Interconnected power system CPS instruction dynamic allocation and optimization method

Abstract

The invention discloses an interconnected power system CPS instruction dynamic allocation and optimization method. The interconnected power system CPS instruction dynamic allocation and optimization method comprises the following steps of (1) determining a control target, (2) determining a state discrete set S, (3) selecting balance units and determining a combined action discrete set A, (4) calculating the momentary value of ACE (k) and the moment value of CPS (k) in an area, (5) obtaining the instant reward value Ri (k) of each intelligent body, (6) calculating related equalization linkage strategies through linear equalization and equalization selection function, (7) carrying out corresponding operation on all units j, and (8) returning to the step (4) when a next control circle arrives. The interconnected power system CPS instruction dynamic allocation optimizing method has the advantages that frequent adjusting times of various units can be reduced effectively, the CPS control performance of an AGC system is improved, and the interconnected power system CPS instruction dynamic allocation optimizing method is especially suitable for the CPS instruction dynamic allocation and optimization of interconnected power systems in which thermal power is dominant and unit combination is complex.

Description

technical field [0001] The present invention relates to the technical field of electric power system automatic power generation control (secondary frequency regulation), in particular to a method for dynamic allocation and optimization of CPS commands in an interconnected grid. Allocation optimization. Background technique [0002] Since the control performance standard (Control Performance Standard, CPS) was proposed in the automatic generation control (AGC) of the interconnected grid, the pass rate of the CPS has become an important factor affecting the AGC control strategy. One of the key steps of the AGC control system is to distribute the CPS general adjustment command to each AGC unit according to a certain optimization algorithm. [0003] The traditional AGC regulation power distribution adopts the average distribution method, which does not consider the differences among units, and cannot meet the needs of CPS regulation. In addition to reinforcement learning, most...

AI Technical Summary

Problems solved by technology

Although classical reinforcement learning can obtain the convergent equilibrium point under the premise of satisfying the CPS assessment standard of the power grid, the allocation factor action strategy with limited unit output combination space is used in the allocation process, so that the equilibrium point found may not be the optimal equilibrium point. point, the adjustment of various units is more frequent, the number of convergence steps is relatively long, and the real-time curves of CPS1 and ACE after convergence are not smooth enough

In addition, Q-learning, Q(λ)-learning, and hierarchical Q-learning are all single-agent reinforcement learning algorithms in essence, and do not involve collaborative learning between agents, and the action combinations of each agent are not necessarily joint optimal action

Images