This work constructs an innovative dynamic energy efficiency optimization model of methane hydrate dissociation by thermal stimulation method base on artificial intelligence predictive control of heat injection strategy. Model can divided into two parts, firstly, the prediction of hydrate decomposition rate of each time step is realized via the supervised learning neural network prediction part of the model. The optimization of the energy consumption of hydrate decomposition by thermal stimulation under different gas recovery situations is realized by the deep reinforcement learning-based policy optimization part of the model. Take the lowest injection/recovery energy consumption ratio as the optimization objective, take the injection temperature and heat injection rate (per unit time step) as optimized variables. Implement evaluation and execution for each time step, updating and correcting the prediction errors of successive time steps to achieve dynamic optimization of energy efficiency. The application results of the model showed that under the premise of the same deposits conditions and the same injected heat, the recovery time of the model optimization group decreased by 38% compared with the control group; while under the same deposits conditions and recovery time, the energy consumption of the model optimization group decreased by 40% compared with the control group.
Keywords natural gas hydrate, thermal stimulation, energy efficiency, deep reinforcement learning, dynamic optimization