Adversarial strategy generation method based on big data

Abstract

The invention relates to the technical field of big data processing, and discloses a big data-based confrontation strategy generation method, which comprises the following steps of: collecting multi-source data which comprises historical behavior data, environmental influence data and response data; the method comprises the following steps: performing data cleaning, feature extraction and anomaly recognition on collected multi-source data to generate a preliminary data feature set; historical behavior data, environmental influence data and response data are comprehensively integrated through collection and processing of multi-source data, a preliminary data feature set based on feature extraction is established, and strategy generation and pre-evaluation are performed by using a machine learning algorithm. Through the multi-dimensional data support, the generated confrontation strategy is more targeted and can better adapt to different application environments and real-time requirements, meanwhile, strategy parameters can be dynamically adjusted through calculation and optimization analysis of strategy effect coefficients, and the accuracy and applicability of the strategy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and particularly to a method for generating an adversarial strategy based on big data. Background Art

[0002] With the development of informatization and intelligence, big data technology has been widely applied to decision-making and strategy generation in various complex environments. Especially in adversarial scenarios and dynamic environments, generating real-time adjusted adversarial strategies through the analysis of big data has become a key technical means to cope with changing scenarios. However, existing methods for generating adversarial strategies mostly adopt single data sources or static data models, and it is difficult to dynamically adapt to strategy requirements in a multi-dimensional environment. Especially in complex and rapidly changing application environments, there are various deficiencies in the generation and optimization of adversarial strategies.

[0003] For example, traditional methods for generating adversarial strategies lack the ability to integrate multi-source data in data processing. They mainly rely on single-type data input (such as historical behavior data or environmental data), and it is difficult to obtain comprehensive feature information. This single data dependence makes the generated strategies less adaptable to different application environments, and it is difficult to cover factors that have important impacts on strategy generation and optimization in terms of data dimensions. In addition, due to the lack of effective processing of noise data and abnormal data, the adversarial strategies generated by existing methods are easily affected by external interference in complex scenarios, and the stability and reliability of the strategies are poor. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for generating an adversarial strategy based on big data, which solves the problems mentioned in the above background art.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for generating an adversarial strategy based on big data, comprising the following steps:

[0006] Collect multi-source data, where the multi-source data includes historical behavior data, environmental impact data, and response data;

[0007] Perform data cleaning, feature extraction, and anomaly recognition on the collected multi-source data to generate a preliminary data feature set;

[0008] Based on the preliminary data feature set, establish an adversarial strategy pre-evaluation model through a strategy generation model, and calculate the strategy effect coefficient;

[0009] Compare and analyze the strategy effect coefficient with the strategy generation requirements, predict the strategy application effect, optimize and adjust the strategies with defects or risks, and generate the final adversarial strategy.

[0010] Preferably, the historical behavior data includes operation records in different environments, user operation logs, and historical policy response records.

[0011] Preferably, the environmental impact data includes real-time network status, resource consumption, and security threat information.

[0012] Preferably, the response data includes resource load, response time, failure rate, and stability metrics.

[0013] Preferably, the data processing includes elimination of noise data, filling of missing data, and marking of abnormal data, and the feature extraction includes behavior pattern recognition, feature correlation analysis, and high-frequency feature screening.

[0014] Preferably, the policy generation model adopts an adversarial policy generation algorithm based on machine learning algorithms, calculates the policy effect coefficient through reinforcement learning, decision tree, and clustering algorithms, and the calculation formula of the policy effect coefficient is:

[0015] E = w 1 ×F h +w 2 ×F e +w 3 ×F s ;

[0016] where F h is the matching coefficient of the historical behavior data, F e is the adaptation coefficient of the environmental impact data, F s is the efficiency coefficient of the response data, w 1 , w 2 and w 3 are the weight values of each coefficient, and satisfy w 1 +w 2 +w 3 = 1.

[0017] Preferably, the policy optimization steps include the following:

[0018] If the policy effect coefficient is lower than the risk threshold, trigger the optimization algorithm to adjust the policy parameters and regenerate the policy effect coefficient;

[0019] If the policy effect coefficient is higher than the risk threshold but lower than the safety threshold, mark the policy effect coefficient as an inefficient state and generate policy correction suggestions;

[0020] If the policy effect coefficient is higher than the safety threshold, mark the policy as a feasible state and determine it as the final adversarial policy.

[0021] Preferably, the method further includes steps of behavior pattern extraction and policy enhancement based on experimental data to improve the adaptability and response efficiency of the adversarial policy. The steps include:

[0022] During the generation of the adversarial policy, the reinforcement learning unit performs experimental policy simulations on multi-source data to generate a set of policy experimental data under various conditions;

[0023] Based on the causal associations in the set of policy experimental data, extract the behavior patterns that best match the target task conditions to form a policy optimization framework driven by behavior patterns;

[0024] According to the feedback of the behavior patterns, dynamically adjust the policy parameters during the policy generation process to adapt to different task environments and variable conditions, so as to maintain the robustness and application efficiency of the policy in diverse environments.

[0025] The present invention provides an adversarial policy generation method based on big data. It has the following beneficial effects:

[0026] 1. Through the collection and processing of multi-source data, the present invention comprehensively integrates historical behavior data, environmental impact data, and response data, establishes a preliminary data feature set based on feature extraction, and uses machine learning algorithms for policy generation and pre-evaluation. This multi-dimensional data support makes the generated adversarial policy more targeted, better able to adapt to different application environments and real-time requirements. At the same time, by calculating and optimizing the analysis of the policy effect coefficient, it can dynamically adjust the policy parameters, improve the accuracy and applicability of the policy, effectively respond to changes in different environments, and improve the application effect of the policy generation method under changing conditions.

[0027] 2. Through the comparative analysis of the policy effect coefficient and the policy generation requirements, the present invention can achieve a forward-looking prediction of the policy application effect and conduct real-time evaluation of various performances of the policy. Specifically, by calculating and analyzing the policy effect coefficient, the system can timely identify potential defects or risks, thereby predicting the performance of the policy in the actual application environment. This process enables policy generation not to be limited to a single evaluation, but to continuously feedback optimization information by dynamically monitoring the policy effect coefficient.

[0028] 3. During the generation of the adversarial policy, the present invention adds steps of behavior pattern extraction and policy enhancement based on experimental data. The reinforcement learning unit performs experimental policy simulations on multi-source data to generate a set of policy experimental data under various conditions, and uses causal analysis to extract the optimal behavior patterns, enabling the policy generation process to be adaptively adjusted under different conditions, ensuring the robustness of policy generation, making the policy adaptable to a wide range of environmental changes, and thus showing stronger robustness and anti-interference ability in actual applications, effectively avoiding the risk of policy failure under extreme conditions. Brief Description of the Drawings

[0029] Figure 1 This is a flowchart of the method of the present invention. Detailed Embodiments

[0030] Next, in combination with the drawings of the present invention, the technical solutions of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0031] Embodiment:

[0032] Please refer to the attached Figure 1 , the embodiment of the present invention provides a method for generating an adversarial strategy based on big data, including the following steps:

[0033] Collect multi-source data, where the multi-source data includes historical behavior data, environmental impact data, and response data;

[0034] Clean, extract features, and identify anomalies in the collected multi-source data to generate a preliminary data feature set;

[0035] Based on the preliminary data feature set, establish an adversarial strategy pre-evaluation model through a strategy generation model, and calculate the strategy effect coefficient;

[0036] Compare and analyze the strategy effect coefficient with the strategy generation requirements, predict the strategy application effect, optimize and adjust the strategies with defects or risks, and generate the final adversarial strategy.

[0037] Specifically, comprehensive data resources required for generating countermeasure strategies are obtained through multi-source data collection, including historical behavior data, environmental impact data, and response data, to ensure coverage of different dimensions and changing factors, thereby providing comprehensive data support for strategy generation. In the data processing stage, the collected data is systematically cleaned, feature extracted, and anomaly identified to remove noise data, fill in missing data, and mark potential abnormal data, finally generating a preliminary data feature set, laying a reliable data foundation for subsequent strategy generation. Based on this feature set, a strategy generation model is constructed and a countermeasure strategy pre-evaluation model is established to calculate the strategy effect coefficient. Through the calculation and analysis of the strategy effect coefficient, not only can the effect of the strategy be predicted in advance at the data level, but also potential risks or defects can be effectively identified to ensure the applicability and security of the strategy. Finally, the calculated strategy effect coefficient is compared and analyzed with the strategy generation requirements, the application effect of the strategy is evaluated, and targeted optimization and adjustment are carried out, and then the final countermeasure strategy that meets the actual requirements is generated. This method effectively improves the adaptability and stability of countermeasure strategies in complex environments through multi-dimensional data fusion analysis, dynamic strategy evaluation, and optimization process based on the effect coefficient, enabling the strategy generation process to flexibly respond to changing requirements and continuously optimize, thus enhancing the overall effectiveness of countermeasure strategies in diverse application scenarios.

[0038] Historical behavior data includes operation records in different environments, user operation logs, and historical strategy response records.

[0039] Environmental impact data includes real-time network status, resource consumption, and security threat information.

[0040] Response data includes resource load, response time, failure rate, and stability indicators.

[0041] Specifically, historical behavior data includes operation records in different environments, user operation logs, and historical strategy response records. By comprehensively recording the operation of the system in various environments and user interaction behaviors, it can provide more accurate behavior patterns and historical response data support for strategy generation, helping to scientifically evaluate the long-term performance of the strategy. Environmental impact data includes real-time network status, resource consumption, and security threat information. These data can reflect the current external environment and potential risks of the system in real time, providing necessary external context references for countermeasure strategies, enabling the generated strategies to better adapt to dynamic external conditions and ensuring the efficiency and stability of the strategy during execution. Response data includes resource load, response time, failure rate, and stability indicators, which are used to comprehensively monitor the state and performance of the system during operation. It not only helps to identify the bottlenecks and risks of the system but also provides a feedback mechanism for strategy optimization, enabling the generated strategies to be adjusted according to the actual situation of the system, improving the execution effect and reliability of the strategy.

[0042] Data processing includes elimination of noise data, filling of missing data, and marking of abnormal data. Feature extraction includes behavior pattern recognition, feature correlation analysis, and high-frequency feature screening.

[0043] Specifically, the data processing formula is expressed as:

[0044] Elimination of noise data: Let the original data set be D = {x 1 , x 2 , …, x n}, where x i represents a single data point. Define the noise threshold ∈. If the data point x i satisfies |x i - μ| > ∈σ (where μ and σ are the mean and standard deviation of the data respectively), then x i is regarded as noise and eliminated:

[0045] D′ = {x i ∈ D || x i - μ∣ ≤ ∈σ}, where D′ is the data set after eliminating noise.

[0046] Filling of missing data: For the missing value x missing in the data set D′, mean filling or interpolation method can be used for filling. Using mean filling is: x missing = μ D′ , and using interpolation method is:

[0047] Marking of abnormal data: For the abnormal data (outliers) in the remaining data, define the abnormal marking condition |x i - μ| > kσ (where k > ∈) to mark the abnormal points in the data.

[0048] Feature extraction formula is expressed as:

[0049] Behavior pattern recognition: Let the behavior pattern feature vector be f B , and extract the behavior pattern features through clustering method or dimensionality reduction algorithm (such as PCA):

[0050] f B = PCA(or KMeans(D″), where D″ is the data set after data processing.

[0051] Feature correlation analysis: Let the data feature set be {f 1 , f 2 , …, f m}, then the feature correlation analysis can be represented by the correlation coefficient matrix R:

[0052]

[0053] Among them, ρ(f i , f j ) represents the correlation coefficient of feature f i and f j .

[0054] High-frequency feature screening: Let the threshold of high-frequency features be T f . If the frequency of occurrence of a certain feature f i in the dataset satisfies freq(f i ) ≥ T f , then retain this feature:

[0055] F high_freq = {f i | freq(f i ) ≥ T f}, where F high_freq is the set of high-frequency features selected.

[0056] The strategy generation model adopts an adversarial strategy generation algorithm based on machine learning algorithms, calculates the strategy effect coefficient through reinforcement learning, decision tree, and clustering algorithms, and the calculation formula of the strategy effect coefficient is:

[0057] E = w 1 × F h + w 2 × F e + w 3 × F s ;

[0058] Among them, F h is the matching coefficient of historical behavior data, F e is the adaptation coefficient of environmental impact data, F s is the efficiency coefficient of response data, w 1 , w 2 and w 3 are the weight values of each coefficient, and satisfy w 1 + w 2 + w 3 = 1.

[0059] Specifically, the policy effect coefficient is used to quantify the performance of the policy under different environmental conditions, ensuring that the generated adversarial policy can adapt to complex application scenarios. The calculation formula includes multiple core parameters, such as the matching coefficient of historical behavior data, the adaptation coefficient of environmental impact data, and the efficiency coefficient of response data, to comprehensively reflect the effectiveness of the policy in different data dimensions. By introducing the weight values of each coefficient, the policy generation model can flexibly adjust the evaluation focus under different application requirements and environmental conditions. The setting of the weight values satisfies the constraint that the sum is 1 to ensure the balance of data in each dimension, thereby enhancing the overall reliability of the policy effect coefficient.

[0060] The policy optimization steps include the following:

[0061] If the policy effect coefficient is lower than the risk threshold, the optimization algorithm is triggered to adjust the policy parameters and regenerate the policy effect coefficient;

[0062] If the policy effect coefficient is higher than the risk threshold but lower than the safety threshold, the policy effect coefficient is marked as an inefficient state and policy correction suggestions are generated;

[0063] If the policy effect coefficient is higher than the safety threshold, the policy is marked as a feasible state and determined as the final adversarial policy.

[0064] Specifically, through the hierarchical evaluation of the policy effect coefficient, it is ensured that the generated adversarial policy is optimized and effectively adjusted in different risk conditions. When the policy effect coefficient is lower than the preset risk threshold, the optimization algorithm is automatically triggered to adjust the policy parameters to regenerate the policy effect coefficient, enabling the policy to quickly respond to risks and adapt to environmental changes, thereby improving the stability and anti-interference ability of the policy; if the policy effect coefficient is between the risk threshold and the safety threshold, it is marked as an inefficient state and detailed policy correction suggestions are generated to help make the subsequent optimization process more targeted and ensure that the policy meets the required performance standards in the short term; when the policy effect coefficient is higher than the safety threshold, it indicates that the policy has met the application requirements, and it is marked as a feasible state and confirmed as the final adversarial policy. Through the phased policy evaluation and gradual optimization mechanism, the present invention effectively improves the applicability of the adversarial policy, making it have higher reliability and execution efficiency in diverse environments.

[0065] The method further includes steps of behavior pattern extraction and policy enhancement based on experimental data to improve the adaptability and response efficiency of the adversarial policy. The steps include:

[0066] During the generation of the adversarial policy, through the reinforcement learning unit, experimental policy simulations are performed on multi-source data to generate a set of policy experimental data under various conditions;

[0067] Extract the behavior patterns that best match the target task conditions based on the causal associations in the policy experiment data set, and form a policy optimization framework driven by behavior patterns;

[0068] According to the feedback of the behavior patterns, dynamically adjust the policy parameters during the policy generation process to adapt to different task environments and variable conditions, so as to maintain the robustness and application efficiency of the policy in diverse environments.

[0069] Specifically, use the reinforcement learning unit to conduct experimental policy simulations on multi-source data to generate a policy experiment data set containing various environments and conditions, thereby providing richer scenario information and decision-making basis for policy generation. Based on the causal association relationships in this policy experiment data set, the system can extract the behavior patterns that best match the target task conditions and form a policy optimization framework driven by behavior patterns. In this way, policy optimization can accurately focus on the behavior patterns most relevant to the target task, effectively improving the pertinence and adaptability of policy generation. At the same time, according to the real-time feedback of the behavior patterns, dynamically adjust the policy parameters during the policy generation process to cope with the changes in different task environments and variable conditions, so as to maintain the high efficiency of the policy in diverse environments.

[0070] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating adversarial strategies based on big data, characterized in that: The following steps are involved: Collecting multi-source data, wherein the multi-source data includes historical behavior data, environmental impact data, and response data; Perform data cleaning, feature extraction and anomaly identification on the collected multi-source data to generate a preliminary data feature set; Based on the preliminary data feature set, a confrontation strategy pre-evaluation model is established through a strategy generation model to calculate a strategy effect coefficient; Compare and analyze the strategy effect coefficient with the strategy generation requirements, predict the effect of strategy application, optimize and adjust the strategies with defects or risks, and generate the final confrontation strategy.

2. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The historical behavior data includes operation records under different environments, user operation logs and historical policy response records.

3. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The environmental impact data includes real-time network status, resource consumption and security threat information.

4. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The response data includes resource load, response time, failure rate and stability indicators.

5. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The data processing includes the removal of noise data, the filling of missing data and the marking of abnormal data, and the feature extraction includes behavior pattern recognition, feature association analysis and high-frequency feature screening.

6. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The strategy generation model adopts an adversarial strategy generation algorithm based on a machine learning algorithm, and calculates the strategy effect coefficient through reinforcement learning, decision tree and clustering algorithm.

7. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The calculation formula of the strategy effect coefficient is: E=w1×F h +w2×F e +w3×F s ; Among them, F h is the matching coefficient of historical behavior data, F e is the adaptation coefficient of environmental impact data, F s is the efficiency coefficient of the response data, w1, w2 and w3 are the weight values ​​of each coefficient, and w1+w2+w3=1.

8. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The strategy optimization step includes the following: If the strategy effect coefficient is lower than the risk threshold, the optimization algorithm is triggered to adjust the strategy parameters and regenerate the strategy effect coefficient; If the strategy effect coefficient is higher than the risk threshold but lower than the safety threshold, the strategy effect coefficient is marked as inefficient and a strategy correction suggestion is generated; If the strategy effect coefficient is higher than the safety threshold, the strategy is marked as feasible and determined as the final confrontation strategy.

9. The method for generating a confrontation strategy based on big data according to claim 1, characterized in that: The method further includes a behavior pattern extraction and strategy enhancement step based on experimental data to improve the adaptability and response efficiency of the adversarial strategy, the steps including: In the process of generating adversarial strategies, experimental strategy simulation is performed on multi-source data through reinforcement learning units to generate a set of strategy experimental data under various conditions; Based on the causal relationship in the strategy experiment data set, the behavior pattern that best matches the target task conditions is extracted to form a strategy optimization framework driven by the behavior pattern; According to the feedback of behavioral patterns, the policy parameters are dynamically adjusted during the policy generation process to adapt to different task environments and variable conditions, thereby maintaining the robustness and application efficiency of the strategy in diverse environments.

Images