Anomaly Traffic Self-Learning Intrusion Detection Device and Method for Industrial Control Networks
By employing a bidirectional temporal convolution module and a multi-head squeezing excitation mechanism in an abnormal traffic intrusion detection model in industrial control networks, the problem of inaccurate detection of complex attacks in existing technologies is solved, achieving higher detection accuracy and reliability.
Patent Information
- Application Number
- CN202611315418.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies for detecting abnormal traffic in industrial control networks rely solely on historical information, resulting in high false negative and false positive rates for complex attacks, making it difficult to accurately detect sudden and bidirectionally correlated attack behaviors.
An abnormal traffic intrusion detection model employing a bidirectional temporal convolution module and a multi-head squeezing excitation mechanism extracts historical and future dependency features from packet length time-series data and performs feature fusion. It also utilizes an dilated causal convolution module to enhance the ability to capture long-term dependencies.
It improves the accuracy of abnormal traffic intrusion detection in industrial control network environments, can more comprehensively characterize traffic timing patterns, reduce false alarm and missed detection rates, and provide reliable technical support for proactive defense of industrial control networks.
Smart Images

Figure CN123122923A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intrusion detection technology, specifically relating to an abnormal traffic self-learning intrusion detection device and method for industrial control networks. Background Technology
[0002] Industrial Control Networks (ICNs) are the core support for modern critical infrastructure (such as power, petrochemical, and intelligent manufacturing). While their deep integration with traditional information networks improves production efficiency, it also introduces serious cybersecurity threats. Because ICNs have extremely high requirements for real-time performance and availability, abnormal traffic (such as denial-of-service attacks, probe scanning, and malicious tampering) can lead to equipment malfunctions, production stoppages, and even major safety incidents if it intrudes. Therefore, accurate and efficient detection of abnormal traffic in industrial control networks is crucial for ensuring industrial production safety and public safety.
[0003] Currently, deep learning-based temporal modeling methods (such as Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Temporal Convolutional Networks (TCNs)) are widely used for anomaly traffic detection. However, existing technologies have significant limitations when processing traffic sequences in industrial control networks: while RNNs and their variants can capture the dependencies between data packets in chronological order, the models are essentially unidirectional temporal models, meaning they can only utilize historical information prior to the current moment (the influence of past data packets on the present). In industrial control networks, attacks often exhibit suddenness and bidirectional correlation; relying solely on historical information leads to the loss of numerous discriminative clues, resulting in persistently high false negative and false positive rates for complex attacks (such as slow probing and low-rate denial-of-service attacks).
[0004] Therefore, how to provide an effective solution to improve the accuracy of abnormal traffic intrusion detection in industrial control network environments has become an urgent problem to be solved in existing technologies. Summary of the Invention
[0005] The purpose of this invention is to provide an abnormal traffic self-learning intrusion detection device and method for industrial control networks, in order to solve the above-mentioned problems existing in the prior art.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an abnormal traffic intrusion detection method for industrial control networks, comprising: Acquire the time-series data of the packet length of the network flow for the target object in the industrial control network. The time-series data of the packet length includes the length data of all packets in the network flow sorted by time. The data packet length time series data is standardized to obtain standardized data packet length time series data. The standardized packet length time series data is used as input to a pre-trained abnormal traffic intrusion detection model to obtain the abnormal traffic intrusion detection result of the network flow. The abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing incentive mechanism module, and a Softmax module. The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module is used to extract historical dependency features of the standardized packet length time-series data, and the backward temporal convolution module is used to extract future dependency features of the standardized packet length time-series data. The historical dependency features are used to characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features are used to characterize the influence of later data on earlier data in the packet length time-series data. The fusion module is used to fuse the historical dependency features and the future dependency features to obtain fused features. The multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules are used to perform channel weighting on the fused features through different channel attention weights to obtain multiple enhanced features. The splicing module is used to splice the multiple enhanced features to obtain spliced features. The Softmax module is used to perform Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results of the network flow.
[0007] In one possible design, the network flow comprises multiple consecutive data packets containing the same 5-tuple, which includes a source IP address, a destination IP address, a source port number, a destination port number, and a transport layer protocol.
[0008] In one possible design, both the forward temporal convolution module and the backward temporal convolution module include multiple sequentially connected temporal convolution residual modules. Each temporal convolution residual module includes a dilated causal convolution module, a weight normalization module, an ELU activation module, and a regularization module. The dilated causal convolution module is used to perform dilated causal convolution operations on the standardized data packet length temporal data. The weight normalization module is used to reparameterize the convolution kernel weights corresponding to the features output by the dilated causal convolution module. The ELU activation module is used to perform ELU activation on the features output by the weight normalization module. The regularization module is used to randomly set each neuron output by the ELU activation module to zero with a preset probability.
[0009] In one possible design, the activation function used by the ELU activation module to perform ELU activation on the features output by the weight normalization module is: , where x represents the feature output by the weight normalization module, and α represents the adjustable parameter.
[0010] In one possible design, the squeezing and activation module includes a convolution module, a squeezing module, an activation module, and a scaling module. The convolution module is used to perform convolution operations on each channel of the fused feature. The squeezing module is used to perform global average pooling on the fused feature after the convolution operation. The activation module is used to perform dimensionality reduction and dimensionality increase on the feature after global average pooling and then perform Sigmoid activation to obtain the attention weights corresponding to each channel. The scaling module is used to multiply each channel of the fused feature by its corresponding weight to obtain the enhanced feature.
[0011] In one possible design, the method further includes: Obtain a training sample set, wherein the training samples in the training sample set are time-series data of historical data packet lengths corresponding to historical network flows for the target object in the industrial control network; Standardize the time-series data of each historical data packet length; The standardized time-series data of each historical data packet length is used as the input sample, and the abnormal traffic intrusion detection results of the historical network flow corresponding to each historical data packet length are used as the output sample. The abnormal traffic intrusion detection model is trained to obtain the trained abnormal traffic intrusion detection model.
[0012] Secondly, the present invention provides an abnormal traffic intrusion detection device for industrial control networks, comprising: The acquisition unit is used to acquire the packet length time-series data of the network flow for the target object in the industrial control network. The packet length time-series data includes the length data of all packets in the network flow sorted by time. A standardization processing unit is used to standardize the data packet length time-series data to obtain standardized data packet length time-series data. An intrusion detection unit is used to take standardized data packet length time-series data as input to a pre-trained abnormal traffic intrusion detection model and perform calculations to obtain the abnormal traffic intrusion detection result of the network flow. The abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing incentive mechanism module, and a Softmax module. The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module is used to extract historical dependency features of the standardized packet length time-series data, and the backward temporal convolution module is used to extract future dependency features of the standardized packet length time-series data. The historical dependency features are used to characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features are used to characterize the influence of later data on earlier data in the packet length time-series data. The fusion module is used to fuse the historical dependency features and the future dependency features to obtain fused features. The multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules are used to perform channel weighting on the fused features through different channel attention weights to obtain multiple enhanced features. The splicing module is used to splice the multiple enhanced features to obtain spliced features. The Softmax module is used to perform Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results of the network flow.
[0013] Thirdly, the present invention provides an electronic device comprising a memory, a processor, and a transceiver connected in sequence and communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the abnormal traffic intrusion detection method for industrial control networks as described in the first aspect or any possible design of the first aspect.
[0014] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the abnormal traffic intrusion detection method for industrial control networks described in the first aspect or any possible design of the first aspect.
[0015] Fifthly, the present invention provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform an abnormal traffic intrusion detection method for industrial control networks as described in the first aspect or any possible design of the first aspect.
[0016] Beneficial effects: This invention discloses an abnormal traffic intrusion detection scheme for industrial control networks, which can improve the accuracy of abnormal traffic intrusion detection in industrial control network environments. Specifically, it acquires the time-series data of the network flow data of the target object in the industrial control network, which includes the length data of all data packets in the network flow sorted by time; then, it performs standardization processing on the time-series data of the data packet length to obtain standardized time-series data of the data packet length; finally, it uses the standardized time-series data of the data packet length as input to a pre-trained abnormal traffic intrusion detection model for calculation to obtain the abnormal traffic intrusion detection result of the network flow; wherein, the abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing excitation mechanism module, and a Softmax module, the bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module, the forward temporal convolution module is used to extract the historical dependency features of the standardized time-series data of the data packet length, the backward temporal convolution module is used to extract the historical dependency features of the standardized time-series data of the data packet length, the backward temporal convolution module is used to extract the historical dependency features of the standardized time-series data of the data packet length, the backward temporal convolution module is used to extract the historical dependency features of the target object, the forward temporal convolution module is used to extract the historical dependency features of the target object ... The system extracts future dependency features from standardized packet length time-series data. Historical dependency features characterize the influence of earlier data on later data within the packet length time-series data, while future dependency features characterize the influence of later data on earlier data within the packet length time-series data. A fusion module fuses the historical and future dependency features to obtain fused features. A multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. These parallel modules apply channel weights to the fused features using different channel attention weights to obtain multiple enhanced features. The splicing module splices these enhanced features to obtain spliced features. A Softmax module performs Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results for the network flow. Thus, when detecting abnormal traffic intrusions in industrial control networks, the constructed bidirectional temporal convolution module extracts the historical and future dependency features of the packet length time series data, thereby fusing bidirectional contextual information from the past and future. This allows for a more comprehensive characterization of the temporal patterns of industrial control network traffic, more thorough capture of temporal features, and improved accuracy in detecting abnormal traffic intrusions, providing reliable technical support for proactive defense of industrial control networks.
[0017] Furthermore, by utilizing the dilated causal convolution module, the receptive field is expanded exponentially while stabilizing gradient propagation, significantly improving the ability to capture long-term dependencies in data packet length sequences. This avoids the problem of gradient vanishing or exploding when processing long sequences, which limits the ability to model long-range dependencies, and further improves the accuracy of abnormal traffic intrusion detection. Attached Figure Description
[0018] Figure 1 A flowchart of an abnormal traffic intrusion detection method for industrial control networks provided in an embodiment of this application; Figure 2 A block diagram of an abnormal traffic intrusion detection device for industrial control networks provided in an embodiment of this application; Figure 3 This is a block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is 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. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.
[0020] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.
[0021] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.
[0022] It should be understood that specific details are provided in the following description to facilitate a complete understanding of the exemplary embodiments. However, those skilled in the art will understand that the exemplary embodiments can be implemented without these specific details. For example, the system may be shown in block diagrams to avoid obscuring the example with unnecessary details. In other instances, well-known processes, structures, and techniques may be shown without unnecessary details to avoid obscuring the exemplary embodiments.
[0023] To improve the accuracy of abnormal traffic intrusion detection in industrial control network environments, this application provides an abnormal traffic self-learning intrusion detection device and method for industrial control networks. This device and method can improve the accuracy of abnormal traffic intrusion detection and provide reliable technical support for proactive defense of industrial control networks.
[0024] The abnormal traffic intrusion detection method for industrial control networks provided in this application can be applied to routing devices, backend terminals or servers specifically designed for abnormal traffic intrusion detection in industrial control networks. It is understood that the execution entity described herein does not constitute a limitation on the embodiments of this application.
[0025] The abnormal traffic intrusion detection method for industrial control networks provided in the embodiments of this application will be described in detail below.
[0026] like Figure 1 The diagram shown is a flowchart of an abnormal traffic intrusion detection method for industrial control networks provided in the first aspect of the embodiments of this application. The abnormal traffic intrusion detection method for industrial control networks may include, but is not limited to, the following steps S101-S103.
[0027] Step S101. Obtain the packet length time-series data of the network flow for the target object in the industrial control network.
[0028] The target object can be any node device in the industrial control network that generates IP (Internet Protocol) traffic.
[0029] The packet length time-series data includes the length data of all packets in the network flow ordered by time. The network flow includes multiple consecutive packets containing the same 5-tuple, where the 5-tuple includes the source IP address, destination IP address, source port number, destination port number, and transport layer protocol. That is, multiple consecutive packets containing the same IP address, destination IP address, source port number, destination port number, and transport layer protocol can be referred to as a network flow.
[0030] For example, if a network flow consists of three consecutive data packets containing the same 5-tuple, and the lengths of these three data packets are 100, 200, and 300 respectively, then the time-series data of the data packet lengths of this network flow can be represented as (100, 200, 300).
[0031] Step S102. Standardize the data packet length time-series data to obtain standardized data packet length time-series data.
[0032] Through standardization, each data point in the packet length time-series data can be mapped to the [0,1] interval, which facilitates subsequent detection of abnormal traffic intrusion using a detection model.
[0033] Step S103. Use the standardized data packet length time series data as input to the pre-trained abnormal traffic intrusion detection model to perform calculations and obtain the abnormal traffic intrusion detection results of the network flow.
[0034] In this embodiment, an abnormal traffic intrusion detection model is pre-trained for detecting abnormal traffic intrusions. After standardizing the packet length time-series data to obtain standardized packet length time-series data, the standardized packet length time-series data can be used as input to the pre-trained abnormal traffic intrusion detection model for computation, resulting in the abnormal traffic intrusion detection result for the network flow. The abnormal traffic intrusion detection result can be the probability of the existence of various types of abnormal traffic intrusions (i.e., the existence of network attacks) and the absence of abnormal traffic intrusions (i.e., the absence of network attacks).
[0035] When training the abnormal traffic intrusion detection model, a training sample set can be obtained first. The training samples in this set are time-series data of historical data packet lengths corresponding to historical network flows targeting a target object in an industrial control network. Then, the time-series data of each historical data packet length is standardized. Finally, the standardized time-series data of each historical data packet length is used as input samples, and the abnormal traffic intrusion detection results of the historical network flows corresponding to each historical data packet length are used as output samples to train the abnormal traffic intrusion detection model, resulting in a trained abnormal traffic intrusion detection model.
[0036] In this embodiment, the abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing excitation mechanism module, and a Softmax module.
[0037] The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module extracts historical dependency features from the standardized packet length time-series data, and the backward temporal convolution module extracts future dependency features from the standardized packet length time-series data. The historical dependency features characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features characterize the influence of later data on earlier data in the packet length time-series data. The fusion module fuses the historical dependency features and the future dependency features (i.e., feature concatenation) to obtain fused features. The multi-head squeezing incentive mechanism module includes a concatenation module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules apply channel weights to the fused features using different channel attention weights to obtain multiple enhanced features. The concatenation module concatenates the multiple enhanced features to obtain concatenated features. The Softmax module performs Softmax classification on the concatenated features to obtain the abnormal traffic intrusion detection results of the network flow.
[0038] Specifically, both the forward temporal convolution module and the backward temporal convolution module include multiple sequentially connected temporal convolution residual modules. Each temporal convolution residual module includes a dilated causal convolution module, a weight normalization module, an ELU activation module, and a regularization module. The dilated causal convolution module performs dilated causal convolution operations on standardized data packet length temporal data, i.e., inserting fixed intervals between convolution kernel elements to exponentially expand the receptive field without increasing the number of parameters, thereby capturing long-range temporal dependencies. The weight normalization module reparameterizes the convolution kernel weights corresponding to the features output by the dilated causal convolution module, decomposing the kernel weights into two learning parameters: length and direction. This reparameterization accelerates model convergence and improves model training stability. The ELU activation module performs ELU activation on the features output by the weight normalization module. The activation function for ELU activation of the features output by the weight normalization module can be expressed as: Here, x represents the feature output by the weight normalization module, and α represents an adjustable parameter used to control the saturation level of the function in the negative region. Its typical value is 1. When x > 0, the function exhibits a linear form, which helps alleviate the gradient vanishing problem. When x ≤ 0, the function value approaches -α. The regularization module is used to randomly set each neuron output by the ELU activation module to zero with a preset probability, preventing overfitting and enhancing the model's generalization ability.
[0039] The squeezing and activation module includes a convolution module, a squeezing module, an activation module, and a scaling module. The convolution module performs convolution operations on each channel of the fused feature. The squeezing module performs global average pooling on the fused feature after convolution. The activation module performs dimensionality reduction and dimensionality increase on the globally average pooled feature, followed by Sigmoid activation to obtain the attention weights corresponding to each channel. Specifically, the activation module can first perform dimensionality reduction on the globally average pooled feature, then perform ReLU activation, then perform dimensionality increase on the ReLU activation to restore the original channel dimension, and finally normalize the channel weight vector using the Sigmoid activation function to obtain the attention weights for each channel. The scaling module multiplies each channel of the fused feature by its corresponding weight to obtain the enhanced feature.
[0040] Multiple parallel squeezing and excitation modules apply channel-weighted summaries to the fused features using different channel attention weights, resulting in multiple enhanced features corresponding one-to-one with each of the parallel squeezing and excitation modules. Then, a concatenation module concatenates these enhanced features to obtain the concatenated features. Finally, a Softmax module performs Softmax classification on the concatenated features to obtain the abnormal traffic intrusion detection results for the network flow, i.e., the probability of the presence or absence of various network attacks.
[0041] In summary, the abnormal traffic intrusion detection method for industrial control networks provided by this invention acquires the time-series data of packet lengths of network flows targeting a target object in the industrial control network. This time-series data includes the lengths of all packets in the network flow ordered by time. The data is then standardized to obtain standardized time-series data. Finally, the standardized time-series data is used as input to a pre-trained abnormal traffic intrusion detection model to obtain the abnormal traffic intrusion detection result of the network flow. The abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing excitation mechanism module, and a Softmax module. The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module is used to extract historical dependency features from the standardized time-series data of packet lengths. The backward temporal convolution module is used to extract future dependency features from the standardized packet length time-series data. The historical dependency features are used to characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features are used to characterize the influence of later data on earlier data in the packet length time-series data. The fusion module is used to fuse the historical dependency features and the future dependency features to obtain fused features. The multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules are used to channel-weight the fused features through different channel attention weights to obtain multiple enhanced features. The splicing module is used to splice the multiple enhanced features to obtain spliced features. The Softmax module is used to perform Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results of the network flow. Thus, when detecting abnormal traffic intrusions in industrial control networks, the constructed bidirectional temporal convolution module extracts historical and future dependency features from the time-series data of packet lengths. This fusion of past and future contextual information allows for a more comprehensive characterization of the temporal patterns of industrial control network traffic, more thorough capture of temporal features, and improved accuracy in detecting abnormal traffic intrusions. This provides reliable technical support for proactive defense of industrial control networks. Furthermore, the dilated causal convolution module, while stabilizing gradient propagation, exponentially expands the receptive field, significantly enhancing the ability to capture long-term dependencies in packet length sequences. This avoids the gradient vanishing or exploding problems that can easily occur when processing long sequences, limiting the ability to model long-range dependencies and further improving the accuracy of abnormal traffic intrusion detection.
[0042] Please see Figure 2 The second aspect of this application provides an abnormal traffic intrusion detection device for industrial control networks, comprising: The acquisition unit is used to acquire the packet length time-series data of the network flow for the target object in the industrial control network. The packet length time-series data includes the length data of all packets in the network flow sorted by time. A standardization processing unit is used to standardize the data packet length time-series data to obtain standardized data packet length time-series data. An intrusion detection unit is used to take standardized data packet length time-series data as input to a pre-trained abnormal traffic intrusion detection model and perform calculations to obtain the abnormal traffic intrusion detection result of the network flow. The abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing incentive mechanism module, and a Softmax module. The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module is used to extract historical dependency features of the standardized packet length time-series data, and the backward temporal convolution module is used to extract future dependency features of the standardized packet length time-series data. The historical dependency features are used to characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features are used to characterize the influence of later data on earlier data in the packet length time-series data. The fusion module is used to fuse the historical dependency features and the future dependency features to obtain fused features. The multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules are used to perform channel weighting on the fused features through different channel attention weights to obtain multiple enhanced features. The splicing module is used to splice the multiple enhanced features to obtain spliced features. The Softmax module is used to perform Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results of the network flow.
[0043] The working process, working details and technical effects of the abnormal traffic intrusion detection device for industrial control networks provided in the second aspect of this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.
[0044] like Figure 3 As shown, a third aspect of this application provides an electronic device, including a memory, a processor, and a transceiver that are sequentially and communicatively connected. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the abnormal traffic intrusion detection method for industrial control networks as described in the first aspect of the application.
[0045] Specifically, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or last-in-first-out (FILO) memory, etc.; the processor may not be limited to microprocessors of the STM32F105 series, ARM (Advanced RISC Machines), x86 architecture processors, or processors with integrated NPU (neural-network processing units); the transceiver may be, but is not limited to, WiFi (Wireless Fidelity) wireless transceivers, Bluetooth wireless transceivers, General Packet Radio Service (GPRS) wireless transceivers, ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard), 3G transceivers, 4G transceivers, and / or 5G transceivers, etc.
[0046] This fourth aspect of the embodiment provides a computer-readable storage medium storing instructions comprising the abnormal traffic intrusion detection method for industrial control networks described in the first aspect of the embodiment. Specifically, the computer-readable storage medium stores instructions that, when executed on a computer, perform the abnormal traffic intrusion detection method for industrial control networks as described in the first aspect. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
[0047] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the abnormal traffic intrusion detection method for industrial control networks as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.
[0048] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for detecting abnormal traffic intrusion in industrial control networks, characterized in that, include: Acquire the time-series data of the packet length of the network flow for the target object in the industrial control network. The time-series data of the packet length includes the length data of all packets in the network flow sorted by time. The data packet length time series data is standardized to obtain standardized data packet length time series data. The standardized packet length time series data is used as input to a pre-trained abnormal traffic intrusion detection model to obtain the abnormal traffic intrusion detection result of the network flow. The abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing incentive mechanism module, and a Softmax module. The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module is used to extract historical dependency features of the standardized packet length time-series data, and the backward temporal convolution module is used to extract future dependency features of the standardized packet length time-series data. The historical dependency features are used to characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features are used to characterize the influence of later data on earlier data in the packet length time-series data. The fusion module is used to fuse the historical dependency features and the future dependency features to obtain fused features. The multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules are used to perform channel weighting on the fused features through different channel attention weights to obtain multiple enhanced features. The splicing module is used to splice the multiple enhanced features to obtain spliced features. The Softmax module is used to perform Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results of the network flow.
2. The abnormal traffic intrusion detection method for industrial control networks according to claim 1, characterized in that, The network flow comprises multiple consecutive data packets containing the same 5-tuple, which includes the source IP address, destination IP address, source port number, destination port number, and transport layer protocol.
3. The abnormal traffic intrusion detection method for industrial control networks according to claim 1, characterized in that, Both the forward temporal convolution module and the backward temporal convolution module include multiple sequentially connected temporal convolution residual modules. Each temporal convolution residual module includes a dilated causal convolution module, a weight normalization module, an ELU activation module, and a regularization module. The dilated causal convolution module is used to perform dilated causal convolution operations on the standardized data packet length temporal data. The weight normalization module is used to reparameterize the convolution kernel weights corresponding to the features output by the dilated causal convolution module. The ELU activation module is used to perform ELU activation on the features output by the weight normalization module. The regularization module is used to randomly set each neuron output by the ELU activation module to zero with a preset probability.
4. The abnormal traffic intrusion detection method for industrial control networks according to claim 3, characterized in that, The activation function of the ELU activation module for the features output by the weight normalization module is: , where x represents the feature output by the weight normalization module, and α represents the adjustable parameter.
5. The abnormal traffic intrusion detection method for industrial control networks according to claim 1, characterized in that, The squeezing and activation module includes a convolution module, a squeezing module, an activation module, and a scaling module. The convolution module performs convolution operations on each channel of the fused feature. The squeezing module performs global average pooling on the fused feature after convolution operations. The activation module performs dimensionality reduction and dimensionality increase on the feature after global average pooling and then performs Sigmoid activation to obtain the attention weights corresponding to each channel. The scaling module multiplies each channel of the fused feature by its corresponding weight to obtain the enhanced feature.
6. The abnormal traffic intrusion detection method for industrial control networks according to claim 1, characterized in that, The method further includes: Obtain a training sample set, wherein the training samples in the training sample set are time-series data of historical data packet lengths corresponding to historical network flows for the target object in the industrial control network; Standardize the time-series data of each historical data packet length; The standardized time-series data of each historical data packet length is used as the input sample, and the abnormal traffic intrusion detection results of the historical network flow corresponding to each historical data packet length are used as the output sample. The abnormal traffic intrusion detection model is trained to obtain the trained abnormal traffic intrusion detection model.
7. An abnormal traffic intrusion detection device for industrial control networks, characterized in that, include: The acquisition unit is used to acquire the packet length time-series data of the network flow for the target object in the industrial control network. The packet length time-series data includes the length data of all packets in the network flow sorted by time. A standardization processing unit is used to standardize the data packet length time-series data to obtain standardized data packet length time-series data. An intrusion detection unit is used to take standardized data packet length time-series data as input to a pre-trained abnormal traffic intrusion detection model and perform calculations to obtain the abnormal traffic intrusion detection result of the network flow. The abnormal traffic intrusion detection model includes a bidirectional temporal convolution module, a multi-head squeezing incentive mechanism module, and a Softmax module. The bidirectional temporal convolution module includes a forward temporal convolution module, a backward temporal convolution module, and a fusion module. The forward temporal convolution module is used to extract historical dependency features of the standardized packet length time-series data, and the backward temporal convolution module is used to extract future dependency features of the standardized packet length time-series data. The historical dependency features are used to characterize the influence of earlier data on later data in the packet length time-series data, and the future dependency features are used to characterize the influence of later data on earlier data in the packet length time-series data. The fusion module is used to fuse the historical dependency features and the future dependency features to obtain fused features. The multi-head squeezing incentive mechanism module includes a splicing module and multiple parallel squeezing and incentive modules. The multiple parallel squeezing and incentive modules are used to perform channel weighting on the fused features through different channel attention weights to obtain multiple enhanced features. The splicing module is used to splice the multiple enhanced features to obtain spliced features. The Softmax module is used to perform Softmax classification on the spliced features to obtain the abnormal traffic intrusion detection results of the network flow.
8. An electronic device, characterized in that, The device includes a memory, a processor, and a transceiver that are sequentially and communicatively connected. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the abnormal traffic intrusion detection method for industrial control networks as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, perform the abnormal traffic intrusion detection method for industrial control networks as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or the instructions are executed by the computer, they implement the abnormal traffic intrusion detection method for industrial control networks as described in any one of claims 1 to 6.
