A Multi-User Blockchain Framework Integrated With Disentangled Graph Variational Autoencoder for Data Integrity Assurance in Heterogeneous Cloud Environments

Abstract

Heterogeneous cloud environments integrate a variety of platforms, services and data formats for multiple applications and multiple user access. In these systems, maintaining data integrity is crucial for accuracy, consistency, and preventing unauthorized modifications, particularly when data is accessed and stored across various locations. Challenges include detecting data tampering, enforcing access control, and managing heterogeneous data types, highlighting the need for scalable solutions to ensure cloud data integrity. This paper proposes a Multi-User Blockchain Framework with a Disentangled Graph Variational Autoencoder for Enhancing Integrity in Heterogeneous Cloud Data (MUBF-DGVAE). The framework has three phases. In data acquisition, heterogeneous cloud data is processed using a Polynomial Non-Linear Chaotic Function (PNLCF) to create unique hash fingerprints for integrity verification. During encryption and storage, Quantum Hash-Based Attribute-Based Encryption (QHABE) is applied per a defined access policy; encrypted data and the quantum hash are stored in the cloud. A blockchain transaction with the encrypted data URL, initial hash key, timestamp, and access policy is validated and recorded using Non-Interactive Practical Proof-of-Storage (nPPoS). In verification, when a user requests access, Multi-Qubit Quantum Key Distribution Ciphertext-Policy Attribute Based Encryption (QKDCPABE) generates decryption keys by validating user attributes against the access policy. After decryption, PNLCF regenerates a hash from decrypted data, compared to the original blockchain stored hash by the Disentangled Graph Variational Autoencoder (DGVAE) to detect tampering. Experimental evaluations show MUBF-DGVAE achieves a stable hash bit change of 11.9–14.8 ms, a hash time of 0.011–0.080 ms, a cloud encryption runtime of 124.47 ms at 100 users, and a decryption runtime of 83.54 ms at 100 users, outperforming methods like EI-MUBF-HSCM, EPQKG-MUAED-EHR, and ARA-MUCIC-ME across all metrics..