Research On Data Sharing And Decision Model Of Intelligent Greenhouse Group Based On Consortium Blockchain

With the continuous development of smart greenhouses,the issues of data sharing and data security in the decision-making process within a smart greenhouse cluster are becoming increasingly prominent.In traditional smart greenhouses,there is a risk of data tampering in local databases,leading to unreliable shared data and decision-making results.Additionally,the combination of differential privacy and single-key homomorphism in the process of building decision models poses challenges such as the immutability of fused members and error amplification.In response to these issues,the main research contents and results are as follows:(1)A smart greenhouse group data sharing and decision-making collaborative solution is proposed to address the issues of unreliable data sharing and lack of feedback on decision execution in the smart greenhouse group.Based on the characteristics of the distributed architecture of the consortium chain,a network structure for interaction between edge servers in the smart greenhouse group is built using a consortium chain.A reliable data sharing and decision-making collaborative interaction process is designed between entities to enhance the reliability of data sharing and decision-making.Data summaries are uploaded to the chain to enable verification of data hashing in a chain-on and chain-off manner.Key-value pairs are used to separately store decision-making and execution results to provide feedback on decision results.Finally,various smart contracts are designed for participating nodes to upload and download data,ensuring the trustworthiness of the data.(2)A privacy fusion mechanism for federated learning parameters in blockchain group is proposed based on multi-key homomorphic encryption algorithms to address issues such as immutable fusion members when using single-key homomorphic encryption parameters during decision model establishment and error amplification when using differential privacy to protect parameters.Firstly,a blockchain federated learning model is constructed,and nodes are set to complete the target function and loss function of model training locally.Based on the working principle of federated learning and the steps of multi-key homomorphism,a parameter privacy fusion mechanism is designed.According to the federated averaging algorithm,specific steps for encrypting federated fusion are set,and smart contracts involved in the parameter privacy fusion mechanism are written.Finally,an accuracy and security analysis of the privacy fusion mechanism is conducted,and a comparison with the overall solution is made.This solution avoids the central node failure and resulting paralysis and has the function of multiple encryption.Homomorphic results require group node collaboration and flexible selection of group members,improving the security and transparency of federated learning.(3)A prototype system was implemented to test and analyze the proposed solution and privacy fusion mechanism described in the paper.The results indicate that the system successfully implemented the following functionalities for the smart greenhouse cluster: data evidence storage,verification of decision outcomes,generation of public and private keys for participating nodes,on-chain storage of public keys,as well as encryption and decryption capabilities.This implementation facilitated trusted data sharing and decision-making in the greenhouse environment.In terms of performance,when the block packaging time was set to4 seconds,the maximum throughput achieved for storing data on the consortium chain was27.3 transactions per second(tps).Therefore,the block packaging time was fixed at 4seconds.In the federated learning system,when training the MNIST dataset using Fully Connected Neural Network(FCNN)and Convolutional Neural Network(CNN),the paper achieved accuracy rates of 98% and 99%,respectively.In comparison,the accuracy rates obtained using differential privacy were 92% and 96%,respectively.For training the CIFAR-10 dataset,the paper achieved accuracy rates of 75% and 94% using FCNN and CNN,respectively,while the accuracy rates with differential privacy were 56% and 73%,respectively.Finally,when training the greenhouse dataset,the paper achieved accuracy rates of 97% and 93% using FCNN and CNN,respectively,whereas the accuracy rates with differential privacy were 92% and 90%,respectively.The paper’s privacy fusion mechanism based on multi-key homomorphic encryption achieved accuracy rates almost identical to the original models,effectively preserving data privacy while avoiding error amplification.The proposed solution in this article can improve the reliability of data sharing and decision-making in intelligent greenhouse clusters,and ensure the privacy fusion of model parameters.This has important practical significance for enhancing the data security and production efficiency of intelligent greenhouse clusters.