Cross-domain Internet of Things access decision-making method and system based on space-time enhanced neural network
By adopting a spatiotemporal and enhanced neural network access decision model in the cross-domain Internet of Things, the problem of relying on trusted third parties and difficulty in dealing with non-independent and homogeneous data in the existing technology is solved, and more efficient and secure cross-domain Internet of Things access decisions is achieved.
Patent Information
- Application Number
- CN202510102925.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-22
- Publication Date
- 2025-05-30
AI Technical Summary
Existing cross-domain IoT access decision-making methods rely on trusted third parties, are vulnerable to attacks, and are difficult to effectively process non-independent and homogeneously distributed data, resulting in performance degradation and security risks.
The cross-domain IoT access decision model based on spatiotemporal enhancement neural network is adopted to obtain spatial and temporal enhancement features through multi-scale convolution operations, residual connections, BiLSTM and Concat functions and other technologies, and aggregate the model parameters of local devices through personalized federated learning modules to balance local optimization and global generalization.
It improves the performance and effectiveness of access decisions of devices when processing timing data, ensures the security of devices, can process non-independent and homogeneous data, and supports cross-domain collaboration while ensuring data privacy, providing an efficient multi-domain collaboration solution.
Smart Images
Figure CN120074877A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of the Internet of Things, and particularly to a cross - domain Internet of Things access decision - making method and system based on a spatio - temporal enhanced neural network. Background Art
[0002] The cross - domain Internet of Things is an Internet of Things application mode aimed at connecting physical devices, systems, and data in different fields through sensors, actuators, and network technologies to achieve cross - domain information sharing, intelligent control, and collaborative operation. This mode provides rich opportunities for various industries and application scenarios. In a cross - domain Internet of Things environment, each field can be a unit within a single entity or different institutions, and each unit or institution contains internal servers and Internet of Things devices. These devices communicate with servers or gateways within the field through the network and establish connections with external servers to achieve cross - domain data exchange and parameter sharing. This connection method provides higher efficiency and broader development space for various industries. For example, in the industrial Internet of Things, cross - domain can refer to multiple manufacturing plants. The production lines of each plant are equipped with sensors, actuators, and other intelligent devices. Extensive communication is required between these plants to better collaborate in manufacturing products and monitor the manufacturing process through sensing devices. These data can be further used for manufacturing optimization decisions.
[0003] Traditional cross - domain Internet of Things access decision - making methods usually rely on trusted third parties to verify requests and make corresponding access decisions. For example, the National Health Information Network has established a virtual alliance in the Internet of Things field in multiple hospitals by providing a trusted third - party platform. Similarly, companies such as SmartThings and Google Home provide cross - platform access services, which are responsible for making access decisions and converting user rules into requests applicable to different fields. BaiL et al. proposed a cross - domain access decision - making method based on a trusted third party and an attribute mapping center, achieving cross - domain collaboration by converting user rules into requests across different fields. Although cross - domain authorization and access control technologies provided by third - party organizations can help mitigate certain security risks, these organizations are still vulnerable to potential attacks and often lack secure access decision - making strategies for cross - domain collaboration.
[0004] To address these challenges, researchers have begun to explore more intelligent and adaptive cross - domain Internet of Things access decision - making methods. Among them, federated learning is considered one of the key technologies to overcome data barriers between domains. It can perform model training without sharing cross - domain data and helps improve the intelligence level.
[0005] However, there are still some deficiencies in the research on cross-domain Internet of Things intelligent dynamic access decision-making. First, existing research uses the federated learning model to solve the data silo problem in cross-domain access. Although this method helps devices in different domains to jointly learn the model, thus avoiding the situation where data cannot be effectively shared and utilized within each domain, the potential risks that may be brought about by data sharing among devices within the same domain have not been fully considered. Second, most current local machine learning models are based on a single model and are difficult to comprehensively capture the spatio-temporal attributes of data. In access traffic classification, spatio-temporal features include the time and space attributes of data, such as the periodicity, trend, and variability of access patterns, as well as the relationships between different sources, destinations, and network nodes. If only relying on a single model, these important spatio-temporal features may not be fully utilized, thus limiting the accuracy and comprehensiveness of the traffic classification model. Third, in the Non-IID data environment, the problem of the decline in federated learning performance has not been given sufficient attention. In practical applications, the data between different domains often has a high degree of heterogeneity, and their distributions, features, and attributes may vary significantly. This data heterogeneity directly affects the performance and accuracy of traditional federated learning algorithms. Summary of the Invention
[0006] To solve the above problems in the prior art, the present invention provides a cross-domain Internet of Things access decision-making method and system based on a spatio-temporal enhanced neural network. The invention makes access decisions for the cross-domain Internet of Things through a cross-domain Internet of Things access decision-making model based on a spatio-temporal enhanced neural network. The cross-domain Internet of Things access decision-making model based on a spatio-temporal enhanced neural network uses multi-scale convolution operations and residual connection neural networks to obtain spatial features, uses BiLSTM for modeling to obtain spatial features, uses the Concat function for feature fusion to obtain spatio-temporal enhanced features, and performs non-linear mapping of the spatio-temporal enhanced features through an output module to output access decision results, improving the performance of devices in processing time-series data and the effectiveness of device access decisions, and ensuring the security of devices; at the same time, the model aggregates the model parameters of multiple local devices through a personalized federated learning module, effectively balancing local optimization and global generalization, and improving the adaptability and cross-domain robustness of the model in cross-domain Internet of Things scenarios. The invention can process non-independent and identically distributed data, and supports cross-domain collaboration on the premise of ensuring data privacy. By integrating spatial, time, and spatio-temporal feature modeling, it provides an efficient multi-domain collaboration solution. To achieve the above object, the technical solution is as follows:
[0007] On the one hand, the present invention provides a cross-domain Internet of Things access decision-making method based on a spatio-temporal enhanced neural network, and the method includes:
[0008] S1. The cross - domain Internet of Things access decision model based on the spatio - temporal enhanced neural network is sent to multiple online local client devices through the global server, and global models of multiple local client devices are obtained;
[0009] S2. According to the local client device, real - time access data is collected through intelligent sensors to obtain an access data set of the local device;
[0010] S3. The access data set of the local device is input into the global model of the local client device, and feature extraction is performed through the spatial feature extraction module in the global model of the local client device to obtain the spatial features of the access data set;
[0011] S4. The access data set of the local device is input into the global model of the local client device, and feature extraction is performed through the time - feature modeling module in the global model of the local client device to obtain the time features of the access data set;
[0012] S5. According to the spatial features and time features of the access data set, fusion is performed through the spatio - temporal feature fusion module in the global model of the local client device to obtain the spatio - temporal enhanced features of the access data set;
[0013] S6. According to the spatio - temporal enhanced features of the access data set, prediction is performed through the output module in the global model of the local client device to obtain the cross - domain Internet of Things access decision result.
[0014] Optionally, the training method of the cross - domain Internet of Things access decision model based on the spatio - temporal enhanced neural network includes:
[0015] S11. The spatio - temporal enhanced neural network model is deployed on the global server and initialized to obtain an initialized cross - domain Internet of Things access decision model;
[0016] S12. The initialized cross - domain Internet of Things access decision model is randomly sent by the global server to multiple online local client devices to obtain multiple initialized local cross - domain Internet of Things access decision models;
[0017] S13. According to the local client device, real - time access data is collected through intelligent sensors to obtain a training data set;
[0018] S14. According to the training data set, a decision result of the training data set is obtained through judgment;
[0019] S15. The training data set is input into the initialized local cross - domain Internet of Things access decision model, multi - scale convolution operations are adopted and residual connections are introduced to obtain the spatial features of the training data set;
[0020] S16. Input the training data set into the initialized local cross - domain Internet of Things access decision model, and use BiLSTM to model the training data set to obtain the time features of the training data set;
[0021] S17. According to the spatial features and time features of the training data set, use the Concat function to fuse the spatial features and time features of the training data set to obtain the spatio - temporal enhanced features of the training data set;
[0022] S18. According to the spatio - temporal enhanced features of the training data set, perform non - linear mapping through a multi - layer fully - connected layer and the ReLU activation function to obtain the model decision result;
[0023] S19. Repeat steps S12 - S18, compare the decision result of the training data set and the model decision result. If the results are the same, save the model parameters to obtain the local training model parameters;
[0024] S110. Transmit the local training model parameters to the global server through the local client device, and aggregate multiple local training model parameters through the personalized federated learning module to obtain the global training model parameters;
[0025] S111. Load the global training model parameters into the spatio - temporal enhanced neural network and the initialized model to obtain a cross - domain Internet of Things access decision model based on the spatio - temporal enhanced neural network.
[0026] Optionally, the access data set includes: source IP, target IP, access time, and access device information.
[0027] Optionally, the spatial feature extraction module includes: the ResInceptNet module.
[0028] Optionally, the ResInceptNet module includes: three Inception modules and four convolutional layers.
[0029] Optionally, the time feature modeling module includes: the BiLSTM module.
[0030] Optionally, the spatio - temporal feature fusion module includes: the Concat function module and the Transformer module.
[0031] Optionally, the output module includes: three Linear functions and two ReLU functions.
[0032] Optionally, the personalized federated learning module includes: the Per - FedAvg algorithm module.
[0033] On the other hand, the present invention provides a cross-domain Internet of Things access decision-making system based on a spatio-temporal enhanced neural network. This system is applied to the cross-domain Internet of Things access decision-making method based on a spatio-temporal enhanced neural network. The system includes:
[0034] A model deployment module for distributing the cross-domain Internet of Things access decision-making model based on a spatio-temporal enhanced neural network to multiple online local client devices through a global server, obtaining global models of multiple local client devices;
[0035] A data acquisition module for collecting real-time access data through intelligent sensors according to the local client device, obtaining an access data set of the local device;
[0036] A spatial feature extraction module for inputting the access data set of the local device into the global model of the local client device, and performing feature extraction through the spatial feature extraction module in the global model of the local client device to obtain spatial features of the access data set;
[0037] A time feature acquisition module for inputting the access data set of the local device into the global model of the local client device, and performing feature extraction through the time feature modeling module in the global model of the local client device to obtain time features of the access data set;
[0038] A spatio-temporal feature fusion module for fusing according to the spatial features and time features of the access data set through the spatio-temporal feature fusion module in the global model of the local client device to obtain spatio-temporal enhanced features of the access data set;
[0039] A decision result output module for predicting according to the spatio-temporal enhanced features of the access data set through the output module in the global model of the local client device to obtain a cross-domain Internet of Things access decision result.
[0040] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects:
[0041] On the one hand, the above solution makes access decisions for cross - domain Internet of Things through a cross - domain Internet of Things access decision model based on a spatio - temporal enhanced neural network. This cross - domain Internet of Things access decision model based on a spatio - temporal enhanced neural network uses multi - scale convolution operations and residual connection neural networks to obtain spatial features, uses BiLSTM for modeling to obtain spatial features, uses the Concat function for feature fusion to obtain spatio - temporal enhanced features, and performs non - linear mapping of spatio - temporal enhanced features through an output module to output access decision results. This improves the performance of devices in processing time - series data and the effectiveness of device access decisions, ensuring the security of devices. On the second hand, the model aggregates the model parameters of multiple local devices through a personalized federated learning module, effectively balancing local optimization and global generalization, and improving the adaptability and cross - domain robustness of the model in cross - domain Internet of Things scenarios. On the third hand, the invention can process non - independent and identically - distributed data, and support cross - domain collaboration on the premise of ensuring data privacy. By integrating spatial, temporal, and spatio - temporal feature modeling, it provides an efficient multi - domain collaboration solution. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0043] Figure 1 It is a flowchart of an embodiment of the cross - domain Internet of Things access decision method based on a spatio - temporal enhanced neural network of the present invention;
[0044] Figure 2 It is a flowchart of the training method of the cross - domain Internet of Things access decision model based on a spatio - temporal enhanced neural network in the embodiment of the cross - domain Internet of Things access decision method based on a spatio - temporal enhanced neural network of the present invention;
[0045] Figure 3 It is a schematic structural diagram of the cross - domain Internet of Things access decision model based on a spatio - temporal enhanced neural network in the embodiment of the cross - domain Internet of Things access decision method based on a spatio - temporal enhanced neural network of the present invention;
[0046] Figure 4 It is a system block diagram of an embodiment of the cross - domain Internet of Things access decision system based on a spatio - temporal enhanced neural network of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0047] The following will describe the technical solutions in the present invention with reference to the drawings.
[0048] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or more advantageous than other embodiments or design solutions. Rather, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two.
[0049] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.
[0050] As Figure 1 shown in the flowchart of the embodiment of the cross-domain Internet of Things access decision-making method based on the spatio-temporal enhanced neural network of the present invention, the present invention provides a cross-domain Internet of Things access decision-making method based on the spatio-temporal enhanced neural network, which is implemented by a cross-domain Internet of Things access decision-making system based on the spatio-temporal enhanced neural network. The method includes:
[0051] S1. Send the cross-domain Internet of Things access decision-making model based on the spatio-temporal enhanced neural network to multiple online local client devices through the global server to obtain global models of multiple local client devices;
[0052] Specifically, as Figure 2 shown in the flowchart of the training method of the cross-domain Internet of Things access decision-making model based on the spatio-temporal enhanced neural network in the embodiment of the cross-domain Internet of Things access decision-making method based on the spatio-temporal enhanced neural network of the present invention and as Figure 3 shown in the structural schematic diagram of the cross-domain Internet of Things access decision-making model based on the spatio-temporal enhanced neural network in the embodiment of the cross-domain Internet of Things access decision-making method based on the spatio-temporal enhanced neural network of the present invention, the training method of the cross-domain Internet of Things access decision-making model based on the spatio-temporal enhanced neural network includes:
[0053] S11. Deploy the spatio-temporal enhanced neural network model to the global server and initialize it to obtain an initialized cross-domain Internet of Things access decision-making model;
[0054] S12. Randomly send the initialized cross-domain Internet of Things access decision-making model to multiple online local client devices through the global server to obtain multiple initialized local cross-domain Internet of Things access decision-making models;
[0055] S13. Collect real-time access data through intelligent sensors according to the local client device to obtain a training data set;
[0056] S14. Obtain the decision result of the training data set through judgment according to the training data set;
[0057] S15. Input the training data set into the initialized local cross-domain Internet of Things access decision model, adopt multi-scale convolution operations and introduce residual connections to obtain the spatial features of the training data set;
[0058] S16. Input the training data set into the initialized local cross-domain Internet of Things access decision model, and use BiLSTM to model the training data set to obtain the temporal features of the training data set;
[0059] S17. According to the spatial features and temporal features of the training data set, use the Concat function to fuse the spatial features and temporal features of the training data set to obtain the spatio-temporal enhanced features of the training data set;
[0060] S18. According to the spatio-temporal enhanced features of the training data set, perform non-linear mapping through a multi-layer fully connected layer and a ReLU activation function to obtain the model decision result;
[0061] S19. Repeat steps S12 - S18, compare the decision result of the training data set with the model decision result. If the results are the same, save the model parameters to obtain the local training model parameters;
[0062] S110. Transmit the local training model parameters to the global server through the local client device, and aggregate multiple local training model parameters through the personalized federated learning module to obtain the global training model parameters;
[0063] The personalized federated learning module includes: a Per-FedAvg algorithm module.
[0064] S111. Load the global training model parameters into the spatio-temporal enhanced neural network and the initialized model to obtain a cross-domain Internet of Things access decision model based on the spatio-temporal enhanced neural network.
[0065] Furthermore, in order to verify the effectiveness of the cross-domain Internet of Things access decision model based on the spatio-temporal enhanced neural network proposed in the present invention, a comparative experiment was conducted. The CI-CIDS2018 and UNSW-NB15 data sets were selected as the research objects.
[0066] Obtain the experimental results of the model provided by the present invention after 400 rounds on the CI-CIDS2018 and UNSW-NB15 datasets. As shown in Table 1, the performance comparison of the model provided by the present invention and other models on the CICIDS2018 dataset, and as shown in Table 2, the performance comparison of the model provided by the present invention and other models on the UNSW-NB15 dataset. It can be seen that the model provided by the present invention shows good performance on both datasets and is superior to other models in terms of accuracy, F-score, and recall rate. On the CICIDS2018 dataset, the accuracy of the model provided by the present invention reaches 99.99%, the F1-score is 99.30%, and the recall rate is 99.23%. Compared with the FedAvg(STE-Net) model, the model provided by the present invention improves by 1.59% in accuracy, 9.7% in F1-score, and 8.43% in recall rate. Similarly, compared with the Fed-anids model, the model provided by the present invention improves by 5.49% in accuracy and 8.7% in F1-score, while compared with the stacked-unsupervisedFL model, the accuracy improves by 1.99%, the F1-score improves by 9.3%, and the recall rate improves by 11.23%. In addition, compared with the STE-PFL model, the model provided by the present invention improves by 0.59%, 7.9%, and 7.93% in accuracy, F1-score, and recall rate respectively. On the UNSW-NB15 dataset, the accuracy of the model provided by the present invention reaches 97.80%, the F1-score is 80.23%, and the recall rate is 79.45%. Compared with the FedAvg(STE-Net) model, the model provided by the present invention improves by 22.6% in accuracy, 41.63% in F1-score, and 34.95% in recall rate. Compared with the FL-SEResNet model, the accuracy of the model provided by the present invention improves by 17.4%, while compared with the STE-PFL model, the model provided by the present invention improves by 4.9%, 20.23%, and 18.85% in accuracy, F1-score, and recall rate respectively.
[0067] Table 1 Performance Comparison of the Model Provided by the Present Invention and Other Models on the CICIDS2018 Dataset
[0068]
[0069] Table 2 Performance Comparison of the Model Provided by the Present Invention and Other Models on the UNSW-NB15 Dataset
[0070]
[0071] S2. According to the local client device, collect real-time access data through intelligent sensors to obtain the access dataset of the local device;
[0072] Specifically, the access data set includes: source IP, destination IP, access time, and access device information.
[0073] S3. Input the access data set of the local device into the global model of the local client device, and perform feature extraction through the spatial feature extraction module in the global model of the local client device to obtain the spatial features of the access data set.
[0074] Specifically, the spatial feature extraction module includes: a ResInceptNet module.
[0075] Specifically, the ResInceptNet module includes: three Inception modules and four convolutional layers.
[0076] Furthermore, when building a traditional CNN, it is necessary to select the filter size of the convolutional layer and the position of the pooling layer. These subjective selections may affect the performance of the network model. The emergence of the Inception module makes up for this defect. The Inception module is designed based on the multi-scale idea. The Inception module realizes multi-level feature extraction by using convolutional operations at different scales. The Inception module contains convolutional layers composed of three different sizes of convolutional kernels (1×1, 3×3, and 5×5) and a max-pooling layer. This design effectively realizes the in-depth extraction of fine-grained and global context feature expressions. The parameters of each channel of the Inception module can be expressed as shown in formula (1).
[0077] (1)
[0078] Among them, the meaning of (1×1×i, 3×3×i) is that the third channel of the Inception module first uses i (1×1)-sized filters to perform feature extraction on the output of the previous layer, and then uses i (3×3)-sized filters to further extract data features.
[0079] S4. Input the access data set of the local device into the global model of the local client device, and perform feature extraction through the time feature modeling module in the global model of the local client device to obtain the time features of the access data set.
[0080] Specifically, the time feature modeling module includes: a BiLSTM module.
[0081] S5. According to the spatial features and time features of the access data set, perform fusion through the spatio-temporal feature fusion module in the global model of the local client device to obtain the spatio-temporal enhanced features of the access data set.
[0082] Specifically, the spatio-temporal feature fusion module includes a Concat function module and a Transformer module.
[0083] Furthermore, the Transformer module is a neural network model based on the self-attention mechanism. The Transformer module has no recurrent and convolutional structures, so it can process each element in the sequence in parallel, accelerating the training and inference speed. The Transformer module consists of two parts: an encoder and a decoder. The encoder converts the input sequence into a series of high-dimensional vectors, and the decoder gradually generates the target sequence based on the output of the encoder and previous predictions. Both the encoder and the decoder are stacked by multiple identical basic modules. The TransformerBlock is the core component of the Transformer model and consists of two parts: the multi-head attention mechanism and the feed-forward neural network. The role of the multi-head attention mechanism is to model the correlations in the input feature map sequence, interact the information between different positions, and thus capture the long-term dependencies in the sequence. The feed-forward network is usually connected after the multi-head attention mechanism, and its role is to further process the feature representation on the basis of the multi-head attention to enhance the expressive power of the network.
[0084] S6. According to the spatio-temporal enhanced features of the access data set, perform prediction through the output module in the global model of the local client device to obtain the cross-domain Internet of Things access decision result.
[0085] Specifically, the output module includes three Linear functions and two ReLU functions.
[0086] As Figure 4 shown in the system block diagram of the embodiment of the cross-domain Internet of Things access decision system based on the spatio-temporal enhanced neural network of the present invention, the present invention provides a cross-domain Internet of Things access decision system based on the spatio-temporal enhanced neural network. This system is applied to the cross-domain Internet of Things access decision method based on the spatio-temporal enhanced neural network. The system includes a model deployment module, a data acquisition module, a spatial feature extraction module, a temporal feature acquisition module, a spatio-temporal feature fusion module, and a decision result output module. Specifically,
[0087] The model deployment module is used to send the cross-domain Internet of Things access decision model based on the spatio-temporal enhanced neural network to multiple online local client devices through the global server to obtain the global models of multiple local client devices;
[0088] The data acquisition module is used to collect real-time access data through intelligent sensors according to the local client device to obtain the access data set of the local device;
[0089] A spatial feature extraction module, which is used to input the access data set of the local device into the global model of the local client device, and perform feature extraction through the spatial feature extraction module in the global model of the local client device to obtain the spatial features of the access data set;
[0090] A time feature acquisition module, which is used to input the access data set of the local device into the global model of the local client device, and perform feature extraction through the time feature modeling module in the global model of the local client device to obtain the time features of the access data set;
[0091] A spatio-temporal feature fusion module, which is used to fuse the spatial features and the time features of the access data set through the spatio-temporal feature fusion module in the global model of the local client device to obtain the spatio-temporal enhanced features of the access data set;
[0092] A decision result output module, which is used to perform prediction through the output module in the global model of the local client device according to the spatio-temporal enhanced features of the access data set to obtain the cross-domain Internet of Things access decision result.
[0093] The present invention provides a cross-domain Internet of Things access decision method and system based on a spatio-temporal enhanced neural network. The invention performs access decision for the cross-domain Internet of Things through a cross-domain Internet of Things access decision model based on a spatio-temporal enhanced neural network. The cross-domain Internet of Things access decision model based on a spatio-temporal enhanced neural network adopts multi-scale convolution operations and a residual connection neural network to obtain spatial features, uses BiLSTM for modeling to obtain spatial features, uses a Concat function for feature fusion to obtain spatio-temporal enhanced features, performs non-linear mapping of the spatio-temporal enhanced features through an output module, and outputs an access decision result, which improves the performance of the device when processing time-series data and the effectiveness of device access decision-making, and ensures the security of the device; at the same time, the model aggregates the model parameters of multiple local devices through a personalized federated learning module, effectively balances local optimization and global generalization, and improves the adaptability and cross-domain robustness of the model in cross-domain Internet of Things scenarios; the invention can process non-independent and identically distributed data, and supports cross-domain collaboration on the premise of ensuring data privacy. By integrating spatial, time, and spatio-temporal feature modeling, it provides an efficient multi-domain collaboration solution.
[0094] It should be understood that the present invention is described by the above embodiments and should not be construed as a limitation on the embodiments and scope of the present invention. As is known to those skilled in the art, various changes or equivalent replacements can be made to these features and embodiments without departing from the spirit and scope of the present invention. Additionally, under the teaching of the present invention, these features and embodiments can be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.
Claims
1. A cross-domain IoT access decision method based on spatiotemporal enhanced neural network, characterized in that: The method comprises: S1. Sending a cross-domain IoT access decision model based on spatiotemporal enhanced neural network to multiple online local client devices through a global server to obtain a global model of multiple local client devices; S2. According to the local client device, collect real-time access data through intelligent sensors to obtain an access data set of the local device; S3, inputting the access data set of the local device into the global model of the local client device, performing feature extraction through a spatial feature extraction module in the global model of the local client device, and obtaining spatial features of the access data set; S4, inputting the access data set of the local device into the global model of the local client device, performing feature extraction through the time feature modeling module in the global model of the local client device, and obtaining the time feature of the access data set; S5, fusing the spatial features of the access data set and the temporal features of the access data set through a spatiotemporal feature fusion module in the global model of the local client device to obtain spatiotemporal enhanced features of the access data set; S6. According to the spatiotemporal enhancement features of the access data set, prediction is performed through the output module in the global model of the local client device to obtain a cross-domain Internet of Things access decision result.
2. According to claim 1, the cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network is characterized in that: The training method of the cross-domain Internet of Things access decision model based on spatiotemporal enhanced neural network includes: S11, deploying the spatiotemporal enhanced neural network model on the global server and initializing it to obtain an initialized cross-domain Internet of Things access decision model; S12, randomly sending the initialized cross-domain Internet of Things access decision model to multiple online local client devices through the global server to obtain multiple initialized local cross-domain Internet of Things access decision models; S13, according to the local client device, collecting real-time access data through intelligent sensors to obtain a training data set; S14, obtaining a decision result of the training data set through judgment according to the training data set; S15, inputting the training data set into the initialized local cross-domain IoT access decision model, adopting multi-scale convolution operation and introducing residual connection to obtain the spatial features of the training data set; S16, inputting the training data set into the initialized local cross-domain IoT access decision model, using BiLSTM to model the training data set, and obtaining the time characteristics of the training data set; S17, according to the spatial features of the training data set and the temporal features of the training data set, fusing the spatial features of the training data set and the temporal features of the training data set through a Concat function to obtain spatiotemporal enhancement features of the training data set; S18, according to the spatiotemporal enhancement features of the training data set, nonlinear mapping is performed through multiple layers of fully connected layers and ReLU activation functions to obtain a model decision result; S19, repeating steps S12 to S18, comparing the decision result of the training data set with the decision result of the model, and if the results are consistent, saving the model parameters to obtain local training model parameters; S110, transmitting the local training model parameters to the global server through the local client device, and aggregating multiple local training model parameters through a personalized federated learning module to obtain global training model parameters; S111. Load the global training model parameters into the spatiotemporal enhanced neural network and the initialized model to obtain a cross-domain Internet of Things access decision model based on the spatiotemporal enhanced neural network.
3. According to claim 1, the cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network is characterized in that: The access data set includes: source IP, target IP, access time and access device information.
4. According to claim 1, the cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network is characterized in that: The spatial feature extraction module includes: a ResInceptNet module.
5. According to claim 4, the cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network is characterized in that: The ResInceptNet module includes: three Inception modules and four convolutional layers.
6. The cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network according to claim 1 is characterized in that: The time feature modeling module includes: a BiLSTM module.
7. The cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network according to claim 1 is characterized in that: The spatiotemporal feature fusion module includes: a Concat function module and a Transformer module.
8. The cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network according to claim 1 is characterized in that: The output module includes: three Linear functions and two ReLU functions.
9. The cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network according to claim 2 is characterized in that: The personalized federated learning module includes: a Per-FedAvg algorithm module.
10. A cross-domain Internet of Things access decision system based on spatiotemporal enhanced neural network, used to implement the cross-domain Internet of Things access decision method based on spatiotemporal enhanced neural network as claimed in any one of claims 1 to 9, characterized in that: The system comprises: A model deployment module is used to send the cross-domain IoT access decision model based on spatiotemporal enhanced neural network to multiple online local client devices through a global server to obtain a global model of multiple local client devices; A data acquisition module, used to collect real-time access data through intelligent sensors according to the local client device to obtain an access data set of the local device; A spatial feature extraction module, used to input the access data set of the local device into the global model of the local client device, perform feature extraction through the spatial feature extraction module in the global model of the local client device, and obtain the spatial features of the access data set; A time feature acquisition module, used to input the access data set of the local device into the global model of the local client device, perform feature extraction through the time feature modeling module in the global model of the local client device, and obtain the time feature of the access data set; a spatiotemporal feature fusion module, configured to obtain spatiotemporal enhanced features of the access data set by fusing the spatial features of the access data set and the temporal features of the access data set through the spatiotemporal feature fusion module in the global model of the local client device; The decision result output module is used to make predictions through the output module in the global model of the local client device according to the spatiotemporal enhancement features of the access data set to obtain a cross-domain Internet of Things access decision result.