A Reinforcement Learning Algorithm based on Neural Network for Economic Dispatch

Economic Dispatch (ED) problem for power system as an important yet challenging task, is commonly solved by recent methods under the assumption that values stored in Q-table. While with the computational complexity and mass data, it is no longer suitable to calculate the reward in Q-learning. To address the above problem, we propose a new Q-learning model based on neural networks and find the proposed model play an important role in our task. Therefore we adopt the proposed an improved strategy to solve the issue. By using our model, the cost function can be effectively learned and directly predicted the result instead of looking up table stored Q-values. In addition, our model can solve the ED problem with load demand varying with time and have good applications in the power system. So it reveals that our proposed model is time-saving and easily mounted on the power system. The experimental results show that our method is effective and efficient compared with state-of-the-art works. Finally, we argue our method has great potential for further research.