船舶设计任务动态调度预测

Forecasting of Dynamic Scheduling of Ship Design Tasks

在船舶设计过程中经常会出现随机新设计任务,为船舶设计任务调度方案的制订带来一定的困难。基于反向传播(Back Propagation, BP)算法,引入动量-自适应学习率反向传播(Momentum and Self-Adaptive Learning Rate Back Propagation, MSBP)算法预测随机新设计任务是否可加入制订的船舶设计任务调度方案,以解决扰动情况下的船舶设计任务动态调度(Dynamic Scheduling of Ship Design Tasks, DSSDT)问题。为减小求解空间和训练难度,选择对调度结果具有重大影响的属性作为MSBP算法的特征值。基于抽取的特征值构建MSBP算法模型,并采用大量数据完成对模型的训练。对比试验结果表明,MSBP算法的准确性优于未改进的BP算法,某项随机新设计任务的可调度性与其优先级最为密切。

The stochastic new design tasks often appear during ship design,which brings some difficulties to the formulation of scheduling scheme of ship design tasks.Based on the Back Propagation(BP)algorithm,the Momentum and Self-Adaptive Learning Rate Back Propagation(MSBP)algorithm is introduced to forecast whether the stochastic new design tasks can be added into the formulated scheduling scheme of ship design tasks,so as to solve the Dynamic Scheduling of Ship Design Tasks(DSSDT)problem in the case of disturbance.In order to reduce the solving space and training difficulty,the attributes with great influence on the scheduling results are selected as the eigenvalues of MSBP algorithm.The MSBP algorithm model is constructed based on the extracted eigenvalues,and the model is trained with a large amount of data.The contrast test results show that the accuracy of MSBP algorithm is better than that of unimproved BP algorithm,and the schedulability of a stochastic new design task is most closely related to its priority.