Bayesian Heuristic for Multiperiod Control
多期控制的贝叶斯启发式方法AI 解读 ↓
← 返回论文列表与筛选
阅读说明
本页整理公开书目信息，不生成学术质量评分。AI 解读根据题名及所注明的摘要或原文片段生成，未完成全文核验，可能有误；具体假设、证明与结论请核对原文。摘要保留来源文本，未经模型改写。
DOI 首次登记不等于接收或发表日期。部分更正、评论等条目没有独立摘要；缺少可靠材料时保留待补状态。
引用量采用 OpenAlex 单条关联记录，不合计多个记录；关联记录可能包含正式发表版本，不等于此版本单独引用。缺失不视为零，引用指标不代表论文质量。
关注作者按书目中的完整姓名匹配，同名和姓名变体需人工辨别。相关论文先按共享细分方向和具体方法排列，再参考通用方法与发表日期。
分类依据与纠错
AI 根据所提供的论文材料归类：摘要涉及一般回归模型下的多期实验设计和控制，没有明确给出宏观预测、时间序列建模或其他所列领域的证据。
反馈分类问题
正在核对关联版本…
代码、数据与补充材料
正在核对对应版本的资源…
关注这篇论文的作者
多期控制的贝叶斯启发式方法
- 研究问题
- 在多期控制中，如何选择多个输入，使当前输出接近目标，同时获得有助于改善未来控制表现的模型信息？
- 方法与思路
- 作者在同时包含确定性等价设计与短视设计的候选族中寻找最佳设计，并在一般回归和一阶线性回归模型中研究其表现。
- 材料陈述的贡献
- 摘要称所提设计可改善两类基准并协调不同期限下的取舍，同时观察到确定性等价设计可能引入过多模型学习。
- 阅读关注点
- 应核对设计族与优化准则、先验及更新方式、控制期限的作用，以及相对两类基准的改善在哪些模型和指标下成立。
- 阅读边界
- 仅据版本未注明的OpenAlex索引摘要。摘要未说明具体设计公式、比较幅度或最优性保证，不能将启发式搜索视为完整动态最优解。
原文摘要
In multiperiod control problems, one needs to sequentially choose the value of a set of multiple input variables, xn, to keep the value of a single output variable, yn, close to a target, T. At any stage n, the selection of xn poses a difficult dilemma. On the one hand, xn has to be such that (yn(xn) - T)2 is small in expectation. On the other hand, xn has to be such that it maximizes the amount of information in yn(xn) about the model relating yn and xn, to help keep the value of future y's as close to T as possible. The main approach to this problem is through certainty equivalence designs, which select xn such that yn(xn) is estimated to be equal to T. Instead, myopic designs select the xn that would be optimal if yn(xn) was the last observation to be made. We propose designs that improve on certainty equivalence and on myopic designs by searching for the best design in a family that contains both. In that way, the proposed design reaches a good compromise between the certainty equivalence choice, known to have good asymptotic properties, and the myopic choice, which is close to optimal for problems with small horizons. The characteristics and performance of the new designs are explored for regression models in general and for the first order linear regression model in particular. We observe that certainty equivalence designs induce too much learning about the model, just opposite of what is sometimes implied by the engineering literature on adaptive control when certainty equivalence designs are treated as if they were myopic designs. Our approach can be exported to many other sequential and nonsequential design problems where the optimal rule is too difficult to compute.