A method and system for intrusion detection data analysis in distributed networks
By combining local computation at distributed detection nodes and exchange of low-dimensional statistics between adjacent nodes with adaptive consistency verification and deviation accumulation detection, the problem of identifying hidden anomalies in industrial Ethernet loop communication is solved, thereby improving the accuracy and efficiency of intrusion detection.
Patent Information
- Application Number
- CN202611587454.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to effectively identify hidden anomalies in industrial Ethernet cyclic communication where single arrival time or data length deviations fall within the normal jitter range but accumulate over time. Furthermore, it is difficult to distinguish whether slow deviations are caused by intrusion or normal deviations due to global clock adjustments, leading to false alarms or missed alarms.
The arrival time and data length deviation of the circular packets are calculated locally on the distributed detection node to form a joint deviation. Low-dimensional trend statistics are generated by segmented statistics within a sliding window. Combined with adaptive consistency verification of adjacent nodes and deviation accumulation detection, an intrusion alarm is generated.
It reduces network bandwidth consumption, decreases false alarms caused by clock fluctuations or topology changes, improves the ability to detect covert intrusions, and enhances the robustness and scalability of intrusion detection.
Smart Images

Figure CN123123020A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of network security technology, specifically to an intrusion detection data analysis method and system in a distributed network. Background Technology
[0002] A distributed network refers to a network composed of multiple nodes with independent acquisition, computing, and communication capabilities. Nodes exchange data through communication links and can form neighboring node relationships based on the network topology. Intrusion detection refers to monitoring network traffic, communication behavior, or device status to identify activities that deviate from normal behavior. Industrial Ethernet cyclic communication refers to a communication method in industrial control systems where controllers and field devices exchange cyclic messages according to a fixed or configured cycle. This is common in industrial networks such as PROFINET, EtherCAT, and EtherNet / IP. A cyclic message is a message repeatedly sent periodically in this type of communication, typically possessing attributes such as nominal period, nominal data length, message identifier, and priority. The nominal period is the agreed-upon interval for sending cyclic messages between devices. Arrival time deviation is the difference between the actual arrival time interval of a cyclic message and the nominal period. Data length deviation is the difference between the actual data length of a cyclic message and the nominal data length. A sliding window is a data sequence composed of several recent cyclic messages. Trend statistics are a small number of characteristic values obtained by statistically analyzing the deviation changes within the sliding window. Neighbor consistency check compares whether the trend statistics obtained by adjacent detection nodes are consistent.
[0003] In existing technologies, industrial networks commonly employ industrial firewalls, deep packet inspection, signature-based intrusion detection systems, traffic statistics-based anomaly detection, and security information and event management systems for security monitoring. Signature-based intrusion detection systems, such as Snort and Suricata, identify known attack signatures through rule matching; traffic statistics-based anomaly detection identifies anomalies by statistically analyzing indicators such as traffic rate, connection count, and packet length distribution; and security information and event management systems centrally collect logs and alarms from multiple devices for correlation analysis. These technologies are already in practical use at industrial network boundaries, core switch mirror ports, or security monitoring platforms.
[0004] However, in industrial Ethernet loop communication scenarios, the aforementioned existing technologies have a major problem: they struggle to effectively identify hidden anomalies that fall within the normal jitter range for single arrival time deviations or data length deviations but accumulate over time. Furthermore, they cannot distinguish whether such anomalies originate from intrusion behaviors such as local slow delay injection or short-term discard and replay, or from normal deviations caused by global clock adjustments or network reconfiguration. Signature-based detection relies on known features, making it difficult to match slow deviations without obvious signature characteristics; traffic statistics-based detection typically sets a threshold for single deviations, and alarms are not triggered when single deviations are small; centralized detection requires transmitting a large number of raw loop packets to the central node, increasing network bandwidth consumption and detection latency; single-node detection lacks observation consistency comparisons between adjacent nodes, easily misjudging global clock fluctuations as local intrusions, or missing slow deviations that only affect a local area.
[0005] Therefore, how to compress the cyclic communication deviation characteristics locally at distributed detection nodes, exchange low-dimensional statistical information between adjacent nodes with low overhead, and combine deviation accumulation and consistency verification to identify hidden slow anomalies and distinguish global clock fluctuations has become a technical problem that needs to be solved in industrial Ethernet cyclic communication intrusion detection. Summary of the Invention
[0006] To address the aforementioned technical problems, this invention provides a method for analyzing intrusion detection data in a distributed network, comprising the following steps: The detection node collects and parses circular packets in the distributed network, extracting the arrival time and data length of each circular packet; Based on the nominal cycle period and nominal data length, calculate the cycle deviation and length deviation respectively, and then combine the cycle deviation, length deviation and the change in cycle deviation after normalization into a joint deviation. Segmented statistics are performed on the joint deviation within a sliding window to generate low-dimensional trend statistics. The low-dimensional trend statistics are sent to adjacent detection nodes, and consistency verification is performed based on the statistical distance between the low-dimensional trend statistics and the adaptive consistency threshold to obtain local suspicious tags. The combined deviation is cumulatively detected. When the cumulative amount exceeds the cumulative alarm threshold and the local suspicious tag meets the preset conditions, an intrusion alarm is generated. An alarm log is generated based on the intrusion alarm, and response control is executed.
[0007] Preferably, the detection node is deployed at the mirror port of an industrial switch, an edge gateway, or the communication interface of a field device, and has hardware timestamp capability; the detection node reads the local configuration, identifies the cyclic communication relationship, establishes a set of device pairs, and records the nominal cycle period, nominal data length, allowed data length range, message identifier, and priority for each device pair; and calculates the median of the arrival interval within a consecutive number of nominal cycles, and uses this median as the nominal cycle period.
[0008] Preferably, the process of collecting and parsing the cyclic packets includes: parsing the Ethernet frame header, VLAN tag, industrial Ethernet cyclic packet header, and payload length; for cyclic packets belonging to the same device pair, calculating the difference between the current arrival time and the previous arrival time as the arrival interval; filtering non-cyclic packets based on packet identifier, source address, destination address, and priority; maintaining a sliding window for each device pair, and writing the preprocessing results into a local circular buffer in chronological order.
[0009] Preferably, the joint deviation is obtained by: dividing the absolute value of the periodic deviation by the nominal cycle period to obtain the normalized periodic deviation; dividing the absolute value of the length deviation by the nominal data length to obtain the normalized length deviation; dividing the absolute value of the difference between the current periodic deviation and the previous periodic deviation by the nominal cycle period to obtain the normalized periodic deviation change; multiplying the normalized periodic deviation, the normalized length deviation, and the normalized periodic deviation change by non-negative weighting coefficients respectively and then adding them together to obtain the joint deviation; wherein, the weighting coefficients are automatically set according to the device type, message priority, or message type.
[0010] Preferably, generating low-dimensional trend statistics includes: dividing the sliding window into multiple continuous segments; for each segment, calculating the mean of the joint deviation within the segment, the slope of the joint deviation within the segment as the message sequence number changes, and the proportion of the joint deviation within the segment exceeding a uniformly configured threshold; combining the mean, slope, and proportion of each segment in sequence to form low-dimensional trend statistics; the low-dimensional trend statistics do not contain the original message, address, or payload content, and the message authentication code is calculated using the communication key.
[0011] Preferably, the consistency verification includes: for the current detection node and each neighboring detection node, calculating the L1 distance between their low-dimensional trend statistics; collecting all distances and calculating the median distance and the median absolute deviation; determining an adaptive consistency threshold based on the sum of the median distance and the median absolute deviation; determining that the current detection node is inconsistent with its neighboring detection nodes when the distance exceeds the adaptive consistency threshold; and counting the proportion of inconsistent neighbors to the total number of neighbors. When the proportion exceeds a preset proportion, marking the current detection node as locally suspicious.
[0012] Preferably, the method further includes: the current detection node sends the local suspicious marker, low-dimensional trend statistics, and inconsistent neighbor ratio to the distributed intrusion detection management unit; if the management unit finds that multiple neighboring nodes are marked as locally suspicious and the distance between the low-dimensional trend statistics of these nodes is less than the adaptive consistency threshold, it determines that it is a global clock adjustment or network reconfiguration; if only the current detection node is marked as locally suspicious, it determines that it is a slow deviation injection targeting the coverage area of the current detection node.
[0013] Preferably, the cumulative detection includes: maintaining the cumulative deviation amount, where the current cumulative deviation amount is equal to the larger of the previous cumulative deviation amount plus the joint deviation amount minus the adaptive tolerance and zero; the adaptive tolerance is determined based on the mean and standard deviation of the joint deviation amount within the current sliding window; generating a candidate abnormal event when the cumulative deviation amount exceeds the cumulative alarm threshold; generating an intrusion alarm when preset conditions are met by combining local suspicious markers and inconsistent neighbor ratios; and updating the nominal cycle period and resetting the cumulative deviation amount when it is determined to be a global clock adjustment or network reconfiguration.
[0014] Preferably, the alarm record includes node identifier, device pair identifier, time window number, statistical value of joint deviation sequence, cumulative deviation, inconsistent neighbor ratio, and message authentication code verification result; the response control includes at least one of sending a security event notification to the industrial controller, restricting non-critical communication of suspicious field devices, adjusting the switch port mirroring policy, increasing the sampling window length and consistency verification frequency of the relevant device pair.
[0015] An intrusion detection data analysis system in a distributed network includes: The message acquisition module is used to collect and parse circular messages in a distributed network, and extract the arrival time and data length of each circular message. The local joint deviation calculation module is used to calculate the period deviation and length deviation according to the nominal cycle period and nominal data length, and then combine the period deviation, length deviation and the change in period deviation after normalization into a joint deviation. The trend statistics generation module is used to perform segmented statistics on the joint deviation within a sliding window to generate low-dimensional trend statistics. The neighbor consistency verification module is used to send low-dimensional trend statistics to neighboring detection nodes and perform consistency verification based on the statistical distance between the low-dimensional trend statistics and the adaptive consistency threshold to obtain local suspicious tags. The deviation accumulation detection module is used to accumulate and detect the combined deviation. When the accumulated amount exceeds the accumulated alarm threshold and the local suspicious tag meets the preset conditions, an intrusion alarm is generated. The alarm response module is used to generate alarm records based on intrusion alarms and to execute response controls.
[0016] The present invention has the following beneficial effects: By calculating the arrival time deviation and data length deviation of cyclic messages locally at distributed detection nodes, a joint deviation is formed. This joint deviation within a sliding window is then compressed into a low-dimensional trend statistic. The statistic is exchanged only between adjacent nodes without transmitting the original message, which helps reduce network bandwidth consumption in distributed detection and minimizes the security risks associated with the outgoing transmission of original data. Adaptive thresholding and neighbor consistency checks can distinguish between local slow deviation injection and global clock adjustments or network reconfiguration, reducing false alarms caused by clock fluctuations or topology changes. Deviation accumulation detection can identify anomalies with small individual deviations that accumulate over time, improving the ability to detect covert intrusion behaviors in industrial cyclic communications. The overall solution adapts to the periodicity and device correspondence of industrial cyclic communications, facilitating deployment in existing distributed networks and enhancing the robustness and scalability of intrusion detection. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating an intrusion detection data analysis method in a distributed network according to the present invention. Figure 2 This is a schematic diagram of the structure of an intrusion detection data analysis system in a distributed network according to the present invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] Please see Figure 1 As shown, this invention is a method for analyzing intrusion detection data in a distributed network, comprising the following steps: The detection node collects and parses circular packets in the distributed network, extracting the arrival time and data length of each circular packet; Based on the nominal cycle period and nominal data length, calculate the cycle deviation and length deviation respectively, and then combine the cycle deviation, length deviation and the change in cycle deviation after normalization into a joint deviation. Segmented statistics are performed on the joint deviation within a sliding window to generate low-dimensional trend statistics. The low-dimensional trend statistics are sent to adjacent detection nodes, and consistency verification is performed based on the statistical distance between the low-dimensional trend statistics and the adaptive consistency threshold to obtain local suspicious tags. The combined deviation is cumulatively detected. When the cumulative amount exceeds the cumulative alarm threshold and the local suspicious tag meets the preset conditions, an intrusion alarm is generated. An alarm log is generated based on the intrusion alarm, and response control is executed.
[0021] The detection node is deployed at the mirror port of an industrial switch, an edge gateway, or the communication interface of a field device, and has hardware timestamp capability. The detection node reads the local configuration, identifies the cyclic communication relationship, establishes a set of device pairs, and records the nominal cycle period, nominal data length, allowed data length range, message identifier, and priority for each device pair. It calculates the median of the arrival interval within a number of consecutive nominal cycles and uses this median as the nominal cycle period.
[0022] The process of collecting and parsing cyclic packets includes: parsing the Ethernet frame header, VLAN tag, industrial Ethernet cyclic packet header, and payload length; for cyclic packets belonging to the same device pair, calculating the difference between the current arrival time and the previous arrival time as the arrival interval; filtering non-cyclic packets based on packet identifier, source address, destination address, and priority; maintaining a sliding window for each device pair, and writing the preprocessing results into the local circular buffer in chronological order.
[0023] The joint deviation is obtained as follows: the absolute value of the period deviation is divided by the nominal cycle period to obtain the normalized period deviation; the absolute value of the length deviation is divided by the nominal data length to obtain the normalized length deviation; the absolute value of the difference between the current period deviation and the previous period deviation is divided by the nominal cycle period to obtain the normalized period deviation change; the normalized period deviation, normalized length deviation, and normalized period deviation change are multiplied by non-negative weighting coefficients and then added together to obtain the joint deviation; wherein, the weighting coefficients are automatically set according to the device type, message priority, or message type.
[0024] Generating low-dimensional trend statistics involves: dividing the sliding window into multiple continuous segments; for each segment, calculating the mean of the joint deviation within the segment, the slope of the joint deviation within the segment as the message sequence number changes, and the proportion of the joint deviation within the segment exceeding a uniformly configured threshold; combining the mean, slope, and proportion of each segment in sequence to form a low-dimensional trend statistic; the low-dimensional trend statistic does not contain the original message, address, or payload content, and uses the communication key to calculate the message authentication code.
[0025] Consistency verification includes: calculating the L1 distance between the low-dimensional trend statistics of the current detection node and each neighboring detection node; collecting all distances and calculating the median distance and the median absolute deviation; determining an adaptive consistency threshold based on the sum of the median distance and the median absolute deviation; determining that the current detection node is inconsistent with its neighboring detection nodes when the distance exceeds the adaptive consistency threshold; and counting the proportion of inconsistent neighbors to the total number of neighbors. When the proportion exceeds a preset proportion, the current detection node is marked as a local suspicious node.
[0026] It also includes: the current detection node sends local suspicious markers, low-dimensional trend statistics, and inconsistent neighbor ratios to the distributed intrusion detection management unit; if the management unit finds that multiple neighboring nodes are marked as local suspicious and the distance between the low-dimensional trend statistics of these nodes is less than the adaptive consistency threshold, it determines that it is a global clock adjustment or network reconfiguration; if only the current detection node is marked as local suspicious, it determines that it is a slow deviation injection targeting the coverage area of the current detection node.
[0027] Cumulative detection includes: maintaining the cumulative deviation amount, where the current cumulative deviation amount is equal to the larger of the previous cumulative deviation amount plus the joint deviation amount minus the adaptive tolerance and zero; the adaptive tolerance is determined based on the mean and standard deviation of the joint deviation amount within the current sliding window; generating candidate abnormal events when the cumulative deviation amount exceeds the cumulative alarm threshold; generating intrusion alarms when preset conditions are met by combining local suspicious tags and inconsistent neighbor ratios; and updating the nominal cycle period and resetting the cumulative deviation amount when it is determined to be a global clock adjustment or network reconfiguration.
[0028] Alarm logs include node identifier, device pair identifier, time window number, statistical value of joint deviation sequence, cumulative deviation, inconsistent neighbor ratio, and message authentication code verification result; response control includes sending security event notification to industrial controller, restricting non-critical communication of suspicious field devices, adjusting switch port mirroring policy, increasing the sampling window length and consistency verification frequency of relevant device pairs.
[0029] Please see Figure 2 As shown, the present invention is an intrusion detection data analysis system in a distributed network, comprising: The message acquisition module is used to collect and parse circular messages in a distributed network, and extract the arrival time and data length of each circular message. The local joint deviation calculation module is used to calculate the period deviation and length deviation according to the nominal cycle period and nominal data length, and then combine the period deviation, length deviation and the change in period deviation after normalization into a joint deviation. The trend statistics generation module is used to perform segmented statistics on the joint deviation within a sliding window to generate low-dimensional trend statistics. The neighbor consistency verification module is used to send low-dimensional trend statistics to neighboring detection nodes and perform consistency verification based on the statistical distance between the low-dimensional trend statistics and the adaptive consistency threshold to obtain local suspicious tags. The deviation accumulation detection module is used to accumulate and detect the combined deviation. When the accumulated amount exceeds the accumulated alarm threshold and the local suspicious tag meets the preset conditions, an intrusion alarm is generated. The alarm response module is used to generate alarm records based on intrusion alarms and to execute response controls.
[0030] The specific application of this embodiment is as follows: Step S1: Initialization of detection nodes and establishment of cyclic communication baseline Each detection node in the distributed network is deployed at a mirror port of an industrial switch, an edge gateway, or a field device communication interface, and has hardware timestamp capabilities. Each detection node reads its local configuration, identifies cyclic communication relationships, and establishes a device pair set. Each device This corresponds to a loop message stream between a controller and a field device.
[0031] For equipment Record the nominal cycle time Nominal data length Allowed data length range Message identifier and priority. Detection nodes use hardware timestamps to record the arrival time of each cyclic message. ,in Indicates the message sequence number.
[0032] During initialization, the median of arrival intervals over several consecutive nominal periods is used as the initial period estimate, and this median is used as the... Each detection node generates a node identifier, a communication key, and a list of neighboring nodes. The list of neighboring nodes is determined based on the network topology and cyclic communication relationships.
[0033] Step S2: Looping message acquisition, parsing and preprocessing The detection node collects mirrored traffic or interface traffic, parsing Ethernet frame headers, VLAN tags, Industrial Ethernet loop headers, and payload lengths. For each device... Extract hardware timestamps from loop messages. and data length Calculate the arrival interval. ,in This is the hardware timestamp for the previous cycle of messages from the same device. Calculate the cycle deviation. ,in The nominal cycle period obtained in step S1. Calculate the data length deviation. ,in This is the nominal data length.
[0034] like Exceeding If the length exceeds the limit, the event is directly recorded. To avoid interference from non-cyclic traffic, the detection node filters non-cyclic packets based on packet identifier, source address, destination address, and priority. For each device, the maintenance length is... Sliding window, This indicates the number of most recently processed messages that participated in subsequent calculations. The preprocessing results are written to the local circular buffer in chronological order, along with the node identifier and device pair identifier.
[0035] Step S3: Calculation of local joint deviation For the first in the sliding window Calculate the normalized periodicity deviation for each cyclic message. ,in The periodic deviation obtained in step S2, The nominal cycle period. Calculate the normalized length deviation. ,in For data length deviation, Given the nominal data length. Calculate the normalized periodic deviation change. ,in This represents the periodic offset of the previous cycle of messages. The joint offset is defined as: in, , , These are non-negative weighting coefficients, representing the proportions of the changes in normalized periodicity, normalized length, and normalized periodicity in the joint bias, respectively. , , It can automatically set according to the device type, for example, increasing the size of time-sensitive real-time devices. Increased number of devices sensitive to data integrity .
[0036] To adapt to industrial circular communication scenarios, when the message identifier indicates that the message is a high-priority real-time message, Take the higher value; when the message is a diagnostic or parameter message, Take the higher value. (Calculated) The sequence is stored locally and is not sent directly to other nodes.
[0037] Step S4: Generating low-dimensional trend statistics using a sliding window The detection node will have a length of The sliding window is divided into Each segment contains [number] consecutive segments. A joint deviation quantity, satisfying For the first Segment, calculate the mean within the segment: in, The joint deviation obtained in step S3 For the first Number of messages within a segment. Calculate the slope within the segment: in, The segment message sequence number, This represents the average of the sequence numbers within the segment. This is the joint deviation within the segment. This represents the average joint deviation within the segment. Calculate the percentage of segments exceeding the threshold: in, To ensure a uniform threshold for over-threshold configuration, all detection nodes use the same threshold. ; For indicator functions, when Set the value to 1 if the condition is met, otherwise set it to 0. Combine the features of each segment into a low-dimensional trend statistic: in, Indicates the first The trend statistics of each detection node in the current sliding window, with a length of This statistic does not include the original message, address, or payload content. The detection node uses the communication key pair generated in step S1. Calculate the message authentication code and append the node identifier, device pair identifier, and time window number.
[0038] Step S5: Consistency verification of adjacent detection nodes Each detection node, according to its neighbor node list, generates the trend statistics generated in step S4. Send to neighboring detection nodes. For nodes... and nodes Calculate the statistical distance: in, and They are nodes and nodes In the statistics of the first One element, This represents the total number of elements in the statistical measure. Node Collect data with all neighboring nodes. Calculate the median And the median of absolute deviation: in, For all the median of The median of the absolute deviation. The adaptive consistency threshold is defined as: in, This is an adjustment coefficient used to control the stringency of consistency checks; This is a constant used to convert the median of the absolute deviation to the standard deviation under the Gaussian distribution assumption. If Then determine the node With nodes The local observations are inconsistent. Node The proportion of inconsistent neighbors to the total number of neighbors is denoted as . .like Exceeding the preset ratio Then the node Marked as suspicious locally. Node Local suspicious markers, statistics, and Send to the distributed intrusion detection management unit.
[0039] If the management unit finds that multiple adjacent nodes are marked as locally suspicious, and the statistical distance between these nodes is less than [a certain value], then [the following action is taken]. If it is only a node, it is determined to be a global clock adjustment or network reconfiguration; If it is marked as suspicious locally, it is determined to be targeting a node. Slow offset injection within the coverage area. When the number of neighbors is less than two, the node... Calculated using the median of historical distance and the median of absolute deviation. and .
[0040] Step S6: Slow Deviation Accumulation Detection and Global Fluctuation Differentiation For each device pair, the cumulative maintenance deviation of the detection node is: in, This is the cumulative deviation of the previous cycle of messages. The joint deviation obtained in step S3 This is for adaptive tolerance. Adaptive tolerance is defined as: in, This represents the mean of the joint deviation within the current sliding window. The standard deviation of the joint deviation within the current sliding window. is a non-negative coefficient. When At that time, among them To accumulate alarm thresholds, detection nodes generate candidate abnormal events.
[0041] Then, combining the consistency results of step S5: if the node corresponding to the candidate abnormal event is marked as locally suspicious in step S5, and the inconsistent neighbor ratio... Exceed If the alarm is triggered, an intrusion alarm is generated; if the management unit determines it is a global clock adjustment or network reconfiguration, the nominal period is updated. and reset When a circular communication reconfiguration, device restart, or topology change is detected, the detection node automatically resets the accumulated amount to prevent false alarms.
[0042] Step S7: Alarm Generation, Evidence Recording and Response Control When an intrusion alarm is generated in step S6, the detection node generates an alarm record according to the anomaly type and severity. The alarm record includes the node identifier, device pair identifier, time window number, statistical value of the joint deviation sequence, and cumulative value. Inconsistent neighbor ratio The statistical message authentication code verification result is checked. The detection node sends the alarm record to the distributed intrusion detection management unit, while retaining the statistics and statistical evidence in the local circular buffer.
[0043] The management unit merges alarm records from multiple detection nodes: if the same device shows local suspicious activity on multiple nodes and the neighbors are inconsistent, the alarm level is increased; if only a single node is abnormal, it is marked as an event to be confirmed.
[0044] Response control includes: sending security event notifications to industrial controllers, restricting non-critical communication of suspicious field devices, adjusting switch port mirroring policies, and increasing the sampling window length and consistency check frequency of relevant device pairs.
[0045] Throughout the process, the original message does not leave the local detection node, and adjacent nodes only exchange low-dimensional trend statistics, thereby achieving low-bandwidth and low-latency intrusion detection data analysis in a distributed network.
[0046] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A method for intrusion detection data analysis in a distributed network, characterized by, Includes the following steps: The detection node collects and parses circular packets in the distributed network, extracting the arrival time and data length of each circular packet; Based on the nominal cycle period and nominal data length, calculate the cycle deviation and length deviation respectively, and then combine the cycle deviation, length deviation and the change in cycle deviation after normalization into a joint deviation. Segmented statistics are performed on the joint deviation within a sliding window to generate low-dimensional trend statistics. The low-dimensional trend statistics are sent to adjacent detection nodes, and consistency verification is performed based on the statistical distance between the low-dimensional trend statistics and the adaptive consistency threshold to obtain local suspicious tags. The combined deviation is cumulatively detected. When the cumulative amount exceeds the cumulative alarm threshold and the local suspicious tag meets the preset conditions, an intrusion alarm is generated. An alarm log is generated based on the intrusion alarm, and response control is executed.
2. The method of claim 1, wherein, The detection node is deployed at the mirror port of an industrial switch, an edge gateway, or the communication interface of a field device, and has hardware timestamp capability. The detection node reads the local configuration, identifies the cyclic communication relationship, establishes a set of device pairs, and records the nominal cycle period, nominal data length, allowed data length range, message identifier, and priority for each device pair. The median of the arrival interval is calculated within a number of consecutive nominal cycles, and this median is used as the nominal cycle period.
3. The method of claim 1, wherein, The process of collecting and parsing the cyclic packets includes: parsing the Ethernet frame header, VLAN tag, industrial Ethernet cyclic packet header, and payload length; for cyclic packets belonging to the same device pair, calculating the difference between the current arrival time and the previous arrival time as the arrival interval; filtering non-cyclic packets based on packet identifier, source address, destination address, and priority; maintaining a sliding window for each device pair, and writing the preprocessing results into the local circular buffer in chronological order.
4. The method of claim 1, wherein, The joint deviation is obtained as follows: the absolute value of the period deviation is divided by the nominal cycle period to obtain the normalized period deviation; the absolute value of the length deviation is divided by the nominal data length to obtain the normalized length deviation; the absolute value of the difference between the current period deviation and the previous period deviation is divided by the nominal cycle period to obtain the normalized period deviation change; the normalized period deviation, normalized length deviation, and normalized period deviation change are multiplied by non-negative weighting coefficients and then added together to obtain the joint deviation; wherein, the weighting coefficients are automatically set according to the device type, message priority, or message type.
5. The method for intrusion detection data analysis in a distributed network according to claim 1, wherein, The generation of low-dimensional trend statistics includes: dividing the sliding window into multiple continuous segments; for each segment, calculating the mean of the joint deviation within the segment, the slope of the joint deviation within the segment as the message sequence number changes, and the proportion of the joint deviation within the segment exceeding a uniformly configured threshold; combining the mean, slope, and proportion of each segment in sequence to form a low-dimensional trend statistic; the low-dimensional trend statistic does not contain the original message, address, or payload content, and uses the communication key to calculate the message authentication code.
6. The method for intrusion detection data analysis in a distributed network according to claim 1, wherein, The consistency verification includes: for the current detection node and each neighboring detection node, calculating the L1 distance between their low-dimensional trend statistics; collecting all distances and calculating the median distance and the median absolute deviation; determining an adaptive consistency threshold based on the sum of the median distance and the median absolute deviation; determining that the current detection node is inconsistent with its neighboring detection nodes when the distance exceeds the adaptive consistency threshold; and counting the proportion of inconsistent neighbors to the total number of neighbors. When the proportion exceeds a preset proportion, marking the current detection node as a local suspicious node.
7. The method of claim 6, wherein, Also includes: The current detection node sends local suspicious markers, low-dimensional trend statistics, and inconsistent neighbor ratios to the distributed intrusion detection management unit; If the management unit finds that multiple neighboring nodes are marked as locally suspicious, and the distance between the low-dimensional trend statistics of these nodes is less than the adaptive consistency threshold, it determines that it is a global clock adjustment or network reconfiguration; if only the currently detected node is marked as locally suspicious, it determines that it is a slow bias injection targeting the coverage area of the currently detected node.
8. The method of claim 1, wherein, The cumulative detection includes: maintaining the cumulative deviation amount, where the current cumulative deviation amount is equal to the larger of the previous cumulative deviation amount plus the joint deviation amount minus the adaptive tolerance and zero; the adaptive tolerance is determined based on the mean and standard deviation of the joint deviation amount within the current sliding window; generating a candidate abnormal event when the cumulative deviation amount exceeds the cumulative alarm threshold; generating an intrusion alarm when preset conditions are met by combining local suspicious markers and inconsistent neighbor ratios; and updating the nominal cycle period and resetting the cumulative deviation amount when it is determined to be a global clock adjustment or network reconfiguration.
9. The method for intrusion detection data analysis in a distributed network according to claim 1, wherein, The alarm record includes node identifier, device pair identifier, time window number, statistical value of joint deviation sequence, cumulative deviation, inconsistent neighbor ratio, and message authentication code verification result; the response control includes at least one of the following: sending a security event notification to the industrial controller, restricting non-critical communication of suspicious field devices, adjusting the switch port mirroring policy, increasing the sampling window length and consistency verification frequency of the relevant device pair.
10. An intrusion detection data analysis system in a distributed network, characterized by include: The message acquisition module is used to collect and parse circular messages in a distributed network, and extract the arrival time and data length of each circular message. The local joint deviation calculation module is used to calculate the period deviation and length deviation according to the nominal cycle period and nominal data length, and then combine the period deviation, length deviation and the change in period deviation after normalization into a joint deviation. The trend statistics generation module is used to perform segmented statistics on the joint deviation within a sliding window to generate low-dimensional trend statistics. The neighbor consistency verification module is used to send low-dimensional trend statistics to neighboring detection nodes and perform consistency verification based on the statistical distance between the low-dimensional trend statistics and the adaptive consistency threshold to obtain local suspicious tags. The deviation accumulation detection module is used to accumulate and detect the combined deviation. When the accumulated amount exceeds the accumulated alarm threshold and the local suspicious tag meets the preset conditions, an intrusion alarm is generated. The alarm response module is used to generate alarm records based on intrusion alarms and to execute response controls.
