Network security protection method and system

Through multimodal data fusion, heterogeneous feature learning and dynamic model selection, combined with intelligent honeypot and situational perception, the dynamic and performance bottleneck problems of network security protection in the existing technology are solved, and efficient and intelligent threat identification and Defense improves the automation level and response speed of network security.

CN119011238APending Publication Date: 2024-11-22ZHONGTONG SERVICE WANGYING TECH CO LTD
View PDF 0 Cites 14 Cited by

Patent Information

Application Number
CN202411085334.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing network security protection technologies are difficult to deal with dynamically changing network environments and new threats. Static defense strategies are ineffective when facing unknown threats, and performance bottlenecks are serious when processing large-scale multi-source heterogeneous data, resulting in threat identification accuracy and response speed. insufficient.

Method used

Using multimodal data fusion, heterogeneous feature learning, dynamic model selection and fusion technology, combined with intelligent honeypot and situational perception, to capture feature correlation through self-attention mechanism and graph neural network, automated threat analysis and early warning are realized, and dynamic adjustments are achieved. Protection strategy.

Benefits of technology

It improves the accuracy and dimension of threat recognition, achieves a deep understanding of complex network behavior, enhances the system's intelligent decision-making ability and automation level, can quickly and accurately identify and defend against various types of network threats, and reduces the probability of security incidents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119011238A_ABST
    Figure CN119011238A_ABST
Patent Text Reader

Abstract

The invention discloses a network security protection method and system, and relates to the technical field of network security, and the protection method comprises the following specific steps: S100, multi-modal data fusion: collecting network flow data and log files, and integrating multi-source data including DNS query records, user behavior logs and social media information, thereby forming a multi-modal data set, the data fusion technology is utilized, feature level and decision level fusion strategies are combined, the dimensionality and precision of threat recognition are comprehensively improved, the characteristics of network threats are comprehensively captured from multiple dimensionalities and levels, deep understanding and precise recognition of a complex network behavior mode are achieved, and through a self-attention mechanism and a graph neural network technology, the network threat recognition precision is improved. The complex relevance between the features and the complementarity of the cross-modal features are effectively captured, the accuracy and dimensionality of threat identification are improved, the probability of occurrence of security events is effectively reduced, and the security and stability of network resources are guaranteed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of network security technology, and in particular to a network security protection method and system. Background Art

[0002] With the rapid development of the Internet, the network has become an indispensable part of people's daily life and work. However, with the popularization and widespread application of the network, network security issues have become increasingly prominent. Network attack methods are constantly evolving, and attackers use various technical means to infiltrate and destroy target systems, causing huge economic losses and information leakage. Faced with increasingly complex and changeable network attack methods, traditional static defense strategies seem to be powerless and difficult to effectively respond to various new threats.

[0003] Current network security protection technology mainly relies on rule matching and signature library updates. This static defense strategy is powerless in the face of dynamically changing network environments and evolving attack methods. On the one hand, although the rule matching method can effectively intercept known threats, it is helpless against unknown threats. On the other hand, signature library updates often lag behind the occurrence of attacks and cannot respond to new attack modes in a timely manner. In addition, traditional methods have obvious performance bottlenecks when processing large-scale, multi-source heterogeneous data, and cannot efficiently perform threat analysis and early warning. The existence of these problems greatly limits the effectiveness and response speed of existing protection technologies.

[0004] However, due to the lack of autonomous learning and adaptability, existing protection systems are difficult to maintain high efficiency and accuracy when attack modes change frequently. When dealing with complex network behaviors, traditional systems can often only analyze based on single-dimensional features, ignoring the interaction between cross-modal features, resulting in a significant reduction in the accuracy and comprehensiveness of threat identification. In addition, under the existing technical framework, the adjustment of security policies usually relies on manual intervention, which not only increases operation and maintenance costs, but also reduces the timeliness and effectiveness of responding to sudden security incidents. Therefore, developing a method that can overcome the above limitations and achieve intelligent, dynamic and efficient network security protection has become an urgent problem that needs to be solved. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the prior art and to provide a network security protection method and system. It can integrate multimodal data fusion, heterogeneous feature learning, dynamic model selection and fusion technology. The network security protection method and system can automatically learn and identify abnormal behavior patterns in the network without excessive human intervention, thereby improving the automation level of security protection. The introduction of an intelligent model library and a dynamic policy control module enables the system to adaptively select and combine optimal models for threat analysis based on real-time monitoring data and changes in the network environment, thereby enhancing the system's intelligent decision-making ability.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a network security protection method, the specific steps of the protection method are:

[0007] S100, multimodal data fusion: collects network traffic data and log files, and integrates multi-source data, including DNS query records, user behavior logs, and social media information, to form a multimodal data set. It uses data fusion technology, combined with feature-level and decision-level fusion strategies, to comprehensively improve the dimension and accuracy of threat identification.

[0008] S200, heterogeneous feature learning: Design heterogeneous feature learning technology for features from different sources and types, and use self-attention mechanism to effectively capture the complex correlation between features, especially the complementarity of cross-modal features, so that the model can understand the network behavior pattern at a deeper level;

[0009] S300, dynamic model selection and fusion: implements the model warehouse management mechanism, built-in pre-trained multiple machine learning models, adaptively selects and combines the most appropriate models for threat analysis based on the characteristics of real-time monitoring data and the specific needs of the current network environment, and continuously evaluates model performance through online learning strategies to ensure the best protection effect;

[0010] S400, active bait and honeypot technology: Combining the concept of active defense, we design and deploy intelligent honeypot technology to attract and mislead potential attackers by dynamically generating virtual services and simulating real user behaviors. At the same time, we collect attack behavior data, provide the model with more abundant abnormal samples, and accelerate the learning and evolution process of the model.

[0011] S500, Context Awareness and Behavior Prediction: Introduces context awareness to comprehensively analyze network context information, enhances the model's ability to understand network behavior, and predicts network behavior to identify potential attack trends and patterns in advance;

[0012] S600, interactive security dashboard and visualization: Design a user interactive interface to provide a visual display of real-time security situation, including threat maps, risk heat maps, and traffic anomaly trend analysis, allowing security administrators to intuitively understand network status and quickly locate problems. It also supports customized alarm rules and report export functions.

[0013] Furthermore, the S100 includes:

[0014] S101, network data collection: collect network traffic data and log files from multiple data sources, including network devices, servers, terminal devices, and security devices. The data types include network packets, system logs, application logs, user behavior logs, DNS query records, and social media information;

[0015] S102, multi-source data integration: Integrate data from different sources to extract key features for identifying network threats, including network traffic features: traffic size, protocol type, source IP address, destination IP address, port number, user behavior features: login time, access path, operation type, time series features: analyze the access frequency and time interval of the data time dimension features;

[0016] S103, feature-level fusion: combining features from different data sources to form a comprehensive feature vector, and using the feature selection algorithm mutual information MI to select the most representative features;

[0017] S104, decision-level fusion: Summarize the preliminary classification and judgment results of each data source and make a final judgment through weighted average.

[0018] Furthermore, the feature-level fusion in S103 is used to measure the correlation and shared information between different modal data. In multimodal data fusion, the mutual information between different modal data is calculated to understand the degree of dependence between these modal data, thereby guiding the selection of features and determining the weight of inter-modal fusion. The algorithm formula is: I(X;Y) is the mutual information between feature X and target variable Y, p(x,y) is the joint probability distribution, and p(x) and p(y) are marginal probability distributions.

[0019] Furthermore, the decision-level fusion in S104 obtains a comprehensive result by weighted averaging the data of different modalities through a weighted averaging method, standardizes and normalizes the data of each modality, and calculates the weighted average of the feature values ​​of these modalities according to the reliability and importance of the data of each modality, so as to obtain the fused feature representation, that is, X i is the modal data, ω i The weights are used to reflect the importance of different modal data for a specific task.

[0020] Furthermore, the S200 extracts network security-related features from a multimodal data set, including network traffic data, log files, DNS query records, user behavior logs, and social media information, and classifies the features according to their sources and types, and divides the features into network traffic features, user behavior features, social media features, and time series features. The self-attention mechanism is used to capture the dependencies between the features, and the information at different positions in the input sequence is used to calculate a new feature representation based on the information, namely: Q is the query matrix, K is the key matrix, V is the value matrix, d kis the dimension of the key matrix, and the softmax function is used to calculate the attention weights, which fuses the features learned through the self-attention mechanism to form a more comprehensive and accurate feature representation.

[0021] Furthermore, the S300 includes:

[0022] S301, model warehouse management: Establish a model warehouse to store a variety of pre-trained machine learning models, including classification models, clustering models, and anomaly detection models. Each model in the warehouse contains metadata about its training data, hyperparameters, and performance indicators;

[0023] S302, real-time data monitoring: monitor the real-time network traffic and user behavior data, extract the characteristics of the current network environment, and dynamically evaluate the changes in the current network environment based on the characteristics of the real-time data;

[0024] S303, model selection: according to the characteristics of real-time data and the specific requirements of the current network environment, the most appropriate model is selected from the model warehouse, and the applicability of each model is evaluated by a ranking algorithm based on performance indicators, and the model with the best performance is selected;

[0025] S304, model fusion: For complex threat detection tasks, it is necessary to combine multiple models for analysis. Model fusion technology is used to fuse the output results of multiple models to improve the accuracy and robustness of threat detection;

[0026] S305, online learning and model updating: Implement online learning mechanism, continuously update model parameters according to real-time data, adapt to changes in network environment, regularly evaluate model performance, and automatically trigger model retraining and updating when performance degradation is found;

[0027] S306, model performance evaluation: Use multiple performance indicators, including accuracy, recall, and F1 score, to evaluate the model to ensure its effectiveness in practical applications, record and analyze the performance changes of the model, and adjust the model selection and fusion strategy in a timely manner.

[0028] Furthermore, S303 uses a ranking algorithm based on performance indicators to sort the models according to their performance indicators on the training data and the validation data, selects the model with the best performance, and evaluates the performance of each model on the training data and the validation data, including accuracy, precision, recall, F1 score, and mean square error. The algorithm formula is: Score(M i )=α·Accuracy(M i )+β·Recall(M i )+γ·F1(M i ), Score(Mi) is the model M iThe comprehensive score of α, β, and γ are weight coefficients. Accuracy(M i )、Recall(M i )、Recall(M i )、F1(M i ) are the accuracy, recall and F1 score of the model respectively. The models are sorted according to the evaluation results of the nature of the problem and the business objectives. The higher the performance of the model, the higher the ranking. The sorting includes single indicator sorting and multi-indicator comprehensive sorting. One or more models are selected from the sorted model list for subsequent fusion and direct deployment. The selection is based on the ranking, diversity, computational cost and interpretability of the model. If multiple models are selected, model fusion is required.

[0029] Furthermore, the model fusion technology is used in S304 to monitor the characteristics of the data and the specific needs of the current network environment in real time, select multiple machine learning models for threat analysis, evaluate the selected models, determine their performance and credibility, use the simple averaging method to fuse the prediction results of the selected models, and average the prediction results of the multiple models in an unbiased manner, that is: y avg is the final prediction result, n is the number of models, and the final prediction result is obtained as the output of network security protection

[0030] A network security protection system, the protection system comprising the following components:

[0031] Multi-source intelligence convergence engine module: collects data from various data sources, including network traffic, log files, DNS records, user behavior logs, and social media information, and uses data fusion technology to integrate these multi-source data into a unified multi-modal data set to ensure that the characteristics of network threats are captured from multiple angles and levels;

[0032] Cross-domain feature parsing module: It is designed to process data features from different sources and types. It uses a self-attention mechanism to learn and understand these heterogeneous features, discover and mine the complementarity and complex relationships between features of different modalities, and thus build a deep understanding of network behavior patterns.

[0033] Intelligent model library and dynamic policy control module: maintains a model warehouse containing a variety of pre-trained machine learning models. It automatically selects and combines the optimal model for threat analysis based on the characteristics of real-time monitoring data and changes in the network environment. It also has online learning capabilities and can continuously evaluate model performance to ensure optimal protection effects.

[0034] Active defense simulation sandbox module: Integrates intelligent honeypot technology to dynamically generate virtual services and simulate real user behaviors to attract potential attackers. It also collects attack behavior data, increases the amount of data for model training, and accelerates the model learning process.

[0035] Environmental Perception Prediction Center Module: Uses situational awareness to comprehensively analyze network environment context, predict network behavior trends, identify potential threats in advance, and enhance the system's early warning capabilities;

[0036] Security situation visualization control module: provides an interactive user interface to display security situation in real time, namely threat maps, risk heat maps and traffic anomaly trends, helping security personnel to quickly understand network status, locate problems, formulate strategies, and support personalized alarm settings and report output;

[0037] Threat intelligence collaborative learning module: responsible for integrating external threat intelligence, expanding model training data, and ensuring that system protection strategies keep pace with the times and adapt to the ever-changing network security threat environment through a continuous learning mechanism.

[0038] Furthermore, the honeypot technology in the active defense simulation sandbox module is used to deploy virtualized network services, namely Web servers, databases, and file sharing, to simulate a real network environment to attract attackers, create an isolated virtual environment, and record all operations of the attacker in detail, including network connections, command execution, and file access. It analyzes the attacker's behavior patterns, identifies common attack methods and tools, and converts the collected attack data into threat intelligence to help improve protection strategies.

[0039] Compared with the prior art, this network security protection method and system has the following beneficial effects:

[0040] 1. The present invention achieves a deep understanding and accurate identification of complex network behavior patterns by comprehensively capturing the characteristics of network threats from multiple dimensions and levels. Especially when processing large-scale, heterogeneous data, the self-attention mechanism and graph neural network technology can effectively capture the complex correlations between features and the complementarity of cross-modal features, which greatly improves the accuracy and dimension of threat identification, thereby being able to more quickly and accurately identify and defend against various known and unknown network threats, effectively reducing the probability of security incidents, and ensuring the security and stability of network resources.

[0041] 2. The present invention can adaptively select and combine the most appropriate models for threat analysis based on the characteristics of real-time monitoring data and the specific needs of the current network environment through the intelligent model library and dynamic policy control module, and continuously evaluate the model performance through online learning strategies to ensure the best protection effect. In addition, the system also has a continuous learning mechanism that can integrate external threat intelligence and expand model training data to ensure that the system protection strategy keeps pace with the times and adapts to the ever-changing network security threat environment. This dynamic adaptability and intelligent decision-making capability significantly improves the automation level of network security protection and provides strong technical support for network security protection.

[0042] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 The present invention is an operation flow chart of a network security protection method.

[0045] Figure 2 The figure is a flow chart of a network security protection system. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Embodiment 1

[0048] A network security protection method, the method comprising the following steps: multimodal data fusion, heterogeneous feature learning, dynamic model selection and fusion, active bait and honeypot technology, situational awareness and behavior prediction, interactive security dashboard and visualization.

[0049] First, enter the multimodal data fusion stage (S100). When implementing network security protection, multimodal data fusion is first performed. Network traffic data and log files are collected through multiple channels such as network devices, servers, and terminal devices, and multi-source data such as DNS query records, user behavior logs, and social media information are integrated. In the data integration process, key features including traffic size, protocol type, source IP address, user login time, etc. are extracted, and the feature-level fusion strategy is used: The features of different data sources are combined to form a comprehensive feature vector. At the same time, through decision-level fusion: We summarize the preliminary classification and judgment results of various data sources and make a final judgment.

[0050] Next, the heterogeneous feature learning stage (S200) is entered. After obtaining the multimodal data set, the heterogeneous feature learning technology is adopted to use the self-attention mechanism to capture the complex correlation between features, that is, the self-attention mechanism is used to capture the dependency between features, the information at different positions in the input sequence, and the new feature representation is calculated based on this information, that is: The features learned through the self-attention mechanism are fused to form a more comprehensive and accurate feature representation. In particular, attention is paid to the complementarity of cross-modal features to deeply understand the network behavior patterns. The attention weights of the query matrix, key matrix and value matrix are calculated through the self-attention mechanism, and the learned features are fused to form a more comprehensive and accurate feature representation.

[0051] Then, the dynamic model selection and fusion stage (S300) is entered. In order to realize real-time analysis of network threats, a model warehouse management mechanism is established, and a variety of pre-trained machine learning models are built in. According to the characteristics of real-time monitoring data and the specific needs of the current network environment, the most suitable model is adaptively selected and combined for threat analysis. At the same time, the performance indicators of the models on the training data and the verification data are sorted by a sorting algorithm based on performance indicators, and the model with the best performance is selected. The performance of each model on the training data and the verification data is evaluated, including accuracy, precision, recall rate, F1 score, and mean square error. The algorithm formula is: Score(M i )=α·Accuracy(M i )+β·Recall(M i )+γ·F1(M i), sort the models according to the evaluation results of the nature of the problem and the business objectives. The higher the performance of the model, the higher the ranking. The sorting includes single indicator sorting and multi-indicator comprehensive sorting. Select one or more models from the sorted model list for subsequent fusion and direct deployment. The selection is based on the ranking, diversity, computational cost and interpretability of the model. If multiple models are selected, model fusion is required.

[0052] Subsequently, we entered the active bait and honeypot technology stage (S400). In order to further enhance security protection capabilities, we designed and deployed intelligent honeypot technology in combination with the active defense concept. By dynamically generating virtual services and simulating real user behaviors, we successfully attracted potential attackers and collected attack behavior data. These data provided the model with richer abnormal samples, accelerating the learning and evolution process of the model.

[0053] Secondly, enter the situational awareness and behavior prediction stage (S500), using situational awareness technology to comprehensively analyze the network environment context, predict network behavior trends, identify potential threats in advance, enhance early warning capabilities, and enable the security team to respond and formulate strategies more quickly.

[0054] Finally, we enter the interactive security dashboard and visualization stage (S600), which provides an interactive security dashboard to display the security situation in real time, including threat maps, risk heat maps, and traffic anomaly trends, so that security personnel can understand the network status and quickly locate problems. At the same time, the model is evaluated through a variety of performance indicators to ensure its effectiveness, and the model selection and fusion strategy are adjusted in time according to performance changes.

[0055] This embodiment provides a specific implementation of a network security protection method. By integrating multimodal data fusion, heterogeneous feature learning, dynamic model selection and fusion technology, it realizes intelligent, dynamic and efficient network threat protection, and provides strong support for network security protection.

[0056] Embodiment 2

[0057] This embodiment describes in detail the specific implementation process of a network security protection system, including a multi-source intelligence aggregation engine module, a cross-domain feature analysis module, an intelligent model library and dynamic policy regulation module, an active defense simulation sandbox module, an environmental perception prediction center module, a security situation visualization control module and a threat intelligence collaborative learning module. It embodies the system's high degree of automation, intelligence and dynamic adaptability, ensures the effectiveness and timeliness of network security protection, and through continuous learning and model updating, the system can adapt to the ever-changing network security threat environment, providing strong technical support for network protection.

[0058] In the specific implementation, first, the multi-source intelligence convergence engine module collects network traffic data and log files from data sources such as network devices, servers, terminal devices, and security devices, including network packets, system logs, application logs, user behavior logs, DNS query records, and social media information. It integrates data from different sources and extracts key features for identifying network threats, such as network traffic features, user behavior features, and time series features. It uses the feature selection algorithm mutual information MI to combine features from different data sources to form a comprehensive feature vector, and screens out the most representative features. It summarizes the preliminary classification and judgment results from each data source, and makes the final judgment through weighted average.

[0059] The features in the multimodal dataset are divided into network traffic features, user behavior features, social media features and time series features. The embedding technology is used to convert high-dimensional features into low-dimensional vector representations for subsequent processing. The self-attention mechanism is used to capture the complex correlations between features, and the attention weights between features are calculated to generate weighted feature representations. Different feature nodes and the relationships between them are represented as graph structures, and the features are learned to capture the complementarity of cross-modal features. The features processed by the self-attention mechanism and graph neural network are fused to form a comprehensive feature representation.

[0060] The intelligent model library and dynamic policy control module establishes a model warehouse to store a variety of pre-trained machine learning models, including classification models, clustering models, and anomaly detection models. Each model in the warehouse contains metadata of its training data, hyperparameters, and performance indicators. It monitors real-time network traffic and user behavior data, extracts the characteristics of the current network environment, and dynamically evaluates changes in the current network environment based on the characteristics of real-time data. According to the characteristics of real-time data and the specific needs of the current network environment, the most appropriate model is selected from the model warehouse. A ranking algorithm based on performance indicators is used to evaluate the applicability of each model and select the model with the best performance. For complex threat detection tasks, it may be necessary to combine multiple models for analysis. Model fusion technology is used to fuse the output results of multiple models to improve the accuracy and robustness of threat detection, implement an online learning mechanism, continuously update model parameters based on real-time data, adapt to changes in the network environment, regularly evaluate model performance, and automatically trigger model retraining and updating when performance degradation is found.

[0061] Deploy smart honeypot technology in the network to simulate real network environments and services, such as Web servers, databases, and file sharing. Create an isolated virtual environment through virtualization technology (such as virtual machines and containers) to ensure that attack behaviors will not affect the actual production system. Design a variety of bait strategies to attract different types of attackers. For example, simulate vulnerable Web applications and open common weak password accounts. Smart honeypot technology monitors all attacker operations in real time, records attackers' network connections, command executions, file accesses, and other behaviors in detail, analyzes attackers' behavior patterns, identifies common attack methods and tools, and converts the collected attack data into threat intelligence.

[0062] The situational awareness and behavior prediction module introduces situational awareness to comprehensively analyze network context information, enhance the model's ability to understand network behavior, predict network behavior, identify potential attack trends and patterns in advance, and provide an interactive user interface to display security status in real time, such as threat maps, risk heat maps, and traffic anomaly trends, to help quickly understand network conditions, locate problems and formulate strategies. The threat intelligence collaborative learning module integrates external threat intelligence, expands model training data, and ensures that protection strategies keep pace with the times through a continuous learning mechanism to adapt to changes in the network security threat environment.

[0063] According to the above implementation steps, comprehensive monitoring and precise defense of network threats can be achieved. Multimodal data fusion and heterogeneous feature learning technologies improve the dimension and accuracy of threat identification. Dynamic model selection and fusion ensure the best protection effect. Active bait and honeypot technology enhance active defense capabilities. Situational awareness and behavior prediction identify potential attack trends and patterns in advance. Interactive security dashboards and visualization modules provide security administrators with intuitive and convenient management tools.

[0064] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A network security protection method, characterized in that: The specific steps of this protection method are: S100, multimodal data fusion: collects network traffic data and log files, and integrates multi-source data, including DNS query records, user behavior logs, and social media information, to form a multimodal data set. It uses data fusion technology, combined with feature-level and decision-level fusion strategies, to comprehensively improve the dimension and accuracy of threat identification. S200, heterogeneous features Learning: Design heterogeneous feature learning technology for features from different sources and types, and use self-attention mechanism to effectively capture the complex correlations between features, especially the complementarity of cross-modal features, so that the model can understand network behavior patterns at a deeper level; S300, dynamic model selection and fusion: implements the model warehouse management mechanism, built-in pre-trained multiple machine learning models, adaptively selects and combines the most appropriate models for threat analysis based on the characteristics of real-time monitoring data and the specific needs of the current network environment, and continuously evaluates model performance through online learning strategies to ensure the best protection effect; S400, active bait and honeypot technology: Combining the concept of active defense, we design and deploy intelligent honeypot technology to attract and mislead potential attackers by dynamically generating virtual services and simulating real user behaviors. At the same time, we collect attack behavior data, provide the model with more abundant abnormal samples, and accelerate the learning and evolution process of the model. S500, Context Awareness and Behavior Prediction: Introduces context awareness to comprehensively analyze network context information, enhances the model's ability to understand network behavior, and predicts network behavior to identify potential attack trends and patterns in advance; S600, interactive security dashboard and visualization: Design a user interactive interface to provide a visual display of real-time security situation, including threat maps, risk heat maps, and traffic anomaly trend analysis, allowing security administrators to intuitively understand network status and quickly locate problems. It also supports customized alarm rules and report export functions.

2. A network security protection method according to claim 1, characterized in that: The S100 includes: S101, network data collection: collect network traffic data and log files from multiple data sources, including network devices, servers, terminal devices, and security devices. The data types include network packets, system logs, application logs, user behavior logs, DNS query records, and social media information; S102, multi-source data integration: Integrate data from different sources to extract key features for identifying network threats, including network traffic features: traffic size, protocol type, source IP address, destination IP address, port number, user behavior features: login time, access path, operation type, time series features: analyze the access frequency and time interval of the data time dimension features; S103, feature-level fusion: combining features from different data sources to form a comprehensive feature vector, and using the feature selection algorithm mutual information MI to select the most representative features; S104, decision-level fusion: Summarize the preliminary classification and judgment results of each data source and make a final judgment through weighted average.

3. A network security protection method according to claim 1, characterized in that: The S103 feature-level fusion is used to measure the correlation and shared information between different modal data. In multimodal data fusion, the mutual information between different modal data is calculated to understand the degree of dependence between these modal data, thereby guiding the selection of features and determining the weight of inter-modal fusion. The algorithm formula is: I(X;Y) is the mutual information between feature X and target variable Y, p(x,y) is the joint probability distribution, and p(x) and p(y) are marginal probability distributions.

4. A network security protection method according to claim 3, characterized in that: The decision-level fusion in S104 obtains a comprehensive result by weighted averaging the data of different modes through the weighted averaging method, standardizes and normalizes the data of each mode, and calculates the weighted average of the feature values ​​of these modes according to the reliability and importance of the data of each mode, so as to obtain the fused feature representation, that is, X i is the modal data, ω i The weights are used to reflect the importance of different modal data for a specific task.

5. A network security protection method according to claim 3, characterized in that: The S200 extracts network security-related features from a multimodal data set, including network traffic data, log files, DNS query records, user behavior logs, and social media information. The features are classified according to their sources and types, and the features are divided into network traffic features, user behavior features, social media features, and time series features. The self-attention mechanism is used to capture the dependencies between the features, and the information at different positions in the input sequence is used to calculate a new feature representation based on the information, namely: Q is the query matrix, K is the key matrix, V is the value matrix, d k is the dimension of the key matrix, and the softmax function is used to calculate the attention weights, which fuses the features learned through the self-attention mechanism to form a more comprehensive and accurate feature representation.

6. A network security protection method according to claim 1, characterized in that: The S300 includes: S301, model warehouse management: Establish a model warehouse to store a variety of pre-trained machine learning models, including classification models, clustering models, and anomaly detection models. Each model in the warehouse contains metadata about its training data, hyperparameters, and performance indicators; S302, real-time data monitoring: monitor the real-time network traffic and user behavior data, extract the characteristics of the current network environment, and dynamically evaluate the changes in the current network environment based on the characteristics of the real-time data; S303, model selection: according to the characteristics of real-time data and the specific requirements of the current network environment, the most appropriate model is selected from the model warehouse, and the applicability of each model is evaluated by a ranking algorithm based on performance indicators, and the model with the best performance is selected; S304, model fusion: For complex threat detection tasks, it is necessary to combine multiple models for analysis. Model fusion technology is used to fuse the output results of multiple models to improve the accuracy and robustness of threat detection; S305, online learning and model updating: Implement online learning mechanism, continuously update model parameters according to real-time data, adapt to changes in network environment, regularly evaluate model performance, and automatically trigger model retraining and updating when performance degradation is found; S306, model performance evaluation: Use multiple performance indicators, including accuracy, recall, and F1 score, to evaluate the model to ensure its effectiveness in practical applications, record and analyze the performance changes of the model, and adjust the model selection and fusion strategy in a timely manner.

7. A network security protection method according to claim 6, characterized in that: S303 uses a ranking algorithm based on performance indicators to sort the models according to their performance indicators on the training data and the validation data, selects the model with the best performance, and evaluates the performance of each model on the training data and the validation data, including accuracy, precision, recall, F1 score, and mean square error. The algorithm formula is: Score(M i )=α·Accuracy(M i )+β·Recall(M i )+γ·F1(M i ), Score(Mi) is the model M i The comprehensive score of α, β, and γ are weight coefficients. Accuracy(M i )、Recall(M i )、Recall(M i )、F1(M i ) are the accuracy, recall and F1 score of the model respectively. The models are sorted according to the evaluation results of the nature of the problem and the business objectives. The higher the performance of the model, the higher the ranking. The sorting includes single indicator sorting and multi-indicator comprehensive sorting. One or more models are selected from the sorted model list for subsequent fusion and direct deployment. The selection is based on the ranking, diversity, computational cost and interpretability of the model. If multiple models are selected, model fusion is required.

8. A network security protection method according to claim 6, characterized in that: The S304 uses model fusion technology to monitor the characteristics of data and the specific needs of the current network environment in real time, selects multiple machine learning models for threat analysis, evaluates the selected models, determines their performance and credibility, uses a simple averaging method to fuse the prediction results of the selected models, and performs an unbiased average of the prediction results of multiple models, namely: y avg is the final prediction result, n is the number of models, and the final prediction result is obtained as the output of network security protection.

9. A network security protection system, characterized in that: The protection system comprises the following components: Multi-source intelligence convergence engine module: collects data from various data sources, including network traffic, log files, DNS records, user behavior logs, and social media information, and uses data fusion technology to integrate these multi-source data into a unified multi-modal data set to ensure that the characteristics of network threats are captured from multiple angles and levels; Cross-domain feature parsing module: It is designed to process data features from different sources and types. It uses a self-attention mechanism to learn and understand these heterogeneous features, discover and mine the complementarity and complex relationships between features of different modalities, and thus build a deep understanding of network behavior patterns. Intelligent model library and dynamic policy control module: maintains a model warehouse containing a variety of pre-trained machine learning models. It automatically selects and combines the optimal model for threat analysis based on the characteristics of real-time monitoring data and changes in the network environment. It also has online learning capabilities and can continuously evaluate model performance to ensure optimal protection effects. Active defense simulation sandbox module: Integrates intelligent honeypot technology to dynamically generate virtual services and simulate real user behaviors to attract potential attackers. It also collects attack behavior data, increases the amount of data for model training, and accelerates the model learning process. Environmental Perception Prediction Center Module: Uses situational awareness to comprehensively analyze network environment context, predict network behavior trends, identify potential threats in advance, and enhance the system's early warning capabilities; Security situation visualization control module: provides an interactive user interface to display security situation in real time, namely threat maps, risk heat maps and traffic anomaly trends, helping security personnel to quickly understand network status, locate problems, formulate strategies, and support personalized alarm settings and report output; Threat intelligence collaborative learning module: responsible for integrating external threat intelligence, expanding model training data, and ensuring that system protection strategies keep pace with the times and adapt to the ever-changing network security threat environment through a continuous learning mechanism.

10. A network security protection system according to claim 9, characterized in that: The honeypot technology in the active defense simulation sandbox module is used to deploy virtualized network services, namely Web servers, databases, and file sharing, to simulate a real network environment to attract attackers, create an isolated virtual environment, and record all operations of attackers in detail, including network connections, command execution, and file access. It analyzes the attacker's behavior patterns, identifies common attack methods and tools, and converts the collected attack data into threat intelligence to help improve protection strategies.

Citation Information

Cited By

  • Network security protection method, device, equipment, medium and product

    CN119728173A

  • Method and device for detecting threat degree of encrypted traffic and computer equipment

    CN119728189A

  • Methods, devices, and computer equipment for detecting the threat level of encrypted traffic.

    CN119728189B

  • Multi-modal fusion feature-based phishing detection method and system

    CN119814358A

  • Pulse signal monitoring system based on wearable pulse diagnosis bracelet

    CN119896457A