Determining Optimal Incentive Policy for Decentralized Distributed Systems Using Reinforcement Learning

Cryptocurrencies have gained a lot of attention in recent years, mostly due to their decentralized manner of operation and their growth in value. However, a major drawback most of them possess is their high energy consumption. Current solutions to this problem have significant l imitations: bringing back centralization and/or substituting the required energy with, e. g., storage space. This paper aims to address the problem by investigating the use of a two-level deep reinforcement learning (RL) model to design incentive policies for green mining in cryptocurrencies. This is done by modeling one such energy-intensive cryptocurrency system and creating an RL environment. Finally, by running simulations in an RL environment, we develop and test incentive policies, according to which cryptocurrency participants who primarily use renewable energy for their mining operations are more likely to add new blocks to the blockchain. Our results show that even when the green score of each crypto miner (determined by their use of green energy sources) has relatively small importance (up to 0.3) in their selection probability, miners still shift towards green mining in order to increase their chance of being picked to validate cryptocurrency transactions and receive the corresponding rewards.