災(zāi)害應(yīng)急場(chǎng)景下基于多智能體深度強(qiáng)化學(xué)習(xí)的任務(wù)卸載策略

摘要： 針對(duì)傳統(tǒng)深度強(qiáng)化學(xué)習(xí)(deep reinforcement learning, DRL)中收斂速度緩慢、經(jīng)驗(yàn)重放組利用率低的問(wèn)題，提出了災(zāi)害應(yīng)急場(chǎng)景下基于多智能體深度強(qiáng)化學(xué)習(xí)(MADRL)的任務(wù)卸載策略。首先，針對(duì)MEC網(wǎng)絡(luò)環(huán)境隨時(shí)隙變化且當(dāng)災(zāi)害發(fā)生時(shí)傳感器數(shù)據(jù)多跳的問(wèn)題，建立了災(zāi)害應(yīng)急場(chǎng)景下基于MADRL的任務(wù)卸載模型；然后，針對(duì)傳統(tǒng)DRL由高維動(dòng)作空間導(dǎo)致的收斂緩慢問(wèn)題，利...