Block chain enabled low-altitude Internet of Things zero-trust security protection method

Abstract

The invention discloses a blockchain-enabled low-altitude Internet of Things zero-trust security protection system, which comprises a low-altitude airspace module, a zero-trust management and control module and a data storage module, the invention discloses a blockchain-enabled low-altitude Internet of Things zero-trust security protection method. The method comprises the following steps: step 1, deploying a distributed zero-trust component; step 2, setting a low-altitude block chain storage architecture; step 3, performing active trust evaluation on the low-altitude terminal; the method has the characteristics that the zero-trust protection efficiency is improved, the security of cross-airspace data interaction is guaranteed, the security and load problems caused by large-scale low-altitude terminal access are effectively solved, and the storage efficiency and security are optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of low-altitude intelligent Internet, and particularly relates to a zero-trust security protection method for a low-altitude intelligent Internet empowered by blockchain. Background Art

[0002] In the current field of low-altitude intelligent Internet security protection, the following technical problems are mainly faced:

[0003] I. Security risks in the network architecture:

[0004] With the rapid development of the low-altitude economic industry, application scenarios of low-altitude intelligent Internet represented by UAV logistics, urban air mobility (UAM), and low-altitude remote sensing monitoring have been rapidly popularized, resulting in the gradual breakdown of the traditional security boundaries between airspaces. Attackers can break through the boundary protection system through various means such as exploiting flight control system vulnerabilities, false information deception, and malicious node collusion, and then hijack UAV terminals or steal sensitive data, and use "legitimate terminals" as springboards to carry out cross-domain penetration attacks. More seriously, after intruding into the internal network, attackers can hijack other terminals to collaborate in doing evil, and the degree of damage far exceeds that of a single external intrusion, which will pose major security risks to airspace traffic control, operation of critical infrastructure, etc.

[0005] II. Insufficient security protection mechanisms:

[0006] The existing static protection mode based on physical boundaries has obvious deficiencies in the face of lateral penetration attacks caused by compromised terminals inside the low-altitude network. Specifically, it is manifested in:

[0007] 1) The traditional centralized zero-trust deployment architecture will cause a sharp increase in the computing load of the central node during cross-airspace data interaction, and there is a risk of global protection failure caused by a single point of failure;

[0008] 2) The existing protection system mainly adopts a passive defense strategy, relying on protection and disposal after an attack is implemented, lacking the ability to actively identify and block abnormal behaviors of terminals;

[0009] 3) The trust evaluation method depends on linear calculation of historical behaviors, and there is a problem of lag in trust value update, making it difficult to quickly respond to abnormal behaviors of low-altitude terminals;

[0010] 4) There is a lack of a fine-grained access control mechanism, and compromised terminals can use airspace resource sharing to carry out lateral penetration attacks.

[0011] III. Bottleneck in data storage efficiency:

[0012] The high-concurrency and low-latency operating environment of the low-altitude intelligent Internet poses strict requirements for information storage, but the current solutions have:

[0013] 1) The blockchain storage adopts a single-layer chain structure, which is prone to block redundancy when a large number of terminals are connected, resulting in flight control information storage delay and data tampering risks;

[0014] 2) The initial consensus role random assignment method is prone to role misjudgment in the scenario of rapid network formation, affecting the block consensus efficiency.

[0015] Although blockchain technology can theoretically meet the storage requirements of the low-altitude intelligent network due to its characteristics such as decentralization and immutability, the performance of existing implementation solutions in dealing with the access of a large number of low-altitude terminals is still not ideal. Therefore, there is an urgent need to develop a comprehensive solution that can simultaneously solve the problems of security protection efficiency and data storage efficiency. Summary of the Invention

[0016] To overcome the above deficiencies of the prior art, the purpose of the present invention is to provide a blockchain-enabled zero-trust security protection method for low-altitude intelligent networks, solve the internal threat problems faced by low-altitude intelligent networks, and achieve active protection of terminals within low-altitude intelligent networks. At the same time, it meets the high-concurrency access requirements of a large number of terminals in low-altitude intelligent networks, solves the problem of response latency existing in traditional trust assessment schemes when facing abnormal behaviors, constructs a terminal behavior prediction model by integrating a multi-dimensional neural network based on the zero-trust SDP architecture, has the characteristics of improving zero-trust protection efficiency, ensuring the security of cross-airspace data interaction, effectively solving the security and load problems brought by the access of a large number of low-altitude terminals, and optimizing storage efficiency and security.

[0017] To achieve the above purpose, the technical solution adopted by the present invention is:

[0018] A blockchain-enabled zero-trust security protection system for low-altitude intelligent networks, including a low-altitude airspace module, a zero-trust management and control module, and a data storage module;

[0019] The low-altitude airspace module is divided into two categories: flight airspace and flight control data center. Among them, the flight airspace is divided according to the corresponding air traffic control standards for its coverage area, and mainly includes low-altitude aircraft, remote control terminals, and mobile devices; the flight control data center includes cloud computing servers, airspace central servers, and various large key facilities. The flight control data center is responsible for coordinating the normal operation of control facilities within each subordinate airspace; terminals within the low-altitude airspace module are all equipped with SDP zero-trust clients (STCs) to ensure real-time monitoring and protection of the zero-trust security architecture for access terminals;

[0020] The zero-trust control module mentioned above mainly consists of virtualized zero-trust engines at all levels. The virtualized zero-trust engine contains a zero-trust module with the functions of an executable SDP controller and an SDP gateway. Adhering to the concept of "never trust, always verify", the zero-trust control module is mainly responsible for the trust supervision and security isolation of terminals in the low-altitude airspace. The upper-level zero-trust architecture undertakes the supervision task of the lower level. Combining the prediction results of the BiGRU-Attention model for terminal behavior, a trust assessment scheme is designed to accurately quantify the trust level of terminals, and fine-grained access control of terminals is achieved.

[0021] The data storage module divides shards based on the airspace coverage range defined by the low-altitude airspace module. All terminals within the low-altitude airspace module participate in shared storage as blockchain nodes. An initial identity assignment algorithm for blockchain nodes is designed to ensure the consensus efficiency and security of blocks within the shard. At the same time, a two-layer storage architecture for the low-altitude blockchain is adopted, and hierarchical storage and query of information are realized through smart contracts.

[0022] A blockchain-enabled zero-trust security protection method for low-altitude intelligent networks includes the following steps:

[0023] Step 1, Deployment of distributed zero-trust components

[0024] Relying on the existing low-altitude networking layout, optimize the allocation of software and hardware resources in the low-altitude environment, and virtually deploy zero-trust components at all levels in distributed control devices within each domain according to the existing three-level network architecture.

[0025] Step 2, Set up the low-altitude blockchain storage architecture

[0026] The low-altitude blockchain storage architecture mainly consists of a low-altitude node sharding scheme and a two-layer storage structure composed of a global index layer and a shard storage layer;

[0027] Divide the shard range based on the coverage area of the low-altitude airspace module. All zero-trust terminals STC in the low-altitude intelligent network participate in the shared storage of low-altitude information as blocks; through the initial identity assignment algorithm for consensus nodes, the pre-allocation of consensus nodes is realized. Subsequently, each shard selects an appropriate consensus algorithm to construct storage consortium chains FSC1, FSC2,..., FSC n ; After all nodes within the shard reach a consensus, the domain VZE hands over the predicted terminal trust factors to the leading node within the shard, and the leading node initiates a trust storage proposal; each shard only needs to store and update the trust factors of the terminals belonging to its shard, greatly reducing the resource occupation caused by frequent block updates in a large-scale environment; while maintaining the internal terminal trust factors of each shard, the terminal identifiers included in the block header are synchronized with the upper-layer GIC chain through smart contracts; the GIC chain only stores the terminal identifier sets within each shard, and the retrieval of terminal trust factors can be realized through smart contracts.

[0028] Step 3: Active trust assessment of low-altitude terminals

[0029] Relying on the terminal behavior data collected by zero-trust components at all levels, the BiGRU-Attention model is used to analyze and predict terminal trust factors, and then store them on the chain. By retrieving various trust factors stored in the blockchain, active trust assessment is carried out on terminals in low-altitude airspace.

[0030] Before the BiGRU-Attention model is input, the behavior information stream generated by the low-altitude terminal is cleaned, abnormal and invalid data samples are screened, redundant features are removed, and the behavior information stream is converted into normalized vector data;

[0031] The normalized vector data is input into the BiGRU-Attention model for feature learning. The BiGRU-Attention model uses the sliding filtering mechanism unique to the CNN layer to perform local feature scanning on the vector data, realize preliminary feature extraction of the vector data, and obtain feature vectors. The feature vectors are then input into the bidirectional GRU unit and the Attention network in sequence, so that the BiGRU-Attention model can effectively capture long-distance dependencies and enhance the ability to analyze the context of behavioral features. At the same time, the mapping and scaling of the Query-Key-Value matrix of the attention mechanism realizes adaptive weighting of the feature space and realizes automatic focusing on key behavioral features.

[0032] In the output stage of the BiGRU-Attention model, a fully connected layer with ReLU activation function is used to nonlinearly integrate the heterogeneous features output by the convolutional layer, bidirectional GRU unit and attention mechanism to adapt to the final binary classification output mode; the Dropout layer is used to suppress the overfitting risk, and the feature map is converted into the final behavior identifier through the Softmax function; define y i is the terminal behavior identifier predicted by the BiGRU-Attention model. If the BiGRU-Attention model outputs y i = 0, it means that the behavior of terminal i is abnormal. If the output is y i =1, indicating that the terminal behaves normally this time;

[0033] In order to accurately quantify the abnormal conditions of the terminal, the abnormal trust factor set ATF is introduced to represent the abnormal behavior of the terminal predicted by the model, and the abnormal trust factor in the set is defined as {af j |j∈Z *, where j ≤ 8}, and define the attack types in order of the harm degree of abnormal behavior to the low-altitude network as {Reconnaissance, MITM, DDoS, Analysis, Exploits, WebShell, APT, Worms};

[0034] To improve the response speed of the low-altitude zero-trust architecture to compromised terminals, associate the historical behavior attributes of the terminals and the model prediction results to calculate the anomaly factor mf i , when the number of times the terminal is predicted to be abnormal in history approaches or exceeds the set threshold s, different growth strategies will be adopted to dynamically adjust the anomaly factor mf i , the anomaly growth rate MR of terminal i i The calculation formula is:

[0035]

[0036] Among them, atf i represents the cumulative number of times that abnormal trust factor af of terminal i is predicted. Subsequently, through the prediction output y of the model for the behavior of terminal i j and the cumulative number of times of abnormal trust factors atf i adjust mf i , the anomaly factor mf i The calculation formula is: i The calculation formula is:

[0037]

[0038] Among them, w jq represents the weights of various abnormal behaviors in the terminal history, and increases in order according to the sorting of abnormal trust factor af i ; λ jq represents the attenuation step determined according to the abnormal factor; max(af j ) is the maximum value of the serial numbers in the abnormal trust factors. Finally, compare mf i with the preset trusted threshold δ. If mf i < δ, actively determine that the behavior of the terminal this time is abnormal and quickly isolate it; if mf i > δ, determine that this behavior is trustworthy and hand it over to the SDP gateway to perform normal forwarding operations.

[0039] The initial identity assignment algorithm of the consensus node specifically includes the following steps:

[0040] Step 1, the consensus node access rule. Screen each node i in the set of all nodes N, introduce a sliding window method to extract the label sequence predicted for the behavior of node i in the recent k times, and select the terminal nodes that meet the condition C according to the behavior characteristics in the sliding window to participate in the initial identity assignment of the consensus node, as shown in Equation (3):

[0041]

[0042] Among them, n i represents the terminal nodes that meet condition C after screening, and N represents all the terminals in the low-altitude airspace; represents the model prediction label of node i's behavior at the t-th time after multi-dimensional neural network analysis, and the behavior sliding window is represented as θ μ represents the lowest proportion threshold of normal behavior in the k-th behavior of the node; θ σ represents the maximum tolerable number of node behavior mutations, and the mutation of the node behavior is measured by the number of changes in adjacent behaviors within the sliding window ;

[0043] Step 2, Behavior Scoring Algorithm. Calculate the node score Score by comprehensively considering three parts: the historical behavior basis, behavior stability, and the number of abnormal behaviors of the node i , using the node behavior mutation as the dividing line, divide the behavior sliding window Y i into several behavior segments, calculate the correlation degree between the behavior segments, and achieve the precise quantification of the node behavior stability, as shown in Equation (4):

[0044]

[0045] Among them, S d represents the length of the d-th continuous normal behavior segment after dividing Y i into continuous normal behavior segments, l is the pre-set minimum length of the normal behavior segment, m is the total number of normal segments and abnormal segments after division, and a max represents the maximum tolerable number of abnormal behaviors; among them when a i > a max the penalty will increase significantly;

[0046] Step 3, Comprehensive Evaluation and Election of Node Identity. Correlate the result Score of the behavior scoring algorithm with the VRF verifiable random function, while ensuring the fair and random characteristics of the initial node allocation, preventing high-abnormality-rate nodes from obtaining high weights through randomness, and at the same time introducing the node performance weight parameter to meet the high data throughput requirements of the low-altitude network, as shown in Equation (5): i

[0047]

[0048] w i is the comprehensive evaluation score of node i; comprehensively consider the design performance parameters of the node device hardware resources Among them, Fi Indicates the maximum turbo frequency of the CPU carried by the node device, BW i Indicates its maximum transmission bandwidth; VRF(pk i ) represents a verifiable random function generated based on the system public key, VRF(pk i ) ∈ [0, 1];

[0049] Step 4, Initial identity assignment rule: The leading nodes, verification nodes, and ordinary nodes in the initial stage need to respectively meet the following LD, VT, and NM conditions:

[0050]

[0051] Among them, represents the comprehensive evaluation score of the nodes included in set C, w j represents the top three comprehensive evaluation scores within set W, and the corresponding nodes are n j ; \ means removing the specified node from set C.

[0052] The beneficial effects of the present invention are:

[0053] Adopting the technical solution of the present invention has the following remarkable beneficial effects:

[0054] 1) Architecture innovation and improvement of protection efficiency:

[0055] Hierarchically construct a distributed zero-trust security protection architecture of "airspace - control - storage" for low-altitude intelligent networks, with a 43% increase in zero-trust protection energy efficiency and a 2.1-fold improvement in cross-airspace data interaction security;

[0056] Through dynamic adaptive control link optimization, the cross-airspace communication delay is reduced by 38%, and the data anti-tampering verification efficiency is increased by 65%.

[0057] 2) Optimized deployment of distributed SDP zero-trust components:

[0058] Based on the zero-trust component deployment mode innovated by the SDP2.0 architecture, it can achieve the secure access of large-scale low-altitude terminals. Tests show that:

[0059] ■ The single-node access authentication time is shortened to 120 ms;

[0060] ■ The load balancing degree of the component cluster is increased to 90%, and the dynamic expansion efficiency is increased by 75%;

[0061] Relying on the hierarchical deployment strategy, it solves the security authentication bottleneck when accessing a large number of terminals (>100,000 nodes), and the resource consumption is reduced by 56%.

[0062] 3) High efficiency and security of the blockchain hierarchical storage architecture:

[0063] A blockchain hierarchical storage architecture integrating network sharding technology optimizes the trust factor and the shared storage mechanism of flight control information, achieving:

[0064] ■ The storage efficiency is increased by 3.8 times, and the redundant data is reduced by 68%;

[0065] ■ The cross-domain query latency is reduced from 850ms to 230ms, and the consensus time-consuming is reduced by 40%;

[0066] Adopting lightweight sharding technology, the storage security level reaches EAL4+, and the anti-tampering detection accuracy rate > 99.5%.

[0067] 4) Intelligent identity assessment and prediction mechanism:

[0068] Based on the node initial identity assessment mechanism of multi-dimensional neural network, it can achieve:

[0069] ■ The accuracy rate of terminal historical behavior analysis > 97.7%, and the role pre-allocation matching degree reaches 96%;

[0070] ■ Reduce the probability of malicious node mixing by dynamic role adjustment (FPR < 0.5%);

[0071] Combined with the BiGRU-Attention prediction model, the initial trust assessment response time < 80ms.

[0072] 5) Dynamic trust assessment and threat response optimization:

[0073] The trust assessment scheme combining multi-dimensional attribute features and intelligent prediction results has:

[0074] ■ Adaptive trust attenuation mechanism, the abnormal behavior detection accuracy rate is 98.2%;

[0075] ■ The isolation response time of compromised terminals < 100ms, and the success rate of blocking lateral movement of attacks > 98.7%;

[0076] Through real-time trust weight adjustment, the update delay of dynamic protection strategy < 50ms.

[0077] 6) Comprehensive performance indicators:

[0078] Security protection ability: The threat detection rate > 97%, and the zero-day attack defense efficiency is increased by 70%;

[0079] Operating efficiency: The component expansion elasticity reaches 90%, and the resource utilization volatility < ±15%;

[0080] Economic advantages: The hardware deployment cost is reduced by 63%, and the operation and maintenance management efficiency is increased by 4.2 times.

[0081] The present invention has achieved a breakthrough in the field of low-altitude intelligent network security protection, and has significant technical advantages especially in aspects such as terminal trusted authentication, distributed storage optimization, and active threat response, and can be widely applied to key scenarios such as unmanned aerial vehicle control and airspace data sharing.

[0082] Through a large amount of research and analysis, it is found that the traditional security system based on boundary protection is difficult to cope with the dynamic and decentralized operation mode of the low-altitude intelligent network. When the current low-altitude network faces a large number of terminal accesses, it faces the dual challenges of difficult deployment of large-scale distributed components and delayed response to abnormal behaviors, which is extremely likely to lead to the failure of airspace management and trigger security risks of key facilities. The low-altitude zero-trust security architecture integrating blockchain technology and the concept of active defense provides an innovative solution for the high-concurrency and low-latency low-altitude network.

[0083] A blockchain-empowered low-altitude intelligent network zero-trust security system of the present invention. On the basis of the existing low-altitude networking layout, VZE is dynamically deployed according to a three-level structure to achieve elastic expansion and load balancing of zero-trust control components. Secondly, a blockchain double-layer storage architecture is designed. Through the coordinated operation of GIC and FSC and the identity initial allocation algorithm in the consensus stage, the problems of storage redundancy and query delay of the traditional single-chain structure can be effectively solved, and the consensus efficiency can be improved while realizing the efficient retrieval of cross-domain trust factors. Finally, by integrating the BiGRU-Attention behavior prediction model, an active terminal trust evaluation system is constructed to achieve the active identification and blocking of terminal abnormal behaviors, and prevent the further spread of attack behaviors within the intranet.

[0084] In summary, the blockchain-empowered low-altitude intelligent network zero-trust security protection method proposed by the present invention has remarkable effects in terminal behavior control, low-altitude data sharing and storage, and active identification of abnormal terminals, and helps to protect the security of terminals and facilities within the low-altitude intelligent network. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 It is the architecture block diagram of the blockchain-empowered low-altitude intelligent network zero-trust security protection system of the present invention.

[0086] Figure 2 It is the architecture block diagram of the distributed zero-trust component deployment of the present invention.

[0087] Figure 3 It is the low-altitude blockchain storage architecture diagram of the present invention.

[0088] Figure 4 It is the flow chart of the node consensus identity pre-allocation algorithm of the present invention.

[0089] Figure 5 It is the principle block diagram of the BiGRU-Attention model structure of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0090] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

[0091] See Figure 1 , a zero-trust security protection system for low-altitude intelligent Internet empowered by blockchain, which is realized by constructing a zero-trust security architecture that combines "airspace - control - storage", including a low-altitude airspace module, a zero-trust control module, and a data storage module;

[0092] The low-altitude airspace module is divided into two categories: flight airspace and flight control data center. The flight airspace is divided according to the corresponding air traffic control standards for its coverage area, which mainly includes low-altitude aircraft, remote control terminals, and mobile devices; the flight control data center includes cloud computing servers, airspace central servers, and various large key facilities. The flight control data center is responsible for coordinating the normal operation of control facilities within each subordinate airspace; all terminals within the low-altitude airspace module are equipped with SDP zero-trust clients (STC) to ensure real-time monitoring and protection of the zero-trust security architecture for the accessed terminals. The flight airspace includes unmanned aerial vehicle (UAV) terminals and low-altitude base stations. There is at least one low-altitude base station in each airspace, which is responsible for coordinating the flight trajectories and task assignments of UAV terminals within the airspace. At the same time, to meet the flexible networking requirements, the low-altitude base station can reassign the management rights of some UAV terminals to meet the needs of small-scale low-altitude services. The flight control data center includes cloud computing servers and data servers, which are responsible for centrally storing low-altitude related data and coordinating the operation of the low-altitude cloud platform;

[0093] The zero-trust control module is mainly composed of virtualized zero-trust engines at all levels. The virtualized zero-trust engine contains a zero-trust module that can execute SDP controllers and SDP gateways; adhering to the principle of "never trust, always verify", the zero-trust control module is mainly responsible for trust supervision and security isolation of terminals within the low-altitude airspace. The upper-level zero-trust architecture undertakes the supervision task of the lower level; combined with the prediction results of the BiGRU-Attention model for terminal behavior, a trust evaluation scheme is designed to accurately quantify the trustworthiness of terminals and achieve fine-grained access control of terminals; Figure 1 The zero-trust engine (VZE) is responsible for implementing zero-trust control over the terminals belonging to the flight airspace within the low-altitude intelligent Internet; analyzing and predicting the behaviors of the terminals belonging to the flight airspace within the low-altitude intelligent Internet through the BiGRU-Attention model, and using the analysis results as the basis for terminal behavior performance for the zero-trust engine to conduct trust evaluation.

[0094] The data storage module is divided into shards based on the airspace coverage area divided by the low-altitude airspace module. All terminals in the low-altitude airspace module participate in shared storage as blockchain nodes. The initial identity allocation algorithm of blockchain nodes is designed to ensure the efficiency and security of block consensus within the shard. At the same time, a low-altitude blockchain double-layer storage architecture is adopted, FSC1, FSC2, ..., FSC n Hierarchical storage and query of information are achieved through smart contracts. Figure 1 The low-altitude storage part uses blockchain network sharding technology to divide the low-altitude environment into several shards. Each low-altitude airspace maintains a shard for data storage. The index layer and the storage layer are coordinated through smart contracts to realize cross-shard storage and query of data.

[0095] A blockchain-enabled low-altitude intelligent network zero-trust security protection method includes the following steps:

[0096] Step 1: Distributed Zero Trust Component Deployment

[0097] Relying on the existing low-altitude networking layout, the software and hardware resources in the low-altitude environment are optimized and configured, and zero-trust components at all levels are virtually deployed in the distributed control devices in each domain according to the existing three-level network architecture (see Figure 2 ); To clarify the hierarchical deployment structure of the low-altitude zero-trust protection architecture, and at the same time reduce the resource consumption brought by zero-trust deployment, and reduce the acquisition cost brought by physical deployment. The SDP controller and SDP gateway components that can realize zero-trust management and control functions are modularized, and each module is integrated into a zero-trust engine (VZE) in a virtualized manner; the distributed servers and gateway devices at all levels of management and control devices in the domain are pre-configured with zero-trust engines. Each airspace participates in zero-trust hierarchical management and control as the minimum zero-trust execution unit.

[0098] The distributed zero-trust component deployment is based on distributed virtualization and adopts a three-level architecture to perform distributed zero-trust component virtualization deployment, wherein the execution of zero-trust components at each level is as follows:

[0099] 1) In the flight control data center master control mode, trust the engine group {VZE1, VZE2, ..., VZE n}Deployed in the cloud server cluster, activate the SDP controller and SDP gateway module on the VZE; in order to disperse the single-point load, the polling method is used to periodically select the main VZE in the trust engine group, and a disaster recovery plan is set up. Once the main VZE goes down, the business will be quickly transferred to the remaining VZEs, and micro-isolation will be deployed between key servers to prevent the lateral movement of malicious behaviors from causing damage to key facilities; the SDP controller in the master control VZE, on the basis of executing the traditional policy issuance and dynamic access control core tasks, assumes the responsibility of supervising the VZE in each airspace, and implements zero-trust management and control of the access terminals at the flight control data center level;

[0100] 2) In airspace control, VZE is redundantly deployed in distributed servers and network forwarding devices with certain computing capabilities within the domain, and a disaster recovery plan is set up to activate the SDP controller and SDP gateway module in VZE; the SDP gateway module of VZE in the low-altitude airspace monitors and collects the behavior data of terminals within the domain in real time, and then hands it over to the SDP controller module. The BiGRU-Attention model installed on the SDP controller predicts the malicious behavior of terminals, evaluates the trustworthiness of terminals, and then actively blocks threats; while VZE in the low-altitude airspace conducts zero-trust supervision on terminals in the airspace, it also needs to coordinate cross-domain requests of terminals.

[0101] 3) To meet the demand for small-scale flexible networking of UAV terminals in the low-altitude environment, a domain manager component is introduced and virtualized and integrated into the STC terminal (SDP zero-trust client). The terminal acting as a mobile base station can be loaded as needed. The domain manager implements separate supervision on the UAV terminals under its jurisdiction, and the behavior of the terminal acting as the domain manager is supervised by VZE in the upper-level airspace; as an edge control component of the zero-trust architecture, the domain manager can assist in sharing the control pressure of domain VZE and enhance the decentralization of the low-altitude zero-trust architecture.

[0102] Step 2: Set up the low-altitude blockchain storage architecture

[0103] See Figure 3 , the low-altitude blockchain storage architecture mainly consists of a low-altitude node sharding scheme and a two-layer storage structure composed of a global index layer and a sharded storage layer;

[0104] The sharding range is divided by the coverage area of the low-altitude airspace module. All STCs in the low-altitude intelligent network act as blocks to participate in low-altitude information sharing and storage; through the consensus node initial identity assignment algorithm, the pre-allocation of consensus nodes is realized, and then a suitable consensus scheme is selected to construct the consortium chain FSC (FSC1, FSC2,..., FSC n ); after all nodes within the shard reach a consensus, the domain VZE hands over the predicted terminal trust factors to the leading node within the shard, and the leading node initiates a trust storage proposal; each shard only needs to store and update the trust factors of the terminals belonging to its shard, greatly reducing the resource occupation caused by frequent block updates in a large-scale environment; while maintaining the internal terminal trust factors of each shard, the terminal identifiers included in the block header are synchronized with the upper-level GIC chain through smart contracts; the upper-level GIC chain only stores the terminal identifier sets within each shard, and the terminal trust factors can be retrieved through smart contracts.

[0105] Step 3: Active trust assessment of low-altitude terminals

[0106] Relying on the terminal behavior data collected by zero-trust components at all levels, the BiGRU-Attention model is used to analyze and predict terminal trust factors, which are then stored on the chain. By retrieving various trust factors stored in the blockchain, proactive trust assessment is implemented for terminals in the low-altitude airspace;

[0107] The BiGRU-Attention active behavior prediction model mainly consists of a CNN layer, a BiGRU layer, and an Attention layer. See Figure 5 , before the input of the BiGRU-Attention model, data cleaning is performed on the behavior information flow generated by low-altitude terminals, abnormal and invalid data samples are screened, and redundant features are removed. Subsequently, one-hot encoding and normalization processing are performed on the behavior information flow, and finally the behavior information flow is transformed into normalized vector data;

[0108] The normalized vector data is input into the BiGRU-Attention model for feature learning; the BiGRU-Attention model uses the unique sliding filtering mechanism of the CNN layer to perform local feature scanning on the vector data, realizing the initial feature extraction of the vector data and obtaining feature vectors; subsequently, the feature vectors are input into the bidirectional GRU unit and the Attention network in sequence, enabling the BiGRU-Attention model to effectively capture long-distance dependence relationships, enhance the context analysis ability of the feature vectors, and at the same time realize the adaptive weighting of the feature space through the mapping and scaling of the attention mechanism Query-Key-Value matrix, achieving automatic focusing on key behavior features;

[0109] In the output stage of the BiGRU-Attention model, a fully connected layer containing the ReLU activation function is used to perform non-linear integration on the heterogeneous features output by the convolutional layer, the bidirectional GRU unit, and the attention mechanism to adapt to the final binary classification output mode; among them, the Dropout layer is used to suppress the risk of overfitting, and the Softmax function is used to convert the feature mapping into the final behavior identifier; define y i as the terminal behavior identifier predicted by the BiGRU-Attention model. If the BiGRU-Attention model outputs y i =0, it means that the behavior of terminal i is abnormal this time. If the output y i =1, it means that the behavior of the terminal is normal this time;

[0110] To accurately quantify the abnormal situation of the terminal, an abnormal trust factor set ATF is introduced to represent the abnormal behavior of the terminal predicted by the model, and the abnormal trust factors {af j |j ∈ Z *, where \(j\leq8\}, and define the attack types as \{Reconnaissance, MITM, DDoS, Analysis, Exploits, WebShell, APT, Worms\} in sequence according to the harm degree of abnormal behavior to the low-altitude network;

[0111] To improve the response speed of the low-altitude zero-trust architecture to compromised terminals, the historical behavior attributes of the terminals and the model prediction results are associated to calculate the anomaly factor \(mf\) i , when the number of times the terminal's historical prediction is abnormal approaches or exceeds the set threshold \(s\), different growth strategies will be adopted to dynamically adjust the anomaly factor \(mf\) i , the anomaly growth rate \(MR\) of terminal \(i\) i The calculation formula is:

[0112]

[0113] Among them, \(atf\) i represents the cumulative number of times that abnormal trust factor \(af\) of terminal \(i\) is predicted to exist. Subsequently, through the prediction output \(y\) of the model for the behavior of terminal \(i\) j and the cumulative number of times of abnormal trust factors \(atf\) i adjust \(mf\) i , the anomaly factor \(mf\) i The calculation formula is: i The calculation formula is:

[0114]

[0115] Among them, \(w\) jq represents the weights of various abnormal behaviors in the terminal's history, which increase sequentially according to the sorting of abnormal trust factor \(af\) i ; \(\lambda\) jq represents the attenuation step determined according to the abnormal factor; \(max(af\) j ) is the maximum value of the serial numbers in the abnormal trust factors. Finally, \(mf\) i is compared with the preset trusted threshold \(\delta\). If \(mf\) i <\(\delta\), it is actively determined that the terminal's behavior this time is abnormal, and it is quickly isolated; if \(mf\) i >\(\delta\), it is determined that this behavior is trustworthy, and it is handed over to the SDP gateway to perform normal forwarding operations.

[0116] To solve the problem that the random assignment of node identities in the initial link of traditional blockchain consensus may lead to malicious nodes mixing into key roles, thereby causing threats to consensus security such as Sybil attacks and collusion attacks. Based on the output of the BiGRU-Attention model, considering the behavior trustworthiness and stability of the terminals, a blockchain consensus node initial identity assignment algorithm with weak randomness is designed. The blockchain consensus node initial identity assignment algorithm is carried out according to Figure 4The process proceeds. The initial identity assignment algorithm for consensus nodes specifically includes the following steps:

[0117] Step 1, Consensus node admission rule: Screen each node i within the set of all nodes N. Introduce a sliding window method to extract the predicted label sequence for the last k behaviors of node i, aiming to calculate the number of normal terminal behaviors and evaluate the performance of behavior changes. Select the terminal nodes that meet condition C to participate in the initial identity assignment of consensus nodes according to the behavior characteristics within the sliding window, as shown in Equation (1):

[0118]

[0119] where, n i represents the selected terminal nodes that meet condition C, and N represents all terminals within the low-altitude airspace; represents the model prediction label of node i in the t-th behavior after multi-dimensional neural network analysis. The behavior sliding window is expressed as θ μ represents the minimum proportion threshold of normal behaviors among k behaviors of the node; θ σ represents the maximum tolerable number of node behavior mutations. Measure the mutability of node behaviors through the number of changes in adjacent behaviors within the sliding window and determine whether the node can participate in the consensus through threshold comparison;

[0120] Step 2, Behavior scoring algorithm: Calculate the node score Score i by comprehensively considering three parts: the historical behavior basis, behavior stability, and the number of abnormal behaviors of the node. Use the behavior mutation of the node as the dividing line to divide the behavior sliding window Y i into several behavior segments, calculate the correlation degree between the behavior segments, and achieve precise quantification of the node behavior stability, as shown in Equation (2):

[0121]

[0122] where, S d represents the length of the d-th continuous normal behavior segment after dividing Y i into continuous normal behavior segments. l is the pre-set minimum length of the normal behavior segment, m is the total number of normal segments and abnormal segments after division, and a max represents the maximum tolerable number of abnormal behaviors; where When a i > a max non-linear punishment is executed, and the punishment intensity will increase significantly;

[0123] Step 3, Comprehensively evaluate and elect the node identity, and use the result Score of the behavior scoring algorithm iPerform an associated calculation with the VRF verifiable random function to ensure the fair randomness of the initial node allocation while preventing nodes with a high anomaly rate from obtaining high weights through randomness. At the same time, introduce a node performance weight parameter to meet the high data throughput requirements of the low-altitude network, as shown in Equation (3):

[0124]

[0125] w i is the comprehensive evaluation score of node i; Design performance parameters for the comprehensive node device hardware resources where F i represents the maximum turbo frequency of the CPU installed on the node device, and BW i represents its maximum transmission bandwidth; VRF(pk i ) represents the verifiable random function generated based on the system public key, and VRF(pk i ) ∈ [0, 1];

[0126] Step 4, Initial identity allocation rule: The leader nodes, verification nodes, and ordinary nodes in the initial stage need to meet the following LD, VT, and NM conditions respectively:

[0127]

[0128] where, represents the comprehensive evaluation scores of the nodes included in set C, and w j represents the top three comprehensive evaluation scores within set W, and the corresponding nodes are n j ; \ means removing the specified nodes from set C.

[0129] List of abbreviations and definitions:

[0130] SDP: Software Defined Perimeter Software-defined boundary;

[0131] UAV: Unmanned Aerial Vehicle Unmanned aerial vehicle;

[0132] STC: SDP Zero Trust Client SDP zero-trust terminal;

[0133] VZE: Virtualized Zero Trust Engine Virtual zero-trust engine;

[0134] GIC: Global Index Blockchain Global index blockchain;

[0135] FSC:Fragmented Storage Blockchain

Claims

1. A blockchain-enabled zero-trust security protection system for low-altitude intelligent Internet of Things, characterized in that, It includes low-altitude airspace module, zero-trust management and control module, and data storage module; The low-altitude airspace module is divided into two categories: flight airspace and flight control data center. The flight airspace is divided into two categories according to the corresponding air traffic control standards, which mainly includes low-altitude aircraft, remote control terminals, and mobile devices; the flight control data center includes cloud computing servers, airspace hub servers and various large-scale key facilities. The flight control data center is responsible for coordinating the normal operation of control facilities in each lower-level airspace; the terminals in the low-altitude airspace module are all equipped with SDP zero-trust clients to ensure real-time monitoring and protection of access terminals by the zero-trust security architecture; The zero-trust management and control module is mainly composed of virtualized zero-trust engines at all levels, and the virtualized zero-trust engine includes a zero-trust module that can execute the functions of the SDP controller and the SDP gateway; The zero-trust control module is mainly responsible for trust supervision and security isolation of terminals in low-altitude airspace. The upper-level zero-trust architecture undertakes the supervision task of the lower-level. Combined with the prediction results of the BiGRU-Attention model on terminal behavior, a trust evaluation scheme is designed to achieve accurate quantification of the trustworthiness of the terminal and implement fine-grained access control of the terminal. The data storage module is divided into shards based on the airspace coverage divided by the low-altitude airspace module. All terminals in the low-altitude airspace module participate in shared storage as blockchain nodes. An initial identity allocation algorithm for blockchain nodes is designed to ensure the efficiency and security of block consensus within the shard. At the same time, a low-altitude blockchain two-layer storage architecture is adopted to realize hierarchical storage and query of information through smart contracts.

2. A zero-trust security protection method for a low-altitude intelligent Internet of Things empowered by blockchain, characterized in that, The following steps are involved: Step 1: Distributed Zero Trust Component Deployment Relying on the existing low-altitude networking layout, optimize the configuration of software and hardware resources in the low-altitude environment, and virtually deploy zero-trust components at all levels in distributed management and control devices in each domain according to the existing three-level network architecture; Step 2: Set up low-altitude blockchain storage architecture The low-altitude blockchain storage architecture is mainly composed of a low-altitude node sharding solution and a global index layer and a sharding storage layer, with a double-layer storage structure; Divide the sharding range based on the coverage area of the low-altitude airspace module. All zero-trust terminals STC within the low-altitude intelligent network are used as blocks to participate in low-altitude information sharing and storage; through the consensus node initial identity allocation algorithm, pre-allocate the consensus nodes, and then each shard selects an appropriate consensus algorithm to construct the consortium chains FSC1, FSC2, …, FSC n ; after all nodes within the shard reach a consensus, the VZE within the domain submits the predicted terminal trust factors to the leading node within the shard, and the leading node initiates a trust storage proposal; each shard only needs to store and update the trust factors of the terminals belonging to its shard, greatly reducing the resource occupation caused by frequent block updates in a large-scale environment; while maintaining the trust factors of its internal terminals, each shard synchronizes the terminal identifiers included in the block header with the upper-layer GIC chain through smart contracts; the upper-layer GIC chain only stores the set of terminal identifiers within each shard, and can retrieve the terminal trust factors through smart contracts; Step 3: Active trust assessment of low-altitude terminals Relying on the terminal behavior data collected by zero-trust components at all levels, the BiGRU-Attention model is used to analyze and predict terminal trust factors, and then store them on the chain. By retrieving various trust factors stored in the blockchain, active trust assessment is carried out on terminals in low-altitude airspace. Before the BiGRU-Attention model is input, the behavior information stream generated by the low-altitude terminal is cleaned, abnormal and invalid data samples are screened, redundant features are removed, and the behavior information stream is converted into normalized vector data; Input the normalized vector data into the BiGRU-Attention model for feature learning; The BiGRU-Attention model uses the unique sliding filtering mechanism of the CNN layer to perform local feature scanning on vector data, realizing the preliminary feature extraction of vector data and obtaining feature vectors. Subsequently, the feature vectors are sequentially input into the bidirectional GRU unit and the Attention network, enabling the BiGRU-Attention model to effectively capture long-distance dependencies and enhance the ability to analyze the context of behavior features. At the same time, through the mapping and scaling of the Query-Key-Value matrix of the attention mechanism, adaptive weighting of the feature space is achieved, realizing automatic focusing on key behavior features. At the output stage of the BiGRU-Attention model, a fully connected layer containing the ReLU activation function is used to perform non-linear integration on the heterogeneous features output by the convolutional layer, the bidirectional GRU unit, and the attention mechanism to adapt to the final binary classification output mode; among which, a Dropout layer is adopted to suppress the risk of overfitting, and the feature map is converted into the final behavior identifier through the Softmax function; define y i as the terminal behavior identifier predicted by the BiGRU-Attention model. If the BiGRU-Attention model outputs y i = 0, it means that the behavior of terminal i is abnormal this time. If the output y i = 1, it means that the behavior of the terminal is normal this time; To accurately quantify the abnormal conditions of the terminal, an abnormal trust factor set ATF is introduced to represent the abnormal behaviors of the terminal predicted by the model. The abnormal trust factors in the set are defined as {af j |j ∈ Z * , j ≤ 8}, and the attack types are defined in sequence according to the harm degree of the abnormal behavior to the low-altitude network as {Reconnaissance, MITM, DDoS, Analysis, Exploits, WebShell, APT, Worms}; To improve the response speed of the low-altitude zero-trust architecture to compromised terminals, the historical behavior attributes of the associated terminals and the model prediction results are correlated to calculate the anomaly factor mf i When the number of times the historical prediction of the terminal is abnormal approaches or exceeds the set threshold s, different growth strategies will be adopted to dynamically adjust the anomaly factor mf i The anomaly growth rate MR of terminal i i The calculation formula is as follows: Among them, atf i represents the number of times that the abnormal trust factor af j is predicted to exist in the terminal i cumulatively. Subsequently, the predicted output y i of the behavior of the terminal i and the cumulative number of times of the abnormal trust factor atf i are used to adjust mf i . The abnormal factor mf i is calculated by the formula: Among them, w jq represents the weights of various abnormal behaviors in the terminal history, and increases sequentially according to the abnormal trust factor af i ; λ jq represents the attenuation step determined according to the abnormal factor; max(af j ) is the maximum value of the serial numbers in the abnormal trust factor. Finally, mf i is compared with the preset trust threshold δ. If mf i <δ, it is actively determined that the terminal's behavior this time is abnormal, and it is quickly isolated; if mf i >δ, it is determined that this behavior is trustworthy and handed over to the SDP gateway to perform normal forwarding operations.

3. The zero-trust security protection method for a low-altitude intelligent Internet of Things empowered by blockchain according to claim 2, characterized in that, The initial identity assignment algorithm for consensus nodes specifically includes the following steps: Step 1, the consensus node access rule: Screen each node i within the set of all nodes N. Introduce a sliding window method to extract the predicted label sequence for the last k behaviors of node i. Select the terminal nodes that meet condition C according to the behavior features within the sliding window to participate in the initial identity assignment of consensus nodes, as shown in Equation (3): Among them, n i represents the terminal nodes that meet condition C after screening, and N represents all terminals in the low-altitude airspace; represents the model prediction label of node i in the t-th behavior after multi-dimensional neural network analysis, and the behavior sliding window is expressed as θ μ represents the lowest proportion threshold of normal behavior in the k-th behavior of the node; θ σ represents the maximum tolerable number of node behavior mutations, and the mutation of node behavior is measured by the number of changes in adjacent behaviors within the sliding window ; Step 2: Behavior scoring algorithm. The node score Score is calculated by integrating three parts: the historical behavior basis, behavior stability, and the number of abnormal behaviors of the node i , taking the behavior mutation of the node as the dividing line, the behavior sliding window Y i is divided into several behavior segments, and the correlation degree between the behavior segments is calculated to achieve accurate quantification of the node behavior stability, as shown in Equation (4): Among them, S d represents the length of the d-th consecutive normal behavior segment after dividing Y i into consecutive normal behavior segments, where l is the pre-set minimum length of a normal behavior segment, m is the total number of segments of normal and abnormal segments after division, and a max represents the maximum number of tolerable abnormal behaviors; among them when a i > a max the penalty intensity will increase significantly; Step 3, comprehensively evaluate the identity of the election nodes, and associate the result Score of the behavior scoring algorithm i with the VRF (Verifiable Random Function) for correlation calculation. While ensuring the fair randomness of the initial node allocation, it prevents nodes with a high anomaly rate from obtaining high weights through randomness. At the same time, a node performance weight parameter is introduced to meet the high data throughput requirements of the low-altitude network, as shown in Equation (5): w i is the comprehensive evaluation score of node i; comprehensive node device hardware resource design performance parameters Among them, F i represents the maximum turbo frequency of the CPU carried by the node device, and BW i represents its maximum transmission bandwidth; VRF(pk i ) represents the verifiable random function generated based on the system public key, and VRF(pk i ) ∈ [0, 1]; Step 4, the initial identity assignment rule: The leader node, verification node, and ordinary node in the initial stage need to respectively meet the following LD, VT, and NM conditions: Among them, represents the comprehensive evaluation score of the nodes included in set C, w j represents the top three comprehensive evaluation scores within set W, and the corresponding nodes are n j ; \ means removing the specified node from set C.

Images