套用的，是节省同学的题目。反正这里没有人看没有人知道。
慢慢我会努力过渡到这里来，因为，这个博客是一个工作的我。
在EVA里，那台超级电脑由三个部分组成，分别代表发明者的三个自我：作为科学家的她，作为女人的她，作为母亲的她。这个博客，是记录作为researcher的我（没有颜面自称研究人员），校内网，是记录作为“I'm not a girl, but not yet a woman”的我，MSN是记录一个不够成熟的孩子气的我。
傻是没有药医的，我不知道节省同学指的是什么。我的指向性很强——今天模型算出来了。问题终于找到，如此简单的一个错误：把一个地方的符号弄错了。
当时正在和Arthur讨论到底那个本来不该出现的正反馈究竟出现在什么地方。忽然发现有两个应该此消彼长的量竟然同步增加了，赶紧进去一看——乖乖，就是它了。
当时很羞愧，以至于做出了非常搞笑的动作……真的不是故意的。我扭过头，用指头戳了戳Arthur的胳膊，说：I found the problem----a very stupid mistake。
然后数据如愿以偿地收敛了。我还是羞愧的无地自容，崩溃地自责——关于那个弱智的错误。Arthur宽慰说：mistakes are always stupid mistakes。真是富含哲理的语句。不枉我如此崇拜Arthur。
当时想到节省同学的话：傻，是没有药医的。
分明就是预言今天的我。
回来和师兄讨论合作的问题。既然已经有了初步成果，纵深合作也就成了可能。
首先，说一说Arthur帮我定的行动方向：
1.简化模型提高速度
2.两步走方针：用基于物理方程构造的模型来生成数据，构造一个黑盒子模型来学习这些数据，得到一些模糊控制的规则。
3.构造出一个能够基于设计要求来计算系统参数的front-end的程序或者模块
4.模型的验证
5.一些应用方法的研究（？忘记鸟，大概是一些控制方法的应用吧）
6.多区自然通风模型由CFD生成数据，然后用黑盒子模型来迅速学习并进行预测。
今天老板要我把模型变得更通用一些。此外，我觉得还应该把稳态变成太阳辐射随时间变化而变化的模型，这样才有实际价值嘛。然后呢，讨论一下各个参数的区间问题，很多参数不是所有的情况都可以的，一方面是物理意义的区间，一方面是收敛与否的区间。在这些的基础上，写一个报告。那么，基于物理公式的模型构造到此结束。
然后，按照Arthur和师兄的共同指示，下一步应该是设计一个模糊控制的黑盒子，模拟基于物理公式的模型。为此，应该看一下如何建模糊系统的书（师兄啊，偶像啊！！复杂事情中能够一眼看到关键，牛人！）这个模型的作用在于，能够迅速得到和前面模型类似的结果。师兄解释，大概是这样一回事，先用构造好的模型生成无数的数据，得到一个数据表（反正硬盘不贵）。这些数据表呢，就相当于一个选择树的枝头的果子，然后再出现的新数据在选择树中徘徊那个前进啊，到最后关头直接进行各种插值计算什么的，得到一个最可能的新果子。
计算出太阳能部件的控制策略还不算完，难度比较大的可能在于整个系统的模糊控制。师兄的话是：解模糊方程组（额滴神，看来不简单）。如果能够做到这一步，就可以PhD毕业了。
但是师兄另外提了一个比较有意思的方向。比较人工智能的一个方向。我们打算往这个方向努努力，能做出多少做出多少。做的多吧，多发几个paper，搞不好就干脆做这个当论文了。做的一般吧，至少可以在毕业论文里当作一章内容。
我们俩个在整个组里算是比较特殊的背景：其他人都是控制出身，我们俩是暖通出身。（都搞出身问题了）。我们学流体力学的时候，各种无量纲准则数为简化经验公式带来了多少好处我就不说了，每个学了流体力学的人都至少还记得雷诺准则数吧！没有它，怎么判断流动情况？还有毕渥准则数（……忘记作用了，咔咔），还有普朗特准则数……这些无量纲准则数都是无数革命先烈们在无数恶心的数据中发掘出来的——仿若想起那个词DATA MINING。但是，这是手工挖掘，很苦。我们希望的是，能够利用计算机，从一堆数据中，自动生成公式，自动选择准则数。
想想，要是真能做到，该是多么牛逼的事情啊！但是这个有多难呢？首先，现在计算机的发展程度还远不具有这样的学习能力。其次，这在AI中也属于比较前沿的东西，我们这个方向一旦有真正的思路，对于AI学科来说都是开创性的。再次吧，人类如何进行这样的归纳总结都是很难再现的，用计算机如何再现呢？是个问题吧。最后呢……我那么弱，我是水桶里最短的木条啊。
但是，目标现在有了，今年，自己写一个发表在杂志上的paper（系统工程啊，万事开头难啊，还要学习写作方法）。还有，今年，transfer。
师兄给我指出了前进的方向：研究生的论文数量和质量就相当于本科的GPA。大公司来，看这个！
我不是不想前进，是没有量化的指标前进不了。这下好了。嗯。希望3年下来，师兄发6篇，我能发个4篇5篇的，我们互相署名。总数量要有两位数！
2007年5月23日星期三
2007年3月1日星期四
一个月没有写这里
今天seminar之后我跟于震讨论了一下我们俩要纵深合作的问题。一方面可以把项目做大点，一方面可以互相帮忙挂名发论文。总数量多些。
我真的很佩服他！无论是学术还是为人。但是鉴于本博不谈个人看法，仅仅书本与学习。就不多评论师兄了。
我和Zen（就用他自己常常用的签名）做的都是renewable energy building方面，我偏重于组件，他偏重于系统。我偏重于建模，他偏重于控制。他今天提了一点非常有启发性。我们应该继续发挥本小组的传统长项：fuzzy。我来建立PV的fuzzy模型，以及其他组件的fuzzy模型。他则整合模型，进行优化与控制。
这个想法让我很振奋！越想越振奋！暂时目标如下：
放假前把fuzzy model建立方法学习一下。同时，将手头的几篇文章里的传统确定性模型建出来。也就是，一方面了解系统本身，一方面学习新的工具。
对simbad进一步了解一下。并考虑用simbad完成华工的毕业论文……（偷懒了）
EASTER期间，用fuzzy的方法对前阶段建的模进行重新解构。同时，……那本太阳能的书要是到了就好了。这也是重点目标。
合作时处于被领导的角色，一点也没有别的想法。学会工作，学习如何领导研究工作。都是学习过程。和这样好的师兄共事，在这样好的导师下学习，真的是幸运极了！
Enjoy my life!
Enjoy my learning!
Enjoy every piece of happiness and success!
I will prove myself
I will earn self-confidence by every single achivement!
我真的很佩服他！无论是学术还是为人。但是鉴于本博不谈个人看法，仅仅书本与学习。就不多评论师兄了。
我和Zen（就用他自己常常用的签名）做的都是renewable energy building方面，我偏重于组件，他偏重于系统。我偏重于建模，他偏重于控制。他今天提了一点非常有启发性。我们应该继续发挥本小组的传统长项：fuzzy。我来建立PV的fuzzy模型，以及其他组件的fuzzy模型。他则整合模型，进行优化与控制。
这个想法让我很振奋！越想越振奋！暂时目标如下：
放假前把fuzzy model建立方法学习一下。同时，将手头的几篇文章里的传统确定性模型建出来。也就是，一方面了解系统本身，一方面学习新的工具。
对simbad进一步了解一下。并考虑用simbad完成华工的毕业论文……（偷懒了）
EASTER期间，用fuzzy的方法对前阶段建的模进行重新解构。同时，……那本太阳能的书要是到了就好了。这也是重点目标。
合作时处于被领导的角色，一点也没有别的想法。学会工作，学习如何领导研究工作。都是学习过程。和这样好的师兄共事，在这样好的导师下学习，真的是幸运极了！
Enjoy my life!
Enjoy my learning!
Enjoy every piece of happiness and success!
I will prove myself
I will earn self-confidence by every single achivement!
2007年1月17日星期三
model-based control and renewable energy
model-based control of renewable energy systems in buildings
Professor Dexter assigned me an essay to read the day I arrived, I used 6 days to finish it. first of all, I will summarize the main idea of the essay; then, I will state some of my crazy ideas about it; finally, I will conclude my reading.
What the essay discussed about is the model-based control in a building, but the new trend of the use of renewable energy is also concerned in the essay.
In the first part of the essay, the author discussed about the framework of the HVAC system. The HVAC system is devided into 2 parts----water system and air system. Though it seems to be a traditional way, but what makes it different is the use of SA/VPV. the VPV has four operational modes: preheating, storage(closed circulation), venting(open circuit), and bypass(turned off). This system provides heat and cold to 3 zones.I have some problem at the simulation diagram at first, but after consulting some problems with Zhen, and carefully read them with the equations, I think I can understood most of them, but I don't think I can design it with simulink. After then, the author analized the parameters used in the simulation and explain the contrained evolutionary strategy----repeated runs of evolutionary strategy will reduce the errors.
Then, the arthor presented the building and system model. The system is consist with 4 modules: I/O modules, water system modules, air system module and BEMS control module. information and mass are transfered among those module. These modules are not static models only, they are mixture of dynamic and steady-state component models.
Since the arthor have to compare their own strategy with the default BEMS system to show the advantage of their strategy, he explain both strategies.
BEMS system was based on a set of predefined rules and setpoints devised by setpoints. the difference between average space temperature and occupied set point will determine the suppy air temperature, if they need to use external energy, they will check the availability of VPV first, and then goes to the heat recovery, and if there still have difference, then the AHU will be activated. After the monitoring period, the design engineer remove the air quality control. The temperatures in the storage tank are important parameters, the high-level temperature determines the operation of boiler while the deployment of the evacuated tube collectors are determined by the low-level temperture.
The main approach taken by the Optimal Control Strategy is that using the unoccupied hours to devise a supervisory control trajectory for the next day, and during the runing hours, the control trajectory can be updated using a faster optimization algorithm. The objective of this strategy is to minimize the net external energy consumption within some contraints.
Finally, the author discussed the experimental results, from the charts we can find out that the optimal control can prove the indoor comfort in winter because the temperature is between 20'C to 22'C during 8 a.m. to 8 p.m while the BEMS system can only maintain the temperature within the range from 12 a.m. to 8 p.m. another different is about the water temperature. In BEMS systems, there is only one climax while there are two lacal maxmum temperature when Optimal Control is operated. On control sequence, the difference is that the operation in BEMS systems are mostly happened within occupied hours while the other is not. In the optimal control, components were enabled before the start of occupation. and this give me some idea about control schemes which i will discussed later.
The Optimal Control is proved to save a great amount of extenal energy.
After finished my reading, i am still interested in the idea of "plan first, adjust second". i think it is a bit different from prediction. it acts like human behavior, we make a plan in the night, then we try to finish it, when there is some unexpected things happened, we can adjust our plan but not design it immediately. I think maybe we can use matrix to deal with the problem. we can design matrix A,B,C.. for several typical situation, and everynight we can use the history data and the weather forecast to design another matrix a,b,c. we can calculate them and use the result as a plan. when the plan runs and we can gather the information of parameters and state from the I/O modules, and use them to develop a third matrix, but this one should have a limited range, so the calculation and simulation could be easier...
however, this idea hasn't been examined yet, because i think i need more knowledge about matrix before that. maybe i will lend a book from libaray and take the A4 course.
I think i should develop some reading skills because i find myself hard to get the most important ideas about the essay. recently, i find the only way for me to get a better understanding is to write a report about it. but maybe it will improve my writting as well.
Professor Dexter assigned me an essay to read the day I arrived, I used 6 days to finish it. first of all, I will summarize the main idea of the essay; then, I will state some of my crazy ideas about it; finally, I will conclude my reading.
What the essay discussed about is the model-based control in a building, but the new trend of the use of renewable energy is also concerned in the essay.
In the first part of the essay, the author discussed about the framework of the HVAC system. The HVAC system is devided into 2 parts----water system and air system. Though it seems to be a traditional way, but what makes it different is the use of SA/VPV. the VPV has four operational modes: preheating, storage(closed circulation), venting(open circuit), and bypass(turned off). This system provides heat and cold to 3 zones.I have some problem at the simulation diagram at first, but after consulting some problems with Zhen, and carefully read them with the equations, I think I can understood most of them, but I don't think I can design it with simulink. After then, the author analized the parameters used in the simulation and explain the contrained evolutionary strategy----repeated runs of evolutionary strategy will reduce the errors.
Then, the arthor presented the building and system model. The system is consist with 4 modules: I/O modules, water system modules, air system module and BEMS control module. information and mass are transfered among those module. These modules are not static models only, they are mixture of dynamic and steady-state component models.
Since the arthor have to compare their own strategy with the default BEMS system to show the advantage of their strategy, he explain both strategies.
BEMS system was based on a set of predefined rules and setpoints devised by setpoints. the difference between average space temperature and occupied set point will determine the suppy air temperature, if they need to use external energy, they will check the availability of VPV first, and then goes to the heat recovery, and if there still have difference, then the AHU will be activated. After the monitoring period, the design engineer remove the air quality control. The temperatures in the storage tank are important parameters, the high-level temperature determines the operation of boiler while the deployment of the evacuated tube collectors are determined by the low-level temperture.
The main approach taken by the Optimal Control Strategy is that using the unoccupied hours to devise a supervisory control trajectory for the next day, and during the runing hours, the control trajectory can be updated using a faster optimization algorithm. The objective of this strategy is to minimize the net external energy consumption within some contraints.
Finally, the author discussed the experimental results, from the charts we can find out that the optimal control can prove the indoor comfort in winter because the temperature is between 20'C to 22'C during 8 a.m. to 8 p.m while the BEMS system can only maintain the temperature within the range from 12 a.m. to 8 p.m. another different is about the water temperature. In BEMS systems, there is only one climax while there are two lacal maxmum temperature when Optimal Control is operated. On control sequence, the difference is that the operation in BEMS systems are mostly happened within occupied hours while the other is not. In the optimal control, components were enabled before the start of occupation. and this give me some idea about control schemes which i will discussed later.
The Optimal Control is proved to save a great amount of extenal energy.
After finished my reading, i am still interested in the idea of "plan first, adjust second". i think it is a bit different from prediction. it acts like human behavior, we make a plan in the night, then we try to finish it, when there is some unexpected things happened, we can adjust our plan but not design it immediately. I think maybe we can use matrix to deal with the problem. we can design matrix A,B,C.. for several typical situation, and everynight we can use the history data and the weather forecast to design another matrix a,b,c. we can calculate them and use the result as a plan. when the plan runs and we can gather the information of parameters and state from the I/O modules, and use them to develop a third matrix, but this one should have a limited range, so the calculation and simulation could be easier...
however, this idea hasn't been examined yet, because i think i need more knowledge about matrix before that. maybe i will lend a book from libaray and take the A4 course.
I think i should develop some reading skills because i find myself hard to get the most important ideas about the essay. recently, i find the only way for me to get a better understanding is to write a report about it. but maybe it will improve my writting as well.
2006年12月30日星期六
D的论文1
第一次开始写自己读文章的想法,开始难免幼稚肤浅.只能说,或许在space上是真性情,在校内网是嬉笑怒骂,在qq上是无病呻吟,在这里我是诚恳严肃地读书,学习,写下感受,一切与私人感受无关.
前天昨天看了将来的导师D的论文《a simplified physical model for estimating the average air temperature》(预测平均温度的简化物理模型).文章首先介绍了一些模拟辅助控制方案,然后介绍了错误探测判断方案以来于末端模拟技术,但是他们往往麻烦而不容易恰当地模拟.D曾经与同僚做过模糊控制,基因算法也在其中得到过应用,还有反馈等更常规的方法.基于建模的困难性,该文作者Z.L(应该是中国人,我注意到D和许多中国人共事过,我很怕给自己人丢脸,决心过去之后什么也不要想,好好学习与工作)设计了这样一种方案----建模简单,需要进行的计算以及对计算机的要求低,长时间的精确性也能够得到保证.它是通过建立一个"柔性控制",使反馈环路成为闭环环路,从而对热舒适以及锅炉都能达到控制效果.
文章建立了一个物理模型,首先,定义了一些参量,比如平均温度是加权平均温度,定义了加权系数,接着,通过建立循环的基本模型,得到散热器,空气,水等散热情况的控制方程(基本上都是微分方程形式).接着,由物理模型创建电子模型(这个词太不专业了...羞愧).
接着,对物理模型进行分析:模型有三个输入量:燃料消耗能量,外界温度,太阳辐射.方程中有10个物理参量.
对于新建建筑,往往设计资料和数据都比较全面,因此许多方法不适合新建筑.本方法可以通过一段时间对历史数据的学习,利用SIMPLEX技术(某非线性强制优化运算法则)来最小化其预测误差方根.
最后,该模型应用于实验室,居民住宅等,取得的数据与预测比较吻合.
该模型用于控制时,也取得了比较好的对室内温度的模拟结果.
读完了还是不是很了解,基础知识太匮乏了.
大致对D的工作有了初步肤浅的了解.
现在在读D和Z.L的另外一篇,涉及此篇没有讲到的对锅炉的模拟.
明天看了再写.
前天昨天看了将来的导师D的论文《a simplified physical model for estimating the average air temperature》(预测平均温度的简化物理模型).文章首先介绍了一些模拟辅助控制方案,然后介绍了错误探测判断方案以来于末端模拟技术,但是他们往往麻烦而不容易恰当地模拟.D曾经与同僚做过模糊控制,基因算法也在其中得到过应用,还有反馈等更常规的方法.基于建模的困难性,该文作者Z.L(应该是中国人,我注意到D和许多中国人共事过,我很怕给自己人丢脸,决心过去之后什么也不要想,好好学习与工作)设计了这样一种方案----建模简单,需要进行的计算以及对计算机的要求低,长时间的精确性也能够得到保证.它是通过建立一个"柔性控制",使反馈环路成为闭环环路,从而对热舒适以及锅炉都能达到控制效果.
文章建立了一个物理模型,首先,定义了一些参量,比如平均温度是加权平均温度,定义了加权系数,接着,通过建立循环的基本模型,得到散热器,空气,水等散热情况的控制方程(基本上都是微分方程形式).接着,由物理模型创建电子模型(这个词太不专业了...羞愧).
接着,对物理模型进行分析:模型有三个输入量:燃料消耗能量,外界温度,太阳辐射.方程中有10个物理参量.
对于新建建筑,往往设计资料和数据都比较全面,因此许多方法不适合新建筑.本方法可以通过一段时间对历史数据的学习,利用SIMPLEX技术(某非线性强制优化运算法则)来最小化其预测误差方根.
最后,该模型应用于实验室,居民住宅等,取得的数据与预测比较吻合.
该模型用于控制时,也取得了比较好的对室内温度的模拟结果.
读完了还是不是很了解,基础知识太匮乏了.
大致对D的工作有了初步肤浅的了解.
现在在读D和Z.L的另外一篇,涉及此篇没有讲到的对锅炉的模拟.
明天看了再写.
订阅：
博文 (Atom)
