Research On Artificial Intelligence(AI) Model Consensus And Data Preprocessing Algorithm For Smart Blockchain Oracles

With higher execution efficiency and credibility,the new generation of smart contracts based on blockchain has shown great value and potential in various industries.Blockchain Oracle(Oracle)is an interface between the off-chain data and smart contracts on the chain,which provide critical data for execution conditions judgment of smart contracts.The existing blockchain oracle system only works well on simple structured data.It is difficult to implement artificial intelligence(AI)technology in a decentralized manner to process massive amounts of data and reach a consensus on the result.This has resulted in a limited application scenario for blockchain oracles.In order to overcome the above shortcomings,this paper proposes an intelligent blockchain oracle design based on AI model consensus.Compared with the traditional design based on result consensus,the new design has advantages in high-throughput data processing and application scenario migration.However,there are two issues to be solved:distributed consensus of AI models,data storage and privacy protection.This paper has carried out research on the two issues respectively.First,for the AI model consensus problem,this article proposes a new model consensus algorithm M-PBFT(Model-based Practical Byzantine Fault Tolerance)by integrating the classic consensus algorithm PBFT with each section of model training and verification.This paper also demonstrates the convergence of the algorithm and establishes a probability model to evaluate the performance and robustness under nonideal communication conditions.Theoretical and numerical results indicate that the model consensus algorithm proposed has high robustness and execution efficiency.Second,for the problem of data storage and privacy protection,this paper proposes a data preprocessing method based on the feature extraction of mutual information neural networks.This method is independent of specific AI models and tasks.It extracts data features in an unsupervised manner based on the mutual information maximization principle,so as to achieve the purpose of data volume compression and privacy protection of original data.This paper verifies the effectiveness and superiority of this method through comparative experiments on two public datasets.The smart blockchain oracle design proposed in this paper realizes the integration of blockchain oracle technology and AI technology,enhances the data processing capabilities of blockchain smart contracts,and has broad application prospects.