Research On Blockchain Abnormal Behavior Detection Method Based On Deep Convolutional Neural Network

Blockchain technology has gained attention for its decentralized,anonymous,and tamper-proof features.However,due to the complexity of its architecture,blockchain systems are vulnerable to various types of attacks,such as Ethereum Ponzi schemes and smart contract code vulnerabilities.These attacks have resulted in significant economic losses and have severely hindered the development and application of blockchain technology.Therefore,it is crucial to research and develop detection methods for detecting Ethereum Ponzi schemes and smart contract vulnerabilities,which holds great practical significance and value.There are some issues with the current abnormal detection methods for blockchain.In detecting Ethereum Ponzi schemes,existing research still has technical deficiencies in handling data imbalance and extracting critical features.For detecting smart contract vulnerabilities,most of the current methods rely on technologies like symbolic execution,formal verification,fuzz testing,middleware representation,and machine learning.However,these methods heavily depend on expert knowledge and contract source code during feature extraction,which hinders the real-time detection of smart contracts.This thesis proposes a new solution to address the problems in traditional detection methods by combining deep convolutional neural networks and blockchain technology.Our method detects abnormal behavior in the blockchain system and prevents potential security threats.We also developed a blockchain security detection platform to facilitate the practical application of our research.Key contributions and innovations of this article include:(1)This thesis proposes an intelligent detection method for Ethereum Ponzi schemes using PD-SECR.It solves the problems of data overlap during data augmentation and feature selection redundancy in traditional detection methods.These issues not only reduce accuracy but also lower detection efficiency.PD-SECR effectively addresses these problems by utilizing Edited Nearest Neighbor(ENN)and convolutional calculations.Additionally,it achieves efficient detection in imbalanced data sample sets.The proposed method creates a safer and more reliable trading environment for the Ethereum transaction platform.(2)This thesis proposes an automated method to detect vulnerabilities in smart contracts using Bic-RL.It addresses the issues of contract source code dependencies,complex data preprocessing,and incomplete feature extraction in current detection methods that lead to decreased performance.To overcome these issues,the proposed method uses Bicubic interpolation,Residual Neural Networks(Restnet),and Long short-term memory(LSTM)to create an intelligent detection method for smart contract vulnerabilities.Additionally,a labeled dataset for smart contract vulnerability detection is provided.This method simplifies data preprocessing,ensures comprehensive feature extraction,and improves model detection efficiency by enabling real-time acquisition of smart contract data.Feasibility analysis and performance evaluations of the proposed detection model demonstrate an accuracy of up to 84.5%.Comparison experiments with other detection methods also show that the proposed method has higher detection accuracy and generalization ability.(3)This thesis has developed a blockchain security detection platform that uses two detection methods(PD-SECR and Bic-RL)to detect Ethereum Ponzi schemes and smart contract vulnerabilities.To address potential security risks in the local runtime environment,which may affect detection results,the platform includes a runtime environment detection module.This module checks the user’s local runtime environment for security before performing any detection,and sets up an abnormality alarm mechanism to ensure the reliability of the results.The developed platform is a one-stop service detection system that integrates multiple functional modules.Users can perform specific detections as needed,and the runtime environment detection module ensures the reliability of the detection environment and the credibility of the results.