|
Title: |
Reactive Power Optimization Calculation Based on Multi-step Q(λ) Learning Algorithm |
|
Source: |
Power and Energy Engineering Conference (PEEC 2010 E-BOOK)
(pp 449-452)
|
|
Author(s): |
Xi-bing Hu, College of Electric Power, South China of Technology, Guangzhou 510640, Guangdong, China
Tao Yu, College of Electric Power, South China of Technology, Guangzhou 510640, Guangdong, China |
|
Abstract: |
In order to pursue greater economic benefits, the operation of power systems increasingly close to the critical stability, increasing the possibility of instability of the system. Thus security has become the focus of modern power system. Take the security of the power system operation for study and establish a reactive power optimization model aimed at constraint variable stability margin. A multi-step predictable Q(λ) learning algorithm based on Q learning algorithm of reinforcement learning is proposed, which with its good backtracking ability, continuously try and backtrack, getting the long-term maximum value of reward to find the optimal action. It is with advantages of online learning capability and convergence speed. This algorithm is compared with other algorithms in IEEE14 standard example and achieves good results, which proves that multi-step Q(λ) learning algorithm is feasible and efficiency for reactive power optimization.
|