Influence of Discrete and Continuous Action Spaces on Deep Reinforcement Learning-Based Pricing Strategy Optimization for Electricity Retailers

The pricing strategy optimization problem becomes important for electricity retailers in electricity market. Deep reinforcement learning (DRL) has been applied to solve the strategic decision-making problems in electricity market area. However, the influence of discrete and continuous action spaces on optimization results by using DRL-based methods to solve for optimal retail price is unknown. This paper applies two different DRL-based retail pricing strategies through deep Q network (DQN) and deep deterministic policy gradient (DDPG) for the electricity retailers. An in-depth comparative analysis between DQN and DDPG is conducted in terms of convergence and computational performance. The numerical results of optimal retail prices and responding loads show the influence of discrete and continuous actions space on optimization effect.