针对室内光线照度分布的非线性、时变性问题,提出一种改进差分进化(improved differential evolution,IDE)和高斯过程回归(Gaussian process regression,GPR)融合的方法,结合孤立森林(isolationforest,iFroest)来建立室内灯光照度预测模型.首先,通过使用孤立森林算法剔除异常数据并对其余数据进行归一化处理.然后,为克服传统差分进化(DE)算法的早熟收敛问题,提出了一种基于进化状态的概率选择策略,并将变异因子F设定为服从正态分布,以提高算法性能.同时,利用IDE算法对具备不确定量化特性的GPR模型的超参数进行寻优,从而建立最优的室内灯光照度预测模型.最后将所提出的基于孤立森林与IDE-GPR的模型与其它模型进行比较,实验结果表明该模型的R2、δMAE、δRMSE分别为0.999、0.245 lux、0.324 lux优于其他模型,能够更准确的预测室内光环境的照明状态.
A luminance prediction model for indoor lamps based on isolated forest and IDE-GPR
Aiming at the non-linear and time-varying problem of indoor light illuminance distribution,an imporved differential evolution(IDE)and Gaussian process regression(GPR)method combined with isolation forest(iForest)is proposed to build a prediction model for indoor light illuminance.Firstly,the outliers are removed and the rest of the data is normalized by using the isolation forest algorithm.Then,an evolutionary state-based probabilistic selection strategy is proposed to overcome the premature convergence problem of the traditional DE algorithm,and the variation factor F is set to obey the normal distribution to improve the performance of the algorithm.At the same time,the IDE algorithm is used to optimize the hyperparameters of the GPR model with uncertain quantitative characteristics,so as to establish the optimal predictive model for indoor lamp illuminance.Finally,the model based on isolation forest and IDE-GPR proposed in this paper is compared with other models,and the experimental results show that the R2,δMAE,and δRMSE of this model are 0.999,0.245 lux,and 0.324 lux better than the other models,and it can more accurately predict the lighting state of the indoor light environment.
万方数据
