Security of Neural Network-Based Protocolfor Smart Grids

The rise of quantum computing has brought about a significant threat to the security of current asymmetriccryptography methods. To address this issue, neural cryptography has emerged as a potential alternative that islightweight and efficient in resisting known quantum computer algorithms. As the implementation of quantumcomputing could expose IoT sensors and smart grid systems to a range of attack vectors, the need for secure andefficient cryptography solutions is crucial. This paper explores the effectiveness of using integer-valued input vectorsto enhance the synchronization of the Tree Parity Machine, which is a type of neural cryptography. The introductionof a new parameter M, which indicates the minimum and maximum values of input vector elements, plays a key rolein evaluating the nonbinary version of the mutual learning algorithm in a simulated insecure environment. The findingssuggest that while there may be some trade-offs between security and synchronization time with the Nonbinary TreeParity Machine, the speed improvement resulting from the enhancement outweighs the decrease in security. Thisenhancement is particularly impactful for smaller adjustments to the parameter M, highlighting the potential of neuralcryptography for securing IoT sensors and smart grid systems.