Parameter self-adaptive agent method and device suitable for power market simulation

Abstract

The invention discloses a parameter self-adaptive agent method and device suitable for power market simulation. The method comprises the steps of establishing an agent model; performing initializationprocessing on the quotation strategy set, the evaluation value and the selection probability of each agent model; carrying out initialization processing on learning parameters in each agent model; selecting a quotation strategy of each agent model according to a roulette mode; performing clearing calculation according to the selected quotation strategy and the market rule to obtain the income ofeach agent model; adaptively updating learning parameters of each agent model by using a particle swarm algorithm according to the income; updating an evaluation value and a selection probability of each quotation strategy according to the updated learning parameters; and judging whether the current simulation result meets a convergence condition or not, if so, ending simulation and outputting a final result, and if not, returning a quotation strategy for reselecting each agent model. According to the method, the parameters of each agent model can be adaptively updated, so that the simulationresult better conforms to the actual market member behaviors.

Description

technical field [0001] The invention relates to the technical field of power system market transactions, in particular to a parameter adaptive agent method and device suitable for power market simulation. Background technique [0002] At present, the research and platform development of power market simulation have started at home and abroad. The power market simulation mainly has two functions. First, it can review the real market environment and provide real scenarios for the analysis of real market problems; Second, study the influence of individual and group behaviors of market members on market transaction results, analyze the rationality of market mechanism design, and simulate the equilibrium establishment process of this electricity market model for dynamic simulation. [0003] In the current power market simulation research, there is mainly an agent-based computational economics method, which is to embed the power market model into the intelligent agent research fra...

AI Technical Summary

Problems solved by technology

Most of the currently commonly used agent models use reinforcement learning algorithms, such as RE algorithm, Q-learning algorithm, etc. Although these models can simulate the quotation behavior of market members to a certain extent, experience shows that even in the same scenario, the same agent model Different parameter settings will generate different quotation behaviors, which will affect the market simulation results

In addition, if there are a large number of proxy models in the simulation, how to set or adjust the parameters of the proxy models becomes a difficult problem

Images