Research Of Reactive Power Optimization Considering Carbon-energy Combined-flow Based On Consensus Reinforcement Learning Algorithm

Reactive power optimization is to achieve one or multiple optimal performance indicators through the optimization of control variables in certain power system operation mode under the premise of meeting various constraints. The performance indicators include the minimum active power loss and the optimal voltage quality. The constraints include control variables constraint, state variable constraints and power flow constraints. Traditional mathematical methods have their own adaptability and superiority includ ing linear programming, nonlinear programming method, simplified gradient method, quadratic programming method, Newton method and interior point method. With the improvement of computer performance and the development of artificial intelligence, more and more intelligent algorithms, such as genetic algorithm, particle swarm optimization, swarm search algorithm, colony algorithm, reinforcement learning algorithm, are applied into reactive power optimization. However, when deal with such a complex nonlinear optimization problem, many algorithms are susceptible to the restrictions of high dimension and practical factors. The optimization effect and computing time of these algorithms are hard to meet the needs of real power system.On the one hand, with the growing highlight on the problems of energy and climate problems, the establishment of multiple objectives reactive power optimization model considering CO2 emission has become an important strategic need. On the other hand, with large-scale wind power, solar power and electric vehicle connected to the distribution network, more large-capacity extra high voltage power equipment and new power elements are connected together by the grid. Consequently, the power grid becomes more complicated, which brings about the problems of coordination, massive data and communication bottlenecks. Therefore, distributed architecture of the autonomous synergetic energ y management system(EMS) will be the trend of the smart grid. Traditional centralized algorithms for reactive power optimization are difficult to adapt to the decentralized collaborative EMS framework, which makes it an urgent need to propose a decentralized collaborative algorithm for reactive power optimization.In order to meet the trend of "decentralized autonomy, centralized coordination" in smart grid and the needs of environmental protection and energy saving,this paper proposes multi-area decentralized collaborative carbon-power synthesize framework which considers carbon emission of grid-side as one of the optimization objectives. Based on the concept of consensus, consensus interaction among agents are introduced to Q- learning algorithm, single body Q learning algorithm and multi-agent Q-learning algorithm are proposed in this paper. Simulation of the IEEE standard power system indicates that the proposed algorithm can solve the multi-area decentralized collaborative reactive power optimization effectively.