A correlation analysis method for power network alarm information

By designing alarm transaction aggregation and rule filtering algorithms in the power network, combining FP-Growth and dynamic sliding time windows, the problem of massive alarm data screening is solved, and efficient alarm rule generation in the power network is achieved, and security and management efficiency are improved.

CN114281864BActive Publication Date: 2024-11-19SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111554921.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-17
Publication Date
2024-11-19
Estimated Expiration
2041-12-17

AI Technical Summary

Technical Problem

The existing power network alarm information correlation analysis method is difficult to effectively screen out valuable information when facing massive alarm data. The method based on expert knowledge is limited by experience, the rule-based method lacks scalability, and the machine-learning method is Efficient algorithms need to be designed in combination with actual scenarios.

Method used

A method of alarm information correlation analysis for power networks was designed, including alarm transaction aggregation algorithm and alarm rule filtering algorithm, combining preprocessing and classic FP-Growth algorithm, using dynamic sliding time windows and time-based filtering algorithms to generate Accurate alarm rules.

Benefits of technology

It realizes the rapid and accurate generation of power network alarm rules, improves the safety and stability of power network, and improves the troubleshooting and early warning efficiency of managers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114281864B_ABST
    Figure CN114281864B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for correlation analysis of alarm information in power networks, and the specific steps include: pre-processing the alarm logs generated by the safety equipment in the power network to construct alarm information with consistent fields and formats; inputting the information into the alarm transaction construction module to construct a transaction set, including alarm data segmentation and extraction of alarm transaction sets; using the FP‑Growth algorithm to mine association rules for alarm transactions, and designing a rule filtering algorithm based on time sequence to filter the alarm rules to obtain a final rule list. The present invention can quickly and accurately generate association rules for alarm information in power networks, provide a set of detailed alarm rule lists for power grid security personnel, effectively improve the troubleshooting efficiency of power grid security management personnel, and accurately warn of future network attack events.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the intersection field of electric power industry and network security, and in particular relates to a correlation analysis method for electric power network alarm information. Background Art

[0002] In recent years, with the vigorous development of computer technology and information technology, while information network infrastructure has brought convenience to human society, network security issues have become increasingly severe. As an important part of the industrial control field, the importance and significance of ensuring the safe and stable operation of the power network is self-evident.

[0003] At present, the power network security problem is very serious, and the power network situation awareness technology has come into being. The power network situation awareness can fully perceive the vulnerability of the power network and generate alarm information of related equipment in real time. However, in the face of huge amounts of redundant alarm data, if valuable information cannot be effectively screened out, the power network situation awareness will not achieve the expected effect.

[0004] Currently, there are three mainstream directions in the field of alarm information correlation analysis in public literature: methods based on expert knowledge or experience, based on established rules, and based on machine learning.

[0005] (1) Methods based on expert knowledge

[0006] The expert knowledge-based method is mainly used in early power network correlation analysis and troubleshooting. This method can show great advantages when the amount of alarm data is small, and can handle and deal with most network attacks. However, this method is largely limited by the experience and knowledge of experts. With the expansion of the scale of power networks and the sudden increase in alarm information at this stage, the expert knowledge-based method is obviously unable to effectively find the connection between multiple categories of network attacks from massive alarms. Therefore, this method has been basically eliminated or used as an auxiliary means to supplement the analysis.

[0007] (2) Rule-based approach

[0008] Rule-based power network alarm information correlation analysis mainly relies on pre-defined rules. However, this solution lacks scalability, that is, it can only rely on existing rules in the system. The advantage of the rule-based method is that the implementation process is relatively simple, but the formulation of rules always lags behind new attack behaviors, so it is not suitable for the current complex power network environment.

[0009] (3) Machine learning-based methods

[0010] Power network security is the cornerstone of security in the field of industrial control. Power network alarm information analysis not only requires sufficient theoretical support, but also requires real-time analysis and processing to ensure the security of the power network in actual industrial control scenarios. Machine learning has these advantages, so methods based on machine learning are most widely used in this field.

[0011] In summary, the present invention uses a method based on machine learning and combines it with the actual scenarios in the field of power industry control. At the same time, two algorithms are designed for this specific scenario, which can be used to accurately and efficiently generate alarm rules. Summary of the invention

[0012] In view of the above problems, the present invention proposes a correlation analysis method for power network alarm information, designs an alarm transaction aggregation algorithm and an alarm rule filtering algorithm based on real power network scenarios, and combines preprocessing analysis with existing machine learning algorithms for research and analysis, so as to achieve the goal of quickly and accurately generating power network alarm rules.

[0013] In order to achieve the above object, the present invention provides the following technical solutions:

[0014] A correlation analysis method for power network alarm information includes the following steps:

[0015] (1) Normalize the alarm logs collected by the power network situation awareness platform;

[0016] (2) Preprocess and analyze the normalized alarm data;

[0017] (3) The pre-processed alarm data is input into the alarm transaction construction module for segmentation processing and extraction of alarm transaction sets;

[0018] (4) For the transaction set, the classic association analysis algorithm FP-Growth is used to mine alarm rules, and a time-series-based alarm rule filtering algorithm is designed to filter the mined rules to obtain the final rule set.

[0019] As an improvement of the present invention, the specific method of step (2) is as follows:

[0020] (2.1) Select attributes of the original alarm log after normalization in (1) and select some relatively important alarm attribute features;

[0021] (2.2) Clean the alarm data;

[0022] (2.3) Remove duplicate and redundant alarm information and define duplicate alarms as follows:

[0023]

[0024] Among them, alarm i and alarm j Respectively represent the i-th and j-th alarm attack events, including the attributes of alarm name alarm_name, alarm attack target alarm_tgt, and alarm attack source alm_src; alm_time i and alm_time j Respectively represent the occurrence time of the i-th and j-th alarm events; t min Indicates the minimum time interval threshold, which is a custom parameter.

[0025] (2.4) Sort the alarm information in relative time order.

[0026] Furthermore, the relatively important alarm attribute features selected in step 2.1 are shown in the following table:

[0027]

[0028] As an improvement of the present invention, the design method for segmenting and extracting the alarm transaction set in step (3) is as follows:

[0029] (3.1) Segment the sorted alarm sequence set in (2.4), set the segmentation threshold K in advance according to the specific sample size, and divide the alarm sequence into K segments;

[0030] (3.2) For each alarm subsequence, a method based on a dynamic sliding time window (DSTW) is designed to extract alarm transactions, so that alarms with close associations are aggregated into the same transaction set to the greatest extent possible;

[0031] As an improvement of the present invention, the specific method of step (3.2) is as follows:

[0032] (3.2.1) Perform initialization settings, set the original size W of the sliding time window, the sliding step size S, and the minimum time threshold Time_intvl, for example, W = 5, S = 2, Time_intvl = 10min;

[0033] (3.2.2) Slide from the initial position of each subsequence;

[0034] (3.2.3) Under normal circumstances, when the window size reaches W, the alarm events in the window are taken to form a transaction, and the left pointer slides to the right by S steps; when the time distance between any two alarm attack events exceeds a certain time threshold, the window stops sliding and obtains an alarm transaction; at the same time, the sliding start position is adjusted to the next position of the current position, and sliding starts again;

[0035] (3.2.4) Repeat the steps in (3.2.3) until the sliding pointer has slid to the end of each subsequence. Through this setting, the traditional fixed sliding time window algorithm is improved, and the temporal and spatial relationship between each alarm information can be fully considered to maximize the aggregation of closely related alarm information.

[0036] As an improvement of the present invention, the specific method of step (4) is as follows:

[0037] (4.1) For the alarm transaction set, the frequent pattern growth algorithm (FP-Growth) is used to mine the strong association rules of power network alarms;

[0038] (4.2) A time-series-based alarm rule filtering algorithm is proposed, which fully considers the relationship between each alarm event, filters out alarm rules that do not meet the time sequence conditions, and further reduces the number of rules.

[0039] Through this step, the initial alarm rule set is obtained. At the same time, for a part of the set that does not meet the time sequence rules, a designed filtering algorithm is used to filter out the rules that do not meet the time sequence, further compressing the scale of the rules.

[0040] As an improvement of the present invention, the specific sub-steps of step (4.1) are as follows:

[0041] (4.1.1) Set the minimum support threshold minSup and the minimum confidence threshold minConf;

[0042] (4.1.2) Construct frequent pattern tree FP-Tree;

[0043] (4.1.3) Generate frequent itemsets of warning information; this step depends on the specific calculation formulas for support and confidence, as follows:

[0044]

[0045] This formula represents the calculation item set Item i The support degree of Item i represents the i-th item set, T represents the alarm transaction set, q represents the number of alarm transactions, t j Indicates the jth alarm transaction.

[0046] support(X)=α(X) / |T|

[0047] This formula represents the calculation of the support of the item set X, where |T| represents the number of transactions contained in the transaction set T, and α(X) represents the support of the item set X.

[0048]

[0049] This formula calculates the confidence of the rule X->Y, where support(X∪Y) and support(X) represent the support of the item set X∪Y and the item set X, respectively.

[0050] (4.1.4) Generate initial alarm association rules;

[0051] Compared with the prior art, the technical solution of the present invention has the following advantages and beneficial effects:

[0052] (1) The present invention fully considers the actual business scenarios in the power industry network, can quickly and effectively obtain the alarm rules in the power network, and provide a list of alarm rules for power grid managers to assist managers in troubleshooting and future network attack warnings, which can effectively improve the security and stability of the power network.

[0053] (2) The present invention designs an algorithm based on a dynamic sliding time window (Dynamic Sliding Time Window, DSTW), which fully considers the temporal and spatial connections between various alarm events, can efficiently and accurately extract alarm transaction sets, aggregate alarms with high correlation levels in one transaction window to the greatest extent, and improve the accuracy of the alarm rules subsequently mined.

[0054] (3) The present invention uses a frequent pattern growth algorithm (FP-Growth) to generate alarm frequent item sets. This algorithm, with the help of the data structure of the frequent pattern tree FP-Tree, can avoid the shortcomings of Apriori-like algorithms that require multiple scans of the transaction database and the generation of a large number of candidate item sets, thereby improving the efficiency of alarm rule mining.

[0055] (4) As the FP-growth algorithm is a typical disordered algorithm, the present invention proposes an alarm rule filtering algorithm based on time sequence. Combined with the characteristics of the currently popular network attack chain, the alarm rules that do not meet the time sequence are further eliminated, and the scale of alarm rules is compressed, which can effectively improve the safety inspection efficiency of power network security managers. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1A diagram of a method for correlation analysis of power network alarm information proposed by the present invention;

[0057] Figure 2 It is a detailed flow chart of the pre-processing module;

[0058] Figure 3 Prototype diagram of the designed dynamic sliding time window algorithm;

[0059] Figure 4 A flowchart for generating alarm rules;

[0060] Figure 5 The figure is a comparison chart of the number of alarm rules under different sliding step sizes. DETAILED DESCRIPTION

[0061] The technical solution provided by the present invention will be described in detail below in conjunction with specific embodiments and the accompanying drawings. It should be understood that the following specific implementation methods are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0062] The present invention proposes a correlation analysis method for power network alarm information. The correlation analysis comprehensive framework is as follows: Figure 1 As shown in the figure, it includes three modules. The first part is the alarm preprocessing module. Specifically, the alarm logs generated by different safety devices are normalized to obtain alarm data with consistent formats and attributes, and then the processed alarm logs are preprocessed to convert them into alarm data sets that are conducive to subsequent association analysis; the second part is the transaction set construction module, which divides the original alarm sequence into several alarm subsequences, and then uses the proposed dynamic sliding time window (Dynamic Sliding Time Window, DSTW) to process each subsequence, efficiently and accurately propose transactions and form alarm transaction sets; the third part is the alarm rule mining and filtering module. Specifically, for the alarm transaction set generated in the second part, the FP-Growth algorithm based on FP-Tree is used to mine the strong association rules of power network alarms. At the same time, according to the characteristics of the FP-Growth algorithm that does not consider the time sequence of rules, an alarm rule filtering algorithm based on time sequence is designed to further eliminate invalid rules and compress the scale of alarm rules.

[0063] Specifically, the method of the present invention comprises the following steps:

[0064] (1) The alarm logs collected by the power network situation awareness platform are normalized. These alarm logs come from multi-source security devices, and the formats of the logs are different. Combined with the characteristics of the power grid, they are normalized. After parsing, the alarm logs are unified in format and parallel in important fields, which is conducive to further analysis.

[0065] The following table shows a sample of the parsed alarm log:

[0066]

[0067] (2) Preprocess and analyze the normalized alarm data. Figure 2 Shown is a detailed flow chart of the preprocessing analysis.

[0068] The specific process of this step is as follows:

[0069] (2.1) Attribute selection. The normalized alarm data contains a large number of attributes, which takes up a lot of storage space. Too many irrelevant attributes are not conducive to the mining of association patterns. Therefore, feature selection is first performed. Combined with the characteristics of the power network, some relatively important features are selected, as shown in the following table:

[0070]

[0071] (2.2) Data cleaning: It is inevitable that some fields in the alarm data set will have missing values. Considering that the missing values ​​are not enough to constitute a complete attack event, the present invention adopts a simple deletion method to clean the alarm data;

[0072] (2.3) Remove duplicate and redundant alarm information. According to the characteristics of alarm information, duplicate alarm information is often concentrated within a very small time threshold. The definition of duplicate alarms is as follows:

[0073]

[0074] Among them, alarm i and alarm j Respectively represent the i-th and j-th alarm attack events, including the attributes of alarm name alarm_name, alarm attack target alarm_tgt, alarm attack source alm_src; alm_time i and alm_time j Respectively represent the occurrence time of the i-th and j-th alarm events; t min Indicates the minimum time interval threshold, which is a custom parameter.

[0075] (2.4) Data sorting: sort the data set using the alarm time (alm_time) as the keyword. The alarm data set sorted by time has the pattern of association rule mining, which can match the subsequent transaction extraction and association analysis algorithm based on the dynamic sliding time window.

[0076] (3) The alarm data processed by the preprocessing module is input into the alarm transaction construction module, first segmented, and then the alarm transaction set is extracted;

[0077] The specific process of this step is as follows:

[0078] (3.1) Segment the sorted alarm sequence set in (2.4). According to the specific sample size, set the segmentation threshold K in advance, and divide the alarm sequence into K segments. Through parallel computing, the transaction extraction process can be accelerated and the efficiency of the algorithm can be improved.

[0079] (3.2) For each alarm subsequence, a method based on a dynamic sliding time window (DSTW) is designed to extract alarm transactions, and to aggregate closely related alarms into the same transaction set to the greatest extent possible, so as to achieve the goal of accurately and efficiently clustering attack events. The idea of ​​the DSTW algorithm is as follows: Figure 3 As shown, the specific process is as follows:

[0080] (3.2.1) First, perform initialization parameter settings, set the original size W of the sliding time window, the sliding step size S, and the minimum time threshold Time_intvl, for example, W = 5, S = 2, Time_intvl = 10min;

[0081] (3.2.2) Slide from the initial position of each subsequence;

[0082] (3.2.3) Under normal circumstances, when the window size reaches W, the alarm events in the window are taken to form a transaction, and the pointer on the left side slides S steps to the right; when the time distance between any two alarm attack events exceeds a certain time threshold, the window stops sliding, and the alarm events between the two pointers are used to form an alarm transaction; at the same time, the sliding start position is adjusted to the next position of the current position, and sliding starts again.

[0083] (3.2.4) Repeat the steps in (3.2.3) until the sliding pointer has slid to the end of each subsequence.

[0084] (3.2.5) Figure 3 Taking the alarm subsequence S1 as an example, the above process is used to extract transactions, and the resulting transaction set is shown in the following table:

[0085]

[0086] (4) For the transaction set, the classic association analysis algorithm FP-Growth is used to mine alarm rules, and a time-sequence-based alarm rule filtering algorithm is designed to filter the mined rules to obtain the final rule set. The specific process is as follows: Figure 4 As shown;

[0087] This step specifically includes the following processes:

[0088] (4.1) For the alarm transaction set, the frequent pattern growth algorithm (FP-Growth) is used to mine the strong association rules of power network alarms and obtain the initial alarm rule set. The specific process is as follows:

[0089] (4.1.1) Set the minimum support threshold minSup and the minimum confidence threshold minConf;

[0090] (4.1.2) Generate a head linked list, build a frequent pattern tree FP-Tree based on the head linked list, and store the alarm transaction set information in a form similar to a prefix tree;

[0091] (4.1.3) Generate frequent itemsets of warning information based on FP-Tree. This step depends on the specific calculation of support and confidence. The calculation formula is as follows:

[0092]

[0093] This formula represents the calculation item set Item i The support degree of Item i represents the i-th item set, T represents the alarm transaction set, q represents the number of alarm transactions, t j Indicates the jth alarm transaction.

[0094] support(X)=α(X) / |T|

[0095] This formula represents the calculation of the support of the item set X, where |T| represents the number of transactions contained in the transaction set T, and α(X) represents the support of the item set X.

[0096]

[0097] This formula calculates the confidence of the rule X->Y, where support(X∪Y) and support(X) represent the support of the item set X∪Y and the item set X, respectively.

[0098] (4.1.4) Based on the frequent item sets, the initial alarm association rules are finally generated.

[0099] (4.2) A time-sequence-based alarm rule filtering algorithm is proposed. It fully considers the relationship between each alarm event, filters out alarm rules that do not meet the time sequence conditions, further compresses the number of rules, and obtains the final alarm rule list. The following table shows the alarm rule list after filtering obtained according to the above steps when the minimum support minSupp is set to 0.12 and the minimum confidence minConf is set to 0.25:

[0100]

[0101]

[0102] In order to verify the effect of this method under different sliding step sizes, the number of rules under sliding step sizes of 2, 4, 6, 8, and 10 is compared. Figure 5 As shown in the figure, when the sliding step size is set to 2, smaller but more critical alarm rules can be obtained, which improves the troubleshooting efficiency of managers.

[0103] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical solutions composed of any combination of the above-mentioned technical features. It should be pointed out that for ordinary technicians in this technical field, several improvements and polishings can be made without departing from the principle of the present invention, and these improvements and polishings are also regarded as the protection scope of the present invention.

Claims

1. A correlation analysis method for power network alarm information, It is characterized in that The method comprises the following steps: (1) Normalize the alarm logs collected by the power network situation awareness platform; (2) Preprocess and analyze the normalized alarm data; (3) The pre-processed alarm data is input into the alarm transaction construction module for segmentation processing and extraction of alarm transaction sets; (4) For the transaction set, the classic association analysis algorithm FP-Growth is used to mine alarm rules, and a time-series-based alarm rule filtering algorithm is designed to filter the mined rules to obtain the final rule set; The alarm data after preprocessing in step (3) is input into the alarm transaction construction module for segmentation processing and extraction of alarm transaction sets. The specific steps are as follows: (3.1) Segment the sorted alarm sequence set in (2.4), set the segmentation threshold K in advance according to the specific sample size, and divide the alarm sequence into K segments; (3.2) For each alarm subsequence, a method based on a dynamic sliding time window is designed to extract alarm transactions, so that alarms with close correlations are aggregated into the same transaction set to the greatest extent possible; Among them, the specific method of step (3.2) is as follows: (3.2.1) Perform initialization settings, set the original size W of the sliding time window, the sliding step size S, and the minimum time threshold Time_intvl; (3.2.2) Slide from the initial position of each subsequence; (3.2.3) Under normal circumstances, when the window size reaches W, the alarm events in the window are taken to form a transaction, and the pointer on the left side slides S steps to the right. When the time distance between any two alarm attack events exceeds a certain time threshold, the window stops sliding and obtains an alarm transaction. At the same time, the sliding start position is adjusted to the current position and the sliding starts again. (3.2.4) Repeat the steps in (3.2.3) until the sliding pointer has slid to the end of each subsequence.

2. A method for correlation analysis of power network alarm information according to claim 1, It is characterized in that Step (2) pre-processes and analyzes the normalized alarm data, as follows: (2.1) Attribute selection is performed on the original alarm log after normalization in (1), and a portion of relatively important alarm attribute features are selected; (2.2) Clean the alarm data; (2.3) Remove duplicate and redundant alarm information and define duplicate alarms as follows: Among them, alarm i and alarm j Respectively represent the i-th and j-th alarm attack events, including the attributes of alarm name alarm_name, alarm attack target alarm_tgt, alarm attack source alm_src; alm_time i and alm_time j Respectively represent the occurrence time of the i-th and j-th alarm events; t min Indicates the minimum time interval threshold, which is a custom parameter; (2.4) Sort the alarm information in chronological order.

3. A method for correlation analysis of power network alarm information according to claim 2, It is characterized in that The attribute features selected in step (2.1) are shown in the following table:

4. A method for correlation analysis of power network alarm information according to claim 1, It is characterized in that The specific method of step (4) is as follows: (4.1) For the alarm transaction set, the frequent pattern growth algorithm is used to mine the strong association rules of power network alarms; (4.2) A time-series-based alarm rule filtering algorithm is proposed, which fully considers the relationship between each alarm event, filters out alarm rules that do not meet the time sequence conditions, and further reduces the number of rules.

5. A method for correlation analysis of power network alarm information according to claim 4, It is characterized in that The specific method of step (4.1) is as follows: (4.1.1) Set the minimum support threshold minSup and the minimum confidence threshold minConf; (4.1.2) Construct frequent pattern tree FP-Tree; (4.1.3) Generate frequent itemsets of warning information; this step depends on the specific calculation formula of support and confidence, as follows: This formula represents the calculation item set Item i The support degree of Item i represents the i-th item set, T represents the alarm transaction set, q represents the number of alarm transactions, t j Indicates the jth alarm transaction; support(X)=α(X) / |T| This formula represents the calculation of the support of the item set X, where |T| represents the number of transactions contained in the transaction set T, and α(X) represents the support of the item set X; This formula calculates the confidence of the rule X->Y, where support(X∪Y) and support(X) represent the support of the item set X∪Y and the item set X, respectively. (4.1.4) Generate initial alarm association rules.