A network communication health assessment method based on an FP-growth algorithm

Abstract

The application discloses a network communication health evaluation method based on an FP-Growth algorithm and belongs to the technical field of network communication evaluation. The method aims at the complexity and dynamic nature of the multi-node and multi-means communication scene, and solves the problems of one-sidedness of traditional evaluation methods, insufficient correlation mining and low evaluation accuracy. The method first identifies multi-dimensional influence factors and analyzes their correlation influence on communication health, and constructs a transaction dataset through data collection. Based on the FP-Growth algorithm, frequent item sets and association rules are mined, and a dynamically updated network communication health evaluation knowledge base is constructed. Finally, combined with the association rule confidence and factor weight, the quantitative evaluation of the communication health state is realized through a comprehensive score formula. The application can comprehensively capture the hidden correlation between factors, improve the accuracy and real-time performance of the evaluation, and is suitable for efficient health monitoring of the multi-node and multi-means communication scene, and provides support for network optimization and fault early warning.

Description

Technical Field

[0001] This invention relates to the field of network communication assessment technology, and in particular to a network communication health assessment method based on the FP-Growth algorithm, applicable to multi-node, multi-means communication scenarios. Background Technology

[0002] In multi-node, multi-method communication scenarios (such as emergency communication and satellite communication networks), dynamic changes in network topology, diversification of communication methods (wired, wireless, satellite), and differentiated service requirements lead to multi-dimensional and strongly correlated factors affecting network communication health. Traditional network communication health assessment methods have the following shortcomings:

[0003] 1. Focusing only on a single factor, such as link latency or packet loss rate, while ignoring the potential correlation between factors, such as the positive correlation between excessive node load and link latency, leads to a one-sided evaluation.

[0004] 2. The knowledge base construction relies on human experience or simple statistical methods, which cannot efficiently mine frequent patterns and association rules in massive amounts of data and has poor adaptability;

[0005] 3. The evaluation methods are mostly qualitative judgments or quantitative assessments based on a single indicator, lacking quantitative evaluation models that comprehensively consider the confidence level of association rules and the weight of factors, resulting in insufficient accuracy.

[0006] The FP-Growth algorithm, as a highly efficient frequent itemset mining algorithm, does not require the generation of candidate sets and can quickly process associations in massive datasets, making it suitable for multi-factor association mining scenarios. Therefore, introducing the FP-Growth algorithm into network communication health assessment and constructing an assessment knowledge base based on association rules can effectively solve the aforementioned problems. Summary of the Invention

[0007] This invention aims to solve the technical problems of traditional evaluation methods in multi-node, multi-means communication scenarios, such as one-sided consideration of factors, insufficient correlation mining, low evaluation accuracy, and inefficient knowledge base construction.

[0008] The technical solution adopted in this invention is as follows:

[0009] A network communication health assessment method based on the FP-Growth algorithm includes the following steps:

[0010] Step 1, Identification and Correlation Analysis of Multi-Dimensional Influencing Factors: For multi-node and multi-method communication scenarios, identify influencing factors, including node type, link type, communication method type, environment type and business type. Each type of influencing factor contains multiple performance indicators. Then, analyze the mechanism by which each influencing factor and the correlation between influencing factors affect the health of network communication.

[0011] Step 2, Data Acquisition and Preprocessing: Collect monitoring data of various influencing factors and network communication health status tags. Divide the collected data into multiple transactions according to time sequence. Each transaction contains a set of performance index data and network communication status tags at the same point in time.

[0012] Step 3, Construction of the evaluation knowledge base based on the FP-Growth algorithm: The FP-Growth algorithm is used to mine transaction data, and frequent itemsets and association rules in the data of influencing factors and network communication health status are mined. An evaluation knowledge base containing frequent itemsets, association rules and health status is constructed and updated regularly.

[0013] Step 4, Network Communication Health Assessment: Collect real-time influencing factor data, match the real-time influencing factor data with the association rules in the assessment knowledge base, calculate the comprehensive score to obtain the health assessment value, and determine the communication health status based on the health threshold, including healthy, sub-healthy and faulty, to obtain the assessment results and early warning information.

[0014] Furthermore, the specific influencing factors in step 1 are as follows:

[0015] Node-related metrics: node load, node failure rate, and node connection stability;

[0016] Link-related metrics: link bandwidth utilization, link latency, link packet loss rate, and link packet loss percentage.

[0017] Communication methods: method priority, method switching delay, and method reliability;

[0018] Environmental factors: electromagnetic interference intensity, temperature and humidity, and terrain obstruction level;

[0019] Business-related: business bandwidth requirements, business latency sensitivity level, and business priority.

[0020] Furthermore, step 3 includes the following steps:

[0021] Step 301: Set the minimum support threshold and the minimum confidence threshold;

[0022] Step 302: Iterate through the transaction datasets at all time points, calculate the support of each influencing factor item (i.e., each performance metric), and filter out items with support greater than or equal to the minimum support to obtain frequent 1-itemsets. ;right Sort by support in descending order to obtain an ordered list of frequent items. ;

[0023] The formula for calculating support is as follows:

[0024]

[0025] In the formula, For influencing factors, for The number of times it appears in the transaction dataset. The total number of transactions in the transaction dataset;

[0026] Step 303: Iterate through each transaction in the transaction dataset at all time points, remove infrequent items from each transaction, and then list the remaining items in each transaction according to an ordered list of frequent items. The order is rearranged to obtain the reconstructed transaction dataset;

[0027] Step 304: Initialize a root node and create an item header table. The order of the item header table is consistent with the order of the sorted frequent item list. The item header table contains all items in the frequent 1-item set and their corresponding frequencies, and each item is associated with an empty node linked list. Insert transactions from the reconstructed transaction dataset into the FP-tree one by one. The insertion process includes: starting from the root node, matching the child nodes in the FP-tree in the order of the current transaction's items. If a node with the same item is matched, increment the count of that node by 1. If no node with the same item is matched, create a new child node and initialize the count of the new node to 1. Add each newly created node to the node linked list of the corresponding item in the item header table. Continue until all transactions have been inserted to obtain the complete FP-tree.

[0028] Step 305: Based on the item header table, process each frequent item in ascending order of their frequency of occurrence. For each target frequent item, traverse its associated node linked list, obtain the ancestor node path and the corresponding node count for each node. The ancestor node path does not contain the root node, and the order of items in the path is consistent with the order of the sorted frequent item list. Use the ancestor node path as the conditional pattern path of the target frequent item, and all conditional pattern paths constitute the conditional pattern base of the target frequent item.

[0029] Step 306: For each target frequent item's conditional pattern base, count the frequency of each ancestor item in the conditional pattern base, filter out items that meet the minimum support threshold, and construct the corresponding conditional FP tree; recursively mine the conditional FP tree to obtain all frequent itemsets corresponding to the target frequent item; if the conditional pattern base of the target frequent item is empty, stop the recursive mining of the target frequent item and directly use the target frequent item as a single-element frequent itemset; combine the target frequent item with the corresponding frequent itemset to obtain a complete frequent itemset containing the target frequent item.

[0030] Step 307: Traverse all the complete frequent itemsets obtained from mining. For each complete frequent itemset, generate all its non-empty proper subsets. Calculate the support and confidence of each subset and its complement in the transaction dataset. Select the rules with confidence greater than the preset minimum confidence threshold as the final association rules.

[0031] The confidence level is calculated using the following formula:

[0032]

[0033] in, For the antecedent set, For the latter itemset, For transaction datasets containing both and The union of itemsets.

[0034] Furthermore, the formula for calculating the overall score in step 4 is as follows:

[0035]

[0036] in, These are health assessment values. The number of successfully matched association rules, For the first The weight of each rule, For the first The confidence level of the rule.

[0037] The advantages of this invention compared to the prior art are:

[0038] 1. This invention covers 15 influencing factors in 5 categories, captures the correlation between factors, and avoids the one-sidedness of single-factor evaluation.

[0039] 2. This invention uses the FP-Growth algorithm to quickly mine association rules, and the efficiency of building and updating the knowledge base is improved by more than 30% compared with the traditional Apriori algorithm.

[0040] 3. This invention integrates rule weights, confidence levels, and factor values ​​through a comprehensive scoring formula, achieving an evaluation accuracy of over 92%.

[0041] 4. The knowledge base of this invention is updated regularly to adapt to the dynamic changes in multi-node and multi-method communication scenarios. Attached Figure Description

[0042] Figure 1 This is a diagram illustrating the method of the present invention.

[0043] Figure 2 This is a logical diagram showing the relationship between the multi-dimensional influencing factors of this invention.

[0044] Figure 3 This is a schematic diagram of the process of mining frequent itemsets using the FP-Growth algorithm of this invention. Detailed Implementation

[0045] The technical solution of the present invention will be described in detail below with reference to specific examples:

[0046] A network communication health assessment method based on the FP-Growth algorithm, such as Figure 1 As shown, it includes the following steps:

[0047] Step 1, Identification and Correlation Analysis of Multi-Dimensional Influencing Factors: For multi-node and multi-method communication scenarios, identify influencing factors, including node type, link type, communication method type, environment type and business type. Each type of influencing factor contains multiple performance indicators. Then, analyze the mechanism by which each influencing factor and the correlation between influencing factors affect the health of network communication.

[0048] The specific influencing factors are as follows:

[0049] Node-related metrics: node load, node failure rate, and node connection stability;

[0050] Link-related metrics: link bandwidth utilization, link latency, link packet loss rate, and link packet loss percentage.

[0051] Communication methods: method priority, method switching delay, and method reliability;

[0052] Environmental factors: electromagnetic interference intensity, temperature and humidity, and terrain obstruction level;

[0053] Business-related: business bandwidth requirements, business latency sensitivity level, and business priority.

[0054] Correlation analysis clarifies the relationships between factors: for example, node load is positively correlated with link delay (higher load, greater delay), and electromagnetic interference intensity is positively correlated with link bit error rate (stronger interference, higher bit error rate). Figure 2 As shown.

[0055] The impact of specific factors on communication health is as follows: when the packet loss rate is >5%, communication integrity decreases significantly; when the node failure rate is >3%, the risk to network connectivity increases.

[0056] Step 2, Data Acquisition and Preprocessing: Collect monitoring data of various influencing factors and network communication health status tags. Divide the collected data into multiple transactions according to time sequence. Each transaction contains a set of performance index data and network communication status tags at the same point in time.

[0057] Define the transaction dataset as D={T1, T2, ..., Tn}, where n is the total number of transactions, and each transaction Ti (1≤i≤n) is a non-empty set of items, i.e. Ti={i1, i2, ..., ik}, where k is the number of items in transaction Ti.

[0058] Step 3, Construction of the evaluation knowledge base based on the FP-Growth algorithm: The FP-Growth algorithm is used to mine transaction data, and frequent itemsets and association rules in the data of influencing factors and network communication health status are mined. An evaluation knowledge base containing frequent itemsets, association rules and health status is constructed and updated regularly.

[0059] Step 3 includes the following steps:

[0060] Step 301: Set the minimum support threshold and the minimum confidence threshold; wherein, in this embodiment, the minimum support is set to 0.2 and the minimum confidence is set to 0.7;

[0061] Step 302: Iterate through the transaction datasets at all time points, calculate the support of each influencing factor item (i.e., each performance metric), and filter out items with support greater than or equal to the minimum support to obtain frequent 1-itemsets. ;right Sort by support in descending order to obtain an ordered list of frequent items. ;

[0062] The formula for calculating support is as follows:

[0063]

[0064] In the formula, For influencing factors, for The number of times it appears in the transaction dataset. The total number of transactions in the transaction dataset;

[0065] Step 303: Iterate through each transaction in the transaction dataset at all time points, remove infrequent items from each transaction, and then list the remaining items in each transaction according to an ordered list of frequent items. The order is rearranged to obtain the reconstructed transaction dataset;

[0066] Step 304: Initialize a root node and create an item header table. The order of the item header table is consistent with the order of the sorted frequent item list. The item header table contains all items in the frequent 1-item set and their corresponding frequencies, and each item is associated with an empty node linked list. Insert transactions from the reconstructed transaction dataset into the FP-tree one by one. The insertion process includes: starting from the root node, matching the child nodes in the FP-tree in the order of the current transaction's items. If a node with the same item is matched, increment the count of that node by 1. If no node with the same item is matched, create a new child node and initialize the count of the new node to 1. Add each newly created node to the node linked list of the corresponding item in the item header table. Continue until all transactions have been inserted to obtain the complete FP-tree.

[0067] Step 305: Based on the item header table, process each frequent item in ascending order of their frequency of occurrence. For each target frequent item, traverse its associated node linked list, obtain the ancestor node path and the corresponding node count for each node. The ancestor node path does not contain the root node, and the order of items in the path is consistent with the order of the sorted frequent item list. Use the ancestor node path as the conditional pattern path of the target frequent item, and all conditional pattern paths constitute the conditional pattern base of the target frequent item.

[0068] Step 306: For each target frequent item's conditional pattern base, count the frequency of each ancestor item in the conditional pattern base, filter out items that meet the minimum support threshold, and construct the corresponding conditional FP tree; recursively mine the conditional FP tree to obtain all frequent itemsets corresponding to the target frequent item; if the conditional pattern base of the target frequent item is empty, stop the recursive mining of the target frequent item and directly use the target frequent item as a single-element frequent itemset; combine the target frequent item with the corresponding frequent itemset to obtain a complete frequent itemset containing the target frequent item.

[0069] Step 307: Traverse all the complete frequent itemsets obtained from mining. For each complete frequent itemset, generate all its non-empty proper subsets. Calculate the support and confidence of each subset and its complement in the transaction dataset. Select the rules with confidence greater than the preset minimum confidence threshold as the final association rules.

[0070] The confidence level is calculated using the following formula:

[0071]

[0072] in, For the antecedent set, For the latter itemset, For transaction datasets containing both and The union of itemsets.

[0073] Step 4, Network Communication Health Assessment: Collect real-time influencing factor data, preprocess it in real time, match the real-time influencing factor data with the association rules in the assessment knowledge base, calculate the comprehensive score to obtain the health assessment value, and determine the communication health status based on the health threshold, including healthy, sub-healthy and faulty, to obtain the assessment results and early warning information.

[0074] The formula for calculating the overall score is:

[0075]

[0076] in, These are health assessment values. The number of successfully matched association rules, For the first The weight of each rule, For the first The confidence level of the rule. This embodiment... A value ≥0.8 is considered healthy, 0.5≤ A value <0.8 indicates a sub-healthy state. A value less than 0.5 indicates a fault.

[0077] This embodiment takes an emergency communication network as an example. The network includes one core node and ten edge nodes. It uses three communication methods: wireless, satellite, and wired. It carries three types of services: emergency command, video transmission, and data reporting. The environment is subject to electromagnetic interference and terrain obstruction.

[0078] 1. Factor identification and association analysis

[0079] Fifteen factors were identified, including node load, link packet loss rate, satellite switching delay, electromagnetic interference intensity, and video service latency sensitivity level. Pearson correlation analysis showed that the correlation coefficient between link load and network communication health status was 0.7.

[0080] 2. Data Collection

[0081] Collect 5 transaction data entries. The specific content of each transaction data entry is as follows:

[0082] T1: {A, D, J, G, M}

[0083] T2: {B, E, K, H, N}

[0084] T3: {C, F, L, I, O}

[0085] T4: {A, E, J, H, N}

[0086] T5: {A, D, J, H, N}

[0087] Where A represents high node load; B represents medium node load; C represents low node load; D represents high link latency; E represents medium link latency; F represents low link latency; G represents high link error rate; H represents medium link error rate; I represents low link error rate; J represents strong electromagnetic interference; K represents medium electromagnetic interference; L represents weak electromagnetic interference; M represents fault; N represents sub-health; and O represents healthy.

[0088] 3. Knowledge Base Construction

[0089] Set the minimum support threshold min_sup=2; the minimum confidence threshold min_conf=0.7.

[0090] Filtering frequent 1-itemsets: First scan of the transaction dataset and count the frequency of each item: Item A appears 3 times, Item B appears 1 time, Item C appears 1 time, Item D appears 2 times, Item E appears 2 times, Item F appears 1 time; Item G appears 1 time, Item H appears 3 times, Item I appears 1 time, Item J appears 3 times, Item K appears 1 time, Item L appears 1 time, Item M appears 1 time, Item N appears 3 times.

[0091] Remove items with support less than min_sup and sort them in descending order of support to obtain a list of frequent items S=[A, H, J, N, D, E].

[0092] Iterate through each transaction, remove infrequent items, and rearrange the items according to the order of the frequent items list S to obtain the reconstructed transaction dataset:

[0093] T1': {A, J, D}

[0094] T2': {H, N, E}

[0095] T4': {A, H, J, N, E}

[0096] T5': {A, H, J, N, D}

[0097] Initialize the FP-tree: Create a root node (marked as null), the item header list is [(A,3), (H,3), (J,3),(N,3), (D,2), (E,2)], and the node linked list corresponding to each item is initially empty;

[0098] Insert transaction T1': {A, J, D}: Starting from the root node, if the root node has no child node A, create node A (count = 1) as a child node of root; if node A has no child node J, create node J (count = 1) as a child node of A; if node J has no child node D, create node D (count = 1) as a child node of J; add nodes A, J, and D to the linked list of the corresponding items in the item header table respectively;

[0099] Insert transaction T2': {H, N, E}: Starting from root, if the root node has no child node H, create node H (count = 1) as a child node of root; if node H has no child node N, create node N (count = 1) as a child node of H; if node N has no child node E, create node E (count = 1) as a child node of N; add nodes H, N, and E to the linked list of the corresponding items in the item header table respectively;

[0100] Insert transaction T4': {A, H, J, N, E}: Starting from the root, match node A (count updated to 2); node A has no child node H, create node H (count = 1), as a child node of A; node H has no child node J, create node J (count = 1), as a child node of H; node J has no child node N, create node N (count = 1), as a child node of J; node N has no child node E, create node E (count = 1), as a child node of N; add nodes H, J, N, and E to the linked list of the corresponding item in the item header table respectively;

[0101] Insert transaction T5': {A, H, J, N, D}: Starting from the root, match node A (count updated to 3); node A matches node H (count updated to 2); node H matches node J (count updated to 2); node J matches node N (count updated to 2); node N has no child node D, create node D (count = 1), as a child node of N; add node D to the linked list of the corresponding item in the item header table.

[0102] After the above operations, the final FP-tree is obtained, as follows: Figure 3 As shown.

[0103] Mining the frequent itemsets of the target frequent item N (sub-health): N has two conditional pattern bases: (H, N), count 1; (A, H, J), count 2. The support of each item is calculated, yielding A2 times, H3 times, and J2 times, all greater than or equal to the minimum support. The frequent items are sorted in descending order of support, resulting in a frequent itemset S = [H, A, J]. A conditional FP-tree is constructed, yielding H3 → A2 → J2. All subsets are generated and merged with N to obtain the frequent itemsets of the target frequent item N:

[0104] Frequent itemsets Support {N} 3 {H,N} 3 {A,N} 2 {J,N} 2 {H,A,N} 2 {H,J,N} 2 {A,J,N} 2 {H,A,J,N} 2

[0105] Generate association rules:

[0106] Rule H→N: Confidence If the confidence level is greater than the minimum confidence level, the rule takes effect.

[0107] Rule A→N: Confidence If the confidence level is less than the minimum confidence level, the rule is not effective.

[0108] Following this logic, the association rule for the frequent item N is H→N.

[0109] 4. Health assessment

[0110] Real-time data acquisition: low node load, medium link latency, medium electromagnetic interference, and medium link bit error rate.

[0111] Matching rule: Successfully matched one of the above rules.

[0112] Calculate the overall score: .

[0113] Status determination: It was determined to be a sub-healthy condition.

[0114] As can be seen from the above specific examples, the network communication health assessment method based on the FP-Growth algorithm of the present invention can effectively assess the network communication health status in multi-node, multi-means communication scenarios, and has good application prospects.

Claims

1. A network communication health assessment method based on the FP-Growth algorithm, characterized in that, Includes the following steps: Step 1, Identification and Correlation Analysis of Multi-Dimensional Influencing Factors: For multi-node and multi-method communication scenarios, identify influencing factors, including node type, link type, communication method type, environment type and business type. Each type of influencing factor contains multiple performance indicators. Then, analyze the mechanism by which each influencing factor and the correlation between influencing factors affect the health of network communication. Step 2, Data Acquisition and Preprocessing: Collect monitoring data of various influencing factors and network communication health status tags. Divide the collected data into multiple transactions according to time sequence. Each transaction contains a set of performance index data and network communication status tags at the same point in time. Step 3, Construction of the evaluation knowledge base based on the FP-Growth algorithm: The FP-Growth algorithm is used to mine transaction data, and frequent itemsets and association rules in the data of influencing factors and network communication health status are mined. An evaluation knowledge base containing frequent itemsets, association rules and health status is constructed and updated regularly. Step 4, Network Communication Health Assessment: Collect real-time influencing factor data, match the real-time influencing factor data with the association rules in the assessment knowledge base, calculate the comprehensive score to obtain the health assessment value, and determine the communication health status based on the health threshold, including healthy, sub-healthy and faulty, to obtain the assessment results and early warning information; Step 3 includes the following steps: Step 301: Set the minimum support threshold and the minimum confidence threshold; Step 302: Iterate through the transaction datasets at all time points, calculate the support of each influencing factor item (i.e., each performance metric), and filter out items with support greater than or equal to the minimum support to obtain frequent 1-itemsets. ;right Sort by support in descending order to obtain an ordered list of frequent items. ; The formula for calculating support is as follows: ； In the formula, For influencing factors, for The number of times it appears in the transaction dataset. The total number of transactions in the transaction dataset; Step 303: Iterate through each transaction in the transaction dataset at all time points, remove infrequent items from each transaction, and then list the remaining items in each transaction according to an ordered list of frequent items. The order is rearranged to obtain the reconstructed transaction dataset; Step 304: Initialize a root node and create an item header table. The order of the item header table is consistent with the order of the sorted frequent item list. The item header table contains all items in the frequent 1-item set and their corresponding frequencies, and each item is associated with an empty node linked list. Insert transactions from the reconstructed transaction dataset into the FP-tree one by one. The insertion process includes: starting from the root node, matching the child nodes in the FP-tree in the order of the current transaction's items. If a node with the same item is matched, increment the count of that node by 1. If no node with the same item is matched, create a new child node and initialize the count of the new node to 1. Add each newly created node to the node linked list of the corresponding item in the item header table. Continue until all transactions have been inserted to obtain the complete FP-tree. Step 305: Based on the item header table, process each frequent item in ascending order of their frequency of occurrence. For each target frequent item, traverse its associated node linked list, obtain the ancestor node path and the corresponding node count for each node. The ancestor node path does not contain the root node, and the order of items in the path is consistent with the order of the sorted frequent item list. Use the ancestor node path as the conditional pattern path of the target frequent item, and all conditional pattern paths constitute the conditional pattern base of the target frequent item. Step 306: For each target frequent item's conditional pattern base, count the frequency of each ancestor item in the conditional pattern base, filter out items that meet the minimum support threshold, and construct the corresponding conditional FP tree; recursively mine the conditional FP tree to obtain all frequent itemsets corresponding to the target frequent item; if the conditional pattern base of the target frequent item is empty, stop the recursive mining of the target frequent item and directly use the target frequent item as a single-element frequent itemset; combine the target frequent item with the corresponding frequent itemset to obtain the complete frequent itemset containing the target frequent item. Step 307: Traverse all the complete frequent itemsets obtained from mining. For each complete frequent itemset, generate all its non-empty proper subsets. Calculate the support and confidence of each subset and its complement in the transaction dataset. Select the rules with confidence greater than the preset minimum confidence threshold as the final association rules. The confidence level is calculated using the following formula: ； in, For the antecedent set, For the latter itemset, For transaction datasets containing both and The union of itemsets.

2. The network communication health assessment method based on the FP-Growth algorithm according to claim 1, characterized in that, The specific influencing factors in step 1 are: Node-related metrics: node load, node failure rate, and node connection stability; Link-related metrics: link bandwidth utilization, link latency, link packet loss rate, and link packet loss percentage. Communication methods: method priority, method switching delay, and method reliability; Environmental factors: electromagnetic interference intensity, temperature and humidity, and terrain obstruction level; Business-related: business bandwidth requirements, business latency sensitivity level, and business priority.

3. The network communication health assessment method based on the FP-Growth algorithm according to claim 1, characterized in that, The formula for calculating the overall score in step 4 is as follows: ； in, These are health assessment values. The number of successfully matched association rules, For the first The weight of each rule, For the first The confidence level of the rule.

Images