activation function;

[0125] Define hyperparameters, including the number of LSTM layers, number of hidden units, learning rate, and batch size;

[0126] During the training process, grid search is used to select the best hyperparameter combination, and K-fold cross validation is used to evaluate the performance of different hyperparameter combinations to obtain the optimal hyperparameter combination and the optimal prediction model.

[0127] In this embodiment, the prediction model dynamically adjusts the learning rate during the training process, as follows:

[0128] According to the optimal hyperparameter combination, the initial learning rate is obtained η initial ; Define the learning rate scheduling parameters, including the initial stage length t initial , the epoch number t at the end of the mid-term mid ; Total number of training epochs t final, Decay rate constant λ, minimum learning rate η min;

[0129] During the training process, the learning rate is dynamically adjusted according to the following formula:

[0130] ;

[0131] in, is the learning rate at the end of the mid-term;

[0132] In the early stage, a learning rate greater than the threshold is used to converge quickly, so a relatively high learning rate is maintained within the initial epochs:

[0133] ;

[0134] In the medium term, according to the exponential decay method, the learning rate is gradually reduced to achieve stable convergence, and the learning rate gradually decreases in an exponential decay manner:

[0135] ;

[0136] In the later stage, fine-tune with a learning rate less than the threshold to achieve the best performance. Gradually reduce the learning rate and use cosine annealing to further reduce the learning rate;

[0137] .

[0138] In this embodiment, an anomaly detection model is constructed based on the One-Class SVM model, as follows:

[0139] Collect network performance indicators and performance indicator data, where network performance indicators include latency, throughput, and packet loss rate; system performance indicators include CPU load, memory usage, and disk I / O; and use SMOTE for oversampling and undersampling to balance data categories;

[0140] Convert time series data into feature vectors and construct training data sets; for example, extract features using a sliding window method;

[0141] The optimization goal of the One-Class SVM model is:

[0142] ;

[0143] Where W is the weight vector of the model, is the bias term, is a hyperparameter that controls the anomaly rate, is the feature mapping function, It is input feature vectors, n is the number of input feature vectors;

[0144] The decision function is:

[0145] ;

[0146] Solve the optimization objective by Lagrange multiplier method:

[0147] ;

[0148] ;

[0149] in, and They are and Lagrange multipliers, is the kernel function; It is input feature vector;

[0150] The One-Class SVM model is trained based on the training data set, pseudo-label generation and model update are performed in combination with semi-supervised learning, and the parameters of the model are adjusted using Bayesian optimization, including ν and kernel function parameter γ; ν controls the percentage of support vectors and the upper limit of training errors; γ is the parameter of the RBF kernel, which controls the distribution of points in high-dimensional space.

[0151] In this embodiment, feature analysis is performed on historical abnormal samples to determine their original feature data, and K-Means is used to classify abnormal samples to identify different types of abnormal patterns:

[0152] ① Initialize k cluster centers {u1,u2,...,u k};

[0153] ②Assign each abnormal sample to the nearest cluster center:

[0154] ;

[0155] in, is the jth abnormal sample identified by the One-Class SVM model, is the cluster center containing abnormal samples; For the Cluster centers;

[0156] ③ Update cluster center:

[0157] ;

[0158] ④ Repeat steps ②-③ until the clustering no longer changes;

[0159] Use the tree model to perform feature importance analysis to determine which original features contribute most to the classification of abnormal samples;

[0160] Cluster analysis and feature importance analysis are combined to determine different types of abnormal patterns. Each abnormal pattern defines a set of feature thresholds and matching conditions, and matches are performed based on the abnormal samples detected by the anomaly detection model.

[0161] The tree model is used to perform feature importance analysis to determine which original features contribute most to the classification of abnormal samples, as follows:

[0162] Train a random forest model:

[0163] ;

[0164] Where T is the number of decision trees in the random forest, Indicates The predicted value of the decision tree for the input sample x;

[0165] For each node in the decision tree , through the relevant features Perform segmentation and calculate the change in Gini impurity before and after segmentation. Calculate the Gini importance of each feature. :

[0166] For each feature , accumulate the Gini impurity reduction on all nodes, add up the accumulated values ​​on all trees, and get the feature Total Gini importance of:

[0167] ;

[0168] in, is a node in the tree; Is with the node Related features, It is through the characteristics Split Node After the reduction of impurities; is the first Features; N t’ is the set of all nodes in the t'th tree;

[0169] In this embodiment, the network is abnormal:

[0170] By analyzing the network characteristics of abnormal samples (such as latency, packet loss rate, bandwidth utilization, etc.), the following patterns can be identified:

[0171] Abnormal delay: The delay time is significantly higher than normal, which may be caused by network congestion or link problems.

[0172] Abnormal packet loss: The packet loss rate increases significantly, which may be caused by hardware failure, line problems or network attacks.

[0173] Bandwidth anomaly: The bandwidth utilization rate increases or decreases abnormally, which may be caused by DDoS attacks or application configuration errors.

[0174] In this embodiment, the performance bottleneck is:

[0175] By analyzing the performance characteristics of abnormal samples (such as CPU utilization, memory usage, disk I / O, etc.), the following patterns are identified:

[0176] CPU bottleneck: The CPU utilization is continuously higher than the threshold, which may be caused by computationally intensive tasks or program dead loops.

[0177] Memory bottleneck: Memory usage is continuously above the threshold, which may be caused by memory leaks or large amounts of data processing.

[0178] I / O bottleneck: Disk I / O or network I / O is continuously higher than normal, which may be caused by a large amount of data reading and writing or a surge in network traffic.

[0179] The detected abnormal samples are matched by predefined abnormal patterns (such as delay, packet loss, etc.), and each abnormal pattern defines a set of feature thresholds and matching conditions.

[0180] In this embodiment, the dashboard includes a three-layer structure. The top layer displays the network health status, including key performance indicators and an overall abnormal overview; the middle layer displays the main indicators and forecast data, including traffic and latency in time series; the bottom layer displays specific abnormal events and scheduling results, including detailed abnormal causes and scheduling strategies;

[0181] The display of abnormal detection results lists all abnormal events in a table, including time, node, type, severity, and adds hyperlinks. Click to view detailed abnormal reports. A heat map is used to display the distribution of abnormal events within a certain period of time. The number of abnormal events is aggregated by abnormal type and time period through a bar chart.

[0182] The forecast results are displayed by comparing the historical data of traffic and delay indicators with the forecast data based on the time series graph; the area graph shows the difference between the forecast data and the actual data;

[0183] The changes in resource allocation before and after scheduling are displayed through the network topology diagram, color coding is used to indicate the changes in resource utilization of different nodes, and radar charts are used to compare the comprehensive changes in key performance indicators before and after scheduling.

[0184] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0185] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0186] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0187] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0188] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for integrated management of a power distribution communication network, characterized in that: The following steps are involved: S1: Obtain 5G network slice data, SIM card connection status, traffic data and power terminal collection data opened by public network operators, and perform data cleaning, deduplication and abnormal data processing, convert heterogeneous data into a unified format, and store it; S2: Build an anomaly detection model based on the One-Class SVM model, identify abnormal data points based on the anomaly detection model, use clustering algorithm and feature importance analysis to cluster and classify abnormal data points, identify different types of abnormal patterns, and determine specific abnormal patterns based on network and performance characteristics; The anomaly detection model is constructed based on the One-Class SVM model, as follows: Collect network performance indicators and system performance indicators, including latency, throughput, and packet loss rate; system performance indicators include CPU load, memory usage, and disk I / O; and use SMOTE for oversampling and undersampling to balance data categories; Convert time series data into feature vectors and construct training data sets; The optimization goal of the One-Class SVM model is: ; Where W is the weight vector of the model, is the bias term, is a hyperparameter that controls the anomaly rate, is the feature mapping function, It is input feature vectors, n is the number of input feature vectors; The decision function is: ; Solve the optimization objective by Lagrange multiplier method: ; ; in, and They are and Lagrange multipliers, is the kernel function; It is input feature vector; The One-Class SVM model is trained based on the training data set, pseudo-label generation and model update are performed using semi-supervised learning, and the model parameters are adjusted using Bayesian optimization. S3: Build a prediction model based on the LSTM model to predict the future network status, as follows: A multi-layer LSTM network is used, and the specific structure includes an input layer, several LSTM layers and a Dense layer; a dropout layer is provided after each LSTM layer; The LSTM layer is as follows: ; ; ; ; ; ; in, is the input feature vector at the current moment; is the hidden state at the previous moment; For the Gate of Oblivion; is the input gate; is the candidate value, i.e. the candidate memory cell at the current moment; is the cell state, indicating the memory cell state at the current moment; is the state of the memory cell at the previous moment; is the output gate; is the hidden state at the current moment; , , , They are the weight matrices of the forget gate, input gate, candidate value, and output gate respectively; , , , They are the bias vectors of the forget gate, input gate, candidate value, and output gate respectively; is the activation function; Define hyperparameters, including the number of LSTM layers, number of hidden units, learning rate, and batch size; During the training process, grid search is used to select the best hyperparameter combination, and K-fold cross validation is used to evaluate the performance of different hyperparameter combinations to obtain the optimal hyperparameter combination and the optimal prediction model. S4: Combining the anomaly detection results with the prediction results, the dynamic resource scheduling strategy SD-WAN is used to optimize the 5G slice resource allocation and improve network utilization, as follows: Based on the anomaly detection results, the centralized control function of SD-WAN is used to globally schedule network resources to improve resource utilization. The resource allocation of each slice is dynamically adjusted according to the business importance and real-time network status, including bandwidth and priority adjustments: Bandwidth adjustment: Among them, B new is the adjusted bandwidth, B current is the current bandwidth; ΔB is the bandwidth adjustment amount, which is calculated based on the anomaly detection result; Priority Adjustment: ; Among them, P new is the adjusted priority, P current is the current priority; ΔP is the bandwidth priority, which is calculated based on the anomaly detection results; According to the prediction results of the prediction model, adjust network resources in advance to prevent potential congestion: ; in, For the future resource allocation after time units; According to the prediction results Function for resource adjustment; t represents the current time; S5: Perform data visualization and create dashboards to display network status, traffic, and performance information.

2. A method for integrated management of a power distribution communication network according to claim 1, characterized in that: The S1 is specifically: S11: Obtain the 5G network slice configuration, performance indicators and usage data opened by the operator, and obtain the connection status and traffic data of the SIM card from the operator's connection management system; S12: Perform data cleaning, deduplication and abnormal data processing; S13: Perform field mapping and format conversion on the processed data from different sources to make them conform to the unified data format specification.

3. A method for integrated management of a power distribution communication network according to claim 1, characterized in that: The prediction model dynamically adjusts the learning rate during the training process, as follows: According to the optimal hyperparameter combination, the initial learning rate is obtained η initial ;definition Learning rate scheduling parameters, including the initial stage length t initial , the epoch number t at the end of the mid-term mid ; Total number of training epochs t final, Decay rate constant λ, minimum learning rate η min; During the training process, the learning rate is dynamically adjusted according to the following formula: ; in, is the learning rate at the end of the mid-term; In the early stage, a learning rate greater than the threshold is used to converge quickly, so a relatively high learning rate is maintained within the initial epochs: ; In the medium term, according to the exponential decay method, the learning rate is gradually reduced to achieve stable convergence, and the learning rate gradually decreases in an exponential decay manner: ; In the later stage, fine-tuning is performed using a learning rate less than the threshold to achieve the best performance, and the learning rate is gradually reduced, and the cosine annealing method is used to further reduce the learning rate; 。 4. A method for integrated management of a power distribution communication network according to claim 1, characterized in that: The clustering algorithm and feature importance analysis are used to cluster and classify abnormal data points, identify different types of abnormal patterns, and determine specific abnormal patterns in combination with network and performance characteristics, as follows: Perform feature analysis on historical abnormal samples to determine their original feature data, use K-Means to classify abnormal samples, and identify different types of abnormal patterns: ① Initialize k cluster centers {u1,u2,...,u k }; ②Assign each abnormal sample to the nearest cluster center: ; in, The jth abnormal sample identified by the One-Class SVM model; is the cluster center containing abnormal samples; For the Cluster centers; ③ Update cluster center: ; ④ Repeat steps ②-③ until the clustering no longer changes; Use the tree model to perform feature importance analysis to determine which original features contribute most to the classification of abnormal samples; Cluster analysis and feature importance analysis are combined to determine different types of abnormal patterns. Each abnormal pattern defines a set of feature thresholds and matching conditions, and matches are performed based on the abnormal samples detected by the anomaly detection model.

5. A method for integrated management of a power distribution communication network according to claim 4, characterized in that: The tree model is used to perform feature importance analysis to determine which original features contribute most to the classification of abnormal samples, as follows: Train a random forest model: ; Where T is the number of decision trees in the random forest, Indicates The predicted value of the decision tree for the input sample x; For each node in the decision tree , through the relevant features Perform segmentation and calculate the change in Gini impurity before and after segmentation. Calculate the Gini importance of each feature. : For each feature , accumulate the Gini impurity reduction on all nodes, add up the accumulated values ​​on all trees, and get the feature Total Gini importance of: ; in, is a node in the tree; Is with the node Related features, It is through the characteristics Split Node After the reduction of impurities; is the first Features; N t’ is the set of all nodes in the t'th tree.

6. A method for integrated management of a power distribution communication network according to claim 1, characterized in that: The step S5 is specifically as follows: The dashboard includes a three-layer structure. The top layer displays the network health status, including key performance indicators and an overview of overall anomalies. The middle layer displays key indicators and forecast data, including traffic and latency in time series. The bottom layer displays specific abnormal events and scheduling results, including detailed abnormal causes and scheduling strategies; The display of abnormal detection results lists all abnormal events in a table, including time, node, type, severity, and adds hyperlinks. Click to view detailed abnormal reports. A heat map is used to display the distribution of abnormal events within a certain period of time. The number of abnormal events is aggregated by abnormal type and time period through a bar chart. The forecast results are displayed by comparing the historical data of traffic and delay indicators with the forecast data based on the time series graph; the area graph shows the difference between the forecast data and the actual data; The changes in resource allocation before and after scheduling are displayed through the network topology diagram, color coding is used to indicate the changes in resource utilization of different nodes, and radar charts are used to compare the comprehensive changes in key performance indicators before and after scheduling.

7. A power distribution communication network integrated management system, characterized in that: It includes a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically executes the steps in the integrated management method of a distribution communication network as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Large-scale power data anomaly detection method based on ridge regression-k-means clustering-LOF-LSTM fusion algorithm

    CN117633688A

  • Wireless communication network fault prediction method and system

    CN118075791A