AI-based blockchain smart contract cross-domain authentication security enhancement method and system

Abstract

The application relates to the technical field of blockchains and artificial intelligence, and discloses an AI-based cross-domain authentication security enhancement method and system for a blockchain smart contract, wherein the AI-based cross-domain authentication security enhancement method for the blockchain smart contract comprises the following steps: a layered security architecture comprising a perception layer, an analysis layer, a decision layer and an execution layer is constructed, cross-domain authentication data is collected and transmitted to the analysis layer; a threat detection model is realized based on deep learning, feature data collected by the perception layer is received, and threat scores and abnormal indexes are output; a dynamic trust evaluation mechanism is established, the output result of the threat detection model is received, and real-time trust values between entities are calculated; and an adaptive authentication strategy optimization algorithm is designed, the most suitable authentication strategy is dynamically selected based on the trust evaluation result, the application can solve the problem of security authentication between heterogeneous blockchain networks, and an efficient and safe cross-domain authentication solution is provided.

Description

Technical Field

[0001] This invention relates to the fields of blockchain and artificial intelligence technology, and more specifically, to an AI-based method and system for enhancing cross-domain authentication security of blockchain smart contracts. Background Technology

[0002] With the widespread application of blockchain technology in finance, IoT, supply chain, and other fields, the demand for interoperability between different blockchain networks is growing. In this cross-domain scenario, secure authentication of smart contracts has become a key challenge. Existing cross-chain authentication schemes mainly employ mechanisms such as hash locking, relays, and notaries, but these schemes suffer from problems such as protocol incompatibility, low authentication efficiency, and insufficient security guarantees when facing heterogeneous blockchain networks.

[0003] Traditional identity authentication and access control methods face significant challenges in cross-chain environments, primarily manifested in the following ways: a lack of unified standards for identity recognition mechanisms, leading to ineffective mapping between identities on different chains; rigid trust transfer mechanisms, unable to address dynamically changing security needs; and a difficulty in balancing privacy protection and authentication security, often resulting in compromises. Especially when handling the cross-chain transfer of high-value assets or sensitive data, existing solutions fail to provide sufficient security guarantees, posing a risk of authentication vulnerabilities being exploited by attackers.

[0004] Therefore, there is an urgent need for a cross-domain authentication security enhancement method that can adapt to the characteristics of heterogeneous blockchain networks, has dynamic security assessment capabilities, and can intelligently adjust authentication strategies to solve the problems of identity mutual recognition, trust transfer, and dynamic response to security risks in the process of cross-chain smart contract calls, and ensure the security, reliability, and efficiency of cross-domain interactions. Summary of the Invention

[0005] This invention provides an AI-based method and system for enhancing cross-domain authentication security of blockchain smart contracts, addressing technical issues such as protocol incompatibility, difficulty in trust transfer, and insufficient privacy protection in identity mutual recognition and security authentication between different blockchain systems.

[0006] This invention provides an AI-based method for enhancing cross-domain authentication security of blockchain smart contracts, including:

[0007] Construct a layered security architecture comprising a perception layer, an analysis layer, a decision-making layer, and an execution layer, and collect cross-domain authentication data and transmit it to the analysis layer;

[0008] It receives feature data collected by the perception layer, trains a threat detection model through deep learning, and outputs threat scores and anomaly indicators based on the threat detection model.

[0009] Based on the threat score and anomaly indicators output by the threat detection model, a dynamic trust assessment mechanism is established to calculate the real-time trust value between entities.

[0010] Based on the calculated real-time trust value between entities, an adaptive authentication strategy optimization algorithm is designed to dynamically select the optimal authentication strategy.

[0011] Based on the optimized authentication strategy, a cross-chain interoperability security protocol is implemented to perform corresponding cross-chain authentication operations.

[0012] In a preferred embodiment, the perception layer collects authentication requests, network traffic, and node status information through a distributed data acquisition adapter, performs standardized processing to generate a feature vector set, and transmits it to the analysis layer in real time through a message queue system.

[0013] In a preferred embodiment, the threat detection model includes:

[0014] The data preprocessing module performs temporal window construction and spatial feature batch construction on the input features;

[0015] The CNN module extracts the spatial features of the authentication request;

[0016] The LSTM module captures the temporal characteristics of authentication behavior;

[0017] The feature fusion module dynamically adjusts the weights of spatial and temporal features through an attention mechanism to calculate the threat probability output.

[0018] The anomaly detection module integrates the results of anomaly pattern detection, group anomaly detection, context anomaly detection, and sequence anomaly detection.

[0019] In a preferred embodiment, the dynamic trust assessment mechanism includes:

[0020] Trust model initialization and evidence collection: creating trust data structures for participating entities and collecting interaction history;

[0021] The trust evaluation function calculates the trust level by comprehensively considering interaction history, node reputation, network relationships, and real-time behavior.

[0022] A trust decay model is implemented, and an exponential decay function is applied to handle the timeliness of trust.

[0023] A trust transfer mechanism that calculates indirect trust levels through a trust chain;

[0024] Multi-path trust aggregation calculates a comprehensive trust score by combining multiple trust transmission paths.

[0025] In a preferred embodiment, the adaptive authentication strategy optimization algorithm includes:

[0026] The authentication strategy optimization problem is modeled as a Markov decision process, defining the state space, action space, and reward function.

[0027] The reward function comprehensively calculates security, user experience, and system efficiency, and dynamically adjusts the weights according to different scenarios.

[0028] Policy learning is achieved through deep Q-networks, and learning stability is improved by using experience replay and target network techniques.

[0029] A greedy strategy is adopted to balance exploration and exploitation, and the exploration rate is gradually reduced through an exponential decay rule.

[0030] Establish security mechanisms to limit the minimum security level to be used during exploration.

[0031] In a preferred embodiment, the cross-chain interoperability security protocol includes:

[0032] The protocol architecture is designed to construct a three-layer architecture consisting of a trusted relay network layer, a blockchain adaptation layer, and a blockchain network layer.

[0033] The authentication messaging protocol defines a standardized message format and a six-phase authentication process.

[0034] The privacy protection mechanism employs zero-knowledge proof technology to support selective disclosure, attribute proof, and scope proof.

[0035] Cross-chain atomicity is guaranteed, and a two-phase commit protocol and hash time locking contract mechanism are implemented.

[0036] Security measures include distributed key management, anti-replay attack measures, and fault tolerance mechanisms.

[0037] In a preferred embodiment, the AI-based blockchain smart contract cross-domain authentication security enhancement system is used to perform the steps in the above-described AI-based blockchain smart contract cross-domain authentication security enhancement method, including:

[0038] The layered security architecture module is used to build a collaborative system of perception, analysis, decision-making and execution layers;

[0039] The threat detection model module is used to implement a hybrid neural network that integrates CNN and LSTM to detect abnormal behavior;

[0040] The dynamic trust assessment module is used to calculate the real-time trust relationship between entities based on multi-factor trust metrics.

[0041] The authentication strategy optimization module is used to dynamically adjust authentication strategy parameters through reinforcement learning;

[0042] The cross-chain interoperability module is used to implement secure authentication protocols between heterogeneous blockchain networks.

[0043] In a preferred embodiment, the AI-based blockchain smart contract cross-domain authentication security enhancement system further includes:

[0044] The privacy protection and compliance module is used to implement zero-knowledge proofs, federated learning, and secure multi-party computation functions to protect sensitive information; the multi-layered security module is used to provide in-depth defense capabilities for security management, application security, data security, network security, and infrastructure security.

[0045] In a preferred embodiment, the threat detection model module employs a semi-supervised learning approach, simultaneously training with both labeled and unlabeled data, and using a loss function that combines cross-entropy loss, parameter regularization, and consistency loss.

[0046] In a preferred embodiment, the cross-chain interoperability module uses a plug-in adapter to support multiple blockchain platforms, ensures the reliability of the relay network through fault-tolerant consensus, and implements multi-path routing and degradation service strategies for authentication requests.

[0047] The beneficial effects of this invention are as follows:

[0048] It achieves intelligent perception, analysis, and decision-making in the cross-domain authentication process. Compared with existing technologies, it improves the rejection rate of abnormal authentication requests while maintaining the pass rate of normal authentication requests, effectively balancing security and ease of use.

[0049] The constructed adaptive authentication mechanism can dynamically adjust the authentication strategy based on real-time risk assessment, reducing the authentication response time to less than 200ms in low-risk scenarios, while automatically upgrading the security level in high-risk scenarios. This intelligent security protection strategy enhances the security and adaptability of blockchain smart contracts in cross-domain applications. Attached Figure Description

[0050] Figure 1 This is a flowchart of the AI-based blockchain smart contract cross-domain authentication security enhancement method of the present invention. Detailed Implementation

[0051] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0052] At least one embodiment of the present invention discloses an AI-based method for enhancing cross-domain authentication security of blockchain smart contracts, such as... Figure 1As shown, it includes the following steps:

[0053] Step 100: Construct a layered security architecture including a perception layer, an analysis layer, a decision layer, and an execution layer; collect cross-domain authentication data and transmit it to the analysis layer.

[0054] This step adopts a layered design approach, constructing a security architecture comprising four core layers: the perception layer, the analysis layer, the decision-making layer, and the execution layer, as detailed below:

[0055] The perception layer is responsible for collecting various types of data during the cross-domain authentication process, including user authentication requests, network traffic characteristics, node status information, etc.; the specific implementation method is as follows:

[0056] Deploy distributed data acquisition adapters to support log interfaces and event subscription mechanisms for different blockchain networks;

[0057] The collected raw data is standardized to generate a structured feature vector set:

[0058] D = {d1, d2, ..., d} n};

[0059] Where, d i Let represent the i-th eigenvector, and n represent the number of eigenvectors;

[0060] Each feature vector d i Includes a unified metadata field: timestamp t i Data source IDs i , feature type i eigenvalue i and associated IDr i ;

[0061] The processed feature vector set is transmitted to the analysis layer in real time through a message queue system (such as Kafka or RabbitMQ);

[0062] The analysis layer runs threat detection models and trust assessment algorithms to perform real-time analysis and processing of the data collected by the perception layer. The specific implementation is as follows:

[0063] Receive the feature vector set D from the perception layer, and group and aggregate it according to feature type and associated ID;

[0064] Appropriate preprocessing modules are applied to different types of features, including feature selection, missing value handling, normalization, and outlier filtering.

[0065] The preprocessed features are input into the threat detection model and trust assessment algorithm for real-time analysis.

[0066] Generate an analysis result set R = {r1, r2, ..., r m};

[0067] Where, r i This represents the i-th analysis result, and m represents the number of analysis results; each result includes indicators such as threat score, anomaly type, and credibility.

[0068] The analysis results are pushed to the decision-making level in real time via an event bus;

[0069] The decision-making level makes authentication decisions and protection responses based on the analysis results and preset security policies. The specific implementation method is as follows:

[0070] Receive the analysis result set R pushed by the analysis layer, and perform correlation analysis with historical decision data and the current system status;

[0071] The application strategy engine determines the optimal authentication decisions and protection measures based on preset rules and learned strategy models;

[0072] Generate a decision instruction set C = {c1, c2, ..., c k};

[0073] Among them, c i This represents the i-th decision instruction, and k represents the number of decision instructions; each instruction includes operation type, target object, parameter settings, etc.

[0074] Decision instructions are sent to the execution layer via a secure channel;

[0075] The execution layer is responsible for implementing specific authentication operations and security measures, and interfaces with various blockchain networks and smart contract platforms. The specific implementation method is as follows:

[0076] Receive the decision instruction set C sent by the decision-making layer, and distribute it to the corresponding execution module according to the instruction type;

[0077] By using standardized blockchain adaptation interfaces, authentication decisions and protective measures are translated into operations supported by specific blockchain networks;

[0078] Perform authentication operations and security measures, and collect execution results and status information;

[0079] The execution results and newly collected status information are fed back to the perception layer, forming a complete information loop.

[0080] Communication between these four layers utilizes an event bus based on a publish-subscribe pattern, supporting asynchronous communication and real-time push. A unified data exchange format and interface specification are defined between each layer to ensure loosely coupled collaboration among system components. Inter-layer communication is encrypted using the TLS 1.3 protocol and implements a two-way authentication mechanism to guarantee data transmission security.

[0081] Step 200: Receive feature data collected by the perception layer, train a threat detection model through deep learning, and output threat scores and anomaly indicators based on the threat detection model;

[0082] This step integrates a hybrid neural network architecture combining CNN and LSTM for real-time analysis of behavioral patterns and data flow characteristics in cross-domain authentication.

[0083] Data preprocessing and feature engineering:

[0084] Receive the feature vector set D from the perception layer and perform feature extraction and transformation;

[0085] Construct a sliding window sequence X = {x} for time series data t-w+1 ,x t-w+2 ,...,x t The window size w is adaptively adjusted based on time relevance.

[0086] Construct batch samples X based on spatial features batch ={X1,X2,...,X b The dimension is (b,c,h,w), where b is the batch size, c is the number of feature channels, and (h,w) is the dimension of the feature matrix.

[0087] Feature selection algorithms are applied to reduce dimensionality, and principal component analysis (PCA) or autoencoders are used to extract key features;

[0088] The CNN module extracts spatial features of the authentication request:

[0089] f spatial =CNN(X) batch );

[0090] Among them, X batch This represents the batch data of the input authentication request, f spatial The spatial features extracted.

[0091] The specific implementation of the CNN module includes: the input layer receiving the preprocessed feature matrix X. batch The convolutional layer employs a 3-layer design with 3×3 filters of sizes 32, 64, and 128 respectively, a stride of 1, and padding using SAME. Each convolutional layer is followed by batch normalization and ReLU activation. Max pooling layers of size 2×2 with a stride of 2 are used. The final output feature map is then processed by global average pooling to obtain a fixed-dimensional spatial feature vector f. spatial ;

[0092] The LSTM module captures the temporal characteristics of authentication behavior:

[0093] h t =LSTM(x t ,h t-1 );

[0094] Among them, h t Let x represent the hidden state at time t. t h is the input authentication feature vector. t-1 This is the hidden state from the previous moment.

[0095] The specific implementation of the LSTM module includes: the input layer receiving the preprocessed time sequence X = {x t-w+1 ,x t-w+2 ,...,x t Using a bidirectional LSTM network with 128 hidden units, the gating mechanism for capturing the timing-dependent LSTM units of both the forward and backward directions is implemented through the following formula:

[0096] Forgotten Gate: f t =σ(W f ·[h t-1 ,x t ]+b f );

[0097] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i );

[0098] Candidate memory unit: C t =tanh(W C ·[h t-1 ,x t ]+b C );

[0099] Memory unit update: C t =f t ⊙C t-1 +i t ⊙C t ;

[0100] Output gate: o t =σ(W o ·[h t-1 ,x t ]+b o );

[0101] Hidden state: h t =o t ⊙tanh(C t );

[0102] Where σ is the sigmoid activation function, W f Let b be the forget gate weight matrix. f For the bias term of the forget gate; W i Let b be the input gate weight matrix. i For the bias term of the input gate; W C Let b be the weight matrix of the candidate memory cells. C For candidate memory unit bias terms, tanh is the hyperbolic tangent activation function; ⊙ represents element-wise multiplication, C t-1 The state of the memory unit at the previous moment; W o Let b be the output gate weight matrix. o For the bias term of the output gate; h t Output the hidden state at the current time step, and extract the hidden state h at the final time step. t As a representation of temporal characteristics;

[0103] Feature fusion and threat score calculation:

[0104] P threat =σ(W fusion ·[h t ;f spatial ]+b1);

[0105] Among them, P threat The output is the threat probability, σ is the sigmoid activation function, and W is the threat probability output. fusion Here is the weight matrix of the fusion layer, b1 is the feature fusion bias term, [h t ;f spatial [] indicates the combination of temporal and spatial features.

[0106] The specific implementation of feature fusion includes:

[0107] The attention weights α for spatial and temporal features are dynamically adjusted using an attention mechanism:

[0108] α = softmax(v T tanh(W[h t ;f spatial ]));

[0109] Where v is the attention vector parameter, W is the attention weight matrix, tanh is the hyperbolic tangent activation function, softmax function is used to normalize the weights; T represents matrix transpose.

[0110] Weighted feature fusion: f fusion =α⊙[h t ;f spatial ];

[0111] Among them, f fusionα represents the weighted features after fusion; α represents the attention weights for spatial and temporal features; ⊙ represents element-level multiplication;

[0112] The fused features pass through two fully connected network layers:

[0113] The first layer has 64 neurons, and the second layer has 32 neurons, both using the ReLU activation function;

[0114] Use a Dropout layer (dropout rate 0.5) to prevent overfitting;

[0115] Finally, the threat probability score P is output using the Sigmoid function. threat Additional output: anomaly type classification result T anomaly This is used for trust assessment and decision-making in subsequent steps;

[0116] The model training employs a combination of supervised and semi-supervised learning methods, with the loss function being:

[0117] L = L ce +λ1L reg +λ2L con ;

[0118] Among them, L ce For the cross-entropy loss of labeled data, L reg L is the parameter regularization term. con The consistency loss for unlabeled data is represented by λ1 and λ2, which represent the first and second weighting coefficients, respectively.

[0119] In one embodiment of the present invention, the specific implementation of the training process includes:

[0120] Dataset construction: Collect historical authentication data, including labeled normal samples, abnormal samples, and unlabeled samples;

[0121] Data augmentation techniques were used to augment the training set: feature perturbation, temporal transformation, and synthetic minority samples (SMOTE).

[0122] Phase 1: Supervised learning using labeled data to optimize cross-entropy loss L ce and regularization loss L reg ;

[0123] Phase 2: Introduce unlabeled data and use pseudo-labels and consistency regularization for semi-supervised learning;

[0124] The Adam optimizer is used with an initial learning rate of 0.001, and the learning rate is dynamically adjusted using a cosine annealing strategy.

[0125] Use an early stopping strategy to prevent overfitting; stop training when the performance on the validation set no longer improves.

[0126] After the model is trained, a threat detection function F is generated. threat It is used to predict threat scores and anomaly types in real time, and the return value includes the anomaly probability and the corresponding anomaly type identifier;

[0127] In one embodiment of the present invention, the exception types include:

[0128] Credential misuse anomaly: Unauthorized access is performed using stolen legitimate credentials;

[0129] Abnormal behavior patterns: The user's action sequence deviates significantly from historical behavior patterns;

[0130] Anomalies in the environment: Abnormalities in environmental factors such as the source IP address, device characteristics, and geographical location.

[0131] Permission abuse exception: A user attempts to access resources beyond their authorized scope;

[0132] Timing anomalies: Access requests made during unusual time periods;

[0133] Concurrency exception: The same user accesses the service simultaneously from multiple locations.

[0134] Anomaly detection employs a multi-index fusion method to calculate the final anomaly score S. anomaly :

[0135]

[0136] Among them, S i w represents the anomaly score of the i-th anomaly detection mechanism. i Here, N represents the weighting coefficient for the corresponding anomaly, and N is the total number of anomaly detection mechanisms.

[0137] The specific implementation of anomaly detection includes four complementary detection mechanisms:

[0138] Anomaly pattern detection: Maintaining the baseline of entity behavior B e (Baseline data representing the entity's normal behavior pattern), calculate the current behavior b. c The Mahalanobis distance d between (the entity's currently observed behavioral feature vector) and the baseline. M (b c B e ), generate anomaly score S1;

[0139] Group anomaly detection: The DBSCAN clustering algorithm is used to group entities, calculate the outlier degree O(e,G) of an individual relative to its group (representing the degree of outlier of entity e relative to its group G), and generate anomaly score S2;

[0140] Context-sensitive anomaly detection: Construct a context-sensitive conditional probability model P(b|c) to evaluate the degree of anomaly of behavior b in the current context c and generate anomaly score S3;

[0141] Sequence anomaly detection: Analyzing the sequence of the most recent N1 operations using the sliding window technique. Compared with the normal sequence pattern, an anomaly score S4 is generated; where o t N represents the operation at time t, and N1 represents the number of windows;

[0142] Weighting coefficient w i The initial settings are equal, and the system is dynamically adjusted based on the historical accuracy of each detection mechanism during operation.

[0143] Through the above steps, the threat detection model outputs a threat score P. threat Exception type T anomaly and abnormal score S anomaly .

[0144] Step 300: Based on the threat score and anomaly indicators output by the threat detection model, establish a dynamic trust assessment mechanism and calculate the real-time trust value between entities.

[0145] This step uses a dynamic evaluation algorithm based on multi-factor trust metrics, taking into account factors such as historical interactions, node reputation, network relationships, and real-time behavior.

[0146] Trust model initialization and trust evidence collection:

[0147] The system provides each participating entity e i Create a Trust Data Structure (TD) i This includes basic reputation, historical interaction records, relationship network, and behavioral characteristics;

[0148] Based on the threat detection results from step 200, the entity's behavioral characteristics B are updated in real time. i (t), with particular attention to abnormal behavior and the extent of its impact;

[0149] By collecting historical interaction data between entities through a blockchain network, an interaction record set H is constructed. ij ={(t k ,r k ,c k )}, where t k For the interaction time, r k For the interaction result (success / failure), c k For the interaction context;

[0150] Establish an entity relationship network G = (V, E, W), where V is the entity set, E is the relationship edge set, and W is the relationship weight set;

[0151] Periodically obtain the entity's global reputation score R from the reputation system. i (t), this credit rating is calculated based on the comprehensive evaluation of the entity by all parties;

[0152] The trust evaluation function is defined as:

[0153] T ij (t)=α·H ij (t)+β·R i (t)+γ·N ij (t)+δ·B i (t);

[0154] Among them, T ij H(t) represents the trust level of entity i towards entity j at time t. ij (t) represents the historical interaction trust score, R i (t) represents the node reputation, N ij (t) represents the network relation weights, B i (t) represents the real-time behavior score, where α, β, γ, and δ are the third, fourth, fifth, and sixth weight coefficients, respectively, and satisfy α+β+γ+δ=1.

[0155] The calculation methods for each factor are as follows:

[0156] Historical Interaction Trust Score H ij (t) Calculated from the interaction history:

[0157]

[0158] Where, f(r) k ,c k ) is the evaluation function for the interaction results, w k λ3 is the interaction importance weight, λ3 is the time decay factor, M is the number of historical interaction records, and t is the interaction importance weight. k The timestamp of the k-th interaction;

[0159] Node Reputation R i (t) is obtained directly from the global reputation system, with a value range of [0, 1].

[0160] Network relation weight N ij (t) Calculated by analyzing the connections between entities in the relational network:

[0161]

[0162] Among them, w ij C represents the weight of the direct relationship between entities i and j. ij Let |C| be the set of common neighbors connecting i and j. ij | is the number of public neighbors;

[0163] Real-time behavioral score B i (t) Calculated based on the threat detection results of step 200:

[0164]

[0165] Where threshold is a system-defined threshold used to normalize the calculation results, P threat For threat rating, S anomaly These are abnormal scores;

[0166] The initial values ​​of the weighting coefficients α, β, γ, and δ are set to 0.4, 0.2, 0.2, and 0.2, respectively, and are dynamically adjusted according to the system operation and different scenario requirements.

[0167] The trust decay model uses a time-sensitive decreasing function:

[0168]

[0169] Among them, T init λ represents the initial trust value, λ4 is the decay coefficient, and t-t0 is the time interval.

[0170] The specific implementation of the attenuation model includes:

[0171] Different decay coefficients λ4 are set based on the interaction type and importance, with key interactions decaying more slowly and ordinary interactions decaying more quickly.

[0172] A context-sensitive decay adjustment mechanism is introduced to accelerate decay in high-risk environments and slow down decay in low-risk environments;

[0173] Set the trust threshold value T min This ensures that even with prolonged periods of inactivity, the trust level will not fall below this threshold.

[0174] Fluctuations in trust values ​​caused by abnormal interactions are smoothed out to avoid drastic changes in trust values ​​due to a single abnormal event. Trust transfer employs a trust chain calculation method.

[0175] T ik (t)=min{T ij (t),T jk (t)}·d ijk ;

[0176] Among them, T ik (t) represents the degree of indirect trust that entity i has in entity k through intermediate entity j, d ijk T is the trust decay factor. ij (t) represents the trust level of entity i towards entity j, T jk(t) represents the trust level of entity j towards entity k, and min represents the minimum operation.

[0177] The specific implementation of the trust transfer mechanism includes:

[0178] Trust decay factor d ijk Calculation formula: d ijk =1-β1·hop ijk , among which hop ijk This represents the number of hops for trust transfer, where β1 is the attenuation factor per hop.

[0179] Set the maximum trust delivery depth maxDepth; delivery chains exceeding this depth will be truncated.

[0180] For multi-hop propagation, a dynamic programming algorithm is used to calculate the optimal trusted path to avoid redundant calculations.

[0181] Implement a trust transfer loop detection mechanism to prevent trust inflation caused by loop transfer;

[0182] Multi-path trust aggregation:

[0183]

[0184] in Let represent the overall trust level of entity i towards entity k after aggregation, and P represent the set of all possible intermediate nodes. w represents the indirect trust level calculated through node j. j This represents the path weight.

[0185] The specific implementations of multi-path aggregation include:

[0186] Path weight w j Calculation method: w j =T ij (t)·T jk (t)·d ijk This assigns higher weights to more trustworthy intermediary nodes;

[0187] Dijkstra's algorithm is used to identify the top K most reliable paths from entity i to entity k;

[0188] Set trust threshold T threshold Filter out paths below this threshold to reduce noise impact;

[0189] Aggregation using a weighted geometric mean instead of an arithmetic mean imposes a stronger penalty on paths with extremely low trust:

[0190]

[0191] in w represents the overall trust level of entity i over entity k after aggregation using the weighted geometric mean. j ' represents the normalized path weight, ∑w j 'The sum of normalized weights;

[0192] Through the above mechanism, the system calculates a trust value T that is updated in real time for each pair of entities. ij (t).

[0193] Step 400: Based on the calculated real-time trust value between entities, design an adaptive authentication strategy optimization algorithm to dynamically select the optimal authentication strategy.

[0194] This step uses a reinforcement learning algorithm to dynamically adjust the authentication strategy parameters:

[0195] MDP Modeling and State Space Definition:

[0196] The authentication strategy optimization problem is modeled as a Markov decision process (MDP), defining the state space S, action space A, transition function P, reward function R, and discount factor γ1.

[0197] The state space S contains the following key dimensions:

[0198] Inter-entity trust level T from step 300 ij (t);

[0199] Threat score P from step 200 threat and abnormal score S anomaly ;

[0200] The historical behavior feature vector F of the requesting entity hist ;

[0201] Context information C of the authentication request req This includes request type, resource sensitivity, and time characteristics;

[0202] Overall system security status S sys This includes the current attack alert level, system load, etc.

[0203] By extracting features and reducing dimensionality from the high-dimensional state space, a fixed-dimensional state representation s∈R is generated. d ;

[0204] Action space and authentication strategy definition:

[0205] Action space A is defined as the combination of parameters for the authentication policy, including:

[0206] Authentication factor selection vector F auth Determine which combination of certification factors to use;

[0207] Certification strength parameter L auth Set the security strength level for each authentication factor;

[0208] Authentication timeout parameter T auth Determine the maximum waiting time for the authentication operation;

[0209] Authentication frequency parameter F req Set the re-authentication trigger frequency;

[0210] Exception response type R anom : Specifies the response measures when an anomaly is detected;

[0211] To reduce complexity, the continuous action space is discretized into a finite combination of policy levels, resulting in a total of K possible authentication policies;

[0212] The reward function takes into account security, user experience, and system efficiency.

[0213] R(s,a)=w1·Security(s,a)+w2·Usability(s,a)+w3·Efficiency(s,a)

[0214] Where R(s,a) represents the total reward obtained by taking action a in state s, Security(s,a) evaluates the improvement in security, Usability(s,a) evaluates the impact on user experience, Efficiency(s,a) evaluates the change in system efficiency, and w1, w2, w3 are the seventh, eighth, and ninth weight coefficients, respectively, satisfying w1+w2+w3=1.

[0215] The specific calculation methods for each evaluation indicator are as follows:

[0216] Security assessment: Security(s,a)

[0217] Security(s,a) = 1 - P breach (s,a);

[0218] Where P breach (s,a) represents the probability of security risk that may result from taking action a in state s, which is estimated using a threat model.

[0219] User experience evaluation Usability(s,a):

[0220]

[0221] Where C user (a) The user operation cost incurred by action a, C max The maximum acceptable cost;

[0222] System efficiency evaluation: Efficiency(s,a)

[0223]

[0224] Where T proc (a) represents the processing time of strategy a, Resource(a) represents the resource consumption, and T max and R max These represent the maximum acceptable processing time and maximum resource consumption of the system, respectively.

[0225] The reward function weights are dynamically adjusted according to different scenarios:

[0226] High-security scenario: w1 = 0.7, w2 = 0.15, w3 = 0.15;

[0227] Balanced scenario: w1 = 0.5, w2 = 0.3, w3 = 0.2;

[0228] High availability scenario: w1 = 0.4, w2 = 0.4, w3 = 0.2;

[0229] Q-learning algorithm implementation:

[0230] The optimal authentication strategy is learned using the Q-learning algorithm, and the Q-value update formula is:

[0231]

[0232] Where Q(s,a) represents the Q value of the state-action pair (s,a), which represents the long-term expected reward of taking action a in state s, s represents the current system state, a is the authentication policy action taken, r is the immediate reward, s′ represents the next state, a′ represents the possible action in the next state, γ1 is the discount factor, and η is the learning rate.

[0233] Key steps in implementing Q-learning:

[0234] Initialize the Q-table: For all state-action pairs (s, a), set the initial Q-value to 0;

[0235] Set the hyperparameters as follows: learning rate η = 0.1, discount factor γ1 = 0.9;

[0236] In each decision cycle, the system obtains the current state s, and then:

[0237] Choose action a (exploration) with probability ∈ random;

[0238] Choose the action with the highest current Q value with probability 1 - ∈ a = argmax a Q(s,a) (utilization);

[0239] Execute the selected authentication policy a, observe the results, and obtain the reward r and the new state s′;

[0240] Update Q-table: Q(s,a) ← Q(s,a) + α[r + γ1max] a′ Q(s′,a′)-Q(s,a)];

[0241] The experience replay mechanism is used to store (s,a,r,s′) tuples and perform batch updates by random sampling to improve learning efficiency.

[0242] Deep Q-Network (DQN) Implementation:

[0243] To address the high-dimensional state space problem, the system employs a deep Q-network (DQN) to approximate the Q-function and uses empirical replay and target network techniques to improve learning stability.

[0244] DQN network architecture: Input layer: Receives state vector s∈R d The dimension is d;

[0245] Hidden layer 1: 256 neurons, using the ReLU activation function - Hidden layer 2: 128 neurons, using the ReLU activation function;

[0246] Output layer: |A| neurons, corresponding to the Q-value of each possible action;

[0247] DQN training implementation: Use the experience replay buffer D to store interaction samples (s,a,r,s′,done), where done represents a boolean value indicating whether the current round has ended;

[0248] Maintain two networks: the primary network Q and the target network Q′;

[0249] Synchronize the main network parameters to the target network every C steps: θ′←θ;

[0250] Randomly sample a small batch of samples B from the experience replay buffer D;

[0251] For each sample (s, a, r, s′, done), calculate the target value y:

[0252]

[0253] Calculate the loss function:

[0254] Where: L(θ) represents the loss function; |B| represents the batch size; y represents the target Q value; Q(s,a;θ) represents the Q value predicted by the main network; gradient descent is used to update the main network parameters θ;

[0255] Update the main network parameters θ using gradient descent;

[0256] Exploration strategies and model deployment:

[0257] An ε-greedy strategy is adopted to balance exploration and exploitation. The ε value is initially set to 0.3 and gradually reduced to 0.05 according to the exponential decay rule.

[0258] The specific implementation of the exploration strategy: In the initial stage, a relatively high exploration rate ∈ = 0.3 is adopted to encourage the system to try different authentication strategies;

[0259] As learning progresses, according to the formula Decay exploration rate;

[0260] Where: ∈ min This represents the minimum exploration rate (0.05); ∈ max λ represents the maximum exploration rate (0.3); λ5 represents the decay rate parameter; t represents the number of training steps;

[0261] In a production environment, maintain a minimum exploration rate ∈ min =0.05, ensuring the system can adapt to new threat patterns;

[0262] For high-risk operations, establish safety safeguards and limit the minimum safety level that may be adopted during the exploration process;

[0263] Model deployment and real-time optimization:

[0264] The trained DQN model is deployed as an authentication policy service, receiving real-time state input and outputting the optimal authentication policy.

[0265] Establish a strategy evaluation and feedback loop to continuously monitor the effectiveness of strategy execution and collect actual reward data;

[0266] An online learning mechanism is implemented, and the system continuously optimizes model parameters during operation to adapt to environmental changes;

[0267] Set up a safety rollback mechanism to temporarily revert to a conservative strategy when a performance degradation is detected, thus ensuring system security.

[0268] Through the adaptive authentication strategy optimization algorithm in step 400, the system can automatically select the optimal authentication strategy based on the trust assessment results provided in step 300 and the current context. These strategy decisions will serve as inputs to the cross-chain interoperability security protocol in step 500, guiding the execution of actual authentication operations.

[0269] Step 500: Based on the optimized authentication strategy, implement the cross-chain interoperability security protocol and perform the corresponding cross-chain authentication operation;

[0270] This step designs a cross-chain authentication protocol that supports heterogeneous blockchain networks; and performs corresponding cross-chain authentication operations according to different authentication levels and security requirements:

[0271] Protocol architecture design:

[0272] The protocol adopts a three-layer architecture: a trusted relay network layer, a blockchain adaptation layer, and a blockchain network layer;

[0273] Trusted Relay Network Layer: Serves as a communication hub between different blockchain networks, enabling message routing, protocol conversion, and security assurance;

[0274] Deploy a distributed relay node network and use the Byzantine Fault Tolerance (BFT) consensus algorithm to ensure the reliability of the relay network;

[0275] Each relay node implements complete message verification, routing, and forwarding functions, supporting parallel processing of multiple cross-chain requests;

[0276] The relay node implements dynamic trust management through the trust assessment mechanism established in step 300, prioritizing the selection of nodes with high trust levels for request processing;

[0277] Blockchain adaptation layer: Implements standardized interfaces for each blockchain network, resolving protocol differences between heterogeneous blockchains;

[0278] Develop plug-in adapters to support mainstream blockchain platforms such as Hyperledger Fabric, Ethereum, and R3Corda.

[0279] Achieve unified authentication request / response format conversion, shielding the technical differences of the underlying blockchain;

[0280] Provide an authentication policy mapping mechanism to convert the policy generated in step 4 into authentication operations supported by a specific blockchain;

[0281] Blockchain network layer: The native blockchain network to which each participant belongs, maintaining its own characteristics and security mechanisms;

[0282] It connects to a trusted relay network via an adapter layer interface without changing the core architecture of the existing blockchain;

[0283] Implement authentication logic based on smart contracts or chaincode locally to support local verification of cross-chain authentication;

[0284] Cross-chain authentication messaging protocol:

[0285] Define a standardized message format:

[0286] {

[0287] "msg_id": "Unique message identifier",

[0288] "msg_type": "Authentication request / response / status update type",

[0289] "source_chain": "Source blockchain network identifier"

[0290] "target_chain": the identifier of the target blockchain network.

[0291] "timestamp": the timestamp at which the message was generated.

[0292] "expires_at": "Message expiration time"

[0293] “auth_level”: “Authentication level (determined by step 4)”

[0294] "payload": "encrypted authentication data"

[0295] "proof": "proof of message integrity and origin"

[0296]

[0297] Implement a six-stage certification process:

[0298] Phase 1, Pre-authentication: The source chain entity submits an authentication request to the local blockchain, and the local system determines the authentication level and requirements through the strategy in step 4;

[0299] Phase 2, Local Verification: The source chain verifies the local identity validity of the entity, generates authentication credentials, and sends the request to the trusted relay network;

[0300] Phase 3, Cross-chain Request: The relay network receives the request, verifies the source, and routes the request to the adaptation layer of the target chain;

[0301] Phase 4, Target Chain Verification: The target chain performs local verification based on the authentication level, which may include additional challenge-response processes;

[0302] Phase 5, Result Transmission: The target chain returns the authentication result to the source chain through the relay network;

[0303] Phase 6, State Update: The source chain and the target chain update their respective authentication state records to provide feedback for the trust assessment in step 300;

[0304] Privacy protection mechanism:

[0305] It employs zero-knowledge proof (ZKP) technology to protect sensitive information and supports the following privacy protection features:

[0306] Selective disclosure: Entities may selectively disclose identity attributes, proving only the necessary information without revealing all details;

[0307] Attribute verification: Proving that identity attributes meet specific conditions, such as age greater than a certain value, credit score higher than a certain threshold, etc.

[0308] Range proof: Proving that a numerical attribute falls within a specific range without revealing its specific value.

[0309] ZKP implementation uses the zkSNARK protocol. Key steps include:

[0310] Preprocessing stage: Generate common parameter CRS (CommonReferenceString);

[0311] Proof generation: The prover uses private inputs and public parameters to generate a concise zero-knowledge proof π;

[0312] Proof Verification: The verifier uses public parameters and proves π to verify property assertions without needing to know the specific private inputs;

[0313] ZKP proofs are stored and verified on-chain to ensure their immutability and auditability.

[0314] The core mathematical expression of zkSNARK is as follows:

[0315] For the statement x and the proof π, the verification process can be expressed as:

[0316]

[0317] Where e represents a bilinear mapping, π1 and π2 are proof elements, φ(x) and ψ(r) represent the first and second polynomial functions respectively, g1, g2, and h2 represent the generators of the first, second, and third groups of the elliptic curve respectively, and r1 is a random number;

[0318] Cross-chain atomicity guarantee:

[0319] A two-phase commit protocol is designed to ensure the atomicity of cross-chain operations:

[0320] Preparation phase: All participating chains lock relevant resources and vote on whether the operation can be completed;

[0321] Commit / Rollback Phase: Based on the voting results of the preparation phase, coordinate all participating chains to commit or rollback simultaneously;

[0322] Implement a Hash Time Locked Contract (HTLC) mechanism to support condition-triggered authentication status updates;

[0323] State synchronization mechanisms ensure consistency in cross-chain operations:

[0324] Merkle trees are used to store authentication states, supporting efficient state proof and verification;

[0325] State summaries are periodically published on the relay network, allowing participating chains to verify global state consistency;

[0326] Implement an anomaly detection and automatic recovery mechanism to handle inconsistencies in state under abnormal conditions such as network partitions;

[0327] Safety measures:

[0328] Distributed key management:

[0329] Implement distributed key generation and management based on threshold signatures to avoid single points of failure;

[0330] Key operations are performed using secure multi-party computation (MPC) technology, without exposing the full private key;

[0331] It supports key update and rotation mechanisms to reduce the risk of using the same key for a long time;

[0332] Anti-replay attack measures:

[0333] Each authentication message contains a unique nonce and timestamp;

[0334] Maintain a sliding window record of processed message IDs and refuse to process duplicate messages;

[0335] Set a message expiration mechanism so that messages that have exceeded their expiration date will be automatically rejected.

[0336] Fault tolerance mechanism:

[0337] The relay network adopts ByzantineFaultTolerance (BFT) consensus, which tolerates the failure of some nodes or malicious behavior;

[0338] Implement multi-path routing for authentication requests, and automatically switch to an alternative path when a path fails;

[0339] Design a degraded service strategy to provide limited but secure authentication services in extreme situations;

[0340] Through the above protocol design, this solution achieves secure and reliable authentication between heterogeneous blockchain networks, integrating the AI ​​threat detection, dynamic trust assessment, and adaptive policy optimization functions from the previous steps into the cross-chain protocol, forming a complete cross-domain authentication security solution.

[0341] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for enhancing cross-domain authentication security of blockchain smart contracts based on AI, characterized in that, Includes the following steps: Construct a layered security architecture comprising a perception layer, an analysis layer, a decision-making layer, and an execution layer, and collect cross-domain authentication data and transmit it to the analysis layer; It receives feature data collected by the perception layer, trains a threat detection model through deep learning, and outputs threat scores and anomaly indicators based on the threat detection model. Based on the threat score and anomaly indicators output by the threat detection model, a dynamic trust assessment mechanism is established to calculate real-time trust values ​​between entities; the dynamic trust assessment mechanism includes: Trust model initialization and trust evidence collection: The system initializes and collects trust evidence for each participating entity. Create a trust data structure This includes basic reputation, historical interaction records, relationship network, and behavioral characteristics; The trust evaluation function is defined as: ； in, Indicates at time entity For entities Trust level, Trust score based on historical interactions. For node reputation, For network relationship weights, For real-time behavior scoring, The weight coefficients are the third, fourth, fifth, and sixth weights, respectively, and satisfy the following conditions: ; The calculation methods for each factor are as follows: Historical interaction trust score Calculated from interaction history: ； in, For the evaluation function of the interaction results, As the importance weight of the interaction, The time decay factor, The number of historical interaction records. For the first The timestamp of the next interaction; Node Reputation It is obtained directly from the global reputation system, and its value ranges from [0, 1]. Network Relationship Weight Calculations are made by analyzing the connections between entities in the relational network: ； in, For entities and The weight of the direct relationship between them For connection and The collection of public neighbors The number of public neighbors; Real-time behavior scoring Calculations based on the results output by the threat detection model: ； in, The threshold set for the system is used to normalize the calculation results. Threat rating, These are abnormal scores; Weighting coefficient The initial values ​​were set to 0.4, 0.2, 0.2, and 0.2 respectively, and were dynamically adjusted according to the system operation and different scenario requirements; The trust decay model uses a time-sensitive decreasing function: ； in, Indicates the initial trust value. The attenuation coefficient is... For time intervals; Trust transfer uses the chain of trust computation method: ； in, Representing entities Through intermediate entities For entities Indirect trust level, As a trust decay factor, For entities For entities Trust level, For entities For entities Trust level, This indicates the minimum operation; The specific implementation of the trust transfer mechanism includes: Trust decay factor Calculation formula: ,in This indicates the number of hops in the trust transfer. This is the attenuation coefficient per hop; Set maximum trust delivery depth The transmission chain beyond this depth will be cut off; For multi-hop propagation, a dynamic programming algorithm is used to calculate the optimal trusted path to avoid redundant calculations. Implement a trust transfer loop detection mechanism to prevent trust inflation caused by loop transfer; Multi-path trust aggregation: ； in Represents the aggregated entity For entities Overall trust level Represents the set of all possible intermediate nodes. Indicates passing through nodes Calculated indirect trust level, For path weights; The specific implementations of multi-path aggregation include: Path weight Calculation method: This assigns higher weights to more trustworthy intermediary nodes; Dijkstra's algorithm is used to identify entities To the entity The former The most reliable path; Set trust threshold Filter out paths below this threshold to reduce noise impact; Aggregation using a weighted geometric mean instead of an arithmetic mean imposes a stronger penalty on paths with extremely low trust: ； in Represents the aggregated entity using a weighted geometric mean. For entities Overall trust level The normalized path weights This is the sum of the normalized weights; Based on the calculated real-time trust value between entities, an adaptive authentication strategy optimization algorithm is designed to dynamically select the optimal authentication strategy. The adaptive authentication strategy optimization algorithm includes: MDP Modeling and State Space Definition: The authentication strategy optimization problem is modeled as a Markov Decision Process (MDP), and the state space is defined. Action space Transfer function Reward function and discount factor ; The reward function takes into account security, user experience, and system efficiency. ； in Indicates the state Take action below Total rewards received Assess security improvements. Assess the impact on user experience. Assess changes in system efficiency. The weighting coefficients are the seventh, eighth, and ninth, respectively, satisfying... ; The specific calculation methods for each evaluation indicator are as follows: Security assessment : ； in In the state Take action below The probability of potential security risks is estimated using threat models; User experience evaluation : ； in For action The resulting user operation costs The maximum acceptable cost; System efficiency evaluation : ； in For strategy Processing time, For resource consumption, and These represent the maximum acceptable processing time and maximum resource consumption of the system, respectively. The reward function weights are dynamically adjusted according to different scenarios: High-security scenarios: ; Balanced scenario: ; High availability scenarios: ; Q-learning algorithm implementation: The optimal authentication strategy is learned using the Q-learning algorithm, and the Q-value update formula is: ； in Representing state-action pairs The Q value represents the state. Take action below Long-term expected returns Indicates the current system status. The authentication strategy actions to be taken, For instant rewards, Indicates the next state; Indicates the possible actions for the next state. As a discount factor, The learning rate; Use an experience playback mechanism to store Tuples are randomly sampled and updated in batches to improve learning efficiency; Deep Q-Network (DQN) Implementation: DQN training implementation: using an experience replay buffer Store interaction samples , A boolean value indicating whether the current round has ended; Maintain two networks: a primary network and a target network; Every The step is to synchronize the main network parameters to the target network: ; From the experience replay buffer Random sampling of small batches of samples ; For each sample Calculate the target value : ； Calculate the loss function: ; in: Represents the loss function; Indicates the number of samples in the batch; Indicates the target Q value; This represents the Q-value predicted by the main network; the main network parameters are updated using gradient descent. ; Update the main network parameters using gradient descent. ; Exploration strategies and model deployment: An ε-greedy strategy is adopted to balance exploration and exploitation. The initial value of ε is set to 0.3 and gradually reduced to 0.05 according to the exponential decay rule. In a production environment, maintain a minimum exploration rate. To ensure the system can adapt to new threat patterns; For high-risk operations, establish safety safeguards and limit the minimum safety level that may be adopted during the exploration process; Model deployment and real-time optimization: The trained DQN model is deployed as an authentication policy service, receiving real-time state input and outputting the optimal authentication policy. Establish a strategy evaluation and feedback loop to continuously monitor the effectiveness of strategy execution and collect actual reward data; An online learning mechanism is implemented, and the system continuously optimizes model parameters during operation to adapt to environmental changes; Set up a safety rollback mechanism to temporarily revert to a conservative strategy when a performance degradation is detected, thus ensuring system security. Based on the optimized authentication strategy, a cross-chain interoperability security protocol is implemented to perform corresponding cross-chain authentication operations.

2. The AI-based blockchain smart contract cross-domain authentication security enhancement method according to claim 1, characterized in that, The perception layer collects authentication requests, network traffic, and node status information through a distributed data acquisition adapter, performs standardized processing to generate a feature vector set, and transmits it to the analysis layer in real time through a message queue system.

3. The AI-based blockchain smart contract cross-domain authentication security enhancement method according to claim 1, characterized in that, Threat detection models include: The data preprocessing module performs temporal window construction and spatial feature batch construction on the input features; The CNN module extracts the spatial features of the authentication request; The LSTM module captures the temporal characteristics of authentication behavior; The feature fusion module dynamically adjusts the weights of spatial and temporal features through an attention mechanism to calculate the threat probability output. The anomaly detection module integrates the results of anomaly pattern detection, group anomaly detection, context anomaly detection, and sequence anomaly detection.

4. The AI-based blockchain smart contract cross-domain authentication security enhancement method according to claim 1, characterized in that, Cross-chain interoperability security protocols include: The protocol architecture is designed to construct a three-layer architecture consisting of a trusted relay network layer, a blockchain adaptation layer, and a blockchain network layer. The authentication messaging protocol defines a standardized message format and a six-phase authentication process. The privacy protection mechanism employs zero-knowledge proof technology to support selective disclosure, attribute proof, and scope proof. Cross-chain atomicity is guaranteed, and a two-phase commit protocol and hash time locking contract mechanism are implemented. Security measures include distributed key management, anti-replay attack measures, and fault tolerance mechanisms.

5. An AI-based blockchain smart contract cross-domain authentication security enhancement system, used to execute the steps of the AI-based blockchain smart contract cross-domain authentication security enhancement method as described in any one of claims 1-4, characterized in that, include: The layered security architecture module is used to build a collaborative system of perception, analysis, decision-making and execution layers; The threat detection model module is used to implement a hybrid neural network that integrates CNN and LSTM to detect abnormal behavior; The dynamic trust assessment module is used to calculate the real-time trust relationship between entities based on multi-factor trust metrics. The authentication strategy optimization module is used to dynamically adjust authentication strategy parameters through reinforcement learning; The cross-chain interoperability module is used to implement secure authentication protocols between heterogeneous blockchain networks.

6. The AI-based blockchain smart contract cross-domain authentication security enhancement system according to claim 5, characterized in that, Also includes: The privacy protection and compliance module is used to implement zero-knowledge proofs, federated learning, and secure multi-party computation to protect sensitive information. Multi-layered security modules provide in-depth defense capabilities for security management, application security, data security, network security, and infrastructure security.

7. The AI-based blockchain smart contract cross-domain authentication security enhancement system according to claim 5, characterized in that, The threat detection model module adopts a semi-supervised learning approach, using both labeled and unlabeled data for training. The loss function combines cross-entropy loss, parameter regularization, and consistency loss.

8. The AI-based blockchain smart contract cross-domain authentication security enhancement system according to claim 5, characterized in that, The cross-chain interoperability module uses plug-in adapters to support multiple blockchain platforms, ensures the reliability of the relay network through fault-tolerant consensus, and implements multi-path routing and degradation service strategies for authentication requests.

Images