人工神经网络与动态搜索的机组组合算法

为了使机组达到最优组合，减少运行成本，研究了基于修正BP人工神经网络与动态搜索的快速算法在机组组合中的运用．采用修正Levenberg-Marquardt算法训练BP神经网络，并针对该算法占用内存大的缺点，提出了减少内存占用量的修正．由此，根据负荷预测曲线，应用修正BP人工神经网络产生机组的预开停计划，在此基础之上，针对预计划中某些机组状态不确定的阶段，应用动态规划法进行全局调整以确定机组的状态组合．实验数据表明，所提出的算法与传统的动态规划算法相比，可以在有效地减少时间与内存的占用量的前提下，有效地避免动态规划法中的维数灾的问题．

Fast Algorithm About Unit Commitment Based on Revised BP Artificial Neural Network and Dynamic Search

In order to reduce the operation cost and optimize the unit commitment,the fast algorithm about unit commitment based on revised BP ANN(Artificial Neural Network) and dynamic search is discussed.The BP ANN is trained with Levenberg-Marquardt algorithm,which aiming at its drawback of the storage of some matrices that can be quite large for certain problems,and a revised algorithm is presented.The BP ANN is used to generate a pre-schedule according to the input load profile.Then the dynamic search is performed some stages where the commitment states of some of the units are not certain.The experimental results indicate that the proposed algorithm can reduce the execution time and memory space without degrading the quality of the generation schedule.